WSL2-Linux-Kernel/tools/objtool/builtin-check.c

1207 строки
28 KiB
C
Исходник Обычный вид История

objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
/*
* Copyright (C) 2015 Josh Poimboeuf <jpoimboe@redhat.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, see <http://www.gnu.org/licenses/>.
*/
/*
* objtool check:
*
* This command analyzes every .o file and ensures the validity of its stack
* trace metadata. It enforces a set of rules on asm code and C inline
* assembly code so that stack traces can be reliable.
*
* For more information, see tools/objtool/Documentation/stack-validation.txt.
*/
#include <string.h>
#include <subcmd/parse-options.h>
#include "builtin.h"
#include "elf.h"
#include "special.h"
#include "arch.h"
#include "warn.h"
#include <linux/hashtable.h>
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
#define STATE_FP_SAVED 0x1
#define STATE_FP_SETUP 0x2
#define STATE_FENTRY 0x4
struct instruction {
struct list_head list;
struct hlist_node hash;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
struct section *sec;
unsigned long offset;
unsigned int len, state;
unsigned char type;
unsigned long immediate;
bool alt_group, visited;
struct symbol *call_dest;
struct instruction *jump_dest;
struct list_head alts;
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
struct symbol *func;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
};
struct alternative {
struct list_head list;
struct instruction *insn;
};
struct objtool_file {
struct elf *elf;
struct list_head insn_list;
DECLARE_HASHTABLE(insn_hash, 16);
struct section *rodata, *whitelist;
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
bool ignore_unreachables, c_file;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
};
const char *objname;
static bool nofp;
static struct instruction *find_insn(struct objtool_file *file,
struct section *sec, unsigned long offset)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
struct instruction *insn;
hash_for_each_possible(file->insn_hash, insn, hash, offset)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->sec == sec && insn->offset == offset)
return insn;
return NULL;
}
static struct instruction *next_insn_same_sec(struct objtool_file *file,
struct instruction *insn)
{
struct instruction *next = list_next_entry(insn, list);
if (&next->list == &file->insn_list || next->sec != insn->sec)
return NULL;
return next;
}
#define for_each_insn(file, insn) \
list_for_each_entry(insn, &file->insn_list, list)
#define func_for_each_insn(file, func, insn) \
for (insn = find_insn(file, func->sec, func->offset); \
insn && &insn->list != &file->insn_list && \
insn->sec == func->sec && \
insn->offset < func->offset + func->len; \
insn = list_next_entry(insn, list))
#define sec_for_each_insn_from(file, insn) \
for (; insn; insn = next_insn_same_sec(file, insn))
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
/*
* Check if the function has been manually whitelisted with the
* STACK_FRAME_NON_STANDARD macro, or if it should be automatically whitelisted
* due to its use of a context switching instruction.
*/
static bool ignore_func(struct objtool_file *file, struct symbol *func)
{
struct rela *rela;
struct instruction *insn;
/* check for STACK_FRAME_NON_STANDARD */
if (file->whitelist && file->whitelist->rela)
list_for_each_entry(rela, &file->whitelist->rela->rela_list, list)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (rela->sym->sec == func->sec &&
rela->addend == func->offset)
return true;
/* check if it has a context switching instruction */
func_for_each_insn(file, func, insn)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->type == INSN_CONTEXT_SWITCH)
return true;
return false;
}
/*
* This checks to see if the given function is a "noreturn" function.
*
* For global functions which are outside the scope of this object file, we
* have to keep a manual list of them.
*
* For local functions, we have to detect them manually by simply looking for
* the lack of a return instruction.
*
* Returns:
* -1: error
* 0: no dead end
* 1: dead end
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
*/
static int __dead_end_function(struct objtool_file *file, struct symbol *func,
int recursion)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
int i;
struct instruction *insn;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
bool empty = true;
/*
* Unfortunately these have to be hard coded because the noreturn
* attribute isn't provided in ELF data.
*/
static const char * const global_noreturns[] = {
"__stack_chk_fail",
"panic",
"do_exit",
"__module_put_and_exit",
"complete_and_exit",
"kvm_spurious_fault",
"__reiserfs_panic",
"lbug_with_loc"
};
if (func->bind == STB_WEAK)
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (func->bind == STB_GLOBAL)
for (i = 0; i < ARRAY_SIZE(global_noreturns); i++)
if (!strcmp(func->name, global_noreturns[i]))
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (!func->sec)
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
func_for_each_insn(file, func, insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
empty = false;
if (insn->type == INSN_RETURN)
return 0;
objtool: Prevent infinite recursion in noreturn detection Ingo reported an infinite loop in objtool with a certain randconfig [1]. With the given config, two functions in crypto/ablkcipher.o contained sibling calls to each other, which threw the recursive call in dead_end_function() for a loop (literally!). Split the noreturn detection into two passes. In the first pass, check for return instructions. In the second pass, do the potentially recursive sibling call check. In most cases, the first pass will be good enough. In the rare case where a second pass is needed, recursion should hopefully no longer be possible. [1] https://lkml.kernel.org/r/20160308154909.GA20956@gmail.com Reported-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/16afb602640ef43b7782087d6cca17bf6fc13603.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:51 +03:00
}
if (empty)
return 0;
objtool: Prevent infinite recursion in noreturn detection Ingo reported an infinite loop in objtool with a certain randconfig [1]. With the given config, two functions in crypto/ablkcipher.o contained sibling calls to each other, which threw the recursive call in dead_end_function() for a loop (literally!). Split the noreturn detection into two passes. In the first pass, check for return instructions. In the second pass, do the potentially recursive sibling call check. In most cases, the first pass will be good enough. In the rare case where a second pass is needed, recursion should hopefully no longer be possible. [1] https://lkml.kernel.org/r/20160308154909.GA20956@gmail.com Reported-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/16afb602640ef43b7782087d6cca17bf6fc13603.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:51 +03:00
/*
* A function can have a sibling call instead of a return. In that
* case, the function's dead-end status depends on whether the target
* of the sibling call returns.
*/
func_for_each_insn(file, func, insn) {
objtool: Prevent infinite recursion in noreturn detection Ingo reported an infinite loop in objtool with a certain randconfig [1]. With the given config, two functions in crypto/ablkcipher.o contained sibling calls to each other, which threw the recursive call in dead_end_function() for a loop (literally!). Split the noreturn detection into two passes. In the first pass, check for return instructions. In the second pass, do the potentially recursive sibling call check. In most cases, the first pass will be good enough. In the rare case where a second pass is needed, recursion should hopefully no longer be possible. [1] https://lkml.kernel.org/r/20160308154909.GA20956@gmail.com Reported-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/16afb602640ef43b7782087d6cca17bf6fc13603.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:51 +03:00
if (insn->sec != func->sec ||
insn->offset >= func->offset + func->len)
break;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->type == INSN_JUMP_UNCONDITIONAL) {
struct instruction *dest = insn->jump_dest;
struct symbol *dest_func;
if (!dest)
/* sibling call to another file */
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (dest->sec != func->sec ||
dest->offset < func->offset ||
dest->offset >= func->offset + func->len) {
/* local sibling call */
dest_func = find_symbol_by_offset(dest->sec,
dest->offset);
if (!dest_func)
continue;
if (recursion == 5) {
WARN_FUNC("infinite recursion (objtool bug!)",
dest->sec, dest->offset);
return -1;
}
return __dead_end_function(file, dest_func,
recursion + 1);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
}
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
if (insn->type == INSN_JUMP_DYNAMIC && list_empty(&insn->alts))
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
/* sibling call */
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
return 1;
}
static int dead_end_function(struct objtool_file *file, struct symbol *func)
{
return __dead_end_function(file, func, 0);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
/*
* Call the arch-specific instruction decoder for all the instructions and add
* them to the global instruction list.
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
*/
static int decode_instructions(struct objtool_file *file)
{
struct section *sec;
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
struct symbol *func;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
unsigned long offset;
struct instruction *insn;
int ret;
list_for_each_entry(sec, &file->elf->sections, list) {
if (!(sec->sh.sh_flags & SHF_EXECINSTR))
continue;
for (offset = 0; offset < sec->len; offset += insn->len) {
insn = malloc(sizeof(*insn));
memset(insn, 0, sizeof(*insn));
INIT_LIST_HEAD(&insn->alts);
insn->sec = sec;
insn->offset = offset;
ret = arch_decode_instruction(file->elf, sec, offset,
sec->len - offset,
&insn->len, &insn->type,
&insn->immediate);
if (ret)
return ret;
if (!insn->type || insn->type > INSN_LAST) {
WARN_FUNC("invalid instruction type %d",
insn->sec, insn->offset, insn->type);
return -1;
}
hash_add(file->insn_hash, &insn->hash, insn->offset);
list_add_tail(&insn->list, &file->insn_list);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
list_for_each_entry(func, &sec->symbol_list, list) {
if (func->type != STT_FUNC)
continue;
if (!find_insn(file, sec, func->offset)) {
WARN("%s(): can't find starting instruction",
func->name);
return -1;
}
func_for_each_insn(file, func, insn)
if (!insn->func)
insn->func = func;
}
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
return 0;
}
/*
* Warnings shouldn't be reported for ignored functions.
*/
static void add_ignores(struct objtool_file *file)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
struct instruction *insn;
struct section *sec;
struct symbol *func;
list_for_each_entry(sec, &file->elf->sections, list) {
list_for_each_entry(func, &sec->symbol_list, list) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (func->type != STT_FUNC)
continue;
if (!ignore_func(file, func))
continue;
func_for_each_insn(file, func, insn)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
insn->visited = true;
}
}
}
/*
* Find the destination instructions for all jumps.
*/
static int add_jump_destinations(struct objtool_file *file)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
struct instruction *insn;
struct rela *rela;
struct section *dest_sec;
unsigned long dest_off;
for_each_insn(file, insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->type != INSN_JUMP_CONDITIONAL &&
insn->type != INSN_JUMP_UNCONDITIONAL)
continue;
/* skip ignores */
if (insn->visited)
continue;
rela = find_rela_by_dest_range(insn->sec, insn->offset,
insn->len);
if (!rela) {
dest_sec = insn->sec;
dest_off = insn->offset + insn->len + insn->immediate;
} else if (rela->sym->type == STT_SECTION) {
dest_sec = rela->sym->sec;
dest_off = rela->addend + 4;
} else if (rela->sym->sec->idx) {
dest_sec = rela->sym->sec;
dest_off = rela->sym->sym.st_value + rela->addend + 4;
} else {
/* sibling call */
insn->jump_dest = 0;
continue;
}
insn->jump_dest = find_insn(file, dest_sec, dest_off);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (!insn->jump_dest) {
/*
* This is a special case where an alt instruction
* jumps past the end of the section. These are
* handled later in handle_group_alt().
*/
if (!strcmp(insn->sec->name, ".altinstr_replacement"))
continue;
WARN_FUNC("can't find jump dest instruction at %s+0x%lx",
insn->sec, insn->offset, dest_sec->name,
dest_off);
return -1;
}
}
return 0;
}
/*
* Find the destination instructions for all calls.
*/
static int add_call_destinations(struct objtool_file *file)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
struct instruction *insn;
unsigned long dest_off;
struct rela *rela;
for_each_insn(file, insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->type != INSN_CALL)
continue;
rela = find_rela_by_dest_range(insn->sec, insn->offset,
insn->len);
if (!rela) {
dest_off = insn->offset + insn->len + insn->immediate;
insn->call_dest = find_symbol_by_offset(insn->sec,
dest_off);
if (!insn->call_dest) {
WARN_FUNC("can't find call dest symbol at offset 0x%lx",
insn->sec, insn->offset, dest_off);
return -1;
}
} else if (rela->sym->type == STT_SECTION) {
insn->call_dest = find_symbol_by_offset(rela->sym->sec,
rela->addend+4);
if (!insn->call_dest ||
insn->call_dest->type != STT_FUNC) {
WARN_FUNC("can't find call dest symbol at %s+0x%x",
insn->sec, insn->offset,
rela->sym->sec->name,
rela->addend + 4);
return -1;
}
} else
insn->call_dest = rela->sym;
}
return 0;
}
/*
* The .alternatives section requires some extra special care, over and above
* what other special sections require:
*
* 1. Because alternatives are patched in-place, we need to insert a fake jump
* instruction at the end so that validate_branch() skips all the original
* replaced instructions when validating the new instruction path.
*
* 2. An added wrinkle is that the new instruction length might be zero. In
* that case the old instructions are replaced with noops. We simulate that
* by creating a fake jump as the only new instruction.
*
* 3. In some cases, the alternative section includes an instruction which
* conditionally jumps to the _end_ of the entry. We have to modify these
* jumps' destinations to point back to .text rather than the end of the
* entry in .altinstr_replacement.
*
* 4. It has been requested that we don't validate the !POPCNT feature path
* which is a "very very small percentage of machines".
*/
static int handle_group_alt(struct objtool_file *file,
struct special_alt *special_alt,
struct instruction *orig_insn,
struct instruction **new_insn)
{
struct instruction *last_orig_insn, *last_new_insn, *insn, *fake_jump;
unsigned long dest_off;
last_orig_insn = NULL;
insn = orig_insn;
sec_for_each_insn_from(file, insn) {
if (insn->offset >= special_alt->orig_off + special_alt->orig_len)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
break;
if (special_alt->skip_orig)
insn->type = INSN_NOP;
insn->alt_group = true;
last_orig_insn = insn;
}
if (!next_insn_same_sec(file, last_orig_insn)) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
WARN("%s: don't know how to handle alternatives at end of section",
special_alt->orig_sec->name);
return -1;
}
fake_jump = malloc(sizeof(*fake_jump));
if (!fake_jump) {
WARN("malloc failed");
return -1;
}
memset(fake_jump, 0, sizeof(*fake_jump));
INIT_LIST_HEAD(&fake_jump->alts);
fake_jump->sec = special_alt->new_sec;
fake_jump->offset = -1;
fake_jump->type = INSN_JUMP_UNCONDITIONAL;
fake_jump->jump_dest = list_next_entry(last_orig_insn, list);
if (!special_alt->new_len) {
*new_insn = fake_jump;
return 0;
}
last_new_insn = NULL;
insn = *new_insn;
sec_for_each_insn_from(file, insn) {
if (insn->offset >= special_alt->new_off + special_alt->new_len)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
break;
last_new_insn = insn;
if (insn->type != INSN_JUMP_CONDITIONAL &&
insn->type != INSN_JUMP_UNCONDITIONAL)
continue;
if (!insn->immediate)
continue;
dest_off = insn->offset + insn->len + insn->immediate;
if (dest_off == special_alt->new_off + special_alt->new_len)
insn->jump_dest = fake_jump;
if (!insn->jump_dest) {
WARN_FUNC("can't find alternative jump destination",
insn->sec, insn->offset);
return -1;
}
}
if (!last_new_insn) {
WARN_FUNC("can't find last new alternative instruction",
special_alt->new_sec, special_alt->new_off);
return -1;
}
list_add(&fake_jump->list, &last_new_insn->list);
return 0;
}
/*
* A jump table entry can either convert a nop to a jump or a jump to a nop.
* If the original instruction is a jump, make the alt entry an effective nop
* by just skipping the original instruction.
*/
static int handle_jump_alt(struct objtool_file *file,
struct special_alt *special_alt,
struct instruction *orig_insn,
struct instruction **new_insn)
{
if (orig_insn->type == INSN_NOP)
return 0;
if (orig_insn->type != INSN_JUMP_UNCONDITIONAL) {
WARN_FUNC("unsupported instruction at jump label",
orig_insn->sec, orig_insn->offset);
return -1;
}
*new_insn = list_next_entry(orig_insn, list);
return 0;
}
/*
* Read all the special sections which have alternate instructions which can be
* patched in or redirected to at runtime. Each instruction having alternate
* instruction(s) has them added to its insn->alts list, which will be
* traversed in validate_branch().
*/
static int add_special_section_alts(struct objtool_file *file)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
struct list_head special_alts;
struct instruction *orig_insn, *new_insn;
struct special_alt *special_alt, *tmp;
struct alternative *alt;
int ret;
ret = special_get_alts(file->elf, &special_alts);
if (ret)
return ret;
list_for_each_entry_safe(special_alt, tmp, &special_alts, list) {
alt = malloc(sizeof(*alt));
if (!alt) {
WARN("malloc failed");
ret = -1;
goto out;
}
orig_insn = find_insn(file, special_alt->orig_sec,
special_alt->orig_off);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (!orig_insn) {
WARN_FUNC("special: can't find orig instruction",
special_alt->orig_sec, special_alt->orig_off);
ret = -1;
goto out;
}
new_insn = NULL;
if (!special_alt->group || special_alt->new_len) {
new_insn = find_insn(file, special_alt->new_sec,
special_alt->new_off);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (!new_insn) {
WARN_FUNC("special: can't find new instruction",
special_alt->new_sec,
special_alt->new_off);
ret = -1;
goto out;
}
}
if (special_alt->group) {
ret = handle_group_alt(file, special_alt, orig_insn,
&new_insn);
if (ret)
goto out;
} else if (special_alt->jump_or_nop) {
ret = handle_jump_alt(file, special_alt, orig_insn,
&new_insn);
if (ret)
goto out;
}
alt->insn = new_insn;
list_add_tail(&alt->list, &orig_insn->alts);
list_del(&special_alt->list);
free(special_alt);
}
out:
return ret;
}
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
static int add_switch_table(struct objtool_file *file, struct symbol *func,
struct instruction *insn, struct rela *table,
struct rela *next_table)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
struct rela *rela = table;
struct instruction *alt_insn;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
struct alternative *alt;
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
list_for_each_entry_from(rela, &file->rodata->rela->rela_list, list) {
if (rela == next_table)
break;
if (rela->sym->sec != insn->sec ||
rela->addend <= func->offset ||
rela->addend >= func->offset + func->len)
break;
alt_insn = find_insn(file, insn->sec, rela->addend);
if (!alt_insn) {
WARN("%s: can't find instruction at %s+0x%x",
file->rodata->rela->name, insn->sec->name,
rela->addend);
return -1;
}
alt = malloc(sizeof(*alt));
if (!alt) {
WARN("malloc failed");
return -1;
}
alt->insn = alt_insn;
list_add_tail(&alt->list, &insn->alts);
}
return 0;
}
static int add_func_switch_tables(struct objtool_file *file,
struct symbol *func)
{
struct instruction *insn, *prev_jump;
struct rela *text_rela, *rodata_rela, *prev_rela;
int ret;
prev_jump = NULL;
func_for_each_insn(file, func, insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->type != INSN_JUMP_DYNAMIC)
continue;
text_rela = find_rela_by_dest_range(insn->sec, insn->offset,
insn->len);
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
if (!text_rela || text_rela->sym != file->rodata->sym)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
continue;
/* common case: jmpq *[addr](,%rax,8) */
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
rodata_rela = find_rela_by_dest(file->rodata,
text_rela->addend);
/*
* rare case: jmpq *[addr](%rip)
objtool: Add workaround for GCC switch jump table bug GCC has a rare quirk, currently only seen in three driver functions in the kernel, and only with certain obscure non-distro configs, which can cause objtool to produce "unreachable instruction" false positive warnings. As part of an optimization, GCC makes a copy of an existing switch jump table, modifies it, and then hard-codes the jump (albeit with an indirect jump) to use a single entry in the table. The rest of the jump table and some of its jump targets remain as dead code. In such a case we can just crudely ignore all unreachable instruction warnings for the entire object file. Ideally we would just ignore them for the function, but that would require redesigning the code quite a bit. And honestly that's just not worth doing: unreachable instruction warnings are of questionable value anyway, and this is a very rare issue. kbuild reports: https://lkml.kernel.org/r/201603231906.LWcVUpxm%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603271114.K9i45biy%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603291058.zuJ6ben1%25fengguang.wu@intel.com GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70604 Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/700fa029bbb0feff34f03ffc69d666a3c3b57a61.1460663532.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-14 22:52:24 +03:00
*
* This check is for a rare gcc quirk, currently only seen in
* three driver functions in the kernel, only with certain
* obscure non-distro configs.
*
* As part of an optimization, gcc makes a copy of an existing
* switch jump table, modifies it, and then hard-codes the jump
* (albeit with an indirect jump) to use a single entry in the
* table. The rest of the jump table and some of its jump
* targets remain as dead code.
*
* In such a case we can just crudely ignore all unreachable
* instruction warnings for the entire object file. Ideally we
* would just ignore them for the function, but that would
* require redesigning the code quite a bit. And honestly
* that's just not worth doing: unreachable instruction
* warnings are of questionable value anyway, and this is such
* a rare issue.
*
* kbuild reports:
* - https://lkml.kernel.org/r/201603231906.LWcVUpxm%25fengguang.wu@intel.com
* - https://lkml.kernel.org/r/201603271114.K9i45biy%25fengguang.wu@intel.com
* - https://lkml.kernel.org/r/201603291058.zuJ6ben1%25fengguang.wu@intel.com
*
* gcc bug:
* - https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70604
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
*/
objtool: Add workaround for GCC switch jump table bug GCC has a rare quirk, currently only seen in three driver functions in the kernel, and only with certain obscure non-distro configs, which can cause objtool to produce "unreachable instruction" false positive warnings. As part of an optimization, GCC makes a copy of an existing switch jump table, modifies it, and then hard-codes the jump (albeit with an indirect jump) to use a single entry in the table. The rest of the jump table and some of its jump targets remain as dead code. In such a case we can just crudely ignore all unreachable instruction warnings for the entire object file. Ideally we would just ignore them for the function, but that would require redesigning the code quite a bit. And honestly that's just not worth doing: unreachable instruction warnings are of questionable value anyway, and this is a very rare issue. kbuild reports: https://lkml.kernel.org/r/201603231906.LWcVUpxm%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603271114.K9i45biy%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603291058.zuJ6ben1%25fengguang.wu@intel.com GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70604 Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/700fa029bbb0feff34f03ffc69d666a3c3b57a61.1460663532.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-14 22:52:24 +03:00
if (!rodata_rela) {
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
rodata_rela = find_rela_by_dest(file->rodata,
text_rela->addend + 4);
objtool: Add workaround for GCC switch jump table bug GCC has a rare quirk, currently only seen in three driver functions in the kernel, and only with certain obscure non-distro configs, which can cause objtool to produce "unreachable instruction" false positive warnings. As part of an optimization, GCC makes a copy of an existing switch jump table, modifies it, and then hard-codes the jump (albeit with an indirect jump) to use a single entry in the table. The rest of the jump table and some of its jump targets remain as dead code. In such a case we can just crudely ignore all unreachable instruction warnings for the entire object file. Ideally we would just ignore them for the function, but that would require redesigning the code quite a bit. And honestly that's just not worth doing: unreachable instruction warnings are of questionable value anyway, and this is a very rare issue. kbuild reports: https://lkml.kernel.org/r/201603231906.LWcVUpxm%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603271114.K9i45biy%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603291058.zuJ6ben1%25fengguang.wu@intel.com GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70604 Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/700fa029bbb0feff34f03ffc69d666a3c3b57a61.1460663532.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-14 22:52:24 +03:00
if (rodata_rela)
file->ignore_unreachables = true;
}
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
if (!rodata_rela)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
continue;
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
/*
* We found a switch table, but we don't know yet how big it
* is. Don't add it until we reach the end of the function or
* the beginning of another switch table in the same function.
*/
if (prev_jump) {
ret = add_switch_table(file, func, prev_jump, prev_rela,
rodata_rela);
if (ret)
return ret;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
prev_jump = insn;
prev_rela = rodata_rela;
}
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
if (prev_jump) {
ret = add_switch_table(file, func, prev_jump, prev_rela, NULL);
if (ret)
return ret;
}
return 0;
}
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
objtool: Fix false positive warnings for functions with multiple switch statements Ingo reported [1] some false positive objtool warnings: drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2e7: frame pointer state mismatch drivers/net/wireless/realtek/rtlwifi/base.o: warning: objtool: rtlwifi_rate_mapping()+0x2f3: frame pointer state mismatch ... And so did the 0-day bot [2]: drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x6ce: call without frame pointer save/setup drivers/gpu/drm/radeon/cik.o: warning: objtool: cik_tiling_mode_table_init()+0x72b: call without frame pointer save/setup ... Both sets of warnings involve functions which have multiple switch statements. When there's more than one switch statement in a function, objtool interprets all the switch jump tables as a single table. If the targets of one jump table assume a stack frame and the targets of another one don't, it prints false positive warnings. Fix the bug by detecting the size of each switch jump table. For multiple tables, each one ends where the next one begins. [1] https://lkml.kernel.org/r/20160308103716.GA9618@gmail.com [2] https://lists.01.org/pipermail/kbuild-all/2016-March/018124.html Reported-by: Ingo Molnar <mingo@kernel.org> Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/2d7eecc6bc52d301f494b80f5fd62c2b6c895658.1457502970.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-03-09 09:06:58 +03:00
/*
* For some switch statements, gcc generates a jump table in the .rodata
* section which contains a list of addresses within the function to jump to.
* This finds these jump tables and adds them to the insn->alts lists.
*/
static int add_switch_table_alts(struct objtool_file *file)
{
struct section *sec;
struct symbol *func;
int ret;
if (!file->rodata || !file->rodata->rela)
return 0;
list_for_each_entry(sec, &file->elf->sections, list) {
list_for_each_entry(func, &sec->symbol_list, list) {
if (func->type != STT_FUNC)
continue;
ret = add_func_switch_tables(file, func);
if (ret)
return ret;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
}
return 0;
}
static int decode_sections(struct objtool_file *file)
{
int ret;
ret = decode_instructions(file);
if (ret)
return ret;
add_ignores(file);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
ret = add_jump_destinations(file);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (ret)
return ret;
ret = add_call_destinations(file);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (ret)
return ret;
ret = add_special_section_alts(file);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (ret)
return ret;
ret = add_switch_table_alts(file);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (ret)
return ret;
return 0;
}
static bool is_fentry_call(struct instruction *insn)
{
if (insn->type == INSN_CALL &&
insn->call_dest->type == STT_NOTYPE &&
!strcmp(insn->call_dest->name, "__fentry__"))
return true;
return false;
}
static bool has_modified_stack_frame(struct instruction *insn)
{
return (insn->state & STATE_FP_SAVED) ||
(insn->state & STATE_FP_SETUP);
}
static bool has_valid_stack_frame(struct instruction *insn)
{
return (insn->state & STATE_FP_SAVED) &&
(insn->state & STATE_FP_SETUP);
}
static unsigned int frame_state(unsigned long state)
{
return (state & (STATE_FP_SAVED | STATE_FP_SETUP));
}
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
/*
* Follow the branch starting at the given instruction, and recursively follow
* any other branches (jumps). Meanwhile, track the frame pointer state at
* each instruction and validate all the rules described in
* tools/objtool/Documentation/stack-validation.txt.
*/
static int validate_branch(struct objtool_file *file,
struct instruction *first, unsigned char first_state)
{
struct alternative *alt;
struct instruction *insn;
struct section *sec;
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
struct symbol *func = NULL;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
unsigned char state;
int ret;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
insn = first;
sec = insn->sec;
state = first_state;
if (insn->alt_group && list_empty(&insn->alts)) {
WARN_FUNC("don't know how to handle branch to middle of alternative instruction group",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
while (1) {
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
if (file->c_file && insn->func) {
if (func && func != insn->func) {
WARN("%s() falls through to next function %s()",
func->name, insn->func->name);
return 1;
}
func = insn->func;
}
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->visited) {
if (frame_state(insn->state) != frame_state(state)) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
WARN_FUNC("frame pointer state mismatch",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
insn->visited = true;
insn->state = state;
list_for_each_entry(alt, &insn->alts, list) {
ret = validate_branch(file, alt->insn, state);
if (ret)
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
switch (insn->type) {
case INSN_FP_SAVE:
if (!nofp) {
if (state & STATE_FP_SAVED) {
WARN_FUNC("duplicate frame pointer save",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
state |= STATE_FP_SAVED;
}
break;
case INSN_FP_SETUP:
if (!nofp) {
if (state & STATE_FP_SETUP) {
WARN_FUNC("duplicate frame pointer setup",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
state |= STATE_FP_SETUP;
}
break;
case INSN_FP_RESTORE:
if (!nofp) {
if (has_valid_stack_frame(insn))
state &= ~STATE_FP_SETUP;
state &= ~STATE_FP_SAVED;
}
break;
case INSN_RETURN:
if (!nofp && has_modified_stack_frame(insn)) {
WARN_FUNC("return without frame pointer restore",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
case INSN_CALL:
if (is_fentry_call(insn)) {
state |= STATE_FENTRY;
break;
}
ret = dead_end_function(file, insn->call_dest);
if (ret == 1)
return 0;
if (ret == -1)
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
/* fallthrough */
case INSN_CALL_DYNAMIC:
if (!nofp && !has_valid_stack_frame(insn)) {
WARN_FUNC("call without frame pointer save/setup",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
break;
case INSN_JUMP_CONDITIONAL:
case INSN_JUMP_UNCONDITIONAL:
if (insn->jump_dest) {
ret = validate_branch(file, insn->jump_dest,
state);
if (ret)
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
} else if (has_modified_stack_frame(insn)) {
WARN_FUNC("sibling call from callable instruction with changed frame pointer",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
} /* else it's a sibling call */
if (insn->type == INSN_JUMP_UNCONDITIONAL)
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
break;
case INSN_JUMP_DYNAMIC:
if (list_empty(&insn->alts) &&
has_modified_stack_frame(insn)) {
WARN_FUNC("sibling call from callable instruction with changed frame pointer",
sec, insn->offset);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
case INSN_BUG:
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
default:
break;
}
insn = next_insn_same_sec(file, insn);
if (!insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
WARN("%s: unexpected end of section", sec->name);
return 1;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
}
return 0;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
static bool is_gcov_insn(struct instruction *insn)
{
struct rela *rela;
struct section *sec;
struct symbol *sym;
unsigned long offset;
rela = find_rela_by_dest_range(insn->sec, insn->offset, insn->len);
if (!rela)
return false;
if (rela->sym->type != STT_SECTION)
return false;
sec = rela->sym->sec;
offset = rela->addend + insn->offset + insn->len - rela->offset;
list_for_each_entry(sym, &sec->symbol_list, list) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (sym->type != STT_OBJECT)
continue;
if (offset >= sym->offset && offset < sym->offset + sym->len)
return (!memcmp(sym->name, "__gcov0.", 8));
}
return false;
}
static bool is_kasan_insn(struct instruction *insn)
{
return (insn->type == INSN_CALL &&
!strcmp(insn->call_dest->name, "__asan_handle_no_return"));
}
static bool is_ubsan_insn(struct instruction *insn)
{
return (insn->type == INSN_CALL &&
!strcmp(insn->call_dest->name,
"__ubsan_handle_builtin_unreachable"));
}
static bool ignore_unreachable_insn(struct symbol *func,
struct instruction *insn)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
{
int i;
if (insn->type == INSN_NOP)
return true;
if (is_gcov_insn(insn))
return true;
/*
* Check if this (or a subsequent) instruction is related to
* CONFIG_UBSAN or CONFIG_KASAN.
*
* End the search at 5 instructions to avoid going into the weeds.
*/
for (i = 0; i < 5; i++) {
if (is_kasan_insn(insn) || is_ubsan_insn(insn))
return true;
if (insn->type == INSN_JUMP_UNCONDITIONAL && insn->jump_dest) {
insn = insn->jump_dest;
continue;
}
if (insn->offset + insn->len >= func->offset + func->len)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
break;
insn = list_next_entry(insn, list);
}
return false;
}
static int validate_functions(struct objtool_file *file)
{
struct section *sec;
struct symbol *func;
struct instruction *insn;
int ret, warnings = 0;
list_for_each_entry(sec, &file->elf->sections, list) {
list_for_each_entry(func, &sec->symbol_list, list) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (func->type != STT_FUNC)
continue;
insn = find_insn(file, sec, func->offset);
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
if (!insn)
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
continue;
ret = validate_branch(file, insn, 0);
warnings += ret;
}
}
list_for_each_entry(sec, &file->elf->sections, list) {
list_for_each_entry(func, &sec->symbol_list, list) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (func->type != STT_FUNC)
continue;
func_for_each_insn(file, func, insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (insn->visited)
continue;
insn->visited = true;
objtool: Add workaround for GCC switch jump table bug GCC has a rare quirk, currently only seen in three driver functions in the kernel, and only with certain obscure non-distro configs, which can cause objtool to produce "unreachable instruction" false positive warnings. As part of an optimization, GCC makes a copy of an existing switch jump table, modifies it, and then hard-codes the jump (albeit with an indirect jump) to use a single entry in the table. The rest of the jump table and some of its jump targets remain as dead code. In such a case we can just crudely ignore all unreachable instruction warnings for the entire object file. Ideally we would just ignore them for the function, but that would require redesigning the code quite a bit. And honestly that's just not worth doing: unreachable instruction warnings are of questionable value anyway, and this is a very rare issue. kbuild reports: https://lkml.kernel.org/r/201603231906.LWcVUpxm%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603271114.K9i45biy%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603291058.zuJ6ben1%25fengguang.wu@intel.com GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70604 Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/700fa029bbb0feff34f03ffc69d666a3c3b57a61.1460663532.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-14 22:52:24 +03:00
if (file->ignore_unreachables || warnings ||
ignore_unreachable_insn(func, insn))
continue;
WARN_FUNC("function has unreachable instruction", insn->sec, insn->offset);
warnings++;
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
}
}
}
return warnings;
}
static int validate_uncallable_instructions(struct objtool_file *file)
{
struct instruction *insn;
int warnings = 0;
for_each_insn(file, insn) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
if (!insn->visited && insn->type == INSN_RETURN) {
WARN_FUNC("return instruction outside of a callable function",
insn->sec, insn->offset);
warnings++;
}
}
return warnings;
}
static void cleanup(struct objtool_file *file)
{
struct instruction *insn, *tmpinsn;
struct alternative *alt, *tmpalt;
list_for_each_entry_safe(insn, tmpinsn, &file->insn_list, list) {
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
list_for_each_entry_safe(alt, tmpalt, &insn->alts, list) {
list_del(&alt->list);
free(alt);
}
list_del(&insn->list);
hash_del(&insn->hash);
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
free(insn);
}
elf_close(file->elf);
}
const char * const check_usage[] = {
"objtool check [<options>] file.o",
NULL,
};
int cmd_check(int argc, const char **argv)
{
struct objtool_file file;
int ret, warnings = 0;
const struct option options[] = {
OPT_BOOLEAN('f', "no-fp", &nofp, "Skip frame pointer validation"),
OPT_END(),
};
argc = parse_options(argc, argv, options, check_usage, 0);
if (argc != 1)
usage_with_options(check_usage, options);
objname = argv[0];
file.elf = elf_open(objname);
if (!file.elf) {
fprintf(stderr, "error reading elf file %s\n", objname);
return 1;
}
INIT_LIST_HEAD(&file.insn_list);
hash_init(file.insn_hash);
objtool: Add workaround for GCC switch jump table bug GCC has a rare quirk, currently only seen in three driver functions in the kernel, and only with certain obscure non-distro configs, which can cause objtool to produce "unreachable instruction" false positive warnings. As part of an optimization, GCC makes a copy of an existing switch jump table, modifies it, and then hard-codes the jump (albeit with an indirect jump) to use a single entry in the table. The rest of the jump table and some of its jump targets remain as dead code. In such a case we can just crudely ignore all unreachable instruction warnings for the entire object file. Ideally we would just ignore them for the function, but that would require redesigning the code quite a bit. And honestly that's just not worth doing: unreachable instruction warnings are of questionable value anyway, and this is a very rare issue. kbuild reports: https://lkml.kernel.org/r/201603231906.LWcVUpxm%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603271114.K9i45biy%25fengguang.wu@intel.com https://lkml.kernel.org/r/201603291058.zuJ6ben1%25fengguang.wu@intel.com GCC bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70604 Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/700fa029bbb0feff34f03ffc69d666a3c3b57a61.1460663532.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-14 22:52:24 +03:00
file.whitelist = find_section_by_name(file.elf, "__func_stack_frame_non_standard");
file.rodata = find_section_by_name(file.elf, ".rodata");
file.ignore_unreachables = false;
objtool: Detect falling through to the next function There are several cases in compiled C code where a function may not return at the end, and may instead fall through to the next function. That may indicate a bug in the code, or a gcc bug, or even an objtool bug. But in each case, objtool reports an unhelpful warning, something like: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: duplicate frame pointer save drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_fc_host_stats()+0x0: frame pointer state mismatch Detect this situation and print a more useful error message: drivers/scsi/qla2xxx/qla_attr.o: warning: objtool: qla2x00_get_host_fabric_name() falls through to next function qla2x00_get_starget_node_name() Also add some information about this warning and its potential causes to the documentation. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/caa4ec6c687931db805e692d4e4bf06cd87d33e6.1460729697.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-04-15 17:17:10 +03:00
file.c_file = find_section_by_name(file.elf, ".comment");
objtool: Add tool to perform compile-time stack metadata validation This adds a host tool named objtool which has a "check" subcommand which analyzes .o files to ensure the validity of stack metadata. It enforces a set of rules on asm code and C inline assembly code so that stack traces can be reliable. For each function, it recursively follows all possible code paths and validates the correct frame pointer state at each instruction. It also follows code paths involving kernel special sections, like .altinstructions, __jump_table, and __ex_table, which can add alternative execution paths to a given instruction (or set of instructions). Similarly, it knows how to follow switch statements, for which gcc sometimes uses jump tables. Here are some of the benefits of validating stack metadata: a) More reliable stack traces for frame pointer enabled kernels Frame pointers are used for debugging purposes. They allow runtime code and debug tools to be able to walk the stack to determine the chain of function call sites that led to the currently executing code. For some architectures, frame pointers are enabled by CONFIG_FRAME_POINTER. For some other architectures they may be required by the ABI (sometimes referred to as "backchain pointers"). For C code, gcc automatically generates instructions for setting up frame pointers when the -fno-omit-frame-pointer option is used. But for asm code, the frame setup instructions have to be written by hand, which most people don't do. So the end result is that CONFIG_FRAME_POINTER is honored for C code but not for most asm code. For stack traces based on frame pointers to be reliable, all functions which call other functions must first create a stack frame and update the frame pointer. If a first function doesn't properly create a stack frame before calling a second function, the *caller* of the first function will be skipped on the stack trace. For example, consider the following example backtrace with frame pointers enabled: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff8127f568>] seq_read+0x108/0x3e0 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 It correctly shows that the caller of cmdline_proc_show() is seq_read(). If we remove the frame pointer logic from cmdline_proc_show() by replacing the frame pointer related instructions with nops, here's what it looks like instead: [<ffffffff81812584>] dump_stack+0x4b/0x63 [<ffffffff812d6dc2>] cmdline_proc_show+0x12/0x30 [<ffffffff812cce62>] proc_reg_read+0x42/0x70 [<ffffffff81256197>] __vfs_read+0x37/0x100 [<ffffffff81256b16>] vfs_read+0x86/0x130 [<ffffffff81257898>] SyS_read+0x58/0xd0 [<ffffffff8181c1f2>] entry_SYSCALL_64_fastpath+0x12/0x76 Notice that cmdline_proc_show()'s caller, seq_read(), has been skipped. Instead the stack trace seems to show that cmdline_proc_show() was called by proc_reg_read(). The benefit of "objtool check" here is that because it ensures that *all* functions honor CONFIG_FRAME_POINTER, no functions will ever[*] be skipped on a stack trace. [*] unless an interrupt or exception has occurred at the very beginning of a function before the stack frame has been created, or at the very end of the function after the stack frame has been destroyed. This is an inherent limitation of frame pointers. b) 100% reliable stack traces for DWARF enabled kernels This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. c) Higher live patching compatibility rate This is not yet implemented. For more details about what is planned, see tools/objtool/Documentation/stack-validation.txt. To achieve the validation, "objtool check" enforces the following rules: 1. Each callable function must be annotated as such with the ELF function type. In asm code, this is typically done using the ENTRY/ENDPROC macros. If objtool finds a return instruction outside of a function, it flags an error since that usually indicates callable code which should be annotated accordingly. This rule is needed so that objtool can properly identify each callable function in order to analyze its stack metadata. 2. Conversely, each section of code which is *not* callable should *not* be annotated as an ELF function. The ENDPROC macro shouldn't be used in this case. This rule is needed so that objtool can ignore non-callable code. Such code doesn't have to follow any of the other rules. 3. Each callable function which calls another function must have the correct frame pointer logic, if required by CONFIG_FRAME_POINTER or the architecture's back chain rules. This can by done in asm code with the FRAME_BEGIN/FRAME_END macros. This rule ensures that frame pointer based stack traces will work as designed. If function A doesn't create a stack frame before calling function B, the _caller_ of function A will be skipped on the stack trace. 4. Dynamic jumps and jumps to undefined symbols are only allowed if: a) the jump is part of a switch statement; or b) the jump matches sibling call semantics and the frame pointer has the same value it had on function entry. This rule is needed so that objtool can reliably analyze all of a function's code paths. If a function jumps to code in another file, and it's not a sibling call, objtool has no way to follow the jump because it only analyzes a single file at a time. 5. A callable function may not execute kernel entry/exit instructions. The only code which needs such instructions is kernel entry code, which shouldn't be be in callable functions anyway. This rule is just a sanity check to ensure that callable functions return normally. It currently only supports x86_64. I tried to make the code generic so that support for other architectures can hopefully be plugged in relatively easily. On my Lenovo laptop with a i7-4810MQ 4-core/8-thread CPU, building the kernel with objtool checking every .o file adds about three seconds of total build time. It hasn't been optimized for performance yet, so there are probably some opportunities for better build performance. Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Bernd Petrovitsch <bernd@petrovitsch.priv.at> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris J Arges <chris.j.arges@canonical.com> Cc: Jiri Slaby <jslaby@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Michal Marek <mmarek@suse.cz> Cc: Namhyung Kim <namhyung@gmail.com> Cc: Pedro Alves <palves@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: live-patching@vger.kernel.org Link: http://lkml.kernel.org/r/f3efb173de43bd067b060de73f856567c0fa1174.1456719558.git.jpoimboe@redhat.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-29 07:22:41 +03:00
ret = decode_sections(&file);
if (ret < 0)
goto out;
warnings += ret;
ret = validate_functions(&file);
if (ret < 0)
goto out;
warnings += ret;
ret = validate_uncallable_instructions(&file);
if (ret < 0)
goto out;
warnings += ret;
out:
cleanup(&file);
/* ignore warnings for now until we get all the code cleaned up */
if (ret || warnings)
return 0;
return 0;
}