posix_acl_permission() does not care about MAY_NOT_BLOCK, and in fact
the permission logic internally must not check that bit (it's only for
upper layers to decide whether they can block to do IO to look up the
acl information or not).
But the way the code was written, it _looked_ like it cared, since the
function explicitly did not mask that bit off.
But it has exactly two callers: one for when that bit is set, which
first clears the bit before calling posix_acl_permission(), and the
other call site when that bit was clear.
So stop the silly games "saving" the MAY_NOT_BLOCK bit that must not be
used for the actual permission test, and that currently is pointlessly
cleared by the callers when the function itself should just not care.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Rasmus Villemoes points out that the 'in_group_p()' tests can be a
noticeable expense, and often completely unnecessary. A common
situation is that the 'group' bits are the same as the 'other' bits
wrt the permissions we want to test.
So rewrite 'acl_permission_check()' to not bother checking for group
ownership when the permission check doesn't care.
For example, if we're asking for read permissions, and both 'group' and
'other' allow reading, there's really no reason to check if we're part
of the group or not: either way, we'll allow it.
Rasmus says:
"On a bog-standard Ubuntu 20.04 install, a workload consisting of
compiling lots of userspace programs (i.e., calling lots of
short-lived programs that all need to get their shared libs mapped in,
and the compilers poking around looking for system headers - lots of
/usr/lib, /usr/bin, /usr/include/ accesses) puts in_group_p around
0.1% according to perf top.
System-installed files are almost always 0755 (directories and
binaries) or 0644, so in most cases, we can avoid the binary search
and the cost of pulling the cred->groups array and in_group_p() .text
into the cpu cache"
Reported-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Whiteouts, unlike real device node should not require privileges to create.
The general concern with device nodes is that opening them can have side
effects. The kernel already avoids zero major (see
Documentation/admin-guide/devices.txt). To be on the safe side the patch
explicitly forbids registering a char device with 0/0 number (see
cdev_add()).
This guarantees that a non-O_PATH open on a whiteout will fail with ENODEV;
i.e. it won't have any side effect.
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
brown paperbag time... wrong order of arguments ended up confusing
the values to check dentry and mount_lock seqcounts against.
Reported-by: kernel test robot <rong.a.chen@intel.com>
Fixes: 2aa3847085 ("non-RCU analogue of the previous commit")
Tested-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
We fall back to lookup+create (instead of atomic_open) in several cases:
1) we don't have write access to filesystem and O_TRUNC is
present in the flags. It's not something we want ->atomic_open() to
see - it just might go ahead and truncate the file. However, we can
pass it the flags sans O_TRUNC - eventually do_open() will call
handle_truncate() anyway.
2) we have O_CREAT | O_EXCL and we can't write to parent.
That's going to be an error, of course, but we want to know _which_
error should that be - might be EEXIST (if file exists), might be
EACCES or EROFS. Simply stripping O_CREAT (and checking if we see
ENOENT) would suffice, if not for O_EXCL. However, we used to have
->atomic_open() fully responsible for rejecting O_CREAT | O_EXCL
on existing file and just stripping O_CREAT would've disarmed
those checks. With nothing downstream to catch the problem -
FMODE_OPENED used to be "don't bother with EEXIST checks,
->atomic_open() has done those". Now EEXIST checks downstream
are skipped only if FMODE_CREATED is set - FMODE_OPENED alone
is not enough. That has eliminated the need to fall back onto
lookup+create path in this case.
3) O_WRONLY or O_RDWR when we have no write access to
filesystem, with nothing else objectionable. Fallback is
(and had always been) pointless.
IOW, we don't really need that fallback; all we need in such
cases is to trim O_TRUNC and O_CREAT properly.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
argument had been unused since 1643b43fbd (lookup_open(): lift the
"fallback to !O_CREAT" logics from atomic_open()) back in 2016
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Currently path_openat() has "EEXIST on O_EXCL|O_CREAT" checks done on one
of the ways out of open_last_lookups(). There are 4 cases:
1) the last component is . or ..; check is not done.
2) we had FMODE_OPENED or FMODE_CREATED set while in lookup_open();
check is not done.
3) symlink to be traversed is found; check is not done (nor
should it be)
4) everything else: check done (before complete_walk(), even).
In case (1) O_EXCL|O_CREAT ends up failing with -EISDIR - that's
open("/tmp/.", O_CREAT|O_EXCL, 0600)
Note that in the same conditions
open("/tmp", O_CREAT|O_EXCL, 0600)
would have yielded EEXIST. Either error is allowed, switching to -EEXIST
in these cases would've been more consistent.
Case (2) is more subtle; first of all, if we have FMODE_CREATED set, the
object hadn't existed prior to the call. The check should not be done in
such a case. The rest is problematic, though - we have
FMODE_OPENED set (i.e. it went through ->atomic_open() and got
successfully opened there)
FMODE_CREATED is *NOT* set
O_CREAT and O_EXCL are both set.
Any such case is a bug - either we failed to set FMODE_CREATED when we
had, in fact, created an object (no such instances in the tree) or
we have opened a pre-existing file despite having had both O_CREAT and
O_EXCL passed. One of those was, in fact caught (and fixed) while
sorting out this mess (gfs2 on cold dcache). And in such situations
we should fail with EEXIST.
Note that for (1) and (4) FMODE_CREATED is not set - for (1) there's nothing
in handle_dots() to set it, for (4) we'd explicitly checked that.
And (1), (2) and (4) are exactly the cases when we leave the loop in
the caller, with do_open() called immediately after that loop. IOW, we
can move the check over there, and make it
If we have O_CREAT|O_EXCL and after successful pathname resolution
FMODE_CREATED is *not* set, we must have run into a preexisting file and
should fail with EEXIST.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
now we can have open_last_lookups() directly from the loop in
path_openat() - the rest of do_last() never returns a symlink
to follow, so we can bloody well leave the loop first.
Rename the rest of that thing from do_last() to do_open() and
make it return an int.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
pick_link() needs to push onto stack; we start with using two-element
array embedded into struct nameidata and the first time we need
more than that we switch to separately allocated array.
Allocation can fail, of course, and handling of that would be simple
enough - we need to drop 'link' and bugger off. However, the things
get more complicated in RCU mode. There we must do GFP_ATOMIC
allocation. If that fails, we try to switch to non-RCU mode and
repeat the allocation.
To switch to non-RCU mode we need to grab references to 'link' and
to everything in nameidata. The latter done by unlazy_walk();
the former - legitimize_path(). 'link' must go first - after
unlazy_walk() we are out of RCU-critical period and it's too
late to call legitimize_path() since the references in link->mnt
and link->dentry might be pointing to freed and reused memory.
So we do legitimize_path(), then unlazy_walk(). And that's where
it gets too subtle: what to do if the former fails? We MUST
do path_put(link) to avoid leaks. And we can't do that under
rcu_read_lock(). Solution in mainline was to empty then nameidata
manually, drop out of RCU mode and then do put_path().
In effect, we open-code the things eventual terminate_walk()
would've done on error in RCU mode. That looks badly out of place
and confusing. We could add a comment along the lines of the
explanation above, but... there's a simpler solution. Call
unlazy_walk() even if legitimaze_path() fails. It will take
us out of RCU mode, so we'll be able to do path_put(link).
Yes, it will do unnecessary work - attempt to grab references
on the stuff in nameidata, only to have them dropped as soon
as we return the error to upper layer and get terminate_walk()
called there. So what? We are thoroughly off the fast path
by that point - we had GFP_ATOMIC allocation fail, we had
->d_seq or mount_lock mismatch and we are about to try walking
the same path from scratch in non-RCU mode. Which will need
to do the same allocation, this time with GFP_KERNEL, so it will
be able to apply memory pressure for blocking stuff.
Compared to that the cost of several lockref_get_not_dead()
is noise. And the logics become much easier to understand
that way.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
step_into() tries to avoid grabbing and dropping mount references
on the steps that do not involve crossing mountpoints (which is
obviously the majority of cases). So it uses a local struct path
with unusual refcounting rules - path.mnt is pinned if and only if
it's not equal to nd->path.mnt.
We used to have similar beasts all over the place and we had quite
a few bugs crop up in their handling - it's easy to get confused
when changing e.g. cleanup on failure exits (or adding a new check,
etc.)
Now that's mostly gone - the step_into() instance (which is what
we need them for) is the only one left. It is exposed to mount
traversal and it's (shortly) seen by pick_link(). Since pick_link()
needs to store it in link stack, where the normal rules apply,
it has to make sure that mount is pinned regardless of nd->path.mnt
value. That's done on all calls of pick_link() and very early
in those. Let's do that in the caller (step_into()) instead -
that way the fewer places need to be aware of such struct path
instances.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The only remaining caller (path_pts()) should be using follow_down()
anyway. And clean path_pts() a bit.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
new helper: choose_mountpoint(). Wrapper around choose_mountpoint_rcu(),
similar to lookup_mnt() vs. __lookup_mnt(). follow_dotdot() switched to
it. Now we don't grab mount_lock exclusive anymore; note that the
primitive used non-RCU mount traversals in other direction (lookup_mnt())
doesn't bother with that either - it uses mount_lock seqcount instead.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The loops in follow_dotdot{_rcu()} are doing the same thing:
we have a mount and we want to find out how far up the chain
of mounts do we need to go.
We follow the chain of mount until we find one that is not
directly overmounting the root of another mount. If such
a mount is found, we want the location it's mounted upon.
If we run out of chain (i.e. get to a mount that is not
mounted on anything else) or run into process' root, we
report failure.
On success, we want (in RCU case) d_seq of resulting location
sampled or (in non-RCU case) references to that location
acquired.
This commit introduces such primitive for RCU case and
switches follow_dotdot_rcu() to it; non-RCU case will be
go in the next commit.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Change nd->path only after the loop is done and only in case we hadn't
ended up finding ourselves in root. Same for NO_XDEV check.
That separates the "check how far back do we need to go through the
mount stack" logics from the rest of .. traversal.
NOTE: path_get/path_put introduced here are temporary. They will
go away later in the series.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Change nd->path only after the loop is done and only in case we hadn't
ended up finding ourselves in root. Same for NO_XDEV check. Don't
recheck mount_lock on each step either.
That separates the "check how far back do we need to go through the
mount stack" logics from the rest of .. traversal.
Note that the sequence for d_seq/d_inode here is
* sample mount_lock seqcount
...
* sample d_seq
* fetch d_inode
* verify mount_lock seqcount
The last step makes sure that d_inode value we'd got matches d_seq -
it dentry is guaranteed to have been a mountpoint through the
entire thing, so its d_inode must have been stable.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The logics in both of them is the same:
while true
if in process' root // uncommon
break
if *not* in mount root // normal case
find the parent
return
if at absolute root // very uncommon
break
move to underlying mountpoint
report that we are in root
Pull the common path out of the loop:
if in process' root // uncommon
goto in_root
if unlikely(in mount root)
while true
if at absolute root
goto in_root
move to underlying mountpoint
if in process' root
goto in_root
if in mount root
break;
find the parent // we are not in mount root
return
in_root:
report that we are in root
The reason for that transformation is that we get to keep the
common path straight *and* get a separate block for "move
through underlying mountpoints", which will allow to sanitize
NO_XDEV handling there. What's more, the pared-down loops
will be easier to deal with - in particular, non-RCU case
has no need to grab mount_lock and rewriting it to the
form that wouldn't do that is a non-trivial change. Better
do that with less stuff getting in the way...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
lift step_into() into handle_dots() (where they merge with each other);
have follow_... return dentry and pass inode/seq to the caller.
[braino fix folded; kudos to Qian Cai <cai@lca.pw> for reporting it]
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Right now the tail ends of follow_dotdot{,_rcu}() are pretty
much the open-coded analogues of step_into(). The differences:
* the lack of proper LOOKUP_NO_XDEV handling in non-RCU case
(arguably a bug)
* the lack of ->d_manage() handling (again, arguably a bug)
Adjust the calling conventions so that on the next step with could
just switch those functions to returning step_into().
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Behaviour change: LOOKUP_BENEATH lookup of .. in absolute root
yields an error even if it's not the process' root. That's
possible only if you'd managed to escape chroot jail by way of
procfs symlinks, but IMO the resulting behaviour is not worse -
more consistent and easier to describe:
".." in root is "stay where you are", uness LOOKUP_BENEATH
has been given, in which case it's "fail with EXDEV".
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Instead of returning 0, return new dentry; instead of returning
-ENOENT, return NULL. Adjust the callers accordingly.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Don't mess with got_write there - it is guaranteed to be false on
entry and it will be set true if and only if we decide to go for
truncation and manage to get write access for that.
Don't carry acc_mode through the entire thing - it's only used
in that part. And don't bother with gotos in there - compiler is
quite capable of optimizing that.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
it's easier to drop it right after lookup_open() and regain if
needed (i.e. if we will need to truncate). On the non-FMODE_OPENED
path we do that anyway. In case of FMODE_CREATED we won't be
needing it. And it's easier to prove correctness that way,
especially since the initial failure to get write access is not
always fatal; proving that we'll never end up truncating in that
case is rather convoluted.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
there we'll be able to merge it with its counterparts in other
cases, and there's no reason to do it before the parent has
been unlocked
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
->atomic_open() might have used a different alias than the one we'd
passed to it; in "not opened" case we take care of that, in "opened"
one we don't. Currently we don't care downstream of "opened" case
which alias to return; however, that will change shortly when we
get to unifying may_open() calls.
It's not hard to get right in all cases, anyway.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
common guts of follow_down() and follow_managed() taken to a new
helper - traverse_mounts(). The remnants of follow_managed()
are folded into its sole remaining caller (handle_mounts()).
Calling conventions of handle_mounts() slightly sanitized -
instead of the weird "1 for success, -E... for failure" that used
to be imposed by the calling conventions of walk_component() et.al.
we can use the normal "0 for success, -E... for failure".
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
We use nd->stack to store two things: pinning down the symlinks
we are resolving and resuming the name traversal when a nested
symlink is finished.
Currently, nd->depth is used to keep track of both. It's 0 when
we call link_path_walk() for the first time (for the pathname
itself) and 1 on all subsequent calls (for trailing symlinks,
if any). That's fine, as far as pinning symlinks goes - when
handling a trailing symlink, the string we are interpreting
is the body of symlink pinned down in nd->stack[0]. It's
rather inconvenient with respect to handling nested symlinks,
though - when we run out of a string we are currently interpreting,
we need to decide whether it's a nested symlink (in which case
we need to pick the string saved back when we started to interpret
that nested symlink and resume its traversal) or not (in which
case we are done with link_path_walk()).
Current solution is a bit of a kludge - in handling of trailing symlink
(in lookup_last() and open_last_lookups() we clear nd->stack[0].name.
That allows link_path_walk() to use the following rules when
running out of a string to interpret:
* if nd->depth is zero, we are at the end of pathname itself.
* if nd->depth is positive, check the saved string; for
nested symlink it will be non-NULL, for trailing symlink - NULL.
It works, but it's rather non-obvious. Note that we have two sets:
the set of symlinks currently being traversed and the set of postponed
pathname tails. The former is stored in nd->stack[0..nd->depth-1].link
and it's valid throught the pathname resolution; the latter is valid only
during an individual call of link_path_walk() and it occupies
nd->stack[0..nd->depth-1].name for the first call of link_path_walk() and
nd->stack[1..nd->depth-1].name for subsequent ones. The kludge is basically
a way to recognize the second set becoming empty.
The things get simpler if we keep track of the second set's size
explicitly and always store it in nd->stack[0..depth-1].name.
We access the second set only inside link_path_walk(), so its
size can live in a local variable; that way the check becomes
trivial without the need of that kludge.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
old flags & WALK_FOLLOW <=> new !(flags & WALK_TRAILING)
That's what that flag had really been used for.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
->last_type values are set in 3 places: path_init() (sets to LAST_ROOT),
link_path_walk (LAST_NORM/DOT/DOTDOT) and pick_link (LAST_BIND).
The are checked in walk_component(), lookup_last() and do_last().
They also get copied to the caller by filename_parentat(). In the last
3 cases the value is what we had at the return from link_path_walk().
In case of walk_component() it's either directly downstream from
assignment in link_path_walk() or, when called by lookup_last(), the
value we have at the return from link_path_walk().
The value at the entry into link_path_walk() can survive to return only
if the pathname contains nothing but slashes. Note that pick_link()
never returns such - pure jumps are handled directly. So for the calls
of link_path_walk() for trailing symlinks it does not matter what value
had been there at the entry; the value at the return won't depend upon it.
There are 3 call chains that might have pick_link() storing LAST_BIND:
1) pick_link() from step_into() from walk_component() from
link_path_walk(). In that case we will either be parsing the next
component immediately after return into link_path_walk(), which will
overwrite the ->last_type before anyone has a chance to look at it,
or we'll fail, in which case nobody will be looking at ->last_type at all.
2) pick_link() from step_into() from walk_component() from lookup_last().
The value is never looked at due to the above; it won't affect the value
seen at return from any link_path_walk().
3) pick_link() from step_into() from do_last(). Ditto.
In other words, assignemnt in pick_link() is pointless, and so is
LAST_BIND itself; nothing ever looks at that value. Kill it off.
And make link_path_walk() _always_ assign ->last_type - in the only
case when the value at the entry might survive to the return that value
is always LAST_ROOT, inherited from path_init(). Move that assignment
from path_init() into the beginning of link_path_walk(), to consolidate
the things.
Historical note: LAST_BIND used to be used for the kludge with trailing
pure jump symlinks (extra iteration through the top-level loop).
No point keeping it anymore...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Linus Torvalds
Miklos Szeredi
Al Viro