1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 1993 by Theodore Ts'o.
5 #include <linux/module.h>
6 #include <linux/moduleparam.h>
7 #include <linux/sched.h>
9 #include <linux/pagemap.h>
10 #include <linux/file.h>
11 #include <linux/stat.h>
12 #include <linux/errno.h>
13 #include <linux/major.h>
14 #include <linux/wait.h>
15 #include <linux/blkpg.h>
16 #include <linux/init.h>
17 #include <linux/swap.h>
18 #include <linux/slab.h>
19 #include <linux/compat.h>
20 #include <linux/suspend.h>
21 #include <linux/freezer.h>
22 #include <linux/mutex.h>
23 #include <linux/writeback.h>
24 #include <linux/completion.h>
25 #include <linux/highmem.h>
26 #include <linux/splice.h>
27 #include <linux/sysfs.h>
28 #include <linux/miscdevice.h>
29 #include <linux/falloc.h>
30 #include <linux/uio.h>
31 #include <linux/ioprio.h>
32 #include <linux/blk-cgroup.h>
33 #include <linux/sched/mm.h>
34 #include <linux/statfs.h>
35 #include <linux/uaccess.h>
36 #include <linux/blk-mq.h>
37 #include <linux/spinlock.h>
38 #include <uapi/linux/loop.h>
40 /* Possible states of device */
48 struct loop_func_table
;
55 char lo_file_name
[LO_NAME_SIZE
];
57 struct file
* lo_backing_file
;
58 struct block_device
*lo_device
;
64 spinlock_t lo_work_lock
;
65 struct workqueue_struct
*workqueue
;
66 struct work_struct rootcg_work
;
67 struct list_head rootcg_cmd_list
;
68 struct list_head idle_worker_list
;
69 struct rb_root worker_tree
;
70 struct timer_list timer
;
74 struct request_queue
*lo_queue
;
75 struct blk_mq_tag_set tag_set
;
76 struct gendisk
*lo_disk
;
77 struct mutex lo_mutex
;
82 struct list_head list_entry
;
83 bool use_aio
; /* use AIO interface to handle I/O */
84 atomic_t ref
; /* only for aio */
88 struct cgroup_subsys_state
*blkcg_css
;
89 struct cgroup_subsys_state
*memcg_css
;
92 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
93 #define LOOP_DEFAULT_HW_Q_DEPTH 128
95 static DEFINE_IDR(loop_index_idr
);
96 static DEFINE_MUTEX(loop_ctl_mutex
);
97 static DEFINE_MUTEX(loop_validate_mutex
);
100 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
102 * @lo: struct loop_device
103 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
105 * Returns 0 on success, -EINTR otherwise.
107 * Since loop_validate_file() traverses on other "struct loop_device" if
108 * is_loop_device() is true, we need a global lock for serializing concurrent
109 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
111 static int loop_global_lock_killable(struct loop_device
*lo
, bool global
)
116 err
= mutex_lock_killable(&loop_validate_mutex
);
120 err
= mutex_lock_killable(&lo
->lo_mutex
);
122 mutex_unlock(&loop_validate_mutex
);
127 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
129 * @lo: struct loop_device
130 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
132 static void loop_global_unlock(struct loop_device
*lo
, bool global
)
134 mutex_unlock(&lo
->lo_mutex
);
136 mutex_unlock(&loop_validate_mutex
);
140 static int part_shift
;
142 static loff_t
get_size(loff_t offset
, loff_t sizelimit
, struct file
*file
)
146 /* Compute loopsize in bytes */
147 loopsize
= i_size_read(file
->f_mapping
->host
);
150 /* offset is beyond i_size, weird but possible */
154 if (sizelimit
> 0 && sizelimit
< loopsize
)
155 loopsize
= sizelimit
;
157 * Unfortunately, if we want to do I/O on the device,
158 * the number of 512-byte sectors has to fit into a sector_t.
160 return loopsize
>> 9;
163 static loff_t
get_loop_size(struct loop_device
*lo
, struct file
*file
)
165 return get_size(lo
->lo_offset
, lo
->lo_sizelimit
, file
);
168 static void __loop_update_dio(struct loop_device
*lo
, bool dio
)
170 struct file
*file
= lo
->lo_backing_file
;
171 struct address_space
*mapping
= file
->f_mapping
;
172 struct inode
*inode
= mapping
->host
;
173 unsigned short sb_bsize
= 0;
174 unsigned dio_align
= 0;
177 if (inode
->i_sb
->s_bdev
) {
178 sb_bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
179 dio_align
= sb_bsize
- 1;
183 * We support direct I/O only if lo_offset is aligned with the
184 * logical I/O size of backing device, and the logical block
185 * size of loop is bigger than the backing device's.
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane applications should be PAGE_SIZE aligned
192 if (queue_logical_block_size(lo
->lo_queue
) >= sb_bsize
&&
193 !(lo
->lo_offset
& dio_align
) &&
194 (file
->f_mode
& FMODE_CAN_ODIRECT
))
202 if (lo
->use_dio
== use_dio
)
205 /* flush dirty pages before changing direct IO */
209 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
210 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
211 * will get updated by ioctl(LOOP_GET_STATUS)
213 if (lo
->lo_state
== Lo_bound
)
214 blk_mq_freeze_queue(lo
->lo_queue
);
215 lo
->use_dio
= use_dio
;
217 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
218 lo
->lo_flags
|= LO_FLAGS_DIRECT_IO
;
220 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
221 lo
->lo_flags
&= ~LO_FLAGS_DIRECT_IO
;
223 if (lo
->lo_state
== Lo_bound
)
224 blk_mq_unfreeze_queue(lo
->lo_queue
);
228 * loop_set_size() - sets device size and notifies userspace
229 * @lo: struct loop_device to set the size for
230 * @size: new size of the loop device
232 * Callers must validate that the size passed into this function fits into
233 * a sector_t, eg using loop_validate_size()
235 static void loop_set_size(struct loop_device
*lo
, loff_t size
)
237 if (!set_capacity_and_notify(lo
->lo_disk
, size
))
238 kobject_uevent(&disk_to_dev(lo
->lo_disk
)->kobj
, KOBJ_CHANGE
);
241 static int lo_write_bvec(struct file
*file
, struct bio_vec
*bvec
, loff_t
*ppos
)
246 iov_iter_bvec(&i
, ITER_SOURCE
, bvec
, 1, bvec
->bv_len
);
248 bw
= vfs_iter_write(file
, &i
, ppos
, 0);
250 if (likely(bw
== bvec
->bv_len
))
253 printk_ratelimited(KERN_ERR
254 "loop: Write error at byte offset %llu, length %i.\n",
255 (unsigned long long)*ppos
, bvec
->bv_len
);
261 static int lo_write_simple(struct loop_device
*lo
, struct request
*rq
,
265 struct req_iterator iter
;
268 rq_for_each_segment(bvec
, rq
, iter
) {
269 ret
= lo_write_bvec(lo
->lo_backing_file
, &bvec
, &pos
);
278 static int lo_read_simple(struct loop_device
*lo
, struct request
*rq
,
282 struct req_iterator iter
;
286 rq_for_each_segment(bvec
, rq
, iter
) {
287 iov_iter_bvec(&i
, ITER_DEST
, &bvec
, 1, bvec
.bv_len
);
288 len
= vfs_iter_read(lo
->lo_backing_file
, &i
, &pos
, 0);
292 flush_dcache_page(bvec
.bv_page
);
294 if (len
!= bvec
.bv_len
) {
297 __rq_for_each_bio(bio
, rq
)
307 static int lo_fallocate(struct loop_device
*lo
, struct request
*rq
, loff_t pos
,
311 * We use fallocate to manipulate the space mappings used by the image
312 * a.k.a. discard/zerorange.
314 struct file
*file
= lo
->lo_backing_file
;
317 mode
|= FALLOC_FL_KEEP_SIZE
;
319 if (!bdev_max_discard_sectors(lo
->lo_device
))
322 ret
= file
->f_op
->fallocate(file
, mode
, pos
, blk_rq_bytes(rq
));
323 if (unlikely(ret
&& ret
!= -EINVAL
&& ret
!= -EOPNOTSUPP
))
328 static int lo_req_flush(struct loop_device
*lo
, struct request
*rq
)
330 int ret
= vfs_fsync(lo
->lo_backing_file
, 0);
331 if (unlikely(ret
&& ret
!= -EINVAL
))
337 static void lo_complete_rq(struct request
*rq
)
339 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
340 blk_status_t ret
= BLK_STS_OK
;
342 if (!cmd
->use_aio
|| cmd
->ret
< 0 || cmd
->ret
== blk_rq_bytes(rq
) ||
343 req_op(rq
) != REQ_OP_READ
) {
345 ret
= errno_to_blk_status(cmd
->ret
);
350 * Short READ - if we got some data, advance our request and
351 * retry it. If we got no data, end the rest with EIO.
354 blk_update_request(rq
, BLK_STS_OK
, cmd
->ret
);
356 blk_mq_requeue_request(rq
, true);
359 struct bio
*bio
= rq
->bio
;
368 blk_mq_end_request(rq
, ret
);
372 static void lo_rw_aio_do_completion(struct loop_cmd
*cmd
)
374 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
376 if (!atomic_dec_and_test(&cmd
->ref
))
380 if (likely(!blk_should_fake_timeout(rq
->q
)))
381 blk_mq_complete_request(rq
);
384 static void lo_rw_aio_complete(struct kiocb
*iocb
, long ret
)
386 struct loop_cmd
*cmd
= container_of(iocb
, struct loop_cmd
, iocb
);
389 lo_rw_aio_do_completion(cmd
);
392 static int lo_rw_aio(struct loop_device
*lo
, struct loop_cmd
*cmd
,
395 struct iov_iter iter
;
396 struct req_iterator rq_iter
;
397 struct bio_vec
*bvec
;
398 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
399 struct bio
*bio
= rq
->bio
;
400 struct file
*file
= lo
->lo_backing_file
;
406 rq_for_each_bvec(tmp
, rq
, rq_iter
)
409 if (rq
->bio
!= rq
->biotail
) {
411 bvec
= kmalloc_array(nr_bvec
, sizeof(struct bio_vec
),
418 * The bios of the request may be started from the middle of
419 * the 'bvec' because of bio splitting, so we can't directly
420 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
421 * API will take care of all details for us.
423 rq_for_each_bvec(tmp
, rq
, rq_iter
) {
431 * Same here, this bio may be started from the middle of the
432 * 'bvec' because of bio splitting, so offset from the bvec
433 * must be passed to iov iterator
435 offset
= bio
->bi_iter
.bi_bvec_done
;
436 bvec
= __bvec_iter_bvec(bio
->bi_io_vec
, bio
->bi_iter
);
438 atomic_set(&cmd
->ref
, 2);
440 iov_iter_bvec(&iter
, rw
, bvec
, nr_bvec
, blk_rq_bytes(rq
));
441 iter
.iov_offset
= offset
;
443 cmd
->iocb
.ki_pos
= pos
;
444 cmd
->iocb
.ki_filp
= file
;
445 cmd
->iocb
.ki_complete
= lo_rw_aio_complete
;
446 cmd
->iocb
.ki_flags
= IOCB_DIRECT
;
447 cmd
->iocb
.ki_ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
449 if (rw
== ITER_SOURCE
)
450 ret
= call_write_iter(file
, &cmd
->iocb
, &iter
);
452 ret
= call_read_iter(file
, &cmd
->iocb
, &iter
);
454 lo_rw_aio_do_completion(cmd
);
456 if (ret
!= -EIOCBQUEUED
)
457 lo_rw_aio_complete(&cmd
->iocb
, ret
);
461 static int do_req_filebacked(struct loop_device
*lo
, struct request
*rq
)
463 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
464 loff_t pos
= ((loff_t
) blk_rq_pos(rq
) << 9) + lo
->lo_offset
;
467 * lo_write_simple and lo_read_simple should have been covered
468 * by io submit style function like lo_rw_aio(), one blocker
469 * is that lo_read_simple() need to call flush_dcache_page after
470 * the page is written from kernel, and it isn't easy to handle
471 * this in io submit style function which submits all segments
472 * of the req at one time. And direct read IO doesn't need to
473 * run flush_dcache_page().
475 switch (req_op(rq
)) {
477 return lo_req_flush(lo
, rq
);
478 case REQ_OP_WRITE_ZEROES
:
480 * If the caller doesn't want deallocation, call zeroout to
481 * write zeroes the range. Otherwise, punch them out.
483 return lo_fallocate(lo
, rq
, pos
,
484 (rq
->cmd_flags
& REQ_NOUNMAP
) ?
485 FALLOC_FL_ZERO_RANGE
:
486 FALLOC_FL_PUNCH_HOLE
);
488 return lo_fallocate(lo
, rq
, pos
, FALLOC_FL_PUNCH_HOLE
);
491 return lo_rw_aio(lo
, cmd
, pos
, ITER_SOURCE
);
493 return lo_write_simple(lo
, rq
, pos
);
496 return lo_rw_aio(lo
, cmd
, pos
, ITER_DEST
);
498 return lo_read_simple(lo
, rq
, pos
);
505 static inline void loop_update_dio(struct loop_device
*lo
)
507 __loop_update_dio(lo
, (lo
->lo_backing_file
->f_flags
& O_DIRECT
) |
511 static void loop_reread_partitions(struct loop_device
*lo
)
515 mutex_lock(&lo
->lo_disk
->open_mutex
);
516 rc
= bdev_disk_changed(lo
->lo_disk
, false);
517 mutex_unlock(&lo
->lo_disk
->open_mutex
);
519 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
520 __func__
, lo
->lo_number
, lo
->lo_file_name
, rc
);
523 static inline int is_loop_device(struct file
*file
)
525 struct inode
*i
= file
->f_mapping
->host
;
527 return i
&& S_ISBLK(i
->i_mode
) && imajor(i
) == LOOP_MAJOR
;
530 static int loop_validate_file(struct file
*file
, struct block_device
*bdev
)
532 struct inode
*inode
= file
->f_mapping
->host
;
533 struct file
*f
= file
;
535 /* Avoid recursion */
536 while (is_loop_device(f
)) {
537 struct loop_device
*l
;
539 lockdep_assert_held(&loop_validate_mutex
);
540 if (f
->f_mapping
->host
->i_rdev
== bdev
->bd_dev
)
543 l
= I_BDEV(f
->f_mapping
->host
)->bd_disk
->private_data
;
544 if (l
->lo_state
!= Lo_bound
)
546 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
548 f
= l
->lo_backing_file
;
550 if (!S_ISREG(inode
->i_mode
) && !S_ISBLK(inode
->i_mode
))
556 * loop_change_fd switched the backing store of a loopback device to
557 * a new file. This is useful for operating system installers to free up
558 * the original file and in High Availability environments to switch to
559 * an alternative location for the content in case of server meltdown.
560 * This can only work if the loop device is used read-only, and if the
561 * new backing store is the same size and type as the old backing store.
563 static int loop_change_fd(struct loop_device
*lo
, struct block_device
*bdev
,
566 struct file
*file
= fget(arg
);
567 struct file
*old_file
;
575 /* suppress uevents while reconfiguring the device */
576 dev_set_uevent_suppress(disk_to_dev(lo
->lo_disk
), 1);
578 is_loop
= is_loop_device(file
);
579 error
= loop_global_lock_killable(lo
, is_loop
);
583 if (lo
->lo_state
!= Lo_bound
)
586 /* the loop device has to be read-only */
588 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
))
591 error
= loop_validate_file(file
, bdev
);
595 old_file
= lo
->lo_backing_file
;
599 /* size of the new backing store needs to be the same */
600 if (get_loop_size(lo
, file
) != get_loop_size(lo
, old_file
))
604 disk_force_media_change(lo
->lo_disk
);
605 blk_mq_freeze_queue(lo
->lo_queue
);
606 mapping_set_gfp_mask(old_file
->f_mapping
, lo
->old_gfp_mask
);
607 lo
->lo_backing_file
= file
;
608 lo
->old_gfp_mask
= mapping_gfp_mask(file
->f_mapping
);
609 mapping_set_gfp_mask(file
->f_mapping
,
610 lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
612 blk_mq_unfreeze_queue(lo
->lo_queue
);
613 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
614 loop_global_unlock(lo
, is_loop
);
617 * Flush loop_validate_file() before fput(), for l->lo_backing_file
618 * might be pointing at old_file which might be the last reference.
621 mutex_lock(&loop_validate_mutex
);
622 mutex_unlock(&loop_validate_mutex
);
625 * We must drop file reference outside of lo_mutex as dropping
626 * the file ref can take open_mutex which creates circular locking
631 loop_reread_partitions(lo
);
635 /* enable and uncork uevent now that we are done */
636 dev_set_uevent_suppress(disk_to_dev(lo
->lo_disk
), 0);
640 loop_global_unlock(lo
, is_loop
);
646 /* loop sysfs attributes */
648 static ssize_t
loop_attr_show(struct device
*dev
, char *page
,
649 ssize_t (*callback
)(struct loop_device
*, char *))
651 struct gendisk
*disk
= dev_to_disk(dev
);
652 struct loop_device
*lo
= disk
->private_data
;
654 return callback(lo
, page
);
657 #define LOOP_ATTR_RO(_name) \
658 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
659 static ssize_t loop_attr_do_show_##_name(struct device *d, \
660 struct device_attribute *attr, char *b) \
662 return loop_attr_show(d, b, loop_attr_##_name##_show); \
664 static struct device_attribute loop_attr_##_name = \
665 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
667 static ssize_t
loop_attr_backing_file_show(struct loop_device
*lo
, char *buf
)
672 spin_lock_irq(&lo
->lo_lock
);
673 if (lo
->lo_backing_file
)
674 p
= file_path(lo
->lo_backing_file
, buf
, PAGE_SIZE
- 1);
675 spin_unlock_irq(&lo
->lo_lock
);
677 if (IS_ERR_OR_NULL(p
))
681 memmove(buf
, p
, ret
);
689 static ssize_t
loop_attr_offset_show(struct loop_device
*lo
, char *buf
)
691 return sysfs_emit(buf
, "%llu\n", (unsigned long long)lo
->lo_offset
);
694 static ssize_t
loop_attr_sizelimit_show(struct loop_device
*lo
, char *buf
)
696 return sysfs_emit(buf
, "%llu\n", (unsigned long long)lo
->lo_sizelimit
);
699 static ssize_t
loop_attr_autoclear_show(struct loop_device
*lo
, char *buf
)
701 int autoclear
= (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
);
703 return sysfs_emit(buf
, "%s\n", autoclear
? "1" : "0");
706 static ssize_t
loop_attr_partscan_show(struct loop_device
*lo
, char *buf
)
708 int partscan
= (lo
->lo_flags
& LO_FLAGS_PARTSCAN
);
710 return sysfs_emit(buf
, "%s\n", partscan
? "1" : "0");
713 static ssize_t
loop_attr_dio_show(struct loop_device
*lo
, char *buf
)
715 int dio
= (lo
->lo_flags
& LO_FLAGS_DIRECT_IO
);
717 return sysfs_emit(buf
, "%s\n", dio
? "1" : "0");
720 LOOP_ATTR_RO(backing_file
);
721 LOOP_ATTR_RO(offset
);
722 LOOP_ATTR_RO(sizelimit
);
723 LOOP_ATTR_RO(autoclear
);
724 LOOP_ATTR_RO(partscan
);
727 static struct attribute
*loop_attrs
[] = {
728 &loop_attr_backing_file
.attr
,
729 &loop_attr_offset
.attr
,
730 &loop_attr_sizelimit
.attr
,
731 &loop_attr_autoclear
.attr
,
732 &loop_attr_partscan
.attr
,
737 static struct attribute_group loop_attribute_group
= {
742 static void loop_sysfs_init(struct loop_device
*lo
)
744 lo
->sysfs_inited
= !sysfs_create_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
745 &loop_attribute_group
);
748 static void loop_sysfs_exit(struct loop_device
*lo
)
750 if (lo
->sysfs_inited
)
751 sysfs_remove_group(&disk_to_dev(lo
->lo_disk
)->kobj
,
752 &loop_attribute_group
);
755 static void loop_config_discard(struct loop_device
*lo
)
757 struct file
*file
= lo
->lo_backing_file
;
758 struct inode
*inode
= file
->f_mapping
->host
;
759 struct request_queue
*q
= lo
->lo_queue
;
760 u32 granularity
, max_discard_sectors
;
763 * If the backing device is a block device, mirror its zeroing
764 * capability. Set the discard sectors to the block device's zeroing
765 * capabilities because loop discards result in blkdev_issue_zeroout(),
766 * not blkdev_issue_discard(). This maintains consistent behavior with
767 * file-backed loop devices: discarded regions read back as zero.
769 if (S_ISBLK(inode
->i_mode
)) {
770 struct request_queue
*backingq
= bdev_get_queue(I_BDEV(inode
));
772 max_discard_sectors
= backingq
->limits
.max_write_zeroes_sectors
;
773 granularity
= bdev_discard_granularity(I_BDEV(inode
)) ?:
774 queue_physical_block_size(backingq
);
777 * We use punch hole to reclaim the free space used by the
778 * image a.k.a. discard.
780 } else if (!file
->f_op
->fallocate
) {
781 max_discard_sectors
= 0;
787 max_discard_sectors
= UINT_MAX
>> 9;
788 if (!vfs_statfs(&file
->f_path
, &sbuf
))
789 granularity
= sbuf
.f_bsize
;
791 max_discard_sectors
= 0;
794 if (max_discard_sectors
) {
795 q
->limits
.discard_granularity
= granularity
;
796 blk_queue_max_discard_sectors(q
, max_discard_sectors
);
797 blk_queue_max_write_zeroes_sectors(q
, max_discard_sectors
);
799 q
->limits
.discard_granularity
= 0;
800 blk_queue_max_discard_sectors(q
, 0);
801 blk_queue_max_write_zeroes_sectors(q
, 0);
806 struct rb_node rb_node
;
807 struct work_struct work
;
808 struct list_head cmd_list
;
809 struct list_head idle_list
;
810 struct loop_device
*lo
;
811 struct cgroup_subsys_state
*blkcg_css
;
812 unsigned long last_ran_at
;
815 static void loop_workfn(struct work_struct
*work
);
817 #ifdef CONFIG_BLK_CGROUP
818 static inline int queue_on_root_worker(struct cgroup_subsys_state
*css
)
820 return !css
|| css
== blkcg_root_css
;
823 static inline int queue_on_root_worker(struct cgroup_subsys_state
*css
)
829 static void loop_queue_work(struct loop_device
*lo
, struct loop_cmd
*cmd
)
831 struct rb_node
**node
, *parent
= NULL
;
832 struct loop_worker
*cur_worker
, *worker
= NULL
;
833 struct work_struct
*work
;
834 struct list_head
*cmd_list
;
836 spin_lock_irq(&lo
->lo_work_lock
);
838 if (queue_on_root_worker(cmd
->blkcg_css
))
841 node
= &lo
->worker_tree
.rb_node
;
845 cur_worker
= container_of(*node
, struct loop_worker
, rb_node
);
846 if (cur_worker
->blkcg_css
== cmd
->blkcg_css
) {
849 } else if ((long)cur_worker
->blkcg_css
< (long)cmd
->blkcg_css
) {
850 node
= &(*node
)->rb_left
;
852 node
= &(*node
)->rb_right
;
858 worker
= kzalloc(sizeof(struct loop_worker
), GFP_NOWAIT
| __GFP_NOWARN
);
860 * In the event we cannot allocate a worker, just queue on the
861 * rootcg worker and issue the I/O as the rootcg
864 cmd
->blkcg_css
= NULL
;
866 css_put(cmd
->memcg_css
);
867 cmd
->memcg_css
= NULL
;
871 worker
->blkcg_css
= cmd
->blkcg_css
;
872 css_get(worker
->blkcg_css
);
873 INIT_WORK(&worker
->work
, loop_workfn
);
874 INIT_LIST_HEAD(&worker
->cmd_list
);
875 INIT_LIST_HEAD(&worker
->idle_list
);
877 rb_link_node(&worker
->rb_node
, parent
, node
);
878 rb_insert_color(&worker
->rb_node
, &lo
->worker_tree
);
882 * We need to remove from the idle list here while
883 * holding the lock so that the idle timer doesn't
886 if (!list_empty(&worker
->idle_list
))
887 list_del_init(&worker
->idle_list
);
888 work
= &worker
->work
;
889 cmd_list
= &worker
->cmd_list
;
891 work
= &lo
->rootcg_work
;
892 cmd_list
= &lo
->rootcg_cmd_list
;
894 list_add_tail(&cmd
->list_entry
, cmd_list
);
895 queue_work(lo
->workqueue
, work
);
896 spin_unlock_irq(&lo
->lo_work_lock
);
899 static void loop_set_timer(struct loop_device
*lo
)
901 timer_reduce(&lo
->timer
, jiffies
+ LOOP_IDLE_WORKER_TIMEOUT
);
904 static void loop_free_idle_workers(struct loop_device
*lo
, bool delete_all
)
906 struct loop_worker
*pos
, *worker
;
908 spin_lock_irq(&lo
->lo_work_lock
);
909 list_for_each_entry_safe(worker
, pos
, &lo
->idle_worker_list
,
912 time_is_after_jiffies(worker
->last_ran_at
+
913 LOOP_IDLE_WORKER_TIMEOUT
))
915 list_del(&worker
->idle_list
);
916 rb_erase(&worker
->rb_node
, &lo
->worker_tree
);
917 css_put(worker
->blkcg_css
);
920 if (!list_empty(&lo
->idle_worker_list
))
922 spin_unlock_irq(&lo
->lo_work_lock
);
925 static void loop_free_idle_workers_timer(struct timer_list
*timer
)
927 struct loop_device
*lo
= container_of(timer
, struct loop_device
, timer
);
929 return loop_free_idle_workers(lo
, false);
932 static void loop_update_rotational(struct loop_device
*lo
)
934 struct file
*file
= lo
->lo_backing_file
;
935 struct inode
*file_inode
= file
->f_mapping
->host
;
936 struct block_device
*file_bdev
= file_inode
->i_sb
->s_bdev
;
937 struct request_queue
*q
= lo
->lo_queue
;
940 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
942 nonrot
= bdev_nonrot(file_bdev
);
945 blk_queue_flag_set(QUEUE_FLAG_NONROT
, q
);
947 blk_queue_flag_clear(QUEUE_FLAG_NONROT
, q
);
951 * loop_set_status_from_info - configure device from loop_info
952 * @lo: struct loop_device to configure
953 * @info: struct loop_info64 to configure the device with
955 * Configures the loop device parameters according to the passed
956 * in loop_info64 configuration.
959 loop_set_status_from_info(struct loop_device
*lo
,
960 const struct loop_info64
*info
)
962 if ((unsigned int) info
->lo_encrypt_key_size
> LO_KEY_SIZE
)
965 switch (info
->lo_encrypt_type
) {
969 pr_warn("support for the xor transformation has been removed.\n");
971 case LO_CRYPT_CRYPTOAPI
:
972 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
978 /* Avoid assigning overflow values */
979 if (info
->lo_offset
> LLONG_MAX
|| info
->lo_sizelimit
> LLONG_MAX
)
982 lo
->lo_offset
= info
->lo_offset
;
983 lo
->lo_sizelimit
= info
->lo_sizelimit
;
985 memcpy(lo
->lo_file_name
, info
->lo_file_name
, LO_NAME_SIZE
);
986 lo
->lo_file_name
[LO_NAME_SIZE
-1] = 0;
987 lo
->lo_flags
= info
->lo_flags
;
991 static int loop_configure(struct loop_device
*lo
, blk_mode_t mode
,
992 struct block_device
*bdev
,
993 const struct loop_config
*config
)
995 struct file
*file
= fget(config
->fd
);
997 struct address_space
*mapping
;
1001 unsigned short bsize
;
1006 is_loop
= is_loop_device(file
);
1008 /* This is safe, since we have a reference from open(). */
1009 __module_get(THIS_MODULE
);
1012 * If we don't hold exclusive handle for the device, upgrade to it
1013 * here to avoid changing device under exclusive owner.
1015 if (!(mode
& BLK_OPEN_EXCL
)) {
1016 error
= bd_prepare_to_claim(bdev
, loop_configure
, NULL
);
1021 error
= loop_global_lock_killable(lo
, is_loop
);
1026 if (lo
->lo_state
!= Lo_unbound
)
1029 error
= loop_validate_file(file
, bdev
);
1033 mapping
= file
->f_mapping
;
1034 inode
= mapping
->host
;
1036 if ((config
->info
.lo_flags
& ~LOOP_CONFIGURE_SETTABLE_FLAGS
) != 0) {
1041 if (config
->block_size
) {
1042 error
= blk_validate_block_size(config
->block_size
);
1047 error
= loop_set_status_from_info(lo
, &config
->info
);
1051 if (!(file
->f_mode
& FMODE_WRITE
) || !(mode
& BLK_OPEN_WRITE
) ||
1052 !file
->f_op
->write_iter
)
1053 lo
->lo_flags
|= LO_FLAGS_READ_ONLY
;
1055 if (!lo
->workqueue
) {
1056 lo
->workqueue
= alloc_workqueue("loop%d",
1057 WQ_UNBOUND
| WQ_FREEZABLE
,
1059 if (!lo
->workqueue
) {
1065 /* suppress uevents while reconfiguring the device */
1066 dev_set_uevent_suppress(disk_to_dev(lo
->lo_disk
), 1);
1068 disk_force_media_change(lo
->lo_disk
);
1069 set_disk_ro(lo
->lo_disk
, (lo
->lo_flags
& LO_FLAGS_READ_ONLY
) != 0);
1071 lo
->use_dio
= lo
->lo_flags
& LO_FLAGS_DIRECT_IO
;
1072 lo
->lo_device
= bdev
;
1073 lo
->lo_backing_file
= file
;
1074 lo
->old_gfp_mask
= mapping_gfp_mask(mapping
);
1075 mapping_set_gfp_mask(mapping
, lo
->old_gfp_mask
& ~(__GFP_IO
|__GFP_FS
));
1077 if (!(lo
->lo_flags
& LO_FLAGS_READ_ONLY
) && file
->f_op
->fsync
)
1078 blk_queue_write_cache(lo
->lo_queue
, true, false);
1080 if (config
->block_size
)
1081 bsize
= config
->block_size
;
1082 else if ((lo
->lo_backing_file
->f_flags
& O_DIRECT
) && inode
->i_sb
->s_bdev
)
1083 /* In case of direct I/O, match underlying block size */
1084 bsize
= bdev_logical_block_size(inode
->i_sb
->s_bdev
);
1088 blk_queue_logical_block_size(lo
->lo_queue
, bsize
);
1089 blk_queue_physical_block_size(lo
->lo_queue
, bsize
);
1090 blk_queue_io_min(lo
->lo_queue
, bsize
);
1092 loop_config_discard(lo
);
1093 loop_update_rotational(lo
);
1094 loop_update_dio(lo
);
1095 loop_sysfs_init(lo
);
1097 size
= get_loop_size(lo
, file
);
1098 loop_set_size(lo
, size
);
1100 /* Order wrt reading lo_state in loop_validate_file(). */
1103 lo
->lo_state
= Lo_bound
;
1105 lo
->lo_flags
|= LO_FLAGS_PARTSCAN
;
1106 partscan
= lo
->lo_flags
& LO_FLAGS_PARTSCAN
;
1108 clear_bit(GD_SUPPRESS_PART_SCAN
, &lo
->lo_disk
->state
);
1110 /* enable and uncork uevent now that we are done */
1111 dev_set_uevent_suppress(disk_to_dev(lo
->lo_disk
), 0);
1113 loop_global_unlock(lo
, is_loop
);
1115 loop_reread_partitions(lo
);
1117 if (!(mode
& BLK_OPEN_EXCL
))
1118 bd_abort_claiming(bdev
, loop_configure
);
1123 loop_global_unlock(lo
, is_loop
);
1125 if (!(mode
& BLK_OPEN_EXCL
))
1126 bd_abort_claiming(bdev
, loop_configure
);
1129 /* This is safe: open() is still holding a reference. */
1130 module_put(THIS_MODULE
);
1134 static void __loop_clr_fd(struct loop_device
*lo
, bool release
)
1137 gfp_t gfp
= lo
->old_gfp_mask
;
1139 if (test_bit(QUEUE_FLAG_WC
, &lo
->lo_queue
->queue_flags
))
1140 blk_queue_write_cache(lo
->lo_queue
, false, false);
1143 * Freeze the request queue when unbinding on a live file descriptor and
1144 * thus an open device. When called from ->release we are guaranteed
1145 * that there is no I/O in progress already.
1148 blk_mq_freeze_queue(lo
->lo_queue
);
1150 spin_lock_irq(&lo
->lo_lock
);
1151 filp
= lo
->lo_backing_file
;
1152 lo
->lo_backing_file
= NULL
;
1153 spin_unlock_irq(&lo
->lo_lock
);
1155 lo
->lo_device
= NULL
;
1157 lo
->lo_sizelimit
= 0;
1158 memset(lo
->lo_file_name
, 0, LO_NAME_SIZE
);
1159 blk_queue_logical_block_size(lo
->lo_queue
, 512);
1160 blk_queue_physical_block_size(lo
->lo_queue
, 512);
1161 blk_queue_io_min(lo
->lo_queue
, 512);
1162 invalidate_disk(lo
->lo_disk
);
1163 loop_sysfs_exit(lo
);
1164 /* let user-space know about this change */
1165 kobject_uevent(&disk_to_dev(lo
->lo_disk
)->kobj
, KOBJ_CHANGE
);
1166 mapping_set_gfp_mask(filp
->f_mapping
, gfp
);
1167 /* This is safe: open() is still holding a reference. */
1168 module_put(THIS_MODULE
);
1170 blk_mq_unfreeze_queue(lo
->lo_queue
);
1172 disk_force_media_change(lo
->lo_disk
);
1174 if (lo
->lo_flags
& LO_FLAGS_PARTSCAN
) {
1178 * open_mutex has been held already in release path, so don't
1179 * acquire it if this function is called in such case.
1181 * If the reread partition isn't from release path, lo_refcnt
1182 * must be at least one and it can only become zero when the
1183 * current holder is released.
1186 mutex_lock(&lo
->lo_disk
->open_mutex
);
1187 err
= bdev_disk_changed(lo
->lo_disk
, false);
1189 mutex_unlock(&lo
->lo_disk
->open_mutex
);
1191 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1192 __func__
, lo
->lo_number
, err
);
1193 /* Device is gone, no point in returning error */
1197 * lo->lo_state is set to Lo_unbound here after above partscan has
1198 * finished. There cannot be anybody else entering __loop_clr_fd() as
1199 * Lo_rundown state protects us from all the other places trying to
1200 * change the 'lo' device.
1204 set_bit(GD_SUPPRESS_PART_SCAN
, &lo
->lo_disk
->state
);
1205 mutex_lock(&lo
->lo_mutex
);
1206 lo
->lo_state
= Lo_unbound
;
1207 mutex_unlock(&lo
->lo_mutex
);
1210 * Need not hold lo_mutex to fput backing file. Calling fput holding
1211 * lo_mutex triggers a circular lock dependency possibility warning as
1212 * fput can take open_mutex which is usually taken before lo_mutex.
1217 static int loop_clr_fd(struct loop_device
*lo
)
1222 * Since lo_ioctl() is called without locks held, it is possible that
1223 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1225 * Therefore, use global lock when setting Lo_rundown state in order to
1226 * make sure that loop_validate_file() will fail if the "struct file"
1227 * which loop_configure()/loop_change_fd() found via fget() was this
1230 err
= loop_global_lock_killable(lo
, true);
1233 if (lo
->lo_state
!= Lo_bound
) {
1234 loop_global_unlock(lo
, true);
1238 * If we've explicitly asked to tear down the loop device,
1239 * and it has an elevated reference count, set it for auto-teardown when
1240 * the last reference goes away. This stops $!~#$@ udev from
1241 * preventing teardown because it decided that it needs to run blkid on
1242 * the loopback device whenever they appear. xfstests is notorious for
1243 * failing tests because blkid via udev races with a losetup
1244 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1245 * command to fail with EBUSY.
1247 if (disk_openers(lo
->lo_disk
) > 1) {
1248 lo
->lo_flags
|= LO_FLAGS_AUTOCLEAR
;
1249 loop_global_unlock(lo
, true);
1252 lo
->lo_state
= Lo_rundown
;
1253 loop_global_unlock(lo
, true);
1255 __loop_clr_fd(lo
, false);
1260 loop_set_status(struct loop_device
*lo
, const struct loop_info64
*info
)
1264 bool partscan
= false;
1265 bool size_changed
= false;
1267 err
= mutex_lock_killable(&lo
->lo_mutex
);
1270 if (lo
->lo_state
!= Lo_bound
) {
1275 if (lo
->lo_offset
!= info
->lo_offset
||
1276 lo
->lo_sizelimit
!= info
->lo_sizelimit
) {
1277 size_changed
= true;
1278 sync_blockdev(lo
->lo_device
);
1279 invalidate_bdev(lo
->lo_device
);
1282 /* I/O need to be drained during transfer transition */
1283 blk_mq_freeze_queue(lo
->lo_queue
);
1285 prev_lo_flags
= lo
->lo_flags
;
1287 err
= loop_set_status_from_info(lo
, info
);
1291 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1292 lo
->lo_flags
&= LOOP_SET_STATUS_SETTABLE_FLAGS
;
1293 /* For those flags, use the previous values instead */
1294 lo
->lo_flags
|= prev_lo_flags
& ~LOOP_SET_STATUS_SETTABLE_FLAGS
;
1295 /* For flags that can't be cleared, use previous values too */
1296 lo
->lo_flags
|= prev_lo_flags
& ~LOOP_SET_STATUS_CLEARABLE_FLAGS
;
1299 loff_t new_size
= get_size(lo
->lo_offset
, lo
->lo_sizelimit
,
1300 lo
->lo_backing_file
);
1301 loop_set_size(lo
, new_size
);
1304 loop_config_discard(lo
);
1306 /* update dio if lo_offset or transfer is changed */
1307 __loop_update_dio(lo
, lo
->use_dio
);
1310 blk_mq_unfreeze_queue(lo
->lo_queue
);
1312 if (!err
&& (lo
->lo_flags
& LO_FLAGS_PARTSCAN
) &&
1313 !(prev_lo_flags
& LO_FLAGS_PARTSCAN
)) {
1314 clear_bit(GD_SUPPRESS_PART_SCAN
, &lo
->lo_disk
->state
);
1318 mutex_unlock(&lo
->lo_mutex
);
1320 loop_reread_partitions(lo
);
1326 loop_get_status(struct loop_device
*lo
, struct loop_info64
*info
)
1332 ret
= mutex_lock_killable(&lo
->lo_mutex
);
1335 if (lo
->lo_state
!= Lo_bound
) {
1336 mutex_unlock(&lo
->lo_mutex
);
1340 memset(info
, 0, sizeof(*info
));
1341 info
->lo_number
= lo
->lo_number
;
1342 info
->lo_offset
= lo
->lo_offset
;
1343 info
->lo_sizelimit
= lo
->lo_sizelimit
;
1344 info
->lo_flags
= lo
->lo_flags
;
1345 memcpy(info
->lo_file_name
, lo
->lo_file_name
, LO_NAME_SIZE
);
1347 /* Drop lo_mutex while we call into the filesystem. */
1348 path
= lo
->lo_backing_file
->f_path
;
1350 mutex_unlock(&lo
->lo_mutex
);
1351 ret
= vfs_getattr(&path
, &stat
, STATX_INO
, AT_STATX_SYNC_AS_STAT
);
1353 info
->lo_device
= huge_encode_dev(stat
.dev
);
1354 info
->lo_inode
= stat
.ino
;
1355 info
->lo_rdevice
= huge_encode_dev(stat
.rdev
);
1362 loop_info64_from_old(const struct loop_info
*info
, struct loop_info64
*info64
)
1364 memset(info64
, 0, sizeof(*info64
));
1365 info64
->lo_number
= info
->lo_number
;
1366 info64
->lo_device
= info
->lo_device
;
1367 info64
->lo_inode
= info
->lo_inode
;
1368 info64
->lo_rdevice
= info
->lo_rdevice
;
1369 info64
->lo_offset
= info
->lo_offset
;
1370 info64
->lo_sizelimit
= 0;
1371 info64
->lo_flags
= info
->lo_flags
;
1372 memcpy(info64
->lo_file_name
, info
->lo_name
, LO_NAME_SIZE
);
1376 loop_info64_to_old(const struct loop_info64
*info64
, struct loop_info
*info
)
1378 memset(info
, 0, sizeof(*info
));
1379 info
->lo_number
= info64
->lo_number
;
1380 info
->lo_device
= info64
->lo_device
;
1381 info
->lo_inode
= info64
->lo_inode
;
1382 info
->lo_rdevice
= info64
->lo_rdevice
;
1383 info
->lo_offset
= info64
->lo_offset
;
1384 info
->lo_flags
= info64
->lo_flags
;
1385 memcpy(info
->lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1387 /* error in case values were truncated */
1388 if (info
->lo_device
!= info64
->lo_device
||
1389 info
->lo_rdevice
!= info64
->lo_rdevice
||
1390 info
->lo_inode
!= info64
->lo_inode
||
1391 info
->lo_offset
!= info64
->lo_offset
)
1398 loop_set_status_old(struct loop_device
*lo
, const struct loop_info __user
*arg
)
1400 struct loop_info info
;
1401 struct loop_info64 info64
;
1403 if (copy_from_user(&info
, arg
, sizeof (struct loop_info
)))
1405 loop_info64_from_old(&info
, &info64
);
1406 return loop_set_status(lo
, &info64
);
1410 loop_set_status64(struct loop_device
*lo
, const struct loop_info64 __user
*arg
)
1412 struct loop_info64 info64
;
1414 if (copy_from_user(&info64
, arg
, sizeof (struct loop_info64
)))
1416 return loop_set_status(lo
, &info64
);
1420 loop_get_status_old(struct loop_device
*lo
, struct loop_info __user
*arg
) {
1421 struct loop_info info
;
1422 struct loop_info64 info64
;
1427 err
= loop_get_status(lo
, &info64
);
1429 err
= loop_info64_to_old(&info64
, &info
);
1430 if (!err
&& copy_to_user(arg
, &info
, sizeof(info
)))
1437 loop_get_status64(struct loop_device
*lo
, struct loop_info64 __user
*arg
) {
1438 struct loop_info64 info64
;
1443 err
= loop_get_status(lo
, &info64
);
1444 if (!err
&& copy_to_user(arg
, &info64
, sizeof(info64
)))
1450 static int loop_set_capacity(struct loop_device
*lo
)
1454 if (unlikely(lo
->lo_state
!= Lo_bound
))
1457 size
= get_loop_size(lo
, lo
->lo_backing_file
);
1458 loop_set_size(lo
, size
);
1463 static int loop_set_dio(struct loop_device
*lo
, unsigned long arg
)
1466 if (lo
->lo_state
!= Lo_bound
)
1469 __loop_update_dio(lo
, !!arg
);
1470 if (lo
->use_dio
== !!arg
)
1477 static int loop_set_block_size(struct loop_device
*lo
, unsigned long arg
)
1481 if (lo
->lo_state
!= Lo_bound
)
1484 err
= blk_validate_block_size(arg
);
1488 if (lo
->lo_queue
->limits
.logical_block_size
== arg
)
1491 sync_blockdev(lo
->lo_device
);
1492 invalidate_bdev(lo
->lo_device
);
1494 blk_mq_freeze_queue(lo
->lo_queue
);
1495 blk_queue_logical_block_size(lo
->lo_queue
, arg
);
1496 blk_queue_physical_block_size(lo
->lo_queue
, arg
);
1497 blk_queue_io_min(lo
->lo_queue
, arg
);
1498 loop_update_dio(lo
);
1499 blk_mq_unfreeze_queue(lo
->lo_queue
);
1504 static int lo_simple_ioctl(struct loop_device
*lo
, unsigned int cmd
,
1509 err
= mutex_lock_killable(&lo
->lo_mutex
);
1513 case LOOP_SET_CAPACITY
:
1514 err
= loop_set_capacity(lo
);
1516 case LOOP_SET_DIRECT_IO
:
1517 err
= loop_set_dio(lo
, arg
);
1519 case LOOP_SET_BLOCK_SIZE
:
1520 err
= loop_set_block_size(lo
, arg
);
1525 mutex_unlock(&lo
->lo_mutex
);
1529 static int lo_ioctl(struct block_device
*bdev
, blk_mode_t mode
,
1530 unsigned int cmd
, unsigned long arg
)
1532 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1533 void __user
*argp
= (void __user
*) arg
;
1539 * Legacy case - pass in a zeroed out struct loop_config with
1540 * only the file descriptor set , which corresponds with the
1541 * default parameters we'd have used otherwise.
1543 struct loop_config config
;
1545 memset(&config
, 0, sizeof(config
));
1548 return loop_configure(lo
, mode
, bdev
, &config
);
1550 case LOOP_CONFIGURE
: {
1551 struct loop_config config
;
1553 if (copy_from_user(&config
, argp
, sizeof(config
)))
1556 return loop_configure(lo
, mode
, bdev
, &config
);
1558 case LOOP_CHANGE_FD
:
1559 return loop_change_fd(lo
, bdev
, arg
);
1561 return loop_clr_fd(lo
);
1562 case LOOP_SET_STATUS
:
1564 if ((mode
& BLK_OPEN_WRITE
) || capable(CAP_SYS_ADMIN
))
1565 err
= loop_set_status_old(lo
, argp
);
1567 case LOOP_GET_STATUS
:
1568 return loop_get_status_old(lo
, argp
);
1569 case LOOP_SET_STATUS64
:
1571 if ((mode
& BLK_OPEN_WRITE
) || capable(CAP_SYS_ADMIN
))
1572 err
= loop_set_status64(lo
, argp
);
1574 case LOOP_GET_STATUS64
:
1575 return loop_get_status64(lo
, argp
);
1576 case LOOP_SET_CAPACITY
:
1577 case LOOP_SET_DIRECT_IO
:
1578 case LOOP_SET_BLOCK_SIZE
:
1579 if (!(mode
& BLK_OPEN_WRITE
) && !capable(CAP_SYS_ADMIN
))
1583 err
= lo_simple_ioctl(lo
, cmd
, arg
);
1590 #ifdef CONFIG_COMPAT
1591 struct compat_loop_info
{
1592 compat_int_t lo_number
; /* ioctl r/o */
1593 compat_dev_t lo_device
; /* ioctl r/o */
1594 compat_ulong_t lo_inode
; /* ioctl r/o */
1595 compat_dev_t lo_rdevice
; /* ioctl r/o */
1596 compat_int_t lo_offset
;
1597 compat_int_t lo_encrypt_type
; /* obsolete, ignored */
1598 compat_int_t lo_encrypt_key_size
; /* ioctl w/o */
1599 compat_int_t lo_flags
; /* ioctl r/o */
1600 char lo_name
[LO_NAME_SIZE
];
1601 unsigned char lo_encrypt_key
[LO_KEY_SIZE
]; /* ioctl w/o */
1602 compat_ulong_t lo_init
[2];
1607 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1608 * - noinlined to reduce stack space usage in main part of driver
1611 loop_info64_from_compat(const struct compat_loop_info __user
*arg
,
1612 struct loop_info64
*info64
)
1614 struct compat_loop_info info
;
1616 if (copy_from_user(&info
, arg
, sizeof(info
)))
1619 memset(info64
, 0, sizeof(*info64
));
1620 info64
->lo_number
= info
.lo_number
;
1621 info64
->lo_device
= info
.lo_device
;
1622 info64
->lo_inode
= info
.lo_inode
;
1623 info64
->lo_rdevice
= info
.lo_rdevice
;
1624 info64
->lo_offset
= info
.lo_offset
;
1625 info64
->lo_sizelimit
= 0;
1626 info64
->lo_flags
= info
.lo_flags
;
1627 memcpy(info64
->lo_file_name
, info
.lo_name
, LO_NAME_SIZE
);
1632 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1633 * - noinlined to reduce stack space usage in main part of driver
1636 loop_info64_to_compat(const struct loop_info64
*info64
,
1637 struct compat_loop_info __user
*arg
)
1639 struct compat_loop_info info
;
1641 memset(&info
, 0, sizeof(info
));
1642 info
.lo_number
= info64
->lo_number
;
1643 info
.lo_device
= info64
->lo_device
;
1644 info
.lo_inode
= info64
->lo_inode
;
1645 info
.lo_rdevice
= info64
->lo_rdevice
;
1646 info
.lo_offset
= info64
->lo_offset
;
1647 info
.lo_flags
= info64
->lo_flags
;
1648 memcpy(info
.lo_name
, info64
->lo_file_name
, LO_NAME_SIZE
);
1650 /* error in case values were truncated */
1651 if (info
.lo_device
!= info64
->lo_device
||
1652 info
.lo_rdevice
!= info64
->lo_rdevice
||
1653 info
.lo_inode
!= info64
->lo_inode
||
1654 info
.lo_offset
!= info64
->lo_offset
)
1657 if (copy_to_user(arg
, &info
, sizeof(info
)))
1663 loop_set_status_compat(struct loop_device
*lo
,
1664 const struct compat_loop_info __user
*arg
)
1666 struct loop_info64 info64
;
1669 ret
= loop_info64_from_compat(arg
, &info64
);
1672 return loop_set_status(lo
, &info64
);
1676 loop_get_status_compat(struct loop_device
*lo
,
1677 struct compat_loop_info __user
*arg
)
1679 struct loop_info64 info64
;
1684 err
= loop_get_status(lo
, &info64
);
1686 err
= loop_info64_to_compat(&info64
, arg
);
1690 static int lo_compat_ioctl(struct block_device
*bdev
, blk_mode_t mode
,
1691 unsigned int cmd
, unsigned long arg
)
1693 struct loop_device
*lo
= bdev
->bd_disk
->private_data
;
1697 case LOOP_SET_STATUS
:
1698 err
= loop_set_status_compat(lo
,
1699 (const struct compat_loop_info __user
*)arg
);
1701 case LOOP_GET_STATUS
:
1702 err
= loop_get_status_compat(lo
,
1703 (struct compat_loop_info __user
*)arg
);
1705 case LOOP_SET_CAPACITY
:
1707 case LOOP_GET_STATUS64
:
1708 case LOOP_SET_STATUS64
:
1709 case LOOP_CONFIGURE
:
1710 arg
= (unsigned long) compat_ptr(arg
);
1713 case LOOP_CHANGE_FD
:
1714 case LOOP_SET_BLOCK_SIZE
:
1715 case LOOP_SET_DIRECT_IO
:
1716 err
= lo_ioctl(bdev
, mode
, cmd
, arg
);
1726 static void lo_release(struct gendisk
*disk
)
1728 struct loop_device
*lo
= disk
->private_data
;
1730 if (disk_openers(disk
) > 0)
1733 mutex_lock(&lo
->lo_mutex
);
1734 if (lo
->lo_state
== Lo_bound
&& (lo
->lo_flags
& LO_FLAGS_AUTOCLEAR
)) {
1735 lo
->lo_state
= Lo_rundown
;
1736 mutex_unlock(&lo
->lo_mutex
);
1738 * In autoclear mode, stop the loop thread
1739 * and remove configuration after last close.
1741 __loop_clr_fd(lo
, true);
1744 mutex_unlock(&lo
->lo_mutex
);
1747 static void lo_free_disk(struct gendisk
*disk
)
1749 struct loop_device
*lo
= disk
->private_data
;
1752 destroy_workqueue(lo
->workqueue
);
1753 loop_free_idle_workers(lo
, true);
1754 timer_shutdown_sync(&lo
->timer
);
1755 mutex_destroy(&lo
->lo_mutex
);
1759 static const struct block_device_operations lo_fops
= {
1760 .owner
= THIS_MODULE
,
1761 .release
= lo_release
,
1763 #ifdef CONFIG_COMPAT
1764 .compat_ioctl
= lo_compat_ioctl
,
1766 .free_disk
= lo_free_disk
,
1770 * And now the modules code and kernel interface.
1774 * If max_loop is specified, create that many devices upfront.
1775 * This also becomes a hard limit. If max_loop is not specified,
1776 * the default isn't a hard limit (as before commit 85c50197716c
1777 * changed the default value from 0 for max_loop=0 reasons), just
1778 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1779 * init time. Loop devices can be requested on-demand with the
1780 * /dev/loop-control interface, or be instantiated by accessing
1781 * a 'dead' device node.
1783 static int max_loop
= CONFIG_BLK_DEV_LOOP_MIN_COUNT
;
1785 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1786 static bool max_loop_specified
;
1788 static int max_loop_param_set_int(const char *val
,
1789 const struct kernel_param
*kp
)
1793 ret
= param_set_int(val
, kp
);
1797 max_loop_specified
= true;
1801 static const struct kernel_param_ops max_loop_param_ops
= {
1802 .set
= max_loop_param_set_int
,
1803 .get
= param_get_int
,
1806 module_param_cb(max_loop
, &max_loop_param_ops
, &max_loop
, 0444);
1807 MODULE_PARM_DESC(max_loop
, "Maximum number of loop devices");
1809 module_param(max_loop
, int, 0444);
1810 MODULE_PARM_DESC(max_loop
, "Initial number of loop devices");
1813 module_param(max_part
, int, 0444);
1814 MODULE_PARM_DESC(max_part
, "Maximum number of partitions per loop device");
1816 static int hw_queue_depth
= LOOP_DEFAULT_HW_Q_DEPTH
;
1818 static int loop_set_hw_queue_depth(const char *s
, const struct kernel_param
*p
)
1822 ret
= kstrtoint(s
, 0, &qd
);
1827 hw_queue_depth
= qd
;
1831 static const struct kernel_param_ops loop_hw_qdepth_param_ops
= {
1832 .set
= loop_set_hw_queue_depth
,
1833 .get
= param_get_int
,
1836 device_param_cb(hw_queue_depth
, &loop_hw_qdepth_param_ops
, &hw_queue_depth
, 0444);
1837 MODULE_PARM_DESC(hw_queue_depth
, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH
));
1839 MODULE_LICENSE("GPL");
1840 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR
);
1842 static blk_status_t
loop_queue_rq(struct blk_mq_hw_ctx
*hctx
,
1843 const struct blk_mq_queue_data
*bd
)
1845 struct request
*rq
= bd
->rq
;
1846 struct loop_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1847 struct loop_device
*lo
= rq
->q
->queuedata
;
1849 blk_mq_start_request(rq
);
1851 if (lo
->lo_state
!= Lo_bound
)
1852 return BLK_STS_IOERR
;
1854 switch (req_op(rq
)) {
1856 case REQ_OP_DISCARD
:
1857 case REQ_OP_WRITE_ZEROES
:
1858 cmd
->use_aio
= false;
1861 cmd
->use_aio
= lo
->use_dio
;
1865 /* always use the first bio's css */
1866 cmd
->blkcg_css
= NULL
;
1867 cmd
->memcg_css
= NULL
;
1868 #ifdef CONFIG_BLK_CGROUP
1870 cmd
->blkcg_css
= bio_blkcg_css(rq
->bio
);
1872 if (cmd
->blkcg_css
) {
1874 cgroup_get_e_css(cmd
->blkcg_css
->cgroup
,
1875 &memory_cgrp_subsys
);
1880 loop_queue_work(lo
, cmd
);
1885 static void loop_handle_cmd(struct loop_cmd
*cmd
)
1887 struct cgroup_subsys_state
*cmd_blkcg_css
= cmd
->blkcg_css
;
1888 struct cgroup_subsys_state
*cmd_memcg_css
= cmd
->memcg_css
;
1889 struct request
*rq
= blk_mq_rq_from_pdu(cmd
);
1890 const bool write
= op_is_write(req_op(rq
));
1891 struct loop_device
*lo
= rq
->q
->queuedata
;
1893 struct mem_cgroup
*old_memcg
= NULL
;
1894 const bool use_aio
= cmd
->use_aio
;
1896 if (write
&& (lo
->lo_flags
& LO_FLAGS_READ_ONLY
)) {
1902 kthread_associate_blkcg(cmd_blkcg_css
);
1904 old_memcg
= set_active_memcg(
1905 mem_cgroup_from_css(cmd_memcg_css
));
1908 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1909 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1910 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1911 * not yet been completed.
1913 ret
= do_req_filebacked(lo
, rq
);
1916 kthread_associate_blkcg(NULL
);
1918 if (cmd_memcg_css
) {
1919 set_active_memcg(old_memcg
);
1920 css_put(cmd_memcg_css
);
1923 /* complete non-aio request */
1924 if (!use_aio
|| ret
) {
1925 if (ret
== -EOPNOTSUPP
)
1928 cmd
->ret
= ret
? -EIO
: 0;
1929 if (likely(!blk_should_fake_timeout(rq
->q
)))
1930 blk_mq_complete_request(rq
);
1934 static void loop_process_work(struct loop_worker
*worker
,
1935 struct list_head
*cmd_list
, struct loop_device
*lo
)
1937 int orig_flags
= current
->flags
;
1938 struct loop_cmd
*cmd
;
1940 current
->flags
|= PF_LOCAL_THROTTLE
| PF_MEMALLOC_NOIO
;
1941 spin_lock_irq(&lo
->lo_work_lock
);
1942 while (!list_empty(cmd_list
)) {
1944 cmd_list
->next
, struct loop_cmd
, list_entry
);
1945 list_del(cmd_list
->next
);
1946 spin_unlock_irq(&lo
->lo_work_lock
);
1948 loop_handle_cmd(cmd
);
1951 spin_lock_irq(&lo
->lo_work_lock
);
1955 * We only add to the idle list if there are no pending cmds
1956 * *and* the worker will not run again which ensures that it
1957 * is safe to free any worker on the idle list
1959 if (worker
&& !work_pending(&worker
->work
)) {
1960 worker
->last_ran_at
= jiffies
;
1961 list_add_tail(&worker
->idle_list
, &lo
->idle_worker_list
);
1964 spin_unlock_irq(&lo
->lo_work_lock
);
1965 current
->flags
= orig_flags
;
1968 static void loop_workfn(struct work_struct
*work
)
1970 struct loop_worker
*worker
=
1971 container_of(work
, struct loop_worker
, work
);
1972 loop_process_work(worker
, &worker
->cmd_list
, worker
->lo
);
1975 static void loop_rootcg_workfn(struct work_struct
*work
)
1977 struct loop_device
*lo
=
1978 container_of(work
, struct loop_device
, rootcg_work
);
1979 loop_process_work(NULL
, &lo
->rootcg_cmd_list
, lo
);
1982 static const struct blk_mq_ops loop_mq_ops
= {
1983 .queue_rq
= loop_queue_rq
,
1984 .complete
= lo_complete_rq
,
1987 static int loop_add(int i
)
1989 struct loop_device
*lo
;
1990 struct gendisk
*disk
;
1994 lo
= kzalloc(sizeof(*lo
), GFP_KERNEL
);
1997 lo
->worker_tree
= RB_ROOT
;
1998 INIT_LIST_HEAD(&lo
->idle_worker_list
);
1999 timer_setup(&lo
->timer
, loop_free_idle_workers_timer
, TIMER_DEFERRABLE
);
2000 lo
->lo_state
= Lo_unbound
;
2002 err
= mutex_lock_killable(&loop_ctl_mutex
);
2006 /* allocate id, if @id >= 0, we're requesting that specific id */
2008 err
= idr_alloc(&loop_index_idr
, lo
, i
, i
+ 1, GFP_KERNEL
);
2012 err
= idr_alloc(&loop_index_idr
, lo
, 0, 0, GFP_KERNEL
);
2014 mutex_unlock(&loop_ctl_mutex
);
2019 lo
->tag_set
.ops
= &loop_mq_ops
;
2020 lo
->tag_set
.nr_hw_queues
= 1;
2021 lo
->tag_set
.queue_depth
= hw_queue_depth
;
2022 lo
->tag_set
.numa_node
= NUMA_NO_NODE
;
2023 lo
->tag_set
.cmd_size
= sizeof(struct loop_cmd
);
2024 lo
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_STACKING
|
2025 BLK_MQ_F_NO_SCHED_BY_DEFAULT
;
2026 lo
->tag_set
.driver_data
= lo
;
2028 err
= blk_mq_alloc_tag_set(&lo
->tag_set
);
2032 disk
= lo
->lo_disk
= blk_mq_alloc_disk(&lo
->tag_set
, lo
);
2034 err
= PTR_ERR(disk
);
2035 goto out_cleanup_tags
;
2037 lo
->lo_queue
= lo
->lo_disk
->queue
;
2039 blk_queue_max_hw_sectors(lo
->lo_queue
, BLK_DEF_MAX_SECTORS
);
2042 * By default, we do buffer IO, so it doesn't make sense to enable
2043 * merge because the I/O submitted to backing file is handled page by
2044 * page. For directio mode, merge does help to dispatch bigger request
2045 * to underlayer disk. We will enable merge once directio is enabled.
2047 blk_queue_flag_set(QUEUE_FLAG_NOMERGES
, lo
->lo_queue
);
2050 * Disable partition scanning by default. The in-kernel partition
2051 * scanning can be requested individually per-device during its
2052 * setup. Userspace can always add and remove partitions from all
2053 * devices. The needed partition minors are allocated from the
2054 * extended minor space, the main loop device numbers will continue
2055 * to match the loop minors, regardless of the number of partitions
2058 * If max_part is given, partition scanning is globally enabled for
2059 * all loop devices. The minors for the main loop devices will be
2060 * multiples of max_part.
2062 * Note: Global-for-all-devices, set-only-at-init, read-only module
2063 * parameteters like 'max_loop' and 'max_part' make things needlessly
2064 * complicated, are too static, inflexible and may surprise
2065 * userspace tools. Parameters like this in general should be avoided.
2068 set_bit(GD_SUPPRESS_PART_SCAN
, &disk
->state
);
2069 mutex_init(&lo
->lo_mutex
);
2071 spin_lock_init(&lo
->lo_lock
);
2072 spin_lock_init(&lo
->lo_work_lock
);
2073 INIT_WORK(&lo
->rootcg_work
, loop_rootcg_workfn
);
2074 INIT_LIST_HEAD(&lo
->rootcg_cmd_list
);
2075 disk
->major
= LOOP_MAJOR
;
2076 disk
->first_minor
= i
<< part_shift
;
2077 disk
->minors
= 1 << part_shift
;
2078 disk
->fops
= &lo_fops
;
2079 disk
->private_data
= lo
;
2080 disk
->queue
= lo
->lo_queue
;
2081 disk
->events
= DISK_EVENT_MEDIA_CHANGE
;
2082 disk
->event_flags
= DISK_EVENT_FLAG_UEVENT
;
2083 sprintf(disk
->disk_name
, "loop%d", i
);
2084 /* Make this loop device reachable from pathname. */
2085 err
= add_disk(disk
);
2087 goto out_cleanup_disk
;
2089 /* Show this loop device. */
2090 mutex_lock(&loop_ctl_mutex
);
2091 lo
->idr_visible
= true;
2092 mutex_unlock(&loop_ctl_mutex
);
2099 blk_mq_free_tag_set(&lo
->tag_set
);
2101 mutex_lock(&loop_ctl_mutex
);
2102 idr_remove(&loop_index_idr
, i
);
2103 mutex_unlock(&loop_ctl_mutex
);
2110 static void loop_remove(struct loop_device
*lo
)
2112 /* Make this loop device unreachable from pathname. */
2113 del_gendisk(lo
->lo_disk
);
2114 blk_mq_free_tag_set(&lo
->tag_set
);
2116 mutex_lock(&loop_ctl_mutex
);
2117 idr_remove(&loop_index_idr
, lo
->lo_number
);
2118 mutex_unlock(&loop_ctl_mutex
);
2120 put_disk(lo
->lo_disk
);
2123 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2124 static void loop_probe(dev_t dev
)
2126 int idx
= MINOR(dev
) >> part_shift
;
2128 if (max_loop_specified
&& max_loop
&& idx
>= max_loop
)
2133 #define loop_probe NULL
2134 #endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2136 static int loop_control_remove(int idx
)
2138 struct loop_device
*lo
;
2142 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2146 /* Hide this loop device for serialization. */
2147 ret
= mutex_lock_killable(&loop_ctl_mutex
);
2150 lo
= idr_find(&loop_index_idr
, idx
);
2151 if (!lo
|| !lo
->idr_visible
)
2154 lo
->idr_visible
= false;
2155 mutex_unlock(&loop_ctl_mutex
);
2159 /* Check whether this loop device can be removed. */
2160 ret
= mutex_lock_killable(&lo
->lo_mutex
);
2163 if (lo
->lo_state
!= Lo_unbound
|| disk_openers(lo
->lo_disk
) > 0) {
2164 mutex_unlock(&lo
->lo_mutex
);
2168 /* Mark this loop device as no more bound, but not quite unbound yet */
2169 lo
->lo_state
= Lo_deleting
;
2170 mutex_unlock(&lo
->lo_mutex
);
2176 /* Show this loop device again. */
2177 mutex_lock(&loop_ctl_mutex
);
2178 lo
->idr_visible
= true;
2179 mutex_unlock(&loop_ctl_mutex
);
2183 static int loop_control_get_free(int idx
)
2185 struct loop_device
*lo
;
2188 ret
= mutex_lock_killable(&loop_ctl_mutex
);
2191 idr_for_each_entry(&loop_index_idr
, lo
, id
) {
2192 /* Hitting a race results in creating a new loop device which is harmless. */
2193 if (lo
->idr_visible
&& data_race(lo
->lo_state
) == Lo_unbound
)
2196 mutex_unlock(&loop_ctl_mutex
);
2197 return loop_add(-1);
2199 mutex_unlock(&loop_ctl_mutex
);
2203 static long loop_control_ioctl(struct file
*file
, unsigned int cmd
,
2208 return loop_add(parm
);
2209 case LOOP_CTL_REMOVE
:
2210 return loop_control_remove(parm
);
2211 case LOOP_CTL_GET_FREE
:
2212 return loop_control_get_free(parm
);
2218 static const struct file_operations loop_ctl_fops
= {
2219 .open
= nonseekable_open
,
2220 .unlocked_ioctl
= loop_control_ioctl
,
2221 .compat_ioctl
= loop_control_ioctl
,
2222 .owner
= THIS_MODULE
,
2223 .llseek
= noop_llseek
,
2226 static struct miscdevice loop_misc
= {
2227 .minor
= LOOP_CTRL_MINOR
,
2228 .name
= "loop-control",
2229 .fops
= &loop_ctl_fops
,
2232 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR
);
2233 MODULE_ALIAS("devname:loop-control");
2235 static int __init
loop_init(void)
2242 part_shift
= fls(max_part
);
2245 * Adjust max_part according to part_shift as it is exported
2246 * to user space so that user can decide correct minor number
2247 * if [s]he want to create more devices.
2249 * Note that -1 is required because partition 0 is reserved
2250 * for the whole disk.
2252 max_part
= (1UL << part_shift
) - 1;
2255 if ((1UL << part_shift
) > DISK_MAX_PARTS
) {
2260 if (max_loop
> 1UL << (MINORBITS
- part_shift
)) {
2265 err
= misc_register(&loop_misc
);
2270 if (__register_blkdev(LOOP_MAJOR
, "loop", loop_probe
)) {
2275 /* pre-create number of devices given by config or max_loop */
2276 for (i
= 0; i
< max_loop
; i
++)
2279 printk(KERN_INFO
"loop: module loaded\n");
2283 misc_deregister(&loop_misc
);
2288 static void __exit
loop_exit(void)
2290 struct loop_device
*lo
;
2293 unregister_blkdev(LOOP_MAJOR
, "loop");
2294 misc_deregister(&loop_misc
);
2297 * There is no need to use loop_ctl_mutex here, for nobody else can
2298 * access loop_index_idr when this module is unloading (unless forced
2299 * module unloading is requested). If this is not a clean unloading,
2300 * we have no means to avoid kernel crash.
2302 idr_for_each_entry(&loop_index_idr
, lo
, id
)
2305 idr_destroy(&loop_index_idr
);
2308 module_init(loop_init
);
2309 module_exit(loop_exit
);
2312 static int __init
max_loop_setup(char *str
)
2314 max_loop
= simple_strtol(str
, NULL
, 0);
2315 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2316 max_loop_specified
= true;
2321 __setup("max_loop=", max_loop_setup
);