2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* linux/drivers/block/loop.c
|
|
|
|
*
|
|
|
|
* Written by Theodore Ts'o, 3/29/93
|
|
|
|
*
|
|
|
|
* Copyright 1993 by Theodore Ts'o. Redistribution of this file is
|
|
|
|
* permitted under the GNU General Public License.
|
|
|
|
*
|
|
|
|
* DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
|
|
|
|
* more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
|
|
|
|
*
|
|
|
|
* Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
|
|
|
|
* Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
|
|
|
|
*
|
|
|
|
* Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
|
|
|
|
*
|
|
|
|
* Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
|
|
|
|
*
|
|
|
|
* Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
|
|
|
|
*
|
|
|
|
* Loadable modules and other fixes by AK, 1998
|
|
|
|
*
|
|
|
|
* Make real block number available to downstream transfer functions, enables
|
|
|
|
* CBC (and relatives) mode encryption requiring unique IVs per data block.
|
|
|
|
* Reed H. Petty, rhp@draper.net
|
|
|
|
*
|
|
|
|
* Maximum number of loop devices now dynamic via max_loop module parameter.
|
|
|
|
* Russell Kroll <rkroll@exploits.org> 19990701
|
|
|
|
*
|
|
|
|
* Maximum number of loop devices when compiled-in now selectable by passing
|
|
|
|
* max_loop=<1-255> to the kernel on boot.
|
2007-10-20 01:21:04 +04:00
|
|
|
* Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
|
2005-04-17 02:20:36 +04:00
|
|
|
*
|
|
|
|
* Completely rewrite request handling to be make_request_fn style and
|
|
|
|
* non blocking, pushing work to a helper thread. Lots of fixes from
|
|
|
|
* Al Viro too.
|
|
|
|
* Jens Axboe <axboe@suse.de>, Nov 2000
|
|
|
|
*
|
|
|
|
* Support up to 256 loop devices
|
|
|
|
* Heinz Mauelshagen <mge@sistina.com>, Feb 2002
|
|
|
|
*
|
|
|
|
* Support for falling back on the write file operation when the address space
|
2008-10-30 00:00:55 +03:00
|
|
|
* operations write_begin is not available on the backing filesystem.
|
2005-04-17 02:20:36 +04:00
|
|
|
* Anton Altaparmakov, 16 Feb 2005
|
|
|
|
*
|
|
|
|
* Still To Fix:
|
|
|
|
* - Advisory locking is ignored here.
|
|
|
|
* - Should use an own CAP_* category instead of CAP_SYS_ADMIN
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/moduleparam.h>
|
|
|
|
#include <linux/sched.h>
|
|
|
|
#include <linux/fs.h>
|
2021-05-07 04:02:27 +03:00
|
|
|
#include <linux/pagemap.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
#include <linux/file.h>
|
|
|
|
#include <linux/stat.h>
|
|
|
|
#include <linux/errno.h>
|
|
|
|
#include <linux/major.h>
|
|
|
|
#include <linux/wait.h>
|
|
|
|
#include <linux/blkdev.h>
|
|
|
|
#include <linux/blkpg.h>
|
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/swap.h>
|
|
|
|
#include <linux/slab.h>
|
2006-08-29 22:06:14 +04:00
|
|
|
#include <linux/compat.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
#include <linux/suspend.h>
|
2007-07-17 15:03:35 +04:00
|
|
|
#include <linux/freezer.h>
|
2010-06-02 16:28:52 +04:00
|
|
|
#include <linux/mutex.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
#include <linux/writeback.h>
|
|
|
|
#include <linux/completion.h>
|
|
|
|
#include <linux/highmem.h>
|
2007-06-04 11:59:47 +04:00
|
|
|
#include <linux/splice.h>
|
2010-08-23 17:16:00 +04:00
|
|
|
#include <linux/sysfs.h>
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
#include <linux/miscdevice.h>
|
2011-08-19 16:50:46 +04:00
|
|
|
#include <linux/falloc.h>
|
2015-04-03 22:21:59 +03:00
|
|
|
#include <linux/uio.h>
|
2018-05-22 20:52:19 +03:00
|
|
|
#include <linux/ioprio.h>
|
2018-12-05 20:10:35 +03:00
|
|
|
#include <linux/blk-cgroup.h>
|
2021-06-29 05:38:21 +03:00
|
|
|
#include <linux/sched/mm.h>
|
block: loop:use kstatfs.f_bsize of backing file to set discard granularity
[ Upstream commit 06582bc86d7f48d35cd044098ca1e246e8c7c52e ]
If backing file's filesystem has implemented ->fallocate(), we think the
loop device can support discard, then pass sb->s_blocksize as
discard_granularity. However, some underlying FS, such as overlayfs,
doesn't set sb->s_blocksize, and causes discard_granularity to be set as
zero, then the warning in __blkdev_issue_discard() is triggered.
Christoph suggested to pass kstatfs.f_bsize as discard granularity, and
this way is fine because kstatfs.f_bsize means 'Optimal transfer block
size', which still matches with definition of discard granularity.
So fix the issue by setting discard_granularity as kstatfs.f_bsize if it
is available, otherwise claims discard isn't supported.
Cc: Christoph Hellwig <hch@lst.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Reported-by: Pei Zhang <pezhang@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220126035830.296465-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2022-01-26 06:58:30 +03:00
|
|
|
#include <linux/statfs.h>
|
2018-05-22 20:52:19 +03:00
|
|
|
|
2013-05-12 18:14:07 +04:00
|
|
|
#include "loop.h"
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2016-12-24 22:46:01 +03:00
|
|
|
#include <linux/uaccess.h>
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
|
|
|
|
|
2011-08-01 00:08:04 +04:00
|
|
|
static DEFINE_IDR(loop_index_idr);
|
2018-11-08 16:01:02 +03:00
|
|
|
static DEFINE_MUTEX(loop_ctl_mutex);
|
2021-07-06 17:40:34 +03:00
|
|
|
static DEFINE_MUTEX(loop_validate_mutex);
|
|
|
|
|
|
|
|
/**
|
|
|
|
* loop_global_lock_killable() - take locks for safe loop_validate_file() test
|
|
|
|
*
|
|
|
|
* @lo: struct loop_device
|
|
|
|
* @global: true if @lo is about to bind another "struct loop_device", false otherwise
|
|
|
|
*
|
|
|
|
* Returns 0 on success, -EINTR otherwise.
|
|
|
|
*
|
|
|
|
* Since loop_validate_file() traverses on other "struct loop_device" if
|
|
|
|
* is_loop_device() is true, we need a global lock for serializing concurrent
|
|
|
|
* loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
|
|
|
|
*/
|
|
|
|
static int loop_global_lock_killable(struct loop_device *lo, bool global)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
|
|
|
if (global) {
|
|
|
|
err = mutex_lock_killable(&loop_validate_mutex);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
err = mutex_lock_killable(&lo->lo_mutex);
|
|
|
|
if (err && global)
|
|
|
|
mutex_unlock(&loop_validate_mutex);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* loop_global_unlock() - release locks taken by loop_global_lock_killable()
|
|
|
|
*
|
|
|
|
* @lo: struct loop_device
|
|
|
|
* @global: true if @lo was about to bind another "struct loop_device", false otherwise
|
|
|
|
*/
|
|
|
|
static void loop_global_unlock(struct loop_device *lo, bool global)
|
|
|
|
{
|
|
|
|
mutex_unlock(&lo->lo_mutex);
|
|
|
|
if (global)
|
|
|
|
mutex_unlock(&loop_validate_mutex);
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
loop: manage partitions in disk image
This patch allows to use loop device with partitionned disk image.
Original behavior of loop is not modified.
A new parameter is introduced to define how many partition we want to be
able to manage per loop device. This parameter is "max_part".
For instance, to manage 63 partitions / loop device, we will do:
# modprobe loop max_part=63
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
And to attach a raw partitionned disk image, the original losetup is used:
# losetup -f etch.img
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 1 2008-03-05 14:57 /dev/loop0p1
brw-rw---- 1 root disk 7, 2 2008-03-05 14:57 /dev/loop0p2
brw-rw---- 1 root disk 7, 5 2008-03-05 14:57 /dev/loop0p5
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
# mount /dev/loop0p1 /mnt
# ls /mnt
bench cdrom home lib mnt root srv usr
bin dev initrd lost+found opt sbin sys var
boot etc initrd.img media proc selinux tmp vmlinuz
# umount /mnt
# losetup -d /dev/loop0
Of course, the same behavior can be done using kpartx on a loop device,
but modifying loop avoids to stack several layers of block device (loop +
device mapper), this is a very light modification (40% of modifications
are to manage the new parameter).
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2008-03-26 14:11:53 +03:00
|
|
|
static int max_part;
|
|
|
|
static int part_shift;
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
static int transfer_xor(struct loop_device *lo, int cmd,
|
|
|
|
struct page *raw_page, unsigned raw_off,
|
|
|
|
struct page *loop_page, unsigned loop_off,
|
|
|
|
int size, sector_t real_block)
|
|
|
|
{
|
2011-11-25 19:14:18 +04:00
|
|
|
char *raw_buf = kmap_atomic(raw_page) + raw_off;
|
|
|
|
char *loop_buf = kmap_atomic(loop_page) + loop_off;
|
2005-04-17 02:20:36 +04:00
|
|
|
char *in, *out, *key;
|
|
|
|
int i, keysize;
|
|
|
|
|
|
|
|
if (cmd == READ) {
|
|
|
|
in = raw_buf;
|
|
|
|
out = loop_buf;
|
|
|
|
} else {
|
|
|
|
in = loop_buf;
|
|
|
|
out = raw_buf;
|
|
|
|
}
|
|
|
|
|
|
|
|
key = lo->lo_encrypt_key;
|
|
|
|
keysize = lo->lo_encrypt_key_size;
|
|
|
|
for (i = 0; i < size; i++)
|
|
|
|
*out++ = *in++ ^ key[(i & 511) % keysize];
|
|
|
|
|
2011-11-25 19:14:18 +04:00
|
|
|
kunmap_atomic(loop_buf);
|
|
|
|
kunmap_atomic(raw_buf);
|
2005-04-17 02:20:36 +04:00
|
|
|
cond_resched();
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
|
|
|
|
{
|
|
|
|
if (unlikely(info->lo_encrypt_key_size <= 0))
|
|
|
|
return -EINVAL;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct loop_func_table none_funcs = {
|
|
|
|
.number = LO_CRYPT_NONE,
|
2015-04-07 19:23:29 +03:00
|
|
|
};
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
static struct loop_func_table xor_funcs = {
|
|
|
|
.number = LO_CRYPT_XOR,
|
|
|
|
.transfer = transfer_xor,
|
|
|
|
.init = xor_init
|
2015-04-07 19:23:29 +03:00
|
|
|
};
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/* xfer_funcs[0] is special - its release function is never called */
|
|
|
|
static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
|
|
|
|
&none_funcs,
|
|
|
|
&xor_funcs
|
|
|
|
};
|
|
|
|
|
2011-11-16 12:21:49 +04:00
|
|
|
static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2013-02-22 03:16:50 +04:00
|
|
|
loff_t loopsize;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/* Compute loopsize in bytes */
|
2013-02-22 03:16:50 +04:00
|
|
|
loopsize = i_size_read(file->f_mapping->host);
|
|
|
|
if (offset > 0)
|
|
|
|
loopsize -= offset;
|
|
|
|
/* offset is beyond i_size, weird but possible */
|
2011-11-16 12:21:49 +04:00
|
|
|
if (loopsize < 0)
|
|
|
|
return 0;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2011-11-16 12:21:49 +04:00
|
|
|
if (sizelimit > 0 && sizelimit < loopsize)
|
|
|
|
loopsize = sizelimit;
|
2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* Unfortunately, if we want to do I/O on the device,
|
|
|
|
* the number of 512-byte sectors has to fit into a sector_t.
|
|
|
|
*/
|
|
|
|
return loopsize >> 9;
|
|
|
|
}
|
|
|
|
|
2011-11-16 12:21:49 +04:00
|
|
|
static loff_t get_loop_size(struct loop_device *lo, struct file *file)
|
|
|
|
{
|
|
|
|
return get_size(lo->lo_offset, lo->lo_sizelimit, file);
|
|
|
|
}
|
|
|
|
|
2015-08-17 05:31:49 +03:00
|
|
|
static void __loop_update_dio(struct loop_device *lo, bool dio)
|
|
|
|
{
|
|
|
|
struct file *file = lo->lo_backing_file;
|
|
|
|
struct address_space *mapping = file->f_mapping;
|
|
|
|
struct inode *inode = mapping->host;
|
|
|
|
unsigned short sb_bsize = 0;
|
|
|
|
unsigned dio_align = 0;
|
|
|
|
bool use_dio;
|
|
|
|
|
|
|
|
if (inode->i_sb->s_bdev) {
|
|
|
|
sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
|
|
|
|
dio_align = sb_bsize - 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We support direct I/O only if lo_offset is aligned with the
|
|
|
|
* logical I/O size of backing device, and the logical block
|
|
|
|
* size of loop is bigger than the backing device's and the loop
|
|
|
|
* needn't transform transfer.
|
|
|
|
*
|
|
|
|
* TODO: the above condition may be loosed in the future, and
|
|
|
|
* direct I/O may be switched runtime at that time because most
|
2017-02-28 01:29:01 +03:00
|
|
|
* of requests in sane applications should be PAGE_SIZE aligned
|
2015-08-17 05:31:49 +03:00
|
|
|
*/
|
|
|
|
if (dio) {
|
|
|
|
if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
|
|
|
|
!(lo->lo_offset & dio_align) &&
|
|
|
|
mapping->a_ops->direct_IO &&
|
|
|
|
!lo->transfer)
|
|
|
|
use_dio = true;
|
|
|
|
else
|
|
|
|
use_dio = false;
|
|
|
|
} else {
|
|
|
|
use_dio = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (lo->use_dio == use_dio)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* flush dirty pages before changing direct IO */
|
|
|
|
vfs_fsync(file, 0);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The flag of LO_FLAGS_DIRECT_IO is handled similarly with
|
|
|
|
* LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
|
|
|
|
* will get updated by ioctl(LOOP_GET_STATUS)
|
|
|
|
*/
|
2020-03-10 16:06:54 +03:00
|
|
|
if (lo->lo_state == Lo_bound)
|
|
|
|
blk_mq_freeze_queue(lo->lo_queue);
|
2015-08-17 05:31:49 +03:00
|
|
|
lo->use_dio = use_dio;
|
2017-09-01 08:09:46 +03:00
|
|
|
if (use_dio) {
|
2018-03-08 04:10:10 +03:00
|
|
|
blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
|
2015-08-17 05:31:49 +03:00
|
|
|
lo->lo_flags |= LO_FLAGS_DIRECT_IO;
|
2017-09-01 08:09:46 +03:00
|
|
|
} else {
|
2018-03-08 04:10:10 +03:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
|
2015-08-17 05:31:49 +03:00
|
|
|
lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
|
2017-09-01 08:09:46 +03:00
|
|
|
}
|
2020-03-10 16:06:54 +03:00
|
|
|
if (lo->lo_state == Lo_bound)
|
|
|
|
blk_mq_unfreeze_queue(lo->lo_queue);
|
2015-08-17 05:31:49 +03:00
|
|
|
}
|
|
|
|
|
2020-05-13 16:38:37 +03:00
|
|
|
/**
|
|
|
|
* loop_set_size() - sets device size and notifies userspace
|
|
|
|
* @lo: struct loop_device to set the size for
|
|
|
|
* @size: new size of the loop device
|
|
|
|
*
|
|
|
|
* Callers must validate that the size passed into this function fits into
|
|
|
|
* a sector_t, eg using loop_validate_size()
|
|
|
|
*/
|
|
|
|
static void loop_set_size(struct loop_device *lo, loff_t size)
|
|
|
|
{
|
2020-11-16 17:56:56 +03:00
|
|
|
if (!set_capacity_and_notify(lo->lo_disk, size))
|
2020-11-16 17:56:53 +03:00
|
|
|
kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
|
2020-05-13 16:38:37 +03:00
|
|
|
}
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
static inline int
|
|
|
|
lo_do_transfer(struct loop_device *lo, int cmd,
|
|
|
|
struct page *rpage, unsigned roffs,
|
|
|
|
struct page *lpage, unsigned loffs,
|
|
|
|
int size, sector_t rblock)
|
|
|
|
{
|
2015-04-07 19:23:29 +03:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
|
|
|
|
if (likely(!ret))
|
2005-04-17 02:20:36 +04:00
|
|
|
return 0;
|
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
printk_ratelimited(KERN_ERR
|
|
|
|
"loop: Transfer error at byte offset %llu, length %i.\n",
|
|
|
|
(unsigned long long)rblock << 9, size);
|
|
|
|
return ret;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2015-04-07 19:23:29 +03:00
|
|
|
struct iov_iter i;
|
2005-04-17 02:20:36 +04:00
|
|
|
ssize_t bw;
|
2015-04-03 22:21:59 +03:00
|
|
|
|
2019-06-26 16:49:28 +03:00
|
|
|
iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2013-03-20 21:04:20 +04:00
|
|
|
file_start_write(file);
|
2017-05-27 11:16:52 +03:00
|
|
|
bw = vfs_iter_write(file, &i, ppos, 0);
|
2013-03-20 21:04:20 +04:00
|
|
|
file_end_write(file);
|
2015-04-07 19:23:29 +03:00
|
|
|
|
|
|
|
if (likely(bw == bvec->bv_len))
|
2005-04-17 02:20:36 +04:00
|
|
|
return 0;
|
2015-04-07 19:23:29 +03:00
|
|
|
|
|
|
|
printk_ratelimited(KERN_ERR
|
|
|
|
"loop: Write error at byte offset %llu, length %i.\n",
|
|
|
|
(unsigned long long)*ppos, bvec->bv_len);
|
2005-04-17 02:20:36 +04:00
|
|
|
if (bw >= 0)
|
|
|
|
bw = -EIO;
|
|
|
|
return bw;
|
|
|
|
}
|
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
static int lo_write_simple(struct loop_device *lo, struct request *rq,
|
|
|
|
loff_t pos)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2015-04-07 19:23:29 +03:00
|
|
|
struct bio_vec bvec;
|
|
|
|
struct req_iterator iter;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
rq_for_each_segment(bvec, rq, iter) {
|
|
|
|
ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
|
|
|
|
if (ret < 0)
|
|
|
|
break;
|
|
|
|
cond_resched();
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
/*
|
2011-10-17 14:57:20 +04:00
|
|
|
* This is the slow, transforming version that needs to double buffer the
|
|
|
|
* data as it cannot do the transformations in place without having direct
|
|
|
|
* access to the destination pages of the backing file.
|
2005-04-17 02:20:36 +04:00
|
|
|
*/
|
2015-04-07 19:23:29 +03:00
|
|
|
static int lo_write_transfer(struct loop_device *lo, struct request *rq,
|
|
|
|
loff_t pos)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2015-04-07 19:23:29 +03:00
|
|
|
struct bio_vec bvec, b;
|
2014-12-31 16:22:58 +03:00
|
|
|
struct req_iterator iter;
|
2015-04-07 19:23:29 +03:00
|
|
|
struct page *page;
|
2013-11-24 05:19:00 +04:00
|
|
|
int ret = 0;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
page = alloc_page(GFP_NOIO);
|
|
|
|
if (unlikely(!page))
|
|
|
|
return -ENOMEM;
|
2011-10-17 14:57:20 +04:00
|
|
|
|
2014-12-31 16:22:58 +03:00
|
|
|
rq_for_each_segment(bvec, rq, iter) {
|
2015-04-07 19:23:29 +03:00
|
|
|
ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
|
|
|
|
bvec.bv_offset, bvec.bv_len, pos >> 9);
|
|
|
|
if (unlikely(ret))
|
|
|
|
break;
|
|
|
|
|
|
|
|
b.bv_page = page;
|
|
|
|
b.bv_offset = 0;
|
|
|
|
b.bv_len = bvec.bv_len;
|
|
|
|
ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
|
2005-04-17 02:20:36 +04:00
|
|
|
if (ret < 0)
|
|
|
|
break;
|
|
|
|
}
|
2015-04-07 19:23:29 +03:00
|
|
|
|
|
|
|
__free_page(page);
|
2005-04-17 02:20:36 +04:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
static int lo_read_simple(struct loop_device *lo, struct request *rq,
|
|
|
|
loff_t pos)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2015-04-07 19:23:29 +03:00
|
|
|
struct bio_vec bvec;
|
|
|
|
struct req_iterator iter;
|
|
|
|
struct iov_iter i;
|
|
|
|
ssize_t len;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
rq_for_each_segment(bvec, rq, iter) {
|
2019-06-26 16:49:28 +03:00
|
|
|
iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
|
2017-05-27 11:16:51 +03:00
|
|
|
len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
|
2015-04-07 19:23:29 +03:00
|
|
|
if (len < 0)
|
|
|
|
return len;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
flush_dcache_page(bvec.bv_page);
|
2007-06-12 23:20:37 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
if (len != bvec.bv_len) {
|
|
|
|
struct bio *bio;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
__rq_for_each_bio(bio, rq)
|
|
|
|
zero_fill_bio(bio);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
cond_resched();
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
2007-06-12 23:20:37 +04:00
|
|
|
}
|
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
static int lo_read_transfer(struct loop_device *lo, struct request *rq,
|
|
|
|
loff_t pos)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2015-04-07 19:23:29 +03:00
|
|
|
struct bio_vec bvec, b;
|
|
|
|
struct req_iterator iter;
|
|
|
|
struct iov_iter i;
|
|
|
|
struct page *page;
|
|
|
|
ssize_t len;
|
|
|
|
int ret = 0;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
page = alloc_page(GFP_NOIO);
|
|
|
|
if (unlikely(!page))
|
|
|
|
return -ENOMEM;
|
2007-06-12 23:20:37 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
rq_for_each_segment(bvec, rq, iter) {
|
|
|
|
loff_t offset = pos;
|
2007-06-12 23:20:37 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
b.bv_page = page;
|
|
|
|
b.bv_offset = 0;
|
|
|
|
b.bv_len = bvec.bv_len;
|
2007-06-12 23:20:37 +04:00
|
|
|
|
2019-06-26 16:49:28 +03:00
|
|
|
iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
|
2017-05-27 11:16:51 +03:00
|
|
|
len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
|
2015-04-07 19:23:29 +03:00
|
|
|
if (len < 0) {
|
|
|
|
ret = len;
|
|
|
|
goto out_free_page;
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
|
|
|
|
bvec.bv_offset, len, offset >> 9);
|
|
|
|
if (ret)
|
|
|
|
goto out_free_page;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
flush_dcache_page(bvec.bv_page);
|
2012-02-09 01:07:19 +04:00
|
|
|
|
2015-04-07 19:23:29 +03:00
|
|
|
if (len != bvec.bv_len) {
|
2014-12-31 16:22:58 +03:00
|
|
|
struct bio *bio;
|
|
|
|
|
|
|
|
__rq_for_each_bio(bio, rq)
|
|
|
|
zero_fill_bio(bio);
|
2005-04-17 02:20:36 +04:00
|
|
|
break;
|
2012-02-09 01:07:19 +04:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
2015-04-07 19:23:29 +03:00
|
|
|
|
|
|
|
ret = 0;
|
|
|
|
out_free_page:
|
|
|
|
__free_page(page);
|
|
|
|
return ret;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2019-10-31 06:29:48 +03:00
|
|
|
static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
|
|
|
|
int mode)
|
2014-12-31 16:22:59 +03:00
|
|
|
{
|
|
|
|
/*
|
2019-10-31 06:29:48 +03:00
|
|
|
* We use fallocate to manipulate the space mappings used by the image
|
|
|
|
* a.k.a. discard/zerorange. However we do not support this if
|
|
|
|
* encryption is enabled, because it may give an attacker useful
|
|
|
|
* information.
|
2014-12-31 16:22:59 +03:00
|
|
|
*/
|
|
|
|
struct file *file = lo->lo_backing_file;
|
2020-04-03 17:43:04 +03:00
|
|
|
struct request_queue *q = lo->lo_queue;
|
2014-12-31 16:22:59 +03:00
|
|
|
int ret;
|
|
|
|
|
2019-10-31 06:29:48 +03:00
|
|
|
mode |= FALLOC_FL_KEEP_SIZE;
|
|
|
|
|
2020-04-03 17:43:04 +03:00
|
|
|
if (!blk_queue_discard(q)) {
|
2014-12-31 16:22:59 +03:00
|
|
|
ret = -EOPNOTSUPP;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
|
|
|
|
if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
|
|
|
|
ret = -EIO;
|
|
|
|
out:
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int lo_req_flush(struct loop_device *lo, struct request *rq)
|
|
|
|
{
|
|
|
|
struct file *file = lo->lo_backing_file;
|
|
|
|
int ret = vfs_fsync(file, 0);
|
|
|
|
if (unlikely(ret && ret != -EINVAL))
|
|
|
|
ret = -EIO;
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-04-20 17:03:02 +03:00
|
|
|
static void lo_complete_rq(struct request *rq)
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
{
|
2017-04-20 17:03:02 +03:00
|
|
|
struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
|
2018-04-14 01:25:57 +03:00
|
|
|
blk_status_t ret = BLK_STS_OK;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
2018-04-14 01:25:57 +03:00
|
|
|
if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
|
|
|
|
req_op(rq) != REQ_OP_READ) {
|
|
|
|
if (cmd->ret < 0)
|
2020-04-03 17:43:03 +03:00
|
|
|
ret = errno_to_blk_status(cmd->ret);
|
2018-04-14 01:25:57 +03:00
|
|
|
goto end_io;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
}
|
2017-04-20 17:03:02 +03:00
|
|
|
|
2018-04-14 01:25:57 +03:00
|
|
|
/*
|
|
|
|
* Short READ - if we got some data, advance our request and
|
|
|
|
* retry it. If we got no data, end the rest with EIO.
|
|
|
|
*/
|
|
|
|
if (cmd->ret) {
|
|
|
|
blk_update_request(rq, BLK_STS_OK, cmd->ret);
|
|
|
|
cmd->ret = 0;
|
|
|
|
blk_mq_requeue_request(rq, true);
|
|
|
|
} else {
|
|
|
|
if (cmd->use_aio) {
|
|
|
|
struct bio *bio = rq->bio;
|
|
|
|
|
|
|
|
while (bio) {
|
|
|
|
zero_fill_bio(bio);
|
|
|
|
bio = bio->bi_next;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ret = BLK_STS_IOERR;
|
|
|
|
end_io:
|
|
|
|
blk_mq_end_request(rq, ret);
|
|
|
|
}
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
}
|
|
|
|
|
2017-09-01 21:15:17 +03:00
|
|
|
static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
|
|
|
|
{
|
2018-04-14 01:24:29 +03:00
|
|
|
struct request *rq = blk_mq_rq_from_pdu(cmd);
|
|
|
|
|
2017-09-01 21:15:17 +03:00
|
|
|
if (!atomic_dec_and_test(&cmd->ref))
|
|
|
|
return;
|
|
|
|
kfree(cmd->bvec);
|
|
|
|
cmd->bvec = NULL;
|
2020-06-11 09:44:47 +03:00
|
|
|
if (likely(!blk_should_fake_timeout(rq->q)))
|
|
|
|
blk_mq_complete_request(rq);
|
2017-09-01 21:15:17 +03:00
|
|
|
}
|
|
|
|
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
|
|
|
|
{
|
|
|
|
struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
|
|
|
|
|
2017-04-20 17:03:02 +03:00
|
|
|
cmd->ret = ret;
|
2017-09-01 21:15:17 +03:00
|
|
|
lo_rw_aio_do_completion(cmd);
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
|
|
|
|
loff_t pos, bool rw)
|
|
|
|
{
|
|
|
|
struct iov_iter iter;
|
2019-02-15 14:13:17 +03:00
|
|
|
struct req_iterator rq_iter;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
struct bio_vec *bvec;
|
2018-04-14 01:24:29 +03:00
|
|
|
struct request *rq = blk_mq_rq_from_pdu(cmd);
|
2017-09-01 08:09:46 +03:00
|
|
|
struct bio *bio = rq->bio;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
struct file *file = lo->lo_backing_file;
|
2019-02-15 14:13:17 +03:00
|
|
|
struct bio_vec tmp;
|
2017-09-01 08:09:46 +03:00
|
|
|
unsigned int offset;
|
2019-02-15 14:13:17 +03:00
|
|
|
int nr_bvec = 0;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
int ret;
|
|
|
|
|
2019-02-15 14:13:17 +03:00
|
|
|
rq_for_each_bvec(tmp, rq, rq_iter)
|
|
|
|
nr_bvec++;
|
|
|
|
|
2017-09-01 08:09:46 +03:00
|
|
|
if (rq->bio != rq->biotail) {
|
|
|
|
|
2019-02-15 14:13:17 +03:00
|
|
|
bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
|
treewide: kmalloc() -> kmalloc_array()
The kmalloc() function has a 2-factor argument form, kmalloc_array(). This
patch replaces cases of:
kmalloc(a * b, gfp)
with:
kmalloc_array(a * b, gfp)
as well as handling cases of:
kmalloc(a * b * c, gfp)
with:
kmalloc(array3_size(a, b, c), gfp)
as it's slightly less ugly than:
kmalloc_array(array_size(a, b), c, gfp)
This does, however, attempt to ignore constant size factors like:
kmalloc(4 * 1024, gfp)
though any constants defined via macros get caught up in the conversion.
Any factors with a sizeof() of "unsigned char", "char", and "u8" were
dropped, since they're redundant.
The tools/ directory was manually excluded, since it has its own
implementation of kmalloc().
The Coccinelle script used for this was:
// Fix redundant parens around sizeof().
@@
type TYPE;
expression THING, E;
@@
(
kmalloc(
- (sizeof(TYPE)) * E
+ sizeof(TYPE) * E
, ...)
|
kmalloc(
- (sizeof(THING)) * E
+ sizeof(THING) * E
, ...)
)
// Drop single-byte sizes and redundant parens.
@@
expression COUNT;
typedef u8;
typedef __u8;
@@
(
kmalloc(
- sizeof(u8) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(__u8) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(char) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(unsigned char) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(u8) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(__u8) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(char) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(unsigned char) * COUNT
+ COUNT
, ...)
)
// 2-factor product with sizeof(type/expression) and identifier or constant.
@@
type TYPE;
expression THING;
identifier COUNT_ID;
constant COUNT_CONST;
@@
(
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (COUNT_ID)
+ COUNT_ID, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * COUNT_ID
+ COUNT_ID, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (COUNT_CONST)
+ COUNT_CONST, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * COUNT_CONST
+ COUNT_CONST, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (COUNT_ID)
+ COUNT_ID, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * COUNT_ID
+ COUNT_ID, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (COUNT_CONST)
+ COUNT_CONST, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * COUNT_CONST
+ COUNT_CONST, sizeof(THING)
, ...)
)
// 2-factor product, only identifiers.
@@
identifier SIZE, COUNT;
@@
- kmalloc
+ kmalloc_array
(
- SIZE * COUNT
+ COUNT, SIZE
, ...)
// 3-factor product with 1 sizeof(type) or sizeof(expression), with
// redundant parens removed.
@@
expression THING;
identifier STRIDE, COUNT;
type TYPE;
@@
(
kmalloc(
- sizeof(TYPE) * (COUNT) * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * (COUNT) * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * COUNT * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * COUNT * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(THING) * (COUNT) * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * (COUNT) * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * COUNT * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * COUNT * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
)
// 3-factor product with 2 sizeof(variable), with redundant parens removed.
@@
expression THING1, THING2;
identifier COUNT;
type TYPE1, TYPE2;
@@
(
kmalloc(
- sizeof(TYPE1) * sizeof(TYPE2) * COUNT
+ array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2))
, ...)
|
kmalloc(
- sizeof(THING1) * sizeof(THING2) * COUNT
+ array3_size(COUNT, sizeof(THING1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(THING1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(THING1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * COUNT
+ array3_size(COUNT, sizeof(TYPE1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(TYPE1), sizeof(THING2))
, ...)
)
// 3-factor product, only identifiers, with redundant parens removed.
@@
identifier STRIDE, SIZE, COUNT;
@@
(
kmalloc(
- (COUNT) * STRIDE * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * (STRIDE) * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * STRIDE * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * (STRIDE) * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * (STRIDE) * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * STRIDE * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * (STRIDE) * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * STRIDE * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
)
// Any remaining multi-factor products, first at least 3-factor products,
// when they're not all constants...
@@
expression E1, E2, E3;
constant C1, C2, C3;
@@
(
kmalloc(C1 * C2 * C3, ...)
|
kmalloc(
- (E1) * E2 * E3
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- (E1) * (E2) * E3
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- (E1) * (E2) * (E3)
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- E1 * E2 * E3
+ array3_size(E1, E2, E3)
, ...)
)
// And then all remaining 2 factors products when they're not all constants,
// keeping sizeof() as the second factor argument.
@@
expression THING, E1, E2;
type TYPE;
constant C1, C2, C3;
@@
(
kmalloc(sizeof(THING) * C2, ...)
|
kmalloc(sizeof(TYPE) * C2, ...)
|
kmalloc(C1 * C2 * C3, ...)
|
kmalloc(C1 * C2, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (E2)
+ E2, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * E2
+ E2, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (E2)
+ E2, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * E2
+ E2, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- (E1) * E2
+ E1, E2
, ...)
|
- kmalloc
+ kmalloc_array
(
- (E1) * (E2)
+ E1, E2
, ...)
|
- kmalloc
+ kmalloc_array
(
- E1 * E2
+ E1, E2
, ...)
)
Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-12 23:55:00 +03:00
|
|
|
GFP_NOIO);
|
2017-09-01 08:09:46 +03:00
|
|
|
if (!bvec)
|
|
|
|
return -EIO;
|
|
|
|
cmd->bvec = bvec;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The bios of the request may be started from the middle of
|
|
|
|
* the 'bvec' because of bio splitting, so we can't directly
|
2019-02-15 14:13:17 +03:00
|
|
|
* copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
|
2017-09-01 08:09:46 +03:00
|
|
|
* API will take care of all details for us.
|
|
|
|
*/
|
2019-02-15 14:13:17 +03:00
|
|
|
rq_for_each_bvec(tmp, rq, rq_iter) {
|
2017-09-01 08:09:46 +03:00
|
|
|
*bvec = tmp;
|
|
|
|
bvec++;
|
|
|
|
}
|
|
|
|
bvec = cmd->bvec;
|
|
|
|
offset = 0;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Same here, this bio may be started from the middle of the
|
|
|
|
* 'bvec' because of bio splitting, so offset from the bvec
|
|
|
|
* must be passed to iov iterator
|
|
|
|
*/
|
|
|
|
offset = bio->bi_iter.bi_bvec_done;
|
|
|
|
bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
|
|
|
|
}
|
2017-09-01 21:15:17 +03:00
|
|
|
atomic_set(&cmd->ref, 2);
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
2019-06-26 16:49:28 +03:00
|
|
|
iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
|
2017-09-01 08:09:46 +03:00
|
|
|
iter.iov_offset = offset;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
|
|
|
cmd->iocb.ki_pos = pos;
|
|
|
|
cmd->iocb.ki_filp = file;
|
|
|
|
cmd->iocb.ki_complete = lo_rw_aio_complete;
|
|
|
|
cmd->iocb.ki_flags = IOCB_DIRECT;
|
2018-05-22 20:52:19 +03:00
|
|
|
cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
|
|
|
if (rw == WRITE)
|
2017-02-20 18:51:23 +03:00
|
|
|
ret = call_write_iter(file, &cmd->iocb, &iter);
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
else
|
2017-02-20 18:51:23 +03:00
|
|
|
ret = call_read_iter(file, &cmd->iocb, &iter);
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
2017-09-01 21:15:17 +03:00
|
|
|
lo_rw_aio_do_completion(cmd);
|
|
|
|
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
if (ret != -EIOCBQUEUED)
|
|
|
|
cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2016-08-04 17:10:01 +03:00
|
|
|
static int do_req_filebacked(struct loop_device *lo, struct request *rq)
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
{
|
|
|
|
struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
|
2016-08-04 17:10:01 +03:00
|
|
|
loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* lo_write_simple and lo_read_simple should have been covered
|
|
|
|
* by io submit style function like lo_rw_aio(), one blocker
|
|
|
|
* is that lo_read_simple() need to call flush_dcache_page after
|
|
|
|
* the page is written from kernel, and it isn't easy to handle
|
|
|
|
* this in io submit style function which submits all segments
|
|
|
|
* of the req at one time. And direct read IO doesn't need to
|
|
|
|
* run flush_dcache_page().
|
|
|
|
*/
|
2016-08-04 17:10:01 +03:00
|
|
|
switch (req_op(rq)) {
|
|
|
|
case REQ_OP_FLUSH:
|
|
|
|
return lo_req_flush(lo, rq);
|
2017-04-05 20:21:15 +03:00
|
|
|
case REQ_OP_WRITE_ZEROES:
|
2019-10-31 06:29:48 +03:00
|
|
|
/*
|
|
|
|
* If the caller doesn't want deallocation, call zeroout to
|
|
|
|
* write zeroes the range. Otherwise, punch them out.
|
|
|
|
*/
|
|
|
|
return lo_fallocate(lo, rq, pos,
|
|
|
|
(rq->cmd_flags & REQ_NOUNMAP) ?
|
|
|
|
FALLOC_FL_ZERO_RANGE :
|
|
|
|
FALLOC_FL_PUNCH_HOLE);
|
|
|
|
case REQ_OP_DISCARD:
|
|
|
|
return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
|
2016-08-04 17:10:01 +03:00
|
|
|
case REQ_OP_WRITE:
|
|
|
|
if (lo->transfer)
|
|
|
|
return lo_write_transfer(lo, rq, pos);
|
|
|
|
else if (cmd->use_aio)
|
|
|
|
return lo_rw_aio(lo, cmd, pos, WRITE);
|
2014-12-31 16:23:00 +03:00
|
|
|
else
|
2016-08-04 17:10:01 +03:00
|
|
|
return lo_write_simple(lo, rq, pos);
|
|
|
|
case REQ_OP_READ:
|
2015-04-07 19:23:29 +03:00
|
|
|
if (lo->transfer)
|
2016-08-04 17:10:01 +03:00
|
|
|
return lo_read_transfer(lo, rq, pos);
|
|
|
|
else if (cmd->use_aio)
|
|
|
|
return lo_rw_aio(lo, cmd, pos, READ);
|
2015-04-07 19:23:29 +03:00
|
|
|
else
|
2016-08-04 17:10:01 +03:00
|
|
|
return lo_read_simple(lo, rq, pos);
|
|
|
|
default:
|
|
|
|
WARN_ON_ONCE(1);
|
|
|
|
return -EIO;
|
2015-04-07 19:23:29 +03:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2015-08-17 05:31:49 +03:00
|
|
|
static inline void loop_update_dio(struct loop_device *lo)
|
|
|
|
{
|
2020-04-30 17:41:33 +03:00
|
|
|
__loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
|
|
|
|
lo->use_dio);
|
2015-08-17 05:31:49 +03:00
|
|
|
}
|
|
|
|
|
2021-06-24 15:32:40 +03:00
|
|
|
static void loop_reread_partitions(struct loop_device *lo)
|
2015-05-06 07:26:24 +03:00
|
|
|
{
|
|
|
|
int rc;
|
|
|
|
|
2021-06-24 15:32:40 +03:00
|
|
|
mutex_lock(&lo->lo_disk->open_mutex);
|
|
|
|
rc = bdev_disk_changed(lo->lo_disk, false);
|
|
|
|
mutex_unlock(&lo->lo_disk->open_mutex);
|
2015-05-06 07:26:24 +03:00
|
|
|
if (rc)
|
|
|
|
pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
|
|
|
|
__func__, lo->lo_number, lo->lo_file_name, rc);
|
|
|
|
}
|
|
|
|
|
2018-05-07 18:37:58 +03:00
|
|
|
static inline int is_loop_device(struct file *file)
|
|
|
|
{
|
|
|
|
struct inode *i = file->f_mapping->host;
|
|
|
|
|
2021-02-01 03:23:55 +03:00
|
|
|
return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
|
2018-05-07 18:37:58 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
static int loop_validate_file(struct file *file, struct block_device *bdev)
|
|
|
|
{
|
|
|
|
struct inode *inode = file->f_mapping->host;
|
|
|
|
struct file *f = file;
|
|
|
|
|
|
|
|
/* Avoid recursion */
|
|
|
|
while (is_loop_device(f)) {
|
|
|
|
struct loop_device *l;
|
|
|
|
|
2021-07-06 17:40:34 +03:00
|
|
|
lockdep_assert_held(&loop_validate_mutex);
|
2020-11-23 15:38:40 +03:00
|
|
|
if (f->f_mapping->host->i_rdev == bdev->bd_dev)
|
2018-05-07 18:37:58 +03:00
|
|
|
return -EBADF;
|
|
|
|
|
2020-11-23 15:38:40 +03:00
|
|
|
l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
|
2021-07-06 17:40:34 +03:00
|
|
|
if (l->lo_state != Lo_bound)
|
2018-05-07 18:37:58 +03:00
|
|
|
return -EINVAL;
|
2021-07-06 17:40:34 +03:00
|
|
|
/* Order wrt setting lo->lo_backing_file in loop_configure(). */
|
|
|
|
rmb();
|
2018-05-07 18:37:58 +03:00
|
|
|
f = l->lo_backing_file;
|
|
|
|
}
|
|
|
|
if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
|
|
|
|
return -EINVAL;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
/*
|
|
|
|
* loop_change_fd switched the backing store of a loopback device to
|
|
|
|
* a new file. This is useful for operating system installers to free up
|
|
|
|
* the original file and in High Availability environments to switch to
|
|
|
|
* an alternative location for the content in case of server meltdown.
|
|
|
|
* This can only work if the loop device is used read-only, and if the
|
|
|
|
* new backing store is the same size and type as the old backing store.
|
|
|
|
*/
|
2008-03-02 17:29:48 +03:00
|
|
|
static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
|
|
|
|
unsigned int arg)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2021-07-06 17:40:34 +03:00
|
|
|
struct file *file = fget(arg);
|
|
|
|
struct file *old_file;
|
|
|
|
int error;
|
|
|
|
bool partscan;
|
|
|
|
bool is_loop;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-07-06 17:40:34 +03:00
|
|
|
if (!file)
|
|
|
|
return -EBADF;
|
|
|
|
is_loop = is_loop_device(file);
|
|
|
|
error = loop_global_lock_killable(lo, is_loop);
|
2018-11-08 16:01:11 +03:00
|
|
|
if (error)
|
2021-07-06 17:40:34 +03:00
|
|
|
goto out_putf;
|
2005-04-17 02:20:36 +04:00
|
|
|
error = -ENXIO;
|
|
|
|
if (lo->lo_state != Lo_bound)
|
2018-11-08 16:01:15 +03:00
|
|
|
goto out_err;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/* the loop device has to be read-only */
|
|
|
|
error = -EINVAL;
|
|
|
|
if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
|
2018-11-08 16:01:15 +03:00
|
|
|
goto out_err;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2018-05-07 18:37:58 +03:00
|
|
|
error = loop_validate_file(file, bdev);
|
|
|
|
if (error)
|
2018-11-08 16:01:15 +03:00
|
|
|
goto out_err;
|
2018-05-07 18:37:58 +03:00
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
old_file = lo->lo_backing_file;
|
|
|
|
|
|
|
|
error = -EINVAL;
|
|
|
|
|
|
|
|
/* size of the new backing store needs to be the same */
|
|
|
|
if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
|
2018-11-08 16:01:15 +03:00
|
|
|
goto out_err;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/* and ... switch */
|
2021-07-13 02:05:30 +03:00
|
|
|
disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
|
2017-08-24 10:03:44 +03:00
|
|
|
blk_mq_freeze_queue(lo->lo_queue);
|
|
|
|
mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
|
|
|
|
lo->lo_backing_file = file;
|
|
|
|
lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
|
|
|
|
mapping_set_gfp_mask(file->f_mapping,
|
|
|
|
lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
|
|
|
|
loop_update_dio(lo);
|
|
|
|
blk_mq_unfreeze_queue(lo->lo_queue);
|
2018-11-08 16:01:13 +03:00
|
|
|
partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
|
2021-07-06 17:40:34 +03:00
|
|
|
loop_global_unlock(lo, is_loop);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Flush loop_validate_file() before fput(), for l->lo_backing_file
|
|
|
|
* might be pointing at old_file which might be the last reference.
|
|
|
|
*/
|
|
|
|
if (!is_loop) {
|
|
|
|
mutex_lock(&loop_validate_mutex);
|
|
|
|
mutex_unlock(&loop_validate_mutex);
|
|
|
|
}
|
2018-11-08 16:01:15 +03:00
|
|
|
/*
|
2021-01-26 17:46:30 +03:00
|
|
|
* We must drop file reference outside of lo_mutex as dropping
|
2021-05-25 09:12:56 +03:00
|
|
|
* the file ref can take open_mutex which creates circular locking
|
2018-11-08 16:01:15 +03:00
|
|
|
* dependency.
|
|
|
|
*/
|
|
|
|
fput(old_file);
|
2018-11-08 16:01:13 +03:00
|
|
|
if (partscan)
|
2021-06-24 15:32:40 +03:00
|
|
|
loop_reread_partitions(lo);
|
2005-04-17 02:20:36 +04:00
|
|
|
return 0;
|
|
|
|
|
2018-11-08 16:01:15 +03:00
|
|
|
out_err:
|
2021-07-06 17:40:34 +03:00
|
|
|
loop_global_unlock(lo, is_loop);
|
|
|
|
out_putf:
|
|
|
|
fput(file);
|
2005-04-17 02:20:36 +04:00
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2010-08-23 17:16:00 +04:00
|
|
|
/* loop sysfs attributes */
|
|
|
|
|
|
|
|
static ssize_t loop_attr_show(struct device *dev, char *page,
|
|
|
|
ssize_t (*callback)(struct loop_device *, char *))
|
|
|
|
{
|
2011-08-01 00:08:04 +04:00
|
|
|
struct gendisk *disk = dev_to_disk(dev);
|
|
|
|
struct loop_device *lo = disk->private_data;
|
2010-08-23 17:16:00 +04:00
|
|
|
|
2011-08-01 00:08:04 +04:00
|
|
|
return callback(lo, page);
|
2010-08-23 17:16:00 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
#define LOOP_ATTR_RO(_name) \
|
|
|
|
static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
|
|
|
|
static ssize_t loop_attr_do_show_##_name(struct device *d, \
|
|
|
|
struct device_attribute *attr, char *b) \
|
|
|
|
{ \
|
|
|
|
return loop_attr_show(d, b, loop_attr_##_name##_show); \
|
|
|
|
} \
|
|
|
|
static struct device_attribute loop_attr_##_name = \
|
2018-05-24 22:38:59 +03:00
|
|
|
__ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
|
2010-08-23 17:16:00 +04:00
|
|
|
|
|
|
|
static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
|
|
|
|
{
|
|
|
|
ssize_t ret;
|
|
|
|
char *p = NULL;
|
|
|
|
|
2011-08-01 00:21:35 +04:00
|
|
|
spin_lock_irq(&lo->lo_lock);
|
2010-08-23 17:16:00 +04:00
|
|
|
if (lo->lo_backing_file)
|
2015-06-19 11:29:13 +03:00
|
|
|
p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
|
2011-08-01 00:21:35 +04:00
|
|
|
spin_unlock_irq(&lo->lo_lock);
|
2010-08-23 17:16:00 +04:00
|
|
|
|
|
|
|
if (IS_ERR_OR_NULL(p))
|
|
|
|
ret = PTR_ERR(p);
|
|
|
|
else {
|
|
|
|
ret = strlen(p);
|
|
|
|
memmove(buf, p, ret);
|
|
|
|
buf[ret++] = '\n';
|
|
|
|
buf[ret] = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
|
|
|
|
{
|
|
|
|
return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
|
|
|
|
{
|
|
|
|
return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
|
|
|
|
{
|
|
|
|
int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
|
|
|
|
|
|
|
|
return sprintf(buf, "%s\n", autoclear ? "1" : "0");
|
|
|
|
}
|
|
|
|
|
loop: always allow userspace partitions and optionally support automatic scanning
Automatic partition scanning can be requested individually per loop
device during its setup by setting LO_FLAGS_PARTSCAN. By default, no
partition tables are scanned.
Userspace can now always add and remove partitions from all loop
devices, regardless if the in-kernel partition scanner is enabled or
not.
The needed partition minor numbers are allocated from the extended
minors space, the main loop device numbers will continue to match the
loop minors, regardless of the number of partitions used.
# grep . /sys/class/block/loop1/loop/*
/sys/block/loop1/loop/autoclear:0
/sys/block/loop1/loop/backing_file:/home/kay/data/stuff/part.img
/sys/block/loop1/loop/offset:0
/sys/block/loop1/loop/partscan:1
/sys/block/loop1/loop/sizelimit:0
# ls -l /dev/loop*
brw-rw---- 1 root disk 7, 0 Aug 14 20:22 /dev/loop0
brw-rw---- 1 root disk 7, 1 Aug 14 20:23 /dev/loop1
brw-rw---- 1 root disk 259, 0 Aug 14 20:23 /dev/loop1p1
brw-rw---- 1 root disk 259, 1 Aug 14 20:23 /dev/loop1p2
brw-rw---- 1 root disk 7, 99 Aug 14 20:23 /dev/loop99
brw-rw---- 1 root disk 259, 2 Aug 14 20:23 /dev/loop99p1
brw-rw---- 1 root disk 259, 3 Aug 14 20:23 /dev/loop99p2
crw------T 1 root root 10, 237 Aug 14 20:22 /dev/loop-control
Cc: Karel Zak <kzak@redhat.com>
Cc: Davidlohr Bueso <dave@gnu.org>
Acked-By: Tejun Heo <tj@kernel.org>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-23 22:12:04 +04:00
|
|
|
static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
|
|
|
|
{
|
|
|
|
int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
|
|
|
|
|
|
|
|
return sprintf(buf, "%s\n", partscan ? "1" : "0");
|
|
|
|
}
|
|
|
|
|
2015-08-17 05:31:49 +03:00
|
|
|
static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
|
|
|
|
{
|
|
|
|
int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
|
|
|
|
|
|
|
|
return sprintf(buf, "%s\n", dio ? "1" : "0");
|
|
|
|
}
|
|
|
|
|
2010-08-23 17:16:00 +04:00
|
|
|
LOOP_ATTR_RO(backing_file);
|
|
|
|
LOOP_ATTR_RO(offset);
|
|
|
|
LOOP_ATTR_RO(sizelimit);
|
|
|
|
LOOP_ATTR_RO(autoclear);
|
loop: always allow userspace partitions and optionally support automatic scanning
Automatic partition scanning can be requested individually per loop
device during its setup by setting LO_FLAGS_PARTSCAN. By default, no
partition tables are scanned.
Userspace can now always add and remove partitions from all loop
devices, regardless if the in-kernel partition scanner is enabled or
not.
The needed partition minor numbers are allocated from the extended
minors space, the main loop device numbers will continue to match the
loop minors, regardless of the number of partitions used.
# grep . /sys/class/block/loop1/loop/*
/sys/block/loop1/loop/autoclear:0
/sys/block/loop1/loop/backing_file:/home/kay/data/stuff/part.img
/sys/block/loop1/loop/offset:0
/sys/block/loop1/loop/partscan:1
/sys/block/loop1/loop/sizelimit:0
# ls -l /dev/loop*
brw-rw---- 1 root disk 7, 0 Aug 14 20:22 /dev/loop0
brw-rw---- 1 root disk 7, 1 Aug 14 20:23 /dev/loop1
brw-rw---- 1 root disk 259, 0 Aug 14 20:23 /dev/loop1p1
brw-rw---- 1 root disk 259, 1 Aug 14 20:23 /dev/loop1p2
brw-rw---- 1 root disk 7, 99 Aug 14 20:23 /dev/loop99
brw-rw---- 1 root disk 259, 2 Aug 14 20:23 /dev/loop99p1
brw-rw---- 1 root disk 259, 3 Aug 14 20:23 /dev/loop99p2
crw------T 1 root root 10, 237 Aug 14 20:22 /dev/loop-control
Cc: Karel Zak <kzak@redhat.com>
Cc: Davidlohr Bueso <dave@gnu.org>
Acked-By: Tejun Heo <tj@kernel.org>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-23 22:12:04 +04:00
|
|
|
LOOP_ATTR_RO(partscan);
|
2015-08-17 05:31:49 +03:00
|
|
|
LOOP_ATTR_RO(dio);
|
2010-08-23 17:16:00 +04:00
|
|
|
|
|
|
|
static struct attribute *loop_attrs[] = {
|
|
|
|
&loop_attr_backing_file.attr,
|
|
|
|
&loop_attr_offset.attr,
|
|
|
|
&loop_attr_sizelimit.attr,
|
|
|
|
&loop_attr_autoclear.attr,
|
loop: always allow userspace partitions and optionally support automatic scanning
Automatic partition scanning can be requested individually per loop
device during its setup by setting LO_FLAGS_PARTSCAN. By default, no
partition tables are scanned.
Userspace can now always add and remove partitions from all loop
devices, regardless if the in-kernel partition scanner is enabled or
not.
The needed partition minor numbers are allocated from the extended
minors space, the main loop device numbers will continue to match the
loop minors, regardless of the number of partitions used.
# grep . /sys/class/block/loop1/loop/*
/sys/block/loop1/loop/autoclear:0
/sys/block/loop1/loop/backing_file:/home/kay/data/stuff/part.img
/sys/block/loop1/loop/offset:0
/sys/block/loop1/loop/partscan:1
/sys/block/loop1/loop/sizelimit:0
# ls -l /dev/loop*
brw-rw---- 1 root disk 7, 0 Aug 14 20:22 /dev/loop0
brw-rw---- 1 root disk 7, 1 Aug 14 20:23 /dev/loop1
brw-rw---- 1 root disk 259, 0 Aug 14 20:23 /dev/loop1p1
brw-rw---- 1 root disk 259, 1 Aug 14 20:23 /dev/loop1p2
brw-rw---- 1 root disk 7, 99 Aug 14 20:23 /dev/loop99
brw-rw---- 1 root disk 259, 2 Aug 14 20:23 /dev/loop99p1
brw-rw---- 1 root disk 259, 3 Aug 14 20:23 /dev/loop99p2
crw------T 1 root root 10, 237 Aug 14 20:22 /dev/loop-control
Cc: Karel Zak <kzak@redhat.com>
Cc: Davidlohr Bueso <dave@gnu.org>
Acked-By: Tejun Heo <tj@kernel.org>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-23 22:12:04 +04:00
|
|
|
&loop_attr_partscan.attr,
|
2015-08-17 05:31:49 +03:00
|
|
|
&loop_attr_dio.attr,
|
2010-08-23 17:16:00 +04:00
|
|
|
NULL,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct attribute_group loop_attribute_group = {
|
|
|
|
.name = "loop",
|
|
|
|
.attrs= loop_attrs,
|
|
|
|
};
|
|
|
|
|
2018-05-04 19:58:09 +03:00
|
|
|
static void loop_sysfs_init(struct loop_device *lo)
|
2010-08-23 17:16:00 +04:00
|
|
|
{
|
2018-05-04 19:58:09 +03:00
|
|
|
lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
|
|
|
|
&loop_attribute_group);
|
2010-08-23 17:16:00 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
static void loop_sysfs_exit(struct loop_device *lo)
|
|
|
|
{
|
2018-05-04 19:58:09 +03:00
|
|
|
if (lo->sysfs_inited)
|
|
|
|
sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
|
|
|
|
&loop_attribute_group);
|
2010-08-23 17:16:00 +04:00
|
|
|
}
|
|
|
|
|
2011-08-19 16:50:46 +04:00
|
|
|
static void loop_config_discard(struct loop_device *lo)
|
|
|
|
{
|
|
|
|
struct file *file = lo->lo_backing_file;
|
|
|
|
struct inode *inode = file->f_mapping->host;
|
|
|
|
struct request_queue *q = lo->lo_queue;
|
2020-08-17 13:01:30 +03:00
|
|
|
u32 granularity, max_discard_sectors;
|
2011-08-19 16:50:46 +04:00
|
|
|
|
2020-04-03 17:43:04 +03:00
|
|
|
/*
|
|
|
|
* If the backing device is a block device, mirror its zeroing
|
|
|
|
* capability. Set the discard sectors to the block device's zeroing
|
|
|
|
* capabilities because loop discards result in blkdev_issue_zeroout(),
|
|
|
|
* not blkdev_issue_discard(). This maintains consistent behavior with
|
|
|
|
* file-backed loop devices: discarded regions read back as zero.
|
|
|
|
*/
|
|
|
|
if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
|
2020-11-23 15:38:40 +03:00
|
|
|
struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
|
2020-04-03 17:43:04 +03:00
|
|
|
|
2020-08-17 13:01:30 +03:00
|
|
|
max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
|
|
|
|
granularity = backingq->limits.discard_granularity ?:
|
|
|
|
queue_physical_block_size(backingq);
|
2020-04-03 17:43:04 +03:00
|
|
|
|
2011-08-19 16:50:46 +04:00
|
|
|
/*
|
|
|
|
* We use punch hole to reclaim the free space used by the
|
2014-01-22 02:39:24 +04:00
|
|
|
* image a.k.a. discard. However we do not support discard if
|
2011-08-19 16:50:46 +04:00
|
|
|
* encryption is enabled, because it may give an attacker
|
|
|
|
* useful information.
|
|
|
|
*/
|
2020-04-03 17:43:04 +03:00
|
|
|
} else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
|
2020-08-17 13:01:30 +03:00
|
|
|
max_discard_sectors = 0;
|
|
|
|
granularity = 0;
|
2011-08-19 16:50:46 +04:00
|
|
|
|
2020-04-03 17:43:04 +03:00
|
|
|
} else {
|
block: loop:use kstatfs.f_bsize of backing file to set discard granularity
[ Upstream commit 06582bc86d7f48d35cd044098ca1e246e8c7c52e ]
If backing file's filesystem has implemented ->fallocate(), we think the
loop device can support discard, then pass sb->s_blocksize as
discard_granularity. However, some underlying FS, such as overlayfs,
doesn't set sb->s_blocksize, and causes discard_granularity to be set as
zero, then the warning in __blkdev_issue_discard() is triggered.
Christoph suggested to pass kstatfs.f_bsize as discard granularity, and
this way is fine because kstatfs.f_bsize means 'Optimal transfer block
size', which still matches with definition of discard granularity.
So fix the issue by setting discard_granularity as kstatfs.f_bsize if it
is available, otherwise claims discard isn't supported.
Cc: Christoph Hellwig <hch@lst.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Reported-by: Pei Zhang <pezhang@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220126035830.296465-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2022-01-26 06:58:30 +03:00
|
|
|
struct kstatfs sbuf;
|
|
|
|
|
2020-08-17 13:01:30 +03:00
|
|
|
max_discard_sectors = UINT_MAX >> 9;
|
block: loop:use kstatfs.f_bsize of backing file to set discard granularity
[ Upstream commit 06582bc86d7f48d35cd044098ca1e246e8c7c52e ]
If backing file's filesystem has implemented ->fallocate(), we think the
loop device can support discard, then pass sb->s_blocksize as
discard_granularity. However, some underlying FS, such as overlayfs,
doesn't set sb->s_blocksize, and causes discard_granularity to be set as
zero, then the warning in __blkdev_issue_discard() is triggered.
Christoph suggested to pass kstatfs.f_bsize as discard granularity, and
this way is fine because kstatfs.f_bsize means 'Optimal transfer block
size', which still matches with definition of discard granularity.
So fix the issue by setting discard_granularity as kstatfs.f_bsize if it
is available, otherwise claims discard isn't supported.
Cc: Christoph Hellwig <hch@lst.de>
Cc: Vivek Goyal <vgoyal@redhat.com>
Reported-by: Pei Zhang <pezhang@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20220126035830.296465-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
2022-01-26 06:58:30 +03:00
|
|
|
if (!vfs_statfs(&file->f_path, &sbuf))
|
|
|
|
granularity = sbuf.f_bsize;
|
|
|
|
else
|
|
|
|
max_discard_sectors = 0;
|
2020-04-03 17:43:04 +03:00
|
|
|
}
|
|
|
|
|
2020-08-17 13:01:30 +03:00
|
|
|
if (max_discard_sectors) {
|
|
|
|
q->limits.discard_granularity = granularity;
|
|
|
|
blk_queue_max_discard_sectors(q, max_discard_sectors);
|
|
|
|
blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
|
2020-04-03 17:43:04 +03:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
|
2020-08-17 13:01:30 +03:00
|
|
|
} else {
|
|
|
|
q->limits.discard_granularity = 0;
|
|
|
|
blk_queue_max_discard_sectors(q, 0);
|
|
|
|
blk_queue_max_write_zeroes_sectors(q, 0);
|
2020-04-03 17:43:04 +03:00
|
|
|
blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
|
2020-08-17 13:01:30 +03:00
|
|
|
}
|
|
|
|
q->limits.discard_alignment = 0;
|
2011-08-19 16:50:46 +04:00
|
|
|
}
|
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
struct loop_worker {
|
|
|
|
struct rb_node rb_node;
|
|
|
|
struct work_struct work;
|
|
|
|
struct list_head cmd_list;
|
|
|
|
struct list_head idle_list;
|
|
|
|
struct loop_device *lo;
|
2021-06-29 05:38:21 +03:00
|
|
|
struct cgroup_subsys_state *blkcg_css;
|
2021-06-29 05:38:15 +03:00
|
|
|
unsigned long last_ran_at;
|
|
|
|
};
|
|
|
|
|
|
|
|
static void loop_workfn(struct work_struct *work);
|
|
|
|
static void loop_rootcg_workfn(struct work_struct *work);
|
|
|
|
static void loop_free_idle_workers(struct timer_list *timer);
|
|
|
|
|
|
|
|
#ifdef CONFIG_BLK_CGROUP
|
|
|
|
static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
|
2015-08-17 05:31:48 +03:00
|
|
|
{
|
2021-06-29 05:38:15 +03:00
|
|
|
return !css || css == blkcg_root_css;
|
2015-08-17 05:31:48 +03:00
|
|
|
}
|
2021-06-29 05:38:15 +03:00
|
|
|
#else
|
|
|
|
static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
|
loop: Add PF_LESS_THROTTLE to block/loop device thread.
When a filesystem is mounted from a loop device, writes are
throttled by balance_dirty_pages() twice: once when writing
to the filesystem and once when the loop_handle_cmd() writes
to the backing file. This double-throttling can trigger
positive feedback loops that create significant delays. The
throttling at the lower level is seen by the upper level as
a slow device, so it throttles extra hard.
The PF_LESS_THROTTLE flag was created to handle exactly this
circumstance, though with an NFS filesystem mounted from a
local NFS server. It reduces the throttling on the lower
layer so that it can proceed largely unthrottled.
To demonstrate this, create a filesystem on a loop device
and write (e.g. with dd) several large files which combine
to consume significantly more than the limit set by
/proc/sys/vm/dirty_ratio or dirty_bytes. Measure the total
time taken.
When I do this directly on a device (no loop device) the
total time for several runs (mkfs, mount, write 200 files,
umount) is fairly stable: 28-35 seconds.
When I do this over a loop device the times are much worse
and less stable. 52-460 seconds. Half below 100seconds,
half above.
When I apply this patch, the times become stable again,
though not as fast as the no-loop-back case: 53-72 seconds.
There may be room for further improvement as the total overhead still
seems too high, but this is a big improvement.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2017-06-16 08:02:09 +03:00
|
|
|
{
|
2021-06-29 05:38:15 +03:00
|
|
|
return !css;
|
loop: Add PF_LESS_THROTTLE to block/loop device thread.
When a filesystem is mounted from a loop device, writes are
throttled by balance_dirty_pages() twice: once when writing
to the filesystem and once when the loop_handle_cmd() writes
to the backing file. This double-throttling can trigger
positive feedback loops that create significant delays. The
throttling at the lower level is seen by the upper level as
a slow device, so it throttles extra hard.
The PF_LESS_THROTTLE flag was created to handle exactly this
circumstance, though with an NFS filesystem mounted from a
local NFS server. It reduces the throttling on the lower
layer so that it can proceed largely unthrottled.
To demonstrate this, create a filesystem on a loop device
and write (e.g. with dd) several large files which combine
to consume significantly more than the limit set by
/proc/sys/vm/dirty_ratio or dirty_bytes. Measure the total
time taken.
When I do this directly on a device (no loop device) the
total time for several runs (mkfs, mount, write 200 files,
umount) is fairly stable: 28-35 seconds.
When I do this over a loop device the times are much worse
and less stable. 52-460 seconds. Half below 100seconds,
half above.
When I apply this patch, the times become stable again,
though not as fast as the no-loop-back case: 53-72 seconds.
There may be room for further improvement as the total overhead still
seems too high, but this is a big improvement.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2017-06-16 08:02:09 +03:00
|
|
|
}
|
2021-06-29 05:38:15 +03:00
|
|
|
#endif
|
loop: Add PF_LESS_THROTTLE to block/loop device thread.
When a filesystem is mounted from a loop device, writes are
throttled by balance_dirty_pages() twice: once when writing
to the filesystem and once when the loop_handle_cmd() writes
to the backing file. This double-throttling can trigger
positive feedback loops that create significant delays. The
throttling at the lower level is seen by the upper level as
a slow device, so it throttles extra hard.
The PF_LESS_THROTTLE flag was created to handle exactly this
circumstance, though with an NFS filesystem mounted from a
local NFS server. It reduces the throttling on the lower
layer so that it can proceed largely unthrottled.
To demonstrate this, create a filesystem on a loop device
and write (e.g. with dd) several large files which combine
to consume significantly more than the limit set by
/proc/sys/vm/dirty_ratio or dirty_bytes. Measure the total
time taken.
When I do this directly on a device (no loop device) the
total time for several runs (mkfs, mount, write 200 files,
umount) is fairly stable: 28-35 seconds.
When I do this over a loop device the times are much worse
and less stable. 52-460 seconds. Half below 100seconds,
half above.
When I apply this patch, the times become stable again,
though not as fast as the no-loop-back case: 53-72 seconds.
There may be room for further improvement as the total overhead still
seems too high, but this is a big improvement.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <tom.leiming@gmail.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2017-06-16 08:02:09 +03:00
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
|
2015-08-17 05:31:48 +03:00
|
|
|
{
|
2021-06-29 05:38:15 +03:00
|
|
|
struct rb_node **node = &(lo->worker_tree.rb_node), *parent = NULL;
|
|
|
|
struct loop_worker *cur_worker, *worker = NULL;
|
|
|
|
struct work_struct *work;
|
|
|
|
struct list_head *cmd_list;
|
|
|
|
|
|
|
|
spin_lock_irq(&lo->lo_work_lock);
|
|
|
|
|
2021-06-29 05:38:21 +03:00
|
|
|
if (queue_on_root_worker(cmd->blkcg_css))
|
2021-06-29 05:38:15 +03:00
|
|
|
goto queue_work;
|
|
|
|
|
|
|
|
node = &lo->worker_tree.rb_node;
|
|
|
|
|
|
|
|
while (*node) {
|
|
|
|
parent = *node;
|
|
|
|
cur_worker = container_of(*node, struct loop_worker, rb_node);
|
2021-06-29 05:38:21 +03:00
|
|
|
if (cur_worker->blkcg_css == cmd->blkcg_css) {
|
2021-06-29 05:38:15 +03:00
|
|
|
worker = cur_worker;
|
|
|
|
break;
|
2021-06-29 05:38:21 +03:00
|
|
|
} else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
|
2021-06-29 05:38:15 +03:00
|
|
|
node = &(*node)->rb_left;
|
|
|
|
} else {
|
|
|
|
node = &(*node)->rb_right;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (worker)
|
|
|
|
goto queue_work;
|
|
|
|
|
|
|
|
worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
|
|
|
|
/*
|
|
|
|
* In the event we cannot allocate a worker, just queue on the
|
2021-06-29 05:38:21 +03:00
|
|
|
* rootcg worker and issue the I/O as the rootcg
|
2021-06-29 05:38:15 +03:00
|
|
|
*/
|
2021-06-29 05:38:21 +03:00
|
|
|
if (!worker) {
|
|
|
|
cmd->blkcg_css = NULL;
|
|
|
|
if (cmd->memcg_css)
|
|
|
|
css_put(cmd->memcg_css);
|
|
|
|
cmd->memcg_css = NULL;
|
2021-06-29 05:38:15 +03:00
|
|
|
goto queue_work;
|
2021-06-29 05:38:21 +03:00
|
|
|
}
|
2021-06-29 05:38:15 +03:00
|
|
|
|
2021-06-29 05:38:21 +03:00
|
|
|
worker->blkcg_css = cmd->blkcg_css;
|
|
|
|
css_get(worker->blkcg_css);
|
2021-06-29 05:38:15 +03:00
|
|
|
INIT_WORK(&worker->work, loop_workfn);
|
|
|
|
INIT_LIST_HEAD(&worker->cmd_list);
|
|
|
|
INIT_LIST_HEAD(&worker->idle_list);
|
|
|
|
worker->lo = lo;
|
|
|
|
rb_link_node(&worker->rb_node, parent, node);
|
|
|
|
rb_insert_color(&worker->rb_node, &lo->worker_tree);
|
|
|
|
queue_work:
|
|
|
|
if (worker) {
|
|
|
|
/*
|
|
|
|
* We need to remove from the idle list here while
|
|
|
|
* holding the lock so that the idle timer doesn't
|
|
|
|
* free the worker
|
|
|
|
*/
|
|
|
|
if (!list_empty(&worker->idle_list))
|
|
|
|
list_del_init(&worker->idle_list);
|
|
|
|
work = &worker->work;
|
|
|
|
cmd_list = &worker->cmd_list;
|
|
|
|
} else {
|
|
|
|
work = &lo->rootcg_work;
|
|
|
|
cmd_list = &lo->rootcg_cmd_list;
|
|
|
|
}
|
|
|
|
list_add_tail(&cmd->list_entry, cmd_list);
|
|
|
|
queue_work(lo->workqueue, work);
|
|
|
|
spin_unlock_irq(&lo->lo_work_lock);
|
2015-08-17 05:31:48 +03:00
|
|
|
}
|
|
|
|
|
2019-02-13 01:54:24 +03:00
|
|
|
static void loop_update_rotational(struct loop_device *lo)
|
|
|
|
{
|
|
|
|
struct file *file = lo->lo_backing_file;
|
|
|
|
struct inode *file_inode = file->f_mapping->host;
|
|
|
|
struct block_device *file_bdev = file_inode->i_sb->s_bdev;
|
|
|
|
struct request_queue *q = lo->lo_queue;
|
|
|
|
bool nonrot = true;
|
|
|
|
|
|
|
|
/* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
|
|
|
|
if (file_bdev)
|
|
|
|
nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
|
|
|
|
|
|
|
|
if (nonrot)
|
|
|
|
blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
|
|
|
|
else
|
|
|
|
blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
|
|
|
|
}
|
|
|
|
|
2020-05-13 16:38:42 +03:00
|
|
|
static int
|
|
|
|
loop_release_xfer(struct loop_device *lo)
|
|
|
|
{
|
|
|
|
int err = 0;
|
|
|
|
struct loop_func_table *xfer = lo->lo_encryption;
|
|
|
|
|
|
|
|
if (xfer) {
|
|
|
|
if (xfer->release)
|
|
|
|
err = xfer->release(lo);
|
|
|
|
lo->transfer = NULL;
|
|
|
|
lo->lo_encryption = NULL;
|
|
|
|
module_put(xfer->owner);
|
|
|
|
}
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
|
|
|
|
const struct loop_info64 *i)
|
|
|
|
{
|
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
if (xfer) {
|
|
|
|
struct module *owner = xfer->owner;
|
|
|
|
|
|
|
|
if (!try_module_get(owner))
|
|
|
|
return -EINVAL;
|
|
|
|
if (xfer->init)
|
|
|
|
err = xfer->init(lo, i);
|
|
|
|
if (err)
|
|
|
|
module_put(owner);
|
|
|
|
else
|
|
|
|
lo->lo_encryption = xfer;
|
|
|
|
}
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* loop_set_status_from_info - configure device from loop_info
|
|
|
|
* @lo: struct loop_device to configure
|
|
|
|
* @info: struct loop_info64 to configure the device with
|
|
|
|
*
|
|
|
|
* Configures the loop device parameters according to the passed
|
|
|
|
* in loop_info64 configuration.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
loop_set_status_from_info(struct loop_device *lo,
|
|
|
|
const struct loop_info64 *info)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
struct loop_func_table *xfer;
|
|
|
|
kuid_t uid = current_uid();
|
|
|
|
|
|
|
|
if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
err = loop_release_xfer(lo);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
if (info->lo_encrypt_type) {
|
|
|
|
unsigned int type = info->lo_encrypt_type;
|
|
|
|
|
|
|
|
if (type >= MAX_LO_CRYPT)
|
|
|
|
return -EINVAL;
|
|
|
|
xfer = xfer_funcs[type];
|
|
|
|
if (xfer == NULL)
|
|
|
|
return -EINVAL;
|
|
|
|
} else
|
|
|
|
xfer = NULL;
|
|
|
|
|
|
|
|
err = loop_init_xfer(lo, xfer, info);
|
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
|
|
|
|
lo->lo_offset = info->lo_offset;
|
|
|
|
lo->lo_sizelimit = info->lo_sizelimit;
|
|
|
|
memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
|
|
|
|
memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
|
|
|
|
lo->lo_file_name[LO_NAME_SIZE-1] = 0;
|
|
|
|
lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
|
|
|
|
|
|
|
|
if (!xfer)
|
|
|
|
xfer = &none_funcs;
|
|
|
|
lo->transfer = xfer->transfer;
|
|
|
|
lo->ioctl = xfer->ioctl;
|
|
|
|
|
2020-05-13 16:38:44 +03:00
|
|
|
lo->lo_flags = info->lo_flags;
|
2020-05-13 16:38:42 +03:00
|
|
|
|
|
|
|
lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
|
|
|
|
lo->lo_init[0] = info->lo_init[0];
|
|
|
|
lo->lo_init[1] = info->lo_init[1];
|
|
|
|
if (info->lo_encrypt_key_size) {
|
|
|
|
memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
|
|
|
|
info->lo_encrypt_key_size);
|
|
|
|
lo->lo_key_owner = uid;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2020-05-13 16:38:45 +03:00
|
|
|
static int loop_configure(struct loop_device *lo, fmode_t mode,
|
|
|
|
struct block_device *bdev,
|
|
|
|
const struct loop_config *config)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2021-07-06 17:40:34 +03:00
|
|
|
struct file *file = fget(config->fd);
|
|
|
|
struct inode *inode;
|
2005-04-17 02:20:36 +04:00
|
|
|
struct address_space *mapping;
|
2021-07-06 17:40:34 +03:00
|
|
|
int error;
|
|
|
|
loff_t size;
|
|
|
|
bool partscan;
|
|
|
|
unsigned short bsize;
|
|
|
|
bool is_loop;
|
|
|
|
|
|
|
|
if (!file)
|
|
|
|
return -EBADF;
|
|
|
|
is_loop = is_loop_device(file);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
/* This is safe, since we have a reference from open(). */
|
|
|
|
__module_get(THIS_MODULE);
|
|
|
|
|
2019-05-16 17:01:27 +03:00
|
|
|
/*
|
|
|
|
* If we don't hold exclusive handle for the device, upgrade to it
|
|
|
|
* here to avoid changing device under exclusive owner.
|
|
|
|
*/
|
|
|
|
if (!(mode & FMODE_EXCL)) {
|
2020-11-25 23:20:08 +03:00
|
|
|
error = bd_prepare_to_claim(bdev, loop_configure);
|
2020-07-16 17:33:09 +03:00
|
|
|
if (error)
|
2019-05-16 17:01:27 +03:00
|
|
|
goto out_putf;
|
|
|
|
}
|
|
|
|
|
2021-07-06 17:40:34 +03:00
|
|
|
error = loop_global_lock_killable(lo, is_loop);
|
2018-11-08 16:01:10 +03:00
|
|
|
if (error)
|
2019-05-16 17:01:27 +03:00
|
|
|
goto out_bdev;
|
2018-11-08 16:01:10 +03:00
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
error = -EBUSY;
|
|
|
|
if (lo->lo_state != Lo_unbound)
|
2018-11-08 16:01:10 +03:00
|
|
|
goto out_unlock;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2018-05-07 18:37:58 +03:00
|
|
|
error = loop_validate_file(file, bdev);
|
|
|
|
if (error)
|
2018-11-08 16:01:10 +03:00
|
|
|
goto out_unlock;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
mapping = file->f_mapping;
|
|
|
|
inode = mapping->host;
|
|
|
|
|
2020-05-13 16:38:45 +03:00
|
|
|
if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
|
|
|
|
error = -EINVAL;
|
|
|
|
goto out_unlock;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (config->block_size) {
|
2021-10-26 17:40:14 +03:00
|
|
|
error = blk_validate_block_size(config->block_size);
|
2020-05-13 16:38:45 +03:00
|
|
|
if (error)
|
|
|
|
goto out_unlock;
|
|
|
|
}
|
|
|
|
|
|
|
|
error = loop_set_status_from_info(lo, &config->info);
|
|
|
|
if (error)
|
|
|
|
goto out_unlock;
|
|
|
|
|
2011-10-17 14:57:20 +04:00
|
|
|
if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
|
2015-04-03 22:21:59 +03:00
|
|
|
!file->f_op->write_iter)
|
2020-05-13 16:38:45 +03:00
|
|
|
lo->lo_flags |= LO_FLAGS_READ_ONLY;
|
2006-09-27 12:50:49 +04:00
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
lo->workqueue = alloc_workqueue("loop%d",
|
|
|
|
WQ_UNBOUND | WQ_FREEZABLE,
|
|
|
|
0,
|
|
|
|
lo->lo_number);
|
|
|
|
if (!lo->workqueue) {
|
|
|
|
error = -ENOMEM;
|
2018-11-08 16:01:10 +03:00
|
|
|
goto out_unlock;
|
2021-06-29 05:38:15 +03:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-07-13 02:05:30 +03:00
|
|
|
disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
|
2020-11-03 13:00:16 +03:00
|
|
|
set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
|
|
|
|
INIT_LIST_HEAD(&lo->rootcg_cmd_list);
|
|
|
|
INIT_LIST_HEAD(&lo->idle_worker_list);
|
|
|
|
lo->worker_tree = RB_ROOT;
|
|
|
|
timer_setup(&lo->timer, loop_free_idle_workers,
|
|
|
|
TIMER_DEFERRABLE);
|
2020-05-13 16:38:45 +03:00
|
|
|
lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
|
2005-04-17 02:20:36 +04:00
|
|
|
lo->lo_device = bdev;
|
|
|
|
lo->lo_backing_file = file;
|
|
|
|
lo->old_gfp_mask = mapping_gfp_mask(mapping);
|
|
|
|
mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
|
|
|
|
|
2020-05-13 16:38:45 +03:00
|
|
|
if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
|
2016-03-30 19:09:35 +03:00
|
|
|
blk_queue_write_cache(lo->lo_queue, true, false);
|
2009-03-24 14:29:54 +03:00
|
|
|
|
2020-05-13 16:38:45 +03:00
|
|
|
if (config->block_size)
|
|
|
|
bsize = config->block_size;
|
2020-06-03 05:45:12 +03:00
|
|
|
else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
|
2019-09-04 22:49:01 +03:00
|
|
|
/* In case of direct I/O, match underlying block size */
|
2020-05-13 16:38:45 +03:00
|
|
|
bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
|
|
|
|
else
|
|
|
|
bsize = 512;
|
2019-09-04 22:49:01 +03:00
|
|
|
|
2020-05-13 16:38:45 +03:00
|
|
|
blk_queue_logical_block_size(lo->lo_queue, bsize);
|
|
|
|
blk_queue_physical_block_size(lo->lo_queue, bsize);
|
|
|
|
blk_queue_io_min(lo->lo_queue, bsize);
|
2019-09-04 22:49:01 +03:00
|
|
|
|
2021-06-18 14:51:57 +03:00
|
|
|
loop_config_discard(lo);
|
2019-02-13 01:54:24 +03:00
|
|
|
loop_update_rotational(lo);
|
2015-08-17 05:31:49 +03:00
|
|
|
loop_update_dio(lo);
|
2010-08-23 17:16:00 +04:00
|
|
|
loop_sysfs_init(lo);
|
2020-08-25 10:18:29 +03:00
|
|
|
|
|
|
|
size = get_loop_size(lo, file);
|
2020-05-13 16:38:37 +03:00
|
|
|
loop_set_size(lo, size);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-07-06 17:40:34 +03:00
|
|
|
/* Order wrt reading lo_state in loop_validate_file(). */
|
|
|
|
wmb();
|
|
|
|
|
2006-09-29 12:59:11 +04:00
|
|
|
lo->lo_state = Lo_bound;
|
loop: always allow userspace partitions and optionally support automatic scanning
Automatic partition scanning can be requested individually per loop
device during its setup by setting LO_FLAGS_PARTSCAN. By default, no
partition tables are scanned.
Userspace can now always add and remove partitions from all loop
devices, regardless if the in-kernel partition scanner is enabled or
not.
The needed partition minor numbers are allocated from the extended
minors space, the main loop device numbers will continue to match the
loop minors, regardless of the number of partitions used.
# grep . /sys/class/block/loop1/loop/*
/sys/block/loop1/loop/autoclear:0
/sys/block/loop1/loop/backing_file:/home/kay/data/stuff/part.img
/sys/block/loop1/loop/offset:0
/sys/block/loop1/loop/partscan:1
/sys/block/loop1/loop/sizelimit:0
# ls -l /dev/loop*
brw-rw---- 1 root disk 7, 0 Aug 14 20:22 /dev/loop0
brw-rw---- 1 root disk 7, 1 Aug 14 20:23 /dev/loop1
brw-rw---- 1 root disk 259, 0 Aug 14 20:23 /dev/loop1p1
brw-rw---- 1 root disk 259, 1 Aug 14 20:23 /dev/loop1p2
brw-rw---- 1 root disk 7, 99 Aug 14 20:23 /dev/loop99
brw-rw---- 1 root disk 259, 2 Aug 14 20:23 /dev/loop99p1
brw-rw---- 1 root disk 259, 3 Aug 14 20:23 /dev/loop99p2
crw------T 1 root root 10, 237 Aug 14 20:22 /dev/loop-control
Cc: Karel Zak <kzak@redhat.com>
Cc: Davidlohr Bueso <dave@gnu.org>
Acked-By: Tejun Heo <tj@kernel.org>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-23 22:12:04 +04:00
|
|
|
if (part_shift)
|
|
|
|
lo->lo_flags |= LO_FLAGS_PARTSCAN;
|
2018-11-08 16:01:13 +03:00
|
|
|
partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
|
2020-08-10 20:16:32 +03:00
|
|
|
if (partscan)
|
|
|
|
lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
|
2013-04-01 20:47:56 +04:00
|
|
|
|
2021-07-06 17:40:34 +03:00
|
|
|
loop_global_unlock(lo, is_loop);
|
2018-11-08 16:01:13 +03:00
|
|
|
if (partscan)
|
2021-06-24 15:32:40 +03:00
|
|
|
loop_reread_partitions(lo);
|
2020-11-25 23:20:08 +03:00
|
|
|
if (!(mode & FMODE_EXCL))
|
|
|
|
bd_abort_claiming(bdev, loop_configure);
|
2005-04-17 02:20:36 +04:00
|
|
|
return 0;
|
|
|
|
|
2018-11-08 16:01:10 +03:00
|
|
|
out_unlock:
|
2021-07-06 17:40:34 +03:00
|
|
|
loop_global_unlock(lo, is_loop);
|
2019-05-16 17:01:27 +03:00
|
|
|
out_bdev:
|
2020-11-25 23:20:08 +03:00
|
|
|
if (!(mode & FMODE_EXCL))
|
|
|
|
bd_abort_claiming(bdev, loop_configure);
|
2018-11-08 16:01:10 +03:00
|
|
|
out_putf:
|
2005-04-17 02:20:36 +04:00
|
|
|
fput(file);
|
|
|
|
/* This is safe: open() is still holding a reference. */
|
|
|
|
module_put(THIS_MODULE);
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2018-11-08 16:01:14 +03:00
|
|
|
static int __loop_clr_fd(struct loop_device *lo, bool release)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2018-11-08 16:01:07 +03:00
|
|
|
struct file *filp = NULL;
|
2005-10-21 11:22:34 +04:00
|
|
|
gfp_t gfp = lo->old_gfp_mask;
|
2011-09-21 12:02:13 +04:00
|
|
|
struct block_device *bdev = lo->lo_device;
|
2018-11-08 16:01:07 +03:00
|
|
|
int err = 0;
|
2018-11-08 16:01:14 +03:00
|
|
|
bool partscan = false;
|
|
|
|
int lo_number;
|
2021-06-29 05:38:15 +03:00
|
|
|
struct loop_worker *pos, *worker;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-07-06 17:40:34 +03:00
|
|
|
/*
|
|
|
|
* Flush loop_configure() and loop_change_fd(). It is acceptable for
|
|
|
|
* loop_validate_file() to succeed, for actual clear operation has not
|
|
|
|
* started yet.
|
|
|
|
*/
|
|
|
|
mutex_lock(&loop_validate_mutex);
|
|
|
|
mutex_unlock(&loop_validate_mutex);
|
|
|
|
/*
|
|
|
|
* loop_validate_file() now fails because l->lo_state != Lo_bound
|
|
|
|
* became visible.
|
|
|
|
*/
|
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_lock(&lo->lo_mutex);
|
2018-11-08 16:01:07 +03:00
|
|
|
if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
|
|
|
|
err = -ENXIO;
|
|
|
|
goto out_unlock;
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2018-11-08 16:01:07 +03:00
|
|
|
filp = lo->lo_backing_file;
|
|
|
|
if (filp == NULL) {
|
|
|
|
err = -EINVAL;
|
|
|
|
goto out_unlock;
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-02-22 18:41:23 +03:00
|
|
|
if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
|
|
|
|
blk_queue_write_cache(lo->lo_queue, false, false);
|
|
|
|
|
2015-05-06 07:26:23 +03:00
|
|
|
/* freeze request queue during the transition */
|
|
|
|
blk_mq_freeze_queue(lo->lo_queue);
|
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
destroy_workqueue(lo->workqueue);
|
|
|
|
spin_lock_irq(&lo->lo_work_lock);
|
|
|
|
list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
|
|
|
|
idle_list) {
|
|
|
|
list_del(&worker->idle_list);
|
|
|
|
rb_erase(&worker->rb_node, &lo->worker_tree);
|
2021-06-29 05:38:21 +03:00
|
|
|
css_put(worker->blkcg_css);
|
2021-06-29 05:38:15 +03:00
|
|
|
kfree(worker);
|
|
|
|
}
|
|
|
|
spin_unlock_irq(&lo->lo_work_lock);
|
|
|
|
del_timer_sync(&lo->timer);
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
spin_lock_irq(&lo->lo_lock);
|
|
|
|
lo->lo_backing_file = NULL;
|
2011-08-01 00:21:35 +04:00
|
|
|
spin_unlock_irq(&lo->lo_lock);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
loop_release_xfer(lo);
|
|
|
|
lo->transfer = NULL;
|
|
|
|
lo->ioctl = NULL;
|
|
|
|
lo->lo_device = NULL;
|
|
|
|
lo->lo_encryption = NULL;
|
|
|
|
lo->lo_offset = 0;
|
|
|
|
lo->lo_sizelimit = 0;
|
|
|
|
lo->lo_encrypt_key_size = 0;
|
|
|
|
memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
|
|
|
|
memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
|
|
|
|
memset(lo->lo_file_name, 0, LO_NAME_SIZE);
|
2017-08-24 10:03:43 +03:00
|
|
|
blk_queue_logical_block_size(lo->lo_queue, 512);
|
2017-09-06 00:24:47 +03:00
|
|
|
blk_queue_physical_block_size(lo->lo_queue, 512);
|
|
|
|
blk_queue_io_min(lo->lo_queue, 512);
|
2013-04-01 20:47:56 +04:00
|
|
|
if (bdev) {
|
2008-03-02 17:29:48 +03:00
|
|
|
invalidate_bdev(bdev);
|
2018-05-21 21:35:03 +03:00
|
|
|
bdev->bd_inode->i_mapping->wb_err = 0;
|
2013-04-01 20:47:56 +04:00
|
|
|
}
|
2007-05-08 11:28:20 +04:00
|
|
|
set_capacity(lo->lo_disk, 0);
|
2010-10-28 05:51:30 +04:00
|
|
|
loop_sysfs_exit(lo);
|
2010-05-03 16:08:59 +04:00
|
|
|
if (bdev) {
|
|
|
|
/* let user-space know about this change */
|
|
|
|
kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
mapping_set_gfp_mask(filp->f_mapping, gfp);
|
|
|
|
/* This is safe: open() is still holding a reference. */
|
|
|
|
module_put(THIS_MODULE);
|
2015-05-06 07:26:23 +03:00
|
|
|
blk_mq_unfreeze_queue(lo->lo_queue);
|
|
|
|
|
2018-11-08 16:01:14 +03:00
|
|
|
partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
|
|
|
|
lo_number = lo->lo_number;
|
2021-07-13 02:05:30 +03:00
|
|
|
disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
|
2018-11-08 16:01:14 +03:00
|
|
|
out_unlock:
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:14 +03:00
|
|
|
if (partscan) {
|
2018-11-08 16:01:12 +03:00
|
|
|
/*
|
2021-05-25 09:12:56 +03:00
|
|
|
* open_mutex has been held already in release path, so don't
|
2018-11-08 16:01:12 +03:00
|
|
|
* acquire it if this function is called in such case.
|
|
|
|
*
|
|
|
|
* If the reread partition isn't from release path, lo_refcnt
|
|
|
|
* must be at least one and it can only become zero when the
|
|
|
|
* current holder is released.
|
|
|
|
*/
|
2019-11-14 17:34:36 +03:00
|
|
|
if (!release)
|
2021-06-24 15:32:40 +03:00
|
|
|
mutex_lock(&lo->lo_disk->open_mutex);
|
|
|
|
err = bdev_disk_changed(lo->lo_disk, false);
|
2019-11-14 17:34:36 +03:00
|
|
|
if (!release)
|
2021-06-24 15:32:40 +03:00
|
|
|
mutex_unlock(&lo->lo_disk->open_mutex);
|
2019-02-22 17:10:19 +03:00
|
|
|
if (err)
|
|
|
|
pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
|
|
|
|
__func__, lo_number, err);
|
2018-11-08 16:01:12 +03:00
|
|
|
/* Device is gone, no point in returning error */
|
|
|
|
err = 0;
|
|
|
|
}
|
2019-02-22 17:10:20 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* lo->lo_state is set to Lo_unbound here after above partscan has
|
|
|
|
* finished.
|
|
|
|
*
|
|
|
|
* There cannot be anybody else entering __loop_clr_fd() as
|
|
|
|
* lo->lo_backing_file is already cleared and Lo_rundown state
|
|
|
|
* protects us from all the other places trying to change the 'lo'
|
|
|
|
* device.
|
|
|
|
*/
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_lock(&lo->lo_mutex);
|
2019-02-22 17:10:20 +03:00
|
|
|
lo->lo_flags = 0;
|
|
|
|
if (!part_shift)
|
|
|
|
lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
|
|
|
|
lo->lo_state = Lo_unbound;
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2019-02-22 17:10:20 +03:00
|
|
|
|
loop: fix circular locking in loop_clr_fd()
With CONFIG_PROVE_LOCKING enabled
$ losetup /dev/loop0 file
$ losetup -o 32256 /dev/loop1 /dev/loop0
$ losetup -d /dev/loop1
$ losetup -d /dev/loop0
triggers a [ INFO: possible circular locking dependency detected ]
I think this warning is a false positive.
Open/close on a loop device acquires bd_mutex of the device before
acquiring lo_ctl_mutex of the same device. For ioctl(LOOP_CLR_FD) after
acquiring lo_ctl_mutex, fput on the backing_file might acquire the bd_mutex of
a device, if backing file is a device and this is the last reference to the
file being dropped . But it is guaranteed that it is impossible to have a
circular list of backing devices.(say loop2->loop1->loop0->loop2 is not
possible), which guarantees that this can never deadlock.
So this warning should be suppressed. It is very difficult to annotate lockdep
not to warn here in the correct way. A simple way to silence lockdep could be
to mark the lo_ctl_mutex in ioctl to be a sub class, but this might mask some
other real bugs.
@@ -1164,7 +1164,7 @@ static int lo_ioctl(struct block_device *bdev, fmode_t mode,
struct loop_device *lo = bdev->bd_disk->private_data;
int err;
- mutex_lock(&lo->lo_ctl_mutex);
+ mutex_lock_nested(&lo->lo_ctl_mutex, 1);
switch (cmd) {
case LOOP_SET_FD:
err = loop_set_fd(lo, mode, bdev, arg);
Or actually marking the bd_mutex after lo_ctl_mutex as a sub class could be
a better solution.
Luckily it is easy to avoid calling fput on backing file with lo_ctl_mutex
held, so no lockdep annotation is required.
If you do not like the special handling of the lo_ctl_mutex just for the
LOOP_CLR_FD ioctl in lo_ioctl(), the mutex handling could be moved inside
each of the individual ioctl handlers and I could send you another patch.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2009-03-24 14:33:41 +03:00
|
|
|
/*
|
2021-01-26 17:46:30 +03:00
|
|
|
* Need not hold lo_mutex to fput backing file. Calling fput holding
|
|
|
|
* lo_mutex triggers a circular lock dependency possibility warning as
|
2021-05-25 09:12:56 +03:00
|
|
|
* fput can take open_mutex which is usually taken before lo_mutex.
|
loop: fix circular locking in loop_clr_fd()
With CONFIG_PROVE_LOCKING enabled
$ losetup /dev/loop0 file
$ losetup -o 32256 /dev/loop1 /dev/loop0
$ losetup -d /dev/loop1
$ losetup -d /dev/loop0
triggers a [ INFO: possible circular locking dependency detected ]
I think this warning is a false positive.
Open/close on a loop device acquires bd_mutex of the device before
acquiring lo_ctl_mutex of the same device. For ioctl(LOOP_CLR_FD) after
acquiring lo_ctl_mutex, fput on the backing_file might acquire the bd_mutex of
a device, if backing file is a device and this is the last reference to the
file being dropped . But it is guaranteed that it is impossible to have a
circular list of backing devices.(say loop2->loop1->loop0->loop2 is not
possible), which guarantees that this can never deadlock.
So this warning should be suppressed. It is very difficult to annotate lockdep
not to warn here in the correct way. A simple way to silence lockdep could be
to mark the lo_ctl_mutex in ioctl to be a sub class, but this might mask some
other real bugs.
@@ -1164,7 +1164,7 @@ static int lo_ioctl(struct block_device *bdev, fmode_t mode,
struct loop_device *lo = bdev->bd_disk->private_data;
int err;
- mutex_lock(&lo->lo_ctl_mutex);
+ mutex_lock_nested(&lo->lo_ctl_mutex, 1);
switch (cmd) {
case LOOP_SET_FD:
err = loop_set_fd(lo, mode, bdev, arg);
Or actually marking the bd_mutex after lo_ctl_mutex as a sub class could be
a better solution.
Luckily it is easy to avoid calling fput on backing file with lo_ctl_mutex
held, so no lockdep annotation is required.
If you do not like the special handling of the lo_ctl_mutex just for the
LOOP_CLR_FD ioctl in lo_ioctl(), the mutex handling could be moved inside
each of the individual ioctl handlers and I could send you another patch.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2009-03-24 14:33:41 +03:00
|
|
|
*/
|
2018-11-08 16:01:07 +03:00
|
|
|
if (filp)
|
|
|
|
fput(filp);
|
|
|
|
return err;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2018-11-08 16:01:06 +03:00
|
|
|
static int loop_clr_fd(struct loop_device *lo)
|
|
|
|
{
|
2018-11-08 16:01:07 +03:00
|
|
|
int err;
|
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
err = mutex_lock_killable(&lo->lo_mutex);
|
2018-11-08 16:01:07 +03:00
|
|
|
if (err)
|
|
|
|
return err;
|
|
|
|
if (lo->lo_state != Lo_bound) {
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:06 +03:00
|
|
|
return -ENXIO;
|
2018-11-08 16:01:07 +03:00
|
|
|
}
|
2018-11-08 16:01:06 +03:00
|
|
|
/*
|
|
|
|
* If we've explicitly asked to tear down the loop device,
|
|
|
|
* and it has an elevated reference count, set it for auto-teardown when
|
|
|
|
* the last reference goes away. This stops $!~#$@ udev from
|
|
|
|
* preventing teardown because it decided that it needs to run blkid on
|
|
|
|
* the loopback device whenever they appear. xfstests is notorious for
|
|
|
|
* failing tests because blkid via udev races with a losetup
|
|
|
|
* <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
|
|
|
|
* command to fail with EBUSY.
|
|
|
|
*/
|
|
|
|
if (atomic_read(&lo->lo_refcnt) > 1) {
|
|
|
|
lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:06 +03:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
lo->lo_state = Lo_rundown;
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:06 +03:00
|
|
|
|
2018-11-08 16:01:14 +03:00
|
|
|
return __loop_clr_fd(lo, false);
|
2018-11-08 16:01:06 +03:00
|
|
|
}
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
static int
|
|
|
|
loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
|
|
|
|
{
|
|
|
|
int err;
|
2020-05-13 16:38:41 +03:00
|
|
|
kuid_t uid = current_uid();
|
2020-05-13 16:38:44 +03:00
|
|
|
int prev_lo_flags;
|
2018-11-08 16:01:13 +03:00
|
|
|
bool partscan = false;
|
2020-05-13 16:38:41 +03:00
|
|
|
bool size_changed = false;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
err = mutex_lock_killable(&lo->lo_mutex);
|
2018-11-08 16:01:09 +03:00
|
|
|
if (err)
|
|
|
|
return err;
|
2008-11-14 02:38:41 +03:00
|
|
|
if (lo->lo_encrypt_key_size &&
|
2012-02-11 23:23:51 +04:00
|
|
|
!uid_eq(lo->lo_key_owner, uid) &&
|
2018-11-08 16:01:09 +03:00
|
|
|
!capable(CAP_SYS_ADMIN)) {
|
|
|
|
err = -EPERM;
|
|
|
|
goto out_unlock;
|
|
|
|
}
|
|
|
|
if (lo->lo_state != Lo_bound) {
|
|
|
|
err = -ENXIO;
|
|
|
|
goto out_unlock;
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2019-01-10 06:17:14 +03:00
|
|
|
if (lo->lo_offset != info->lo_offset ||
|
|
|
|
lo->lo_sizelimit != info->lo_sizelimit) {
|
2020-05-13 16:38:41 +03:00
|
|
|
size_changed = true;
|
2019-01-10 06:17:14 +03:00
|
|
|
sync_blockdev(lo->lo_device);
|
2020-06-18 07:21:37 +03:00
|
|
|
invalidate_bdev(lo->lo_device);
|
2019-01-10 06:17:14 +03:00
|
|
|
}
|
|
|
|
|
2017-02-11 06:40:45 +03:00
|
|
|
/* I/O need to be drained during transfer transition */
|
|
|
|
blk_mq_freeze_queue(lo->lo_queue);
|
|
|
|
|
2020-05-13 16:38:41 +03:00
|
|
|
if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
|
2020-06-18 07:21:37 +03:00
|
|
|
/* If any pages were dirtied after invalidate_bdev(), try again */
|
2020-05-13 16:38:41 +03:00
|
|
|
err = -EAGAIN;
|
|
|
|
pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
|
|
|
|
__func__, lo->lo_number, lo->lo_file_name,
|
|
|
|
lo->lo_device->bd_inode->i_mapping->nrpages);
|
2018-11-08 16:01:09 +03:00
|
|
|
goto out_unfreeze;
|
2020-05-13 16:38:41 +03:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2020-05-13 16:38:44 +03:00
|
|
|
prev_lo_flags = lo->lo_flags;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2020-05-13 16:38:41 +03:00
|
|
|
err = loop_set_status_from_info(lo, info);
|
2005-04-17 02:20:36 +04:00
|
|
|
if (err)
|
2018-11-08 16:01:09 +03:00
|
|
|
goto out_unfreeze;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2020-05-13 16:38:44 +03:00
|
|
|
/* Mask out flags that can't be set using LOOP_SET_STATUS. */
|
2020-06-04 23:25:20 +03:00
|
|
|
lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
|
2020-05-13 16:38:44 +03:00
|
|
|
/* For those flags, use the previous values instead */
|
|
|
|
lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
|
|
|
|
/* For flags that can't be cleared, use previous values too */
|
|
|
|
lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
|
|
|
|
|
2020-05-13 16:38:39 +03:00
|
|
|
if (size_changed) {
|
|
|
|
loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
|
|
|
|
lo->lo_backing_file);
|
|
|
|
loop_set_size(lo, new_size);
|
2017-06-09 13:19:18 +03:00
|
|
|
}
|
2013-02-22 03:16:46 +04:00
|
|
|
|
2011-08-19 16:50:46 +04:00
|
|
|
loop_config_discard(lo);
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2015-08-17 05:31:49 +03:00
|
|
|
/* update dio if lo_offset or transfer is changed */
|
|
|
|
__loop_update_dio(lo, lo->use_dio);
|
|
|
|
|
2018-11-08 16:01:09 +03:00
|
|
|
out_unfreeze:
|
2017-02-11 06:40:45 +03:00
|
|
|
blk_mq_unfreeze_queue(lo->lo_queue);
|
2017-03-01 21:42:38 +03:00
|
|
|
|
2020-05-13 16:38:44 +03:00
|
|
|
if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
|
|
|
|
!(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
|
2017-03-01 21:42:38 +03:00
|
|
|
lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
|
2018-11-08 16:01:13 +03:00
|
|
|
partscan = true;
|
2017-03-01 21:42:38 +03:00
|
|
|
}
|
2018-11-08 16:01:09 +03:00
|
|
|
out_unlock:
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:13 +03:00
|
|
|
if (partscan)
|
2021-06-24 15:32:40 +03:00
|
|
|
loop_reread_partitions(lo);
|
2017-03-01 21:42:38 +03:00
|
|
|
|
2017-02-11 06:40:45 +03:00
|
|
|
return err;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_get_status(struct loop_device *lo, struct loop_info64 *info)
|
|
|
|
{
|
2018-11-08 16:01:01 +03:00
|
|
|
struct path path;
|
2005-04-17 02:20:36 +04:00
|
|
|
struct kstat stat;
|
2018-03-27 07:39:11 +03:00
|
|
|
int ret;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
ret = mutex_lock_killable(&lo->lo_mutex);
|
2018-11-08 16:01:08 +03:00
|
|
|
if (ret)
|
|
|
|
return ret;
|
2018-03-27 07:39:11 +03:00
|
|
|
if (lo->lo_state != Lo_bound) {
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2005-04-17 02:20:36 +04:00
|
|
|
return -ENXIO;
|
2018-03-27 07:39:11 +03:00
|
|
|
}
|
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
memset(info, 0, sizeof(*info));
|
|
|
|
info->lo_number = lo->lo_number;
|
|
|
|
info->lo_offset = lo->lo_offset;
|
|
|
|
info->lo_sizelimit = lo->lo_sizelimit;
|
|
|
|
info->lo_flags = lo->lo_flags;
|
|
|
|
memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
|
|
|
|
memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
|
|
|
|
info->lo_encrypt_type =
|
|
|
|
lo->lo_encryption ? lo->lo_encryption->number : 0;
|
|
|
|
if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
|
|
|
|
info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
|
|
|
|
memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
|
|
|
|
lo->lo_encrypt_key_size);
|
|
|
|
}
|
2018-03-27 07:39:11 +03:00
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
/* Drop lo_mutex while we call into the filesystem. */
|
2018-11-08 16:01:01 +03:00
|
|
|
path = lo->lo_backing_file->f_path;
|
|
|
|
path_get(&path);
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:01 +03:00
|
|
|
ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
|
2018-03-27 07:39:11 +03:00
|
|
|
if (!ret) {
|
|
|
|
info->lo_device = huge_encode_dev(stat.dev);
|
|
|
|
info->lo_inode = stat.ino;
|
|
|
|
info->lo_rdevice = huge_encode_dev(stat.rdev);
|
|
|
|
}
|
2018-11-08 16:01:01 +03:00
|
|
|
path_put(&path);
|
2018-03-27 07:39:11 +03:00
|
|
|
return ret;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
|
|
|
loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
|
|
|
|
{
|
|
|
|
memset(info64, 0, sizeof(*info64));
|
|
|
|
info64->lo_number = info->lo_number;
|
|
|
|
info64->lo_device = info->lo_device;
|
|
|
|
info64->lo_inode = info->lo_inode;
|
|
|
|
info64->lo_rdevice = info->lo_rdevice;
|
|
|
|
info64->lo_offset = info->lo_offset;
|
|
|
|
info64->lo_sizelimit = 0;
|
|
|
|
info64->lo_encrypt_type = info->lo_encrypt_type;
|
|
|
|
info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
|
|
|
|
info64->lo_flags = info->lo_flags;
|
|
|
|
info64->lo_init[0] = info->lo_init[0];
|
|
|
|
info64->lo_init[1] = info->lo_init[1];
|
|
|
|
if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
|
|
|
memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
|
|
|
|
else
|
|
|
|
memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
|
|
|
|
memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
|
|
|
|
{
|
|
|
|
memset(info, 0, sizeof(*info));
|
|
|
|
info->lo_number = info64->lo_number;
|
|
|
|
info->lo_device = info64->lo_device;
|
|
|
|
info->lo_inode = info64->lo_inode;
|
|
|
|
info->lo_rdevice = info64->lo_rdevice;
|
|
|
|
info->lo_offset = info64->lo_offset;
|
|
|
|
info->lo_encrypt_type = info64->lo_encrypt_type;
|
|
|
|
info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
|
|
|
|
info->lo_flags = info64->lo_flags;
|
|
|
|
info->lo_init[0] = info64->lo_init[0];
|
|
|
|
info->lo_init[1] = info64->lo_init[1];
|
|
|
|
if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
|
|
|
memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
|
|
|
|
else
|
|
|
|
memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
|
|
|
|
memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
|
|
|
|
|
|
|
|
/* error in case values were truncated */
|
|
|
|
if (info->lo_device != info64->lo_device ||
|
|
|
|
info->lo_rdevice != info64->lo_rdevice ||
|
|
|
|
info->lo_inode != info64->lo_inode ||
|
|
|
|
info->lo_offset != info64->lo_offset)
|
|
|
|
return -EOVERFLOW;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
|
|
|
|
{
|
|
|
|
struct loop_info info;
|
|
|
|
struct loop_info64 info64;
|
|
|
|
|
|
|
|
if (copy_from_user(&info, arg, sizeof (struct loop_info)))
|
|
|
|
return -EFAULT;
|
|
|
|
loop_info64_from_old(&info, &info64);
|
|
|
|
return loop_set_status(lo, &info64);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
|
|
|
|
{
|
|
|
|
struct loop_info64 info64;
|
|
|
|
|
|
|
|
if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
|
|
|
|
return -EFAULT;
|
|
|
|
return loop_set_status(lo, &info64);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
|
|
|
|
struct loop_info info;
|
|
|
|
struct loop_info64 info64;
|
2018-04-06 19:57:03 +03:00
|
|
|
int err;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2018-11-08 16:01:08 +03:00
|
|
|
if (!arg)
|
2018-04-06 19:57:03 +03:00
|
|
|
return -EINVAL;
|
|
|
|
err = loop_get_status(lo, &info64);
|
2005-04-17 02:20:36 +04:00
|
|
|
if (!err)
|
|
|
|
err = loop_info64_to_old(&info64, &info);
|
|
|
|
if (!err && copy_to_user(arg, &info, sizeof(info)))
|
|
|
|
err = -EFAULT;
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
|
|
|
|
struct loop_info64 info64;
|
2018-04-06 19:57:03 +03:00
|
|
|
int err;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2018-11-08 16:01:08 +03:00
|
|
|
if (!arg)
|
2018-04-06 19:57:03 +03:00
|
|
|
return -EINVAL;
|
|
|
|
err = loop_get_status(lo, &info64);
|
2005-04-17 02:20:36 +04:00
|
|
|
if (!err && copy_to_user(arg, &info64, sizeof(info64)))
|
|
|
|
err = -EFAULT;
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2017-06-08 14:46:44 +03:00
|
|
|
static int loop_set_capacity(struct loop_device *lo)
|
loop: add ioctl to resize a loop device
Add the ability to 'resize' the loop device on the fly.
One practical application is a loop file with XFS filesystem, already
mounted: You can easily enlarge the file (append some bytes) and then call
ioctl(fd, LOOP_SET_CAPACITY, new); The loop driver will learn about the
new size and you can use xfs_growfs later on, which will allow you to use
full capacity of the loop file without the need to unmount.
Test app:
#include <linux/fs.h>
#include <linux/loop.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define _GNU_SOURCE
#include <getopt.h>
char *me;
void usage(FILE *f)
{
fprintf(f, "%s [options] loop_dev [backend_file]\n"
"-s, --set new_size_in_bytes\n"
"\twhen backend_file is given, "
"it will be expanded too while keeping the original contents\n",
me);
}
struct option opts[] = {
{
.name = "set",
.has_arg = 1,
.flag = NULL,
.val = 's'
},
{
.name = "help",
.has_arg = 0,
.flag = NULL,
.val = 'h'
}
};
void err_size(char *name, __u64 old)
{
fprintf(stderr, "size must be larger than current %s (%llu)\n",
name, old);
}
int main(int argc, char *argv[])
{
int fd, err, c, i, bfd;
ssize_t ssz;
size_t sz;
__u64 old, new, append;
char a[BUFSIZ];
struct stat st;
FILE *out;
char *backend, *dev;
err = EINVAL;
out = stderr;
me = argv[0];
new = 0;
while ((c = getopt_long(argc, argv, "s:h", opts, &i)) != -1) {
switch (c) {
case 's':
errno = 0;
new = strtoull(optarg, NULL, 0);
if (errno) {
err = errno;
perror(argv[i]);
goto out;
}
break;
case 'h':
err = 0;
out = stdout;
goto err;
default:
perror(argv[i]);
goto err;
}
}
if (optind < argc)
dev = argv[optind++];
else
goto err;
fd = open(dev, O_RDONLY);
if (fd < 0) {
err = errno;
perror(dev);
goto out;
}
err = ioctl(fd, BLKGETSIZE64, &old);
if (err) {
err = errno;
perror("ioctl BLKGETSIZE64");
goto out;
}
if (!new) {
printf("%llu\n", old);
goto out;
}
if (new < old) {
err = EINVAL;
err_size(dev, old);
goto out;
}
if (optind < argc) {
backend = argv[optind++];
bfd = open(backend, O_WRONLY|O_APPEND);
if (bfd < 0) {
err = errno;
perror(backend);
goto out;
}
err = fstat(bfd, &st);
if (err) {
err = errno;
perror(backend);
goto out;
}
if (new < st.st_size) {
err = EINVAL;
err_size(backend, st.st_size);
goto out;
}
append = new - st.st_size;
sz = sizeof(a);
while (append > 0) {
if (append < sz)
sz = append;
ssz = write(bfd, a, sz);
if (ssz != sz) {
err = errno;
perror(backend);
goto out;
}
append -= sz;
}
err = fsync(bfd);
if (err) {
err = errno;
perror(backend);
goto out;
}
}
err = ioctl(fd, LOOP_SET_CAPACITY, new);
if (err) {
err = errno;
perror("ioctl LOOP_SET_CAPACITY");
}
goto out;
err:
usage(out);
out:
return err;
}
Signed-off-by: J. R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Tomas Matejicek <tomas@slax.org>
Cc: <util-linux-ng@vger.kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-01 02:23:43 +04:00
|
|
|
{
|
2020-05-13 16:38:40 +03:00
|
|
|
loff_t size;
|
|
|
|
|
loop: add ioctl to resize a loop device
Add the ability to 'resize' the loop device on the fly.
One practical application is a loop file with XFS filesystem, already
mounted: You can easily enlarge the file (append some bytes) and then call
ioctl(fd, LOOP_SET_CAPACITY, new); The loop driver will learn about the
new size and you can use xfs_growfs later on, which will allow you to use
full capacity of the loop file without the need to unmount.
Test app:
#include <linux/fs.h>
#include <linux/loop.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define _GNU_SOURCE
#include <getopt.h>
char *me;
void usage(FILE *f)
{
fprintf(f, "%s [options] loop_dev [backend_file]\n"
"-s, --set new_size_in_bytes\n"
"\twhen backend_file is given, "
"it will be expanded too while keeping the original contents\n",
me);
}
struct option opts[] = {
{
.name = "set",
.has_arg = 1,
.flag = NULL,
.val = 's'
},
{
.name = "help",
.has_arg = 0,
.flag = NULL,
.val = 'h'
}
};
void err_size(char *name, __u64 old)
{
fprintf(stderr, "size must be larger than current %s (%llu)\n",
name, old);
}
int main(int argc, char *argv[])
{
int fd, err, c, i, bfd;
ssize_t ssz;
size_t sz;
__u64 old, new, append;
char a[BUFSIZ];
struct stat st;
FILE *out;
char *backend, *dev;
err = EINVAL;
out = stderr;
me = argv[0];
new = 0;
while ((c = getopt_long(argc, argv, "s:h", opts, &i)) != -1) {
switch (c) {
case 's':
errno = 0;
new = strtoull(optarg, NULL, 0);
if (errno) {
err = errno;
perror(argv[i]);
goto out;
}
break;
case 'h':
err = 0;
out = stdout;
goto err;
default:
perror(argv[i]);
goto err;
}
}
if (optind < argc)
dev = argv[optind++];
else
goto err;
fd = open(dev, O_RDONLY);
if (fd < 0) {
err = errno;
perror(dev);
goto out;
}
err = ioctl(fd, BLKGETSIZE64, &old);
if (err) {
err = errno;
perror("ioctl BLKGETSIZE64");
goto out;
}
if (!new) {
printf("%llu\n", old);
goto out;
}
if (new < old) {
err = EINVAL;
err_size(dev, old);
goto out;
}
if (optind < argc) {
backend = argv[optind++];
bfd = open(backend, O_WRONLY|O_APPEND);
if (bfd < 0) {
err = errno;
perror(backend);
goto out;
}
err = fstat(bfd, &st);
if (err) {
err = errno;
perror(backend);
goto out;
}
if (new < st.st_size) {
err = EINVAL;
err_size(backend, st.st_size);
goto out;
}
append = new - st.st_size;
sz = sizeof(a);
while (append > 0) {
if (append < sz)
sz = append;
ssz = write(bfd, a, sz);
if (ssz != sz) {
err = errno;
perror(backend);
goto out;
}
append -= sz;
}
err = fsync(bfd);
if (err) {
err = errno;
perror(backend);
goto out;
}
}
err = ioctl(fd, LOOP_SET_CAPACITY, new);
if (err) {
err = errno;
perror("ioctl LOOP_SET_CAPACITY");
}
goto out;
err:
usage(out);
out:
return err;
}
Signed-off-by: J. R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Tomas Matejicek <tomas@slax.org>
Cc: <util-linux-ng@vger.kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-01 02:23:43 +04:00
|
|
|
if (unlikely(lo->lo_state != Lo_bound))
|
2013-02-22 03:16:47 +04:00
|
|
|
return -ENXIO;
|
loop: add ioctl to resize a loop device
Add the ability to 'resize' the loop device on the fly.
One practical application is a loop file with XFS filesystem, already
mounted: You can easily enlarge the file (append some bytes) and then call
ioctl(fd, LOOP_SET_CAPACITY, new); The loop driver will learn about the
new size and you can use xfs_growfs later on, which will allow you to use
full capacity of the loop file without the need to unmount.
Test app:
#include <linux/fs.h>
#include <linux/loop.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define _GNU_SOURCE
#include <getopt.h>
char *me;
void usage(FILE *f)
{
fprintf(f, "%s [options] loop_dev [backend_file]\n"
"-s, --set new_size_in_bytes\n"
"\twhen backend_file is given, "
"it will be expanded too while keeping the original contents\n",
me);
}
struct option opts[] = {
{
.name = "set",
.has_arg = 1,
.flag = NULL,
.val = 's'
},
{
.name = "help",
.has_arg = 0,
.flag = NULL,
.val = 'h'
}
};
void err_size(char *name, __u64 old)
{
fprintf(stderr, "size must be larger than current %s (%llu)\n",
name, old);
}
int main(int argc, char *argv[])
{
int fd, err, c, i, bfd;
ssize_t ssz;
size_t sz;
__u64 old, new, append;
char a[BUFSIZ];
struct stat st;
FILE *out;
char *backend, *dev;
err = EINVAL;
out = stderr;
me = argv[0];
new = 0;
while ((c = getopt_long(argc, argv, "s:h", opts, &i)) != -1) {
switch (c) {
case 's':
errno = 0;
new = strtoull(optarg, NULL, 0);
if (errno) {
err = errno;
perror(argv[i]);
goto out;
}
break;
case 'h':
err = 0;
out = stdout;
goto err;
default:
perror(argv[i]);
goto err;
}
}
if (optind < argc)
dev = argv[optind++];
else
goto err;
fd = open(dev, O_RDONLY);
if (fd < 0) {
err = errno;
perror(dev);
goto out;
}
err = ioctl(fd, BLKGETSIZE64, &old);
if (err) {
err = errno;
perror("ioctl BLKGETSIZE64");
goto out;
}
if (!new) {
printf("%llu\n", old);
goto out;
}
if (new < old) {
err = EINVAL;
err_size(dev, old);
goto out;
}
if (optind < argc) {
backend = argv[optind++];
bfd = open(backend, O_WRONLY|O_APPEND);
if (bfd < 0) {
err = errno;
perror(backend);
goto out;
}
err = fstat(bfd, &st);
if (err) {
err = errno;
perror(backend);
goto out;
}
if (new < st.st_size) {
err = EINVAL;
err_size(backend, st.st_size);
goto out;
}
append = new - st.st_size;
sz = sizeof(a);
while (append > 0) {
if (append < sz)
sz = append;
ssz = write(bfd, a, sz);
if (ssz != sz) {
err = errno;
perror(backend);
goto out;
}
append -= sz;
}
err = fsync(bfd);
if (err) {
err = errno;
perror(backend);
goto out;
}
}
err = ioctl(fd, LOOP_SET_CAPACITY, new);
if (err) {
err = errno;
perror("ioctl LOOP_SET_CAPACITY");
}
goto out;
err:
usage(out);
out:
return err;
}
Signed-off-by: J. R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Tomas Matejicek <tomas@slax.org>
Cc: <util-linux-ng@vger.kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-01 02:23:43 +04:00
|
|
|
|
2020-05-13 16:38:40 +03:00
|
|
|
size = get_loop_size(lo, lo->lo_backing_file);
|
|
|
|
loop_set_size(lo, size);
|
2020-05-13 16:38:36 +03:00
|
|
|
|
|
|
|
return 0;
|
loop: add ioctl to resize a loop device
Add the ability to 'resize' the loop device on the fly.
One practical application is a loop file with XFS filesystem, already
mounted: You can easily enlarge the file (append some bytes) and then call
ioctl(fd, LOOP_SET_CAPACITY, new); The loop driver will learn about the
new size and you can use xfs_growfs later on, which will allow you to use
full capacity of the loop file without the need to unmount.
Test app:
#include <linux/fs.h>
#include <linux/loop.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define _GNU_SOURCE
#include <getopt.h>
char *me;
void usage(FILE *f)
{
fprintf(f, "%s [options] loop_dev [backend_file]\n"
"-s, --set new_size_in_bytes\n"
"\twhen backend_file is given, "
"it will be expanded too while keeping the original contents\n",
me);
}
struct option opts[] = {
{
.name = "set",
.has_arg = 1,
.flag = NULL,
.val = 's'
},
{
.name = "help",
.has_arg = 0,
.flag = NULL,
.val = 'h'
}
};
void err_size(char *name, __u64 old)
{
fprintf(stderr, "size must be larger than current %s (%llu)\n",
name, old);
}
int main(int argc, char *argv[])
{
int fd, err, c, i, bfd;
ssize_t ssz;
size_t sz;
__u64 old, new, append;
char a[BUFSIZ];
struct stat st;
FILE *out;
char *backend, *dev;
err = EINVAL;
out = stderr;
me = argv[0];
new = 0;
while ((c = getopt_long(argc, argv, "s:h", opts, &i)) != -1) {
switch (c) {
case 's':
errno = 0;
new = strtoull(optarg, NULL, 0);
if (errno) {
err = errno;
perror(argv[i]);
goto out;
}
break;
case 'h':
err = 0;
out = stdout;
goto err;
default:
perror(argv[i]);
goto err;
}
}
if (optind < argc)
dev = argv[optind++];
else
goto err;
fd = open(dev, O_RDONLY);
if (fd < 0) {
err = errno;
perror(dev);
goto out;
}
err = ioctl(fd, BLKGETSIZE64, &old);
if (err) {
err = errno;
perror("ioctl BLKGETSIZE64");
goto out;
}
if (!new) {
printf("%llu\n", old);
goto out;
}
if (new < old) {
err = EINVAL;
err_size(dev, old);
goto out;
}
if (optind < argc) {
backend = argv[optind++];
bfd = open(backend, O_WRONLY|O_APPEND);
if (bfd < 0) {
err = errno;
perror(backend);
goto out;
}
err = fstat(bfd, &st);
if (err) {
err = errno;
perror(backend);
goto out;
}
if (new < st.st_size) {
err = EINVAL;
err_size(backend, st.st_size);
goto out;
}
append = new - st.st_size;
sz = sizeof(a);
while (append > 0) {
if (append < sz)
sz = append;
ssz = write(bfd, a, sz);
if (ssz != sz) {
err = errno;
perror(backend);
goto out;
}
append -= sz;
}
err = fsync(bfd);
if (err) {
err = errno;
perror(backend);
goto out;
}
}
err = ioctl(fd, LOOP_SET_CAPACITY, new);
if (err) {
err = errno;
perror("ioctl LOOP_SET_CAPACITY");
}
goto out;
err:
usage(out);
out:
return err;
}
Signed-off-by: J. R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Tomas Matejicek <tomas@slax.org>
Cc: <util-linux-ng@vger.kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-01 02:23:43 +04:00
|
|
|
}
|
|
|
|
|
2015-08-17 05:31:50 +03:00
|
|
|
static int loop_set_dio(struct loop_device *lo, unsigned long arg)
|
|
|
|
{
|
|
|
|
int error = -ENXIO;
|
|
|
|
if (lo->lo_state != Lo_bound)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
__loop_update_dio(lo, !!arg);
|
|
|
|
if (lo->use_dio == !!arg)
|
|
|
|
return 0;
|
|
|
|
error = -EINVAL;
|
|
|
|
out:
|
|
|
|
return error;
|
|
|
|
}
|
|
|
|
|
2017-08-24 10:03:43 +03:00
|
|
|
static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
|
|
|
|
{
|
2019-01-10 06:17:14 +03:00
|
|
|
int err = 0;
|
|
|
|
|
2017-08-24 10:03:43 +03:00
|
|
|
if (lo->lo_state != Lo_bound)
|
|
|
|
return -ENXIO;
|
|
|
|
|
2021-10-26 17:40:14 +03:00
|
|
|
err = blk_validate_block_size(arg);
|
2020-05-13 16:38:45 +03:00
|
|
|
if (err)
|
|
|
|
return err;
|
2017-08-24 10:03:43 +03:00
|
|
|
|
2020-03-10 16:12:30 +03:00
|
|
|
if (lo->lo_queue->limits.logical_block_size == arg)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
sync_blockdev(lo->lo_device);
|
2020-06-18 07:21:37 +03:00
|
|
|
invalidate_bdev(lo->lo_device);
|
2019-01-10 06:17:14 +03:00
|
|
|
|
2017-08-24 10:03:43 +03:00
|
|
|
blk_mq_freeze_queue(lo->lo_queue);
|
|
|
|
|
2020-06-18 07:21:37 +03:00
|
|
|
/* invalidate_bdev should have truncated all the pages */
|
2020-03-10 16:12:30 +03:00
|
|
|
if (lo->lo_device->bd_inode->i_mapping->nrpages) {
|
2019-01-10 06:17:14 +03:00
|
|
|
err = -EAGAIN;
|
|
|
|
pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
|
|
|
|
__func__, lo->lo_number, lo->lo_file_name,
|
|
|
|
lo->lo_device->bd_inode->i_mapping->nrpages);
|
|
|
|
goto out_unfreeze;
|
|
|
|
}
|
|
|
|
|
2017-08-24 10:03:43 +03:00
|
|
|
blk_queue_logical_block_size(lo->lo_queue, arg);
|
2017-09-06 00:24:47 +03:00
|
|
|
blk_queue_physical_block_size(lo->lo_queue, arg);
|
|
|
|
blk_queue_io_min(lo->lo_queue, arg);
|
2017-08-24 10:03:43 +03:00
|
|
|
loop_update_dio(lo);
|
2019-01-10 06:17:14 +03:00
|
|
|
out_unfreeze:
|
2017-08-24 10:03:43 +03:00
|
|
|
blk_mq_unfreeze_queue(lo->lo_queue);
|
|
|
|
|
2019-01-10 06:17:14 +03:00
|
|
|
return err;
|
2017-08-24 10:03:43 +03:00
|
|
|
}
|
|
|
|
|
2018-11-08 16:01:05 +03:00
|
|
|
static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
|
|
|
|
unsigned long arg)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
err = mutex_lock_killable(&lo->lo_mutex);
|
2018-03-27 07:39:12 +03:00
|
|
|
if (err)
|
2018-11-08 16:01:05 +03:00
|
|
|
return err;
|
|
|
|
switch (cmd) {
|
|
|
|
case LOOP_SET_CAPACITY:
|
|
|
|
err = loop_set_capacity(lo);
|
|
|
|
break;
|
|
|
|
case LOOP_SET_DIRECT_IO:
|
|
|
|
err = loop_set_dio(lo, arg);
|
|
|
|
break;
|
|
|
|
case LOOP_SET_BLOCK_SIZE:
|
|
|
|
err = loop_set_block_size(lo, arg);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
|
|
|
|
}
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-11-08 16:01:05 +03:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int lo_ioctl(struct block_device *bdev, fmode_t mode,
|
|
|
|
unsigned int cmd, unsigned long arg)
|
|
|
|
{
|
|
|
|
struct loop_device *lo = bdev->bd_disk->private_data;
|
2020-05-13 16:38:43 +03:00
|
|
|
void __user *argp = (void __user *) arg;
|
2018-11-08 16:01:05 +03:00
|
|
|
int err;
|
2018-03-27 07:39:12 +03:00
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
switch (cmd) {
|
2020-05-13 16:38:45 +03:00
|
|
|
case LOOP_SET_FD: {
|
|
|
|
/*
|
|
|
|
* Legacy case - pass in a zeroed out struct loop_config with
|
|
|
|
* only the file descriptor set , which corresponds with the
|
|
|
|
* default parameters we'd have used otherwise.
|
|
|
|
*/
|
|
|
|
struct loop_config config;
|
|
|
|
|
|
|
|
memset(&config, 0, sizeof(config));
|
|
|
|
config.fd = arg;
|
|
|
|
|
|
|
|
return loop_configure(lo, mode, bdev, &config);
|
|
|
|
}
|
|
|
|
case LOOP_CONFIGURE: {
|
|
|
|
struct loop_config config;
|
|
|
|
|
|
|
|
if (copy_from_user(&config, argp, sizeof(config)))
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
return loop_configure(lo, mode, bdev, &config);
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
case LOOP_CHANGE_FD:
|
2018-11-08 16:01:11 +03:00
|
|
|
return loop_change_fd(lo, bdev, arg);
|
2005-04-17 02:20:36 +04:00
|
|
|
case LOOP_CLR_FD:
|
2018-11-08 16:01:07 +03:00
|
|
|
return loop_clr_fd(lo);
|
2005-04-17 02:20:36 +04:00
|
|
|
case LOOP_SET_STATUS:
|
2011-11-16 12:21:49 +04:00
|
|
|
err = -EPERM;
|
2018-11-08 16:01:05 +03:00
|
|
|
if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
|
2020-05-13 16:38:43 +03:00
|
|
|
err = loop_set_status_old(lo, argp);
|
2018-11-08 16:01:05 +03:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
break;
|
|
|
|
case LOOP_GET_STATUS:
|
2020-05-13 16:38:43 +03:00
|
|
|
return loop_get_status_old(lo, argp);
|
2005-04-17 02:20:36 +04:00
|
|
|
case LOOP_SET_STATUS64:
|
2011-11-16 12:21:49 +04:00
|
|
|
err = -EPERM;
|
2018-11-08 16:01:05 +03:00
|
|
|
if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
|
2020-05-13 16:38:43 +03:00
|
|
|
err = loop_set_status64(lo, argp);
|
2018-11-08 16:01:05 +03:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
break;
|
|
|
|
case LOOP_GET_STATUS64:
|
2020-05-13 16:38:43 +03:00
|
|
|
return loop_get_status64(lo, argp);
|
2018-11-08 16:01:05 +03:00
|
|
|
case LOOP_SET_CAPACITY:
|
2015-08-17 05:31:50 +03:00
|
|
|
case LOOP_SET_DIRECT_IO:
|
2017-08-24 10:03:43 +03:00
|
|
|
case LOOP_SET_BLOCK_SIZE:
|
2018-11-08 16:01:05 +03:00
|
|
|
if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
|
|
|
|
return -EPERM;
|
2020-08-24 01:36:59 +03:00
|
|
|
fallthrough;
|
2005-04-17 02:20:36 +04:00
|
|
|
default:
|
2018-11-08 16:01:05 +03:00
|
|
|
err = lo_simple_ioctl(lo, cmd, arg);
|
|
|
|
break;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
loop: fix circular locking in loop_clr_fd()
With CONFIG_PROVE_LOCKING enabled
$ losetup /dev/loop0 file
$ losetup -o 32256 /dev/loop1 /dev/loop0
$ losetup -d /dev/loop1
$ losetup -d /dev/loop0
triggers a [ INFO: possible circular locking dependency detected ]
I think this warning is a false positive.
Open/close on a loop device acquires bd_mutex of the device before
acquiring lo_ctl_mutex of the same device. For ioctl(LOOP_CLR_FD) after
acquiring lo_ctl_mutex, fput on the backing_file might acquire the bd_mutex of
a device, if backing file is a device and this is the last reference to the
file being dropped . But it is guaranteed that it is impossible to have a
circular list of backing devices.(say loop2->loop1->loop0->loop2 is not
possible), which guarantees that this can never deadlock.
So this warning should be suppressed. It is very difficult to annotate lockdep
not to warn here in the correct way. A simple way to silence lockdep could be
to mark the lo_ctl_mutex in ioctl to be a sub class, but this might mask some
other real bugs.
@@ -1164,7 +1164,7 @@ static int lo_ioctl(struct block_device *bdev, fmode_t mode,
struct loop_device *lo = bdev->bd_disk->private_data;
int err;
- mutex_lock(&lo->lo_ctl_mutex);
+ mutex_lock_nested(&lo->lo_ctl_mutex, 1);
switch (cmd) {
case LOOP_SET_FD:
err = loop_set_fd(lo, mode, bdev, arg);
Or actually marking the bd_mutex after lo_ctl_mutex as a sub class could be
a better solution.
Luckily it is easy to avoid calling fput on backing file with lo_ctl_mutex
held, so no lockdep annotation is required.
If you do not like the special handling of the lo_ctl_mutex just for the
LOOP_CLR_FD ioctl in lo_ioctl(), the mutex handling could be moved inside
each of the individual ioctl handlers and I could send you another patch.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2009-03-24 14:33:41 +03:00
|
|
|
|
2005-04-17 02:20:36 +04:00
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2006-08-29 22:06:14 +04:00
|
|
|
#ifdef CONFIG_COMPAT
|
|
|
|
struct compat_loop_info {
|
|
|
|
compat_int_t lo_number; /* ioctl r/o */
|
|
|
|
compat_dev_t lo_device; /* ioctl r/o */
|
|
|
|
compat_ulong_t lo_inode; /* ioctl r/o */
|
|
|
|
compat_dev_t lo_rdevice; /* ioctl r/o */
|
|
|
|
compat_int_t lo_offset;
|
|
|
|
compat_int_t lo_encrypt_type;
|
|
|
|
compat_int_t lo_encrypt_key_size; /* ioctl w/o */
|
|
|
|
compat_int_t lo_flags; /* ioctl r/o */
|
|
|
|
char lo_name[LO_NAME_SIZE];
|
|
|
|
unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
|
|
|
|
compat_ulong_t lo_init[2];
|
|
|
|
char reserved[4];
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Transfer 32-bit compatibility structure in userspace to 64-bit loop info
|
|
|
|
* - noinlined to reduce stack space usage in main part of driver
|
|
|
|
*/
|
|
|
|
static noinline int
|
2006-10-11 01:48:27 +04:00
|
|
|
loop_info64_from_compat(const struct compat_loop_info __user *arg,
|
2006-08-29 22:06:14 +04:00
|
|
|
struct loop_info64 *info64)
|
|
|
|
{
|
|
|
|
struct compat_loop_info info;
|
|
|
|
|
|
|
|
if (copy_from_user(&info, arg, sizeof(info)))
|
|
|
|
return -EFAULT;
|
|
|
|
|
|
|
|
memset(info64, 0, sizeof(*info64));
|
|
|
|
info64->lo_number = info.lo_number;
|
|
|
|
info64->lo_device = info.lo_device;
|
|
|
|
info64->lo_inode = info.lo_inode;
|
|
|
|
info64->lo_rdevice = info.lo_rdevice;
|
|
|
|
info64->lo_offset = info.lo_offset;
|
|
|
|
info64->lo_sizelimit = 0;
|
|
|
|
info64->lo_encrypt_type = info.lo_encrypt_type;
|
|
|
|
info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
|
|
|
|
info64->lo_flags = info.lo_flags;
|
|
|
|
info64->lo_init[0] = info.lo_init[0];
|
|
|
|
info64->lo_init[1] = info.lo_init[1];
|
|
|
|
if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
|
|
|
memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
|
|
|
|
else
|
|
|
|
memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
|
|
|
|
memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Transfer 64-bit loop info to 32-bit compatibility structure in userspace
|
|
|
|
* - noinlined to reduce stack space usage in main part of driver
|
|
|
|
*/
|
|
|
|
static noinline int
|
|
|
|
loop_info64_to_compat(const struct loop_info64 *info64,
|
|
|
|
struct compat_loop_info __user *arg)
|
|
|
|
{
|
|
|
|
struct compat_loop_info info;
|
|
|
|
|
|
|
|
memset(&info, 0, sizeof(info));
|
|
|
|
info.lo_number = info64->lo_number;
|
|
|
|
info.lo_device = info64->lo_device;
|
|
|
|
info.lo_inode = info64->lo_inode;
|
|
|
|
info.lo_rdevice = info64->lo_rdevice;
|
|
|
|
info.lo_offset = info64->lo_offset;
|
|
|
|
info.lo_encrypt_type = info64->lo_encrypt_type;
|
|
|
|
info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
|
|
|
|
info.lo_flags = info64->lo_flags;
|
|
|
|
info.lo_init[0] = info64->lo_init[0];
|
|
|
|
info.lo_init[1] = info64->lo_init[1];
|
|
|
|
if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
|
|
|
|
memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
|
|
|
|
else
|
|
|
|
memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
|
|
|
|
memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
|
|
|
|
|
|
|
|
/* error in case values were truncated */
|
|
|
|
if (info.lo_device != info64->lo_device ||
|
|
|
|
info.lo_rdevice != info64->lo_rdevice ||
|
|
|
|
info.lo_inode != info64->lo_inode ||
|
|
|
|
info.lo_offset != info64->lo_offset ||
|
|
|
|
info.lo_init[0] != info64->lo_init[0] ||
|
|
|
|
info.lo_init[1] != info64->lo_init[1])
|
|
|
|
return -EOVERFLOW;
|
|
|
|
|
|
|
|
if (copy_to_user(arg, &info, sizeof(info)))
|
|
|
|
return -EFAULT;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_set_status_compat(struct loop_device *lo,
|
|
|
|
const struct compat_loop_info __user *arg)
|
|
|
|
{
|
|
|
|
struct loop_info64 info64;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
ret = loop_info64_from_compat(arg, &info64);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
return loop_set_status(lo, &info64);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
loop_get_status_compat(struct loop_device *lo,
|
|
|
|
struct compat_loop_info __user *arg)
|
|
|
|
{
|
|
|
|
struct loop_info64 info64;
|
2018-04-06 19:57:03 +03:00
|
|
|
int err;
|
2006-08-29 22:06:14 +04:00
|
|
|
|
2018-11-08 16:01:08 +03:00
|
|
|
if (!arg)
|
2018-04-06 19:57:03 +03:00
|
|
|
return -EINVAL;
|
|
|
|
err = loop_get_status(lo, &info64);
|
2006-08-29 22:06:14 +04:00
|
|
|
if (!err)
|
|
|
|
err = loop_info64_to_compat(&info64, arg);
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2008-03-02 17:29:48 +03:00
|
|
|
static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
|
|
|
|
unsigned int cmd, unsigned long arg)
|
2006-08-29 22:06:14 +04:00
|
|
|
{
|
2008-03-02 17:29:48 +03:00
|
|
|
struct loop_device *lo = bdev->bd_disk->private_data;
|
2006-08-29 22:06:14 +04:00
|
|
|
int err;
|
|
|
|
|
|
|
|
switch(cmd) {
|
|
|
|
case LOOP_SET_STATUS:
|
2018-11-08 16:01:09 +03:00
|
|
|
err = loop_set_status_compat(lo,
|
|
|
|
(const struct compat_loop_info __user *)arg);
|
2006-08-29 22:06:14 +04:00
|
|
|
break;
|
|
|
|
case LOOP_GET_STATUS:
|
2018-11-08 16:01:08 +03:00
|
|
|
err = loop_get_status_compat(lo,
|
|
|
|
(struct compat_loop_info __user *)arg);
|
2006-08-29 22:06:14 +04:00
|
|
|
break;
|
loop: add ioctl to resize a loop device
Add the ability to 'resize' the loop device on the fly.
One practical application is a loop file with XFS filesystem, already
mounted: You can easily enlarge the file (append some bytes) and then call
ioctl(fd, LOOP_SET_CAPACITY, new); The loop driver will learn about the
new size and you can use xfs_growfs later on, which will allow you to use
full capacity of the loop file without the need to unmount.
Test app:
#include <linux/fs.h>
#include <linux/loop.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <assert.h>
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#define _GNU_SOURCE
#include <getopt.h>
char *me;
void usage(FILE *f)
{
fprintf(f, "%s [options] loop_dev [backend_file]\n"
"-s, --set new_size_in_bytes\n"
"\twhen backend_file is given, "
"it will be expanded too while keeping the original contents\n",
me);
}
struct option opts[] = {
{
.name = "set",
.has_arg = 1,
.flag = NULL,
.val = 's'
},
{
.name = "help",
.has_arg = 0,
.flag = NULL,
.val = 'h'
}
};
void err_size(char *name, __u64 old)
{
fprintf(stderr, "size must be larger than current %s (%llu)\n",
name, old);
}
int main(int argc, char *argv[])
{
int fd, err, c, i, bfd;
ssize_t ssz;
size_t sz;
__u64 old, new, append;
char a[BUFSIZ];
struct stat st;
FILE *out;
char *backend, *dev;
err = EINVAL;
out = stderr;
me = argv[0];
new = 0;
while ((c = getopt_long(argc, argv, "s:h", opts, &i)) != -1) {
switch (c) {
case 's':
errno = 0;
new = strtoull(optarg, NULL, 0);
if (errno) {
err = errno;
perror(argv[i]);
goto out;
}
break;
case 'h':
err = 0;
out = stdout;
goto err;
default:
perror(argv[i]);
goto err;
}
}
if (optind < argc)
dev = argv[optind++];
else
goto err;
fd = open(dev, O_RDONLY);
if (fd < 0) {
err = errno;
perror(dev);
goto out;
}
err = ioctl(fd, BLKGETSIZE64, &old);
if (err) {
err = errno;
perror("ioctl BLKGETSIZE64");
goto out;
}
if (!new) {
printf("%llu\n", old);
goto out;
}
if (new < old) {
err = EINVAL;
err_size(dev, old);
goto out;
}
if (optind < argc) {
backend = argv[optind++];
bfd = open(backend, O_WRONLY|O_APPEND);
if (bfd < 0) {
err = errno;
perror(backend);
goto out;
}
err = fstat(bfd, &st);
if (err) {
err = errno;
perror(backend);
goto out;
}
if (new < st.st_size) {
err = EINVAL;
err_size(backend, st.st_size);
goto out;
}
append = new - st.st_size;
sz = sizeof(a);
while (append > 0) {
if (append < sz)
sz = append;
ssz = write(bfd, a, sz);
if (ssz != sz) {
err = errno;
perror(backend);
goto out;
}
append -= sz;
}
err = fsync(bfd);
if (err) {
err = errno;
perror(backend);
goto out;
}
}
err = ioctl(fd, LOOP_SET_CAPACITY, new);
if (err) {
err = errno;
perror("ioctl LOOP_SET_CAPACITY");
}
goto out;
err:
usage(out);
out:
return err;
}
Signed-off-by: J. R. Okajima <hooanon05@yahoo.co.jp>
Signed-off-by: Tomas Matejicek <tomas@slax.org>
Cc: <util-linux-ng@vger.kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: <linux-api@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-04-01 02:23:43 +04:00
|
|
|
case LOOP_SET_CAPACITY:
|
2006-08-29 22:06:14 +04:00
|
|
|
case LOOP_CLR_FD:
|
|
|
|
case LOOP_GET_STATUS64:
|
|
|
|
case LOOP_SET_STATUS64:
|
2020-05-13 16:38:45 +03:00
|
|
|
case LOOP_CONFIGURE:
|
2006-08-29 22:06:14 +04:00
|
|
|
arg = (unsigned long) compat_ptr(arg);
|
2020-08-24 01:36:59 +03:00
|
|
|
fallthrough;
|
2006-08-29 22:06:14 +04:00
|
|
|
case LOOP_SET_FD:
|
|
|
|
case LOOP_CHANGE_FD:
|
2018-07-03 02:03:46 +03:00
|
|
|
case LOOP_SET_BLOCK_SIZE:
|
2019-08-07 03:48:28 +03:00
|
|
|
case LOOP_SET_DIRECT_IO:
|
2008-03-02 17:29:48 +03:00
|
|
|
err = lo_ioctl(bdev, mode, cmd, arg);
|
2006-08-29 22:06:14 +04:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
err = -ENOIOCTLCMD;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2008-03-02 17:29:48 +03:00
|
|
|
static int lo_open(struct block_device *bdev, fmode_t mode)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2021-06-05 17:09:50 +03:00
|
|
|
struct loop_device *lo = bdev->bd_disk->private_data;
|
2018-11-08 16:01:04 +03:00
|
|
|
int err;
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
err = mutex_lock_killable(&lo->lo_mutex);
|
|
|
|
if (err)
|
|
|
|
return err;
|
2021-06-05 17:09:50 +03:00
|
|
|
if (lo->lo_state == Lo_deleting)
|
|
|
|
err = -ENXIO;
|
|
|
|
else
|
|
|
|
atomic_inc(&lo->lo_refcnt);
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2021-06-05 17:09:50 +03:00
|
|
|
return err;
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
2018-11-08 16:01:03 +03:00
|
|
|
static void lo_release(struct gendisk *disk, fmode_t mode)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2021-01-26 17:46:30 +03:00
|
|
|
struct loop_device *lo = disk->private_data;
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_lock(&lo->lo_mutex);
|
2015-05-06 07:26:23 +03:00
|
|
|
if (atomic_dec_return(&lo->lo_refcnt))
|
2018-11-08 16:01:04 +03:00
|
|
|
goto out_unlock;
|
2008-12-12 16:48:27 +03:00
|
|
|
|
|
|
|
if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
|
2018-11-08 16:01:06 +03:00
|
|
|
if (lo->lo_state != Lo_bound)
|
|
|
|
goto out_unlock;
|
|
|
|
lo->lo_state = Lo_rundown;
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2008-12-12 16:48:27 +03:00
|
|
|
/*
|
|
|
|
* In autoclear mode, stop the loop thread
|
|
|
|
* and remove configuration after last close.
|
|
|
|
*/
|
2018-11-08 16:01:14 +03:00
|
|
|
__loop_clr_fd(lo, true);
|
2018-11-08 16:01:07 +03:00
|
|
|
return;
|
2017-08-24 10:03:44 +03:00
|
|
|
} else if (lo->lo_state == Lo_bound) {
|
2008-12-12 16:48:27 +03:00
|
|
|
/*
|
|
|
|
* Otherwise keep thread (if running) and config,
|
|
|
|
* but flush possible ongoing bios in thread.
|
|
|
|
*/
|
2017-08-24 10:03:44 +03:00
|
|
|
blk_mq_freeze_queue(lo->lo_queue);
|
|
|
|
blk_mq_unfreeze_queue(lo->lo_queue);
|
2008-12-12 16:48:27 +03:00
|
|
|
}
|
2008-02-06 12:36:27 +03:00
|
|
|
|
2018-11-08 16:01:04 +03:00
|
|
|
out_unlock:
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_unlock(&lo->lo_mutex);
|
2018-01-06 03:26:00 +03:00
|
|
|
}
|
|
|
|
|
2009-09-22 04:01:13 +04:00
|
|
|
static const struct block_device_operations lo_fops = {
|
2005-04-17 02:20:36 +04:00
|
|
|
.owner = THIS_MODULE,
|
2008-03-02 17:29:48 +03:00
|
|
|
.open = lo_open,
|
|
|
|
.release = lo_release,
|
|
|
|
.ioctl = lo_ioctl,
|
2006-08-29 22:06:14 +04:00
|
|
|
#ifdef CONFIG_COMPAT
|
2008-03-02 17:29:48 +03:00
|
|
|
.compat_ioctl = lo_compat_ioctl,
|
2006-08-29 22:06:14 +04:00
|
|
|
#endif
|
2005-04-17 02:20:36 +04:00
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* And now the modules code and kernel interface.
|
|
|
|
*/
|
2007-05-08 11:28:20 +04:00
|
|
|
static int max_loop;
|
2018-05-24 22:38:59 +03:00
|
|
|
module_param(max_loop, int, 0444);
|
2007-06-09 00:46:44 +04:00
|
|
|
MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
|
2018-05-24 22:38:59 +03:00
|
|
|
module_param(max_part, int, 0444);
|
loop: manage partitions in disk image
This patch allows to use loop device with partitionned disk image.
Original behavior of loop is not modified.
A new parameter is introduced to define how many partition we want to be
able to manage per loop device. This parameter is "max_part".
For instance, to manage 63 partitions / loop device, we will do:
# modprobe loop max_part=63
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
And to attach a raw partitionned disk image, the original losetup is used:
# losetup -f etch.img
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 1 2008-03-05 14:57 /dev/loop0p1
brw-rw---- 1 root disk 7, 2 2008-03-05 14:57 /dev/loop0p2
brw-rw---- 1 root disk 7, 5 2008-03-05 14:57 /dev/loop0p5
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
# mount /dev/loop0p1 /mnt
# ls /mnt
bench cdrom home lib mnt root srv usr
bin dev initrd lost+found opt sbin sys var
boot etc initrd.img media proc selinux tmp vmlinuz
# umount /mnt
# losetup -d /dev/loop0
Of course, the same behavior can be done using kpartx on a loop device,
but modifying loop avoids to stack several layers of block device (loop +
device mapper), this is a very light modification (40% of modifications
are to manage the new parameter).
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2008-03-26 14:11:53 +03:00
|
|
|
MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
|
2005-04-17 02:20:36 +04:00
|
|
|
MODULE_LICENSE("GPL");
|
|
|
|
MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
|
|
|
|
|
|
|
|
int loop_register_transfer(struct loop_func_table *funcs)
|
|
|
|
{
|
|
|
|
unsigned int n = funcs->number;
|
|
|
|
|
|
|
|
if (n >= MAX_LO_CRYPT || xfer_funcs[n])
|
|
|
|
return -EINVAL;
|
|
|
|
xfer_funcs[n] = funcs;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int loop_unregister_transfer(int number)
|
|
|
|
{
|
|
|
|
unsigned int n = number;
|
|
|
|
struct loop_func_table *xfer;
|
|
|
|
|
|
|
|
if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
|
|
|
|
return -EINVAL;
|
2021-09-02 03:07:35 +03:00
|
|
|
/*
|
|
|
|
* This function is called from only cleanup_cryptoloop().
|
|
|
|
* Given that each loop device that has a transfer enabled holds a
|
|
|
|
* reference to the module implementing it we should never get here
|
|
|
|
* with a transfer that is set (unless forced module unloading is
|
|
|
|
* requested). Thus, check module's refcount and warn if this is
|
|
|
|
* not a clean unloading.
|
|
|
|
*/
|
|
|
|
#ifdef CONFIG_MODULE_UNLOAD
|
|
|
|
if (xfer->owner && module_refcount(xfer->owner) != -1)
|
|
|
|
pr_err("Danger! Unregistering an in use transfer function.\n");
|
|
|
|
#endif
|
2005-04-17 02:20:36 +04:00
|
|
|
|
|
|
|
xfer_funcs[n] = NULL;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
EXPORT_SYMBOL(loop_register_transfer);
|
|
|
|
EXPORT_SYMBOL(loop_unregister_transfer);
|
|
|
|
|
2017-06-03 10:38:05 +03:00
|
|
|
static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
const struct blk_mq_queue_data *bd)
|
|
|
|
{
|
2018-04-14 01:24:29 +03:00
|
|
|
struct request *rq = bd->rq;
|
|
|
|
struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
|
|
|
|
struct loop_device *lo = rq->q->queuedata;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2018-04-14 01:24:29 +03:00
|
|
|
blk_mq_start_request(rq);
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2015-05-05 14:49:54 +03:00
|
|
|
if (lo->lo_state != Lo_bound)
|
2017-06-03 10:38:05 +03:00
|
|
|
return BLK_STS_IOERR;
|
2015-05-05 14:49:54 +03:00
|
|
|
|
2018-04-14 01:24:29 +03:00
|
|
|
switch (req_op(rq)) {
|
2016-08-04 17:10:00 +03:00
|
|
|
case REQ_OP_FLUSH:
|
|
|
|
case REQ_OP_DISCARD:
|
2017-04-05 20:21:15 +03:00
|
|
|
case REQ_OP_WRITE_ZEROES:
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
cmd->use_aio = false;
|
2016-08-04 17:10:00 +03:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
cmd->use_aio = lo->use_dio;
|
|
|
|
break;
|
|
|
|
}
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
|
2017-09-25 22:07:22 +03:00
|
|
|
/* always use the first bio's css */
|
2021-06-29 05:38:21 +03:00
|
|
|
cmd->blkcg_css = NULL;
|
|
|
|
cmd->memcg_css = NULL;
|
2017-09-26 21:02:12 +03:00
|
|
|
#ifdef CONFIG_BLK_CGROUP
|
2021-06-29 05:38:21 +03:00
|
|
|
if (rq->bio && rq->bio->bi_blkg) {
|
|
|
|
cmd->blkcg_css = &bio_blkcg(rq->bio)->css;
|
|
|
|
#ifdef CONFIG_MEMCG
|
|
|
|
cmd->memcg_css =
|
|
|
|
cgroup_get_e_css(cmd->blkcg_css->cgroup,
|
|
|
|
&memory_cgrp_subsys);
|
|
|
|
#endif
|
|
|
|
}
|
2017-09-25 22:07:22 +03:00
|
|
|
#endif
|
2021-06-29 05:38:15 +03:00
|
|
|
loop_queue_work(lo, cmd);
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2017-06-03 10:38:05 +03:00
|
|
|
return BLK_STS_OK;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
static void loop_handle_cmd(struct loop_cmd *cmd)
|
|
|
|
{
|
2018-04-14 01:24:29 +03:00
|
|
|
struct request *rq = blk_mq_rq_from_pdu(cmd);
|
|
|
|
const bool write = op_is_write(req_op(rq));
|
|
|
|
struct loop_device *lo = rq->q->queuedata;
|
2015-09-27 22:01:50 +03:00
|
|
|
int ret = 0;
|
2021-06-29 05:38:21 +03:00
|
|
|
struct mem_cgroup *old_memcg = NULL;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2015-09-27 22:01:50 +03:00
|
|
|
if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
|
|
|
|
ret = -EIO;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
goto failed;
|
2015-09-27 22:01:50 +03:00
|
|
|
}
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2021-06-29 05:38:21 +03:00
|
|
|
if (cmd->blkcg_css)
|
|
|
|
kthread_associate_blkcg(cmd->blkcg_css);
|
|
|
|
if (cmd->memcg_css)
|
|
|
|
old_memcg = set_active_memcg(
|
|
|
|
mem_cgroup_from_css(cmd->memcg_css));
|
|
|
|
|
2018-04-14 01:24:29 +03:00
|
|
|
ret = do_req_filebacked(lo, rq);
|
2021-06-29 05:38:21 +03:00
|
|
|
|
|
|
|
if (cmd->blkcg_css)
|
|
|
|
kthread_associate_blkcg(NULL);
|
|
|
|
|
|
|
|
if (cmd->memcg_css) {
|
|
|
|
set_active_memcg(old_memcg);
|
|
|
|
css_put(cmd->memcg_css);
|
|
|
|
}
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
failed:
|
block: loop: support DIO & AIO
There are at least 3 advantages to use direct I/O and AIO on
read/write loop's backing file:
1) double cache can be avoided, then memory usage gets
decreased a lot
2) not like user space direct I/O, there isn't cost of
pinning pages
3) avoid context switch for obtaining good throughput
- in buffered file read, random I/O top throughput is often obtained
only if they are submitted concurrently from lots of tasks; but for
sequential I/O, most of times they can be hit from page cache, so
concurrent submissions often introduce unnecessary context switch
and can't improve throughput much. There was such discussion[1]
to use non-blocking I/O to improve the problem for application.
- with direct I/O and AIO, concurrent submissions can be
avoided and random read throughput can't be affected meantime
xfstests(-g auto, ext4) is basically passed when running with
direct I/O(aio), one exception is generic/232, but it failed in
loop buffered I/O(4.2-rc6-next-20150814) too.
Follows the fio test result for performance purpose:
4 jobs fio test inside ext4 file system over loop block
1) How to run
- KVM: 4 VCPUs, 2G RAM
- linux kernel: 4.2-rc6-next-20150814(base) with the patchset
- the loop block is over one image on SSD.
- linux psync, 4 jobs, size 1500M, ext4 over loop block
- test result: IOPS from fio output
2) Throughput(IOPS) becomes a bit better with direct I/O(aio)
-------------------------------------------------------------
test cases |randread |read |randwrite |write |
-------------------------------------------------------------
base |8015 |113811 |67442 |106978
-------------------------------------------------------------
base+loop aio |8136 |125040 |67811 |111376
-------------------------------------------------------------
- somehow, it should be caused by more page cache avaiable for
application or one extra page copy is avoided in case of direct I/O
3) context switch
- context switch decreased by ~50% with loop direct I/O(aio)
compared with loop buffered I/O(4.2-rc6-next-20150814)
4) memory usage from /proc/meminfo
-------------------------------------------------------------
| Buffers | Cached
-------------------------------------------------------------
base | > 760MB | ~950MB
-------------------------------------------------------------
base+loop direct I/O(aio) | < 5MB | ~1.6GB
-------------------------------------------------------------
- so there are much more page caches available for application with
direct I/O
[1] https://lwn.net/Articles/612483/
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
2015-08-17 05:31:51 +03:00
|
|
|
/* complete non-aio request */
|
2017-04-20 17:03:02 +03:00
|
|
|
if (!cmd->use_aio || ret) {
|
2020-04-03 17:43:03 +03:00
|
|
|
if (ret == -EOPNOTSUPP)
|
|
|
|
cmd->ret = ret;
|
|
|
|
else
|
|
|
|
cmd->ret = ret ? -EIO : 0;
|
2020-06-11 09:44:47 +03:00
|
|
|
if (likely(!blk_should_fake_timeout(rq->q)))
|
|
|
|
blk_mq_complete_request(rq);
|
2017-04-20 17:03:02 +03:00
|
|
|
}
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
}
|
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
static void loop_set_timer(struct loop_device *lo)
|
|
|
|
{
|
|
|
|
timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void loop_process_work(struct loop_worker *worker,
|
|
|
|
struct list_head *cmd_list, struct loop_device *lo)
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
{
|
2021-06-29 05:38:15 +03:00
|
|
|
int orig_flags = current->flags;
|
|
|
|
struct loop_cmd *cmd;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
|
|
|
|
spin_lock_irq(&lo->lo_work_lock);
|
|
|
|
while (!list_empty(cmd_list)) {
|
|
|
|
cmd = container_of(
|
|
|
|
cmd_list->next, struct loop_cmd, list_entry);
|
|
|
|
list_del(cmd_list->next);
|
|
|
|
spin_unlock_irq(&lo->lo_work_lock);
|
|
|
|
|
|
|
|
loop_handle_cmd(cmd);
|
|
|
|
cond_resched();
|
|
|
|
|
|
|
|
spin_lock_irq(&lo->lo_work_lock);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We only add to the idle list if there are no pending cmds
|
|
|
|
* *and* the worker will not run again which ensures that it
|
|
|
|
* is safe to free any worker on the idle list
|
|
|
|
*/
|
|
|
|
if (worker && !work_pending(&worker->work)) {
|
|
|
|
worker->last_ran_at = jiffies;
|
|
|
|
list_add_tail(&worker->idle_list, &lo->idle_worker_list);
|
|
|
|
loop_set_timer(lo);
|
|
|
|
}
|
|
|
|
spin_unlock_irq(&lo->lo_work_lock);
|
|
|
|
current->flags = orig_flags;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
}
|
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
static void loop_workfn(struct work_struct *work)
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
{
|
2021-06-29 05:38:15 +03:00
|
|
|
struct loop_worker *worker =
|
|
|
|
container_of(work, struct loop_worker, work);
|
|
|
|
loop_process_work(worker, &worker->cmd_list, worker->lo);
|
|
|
|
}
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
|
2021-06-29 05:38:15 +03:00
|
|
|
static void loop_rootcg_workfn(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct loop_device *lo =
|
|
|
|
container_of(work, struct loop_device, rootcg_work);
|
|
|
|
loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void loop_free_idle_workers(struct timer_list *timer)
|
|
|
|
{
|
|
|
|
struct loop_device *lo = container_of(timer, struct loop_device, timer);
|
|
|
|
struct loop_worker *pos, *worker;
|
|
|
|
|
|
|
|
spin_lock_irq(&lo->lo_work_lock);
|
|
|
|
list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
|
|
|
|
idle_list) {
|
|
|
|
if (time_is_after_jiffies(worker->last_ran_at +
|
|
|
|
LOOP_IDLE_WORKER_TIMEOUT))
|
|
|
|
break;
|
|
|
|
list_del(&worker->idle_list);
|
|
|
|
rb_erase(&worker->rb_node, &lo->worker_tree);
|
2021-06-29 05:38:21 +03:00
|
|
|
css_put(worker->blkcg_css);
|
2021-06-29 05:38:15 +03:00
|
|
|
kfree(worker);
|
|
|
|
}
|
|
|
|
if (!list_empty(&lo->idle_worker_list))
|
|
|
|
loop_set_timer(lo);
|
|
|
|
spin_unlock_irq(&lo->lo_work_lock);
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
}
|
|
|
|
|
2017-03-30 23:39:16 +03:00
|
|
|
static const struct blk_mq_ops loop_mq_ops = {
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
.queue_rq = loop_queue_rq,
|
2017-04-20 17:03:02 +03:00
|
|
|
.complete = lo_complete_rq,
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
};
|
|
|
|
|
2021-06-23 17:59:02 +03:00
|
|
|
static int loop_add(int i)
|
2007-05-08 11:28:20 +04:00
|
|
|
{
|
|
|
|
struct loop_device *lo;
|
|
|
|
struct gendisk *disk;
|
2011-08-01 00:08:04 +04:00
|
|
|
int err;
|
2007-05-08 11:28:20 +04:00
|
|
|
|
2012-07-15 02:39:58 +04:00
|
|
|
err = -ENOMEM;
|
2007-05-08 11:28:20 +04:00
|
|
|
lo = kzalloc(sizeof(*lo), GFP_KERNEL);
|
2012-07-15 02:39:58 +04:00
|
|
|
if (!lo)
|
2007-05-08 11:28:20 +04:00
|
|
|
goto out;
|
2013-10-16 00:14:38 +04:00
|
|
|
lo->lo_state = Lo_unbound;
|
|
|
|
|
2021-06-23 17:59:05 +03:00
|
|
|
err = mutex_lock_killable(&loop_ctl_mutex);
|
|
|
|
if (err)
|
|
|
|
goto out_free_dev;
|
|
|
|
|
2013-02-28 05:03:58 +04:00
|
|
|
/* allocate id, if @id >= 0, we're requesting that specific id */
|
2011-08-01 00:08:04 +04:00
|
|
|
if (i >= 0) {
|
2013-02-28 05:03:58 +04:00
|
|
|
err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
|
|
|
|
if (err == -ENOSPC)
|
2011-08-01 00:08:04 +04:00
|
|
|
err = -EEXIST;
|
|
|
|
} else {
|
2013-02-28 05:03:58 +04:00
|
|
|
err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
|
2011-08-01 00:08:04 +04:00
|
|
|
}
|
2021-09-02 03:07:35 +03:00
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
2011-08-01 00:08:04 +04:00
|
|
|
if (err < 0)
|
2021-09-02 03:07:35 +03:00
|
|
|
goto out_free_dev;
|
2013-02-28 05:03:58 +04:00
|
|
|
i = err;
|
2007-05-08 11:28:20 +04:00
|
|
|
|
2013-03-22 18:59:19 +04:00
|
|
|
err = -ENOMEM;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
lo->tag_set.ops = &loop_mq_ops;
|
|
|
|
lo->tag_set.nr_hw_queues = 1;
|
|
|
|
lo->tag_set.queue_depth = 128;
|
|
|
|
lo->tag_set.numa_node = NUMA_NO_NODE;
|
|
|
|
lo->tag_set.cmd_size = sizeof(struct loop_cmd);
|
2021-08-05 20:42:00 +03:00
|
|
|
lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
|
|
|
|
BLK_MQ_F_NO_SCHED_BY_DEFAULT;
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
lo->tag_set.driver_data = lo;
|
|
|
|
|
|
|
|
err = blk_mq_alloc_tag_set(&lo->tag_set);
|
|
|
|
if (err)
|
2013-10-14 20:12:24 +04:00
|
|
|
goto out_free_idr;
|
2007-05-08 11:28:20 +04:00
|
|
|
|
2021-06-02 09:53:33 +03:00
|
|
|
disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
|
|
|
|
if (IS_ERR(disk)) {
|
|
|
|
err = PTR_ERR(disk);
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
goto out_cleanup_tags;
|
|
|
|
}
|
2021-06-02 09:53:33 +03:00
|
|
|
lo->lo_queue = lo->lo_disk->queue;
|
2013-10-16 00:14:38 +04:00
|
|
|
|
2017-09-01 08:09:45 +03:00
|
|
|
blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
|
2017-09-01 08:09:46 +03:00
|
|
|
|
2015-08-17 05:31:47 +03:00
|
|
|
/*
|
2017-09-01 08:09:46 +03:00
|
|
|
* By default, we do buffer IO, so it doesn't make sense to enable
|
|
|
|
* merge because the I/O submitted to backing file is handled page by
|
|
|
|
* page. For directio mode, merge does help to dispatch bigger request
|
|
|
|
* to underlayer disk. We will enable merge once directio is enabled.
|
2015-08-17 05:31:47 +03:00
|
|
|
*/
|
2018-03-08 04:10:10 +03:00
|
|
|
blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
|
2015-08-17 05:31:47 +03:00
|
|
|
|
loop: always allow userspace partitions and optionally support automatic scanning
Automatic partition scanning can be requested individually per loop
device during its setup by setting LO_FLAGS_PARTSCAN. By default, no
partition tables are scanned.
Userspace can now always add and remove partitions from all loop
devices, regardless if the in-kernel partition scanner is enabled or
not.
The needed partition minor numbers are allocated from the extended
minors space, the main loop device numbers will continue to match the
loop minors, regardless of the number of partitions used.
# grep . /sys/class/block/loop1/loop/*
/sys/block/loop1/loop/autoclear:0
/sys/block/loop1/loop/backing_file:/home/kay/data/stuff/part.img
/sys/block/loop1/loop/offset:0
/sys/block/loop1/loop/partscan:1
/sys/block/loop1/loop/sizelimit:0
# ls -l /dev/loop*
brw-rw---- 1 root disk 7, 0 Aug 14 20:22 /dev/loop0
brw-rw---- 1 root disk 7, 1 Aug 14 20:23 /dev/loop1
brw-rw---- 1 root disk 259, 0 Aug 14 20:23 /dev/loop1p1
brw-rw---- 1 root disk 259, 1 Aug 14 20:23 /dev/loop1p2
brw-rw---- 1 root disk 7, 99 Aug 14 20:23 /dev/loop99
brw-rw---- 1 root disk 259, 2 Aug 14 20:23 /dev/loop99p1
brw-rw---- 1 root disk 259, 3 Aug 14 20:23 /dev/loop99p2
crw------T 1 root root 10, 237 Aug 14 20:22 /dev/loop-control
Cc: Karel Zak <kzak@redhat.com>
Cc: Davidlohr Bueso <dave@gnu.org>
Acked-By: Tejun Heo <tj@kernel.org>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-23 22:12:04 +04:00
|
|
|
/*
|
|
|
|
* Disable partition scanning by default. The in-kernel partition
|
|
|
|
* scanning can be requested individually per-device during its
|
|
|
|
* setup. Userspace can always add and remove partitions from all
|
|
|
|
* devices. The needed partition minors are allocated from the
|
|
|
|
* extended minor space, the main loop device numbers will continue
|
|
|
|
* to match the loop minors, regardless of the number of partitions
|
|
|
|
* used.
|
|
|
|
*
|
|
|
|
* If max_part is given, partition scanning is globally enabled for
|
|
|
|
* all loop devices. The minors for the main loop devices will be
|
|
|
|
* multiples of max_part.
|
|
|
|
*
|
|
|
|
* Note: Global-for-all-devices, set-only-at-init, read-only module
|
|
|
|
* parameteters like 'max_loop' and 'max_part' make things needlessly
|
|
|
|
* complicated, are too static, inflexible and may surprise
|
|
|
|
* userspace tools. Parameters like this in general should be avoided.
|
|
|
|
*/
|
|
|
|
if (!part_shift)
|
|
|
|
disk->flags |= GENHD_FL_NO_PART_SCAN;
|
|
|
|
disk->flags |= GENHD_FL_EXT_DEVT;
|
2015-05-06 07:26:23 +03:00
|
|
|
atomic_set(&lo->lo_refcnt, 0);
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_init(&lo->lo_mutex);
|
2007-05-08 11:28:20 +04:00
|
|
|
lo->lo_number = i;
|
|
|
|
spin_lock_init(&lo->lo_lock);
|
2021-06-29 05:38:15 +03:00
|
|
|
spin_lock_init(&lo->lo_work_lock);
|
2007-05-08 11:28:20 +04:00
|
|
|
disk->major = LOOP_MAJOR;
|
loop: manage partitions in disk image
This patch allows to use loop device with partitionned disk image.
Original behavior of loop is not modified.
A new parameter is introduced to define how many partition we want to be
able to manage per loop device. This parameter is "max_part".
For instance, to manage 63 partitions / loop device, we will do:
# modprobe loop max_part=63
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
And to attach a raw partitionned disk image, the original losetup is used:
# losetup -f etch.img
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 1 2008-03-05 14:57 /dev/loop0p1
brw-rw---- 1 root disk 7, 2 2008-03-05 14:57 /dev/loop0p2
brw-rw---- 1 root disk 7, 5 2008-03-05 14:57 /dev/loop0p5
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
# mount /dev/loop0p1 /mnt
# ls /mnt
bench cdrom home lib mnt root srv usr
bin dev initrd lost+found opt sbin sys var
boot etc initrd.img media proc selinux tmp vmlinuz
# umount /mnt
# losetup -d /dev/loop0
Of course, the same behavior can be done using kpartx on a loop device,
but modifying loop avoids to stack several layers of block device (loop +
device mapper), this is a very light modification (40% of modifications
are to manage the new parameter).
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2008-03-26 14:11:53 +03:00
|
|
|
disk->first_minor = i << part_shift;
|
2021-06-02 09:53:33 +03:00
|
|
|
disk->minors = 1 << part_shift;
|
2007-05-08 11:28:20 +04:00
|
|
|
disk->fops = &lo_fops;
|
|
|
|
disk->private_data = lo;
|
|
|
|
disk->queue = lo->lo_queue;
|
2021-07-13 02:05:30 +03:00
|
|
|
disk->events = DISK_EVENT_MEDIA_CHANGE;
|
|
|
|
disk->event_flags = DISK_EVENT_FLAG_UEVENT;
|
2007-05-08 11:28:20 +04:00
|
|
|
sprintf(disk->disk_name, "loop%d", i);
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Make this loop device reachable from pathname. */
|
2011-08-01 00:08:04 +04:00
|
|
|
add_disk(disk);
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Show this loop device. */
|
|
|
|
mutex_lock(&loop_ctl_mutex);
|
|
|
|
lo->idr_visible = true;
|
2021-06-23 17:59:05 +03:00
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
|
|
|
return i;
|
2007-05-08 11:28:20 +04:00
|
|
|
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
out_cleanup_tags:
|
|
|
|
blk_mq_free_tag_set(&lo->tag_set);
|
2013-10-14 20:12:24 +04:00
|
|
|
out_free_idr:
|
2021-09-02 03:07:35 +03:00
|
|
|
mutex_lock(&loop_ctl_mutex);
|
2013-10-14 20:12:24 +04:00
|
|
|
idr_remove(&loop_index_idr, i);
|
2021-06-23 17:59:05 +03:00
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
2007-05-08 11:28:20 +04:00
|
|
|
out_free_dev:
|
|
|
|
kfree(lo);
|
|
|
|
out:
|
2011-08-01 00:08:04 +04:00
|
|
|
return err;
|
2007-05-08 11:28:20 +04:00
|
|
|
}
|
|
|
|
|
2011-08-01 00:08:04 +04:00
|
|
|
static void loop_remove(struct loop_device *lo)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Make this loop device unreachable from pathname. */
|
2015-04-27 07:12:22 +03:00
|
|
|
del_gendisk(lo->lo_disk);
|
2021-06-02 09:53:33 +03:00
|
|
|
blk_cleanup_disk(lo->lo_disk);
|
block: loop: improve performance via blk-mq
The conversion is a bit straightforward, and use work queue to
dispatch requests of loop block, and one big change is that requests
is submitted to backend file/device concurrently with work queue,
so throughput may get improved much. Given write requests over same
file are often run exclusively, so don't handle them concurrently for
avoiding extra context switch cost, possible lock contention and work
schedule cost. Also with blk-mq, there is opportunity to get loop I/O
merged before submitting to backend file/device.
In the following test:
- base: v3.19-rc2-2041231
- loop over file in ext4 file system on SSD disk
- bs: 4k, libaio, io depth: 64, O_DIRECT, num of jobs: 1
- throughput: IOPS
------------------------------------------------------
| | base | base with loop-mq | delta |
------------------------------------------------------
| randread | 1740 | 25318 | +1355%|
------------------------------------------------------
| read | 42196 | 51771 | +22.6%|
-----------------------------------------------------
| randwrite | 35709 | 34624 | -3% |
-----------------------------------------------------
| write | 39137 | 40326 | +3% |
-----------------------------------------------------
So loop-mq can improve throughput for both read and randread, meantime,
performance of write and randwrite isn't hurted basically.
Another benefit is that loop driver code gets simplified
much after blk-mq conversion, and the patch can be thought as
cleanup too.
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
2014-12-31 16:22:57 +03:00
|
|
|
blk_mq_free_tag_set(&lo->tag_set);
|
2021-09-02 03:07:35 +03:00
|
|
|
mutex_lock(&loop_ctl_mutex);
|
|
|
|
idr_remove(&loop_index_idr, lo->lo_number);
|
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
|
|
|
/* There is no route which can find this loop device. */
|
2021-01-26 17:46:30 +03:00
|
|
|
mutex_destroy(&lo->lo_mutex);
|
2007-05-08 11:28:20 +04:00
|
|
|
kfree(lo);
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2020-10-29 17:58:33 +03:00
|
|
|
static void loop_probe(dev_t dev)
|
2007-05-08 11:28:20 +04:00
|
|
|
{
|
2020-10-29 17:58:33 +03:00
|
|
|
int idx = MINOR(dev) >> part_shift;
|
|
|
|
|
|
|
|
if (max_loop && idx >= max_loop)
|
|
|
|
return;
|
2021-06-23 17:59:03 +03:00
|
|
|
loop_add(idx);
|
2021-06-23 17:59:04 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
static int loop_control_remove(int idx)
|
|
|
|
{
|
|
|
|
struct loop_device *lo;
|
|
|
|
int ret;
|
2021-06-23 17:59:06 +03:00
|
|
|
|
|
|
|
if (idx < 0) {
|
2021-11-29 13:00:43 +03:00
|
|
|
pr_warn_once("deleting an unspecified loop device is not supported.\n");
|
2021-06-23 17:59:06 +03:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
2021-06-23 17:59:04 +03:00
|
|
|
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Hide this loop device for serialization. */
|
2021-06-23 17:59:04 +03:00
|
|
|
ret = mutex_lock_killable(&loop_ctl_mutex);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2021-06-23 17:59:07 +03:00
|
|
|
lo = idr_find(&loop_index_idr, idx);
|
2021-09-02 03:07:35 +03:00
|
|
|
if (!lo || !lo->idr_visible)
|
2021-06-23 17:59:07 +03:00
|
|
|
ret = -ENODEV;
|
2021-09-02 03:07:35 +03:00
|
|
|
else
|
|
|
|
lo->idr_visible = false;
|
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2021-06-23 17:59:04 +03:00
|
|
|
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Check whether this loop device can be removed. */
|
2021-06-23 17:59:04 +03:00
|
|
|
ret = mutex_lock_killable(&lo->lo_mutex);
|
|
|
|
if (ret)
|
2021-09-02 03:07:35 +03:00
|
|
|
goto mark_visible;
|
2021-06-23 17:59:04 +03:00
|
|
|
if (lo->lo_state != Lo_unbound ||
|
|
|
|
atomic_read(&lo->lo_refcnt) > 0) {
|
|
|
|
mutex_unlock(&lo->lo_mutex);
|
|
|
|
ret = -EBUSY;
|
2021-09-02 03:07:35 +03:00
|
|
|
goto mark_visible;
|
2021-06-23 17:59:04 +03:00
|
|
|
}
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Mark this loop device no longer open()-able. */
|
2021-06-23 17:59:04 +03:00
|
|
|
lo->lo_state = Lo_deleting;
|
|
|
|
mutex_unlock(&lo->lo_mutex);
|
|
|
|
|
|
|
|
loop_remove(lo);
|
2021-09-02 03:07:35 +03:00
|
|
|
return 0;
|
|
|
|
|
|
|
|
mark_visible:
|
|
|
|
/* Show this loop device again. */
|
|
|
|
mutex_lock(&loop_ctl_mutex);
|
|
|
|
lo->idr_visible = true;
|
2021-06-23 17:59:04 +03:00
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int loop_control_get_free(int idx)
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
{
|
|
|
|
struct loop_device *lo;
|
2021-06-23 17:59:07 +03:00
|
|
|
int id, ret;
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
|
2018-11-08 16:01:04 +03:00
|
|
|
ret = mutex_lock_killable(&loop_ctl_mutex);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
2021-06-23 17:59:07 +03:00
|
|
|
idr_for_each_entry(&loop_index_idr, lo, id) {
|
2021-09-02 03:07:35 +03:00
|
|
|
/* Hitting a race results in creating a new loop device which is harmless. */
|
|
|
|
if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
|
2021-06-23 17:59:07 +03:00
|
|
|
goto found;
|
|
|
|
}
|
2021-06-23 17:59:04 +03:00
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
2021-06-23 17:59:05 +03:00
|
|
|
return loop_add(-1);
|
2021-06-23 17:59:07 +03:00
|
|
|
found:
|
|
|
|
mutex_unlock(&loop_ctl_mutex);
|
|
|
|
return id;
|
2021-06-23 17:59:04 +03:00
|
|
|
}
|
2018-11-08 16:01:04 +03:00
|
|
|
|
2021-06-23 17:59:04 +03:00
|
|
|
static long loop_control_ioctl(struct file *file, unsigned int cmd,
|
|
|
|
unsigned long parm)
|
|
|
|
{
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
switch (cmd) {
|
|
|
|
case LOOP_CTL_ADD:
|
2021-06-23 17:59:05 +03:00
|
|
|
return loop_add(parm);
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
case LOOP_CTL_REMOVE:
|
2021-06-23 17:59:04 +03:00
|
|
|
return loop_control_remove(parm);
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
case LOOP_CTL_GET_FREE:
|
2021-06-23 17:59:04 +03:00
|
|
|
return loop_control_get_free(parm);
|
|
|
|
default:
|
|
|
|
return -ENOSYS;
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static const struct file_operations loop_ctl_fops = {
|
|
|
|
.open = nonseekable_open,
|
|
|
|
.unlocked_ioctl = loop_control_ioctl,
|
|
|
|
.compat_ioctl = loop_control_ioctl,
|
|
|
|
.owner = THIS_MODULE,
|
|
|
|
.llseek = noop_llseek,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct miscdevice loop_misc = {
|
|
|
|
.minor = LOOP_CTRL_MINOR,
|
|
|
|
.name = "loop-control",
|
|
|
|
.fops = &loop_ctl_fops,
|
|
|
|
};
|
|
|
|
|
|
|
|
MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
|
|
|
|
MODULE_ALIAS("devname:loop-control");
|
|
|
|
|
2007-05-08 11:28:20 +04:00
|
|
|
static int __init loop_init(void)
|
|
|
|
{
|
2007-06-09 00:46:44 +04:00
|
|
|
int i, nr;
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
int err;
|
2007-06-09 00:46:44 +04:00
|
|
|
|
loop: manage partitions in disk image
This patch allows to use loop device with partitionned disk image.
Original behavior of loop is not modified.
A new parameter is introduced to define how many partition we want to be
able to manage per loop device. This parameter is "max_part".
For instance, to manage 63 partitions / loop device, we will do:
# modprobe loop max_part=63
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
And to attach a raw partitionned disk image, the original losetup is used:
# losetup -f etch.img
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 1 2008-03-05 14:57 /dev/loop0p1
brw-rw---- 1 root disk 7, 2 2008-03-05 14:57 /dev/loop0p2
brw-rw---- 1 root disk 7, 5 2008-03-05 14:57 /dev/loop0p5
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
# mount /dev/loop0p1 /mnt
# ls /mnt
bench cdrom home lib mnt root srv usr
bin dev initrd lost+found opt sbin sys var
boot etc initrd.img media proc selinux tmp vmlinuz
# umount /mnt
# losetup -d /dev/loop0
Of course, the same behavior can be done using kpartx on a loop device,
but modifying loop avoids to stack several layers of block device (loop +
device mapper), this is a very light modification (40% of modifications
are to manage the new parameter).
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2008-03-26 14:11:53 +03:00
|
|
|
part_shift = 0;
|
2011-05-27 09:59:25 +04:00
|
|
|
if (max_part > 0) {
|
loop: manage partitions in disk image
This patch allows to use loop device with partitionned disk image.
Original behavior of loop is not modified.
A new parameter is introduced to define how many partition we want to be
able to manage per loop device. This parameter is "max_part".
For instance, to manage 63 partitions / loop device, we will do:
# modprobe loop max_part=63
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
And to attach a raw partitionned disk image, the original losetup is used:
# losetup -f etch.img
# ls -l /dev/loop?*
brw-rw---- 1 root disk 7, 0 2008-03-05 14:55 /dev/loop0
brw-rw---- 1 root disk 7, 1 2008-03-05 14:57 /dev/loop0p1
brw-rw---- 1 root disk 7, 2 2008-03-05 14:57 /dev/loop0p2
brw-rw---- 1 root disk 7, 5 2008-03-05 14:57 /dev/loop0p5
brw-rw---- 1 root disk 7, 64 2008-03-05 14:55 /dev/loop1
brw-rw---- 1 root disk 7, 128 2008-03-05 14:55 /dev/loop2
brw-rw---- 1 root disk 7, 192 2008-03-05 14:55 /dev/loop3
brw-rw---- 1 root disk 7, 256 2008-03-05 14:55 /dev/loop4
brw-rw---- 1 root disk 7, 320 2008-03-05 14:55 /dev/loop5
brw-rw---- 1 root disk 7, 384 2008-03-05 14:55 /dev/loop6
brw-rw---- 1 root disk 7, 448 2008-03-05 14:55 /dev/loop7
# mount /dev/loop0p1 /mnt
# ls /mnt
bench cdrom home lib mnt root srv usr
bin dev initrd lost+found opt sbin sys var
boot etc initrd.img media proc selinux tmp vmlinuz
# umount /mnt
# losetup -d /dev/loop0
Of course, the same behavior can be done using kpartx on a loop device,
but modifying loop avoids to stack several layers of block device (loop +
device mapper), this is a very light modification (40% of modifications
are to manage the new parameter).
Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
2008-03-26 14:11:53 +03:00
|
|
|
part_shift = fls(max_part);
|
|
|
|
|
2011-05-27 09:59:25 +04:00
|
|
|
/*
|
|
|
|
* Adjust max_part according to part_shift as it is exported
|
|
|
|
* to user space so that user can decide correct minor number
|
|
|
|
* if [s]he want to create more devices.
|
|
|
|
*
|
|
|
|
* Note that -1 is required because partition 0 is reserved
|
|
|
|
* for the whole disk.
|
|
|
|
*/
|
|
|
|
max_part = (1UL << part_shift) - 1;
|
|
|
|
}
|
|
|
|
|
2013-02-22 03:16:49 +04:00
|
|
|
if ((1UL << part_shift) > DISK_MAX_PARTS) {
|
|
|
|
err = -EINVAL;
|
2017-08-07 15:37:50 +03:00
|
|
|
goto err_out;
|
2013-02-22 03:16:49 +04:00
|
|
|
}
|
2011-05-24 18:48:54 +04:00
|
|
|
|
2013-02-22 03:16:49 +04:00
|
|
|
if (max_loop > 1UL << (MINORBITS - part_shift)) {
|
|
|
|
err = -EINVAL;
|
2017-08-07 15:37:50 +03:00
|
|
|
goto err_out;
|
2013-02-22 03:16:49 +04:00
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2011-08-01 00:08:04 +04:00
|
|
|
/*
|
|
|
|
* If max_loop is specified, create that many devices upfront.
|
|
|
|
* This also becomes a hard limit. If max_loop is not specified,
|
|
|
|
* create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
|
|
|
|
* init time. Loop devices can be requested on-demand with the
|
|
|
|
* /dev/loop-control interface, or be instantiated by accessing
|
|
|
|
* a 'dead' device node.
|
|
|
|
*/
|
2020-12-12 08:13:02 +03:00
|
|
|
if (max_loop)
|
2007-06-09 00:46:44 +04:00
|
|
|
nr = max_loop;
|
2020-12-12 08:13:02 +03:00
|
|
|
else
|
2011-08-01 00:08:04 +04:00
|
|
|
nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
|
2007-06-09 00:46:44 +04:00
|
|
|
|
2017-08-07 15:37:50 +03:00
|
|
|
err = misc_register(&loop_misc);
|
|
|
|
if (err < 0)
|
|
|
|
goto err_out;
|
|
|
|
|
|
|
|
|
2020-10-29 17:58:33 +03:00
|
|
|
if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
|
2013-02-22 03:16:49 +04:00
|
|
|
err = -EIO;
|
|
|
|
goto misc_out;
|
|
|
|
}
|
2005-04-17 02:20:36 +04:00
|
|
|
|
2011-08-01 00:08:04 +04:00
|
|
|
/* pre-create number of devices given by config or max_loop */
|
2011-08-01 00:08:04 +04:00
|
|
|
for (i = 0; i < nr; i++)
|
2021-06-23 17:59:02 +03:00
|
|
|
loop_add(i);
|
2011-08-01 00:08:04 +04:00
|
|
|
|
2007-05-08 11:28:20 +04:00
|
|
|
printk(KERN_INFO "loop: module loaded\n");
|
2005-04-17 02:20:36 +04:00
|
|
|
return 0;
|
2013-02-22 03:16:49 +04:00
|
|
|
|
|
|
|
misc_out:
|
|
|
|
misc_deregister(&loop_misc);
|
2017-08-07 15:37:50 +03:00
|
|
|
err_out:
|
2013-02-22 03:16:49 +04:00
|
|
|
return err;
|
2011-08-01 00:08:04 +04:00
|
|
|
}
|
2007-06-09 00:46:44 +04:00
|
|
|
|
2007-05-08 11:28:20 +04:00
|
|
|
static void __exit loop_exit(void)
|
2005-04-17 02:20:36 +04:00
|
|
|
{
|
2021-06-23 17:59:08 +03:00
|
|
|
struct loop_device *lo;
|
|
|
|
int id;
|
|
|
|
|
2007-07-17 15:03:46 +04:00
|
|
|
unregister_blkdev(LOOP_MAJOR, "loop");
|
loop: add management interface for on-demand device allocation
Loop devices today have a fixed pre-allocated number of usually 8.
The number can only be changed at module init time. To find a free
device to use, /dev/loop%i needs to be scanned, and all devices need
to be opened until a free one is possibly found.
This adds a new /dev/loop-control device node, that allows to
dynamically find or allocate a free device, and to add and remove loop
devices from the running system:
LOOP_CTL_ADD adds a specific device. Arg is the number
of the device. It returns the device i or a negative
error code.
LOOP_CTL_REMOVE removes a specific device, Arg is the
number the device. It returns the device i or a negative
error code.
LOOP_CTL_GET_FREE finds the next unbound device or allocates
a new one. No arg is given. It returns the device i or a
negative error code.
The loop kernel module gets automatically loaded when
/dev/loop-control is accessed the first time. The alias
specified in the module, instructs udev to create this
'dead' device node, even when the module is not loaded.
Example:
cfd = open("/dev/loop-control", O_RDWR);
# add a new specific loop device
err = ioctl(cfd, LOOP_CTL_ADD, devnr);
# remove a specific loop device
err = ioctl(cfd, LOOP_CTL_REMOVE, devnr);
# find or allocate a free loop device to use
devnr = ioctl(cfd, LOOP_CTL_GET_FREE);
sprintf(loopname, "/dev/loop%i", devnr);
ffd = open("backing-file", O_RDWR);
lfd = open(loopname, O_RDWR);
err = ioctl(lfd, LOOP_SET_FD, ffd);
Cc: Tejun Heo <tj@kernel.org>
Cc: Karel Zak <kzak@redhat.com>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
2011-08-01 00:08:04 +04:00
|
|
|
misc_deregister(&loop_misc);
|
2020-06-19 23:47:27 +03:00
|
|
|
|
2021-09-02 03:07:35 +03:00
|
|
|
/*
|
|
|
|
* There is no need to use loop_ctl_mutex here, for nobody else can
|
|
|
|
* access loop_index_idr when this module is unloading (unless forced
|
|
|
|
* module unloading is requested). If this is not a clean unloading,
|
|
|
|
* we have no means to avoid kernel crash.
|
|
|
|
*/
|
2021-06-23 17:59:08 +03:00
|
|
|
idr_for_each_entry(&loop_index_idr, lo, id)
|
|
|
|
loop_remove(lo);
|
2021-06-23 17:59:01 +03:00
|
|
|
|
|
|
|
idr_destroy(&loop_index_idr);
|
2005-04-17 02:20:36 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
module_init(loop_init);
|
|
|
|
module_exit(loop_exit);
|
|
|
|
|
|
|
|
#ifndef MODULE
|
|
|
|
static int __init max_loop_setup(char *str)
|
|
|
|
{
|
|
|
|
max_loop = simple_strtol(str, NULL, 0);
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
__setup("max_loop=", max_loop_setup);
|
|
|
|
#endif
|