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 */
53 char lo_file_name[LO_NAME_SIZE];
55 struct file *lo_backing_file;
56 unsigned int lo_min_dio_size;
57 struct block_device *lo_device;
63 spinlock_t lo_work_lock;
64 struct workqueue_struct *workqueue;
65 struct work_struct rootcg_work;
66 struct list_head rootcg_cmd_list;
67 struct list_head idle_worker_list;
68 struct rb_root worker_tree;
69 struct timer_list timer;
72 struct request_queue *lo_queue;
73 struct blk_mq_tag_set tag_set;
74 struct gendisk *lo_disk;
75 struct mutex lo_mutex;
80 struct list_head list_entry;
81 bool use_aio; /* use AIO interface to handle I/O */
82 atomic_t ref; /* only for aio */
86 struct cgroup_subsys_state *blkcg_css;
87 struct cgroup_subsys_state *memcg_css;
90 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
91 #define LOOP_DEFAULT_HW_Q_DEPTH 128
93 static DEFINE_IDR(loop_index_idr);
94 static DEFINE_MUTEX(loop_ctl_mutex);
95 static DEFINE_MUTEX(loop_validate_mutex);
98 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
100 * @lo: struct loop_device
101 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
103 * Returns 0 on success, -EINTR otherwise.
105 * Since loop_validate_file() traverses on other "struct loop_device" if
106 * is_loop_device() is true, we need a global lock for serializing concurrent
107 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
109 static int loop_global_lock_killable(struct loop_device *lo, bool global)
114 err = mutex_lock_killable(&loop_validate_mutex);
118 err = mutex_lock_killable(&lo->lo_mutex);
120 mutex_unlock(&loop_validate_mutex);
125 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
127 * @lo: struct loop_device
128 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
130 static void loop_global_unlock(struct loop_device *lo, bool global)
132 mutex_unlock(&lo->lo_mutex);
134 mutex_unlock(&loop_validate_mutex);
138 static int part_shift;
140 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
144 /* Compute loopsize in bytes */
145 loopsize = i_size_read(file->f_mapping->host);
148 /* offset is beyond i_size, weird but possible */
152 if (sizelimit > 0 && sizelimit < loopsize)
153 loopsize = sizelimit;
155 * Unfortunately, if we want to do I/O on the device,
156 * the number of 512-byte sectors has to fit into a sector_t.
158 return loopsize >> 9;
161 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
167 * We support direct I/O only if lo_offset is aligned with the logical I/O size
168 * of backing device, and the logical block size of loop is bigger than that of
169 * the backing device.
171 static bool lo_can_use_dio(struct loop_device *lo)
173 if (!(lo->lo_backing_file->f_mode & FMODE_CAN_ODIRECT))
175 if (queue_logical_block_size(lo->lo_queue) < lo->lo_min_dio_size)
177 if (lo->lo_offset & (lo->lo_min_dio_size - 1))
183 * Direct I/O can be enabled either by using an O_DIRECT file descriptor, or by
184 * passing in the LO_FLAGS_DIRECT_IO flag from userspace. It will be silently
185 * disabled when the device block size is too small or the offset is unaligned.
187 * loop_get_status will always report the effective LO_FLAGS_DIRECT_IO flag and
188 * not the originally passed in one.
190 static inline void loop_update_dio(struct loop_device *lo)
192 lockdep_assert_held(&lo->lo_mutex);
193 WARN_ON_ONCE(lo->lo_state == Lo_bound &&
194 lo->lo_queue->mq_freeze_depth == 0);
196 if ((lo->lo_flags & LO_FLAGS_DIRECT_IO) && !lo_can_use_dio(lo))
197 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
201 * loop_set_size() - sets device size and notifies userspace
202 * @lo: struct loop_device to set the size for
203 * @size: new size of the loop device
205 * Callers must validate that the size passed into this function fits into
206 * a sector_t, eg using loop_validate_size()
208 static void loop_set_size(struct loop_device *lo, loff_t size)
210 if (!set_capacity_and_notify(lo->lo_disk, size))
211 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
214 static void loop_clear_limits(struct loop_device *lo, int mode)
216 struct queue_limits lim = queue_limits_start_update(lo->lo_queue);
218 if (mode & FALLOC_FL_ZERO_RANGE)
219 lim.max_write_zeroes_sectors = 0;
221 if (mode & FALLOC_FL_PUNCH_HOLE) {
222 lim.max_hw_discard_sectors = 0;
223 lim.discard_granularity = 0;
227 * XXX: this updates the queue limits without freezing the queue, which
228 * is against the locking protocol and dangerous. But we can't just
229 * freeze the queue as we're inside the ->queue_rq method here. So this
230 * should move out into a workqueue unless we get the file operations to
231 * advertise if they support specific fallocate operations.
233 queue_limits_commit_update(lo->lo_queue, &lim);
236 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
240 * We use fallocate to manipulate the space mappings used by the image
241 * a.k.a. discard/zerorange.
243 struct file *file = lo->lo_backing_file;
246 mode |= FALLOC_FL_KEEP_SIZE;
248 if (!bdev_max_discard_sectors(lo->lo_device))
251 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
252 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
256 * We initially configure the limits in a hope that fallocate is
257 * supported and clear them here if that turns out not to be true.
259 if (unlikely(ret == -EOPNOTSUPP))
260 loop_clear_limits(lo, mode);
265 static int lo_req_flush(struct loop_device *lo, struct request *rq)
267 int ret = vfs_fsync(lo->lo_backing_file, 0);
268 if (unlikely(ret && ret != -EINVAL))
274 static void lo_complete_rq(struct request *rq)
276 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
277 blk_status_t ret = BLK_STS_OK;
279 if (cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
280 req_op(rq) != REQ_OP_READ) {
282 ret = errno_to_blk_status(cmd->ret);
287 * Short READ - if we got some data, advance our request and
288 * retry it. If we got no data, end the rest with EIO.
291 blk_update_request(rq, BLK_STS_OK, cmd->ret);
293 blk_mq_requeue_request(rq, true);
295 struct bio *bio = rq->bio;
304 blk_mq_end_request(rq, ret);
308 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
310 struct request *rq = blk_mq_rq_from_pdu(cmd);
311 struct loop_device *lo = rq->q->queuedata;
313 if (!atomic_dec_and_test(&cmd->ref))
317 if (req_op(rq) == REQ_OP_WRITE)
318 file_end_write(lo->lo_backing_file);
319 if (likely(!blk_should_fake_timeout(rq->q)))
320 blk_mq_complete_request(rq);
323 static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
325 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
328 lo_rw_aio_do_completion(cmd);
331 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
334 struct iov_iter iter;
335 struct req_iterator rq_iter;
336 struct bio_vec *bvec;
337 struct request *rq = blk_mq_rq_from_pdu(cmd);
338 struct bio *bio = rq->bio;
339 struct file *file = lo->lo_backing_file;
345 rq_for_each_bvec(tmp, rq, rq_iter)
348 if (rq->bio != rq->biotail) {
350 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
357 * The bios of the request may be started from the middle of
358 * the 'bvec' because of bio splitting, so we can't directly
359 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
360 * API will take care of all details for us.
362 rq_for_each_bvec(tmp, rq, rq_iter) {
370 * Same here, this bio may be started from the middle of the
371 * 'bvec' because of bio splitting, so offset from the bvec
372 * must be passed to iov iterator
374 offset = bio->bi_iter.bi_bvec_done;
375 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
377 atomic_set(&cmd->ref, 2);
379 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
380 iter.iov_offset = offset;
382 cmd->iocb.ki_pos = pos;
383 cmd->iocb.ki_filp = file;
384 cmd->iocb.ki_ioprio = req_get_ioprio(rq);
386 cmd->iocb.ki_complete = lo_rw_aio_complete;
387 cmd->iocb.ki_flags = IOCB_DIRECT;
389 cmd->iocb.ki_complete = NULL;
390 cmd->iocb.ki_flags = 0;
393 if (rw == ITER_SOURCE) {
394 file_start_write(lo->lo_backing_file);
395 ret = file->f_op->write_iter(&cmd->iocb, &iter);
397 ret = file->f_op->read_iter(&cmd->iocb, &iter);
399 lo_rw_aio_do_completion(cmd);
401 if (ret != -EIOCBQUEUED)
402 lo_rw_aio_complete(&cmd->iocb, ret);
406 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
408 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
409 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
411 switch (req_op(rq)) {
413 return lo_req_flush(lo, rq);
414 case REQ_OP_WRITE_ZEROES:
416 * If the caller doesn't want deallocation, call zeroout to
417 * write zeroes the range. Otherwise, punch them out.
419 return lo_fallocate(lo, rq, pos,
420 (rq->cmd_flags & REQ_NOUNMAP) ?
421 FALLOC_FL_ZERO_RANGE :
422 FALLOC_FL_PUNCH_HOLE);
424 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
426 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
428 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
435 static void loop_reread_partitions(struct loop_device *lo)
439 mutex_lock(&lo->lo_disk->open_mutex);
440 rc = bdev_disk_changed(lo->lo_disk, false);
441 mutex_unlock(&lo->lo_disk->open_mutex);
443 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
444 __func__, lo->lo_number, lo->lo_file_name, rc);
447 static unsigned int loop_query_min_dio_size(struct loop_device *lo)
449 struct file *file = lo->lo_backing_file;
450 struct block_device *sb_bdev = file->f_mapping->host->i_sb->s_bdev;
454 * Use the minimal dio alignment of the file system if provided.
456 if (!vfs_getattr(&file->f_path, &st, STATX_DIOALIGN, 0) &&
457 (st.result_mask & STATX_DIOALIGN))
458 return st.dio_offset_align;
461 * In a perfect world this wouldn't be needed, but as of Linux 6.13 only
462 * a handful of file systems support the STATX_DIOALIGN flag.
465 return bdev_logical_block_size(sb_bdev);
469 static inline int is_loop_device(struct file *file)
471 struct inode *i = file->f_mapping->host;
473 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
476 static int loop_validate_file(struct file *file, struct block_device *bdev)
478 struct inode *inode = file->f_mapping->host;
479 struct file *f = file;
481 /* Avoid recursion */
482 while (is_loop_device(f)) {
483 struct loop_device *l;
485 lockdep_assert_held(&loop_validate_mutex);
486 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
489 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
490 if (l->lo_state != Lo_bound)
492 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
494 f = l->lo_backing_file;
496 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
501 static void loop_assign_backing_file(struct loop_device *lo, struct file *file)
503 lo->lo_backing_file = file;
504 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
505 mapping_set_gfp_mask(file->f_mapping,
506 lo->old_gfp_mask & ~(__GFP_IO | __GFP_FS));
507 if (lo->lo_backing_file->f_flags & O_DIRECT)
508 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
509 lo->lo_min_dio_size = loop_query_min_dio_size(lo);
512 static int loop_check_backing_file(struct file *file)
514 if (!file->f_op->read_iter)
517 if ((file->f_mode & FMODE_WRITE) && !file->f_op->write_iter)
524 * loop_change_fd switched the backing store of a loopback device to
525 * a new file. This is useful for operating system installers to free up
526 * the original file and in High Availability environments to switch to
527 * an alternative location for the content in case of server meltdown.
528 * This can only work if the loop device is used read-only, and if the
529 * new backing store is the same size and type as the old backing store.
531 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
534 struct file *file = fget(arg);
535 struct file *old_file;
536 unsigned int memflags;
544 error = loop_check_backing_file(file);
548 /* suppress uevents while reconfiguring the device */
549 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
551 is_loop = is_loop_device(file);
552 error = loop_global_lock_killable(lo, is_loop);
556 if (lo->lo_state != Lo_bound)
559 /* the loop device has to be read-only */
561 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
564 error = loop_validate_file(file, bdev);
568 old_file = lo->lo_backing_file;
572 /* size of the new backing store needs to be the same */
573 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
577 * We might switch to direct I/O mode for the loop device, write back
578 * all dirty data the page cache now that so that the individual I/O
579 * operations don't have to do that.
584 disk_force_media_change(lo->lo_disk);
585 memflags = blk_mq_freeze_queue(lo->lo_queue);
586 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
587 loop_assign_backing_file(lo, file);
589 blk_mq_unfreeze_queue(lo->lo_queue, memflags);
590 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
591 loop_global_unlock(lo, is_loop);
594 * Flush loop_validate_file() before fput(), for l->lo_backing_file
595 * might be pointing at old_file which might be the last reference.
598 mutex_lock(&loop_validate_mutex);
599 mutex_unlock(&loop_validate_mutex);
602 * We must drop file reference outside of lo_mutex as dropping
603 * the file ref can take open_mutex which creates circular locking
607 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
609 loop_reread_partitions(lo);
613 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
617 loop_global_unlock(lo, is_loop);
620 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
624 /* loop sysfs attributes */
626 static ssize_t loop_attr_show(struct device *dev, char *page,
627 ssize_t (*callback)(struct loop_device *, char *))
629 struct gendisk *disk = dev_to_disk(dev);
630 struct loop_device *lo = disk->private_data;
632 return callback(lo, page);
635 #define LOOP_ATTR_RO(_name) \
636 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
637 static ssize_t loop_attr_do_show_##_name(struct device *d, \
638 struct device_attribute *attr, char *b) \
640 return loop_attr_show(d, b, loop_attr_##_name##_show); \
642 static struct device_attribute loop_attr_##_name = \
643 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
645 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
650 spin_lock_irq(&lo->lo_lock);
651 if (lo->lo_backing_file)
652 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
653 spin_unlock_irq(&lo->lo_lock);
655 if (IS_ERR_OR_NULL(p))
659 memmove(buf, p, ret);
667 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
669 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
672 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
674 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
677 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
679 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
681 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
684 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
686 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
688 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
691 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
693 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
695 return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
698 LOOP_ATTR_RO(backing_file);
699 LOOP_ATTR_RO(offset);
700 LOOP_ATTR_RO(sizelimit);
701 LOOP_ATTR_RO(autoclear);
702 LOOP_ATTR_RO(partscan);
705 static struct attribute *loop_attrs[] = {
706 &loop_attr_backing_file.attr,
707 &loop_attr_offset.attr,
708 &loop_attr_sizelimit.attr,
709 &loop_attr_autoclear.attr,
710 &loop_attr_partscan.attr,
715 static struct attribute_group loop_attribute_group = {
720 static void loop_sysfs_init(struct loop_device *lo)
722 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
723 &loop_attribute_group);
726 static void loop_sysfs_exit(struct loop_device *lo)
728 if (lo->sysfs_inited)
729 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
730 &loop_attribute_group);
733 static void loop_get_discard_config(struct loop_device *lo,
734 u32 *granularity, u32 *max_discard_sectors)
736 struct file *file = lo->lo_backing_file;
737 struct inode *inode = file->f_mapping->host;
741 * If the backing device is a block device, mirror its zeroing
742 * capability. Set the discard sectors to the block device's zeroing
743 * capabilities because loop discards result in blkdev_issue_zeroout(),
744 * not blkdev_issue_discard(). This maintains consistent behavior with
745 * file-backed loop devices: discarded regions read back as zero.
747 if (S_ISBLK(inode->i_mode)) {
748 struct block_device *bdev = I_BDEV(inode);
750 *max_discard_sectors = bdev_write_zeroes_sectors(bdev);
751 *granularity = bdev_discard_granularity(bdev);
754 * We use punch hole to reclaim the free space used by the
755 * image a.k.a. discard.
757 } else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) {
758 *max_discard_sectors = UINT_MAX >> 9;
759 *granularity = sbuf.f_bsize;
764 struct rb_node rb_node;
765 struct work_struct work;
766 struct list_head cmd_list;
767 struct list_head idle_list;
768 struct loop_device *lo;
769 struct cgroup_subsys_state *blkcg_css;
770 unsigned long last_ran_at;
773 static void loop_workfn(struct work_struct *work);
775 #ifdef CONFIG_BLK_CGROUP
776 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
778 return !css || css == blkcg_root_css;
781 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
787 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
789 struct rb_node **node, *parent = NULL;
790 struct loop_worker *cur_worker, *worker = NULL;
791 struct work_struct *work;
792 struct list_head *cmd_list;
794 spin_lock_irq(&lo->lo_work_lock);
796 if (queue_on_root_worker(cmd->blkcg_css))
799 node = &lo->worker_tree.rb_node;
803 cur_worker = container_of(*node, struct loop_worker, rb_node);
804 if (cur_worker->blkcg_css == cmd->blkcg_css) {
807 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
808 node = &(*node)->rb_left;
810 node = &(*node)->rb_right;
816 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
818 * In the event we cannot allocate a worker, just queue on the
819 * rootcg worker and issue the I/O as the rootcg
822 cmd->blkcg_css = NULL;
824 css_put(cmd->memcg_css);
825 cmd->memcg_css = NULL;
829 worker->blkcg_css = cmd->blkcg_css;
830 css_get(worker->blkcg_css);
831 INIT_WORK(&worker->work, loop_workfn);
832 INIT_LIST_HEAD(&worker->cmd_list);
833 INIT_LIST_HEAD(&worker->idle_list);
835 rb_link_node(&worker->rb_node, parent, node);
836 rb_insert_color(&worker->rb_node, &lo->worker_tree);
840 * We need to remove from the idle list here while
841 * holding the lock so that the idle timer doesn't
844 if (!list_empty(&worker->idle_list))
845 list_del_init(&worker->idle_list);
846 work = &worker->work;
847 cmd_list = &worker->cmd_list;
849 work = &lo->rootcg_work;
850 cmd_list = &lo->rootcg_cmd_list;
852 list_add_tail(&cmd->list_entry, cmd_list);
853 queue_work(lo->workqueue, work);
854 spin_unlock_irq(&lo->lo_work_lock);
857 static void loop_set_timer(struct loop_device *lo)
859 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
862 static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
864 struct loop_worker *pos, *worker;
866 spin_lock_irq(&lo->lo_work_lock);
867 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
870 time_is_after_jiffies(worker->last_ran_at +
871 LOOP_IDLE_WORKER_TIMEOUT))
873 list_del(&worker->idle_list);
874 rb_erase(&worker->rb_node, &lo->worker_tree);
875 css_put(worker->blkcg_css);
878 if (!list_empty(&lo->idle_worker_list))
880 spin_unlock_irq(&lo->lo_work_lock);
883 static void loop_free_idle_workers_timer(struct timer_list *timer)
885 struct loop_device *lo = container_of(timer, struct loop_device, timer);
887 return loop_free_idle_workers(lo, false);
891 * loop_set_status_from_info - configure device from loop_info
892 * @lo: struct loop_device to configure
893 * @info: struct loop_info64 to configure the device with
895 * Configures the loop device parameters according to the passed
896 * in loop_info64 configuration.
899 loop_set_status_from_info(struct loop_device *lo,
900 const struct loop_info64 *info)
902 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
905 switch (info->lo_encrypt_type) {
909 pr_warn("support for the xor transformation has been removed.\n");
911 case LO_CRYPT_CRYPTOAPI:
912 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
918 /* Avoid assigning overflow values */
919 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
922 lo->lo_offset = info->lo_offset;
923 lo->lo_sizelimit = info->lo_sizelimit;
925 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
926 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
930 static unsigned int loop_default_blocksize(struct loop_device *lo)
932 /* In case of direct I/O, match underlying minimum I/O size */
933 if (lo->lo_flags & LO_FLAGS_DIRECT_IO)
934 return lo->lo_min_dio_size;
938 static void loop_update_limits(struct loop_device *lo, struct queue_limits *lim,
941 struct file *file = lo->lo_backing_file;
942 struct inode *inode = file->f_mapping->host;
943 struct block_device *backing_bdev = NULL;
944 u32 granularity = 0, max_discard_sectors = 0;
946 if (S_ISBLK(inode->i_mode))
947 backing_bdev = I_BDEV(inode);
948 else if (inode->i_sb->s_bdev)
949 backing_bdev = inode->i_sb->s_bdev;
952 bsize = loop_default_blocksize(lo);
954 loop_get_discard_config(lo, &granularity, &max_discard_sectors);
956 lim->logical_block_size = bsize;
957 lim->physical_block_size = bsize;
959 lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_ROTATIONAL);
960 if (file->f_op->fsync && !(lo->lo_flags & LO_FLAGS_READ_ONLY))
961 lim->features |= BLK_FEAT_WRITE_CACHE;
962 if (backing_bdev && !bdev_nonrot(backing_bdev))
963 lim->features |= BLK_FEAT_ROTATIONAL;
964 lim->max_hw_discard_sectors = max_discard_sectors;
965 lim->max_write_zeroes_sectors = max_discard_sectors;
966 if (max_discard_sectors)
967 lim->discard_granularity = granularity;
969 lim->discard_granularity = 0;
972 static int loop_configure(struct loop_device *lo, blk_mode_t mode,
973 struct block_device *bdev,
974 const struct loop_config *config)
976 struct file *file = fget(config->fd);
977 struct queue_limits lim;
986 error = loop_check_backing_file(file);
990 is_loop = is_loop_device(file);
992 /* This is safe, since we have a reference from open(). */
993 __module_get(THIS_MODULE);
996 * If we don't hold exclusive handle for the device, upgrade to it
997 * here to avoid changing device under exclusive owner.
999 if (!(mode & BLK_OPEN_EXCL)) {
1000 error = bd_prepare_to_claim(bdev, loop_configure, NULL);
1005 error = loop_global_lock_killable(lo, is_loop);
1010 if (lo->lo_state != Lo_unbound)
1013 error = loop_validate_file(file, bdev);
1017 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1022 error = loop_set_status_from_info(lo, &config->info);
1025 lo->lo_flags = config->info.lo_flags;
1027 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1028 !file->f_op->write_iter)
1029 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1031 if (!lo->workqueue) {
1032 lo->workqueue = alloc_workqueue("loop%d",
1033 WQ_UNBOUND | WQ_FREEZABLE,
1035 if (!lo->workqueue) {
1041 /* suppress uevents while reconfiguring the device */
1042 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1044 disk_force_media_change(lo->lo_disk);
1045 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1047 lo->lo_device = bdev;
1048 loop_assign_backing_file(lo, file);
1050 lim = queue_limits_start_update(lo->lo_queue);
1051 loop_update_limits(lo, &lim, config->block_size);
1052 /* No need to freeze the queue as the device isn't bound yet. */
1053 error = queue_limits_commit_update(lo->lo_queue, &lim);
1058 * We might switch to direct I/O mode for the loop device, write back
1059 * all dirty data the page cache now that so that the individual I/O
1060 * operations don't have to do that.
1064 loop_update_dio(lo);
1065 loop_sysfs_init(lo);
1067 size = get_loop_size(lo, file);
1068 loop_set_size(lo, size);
1070 /* Order wrt reading lo_state in loop_validate_file(). */
1073 lo->lo_state = Lo_bound;
1075 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1076 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1078 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1080 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1081 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1083 loop_global_unlock(lo, is_loop);
1085 loop_reread_partitions(lo);
1087 if (!(mode & BLK_OPEN_EXCL))
1088 bd_abort_claiming(bdev, loop_configure);
1093 loop_global_unlock(lo, is_loop);
1095 if (!(mode & BLK_OPEN_EXCL))
1096 bd_abort_claiming(bdev, loop_configure);
1099 /* This is safe: open() is still holding a reference. */
1100 module_put(THIS_MODULE);
1104 static void __loop_clr_fd(struct loop_device *lo)
1106 struct queue_limits lim;
1108 gfp_t gfp = lo->old_gfp_mask;
1110 spin_lock_irq(&lo->lo_lock);
1111 filp = lo->lo_backing_file;
1112 lo->lo_backing_file = NULL;
1113 spin_unlock_irq(&lo->lo_lock);
1115 lo->lo_device = NULL;
1117 lo->lo_sizelimit = 0;
1118 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1121 * Reset the block size to the default.
1123 * No queue freezing needed because this is called from the final
1124 * ->release call only, so there can't be any outstanding I/O.
1126 lim = queue_limits_start_update(lo->lo_queue);
1127 lim.logical_block_size = SECTOR_SIZE;
1128 lim.physical_block_size = SECTOR_SIZE;
1129 lim.io_min = SECTOR_SIZE;
1130 queue_limits_commit_update(lo->lo_queue, &lim);
1132 invalidate_disk(lo->lo_disk);
1133 loop_sysfs_exit(lo);
1134 /* let user-space know about this change */
1135 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1136 mapping_set_gfp_mask(filp->f_mapping, gfp);
1137 /* This is safe: open() is still holding a reference. */
1138 module_put(THIS_MODULE);
1140 disk_force_media_change(lo->lo_disk);
1142 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1146 * open_mutex has been held already in release path, so don't
1147 * acquire it if this function is called in such case.
1149 * If the reread partition isn't from release path, lo_refcnt
1150 * must be at least one and it can only become zero when the
1151 * current holder is released.
1153 err = bdev_disk_changed(lo->lo_disk, false);
1155 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1156 __func__, lo->lo_number, err);
1157 /* Device is gone, no point in returning error */
1161 * lo->lo_state is set to Lo_unbound here after above partscan has
1162 * finished. There cannot be anybody else entering __loop_clr_fd() as
1163 * Lo_rundown state protects us from all the other places trying to
1164 * change the 'lo' device.
1168 set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1169 mutex_lock(&lo->lo_mutex);
1170 lo->lo_state = Lo_unbound;
1171 mutex_unlock(&lo->lo_mutex);
1174 * Need not hold lo_mutex to fput backing file. Calling fput holding
1175 * lo_mutex triggers a circular lock dependency possibility warning as
1176 * fput can take open_mutex which is usually taken before lo_mutex.
1181 static int loop_clr_fd(struct loop_device *lo)
1186 * Since lo_ioctl() is called without locks held, it is possible that
1187 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1189 * Therefore, use global lock when setting Lo_rundown state in order to
1190 * make sure that loop_validate_file() will fail if the "struct file"
1191 * which loop_configure()/loop_change_fd() found via fget() was this
1194 err = loop_global_lock_killable(lo, true);
1197 if (lo->lo_state != Lo_bound) {
1198 loop_global_unlock(lo, true);
1202 * Mark the device for removing the backing device on last close.
1203 * If we are the only opener, also switch the state to roundown here to
1204 * prevent new openers from coming in.
1207 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1208 if (disk_openers(lo->lo_disk) == 1)
1209 lo->lo_state = Lo_rundown;
1210 loop_global_unlock(lo, true);
1216 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1219 bool partscan = false;
1220 bool size_changed = false;
1221 unsigned int memflags;
1223 err = mutex_lock_killable(&lo->lo_mutex);
1226 if (lo->lo_state != Lo_bound) {
1231 if (lo->lo_offset != info->lo_offset ||
1232 lo->lo_sizelimit != info->lo_sizelimit) {
1233 size_changed = true;
1234 sync_blockdev(lo->lo_device);
1235 invalidate_bdev(lo->lo_device);
1238 /* I/O needs to be drained before changing lo_offset or lo_sizelimit */
1239 memflags = blk_mq_freeze_queue(lo->lo_queue);
1241 err = loop_set_status_from_info(lo, info);
1245 partscan = !(lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1246 (info->lo_flags & LO_FLAGS_PARTSCAN);
1248 lo->lo_flags &= ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1249 lo->lo_flags |= (info->lo_flags & LOOP_SET_STATUS_SETTABLE_FLAGS);
1251 /* update the direct I/O flag if lo_offset changed */
1252 loop_update_dio(lo);
1255 blk_mq_unfreeze_queue(lo->lo_queue, memflags);
1257 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1258 if (!err && size_changed) {
1259 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1260 lo->lo_backing_file);
1261 loop_set_size(lo, new_size);
1264 mutex_unlock(&lo->lo_mutex);
1266 loop_reread_partitions(lo);
1272 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1278 ret = mutex_lock_killable(&lo->lo_mutex);
1281 if (lo->lo_state != Lo_bound) {
1282 mutex_unlock(&lo->lo_mutex);
1286 memset(info, 0, sizeof(*info));
1287 info->lo_number = lo->lo_number;
1288 info->lo_offset = lo->lo_offset;
1289 info->lo_sizelimit = lo->lo_sizelimit;
1290 info->lo_flags = lo->lo_flags;
1291 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1293 /* Drop lo_mutex while we call into the filesystem. */
1294 path = lo->lo_backing_file->f_path;
1296 mutex_unlock(&lo->lo_mutex);
1297 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1299 info->lo_device = huge_encode_dev(stat.dev);
1300 info->lo_inode = stat.ino;
1301 info->lo_rdevice = huge_encode_dev(stat.rdev);
1308 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1310 memset(info64, 0, sizeof(*info64));
1311 info64->lo_number = info->lo_number;
1312 info64->lo_device = info->lo_device;
1313 info64->lo_inode = info->lo_inode;
1314 info64->lo_rdevice = info->lo_rdevice;
1315 info64->lo_offset = info->lo_offset;
1316 info64->lo_sizelimit = 0;
1317 info64->lo_flags = info->lo_flags;
1318 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1322 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1324 memset(info, 0, sizeof(*info));
1325 info->lo_number = info64->lo_number;
1326 info->lo_device = info64->lo_device;
1327 info->lo_inode = info64->lo_inode;
1328 info->lo_rdevice = info64->lo_rdevice;
1329 info->lo_offset = info64->lo_offset;
1330 info->lo_flags = info64->lo_flags;
1331 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1333 /* error in case values were truncated */
1334 if (info->lo_device != info64->lo_device ||
1335 info->lo_rdevice != info64->lo_rdevice ||
1336 info->lo_inode != info64->lo_inode ||
1337 info->lo_offset != info64->lo_offset)
1344 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1346 struct loop_info info;
1347 struct loop_info64 info64;
1349 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1351 loop_info64_from_old(&info, &info64);
1352 return loop_set_status(lo, &info64);
1356 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1358 struct loop_info64 info64;
1360 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1362 return loop_set_status(lo, &info64);
1366 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1367 struct loop_info info;
1368 struct loop_info64 info64;
1373 err = loop_get_status(lo, &info64);
1375 err = loop_info64_to_old(&info64, &info);
1376 if (!err && copy_to_user(arg, &info, sizeof(info)))
1383 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1384 struct loop_info64 info64;
1389 err = loop_get_status(lo, &info64);
1390 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1396 static int loop_set_capacity(struct loop_device *lo)
1400 if (unlikely(lo->lo_state != Lo_bound))
1403 size = get_loop_size(lo, lo->lo_backing_file);
1404 loop_set_size(lo, size);
1409 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1411 bool use_dio = !!arg;
1412 unsigned int memflags;
1414 if (lo->lo_state != Lo_bound)
1416 if (use_dio == !!(lo->lo_flags & LO_FLAGS_DIRECT_IO))
1420 if (!lo_can_use_dio(lo))
1422 /* flush dirty pages before starting to use direct I/O */
1423 vfs_fsync(lo->lo_backing_file, 0);
1426 memflags = blk_mq_freeze_queue(lo->lo_queue);
1428 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
1430 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
1431 blk_mq_unfreeze_queue(lo->lo_queue, memflags);
1435 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1437 struct queue_limits lim;
1438 unsigned int memflags;
1441 if (lo->lo_state != Lo_bound)
1444 if (lo->lo_queue->limits.logical_block_size == arg)
1447 sync_blockdev(lo->lo_device);
1448 invalidate_bdev(lo->lo_device);
1450 lim = queue_limits_start_update(lo->lo_queue);
1451 loop_update_limits(lo, &lim, arg);
1453 memflags = blk_mq_freeze_queue(lo->lo_queue);
1454 err = queue_limits_commit_update(lo->lo_queue, &lim);
1455 loop_update_dio(lo);
1456 blk_mq_unfreeze_queue(lo->lo_queue, memflags);
1461 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1466 err = mutex_lock_killable(&lo->lo_mutex);
1470 case LOOP_SET_CAPACITY:
1471 err = loop_set_capacity(lo);
1473 case LOOP_SET_DIRECT_IO:
1474 err = loop_set_dio(lo, arg);
1476 case LOOP_SET_BLOCK_SIZE:
1477 err = loop_set_block_size(lo, arg);
1482 mutex_unlock(&lo->lo_mutex);
1486 static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1487 unsigned int cmd, unsigned long arg)
1489 struct loop_device *lo = bdev->bd_disk->private_data;
1490 void __user *argp = (void __user *) arg;
1496 * Legacy case - pass in a zeroed out struct loop_config with
1497 * only the file descriptor set , which corresponds with the
1498 * default parameters we'd have used otherwise.
1500 struct loop_config config;
1502 memset(&config, 0, sizeof(config));
1505 return loop_configure(lo, mode, bdev, &config);
1507 case LOOP_CONFIGURE: {
1508 struct loop_config config;
1510 if (copy_from_user(&config, argp, sizeof(config)))
1513 return loop_configure(lo, mode, bdev, &config);
1515 case LOOP_CHANGE_FD:
1516 return loop_change_fd(lo, bdev, arg);
1518 return loop_clr_fd(lo);
1519 case LOOP_SET_STATUS:
1521 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1522 err = loop_set_status_old(lo, argp);
1524 case LOOP_GET_STATUS:
1525 return loop_get_status_old(lo, argp);
1526 case LOOP_SET_STATUS64:
1528 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1529 err = loop_set_status64(lo, argp);
1531 case LOOP_GET_STATUS64:
1532 return loop_get_status64(lo, argp);
1533 case LOOP_SET_CAPACITY:
1534 case LOOP_SET_DIRECT_IO:
1535 case LOOP_SET_BLOCK_SIZE:
1536 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1540 err = lo_simple_ioctl(lo, cmd, arg);
1547 #ifdef CONFIG_COMPAT
1548 struct compat_loop_info {
1549 compat_int_t lo_number; /* ioctl r/o */
1550 compat_dev_t lo_device; /* ioctl r/o */
1551 compat_ulong_t lo_inode; /* ioctl r/o */
1552 compat_dev_t lo_rdevice; /* ioctl r/o */
1553 compat_int_t lo_offset;
1554 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1555 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1556 compat_int_t lo_flags; /* ioctl r/o */
1557 char lo_name[LO_NAME_SIZE];
1558 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1559 compat_ulong_t lo_init[2];
1564 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1565 * - noinlined to reduce stack space usage in main part of driver
1568 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1569 struct loop_info64 *info64)
1571 struct compat_loop_info info;
1573 if (copy_from_user(&info, arg, sizeof(info)))
1576 memset(info64, 0, sizeof(*info64));
1577 info64->lo_number = info.lo_number;
1578 info64->lo_device = info.lo_device;
1579 info64->lo_inode = info.lo_inode;
1580 info64->lo_rdevice = info.lo_rdevice;
1581 info64->lo_offset = info.lo_offset;
1582 info64->lo_sizelimit = 0;
1583 info64->lo_flags = info.lo_flags;
1584 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1589 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1590 * - noinlined to reduce stack space usage in main part of driver
1593 loop_info64_to_compat(const struct loop_info64 *info64,
1594 struct compat_loop_info __user *arg)
1596 struct compat_loop_info info;
1598 memset(&info, 0, sizeof(info));
1599 info.lo_number = info64->lo_number;
1600 info.lo_device = info64->lo_device;
1601 info.lo_inode = info64->lo_inode;
1602 info.lo_rdevice = info64->lo_rdevice;
1603 info.lo_offset = info64->lo_offset;
1604 info.lo_flags = info64->lo_flags;
1605 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1607 /* error in case values were truncated */
1608 if (info.lo_device != info64->lo_device ||
1609 info.lo_rdevice != info64->lo_rdevice ||
1610 info.lo_inode != info64->lo_inode ||
1611 info.lo_offset != info64->lo_offset)
1614 if (copy_to_user(arg, &info, sizeof(info)))
1620 loop_set_status_compat(struct loop_device *lo,
1621 const struct compat_loop_info __user *arg)
1623 struct loop_info64 info64;
1626 ret = loop_info64_from_compat(arg, &info64);
1629 return loop_set_status(lo, &info64);
1633 loop_get_status_compat(struct loop_device *lo,
1634 struct compat_loop_info __user *arg)
1636 struct loop_info64 info64;
1641 err = loop_get_status(lo, &info64);
1643 err = loop_info64_to_compat(&info64, arg);
1647 static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1648 unsigned int cmd, unsigned long arg)
1650 struct loop_device *lo = bdev->bd_disk->private_data;
1654 case LOOP_SET_STATUS:
1655 err = loop_set_status_compat(lo,
1656 (const struct compat_loop_info __user *)arg);
1658 case LOOP_GET_STATUS:
1659 err = loop_get_status_compat(lo,
1660 (struct compat_loop_info __user *)arg);
1662 case LOOP_SET_CAPACITY:
1664 case LOOP_GET_STATUS64:
1665 case LOOP_SET_STATUS64:
1666 case LOOP_CONFIGURE:
1667 arg = (unsigned long) compat_ptr(arg);
1670 case LOOP_CHANGE_FD:
1671 case LOOP_SET_BLOCK_SIZE:
1672 case LOOP_SET_DIRECT_IO:
1673 err = lo_ioctl(bdev, mode, cmd, arg);
1683 static int lo_open(struct gendisk *disk, blk_mode_t mode)
1685 struct loop_device *lo = disk->private_data;
1688 err = mutex_lock_killable(&lo->lo_mutex);
1692 if (lo->lo_state == Lo_deleting || lo->lo_state == Lo_rundown)
1694 mutex_unlock(&lo->lo_mutex);
1698 static void lo_release(struct gendisk *disk)
1700 struct loop_device *lo = disk->private_data;
1701 bool need_clear = false;
1703 if (disk_openers(disk) > 0)
1706 * Clear the backing device information if this is the last close of
1707 * a device that's been marked for auto clear, or on which LOOP_CLR_FD
1711 mutex_lock(&lo->lo_mutex);
1712 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR))
1713 lo->lo_state = Lo_rundown;
1715 need_clear = (lo->lo_state == Lo_rundown);
1716 mutex_unlock(&lo->lo_mutex);
1722 static void lo_free_disk(struct gendisk *disk)
1724 struct loop_device *lo = disk->private_data;
1727 destroy_workqueue(lo->workqueue);
1728 loop_free_idle_workers(lo, true);
1729 timer_shutdown_sync(&lo->timer);
1730 mutex_destroy(&lo->lo_mutex);
1734 static const struct block_device_operations lo_fops = {
1735 .owner = THIS_MODULE,
1737 .release = lo_release,
1739 #ifdef CONFIG_COMPAT
1740 .compat_ioctl = lo_compat_ioctl,
1742 .free_disk = lo_free_disk,
1746 * And now the modules code and kernel interface.
1750 * If max_loop is specified, create that many devices upfront.
1751 * This also becomes a hard limit. If max_loop is not specified,
1752 * the default isn't a hard limit (as before commit 85c50197716c
1753 * changed the default value from 0 for max_loop=0 reasons), just
1754 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1755 * init time. Loop devices can be requested on-demand with the
1756 * /dev/loop-control interface, or be instantiated by accessing
1757 * a 'dead' device node.
1759 static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1761 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1762 static bool max_loop_specified;
1764 static int max_loop_param_set_int(const char *val,
1765 const struct kernel_param *kp)
1769 ret = param_set_int(val, kp);
1773 max_loop_specified = true;
1777 static const struct kernel_param_ops max_loop_param_ops = {
1778 .set = max_loop_param_set_int,
1779 .get = param_get_int,
1782 module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1783 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1785 module_param(max_loop, int, 0444);
1786 MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1789 module_param(max_part, int, 0444);
1790 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1792 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1794 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1798 ret = kstrtoint(s, 0, &qd);
1803 hw_queue_depth = qd;
1807 static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1808 .set = loop_set_hw_queue_depth,
1809 .get = param_get_int,
1812 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1813 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1815 MODULE_DESCRIPTION("Loopback device support");
1816 MODULE_LICENSE("GPL");
1817 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1819 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1820 const struct blk_mq_queue_data *bd)
1822 struct request *rq = bd->rq;
1823 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1824 struct loop_device *lo = rq->q->queuedata;
1826 blk_mq_start_request(rq);
1828 if (lo->lo_state != Lo_bound)
1829 return BLK_STS_IOERR;
1831 switch (req_op(rq)) {
1833 case REQ_OP_DISCARD:
1834 case REQ_OP_WRITE_ZEROES:
1835 cmd->use_aio = false;
1838 cmd->use_aio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1842 /* always use the first bio's css */
1843 cmd->blkcg_css = NULL;
1844 cmd->memcg_css = NULL;
1845 #ifdef CONFIG_BLK_CGROUP
1847 cmd->blkcg_css = bio_blkcg_css(rq->bio);
1849 if (cmd->blkcg_css) {
1851 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1852 &memory_cgrp_subsys);
1857 loop_queue_work(lo, cmd);
1862 static void loop_handle_cmd(struct loop_cmd *cmd)
1864 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1865 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1866 struct request *rq = blk_mq_rq_from_pdu(cmd);
1867 const bool write = op_is_write(req_op(rq));
1868 struct loop_device *lo = rq->q->queuedata;
1870 struct mem_cgroup *old_memcg = NULL;
1872 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1878 kthread_associate_blkcg(cmd_blkcg_css);
1880 old_memcg = set_active_memcg(
1881 mem_cgroup_from_css(cmd_memcg_css));
1884 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1885 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1886 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1887 * not yet been completed.
1889 ret = do_req_filebacked(lo, rq);
1892 kthread_associate_blkcg(NULL);
1894 if (cmd_memcg_css) {
1895 set_active_memcg(old_memcg);
1896 css_put(cmd_memcg_css);
1899 /* complete non-aio request */
1900 if (ret != -EIOCBQUEUED) {
1901 if (ret == -EOPNOTSUPP)
1904 cmd->ret = ret ? -EIO : 0;
1905 if (likely(!blk_should_fake_timeout(rq->q)))
1906 blk_mq_complete_request(rq);
1910 static void loop_process_work(struct loop_worker *worker,
1911 struct list_head *cmd_list, struct loop_device *lo)
1913 int orig_flags = current->flags;
1914 struct loop_cmd *cmd;
1916 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1917 spin_lock_irq(&lo->lo_work_lock);
1918 while (!list_empty(cmd_list)) {
1920 cmd_list->next, struct loop_cmd, list_entry);
1921 list_del(cmd_list->next);
1922 spin_unlock_irq(&lo->lo_work_lock);
1924 loop_handle_cmd(cmd);
1927 spin_lock_irq(&lo->lo_work_lock);
1931 * We only add to the idle list if there are no pending cmds
1932 * *and* the worker will not run again which ensures that it
1933 * is safe to free any worker on the idle list
1935 if (worker && !work_pending(&worker->work)) {
1936 worker->last_ran_at = jiffies;
1937 list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1940 spin_unlock_irq(&lo->lo_work_lock);
1941 current->flags = orig_flags;
1944 static void loop_workfn(struct work_struct *work)
1946 struct loop_worker *worker =
1947 container_of(work, struct loop_worker, work);
1948 loop_process_work(worker, &worker->cmd_list, worker->lo);
1951 static void loop_rootcg_workfn(struct work_struct *work)
1953 struct loop_device *lo =
1954 container_of(work, struct loop_device, rootcg_work);
1955 loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1958 static const struct blk_mq_ops loop_mq_ops = {
1959 .queue_rq = loop_queue_rq,
1960 .complete = lo_complete_rq,
1963 static int loop_add(int i)
1965 struct queue_limits lim = {
1967 * Random number picked from the historic block max_sectors cap.
1969 .max_hw_sectors = 2560u,
1971 struct loop_device *lo;
1972 struct gendisk *disk;
1976 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1979 lo->worker_tree = RB_ROOT;
1980 INIT_LIST_HEAD(&lo->idle_worker_list);
1981 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
1982 lo->lo_state = Lo_unbound;
1984 err = mutex_lock_killable(&loop_ctl_mutex);
1988 /* allocate id, if @id >= 0, we're requesting that specific id */
1990 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1994 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1996 mutex_unlock(&loop_ctl_mutex);
2001 lo->tag_set.ops = &loop_mq_ops;
2002 lo->tag_set.nr_hw_queues = 1;
2003 lo->tag_set.queue_depth = hw_queue_depth;
2004 lo->tag_set.numa_node = NUMA_NO_NODE;
2005 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2006 lo->tag_set.flags = BLK_MQ_F_STACKING | BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2007 lo->tag_set.driver_data = lo;
2009 err = blk_mq_alloc_tag_set(&lo->tag_set);
2013 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo);
2015 err = PTR_ERR(disk);
2016 goto out_cleanup_tags;
2018 lo->lo_queue = lo->lo_disk->queue;
2021 * Disable partition scanning by default. The in-kernel partition
2022 * scanning can be requested individually per-device during its
2023 * setup. Userspace can always add and remove partitions from all
2024 * devices. The needed partition minors are allocated from the
2025 * extended minor space, the main loop device numbers will continue
2026 * to match the loop minors, regardless of the number of partitions
2029 * If max_part is given, partition scanning is globally enabled for
2030 * all loop devices. The minors for the main loop devices will be
2031 * multiples of max_part.
2033 * Note: Global-for-all-devices, set-only-at-init, read-only module
2034 * parameteters like 'max_loop' and 'max_part' make things needlessly
2035 * complicated, are too static, inflexible and may surprise
2036 * userspace tools. Parameters like this in general should be avoided.
2039 set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2040 mutex_init(&lo->lo_mutex);
2042 spin_lock_init(&lo->lo_lock);
2043 spin_lock_init(&lo->lo_work_lock);
2044 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2045 INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2046 disk->major = LOOP_MAJOR;
2047 disk->first_minor = i << part_shift;
2048 disk->minors = 1 << part_shift;
2049 disk->fops = &lo_fops;
2050 disk->private_data = lo;
2051 disk->queue = lo->lo_queue;
2052 disk->events = DISK_EVENT_MEDIA_CHANGE;
2053 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2054 sprintf(disk->disk_name, "loop%d", i);
2055 /* Make this loop device reachable from pathname. */
2056 err = add_disk(disk);
2058 goto out_cleanup_disk;
2060 /* Show this loop device. */
2061 mutex_lock(&loop_ctl_mutex);
2062 lo->idr_visible = true;
2063 mutex_unlock(&loop_ctl_mutex);
2070 blk_mq_free_tag_set(&lo->tag_set);
2072 mutex_lock(&loop_ctl_mutex);
2073 idr_remove(&loop_index_idr, i);
2074 mutex_unlock(&loop_ctl_mutex);
2081 static void loop_remove(struct loop_device *lo)
2083 /* Make this loop device unreachable from pathname. */
2084 del_gendisk(lo->lo_disk);
2085 blk_mq_free_tag_set(&lo->tag_set);
2087 mutex_lock(&loop_ctl_mutex);
2088 idr_remove(&loop_index_idr, lo->lo_number);
2089 mutex_unlock(&loop_ctl_mutex);
2091 put_disk(lo->lo_disk);
2094 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2095 static void loop_probe(dev_t dev)
2097 int idx = MINOR(dev) >> part_shift;
2099 if (max_loop_specified && max_loop && idx >= max_loop)
2104 #define loop_probe NULL
2105 #endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2107 static int loop_control_remove(int idx)
2109 struct loop_device *lo;
2113 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2117 /* Hide this loop device for serialization. */
2118 ret = mutex_lock_killable(&loop_ctl_mutex);
2121 lo = idr_find(&loop_index_idr, idx);
2122 if (!lo || !lo->idr_visible)
2125 lo->idr_visible = false;
2126 mutex_unlock(&loop_ctl_mutex);
2130 /* Check whether this loop device can be removed. */
2131 ret = mutex_lock_killable(&lo->lo_mutex);
2134 if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2135 mutex_unlock(&lo->lo_mutex);
2139 /* Mark this loop device as no more bound, but not quite unbound yet */
2140 lo->lo_state = Lo_deleting;
2141 mutex_unlock(&lo->lo_mutex);
2147 /* Show this loop device again. */
2148 mutex_lock(&loop_ctl_mutex);
2149 lo->idr_visible = true;
2150 mutex_unlock(&loop_ctl_mutex);
2154 static int loop_control_get_free(int idx)
2156 struct loop_device *lo;
2159 ret = mutex_lock_killable(&loop_ctl_mutex);
2162 idr_for_each_entry(&loop_index_idr, lo, id) {
2163 /* Hitting a race results in creating a new loop device which is harmless. */
2164 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2167 mutex_unlock(&loop_ctl_mutex);
2168 return loop_add(-1);
2170 mutex_unlock(&loop_ctl_mutex);
2174 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2179 return loop_add(parm);
2180 case LOOP_CTL_REMOVE:
2181 return loop_control_remove(parm);
2182 case LOOP_CTL_GET_FREE:
2183 return loop_control_get_free(parm);
2189 static const struct file_operations loop_ctl_fops = {
2190 .open = nonseekable_open,
2191 .unlocked_ioctl = loop_control_ioctl,
2192 .compat_ioctl = loop_control_ioctl,
2193 .owner = THIS_MODULE,
2194 .llseek = noop_llseek,
2197 static struct miscdevice loop_misc = {
2198 .minor = LOOP_CTRL_MINOR,
2199 .name = "loop-control",
2200 .fops = &loop_ctl_fops,
2203 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2204 MODULE_ALIAS("devname:loop-control");
2206 static int __init loop_init(void)
2213 part_shift = fls(max_part);
2216 * Adjust max_part according to part_shift as it is exported
2217 * to user space so that user can decide correct minor number
2218 * if [s]he want to create more devices.
2220 * Note that -1 is required because partition 0 is reserved
2221 * for the whole disk.
2223 max_part = (1UL << part_shift) - 1;
2226 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2231 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2236 err = misc_register(&loop_misc);
2241 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2246 /* pre-create number of devices given by config or max_loop */
2247 for (i = 0; i < max_loop; i++)
2250 printk(KERN_INFO "loop: module loaded\n");
2254 misc_deregister(&loop_misc);
2259 static void __exit loop_exit(void)
2261 struct loop_device *lo;
2264 unregister_blkdev(LOOP_MAJOR, "loop");
2265 misc_deregister(&loop_misc);
2268 * There is no need to use loop_ctl_mutex here, for nobody else can
2269 * access loop_index_idr when this module is unloading (unless forced
2270 * module unloading is requested). If this is not a clean unloading,
2271 * we have no means to avoid kernel crash.
2273 idr_for_each_entry(&loop_index_idr, lo, id)
2276 idr_destroy(&loop_index_idr);
2279 module_init(loop_init);
2280 module_exit(loop_exit);
2283 static int __init max_loop_setup(char *str)
2285 max_loop = simple_strtol(str, NULL, 0);
2286 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2287 max_loop_specified = true;
2292 __setup("max_loop=", max_loop_setup);