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 file_start_write(file);
249 bw = vfs_iter_write(file, &i, ppos, 0);
250 file_end_write(file);
252 if (likely(bw == bvec->bv_len))
255 printk_ratelimited(KERN_ERR
256 "loop: Write error at byte offset %llu, length %i.\n",
257 (unsigned long long)*ppos, bvec->bv_len);
263 static int lo_write_simple(struct loop_device *lo, struct request *rq,
267 struct req_iterator iter;
270 rq_for_each_segment(bvec, rq, iter) {
271 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
280 static int lo_read_simple(struct loop_device *lo, struct request *rq,
284 struct req_iterator iter;
288 rq_for_each_segment(bvec, rq, iter) {
289 iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len);
290 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
294 flush_dcache_page(bvec.bv_page);
296 if (len != bvec.bv_len) {
299 __rq_for_each_bio(bio, rq)
309 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
313 * We use fallocate to manipulate the space mappings used by the image
314 * a.k.a. discard/zerorange.
316 struct file *file = lo->lo_backing_file;
319 mode |= FALLOC_FL_KEEP_SIZE;
321 if (!bdev_max_discard_sectors(lo->lo_device))
324 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
325 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
330 static int lo_req_flush(struct loop_device *lo, struct request *rq)
332 int ret = vfs_fsync(lo->lo_backing_file, 0);
333 if (unlikely(ret && ret != -EINVAL))
339 static void lo_complete_rq(struct request *rq)
341 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
342 blk_status_t ret = BLK_STS_OK;
344 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
345 req_op(rq) != REQ_OP_READ) {
347 ret = errno_to_blk_status(cmd->ret);
352 * Short READ - if we got some data, advance our request and
353 * retry it. If we got no data, end the rest with EIO.
356 blk_update_request(rq, BLK_STS_OK, cmd->ret);
358 blk_mq_requeue_request(rq, true);
361 struct bio *bio = rq->bio;
370 blk_mq_end_request(rq, ret);
374 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
376 struct request *rq = blk_mq_rq_from_pdu(cmd);
378 if (!atomic_dec_and_test(&cmd->ref))
382 if (likely(!blk_should_fake_timeout(rq->q)))
383 blk_mq_complete_request(rq);
386 static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
388 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
391 lo_rw_aio_do_completion(cmd);
394 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
397 struct iov_iter iter;
398 struct req_iterator rq_iter;
399 struct bio_vec *bvec;
400 struct request *rq = blk_mq_rq_from_pdu(cmd);
401 struct bio *bio = rq->bio;
402 struct file *file = lo->lo_backing_file;
408 rq_for_each_bvec(tmp, rq, rq_iter)
411 if (rq->bio != rq->biotail) {
413 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
420 * The bios of the request may be started from the middle of
421 * the 'bvec' because of bio splitting, so we can't directly
422 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
423 * API will take care of all details for us.
425 rq_for_each_bvec(tmp, rq, rq_iter) {
433 * Same here, this bio may be started from the middle of the
434 * 'bvec' because of bio splitting, so offset from the bvec
435 * must be passed to iov iterator
437 offset = bio->bi_iter.bi_bvec_done;
438 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
440 atomic_set(&cmd->ref, 2);
442 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
443 iter.iov_offset = offset;
445 cmd->iocb.ki_pos = pos;
446 cmd->iocb.ki_filp = file;
447 cmd->iocb.ki_complete = lo_rw_aio_complete;
448 cmd->iocb.ki_flags = IOCB_DIRECT;
449 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
451 if (rw == ITER_SOURCE)
452 ret = call_write_iter(file, &cmd->iocb, &iter);
454 ret = call_read_iter(file, &cmd->iocb, &iter);
456 lo_rw_aio_do_completion(cmd);
458 if (ret != -EIOCBQUEUED)
459 lo_rw_aio_complete(&cmd->iocb, ret);
463 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
465 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
466 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
469 * lo_write_simple and lo_read_simple should have been covered
470 * by io submit style function like lo_rw_aio(), one blocker
471 * is that lo_read_simple() need to call flush_dcache_page after
472 * the page is written from kernel, and it isn't easy to handle
473 * this in io submit style function which submits all segments
474 * of the req at one time. And direct read IO doesn't need to
475 * run flush_dcache_page().
477 switch (req_op(rq)) {
479 return lo_req_flush(lo, rq);
480 case REQ_OP_WRITE_ZEROES:
482 * If the caller doesn't want deallocation, call zeroout to
483 * write zeroes the range. Otherwise, punch them out.
485 return lo_fallocate(lo, rq, pos,
486 (rq->cmd_flags & REQ_NOUNMAP) ?
487 FALLOC_FL_ZERO_RANGE :
488 FALLOC_FL_PUNCH_HOLE);
490 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
493 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
495 return lo_write_simple(lo, rq, pos);
498 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
500 return lo_read_simple(lo, rq, pos);
507 static inline void loop_update_dio(struct loop_device *lo)
509 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
513 static void loop_reread_partitions(struct loop_device *lo)
517 mutex_lock(&lo->lo_disk->open_mutex);
518 rc = bdev_disk_changed(lo->lo_disk, false);
519 mutex_unlock(&lo->lo_disk->open_mutex);
521 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
522 __func__, lo->lo_number, lo->lo_file_name, rc);
525 static inline int is_loop_device(struct file *file)
527 struct inode *i = file->f_mapping->host;
529 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
532 static int loop_validate_file(struct file *file, struct block_device *bdev)
534 struct inode *inode = file->f_mapping->host;
535 struct file *f = file;
537 /* Avoid recursion */
538 while (is_loop_device(f)) {
539 struct loop_device *l;
541 lockdep_assert_held(&loop_validate_mutex);
542 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
545 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
546 if (l->lo_state != Lo_bound)
548 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
550 f = l->lo_backing_file;
552 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
558 * loop_change_fd switched the backing store of a loopback device to
559 * a new file. This is useful for operating system installers to free up
560 * the original file and in High Availability environments to switch to
561 * an alternative location for the content in case of server meltdown.
562 * This can only work if the loop device is used read-only, and if the
563 * new backing store is the same size and type as the old backing store.
565 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
568 struct file *file = fget(arg);
569 struct file *old_file;
577 /* suppress uevents while reconfiguring the device */
578 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
580 is_loop = is_loop_device(file);
581 error = loop_global_lock_killable(lo, is_loop);
585 if (lo->lo_state != Lo_bound)
588 /* the loop device has to be read-only */
590 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
593 error = loop_validate_file(file, bdev);
597 old_file = lo->lo_backing_file;
601 /* size of the new backing store needs to be the same */
602 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
606 disk_force_media_change(lo->lo_disk);
607 blk_mq_freeze_queue(lo->lo_queue);
608 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
609 lo->lo_backing_file = file;
610 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
611 mapping_set_gfp_mask(file->f_mapping,
612 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
614 blk_mq_unfreeze_queue(lo->lo_queue);
615 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
616 loop_global_unlock(lo, is_loop);
619 * Flush loop_validate_file() before fput(), for l->lo_backing_file
620 * might be pointing at old_file which might be the last reference.
623 mutex_lock(&loop_validate_mutex);
624 mutex_unlock(&loop_validate_mutex);
627 * We must drop file reference outside of lo_mutex as dropping
628 * the file ref can take open_mutex which creates circular locking
633 loop_reread_partitions(lo);
637 /* enable and uncork uevent now that we are done */
638 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
642 loop_global_unlock(lo, is_loop);
648 /* loop sysfs attributes */
650 static ssize_t loop_attr_show(struct device *dev, char *page,
651 ssize_t (*callback)(struct loop_device *, char *))
653 struct gendisk *disk = dev_to_disk(dev);
654 struct loop_device *lo = disk->private_data;
656 return callback(lo, page);
659 #define LOOP_ATTR_RO(_name) \
660 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
661 static ssize_t loop_attr_do_show_##_name(struct device *d, \
662 struct device_attribute *attr, char *b) \
664 return loop_attr_show(d, b, loop_attr_##_name##_show); \
666 static struct device_attribute loop_attr_##_name = \
667 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
669 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
674 spin_lock_irq(&lo->lo_lock);
675 if (lo->lo_backing_file)
676 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
677 spin_unlock_irq(&lo->lo_lock);
679 if (IS_ERR_OR_NULL(p))
683 memmove(buf, p, ret);
691 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
693 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
696 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
698 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
701 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
703 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
705 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
708 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
710 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
712 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
715 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
717 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
719 return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
722 LOOP_ATTR_RO(backing_file);
723 LOOP_ATTR_RO(offset);
724 LOOP_ATTR_RO(sizelimit);
725 LOOP_ATTR_RO(autoclear);
726 LOOP_ATTR_RO(partscan);
729 static struct attribute *loop_attrs[] = {
730 &loop_attr_backing_file.attr,
731 &loop_attr_offset.attr,
732 &loop_attr_sizelimit.attr,
733 &loop_attr_autoclear.attr,
734 &loop_attr_partscan.attr,
739 static struct attribute_group loop_attribute_group = {
744 static void loop_sysfs_init(struct loop_device *lo)
746 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
747 &loop_attribute_group);
750 static void loop_sysfs_exit(struct loop_device *lo)
752 if (lo->sysfs_inited)
753 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
754 &loop_attribute_group);
757 static void loop_config_discard(struct loop_device *lo)
759 struct file *file = lo->lo_backing_file;
760 struct inode *inode = file->f_mapping->host;
761 struct request_queue *q = lo->lo_queue;
762 u32 granularity, max_discard_sectors;
765 * If the backing device is a block device, mirror its zeroing
766 * capability. Set the discard sectors to the block device's zeroing
767 * capabilities because loop discards result in blkdev_issue_zeroout(),
768 * not blkdev_issue_discard(). This maintains consistent behavior with
769 * file-backed loop devices: discarded regions read back as zero.
771 if (S_ISBLK(inode->i_mode)) {
772 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
774 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
775 granularity = bdev_discard_granularity(I_BDEV(inode)) ?:
776 queue_physical_block_size(backingq);
779 * We use punch hole to reclaim the free space used by the
780 * image a.k.a. discard.
782 } else if (!file->f_op->fallocate) {
783 max_discard_sectors = 0;
789 max_discard_sectors = UINT_MAX >> 9;
790 if (!vfs_statfs(&file->f_path, &sbuf))
791 granularity = sbuf.f_bsize;
793 max_discard_sectors = 0;
796 if (max_discard_sectors) {
797 q->limits.discard_granularity = granularity;
798 blk_queue_max_discard_sectors(q, max_discard_sectors);
799 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
801 q->limits.discard_granularity = 0;
802 blk_queue_max_discard_sectors(q, 0);
803 blk_queue_max_write_zeroes_sectors(q, 0);
808 struct rb_node rb_node;
809 struct work_struct work;
810 struct list_head cmd_list;
811 struct list_head idle_list;
812 struct loop_device *lo;
813 struct cgroup_subsys_state *blkcg_css;
814 unsigned long last_ran_at;
817 static void loop_workfn(struct work_struct *work);
819 #ifdef CONFIG_BLK_CGROUP
820 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
822 return !css || css == blkcg_root_css;
825 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
831 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
833 struct rb_node **node, *parent = NULL;
834 struct loop_worker *cur_worker, *worker = NULL;
835 struct work_struct *work;
836 struct list_head *cmd_list;
838 spin_lock_irq(&lo->lo_work_lock);
840 if (queue_on_root_worker(cmd->blkcg_css))
843 node = &lo->worker_tree.rb_node;
847 cur_worker = container_of(*node, struct loop_worker, rb_node);
848 if (cur_worker->blkcg_css == cmd->blkcg_css) {
851 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
852 node = &(*node)->rb_left;
854 node = &(*node)->rb_right;
860 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
862 * In the event we cannot allocate a worker, just queue on the
863 * rootcg worker and issue the I/O as the rootcg
866 cmd->blkcg_css = NULL;
868 css_put(cmd->memcg_css);
869 cmd->memcg_css = NULL;
873 worker->blkcg_css = cmd->blkcg_css;
874 css_get(worker->blkcg_css);
875 INIT_WORK(&worker->work, loop_workfn);
876 INIT_LIST_HEAD(&worker->cmd_list);
877 INIT_LIST_HEAD(&worker->idle_list);
879 rb_link_node(&worker->rb_node, parent, node);
880 rb_insert_color(&worker->rb_node, &lo->worker_tree);
884 * We need to remove from the idle list here while
885 * holding the lock so that the idle timer doesn't
888 if (!list_empty(&worker->idle_list))
889 list_del_init(&worker->idle_list);
890 work = &worker->work;
891 cmd_list = &worker->cmd_list;
893 work = &lo->rootcg_work;
894 cmd_list = &lo->rootcg_cmd_list;
896 list_add_tail(&cmd->list_entry, cmd_list);
897 queue_work(lo->workqueue, work);
898 spin_unlock_irq(&lo->lo_work_lock);
901 static void loop_set_timer(struct loop_device *lo)
903 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
906 static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
908 struct loop_worker *pos, *worker;
910 spin_lock_irq(&lo->lo_work_lock);
911 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
914 time_is_after_jiffies(worker->last_ran_at +
915 LOOP_IDLE_WORKER_TIMEOUT))
917 list_del(&worker->idle_list);
918 rb_erase(&worker->rb_node, &lo->worker_tree);
919 css_put(worker->blkcg_css);
922 if (!list_empty(&lo->idle_worker_list))
924 spin_unlock_irq(&lo->lo_work_lock);
927 static void loop_free_idle_workers_timer(struct timer_list *timer)
929 struct loop_device *lo = container_of(timer, struct loop_device, timer);
931 return loop_free_idle_workers(lo, false);
934 static void loop_update_rotational(struct loop_device *lo)
936 struct file *file = lo->lo_backing_file;
937 struct inode *file_inode = file->f_mapping->host;
938 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
939 struct request_queue *q = lo->lo_queue;
942 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
944 nonrot = bdev_nonrot(file_bdev);
947 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
949 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
953 * loop_set_status_from_info - configure device from loop_info
954 * @lo: struct loop_device to configure
955 * @info: struct loop_info64 to configure the device with
957 * Configures the loop device parameters according to the passed
958 * in loop_info64 configuration.
961 loop_set_status_from_info(struct loop_device *lo,
962 const struct loop_info64 *info)
964 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
967 switch (info->lo_encrypt_type) {
971 pr_warn("support for the xor transformation has been removed.\n");
973 case LO_CRYPT_CRYPTOAPI:
974 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
980 /* Avoid assigning overflow values */
981 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
984 lo->lo_offset = info->lo_offset;
985 lo->lo_sizelimit = info->lo_sizelimit;
987 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
988 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
989 lo->lo_flags = info->lo_flags;
993 static int loop_configure(struct loop_device *lo, blk_mode_t mode,
994 struct block_device *bdev,
995 const struct loop_config *config)
997 struct file *file = fget(config->fd);
999 struct address_space *mapping;
1003 unsigned short bsize;
1008 is_loop = is_loop_device(file);
1010 /* This is safe, since we have a reference from open(). */
1011 __module_get(THIS_MODULE);
1014 * If we don't hold exclusive handle for the device, upgrade to it
1015 * here to avoid changing device under exclusive owner.
1017 if (!(mode & BLK_OPEN_EXCL)) {
1018 error = bd_prepare_to_claim(bdev, loop_configure, NULL);
1023 error = loop_global_lock_killable(lo, is_loop);
1028 if (lo->lo_state != Lo_unbound)
1031 error = loop_validate_file(file, bdev);
1035 mapping = file->f_mapping;
1036 inode = mapping->host;
1038 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1043 if (config->block_size) {
1044 error = blk_validate_block_size(config->block_size);
1049 error = loop_set_status_from_info(lo, &config->info);
1053 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1054 !file->f_op->write_iter)
1055 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1057 if (!lo->workqueue) {
1058 lo->workqueue = alloc_workqueue("loop%d",
1059 WQ_UNBOUND | WQ_FREEZABLE,
1061 if (!lo->workqueue) {
1067 /* suppress uevents while reconfiguring the device */
1068 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1070 disk_force_media_change(lo->lo_disk);
1071 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1073 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1074 lo->lo_device = bdev;
1075 lo->lo_backing_file = file;
1076 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1077 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1079 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1080 blk_queue_write_cache(lo->lo_queue, true, false);
1082 if (config->block_size)
1083 bsize = config->block_size;
1084 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1085 /* In case of direct I/O, match underlying block size */
1086 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1090 blk_queue_logical_block_size(lo->lo_queue, bsize);
1091 blk_queue_physical_block_size(lo->lo_queue, bsize);
1092 blk_queue_io_min(lo->lo_queue, bsize);
1094 loop_config_discard(lo);
1095 loop_update_rotational(lo);
1096 loop_update_dio(lo);
1097 loop_sysfs_init(lo);
1099 size = get_loop_size(lo, file);
1100 loop_set_size(lo, size);
1102 /* Order wrt reading lo_state in loop_validate_file(). */
1105 lo->lo_state = Lo_bound;
1107 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1108 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1110 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1112 /* enable and uncork uevent now that we are done */
1113 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1115 loop_global_unlock(lo, is_loop);
1117 loop_reread_partitions(lo);
1119 if (!(mode & BLK_OPEN_EXCL))
1120 bd_abort_claiming(bdev, loop_configure);
1125 loop_global_unlock(lo, is_loop);
1127 if (!(mode & BLK_OPEN_EXCL))
1128 bd_abort_claiming(bdev, loop_configure);
1131 /* This is safe: open() is still holding a reference. */
1132 module_put(THIS_MODULE);
1136 static void __loop_clr_fd(struct loop_device *lo, bool release)
1139 gfp_t gfp = lo->old_gfp_mask;
1141 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1142 blk_queue_write_cache(lo->lo_queue, false, false);
1145 * Freeze the request queue when unbinding on a live file descriptor and
1146 * thus an open device. When called from ->release we are guaranteed
1147 * that there is no I/O in progress already.
1150 blk_mq_freeze_queue(lo->lo_queue);
1152 spin_lock_irq(&lo->lo_lock);
1153 filp = lo->lo_backing_file;
1154 lo->lo_backing_file = NULL;
1155 spin_unlock_irq(&lo->lo_lock);
1157 lo->lo_device = NULL;
1159 lo->lo_sizelimit = 0;
1160 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1161 blk_queue_logical_block_size(lo->lo_queue, 512);
1162 blk_queue_physical_block_size(lo->lo_queue, 512);
1163 blk_queue_io_min(lo->lo_queue, 512);
1164 invalidate_disk(lo->lo_disk);
1165 loop_sysfs_exit(lo);
1166 /* let user-space know about this change */
1167 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1168 mapping_set_gfp_mask(filp->f_mapping, gfp);
1169 /* This is safe: open() is still holding a reference. */
1170 module_put(THIS_MODULE);
1172 blk_mq_unfreeze_queue(lo->lo_queue);
1174 disk_force_media_change(lo->lo_disk);
1176 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1180 * open_mutex has been held already in release path, so don't
1181 * acquire it if this function is called in such case.
1183 * If the reread partition isn't from release path, lo_refcnt
1184 * must be at least one and it can only become zero when the
1185 * current holder is released.
1188 mutex_lock(&lo->lo_disk->open_mutex);
1189 err = bdev_disk_changed(lo->lo_disk, false);
1191 mutex_unlock(&lo->lo_disk->open_mutex);
1193 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1194 __func__, lo->lo_number, err);
1195 /* Device is gone, no point in returning error */
1199 * lo->lo_state is set to Lo_unbound here after above partscan has
1200 * finished. There cannot be anybody else entering __loop_clr_fd() as
1201 * Lo_rundown state protects us from all the other places trying to
1202 * change the 'lo' device.
1206 set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1207 mutex_lock(&lo->lo_mutex);
1208 lo->lo_state = Lo_unbound;
1209 mutex_unlock(&lo->lo_mutex);
1212 * Need not hold lo_mutex to fput backing file. Calling fput holding
1213 * lo_mutex triggers a circular lock dependency possibility warning as
1214 * fput can take open_mutex which is usually taken before lo_mutex.
1219 static int loop_clr_fd(struct loop_device *lo)
1224 * Since lo_ioctl() is called without locks held, it is possible that
1225 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1227 * Therefore, use global lock when setting Lo_rundown state in order to
1228 * make sure that loop_validate_file() will fail if the "struct file"
1229 * which loop_configure()/loop_change_fd() found via fget() was this
1232 err = loop_global_lock_killable(lo, true);
1235 if (lo->lo_state != Lo_bound) {
1236 loop_global_unlock(lo, true);
1240 * If we've explicitly asked to tear down the loop device,
1241 * and it has an elevated reference count, set it for auto-teardown when
1242 * the last reference goes away. This stops $!~#$@ udev from
1243 * preventing teardown because it decided that it needs to run blkid on
1244 * the loopback device whenever they appear. xfstests is notorious for
1245 * failing tests because blkid via udev races with a losetup
1246 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1247 * command to fail with EBUSY.
1249 if (disk_openers(lo->lo_disk) > 1) {
1250 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1251 loop_global_unlock(lo, true);
1254 lo->lo_state = Lo_rundown;
1255 loop_global_unlock(lo, true);
1257 __loop_clr_fd(lo, false);
1262 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1266 bool partscan = false;
1267 bool size_changed = false;
1269 err = mutex_lock_killable(&lo->lo_mutex);
1272 if (lo->lo_state != Lo_bound) {
1277 if (lo->lo_offset != info->lo_offset ||
1278 lo->lo_sizelimit != info->lo_sizelimit) {
1279 size_changed = true;
1280 sync_blockdev(lo->lo_device);
1281 invalidate_bdev(lo->lo_device);
1284 /* I/O need to be drained during transfer transition */
1285 blk_mq_freeze_queue(lo->lo_queue);
1287 prev_lo_flags = lo->lo_flags;
1289 err = loop_set_status_from_info(lo, info);
1293 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1294 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1295 /* For those flags, use the previous values instead */
1296 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1297 /* For flags that can't be cleared, use previous values too */
1298 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1301 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1302 lo->lo_backing_file);
1303 loop_set_size(lo, new_size);
1306 loop_config_discard(lo);
1308 /* update dio if lo_offset or transfer is changed */
1309 __loop_update_dio(lo, lo->use_dio);
1312 blk_mq_unfreeze_queue(lo->lo_queue);
1314 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1315 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1316 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1320 mutex_unlock(&lo->lo_mutex);
1322 loop_reread_partitions(lo);
1328 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1334 ret = mutex_lock_killable(&lo->lo_mutex);
1337 if (lo->lo_state != Lo_bound) {
1338 mutex_unlock(&lo->lo_mutex);
1342 memset(info, 0, sizeof(*info));
1343 info->lo_number = lo->lo_number;
1344 info->lo_offset = lo->lo_offset;
1345 info->lo_sizelimit = lo->lo_sizelimit;
1346 info->lo_flags = lo->lo_flags;
1347 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1349 /* Drop lo_mutex while we call into the filesystem. */
1350 path = lo->lo_backing_file->f_path;
1352 mutex_unlock(&lo->lo_mutex);
1353 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1355 info->lo_device = huge_encode_dev(stat.dev);
1356 info->lo_inode = stat.ino;
1357 info->lo_rdevice = huge_encode_dev(stat.rdev);
1364 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1366 memset(info64, 0, sizeof(*info64));
1367 info64->lo_number = info->lo_number;
1368 info64->lo_device = info->lo_device;
1369 info64->lo_inode = info->lo_inode;
1370 info64->lo_rdevice = info->lo_rdevice;
1371 info64->lo_offset = info->lo_offset;
1372 info64->lo_sizelimit = 0;
1373 info64->lo_flags = info->lo_flags;
1374 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1378 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1380 memset(info, 0, sizeof(*info));
1381 info->lo_number = info64->lo_number;
1382 info->lo_device = info64->lo_device;
1383 info->lo_inode = info64->lo_inode;
1384 info->lo_rdevice = info64->lo_rdevice;
1385 info->lo_offset = info64->lo_offset;
1386 info->lo_flags = info64->lo_flags;
1387 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1389 /* error in case values were truncated */
1390 if (info->lo_device != info64->lo_device ||
1391 info->lo_rdevice != info64->lo_rdevice ||
1392 info->lo_inode != info64->lo_inode ||
1393 info->lo_offset != info64->lo_offset)
1400 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1402 struct loop_info info;
1403 struct loop_info64 info64;
1405 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1407 loop_info64_from_old(&info, &info64);
1408 return loop_set_status(lo, &info64);
1412 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1414 struct loop_info64 info64;
1416 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1418 return loop_set_status(lo, &info64);
1422 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1423 struct loop_info info;
1424 struct loop_info64 info64;
1429 err = loop_get_status(lo, &info64);
1431 err = loop_info64_to_old(&info64, &info);
1432 if (!err && copy_to_user(arg, &info, sizeof(info)))
1439 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1440 struct loop_info64 info64;
1445 err = loop_get_status(lo, &info64);
1446 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1452 static int loop_set_capacity(struct loop_device *lo)
1456 if (unlikely(lo->lo_state != Lo_bound))
1459 size = get_loop_size(lo, lo->lo_backing_file);
1460 loop_set_size(lo, size);
1465 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1468 if (lo->lo_state != Lo_bound)
1471 __loop_update_dio(lo, !!arg);
1472 if (lo->use_dio == !!arg)
1479 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1483 if (lo->lo_state != Lo_bound)
1486 err = blk_validate_block_size(arg);
1490 if (lo->lo_queue->limits.logical_block_size == arg)
1493 sync_blockdev(lo->lo_device);
1494 invalidate_bdev(lo->lo_device);
1496 blk_mq_freeze_queue(lo->lo_queue);
1497 blk_queue_logical_block_size(lo->lo_queue, arg);
1498 blk_queue_physical_block_size(lo->lo_queue, arg);
1499 blk_queue_io_min(lo->lo_queue, arg);
1500 loop_update_dio(lo);
1501 blk_mq_unfreeze_queue(lo->lo_queue);
1506 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1511 err = mutex_lock_killable(&lo->lo_mutex);
1515 case LOOP_SET_CAPACITY:
1516 err = loop_set_capacity(lo);
1518 case LOOP_SET_DIRECT_IO:
1519 err = loop_set_dio(lo, arg);
1521 case LOOP_SET_BLOCK_SIZE:
1522 err = loop_set_block_size(lo, arg);
1527 mutex_unlock(&lo->lo_mutex);
1531 static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1532 unsigned int cmd, unsigned long arg)
1534 struct loop_device *lo = bdev->bd_disk->private_data;
1535 void __user *argp = (void __user *) arg;
1541 * Legacy case - pass in a zeroed out struct loop_config with
1542 * only the file descriptor set , which corresponds with the
1543 * default parameters we'd have used otherwise.
1545 struct loop_config config;
1547 memset(&config, 0, sizeof(config));
1550 return loop_configure(lo, mode, bdev, &config);
1552 case LOOP_CONFIGURE: {
1553 struct loop_config config;
1555 if (copy_from_user(&config, argp, sizeof(config)))
1558 return loop_configure(lo, mode, bdev, &config);
1560 case LOOP_CHANGE_FD:
1561 return loop_change_fd(lo, bdev, arg);
1563 return loop_clr_fd(lo);
1564 case LOOP_SET_STATUS:
1566 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1567 err = loop_set_status_old(lo, argp);
1569 case LOOP_GET_STATUS:
1570 return loop_get_status_old(lo, argp);
1571 case LOOP_SET_STATUS64:
1573 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1574 err = loop_set_status64(lo, argp);
1576 case LOOP_GET_STATUS64:
1577 return loop_get_status64(lo, argp);
1578 case LOOP_SET_CAPACITY:
1579 case LOOP_SET_DIRECT_IO:
1580 case LOOP_SET_BLOCK_SIZE:
1581 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1585 err = lo_simple_ioctl(lo, cmd, arg);
1592 #ifdef CONFIG_COMPAT
1593 struct compat_loop_info {
1594 compat_int_t lo_number; /* ioctl r/o */
1595 compat_dev_t lo_device; /* ioctl r/o */
1596 compat_ulong_t lo_inode; /* ioctl r/o */
1597 compat_dev_t lo_rdevice; /* ioctl r/o */
1598 compat_int_t lo_offset;
1599 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1600 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1601 compat_int_t lo_flags; /* ioctl r/o */
1602 char lo_name[LO_NAME_SIZE];
1603 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1604 compat_ulong_t lo_init[2];
1609 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1610 * - noinlined to reduce stack space usage in main part of driver
1613 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1614 struct loop_info64 *info64)
1616 struct compat_loop_info info;
1618 if (copy_from_user(&info, arg, sizeof(info)))
1621 memset(info64, 0, sizeof(*info64));
1622 info64->lo_number = info.lo_number;
1623 info64->lo_device = info.lo_device;
1624 info64->lo_inode = info.lo_inode;
1625 info64->lo_rdevice = info.lo_rdevice;
1626 info64->lo_offset = info.lo_offset;
1627 info64->lo_sizelimit = 0;
1628 info64->lo_flags = info.lo_flags;
1629 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1634 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1635 * - noinlined to reduce stack space usage in main part of driver
1638 loop_info64_to_compat(const struct loop_info64 *info64,
1639 struct compat_loop_info __user *arg)
1641 struct compat_loop_info info;
1643 memset(&info, 0, sizeof(info));
1644 info.lo_number = info64->lo_number;
1645 info.lo_device = info64->lo_device;
1646 info.lo_inode = info64->lo_inode;
1647 info.lo_rdevice = info64->lo_rdevice;
1648 info.lo_offset = info64->lo_offset;
1649 info.lo_flags = info64->lo_flags;
1650 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1652 /* error in case values were truncated */
1653 if (info.lo_device != info64->lo_device ||
1654 info.lo_rdevice != info64->lo_rdevice ||
1655 info.lo_inode != info64->lo_inode ||
1656 info.lo_offset != info64->lo_offset)
1659 if (copy_to_user(arg, &info, sizeof(info)))
1665 loop_set_status_compat(struct loop_device *lo,
1666 const struct compat_loop_info __user *arg)
1668 struct loop_info64 info64;
1671 ret = loop_info64_from_compat(arg, &info64);
1674 return loop_set_status(lo, &info64);
1678 loop_get_status_compat(struct loop_device *lo,
1679 struct compat_loop_info __user *arg)
1681 struct loop_info64 info64;
1686 err = loop_get_status(lo, &info64);
1688 err = loop_info64_to_compat(&info64, arg);
1692 static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1693 unsigned int cmd, unsigned long arg)
1695 struct loop_device *lo = bdev->bd_disk->private_data;
1699 case LOOP_SET_STATUS:
1700 err = loop_set_status_compat(lo,
1701 (const struct compat_loop_info __user *)arg);
1703 case LOOP_GET_STATUS:
1704 err = loop_get_status_compat(lo,
1705 (struct compat_loop_info __user *)arg);
1707 case LOOP_SET_CAPACITY:
1709 case LOOP_GET_STATUS64:
1710 case LOOP_SET_STATUS64:
1711 case LOOP_CONFIGURE:
1712 arg = (unsigned long) compat_ptr(arg);
1715 case LOOP_CHANGE_FD:
1716 case LOOP_SET_BLOCK_SIZE:
1717 case LOOP_SET_DIRECT_IO:
1718 err = lo_ioctl(bdev, mode, cmd, arg);
1728 static void lo_release(struct gendisk *disk)
1730 struct loop_device *lo = disk->private_data;
1732 if (disk_openers(disk) > 0)
1735 mutex_lock(&lo->lo_mutex);
1736 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) {
1737 lo->lo_state = Lo_rundown;
1738 mutex_unlock(&lo->lo_mutex);
1740 * In autoclear mode, stop the loop thread
1741 * and remove configuration after last close.
1743 __loop_clr_fd(lo, true);
1746 mutex_unlock(&lo->lo_mutex);
1749 static void lo_free_disk(struct gendisk *disk)
1751 struct loop_device *lo = disk->private_data;
1754 destroy_workqueue(lo->workqueue);
1755 loop_free_idle_workers(lo, true);
1756 timer_shutdown_sync(&lo->timer);
1757 mutex_destroy(&lo->lo_mutex);
1761 static const struct block_device_operations lo_fops = {
1762 .owner = THIS_MODULE,
1763 .release = lo_release,
1765 #ifdef CONFIG_COMPAT
1766 .compat_ioctl = lo_compat_ioctl,
1768 .free_disk = lo_free_disk,
1772 * And now the modules code and kernel interface.
1776 * If max_loop is specified, create that many devices upfront.
1777 * This also becomes a hard limit. If max_loop is not specified,
1778 * the default isn't a hard limit (as before commit 85c50197716c
1779 * changed the default value from 0 for max_loop=0 reasons), just
1780 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1781 * init time. Loop devices can be requested on-demand with the
1782 * /dev/loop-control interface, or be instantiated by accessing
1783 * a 'dead' device node.
1785 static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1787 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1788 static bool max_loop_specified;
1790 static int max_loop_param_set_int(const char *val,
1791 const struct kernel_param *kp)
1795 ret = param_set_int(val, kp);
1799 max_loop_specified = true;
1803 static const struct kernel_param_ops max_loop_param_ops = {
1804 .set = max_loop_param_set_int,
1805 .get = param_get_int,
1808 module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1809 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1811 module_param(max_loop, int, 0444);
1812 MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1815 module_param(max_part, int, 0444);
1816 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1818 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1820 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1824 ret = kstrtoint(s, 0, &qd);
1829 hw_queue_depth = qd;
1833 static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1834 .set = loop_set_hw_queue_depth,
1835 .get = param_get_int,
1838 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1839 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1841 MODULE_LICENSE("GPL");
1842 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1844 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1845 const struct blk_mq_queue_data *bd)
1847 struct request *rq = bd->rq;
1848 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1849 struct loop_device *lo = rq->q->queuedata;
1851 blk_mq_start_request(rq);
1853 if (lo->lo_state != Lo_bound)
1854 return BLK_STS_IOERR;
1856 switch (req_op(rq)) {
1858 case REQ_OP_DISCARD:
1859 case REQ_OP_WRITE_ZEROES:
1860 cmd->use_aio = false;
1863 cmd->use_aio = lo->use_dio;
1867 /* always use the first bio's css */
1868 cmd->blkcg_css = NULL;
1869 cmd->memcg_css = NULL;
1870 #ifdef CONFIG_BLK_CGROUP
1872 cmd->blkcg_css = bio_blkcg_css(rq->bio);
1874 if (cmd->blkcg_css) {
1876 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1877 &memory_cgrp_subsys);
1882 loop_queue_work(lo, cmd);
1887 static void loop_handle_cmd(struct loop_cmd *cmd)
1889 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1890 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1891 struct request *rq = blk_mq_rq_from_pdu(cmd);
1892 const bool write = op_is_write(req_op(rq));
1893 struct loop_device *lo = rq->q->queuedata;
1895 struct mem_cgroup *old_memcg = NULL;
1896 const bool use_aio = cmd->use_aio;
1898 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1904 kthread_associate_blkcg(cmd_blkcg_css);
1906 old_memcg = set_active_memcg(
1907 mem_cgroup_from_css(cmd_memcg_css));
1910 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1911 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1912 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1913 * not yet been completed.
1915 ret = do_req_filebacked(lo, rq);
1918 kthread_associate_blkcg(NULL);
1920 if (cmd_memcg_css) {
1921 set_active_memcg(old_memcg);
1922 css_put(cmd_memcg_css);
1925 /* complete non-aio request */
1926 if (!use_aio || ret) {
1927 if (ret == -EOPNOTSUPP)
1930 cmd->ret = ret ? -EIO : 0;
1931 if (likely(!blk_should_fake_timeout(rq->q)))
1932 blk_mq_complete_request(rq);
1936 static void loop_process_work(struct loop_worker *worker,
1937 struct list_head *cmd_list, struct loop_device *lo)
1939 int orig_flags = current->flags;
1940 struct loop_cmd *cmd;
1942 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1943 spin_lock_irq(&lo->lo_work_lock);
1944 while (!list_empty(cmd_list)) {
1946 cmd_list->next, struct loop_cmd, list_entry);
1947 list_del(cmd_list->next);
1948 spin_unlock_irq(&lo->lo_work_lock);
1950 loop_handle_cmd(cmd);
1953 spin_lock_irq(&lo->lo_work_lock);
1957 * We only add to the idle list if there are no pending cmds
1958 * *and* the worker will not run again which ensures that it
1959 * is safe to free any worker on the idle list
1961 if (worker && !work_pending(&worker->work)) {
1962 worker->last_ran_at = jiffies;
1963 list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1966 spin_unlock_irq(&lo->lo_work_lock);
1967 current->flags = orig_flags;
1970 static void loop_workfn(struct work_struct *work)
1972 struct loop_worker *worker =
1973 container_of(work, struct loop_worker, work);
1974 loop_process_work(worker, &worker->cmd_list, worker->lo);
1977 static void loop_rootcg_workfn(struct work_struct *work)
1979 struct loop_device *lo =
1980 container_of(work, struct loop_device, rootcg_work);
1981 loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1984 static const struct blk_mq_ops loop_mq_ops = {
1985 .queue_rq = loop_queue_rq,
1986 .complete = lo_complete_rq,
1989 static int loop_add(int i)
1991 struct loop_device *lo;
1992 struct gendisk *disk;
1996 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1999 lo->worker_tree = RB_ROOT;
2000 INIT_LIST_HEAD(&lo->idle_worker_list);
2001 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
2002 lo->lo_state = Lo_unbound;
2004 err = mutex_lock_killable(&loop_ctl_mutex);
2008 /* allocate id, if @id >= 0, we're requesting that specific id */
2010 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2014 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2016 mutex_unlock(&loop_ctl_mutex);
2021 lo->tag_set.ops = &loop_mq_ops;
2022 lo->tag_set.nr_hw_queues = 1;
2023 lo->tag_set.queue_depth = hw_queue_depth;
2024 lo->tag_set.numa_node = NUMA_NO_NODE;
2025 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2026 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
2027 BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2028 lo->tag_set.driver_data = lo;
2030 err = blk_mq_alloc_tag_set(&lo->tag_set);
2034 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
2036 err = PTR_ERR(disk);
2037 goto out_cleanup_tags;
2039 lo->lo_queue = lo->lo_disk->queue;
2041 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2044 * By default, we do buffer IO, so it doesn't make sense to enable
2045 * merge because the I/O submitted to backing file is handled page by
2046 * page. For directio mode, merge does help to dispatch bigger request
2047 * to underlayer disk. We will enable merge once directio is enabled.
2049 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2052 * Disable partition scanning by default. The in-kernel partition
2053 * scanning can be requested individually per-device during its
2054 * setup. Userspace can always add and remove partitions from all
2055 * devices. The needed partition minors are allocated from the
2056 * extended minor space, the main loop device numbers will continue
2057 * to match the loop minors, regardless of the number of partitions
2060 * If max_part is given, partition scanning is globally enabled for
2061 * all loop devices. The minors for the main loop devices will be
2062 * multiples of max_part.
2064 * Note: Global-for-all-devices, set-only-at-init, read-only module
2065 * parameteters like 'max_loop' and 'max_part' make things needlessly
2066 * complicated, are too static, inflexible and may surprise
2067 * userspace tools. Parameters like this in general should be avoided.
2070 set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2071 mutex_init(&lo->lo_mutex);
2073 spin_lock_init(&lo->lo_lock);
2074 spin_lock_init(&lo->lo_work_lock);
2075 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2076 INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2077 disk->major = LOOP_MAJOR;
2078 disk->first_minor = i << part_shift;
2079 disk->minors = 1 << part_shift;
2080 disk->fops = &lo_fops;
2081 disk->private_data = lo;
2082 disk->queue = lo->lo_queue;
2083 disk->events = DISK_EVENT_MEDIA_CHANGE;
2084 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2085 sprintf(disk->disk_name, "loop%d", i);
2086 /* Make this loop device reachable from pathname. */
2087 err = add_disk(disk);
2089 goto out_cleanup_disk;
2091 /* Show this loop device. */
2092 mutex_lock(&loop_ctl_mutex);
2093 lo->idr_visible = true;
2094 mutex_unlock(&loop_ctl_mutex);
2101 blk_mq_free_tag_set(&lo->tag_set);
2103 mutex_lock(&loop_ctl_mutex);
2104 idr_remove(&loop_index_idr, i);
2105 mutex_unlock(&loop_ctl_mutex);
2112 static void loop_remove(struct loop_device *lo)
2114 /* Make this loop device unreachable from pathname. */
2115 del_gendisk(lo->lo_disk);
2116 blk_mq_free_tag_set(&lo->tag_set);
2118 mutex_lock(&loop_ctl_mutex);
2119 idr_remove(&loop_index_idr, lo->lo_number);
2120 mutex_unlock(&loop_ctl_mutex);
2122 put_disk(lo->lo_disk);
2125 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2126 static void loop_probe(dev_t dev)
2128 int idx = MINOR(dev) >> part_shift;
2130 if (max_loop_specified && max_loop && idx >= max_loop)
2135 #define loop_probe NULL
2136 #endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2138 static int loop_control_remove(int idx)
2140 struct loop_device *lo;
2144 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2148 /* Hide this loop device for serialization. */
2149 ret = mutex_lock_killable(&loop_ctl_mutex);
2152 lo = idr_find(&loop_index_idr, idx);
2153 if (!lo || !lo->idr_visible)
2156 lo->idr_visible = false;
2157 mutex_unlock(&loop_ctl_mutex);
2161 /* Check whether this loop device can be removed. */
2162 ret = mutex_lock_killable(&lo->lo_mutex);
2165 if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2166 mutex_unlock(&lo->lo_mutex);
2170 /* Mark this loop device as no more bound, but not quite unbound yet */
2171 lo->lo_state = Lo_deleting;
2172 mutex_unlock(&lo->lo_mutex);
2178 /* Show this loop device again. */
2179 mutex_lock(&loop_ctl_mutex);
2180 lo->idr_visible = true;
2181 mutex_unlock(&loop_ctl_mutex);
2185 static int loop_control_get_free(int idx)
2187 struct loop_device *lo;
2190 ret = mutex_lock_killable(&loop_ctl_mutex);
2193 idr_for_each_entry(&loop_index_idr, lo, id) {
2194 /* Hitting a race results in creating a new loop device which is harmless. */
2195 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2198 mutex_unlock(&loop_ctl_mutex);
2199 return loop_add(-1);
2201 mutex_unlock(&loop_ctl_mutex);
2205 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2210 return loop_add(parm);
2211 case LOOP_CTL_REMOVE:
2212 return loop_control_remove(parm);
2213 case LOOP_CTL_GET_FREE:
2214 return loop_control_get_free(parm);
2220 static const struct file_operations loop_ctl_fops = {
2221 .open = nonseekable_open,
2222 .unlocked_ioctl = loop_control_ioctl,
2223 .compat_ioctl = loop_control_ioctl,
2224 .owner = THIS_MODULE,
2225 .llseek = noop_llseek,
2228 static struct miscdevice loop_misc = {
2229 .minor = LOOP_CTRL_MINOR,
2230 .name = "loop-control",
2231 .fops = &loop_ctl_fops,
2234 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2235 MODULE_ALIAS("devname:loop-control");
2237 static int __init loop_init(void)
2244 part_shift = fls(max_part);
2247 * Adjust max_part according to part_shift as it is exported
2248 * to user space so that user can decide correct minor number
2249 * if [s]he want to create more devices.
2251 * Note that -1 is required because partition 0 is reserved
2252 * for the whole disk.
2254 max_part = (1UL << part_shift) - 1;
2257 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2262 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2267 err = misc_register(&loop_misc);
2272 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2277 /* pre-create number of devices given by config or max_loop */
2278 for (i = 0; i < max_loop; i++)
2281 printk(KERN_INFO "loop: module loaded\n");
2285 misc_deregister(&loop_misc);
2290 static void __exit loop_exit(void)
2292 struct loop_device *lo;
2295 unregister_blkdev(LOOP_MAJOR, "loop");
2296 misc_deregister(&loop_misc);
2299 * There is no need to use loop_ctl_mutex here, for nobody else can
2300 * access loop_index_idr when this module is unloading (unless forced
2301 * module unloading is requested). If this is not a clean unloading,
2302 * we have no means to avoid kernel crash.
2304 idr_for_each_entry(&loop_index_idr, lo, id)
2307 idr_destroy(&loop_index_idr);
2310 module_init(loop_init);
2311 module_exit(loop_exit);
2314 static int __init max_loop_setup(char *str)
2316 max_loop = simple_strtol(str, NULL, 0);
2317 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2318 max_loop_specified = true;
2323 __setup("max_loop=", max_loop_setup);