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);
169 * We support direct I/O only if lo_offset is aligned with the logical I/O size
170 * of backing device, and the logical block size of loop is bigger than that of
171 * the backing device.
173 static bool lo_bdev_can_use_dio(struct loop_device *lo,
174 struct block_device *backing_bdev)
176 unsigned short sb_bsize = bdev_logical_block_size(backing_bdev);
178 if (queue_logical_block_size(lo->lo_queue) < sb_bsize)
180 if (lo->lo_offset & (sb_bsize - 1))
185 static void __loop_update_dio(struct loop_device *lo, bool dio)
187 struct file *file = lo->lo_backing_file;
188 struct inode *inode = file->f_mapping->host;
189 struct block_device *backing_bdev = NULL;
192 if (S_ISBLK(inode->i_mode))
193 backing_bdev = I_BDEV(inode);
194 else if (inode->i_sb->s_bdev)
195 backing_bdev = inode->i_sb->s_bdev;
197 use_dio = dio && (file->f_mode & FMODE_CAN_ODIRECT) &&
198 (!backing_bdev || lo_bdev_can_use_dio(lo, backing_bdev));
200 if (lo->use_dio == use_dio)
203 /* flush dirty pages before changing direct IO */
207 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
208 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
209 * will get updated by ioctl(LOOP_GET_STATUS)
211 if (lo->lo_state == Lo_bound)
212 blk_mq_freeze_queue(lo->lo_queue);
213 lo->use_dio = use_dio;
215 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
216 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
218 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
221 if (lo->lo_state == Lo_bound)
222 blk_mq_unfreeze_queue(lo->lo_queue);
226 * loop_set_size() - sets device size and notifies userspace
227 * @lo: struct loop_device to set the size for
228 * @size: new size of the loop device
230 * Callers must validate that the size passed into this function fits into
231 * a sector_t, eg using loop_validate_size()
233 static void loop_set_size(struct loop_device *lo, loff_t size)
235 if (!set_capacity_and_notify(lo->lo_disk, size))
236 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
239 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
244 iov_iter_bvec(&i, ITER_SOURCE, bvec, 1, bvec->bv_len);
246 bw = vfs_iter_write(file, &i, ppos, 0);
248 if (likely(bw == bvec->bv_len))
251 printk_ratelimited(KERN_ERR
252 "loop: Write error at byte offset %llu, length %i.\n",
253 (unsigned long long)*ppos, bvec->bv_len);
259 static int lo_write_simple(struct loop_device *lo, struct request *rq,
263 struct req_iterator iter;
266 rq_for_each_segment(bvec, rq, iter) {
267 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
276 static int lo_read_simple(struct loop_device *lo, struct request *rq,
280 struct req_iterator iter;
284 rq_for_each_segment(bvec, rq, iter) {
285 iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len);
286 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
290 flush_dcache_page(bvec.bv_page);
292 if (len != bvec.bv_len) {
295 __rq_for_each_bio(bio, rq)
305 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
309 * We use fallocate to manipulate the space mappings used by the image
310 * a.k.a. discard/zerorange.
312 struct file *file = lo->lo_backing_file;
315 mode |= FALLOC_FL_KEEP_SIZE;
317 if (!bdev_max_discard_sectors(lo->lo_device))
320 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
321 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
326 static int lo_req_flush(struct loop_device *lo, struct request *rq)
328 int ret = vfs_fsync(lo->lo_backing_file, 0);
329 if (unlikely(ret && ret != -EINVAL))
335 static void lo_complete_rq(struct request *rq)
337 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
338 blk_status_t ret = BLK_STS_OK;
340 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
341 req_op(rq) != REQ_OP_READ) {
343 ret = errno_to_blk_status(cmd->ret);
348 * Short READ - if we got some data, advance our request and
349 * retry it. If we got no data, end the rest with EIO.
352 blk_update_request(rq, BLK_STS_OK, cmd->ret);
354 blk_mq_requeue_request(rq, true);
357 struct bio *bio = rq->bio;
366 blk_mq_end_request(rq, ret);
370 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
372 struct request *rq = blk_mq_rq_from_pdu(cmd);
374 if (!atomic_dec_and_test(&cmd->ref))
378 if (likely(!blk_should_fake_timeout(rq->q)))
379 blk_mq_complete_request(rq);
382 static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
384 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
387 lo_rw_aio_do_completion(cmd);
390 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
393 struct iov_iter iter;
394 struct req_iterator rq_iter;
395 struct bio_vec *bvec;
396 struct request *rq = blk_mq_rq_from_pdu(cmd);
397 struct bio *bio = rq->bio;
398 struct file *file = lo->lo_backing_file;
404 rq_for_each_bvec(tmp, rq, rq_iter)
407 if (rq->bio != rq->biotail) {
409 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
416 * The bios of the request may be started from the middle of
417 * the 'bvec' because of bio splitting, so we can't directly
418 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
419 * API will take care of all details for us.
421 rq_for_each_bvec(tmp, rq, rq_iter) {
429 * Same here, this bio may be started from the middle of the
430 * 'bvec' because of bio splitting, so offset from the bvec
431 * must be passed to iov iterator
433 offset = bio->bi_iter.bi_bvec_done;
434 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
436 atomic_set(&cmd->ref, 2);
438 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
439 iter.iov_offset = offset;
441 cmd->iocb.ki_pos = pos;
442 cmd->iocb.ki_filp = file;
443 cmd->iocb.ki_complete = lo_rw_aio_complete;
444 cmd->iocb.ki_flags = IOCB_DIRECT;
445 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
447 if (rw == ITER_SOURCE)
448 ret = call_write_iter(file, &cmd->iocb, &iter);
450 ret = call_read_iter(file, &cmd->iocb, &iter);
452 lo_rw_aio_do_completion(cmd);
454 if (ret != -EIOCBQUEUED)
455 lo_rw_aio_complete(&cmd->iocb, ret);
459 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
461 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
462 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
465 * lo_write_simple and lo_read_simple should have been covered
466 * by io submit style function like lo_rw_aio(), one blocker
467 * is that lo_read_simple() need to call flush_dcache_page after
468 * the page is written from kernel, and it isn't easy to handle
469 * this in io submit style function which submits all segments
470 * of the req at one time. And direct read IO doesn't need to
471 * run flush_dcache_page().
473 switch (req_op(rq)) {
475 return lo_req_flush(lo, rq);
476 case REQ_OP_WRITE_ZEROES:
478 * If the caller doesn't want deallocation, call zeroout to
479 * write zeroes the range. Otherwise, punch them out.
481 return lo_fallocate(lo, rq, pos,
482 (rq->cmd_flags & REQ_NOUNMAP) ?
483 FALLOC_FL_ZERO_RANGE :
484 FALLOC_FL_PUNCH_HOLE);
486 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
489 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
491 return lo_write_simple(lo, rq, pos);
494 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
496 return lo_read_simple(lo, rq, pos);
503 static inline void loop_update_dio(struct loop_device *lo)
505 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
509 static void loop_reread_partitions(struct loop_device *lo)
513 mutex_lock(&lo->lo_disk->open_mutex);
514 rc = bdev_disk_changed(lo->lo_disk, false);
515 mutex_unlock(&lo->lo_disk->open_mutex);
517 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
518 __func__, lo->lo_number, lo->lo_file_name, rc);
521 static inline int is_loop_device(struct file *file)
523 struct inode *i = file->f_mapping->host;
525 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
528 static int loop_validate_file(struct file *file, struct block_device *bdev)
530 struct inode *inode = file->f_mapping->host;
531 struct file *f = file;
533 /* Avoid recursion */
534 while (is_loop_device(f)) {
535 struct loop_device *l;
537 lockdep_assert_held(&loop_validate_mutex);
538 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
541 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
542 if (l->lo_state != Lo_bound)
544 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
546 f = l->lo_backing_file;
548 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
554 * loop_change_fd switched the backing store of a loopback device to
555 * a new file. This is useful for operating system installers to free up
556 * the original file and in High Availability environments to switch to
557 * an alternative location for the content in case of server meltdown.
558 * This can only work if the loop device is used read-only, and if the
559 * new backing store is the same size and type as the old backing store.
561 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
564 struct file *file = fget(arg);
565 struct file *old_file;
573 /* suppress uevents while reconfiguring the device */
574 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
576 is_loop = is_loop_device(file);
577 error = loop_global_lock_killable(lo, is_loop);
581 if (lo->lo_state != Lo_bound)
584 /* the loop device has to be read-only */
586 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
589 error = loop_validate_file(file, bdev);
593 old_file = lo->lo_backing_file;
597 /* size of the new backing store needs to be the same */
598 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
602 disk_force_media_change(lo->lo_disk);
603 blk_mq_freeze_queue(lo->lo_queue);
604 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
605 lo->lo_backing_file = file;
606 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
607 mapping_set_gfp_mask(file->f_mapping,
608 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
610 blk_mq_unfreeze_queue(lo->lo_queue);
611 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
612 loop_global_unlock(lo, is_loop);
615 * Flush loop_validate_file() before fput(), for l->lo_backing_file
616 * might be pointing at old_file which might be the last reference.
619 mutex_lock(&loop_validate_mutex);
620 mutex_unlock(&loop_validate_mutex);
623 * We must drop file reference outside of lo_mutex as dropping
624 * the file ref can take open_mutex which creates circular locking
629 loop_reread_partitions(lo);
633 /* enable and uncork uevent now that we are done */
634 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
638 loop_global_unlock(lo, is_loop);
644 /* loop sysfs attributes */
646 static ssize_t loop_attr_show(struct device *dev, char *page,
647 ssize_t (*callback)(struct loop_device *, char *))
649 struct gendisk *disk = dev_to_disk(dev);
650 struct loop_device *lo = disk->private_data;
652 return callback(lo, page);
655 #define LOOP_ATTR_RO(_name) \
656 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
657 static ssize_t loop_attr_do_show_##_name(struct device *d, \
658 struct device_attribute *attr, char *b) \
660 return loop_attr_show(d, b, loop_attr_##_name##_show); \
662 static struct device_attribute loop_attr_##_name = \
663 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
665 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
670 spin_lock_irq(&lo->lo_lock);
671 if (lo->lo_backing_file)
672 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
673 spin_unlock_irq(&lo->lo_lock);
675 if (IS_ERR_OR_NULL(p))
679 memmove(buf, p, ret);
687 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
689 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
692 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
694 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
697 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
699 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
701 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
704 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
706 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
708 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
711 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
713 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
715 return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
718 LOOP_ATTR_RO(backing_file);
719 LOOP_ATTR_RO(offset);
720 LOOP_ATTR_RO(sizelimit);
721 LOOP_ATTR_RO(autoclear);
722 LOOP_ATTR_RO(partscan);
725 static struct attribute *loop_attrs[] = {
726 &loop_attr_backing_file.attr,
727 &loop_attr_offset.attr,
728 &loop_attr_sizelimit.attr,
729 &loop_attr_autoclear.attr,
730 &loop_attr_partscan.attr,
735 static struct attribute_group loop_attribute_group = {
740 static void loop_sysfs_init(struct loop_device *lo)
742 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
743 &loop_attribute_group);
746 static void loop_sysfs_exit(struct loop_device *lo)
748 if (lo->sysfs_inited)
749 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
750 &loop_attribute_group);
753 static void loop_config_discard(struct loop_device *lo)
755 struct file *file = lo->lo_backing_file;
756 struct inode *inode = file->f_mapping->host;
757 struct request_queue *q = lo->lo_queue;
758 u32 granularity, max_discard_sectors;
761 * If the backing device is a block device, mirror its zeroing
762 * capability. Set the discard sectors to the block device's zeroing
763 * capabilities because loop discards result in blkdev_issue_zeroout(),
764 * not blkdev_issue_discard(). This maintains consistent behavior with
765 * file-backed loop devices: discarded regions read back as zero.
767 if (S_ISBLK(inode->i_mode)) {
768 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
770 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
771 granularity = bdev_discard_granularity(I_BDEV(inode)) ?:
772 queue_physical_block_size(backingq);
775 * We use punch hole to reclaim the free space used by the
776 * image a.k.a. discard.
778 } else if (!file->f_op->fallocate) {
779 max_discard_sectors = 0;
785 max_discard_sectors = UINT_MAX >> 9;
786 if (!vfs_statfs(&file->f_path, &sbuf))
787 granularity = sbuf.f_bsize;
789 max_discard_sectors = 0;
792 if (max_discard_sectors) {
793 q->limits.discard_granularity = granularity;
794 blk_queue_max_discard_sectors(q, max_discard_sectors);
795 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
797 q->limits.discard_granularity = 0;
798 blk_queue_max_discard_sectors(q, 0);
799 blk_queue_max_write_zeroes_sectors(q, 0);
804 struct rb_node rb_node;
805 struct work_struct work;
806 struct list_head cmd_list;
807 struct list_head idle_list;
808 struct loop_device *lo;
809 struct cgroup_subsys_state *blkcg_css;
810 unsigned long last_ran_at;
813 static void loop_workfn(struct work_struct *work);
815 #ifdef CONFIG_BLK_CGROUP
816 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
818 return !css || css == blkcg_root_css;
821 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
827 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
829 struct rb_node **node, *parent = NULL;
830 struct loop_worker *cur_worker, *worker = NULL;
831 struct work_struct *work;
832 struct list_head *cmd_list;
834 spin_lock_irq(&lo->lo_work_lock);
836 if (queue_on_root_worker(cmd->blkcg_css))
839 node = &lo->worker_tree.rb_node;
843 cur_worker = container_of(*node, struct loop_worker, rb_node);
844 if (cur_worker->blkcg_css == cmd->blkcg_css) {
847 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
848 node = &(*node)->rb_left;
850 node = &(*node)->rb_right;
856 worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
858 * In the event we cannot allocate a worker, just queue on the
859 * rootcg worker and issue the I/O as the rootcg
862 cmd->blkcg_css = NULL;
864 css_put(cmd->memcg_css);
865 cmd->memcg_css = NULL;
869 worker->blkcg_css = cmd->blkcg_css;
870 css_get(worker->blkcg_css);
871 INIT_WORK(&worker->work, loop_workfn);
872 INIT_LIST_HEAD(&worker->cmd_list);
873 INIT_LIST_HEAD(&worker->idle_list);
875 rb_link_node(&worker->rb_node, parent, node);
876 rb_insert_color(&worker->rb_node, &lo->worker_tree);
880 * We need to remove from the idle list here while
881 * holding the lock so that the idle timer doesn't
884 if (!list_empty(&worker->idle_list))
885 list_del_init(&worker->idle_list);
886 work = &worker->work;
887 cmd_list = &worker->cmd_list;
889 work = &lo->rootcg_work;
890 cmd_list = &lo->rootcg_cmd_list;
892 list_add_tail(&cmd->list_entry, cmd_list);
893 queue_work(lo->workqueue, work);
894 spin_unlock_irq(&lo->lo_work_lock);
897 static void loop_set_timer(struct loop_device *lo)
899 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
902 static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
904 struct loop_worker *pos, *worker;
906 spin_lock_irq(&lo->lo_work_lock);
907 list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
910 time_is_after_jiffies(worker->last_ran_at +
911 LOOP_IDLE_WORKER_TIMEOUT))
913 list_del(&worker->idle_list);
914 rb_erase(&worker->rb_node, &lo->worker_tree);
915 css_put(worker->blkcg_css);
918 if (!list_empty(&lo->idle_worker_list))
920 spin_unlock_irq(&lo->lo_work_lock);
923 static void loop_free_idle_workers_timer(struct timer_list *timer)
925 struct loop_device *lo = container_of(timer, struct loop_device, timer);
927 return loop_free_idle_workers(lo, false);
930 static void loop_update_rotational(struct loop_device *lo)
932 struct file *file = lo->lo_backing_file;
933 struct inode *file_inode = file->f_mapping->host;
934 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
935 struct request_queue *q = lo->lo_queue;
938 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
940 nonrot = bdev_nonrot(file_bdev);
943 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
945 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
949 * loop_set_status_from_info - configure device from loop_info
950 * @lo: struct loop_device to configure
951 * @info: struct loop_info64 to configure the device with
953 * Configures the loop device parameters according to the passed
954 * in loop_info64 configuration.
957 loop_set_status_from_info(struct loop_device *lo,
958 const struct loop_info64 *info)
960 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
963 switch (info->lo_encrypt_type) {
967 pr_warn("support for the xor transformation has been removed.\n");
969 case LO_CRYPT_CRYPTOAPI:
970 pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n");
976 /* Avoid assigning overflow values */
977 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
980 lo->lo_offset = info->lo_offset;
981 lo->lo_sizelimit = info->lo_sizelimit;
983 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
984 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
985 lo->lo_flags = info->lo_flags;
989 static int loop_configure(struct loop_device *lo, blk_mode_t mode,
990 struct block_device *bdev,
991 const struct loop_config *config)
993 struct file *file = fget(config->fd);
995 struct address_space *mapping;
999 unsigned short bsize;
1004 is_loop = is_loop_device(file);
1006 /* This is safe, since we have a reference from open(). */
1007 __module_get(THIS_MODULE);
1010 * If we don't hold exclusive handle for the device, upgrade to it
1011 * here to avoid changing device under exclusive owner.
1013 if (!(mode & BLK_OPEN_EXCL)) {
1014 error = bd_prepare_to_claim(bdev, loop_configure, NULL);
1019 error = loop_global_lock_killable(lo, is_loop);
1024 if (lo->lo_state != Lo_unbound)
1027 error = loop_validate_file(file, bdev);
1031 mapping = file->f_mapping;
1032 inode = mapping->host;
1034 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1039 if (config->block_size) {
1040 error = blk_validate_block_size(config->block_size);
1045 error = loop_set_status_from_info(lo, &config->info);
1049 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1050 !file->f_op->write_iter)
1051 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1053 if (!lo->workqueue) {
1054 lo->workqueue = alloc_workqueue("loop%d",
1055 WQ_UNBOUND | WQ_FREEZABLE,
1057 if (!lo->workqueue) {
1063 /* suppress uevents while reconfiguring the device */
1064 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1066 disk_force_media_change(lo->lo_disk);
1067 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1069 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1070 lo->lo_device = bdev;
1071 lo->lo_backing_file = file;
1072 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1073 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1075 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1076 blk_queue_write_cache(lo->lo_queue, true, false);
1078 if (config->block_size)
1079 bsize = config->block_size;
1080 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1081 /* In case of direct I/O, match underlying block size */
1082 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1086 blk_queue_logical_block_size(lo->lo_queue, bsize);
1087 blk_queue_physical_block_size(lo->lo_queue, bsize);
1088 blk_queue_io_min(lo->lo_queue, bsize);
1090 loop_config_discard(lo);
1091 loop_update_rotational(lo);
1092 loop_update_dio(lo);
1093 loop_sysfs_init(lo);
1095 size = get_loop_size(lo, file);
1096 loop_set_size(lo, size);
1098 /* Order wrt reading lo_state in loop_validate_file(). */
1101 lo->lo_state = Lo_bound;
1103 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1104 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1106 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1108 /* enable and uncork uevent now that we are done */
1109 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1111 loop_global_unlock(lo, is_loop);
1113 loop_reread_partitions(lo);
1115 if (!(mode & BLK_OPEN_EXCL))
1116 bd_abort_claiming(bdev, loop_configure);
1121 loop_global_unlock(lo, is_loop);
1123 if (!(mode & BLK_OPEN_EXCL))
1124 bd_abort_claiming(bdev, loop_configure);
1127 /* This is safe: open() is still holding a reference. */
1128 module_put(THIS_MODULE);
1132 static void __loop_clr_fd(struct loop_device *lo, bool release)
1135 gfp_t gfp = lo->old_gfp_mask;
1137 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1138 blk_queue_write_cache(lo->lo_queue, false, false);
1141 * Freeze the request queue when unbinding on a live file descriptor and
1142 * thus an open device. When called from ->release we are guaranteed
1143 * that there is no I/O in progress already.
1146 blk_mq_freeze_queue(lo->lo_queue);
1148 spin_lock_irq(&lo->lo_lock);
1149 filp = lo->lo_backing_file;
1150 lo->lo_backing_file = NULL;
1151 spin_unlock_irq(&lo->lo_lock);
1153 lo->lo_device = NULL;
1155 lo->lo_sizelimit = 0;
1156 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1157 blk_queue_logical_block_size(lo->lo_queue, 512);
1158 blk_queue_physical_block_size(lo->lo_queue, 512);
1159 blk_queue_io_min(lo->lo_queue, 512);
1160 invalidate_disk(lo->lo_disk);
1161 loop_sysfs_exit(lo);
1162 /* let user-space know about this change */
1163 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1164 mapping_set_gfp_mask(filp->f_mapping, gfp);
1165 /* This is safe: open() is still holding a reference. */
1166 module_put(THIS_MODULE);
1168 blk_mq_unfreeze_queue(lo->lo_queue);
1170 disk_force_media_change(lo->lo_disk);
1172 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1176 * open_mutex has been held already in release path, so don't
1177 * acquire it if this function is called in such case.
1179 * If the reread partition isn't from release path, lo_refcnt
1180 * must be at least one and it can only become zero when the
1181 * current holder is released.
1184 mutex_lock(&lo->lo_disk->open_mutex);
1185 err = bdev_disk_changed(lo->lo_disk, false);
1187 mutex_unlock(&lo->lo_disk->open_mutex);
1189 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1190 __func__, lo->lo_number, err);
1191 /* Device is gone, no point in returning error */
1195 * lo->lo_state is set to Lo_unbound here after above partscan has
1196 * finished. There cannot be anybody else entering __loop_clr_fd() as
1197 * Lo_rundown state protects us from all the other places trying to
1198 * change the 'lo' device.
1202 set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1203 mutex_lock(&lo->lo_mutex);
1204 lo->lo_state = Lo_unbound;
1205 mutex_unlock(&lo->lo_mutex);
1208 * Need not hold lo_mutex to fput backing file. Calling fput holding
1209 * lo_mutex triggers a circular lock dependency possibility warning as
1210 * fput can take open_mutex which is usually taken before lo_mutex.
1215 static int loop_clr_fd(struct loop_device *lo)
1220 * Since lo_ioctl() is called without locks held, it is possible that
1221 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1223 * Therefore, use global lock when setting Lo_rundown state in order to
1224 * make sure that loop_validate_file() will fail if the "struct file"
1225 * which loop_configure()/loop_change_fd() found via fget() was this
1228 err = loop_global_lock_killable(lo, true);
1231 if (lo->lo_state != Lo_bound) {
1232 loop_global_unlock(lo, true);
1236 * If we've explicitly asked to tear down the loop device,
1237 * and it has an elevated reference count, set it for auto-teardown when
1238 * the last reference goes away. This stops $!~#$@ udev from
1239 * preventing teardown because it decided that it needs to run blkid on
1240 * the loopback device whenever they appear. xfstests is notorious for
1241 * failing tests because blkid via udev races with a losetup
1242 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1243 * command to fail with EBUSY.
1245 if (disk_openers(lo->lo_disk) > 1) {
1246 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1247 loop_global_unlock(lo, true);
1250 lo->lo_state = Lo_rundown;
1251 loop_global_unlock(lo, true);
1253 __loop_clr_fd(lo, false);
1258 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1262 bool partscan = false;
1263 bool size_changed = false;
1265 err = mutex_lock_killable(&lo->lo_mutex);
1268 if (lo->lo_state != Lo_bound) {
1273 if (lo->lo_offset != info->lo_offset ||
1274 lo->lo_sizelimit != info->lo_sizelimit) {
1275 size_changed = true;
1276 sync_blockdev(lo->lo_device);
1277 invalidate_bdev(lo->lo_device);
1280 /* I/O need to be drained during transfer transition */
1281 blk_mq_freeze_queue(lo->lo_queue);
1283 prev_lo_flags = lo->lo_flags;
1285 err = loop_set_status_from_info(lo, info);
1289 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1290 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1291 /* For those flags, use the previous values instead */
1292 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1293 /* For flags that can't be cleared, use previous values too */
1294 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1297 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1298 lo->lo_backing_file);
1299 loop_set_size(lo, new_size);
1302 /* update dio if lo_offset or transfer is changed */
1303 __loop_update_dio(lo, lo->use_dio);
1306 blk_mq_unfreeze_queue(lo->lo_queue);
1308 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1309 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1310 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1314 mutex_unlock(&lo->lo_mutex);
1316 loop_reread_partitions(lo);
1322 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1328 ret = mutex_lock_killable(&lo->lo_mutex);
1331 if (lo->lo_state != Lo_bound) {
1332 mutex_unlock(&lo->lo_mutex);
1336 memset(info, 0, sizeof(*info));
1337 info->lo_number = lo->lo_number;
1338 info->lo_offset = lo->lo_offset;
1339 info->lo_sizelimit = lo->lo_sizelimit;
1340 info->lo_flags = lo->lo_flags;
1341 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1343 /* Drop lo_mutex while we call into the filesystem. */
1344 path = lo->lo_backing_file->f_path;
1346 mutex_unlock(&lo->lo_mutex);
1347 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1349 info->lo_device = huge_encode_dev(stat.dev);
1350 info->lo_inode = stat.ino;
1351 info->lo_rdevice = huge_encode_dev(stat.rdev);
1358 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1360 memset(info64, 0, sizeof(*info64));
1361 info64->lo_number = info->lo_number;
1362 info64->lo_device = info->lo_device;
1363 info64->lo_inode = info->lo_inode;
1364 info64->lo_rdevice = info->lo_rdevice;
1365 info64->lo_offset = info->lo_offset;
1366 info64->lo_sizelimit = 0;
1367 info64->lo_flags = info->lo_flags;
1368 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1372 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1374 memset(info, 0, sizeof(*info));
1375 info->lo_number = info64->lo_number;
1376 info->lo_device = info64->lo_device;
1377 info->lo_inode = info64->lo_inode;
1378 info->lo_rdevice = info64->lo_rdevice;
1379 info->lo_offset = info64->lo_offset;
1380 info->lo_flags = info64->lo_flags;
1381 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1383 /* error in case values were truncated */
1384 if (info->lo_device != info64->lo_device ||
1385 info->lo_rdevice != info64->lo_rdevice ||
1386 info->lo_inode != info64->lo_inode ||
1387 info->lo_offset != info64->lo_offset)
1394 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1396 struct loop_info info;
1397 struct loop_info64 info64;
1399 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1401 loop_info64_from_old(&info, &info64);
1402 return loop_set_status(lo, &info64);
1406 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1408 struct loop_info64 info64;
1410 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1412 return loop_set_status(lo, &info64);
1416 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1417 struct loop_info info;
1418 struct loop_info64 info64;
1423 err = loop_get_status(lo, &info64);
1425 err = loop_info64_to_old(&info64, &info);
1426 if (!err && copy_to_user(arg, &info, sizeof(info)))
1433 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1434 struct loop_info64 info64;
1439 err = loop_get_status(lo, &info64);
1440 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1446 static int loop_set_capacity(struct loop_device *lo)
1450 if (unlikely(lo->lo_state != Lo_bound))
1453 size = get_loop_size(lo, lo->lo_backing_file);
1454 loop_set_size(lo, size);
1459 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1462 if (lo->lo_state != Lo_bound)
1465 __loop_update_dio(lo, !!arg);
1466 if (lo->use_dio == !!arg)
1473 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1477 if (lo->lo_state != Lo_bound)
1480 err = blk_validate_block_size(arg);
1484 if (lo->lo_queue->limits.logical_block_size == arg)
1487 sync_blockdev(lo->lo_device);
1488 invalidate_bdev(lo->lo_device);
1490 blk_mq_freeze_queue(lo->lo_queue);
1491 blk_queue_logical_block_size(lo->lo_queue, arg);
1492 blk_queue_physical_block_size(lo->lo_queue, arg);
1493 blk_queue_io_min(lo->lo_queue, arg);
1494 loop_update_dio(lo);
1495 blk_mq_unfreeze_queue(lo->lo_queue);
1500 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1505 err = mutex_lock_killable(&lo->lo_mutex);
1509 case LOOP_SET_CAPACITY:
1510 err = loop_set_capacity(lo);
1512 case LOOP_SET_DIRECT_IO:
1513 err = loop_set_dio(lo, arg);
1515 case LOOP_SET_BLOCK_SIZE:
1516 err = loop_set_block_size(lo, arg);
1521 mutex_unlock(&lo->lo_mutex);
1525 static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1526 unsigned int cmd, unsigned long arg)
1528 struct loop_device *lo = bdev->bd_disk->private_data;
1529 void __user *argp = (void __user *) arg;
1535 * Legacy case - pass in a zeroed out struct loop_config with
1536 * only the file descriptor set , which corresponds with the
1537 * default parameters we'd have used otherwise.
1539 struct loop_config config;
1541 memset(&config, 0, sizeof(config));
1544 return loop_configure(lo, mode, bdev, &config);
1546 case LOOP_CONFIGURE: {
1547 struct loop_config config;
1549 if (copy_from_user(&config, argp, sizeof(config)))
1552 return loop_configure(lo, mode, bdev, &config);
1554 case LOOP_CHANGE_FD:
1555 return loop_change_fd(lo, bdev, arg);
1557 return loop_clr_fd(lo);
1558 case LOOP_SET_STATUS:
1560 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1561 err = loop_set_status_old(lo, argp);
1563 case LOOP_GET_STATUS:
1564 return loop_get_status_old(lo, argp);
1565 case LOOP_SET_STATUS64:
1567 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1568 err = loop_set_status64(lo, argp);
1570 case LOOP_GET_STATUS64:
1571 return loop_get_status64(lo, argp);
1572 case LOOP_SET_CAPACITY:
1573 case LOOP_SET_DIRECT_IO:
1574 case LOOP_SET_BLOCK_SIZE:
1575 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1579 err = lo_simple_ioctl(lo, cmd, arg);
1586 #ifdef CONFIG_COMPAT
1587 struct compat_loop_info {
1588 compat_int_t lo_number; /* ioctl r/o */
1589 compat_dev_t lo_device; /* ioctl r/o */
1590 compat_ulong_t lo_inode; /* ioctl r/o */
1591 compat_dev_t lo_rdevice; /* ioctl r/o */
1592 compat_int_t lo_offset;
1593 compat_int_t lo_encrypt_type; /* obsolete, ignored */
1594 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1595 compat_int_t lo_flags; /* ioctl r/o */
1596 char lo_name[LO_NAME_SIZE];
1597 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1598 compat_ulong_t lo_init[2];
1603 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1604 * - noinlined to reduce stack space usage in main part of driver
1607 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1608 struct loop_info64 *info64)
1610 struct compat_loop_info info;
1612 if (copy_from_user(&info, arg, sizeof(info)))
1615 memset(info64, 0, sizeof(*info64));
1616 info64->lo_number = info.lo_number;
1617 info64->lo_device = info.lo_device;
1618 info64->lo_inode = info.lo_inode;
1619 info64->lo_rdevice = info.lo_rdevice;
1620 info64->lo_offset = info.lo_offset;
1621 info64->lo_sizelimit = 0;
1622 info64->lo_flags = info.lo_flags;
1623 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1628 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1629 * - noinlined to reduce stack space usage in main part of driver
1632 loop_info64_to_compat(const struct loop_info64 *info64,
1633 struct compat_loop_info __user *arg)
1635 struct compat_loop_info info;
1637 memset(&info, 0, sizeof(info));
1638 info.lo_number = info64->lo_number;
1639 info.lo_device = info64->lo_device;
1640 info.lo_inode = info64->lo_inode;
1641 info.lo_rdevice = info64->lo_rdevice;
1642 info.lo_offset = info64->lo_offset;
1643 info.lo_flags = info64->lo_flags;
1644 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1646 /* error in case values were truncated */
1647 if (info.lo_device != info64->lo_device ||
1648 info.lo_rdevice != info64->lo_rdevice ||
1649 info.lo_inode != info64->lo_inode ||
1650 info.lo_offset != info64->lo_offset)
1653 if (copy_to_user(arg, &info, sizeof(info)))
1659 loop_set_status_compat(struct loop_device *lo,
1660 const struct compat_loop_info __user *arg)
1662 struct loop_info64 info64;
1665 ret = loop_info64_from_compat(arg, &info64);
1668 return loop_set_status(lo, &info64);
1672 loop_get_status_compat(struct loop_device *lo,
1673 struct compat_loop_info __user *arg)
1675 struct loop_info64 info64;
1680 err = loop_get_status(lo, &info64);
1682 err = loop_info64_to_compat(&info64, arg);
1686 static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1687 unsigned int cmd, unsigned long arg)
1689 struct loop_device *lo = bdev->bd_disk->private_data;
1693 case LOOP_SET_STATUS:
1694 err = loop_set_status_compat(lo,
1695 (const struct compat_loop_info __user *)arg);
1697 case LOOP_GET_STATUS:
1698 err = loop_get_status_compat(lo,
1699 (struct compat_loop_info __user *)arg);
1701 case LOOP_SET_CAPACITY:
1703 case LOOP_GET_STATUS64:
1704 case LOOP_SET_STATUS64:
1705 case LOOP_CONFIGURE:
1706 arg = (unsigned long) compat_ptr(arg);
1709 case LOOP_CHANGE_FD:
1710 case LOOP_SET_BLOCK_SIZE:
1711 case LOOP_SET_DIRECT_IO:
1712 err = lo_ioctl(bdev, mode, cmd, arg);
1722 static void lo_release(struct gendisk *disk)
1724 struct loop_device *lo = disk->private_data;
1726 if (disk_openers(disk) > 0)
1729 mutex_lock(&lo->lo_mutex);
1730 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) {
1731 lo->lo_state = Lo_rundown;
1732 mutex_unlock(&lo->lo_mutex);
1734 * In autoclear mode, stop the loop thread
1735 * and remove configuration after last close.
1737 __loop_clr_fd(lo, true);
1740 mutex_unlock(&lo->lo_mutex);
1743 static void lo_free_disk(struct gendisk *disk)
1745 struct loop_device *lo = disk->private_data;
1748 destroy_workqueue(lo->workqueue);
1749 loop_free_idle_workers(lo, true);
1750 timer_shutdown_sync(&lo->timer);
1751 mutex_destroy(&lo->lo_mutex);
1755 static const struct block_device_operations lo_fops = {
1756 .owner = THIS_MODULE,
1757 .release = lo_release,
1759 #ifdef CONFIG_COMPAT
1760 .compat_ioctl = lo_compat_ioctl,
1762 .free_disk = lo_free_disk,
1766 * And now the modules code and kernel interface.
1770 * If max_loop is specified, create that many devices upfront.
1771 * This also becomes a hard limit. If max_loop is not specified,
1772 * the default isn't a hard limit (as before commit 85c50197716c
1773 * changed the default value from 0 for max_loop=0 reasons), just
1774 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1775 * init time. Loop devices can be requested on-demand with the
1776 * /dev/loop-control interface, or be instantiated by accessing
1777 * a 'dead' device node.
1779 static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1781 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1782 static bool max_loop_specified;
1784 static int max_loop_param_set_int(const char *val,
1785 const struct kernel_param *kp)
1789 ret = param_set_int(val, kp);
1793 max_loop_specified = true;
1797 static const struct kernel_param_ops max_loop_param_ops = {
1798 .set = max_loop_param_set_int,
1799 .get = param_get_int,
1802 module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1803 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1805 module_param(max_loop, int, 0444);
1806 MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1809 module_param(max_part, int, 0444);
1810 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1812 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1814 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1818 ret = kstrtoint(s, 0, &qd);
1823 hw_queue_depth = qd;
1827 static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1828 .set = loop_set_hw_queue_depth,
1829 .get = param_get_int,
1832 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1833 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1835 MODULE_LICENSE("GPL");
1836 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1838 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1839 const struct blk_mq_queue_data *bd)
1841 struct request *rq = bd->rq;
1842 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1843 struct loop_device *lo = rq->q->queuedata;
1845 blk_mq_start_request(rq);
1847 if (lo->lo_state != Lo_bound)
1848 return BLK_STS_IOERR;
1850 switch (req_op(rq)) {
1852 case REQ_OP_DISCARD:
1853 case REQ_OP_WRITE_ZEROES:
1854 cmd->use_aio = false;
1857 cmd->use_aio = lo->use_dio;
1861 /* always use the first bio's css */
1862 cmd->blkcg_css = NULL;
1863 cmd->memcg_css = NULL;
1864 #ifdef CONFIG_BLK_CGROUP
1866 cmd->blkcg_css = bio_blkcg_css(rq->bio);
1868 if (cmd->blkcg_css) {
1870 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1871 &memory_cgrp_subsys);
1876 loop_queue_work(lo, cmd);
1881 static void loop_handle_cmd(struct loop_cmd *cmd)
1883 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1884 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1885 struct request *rq = blk_mq_rq_from_pdu(cmd);
1886 const bool write = op_is_write(req_op(rq));
1887 struct loop_device *lo = rq->q->queuedata;
1889 struct mem_cgroup *old_memcg = NULL;
1890 const bool use_aio = cmd->use_aio;
1892 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1898 kthread_associate_blkcg(cmd_blkcg_css);
1900 old_memcg = set_active_memcg(
1901 mem_cgroup_from_css(cmd_memcg_css));
1904 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1905 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1906 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1907 * not yet been completed.
1909 ret = do_req_filebacked(lo, rq);
1912 kthread_associate_blkcg(NULL);
1914 if (cmd_memcg_css) {
1915 set_active_memcg(old_memcg);
1916 css_put(cmd_memcg_css);
1919 /* complete non-aio request */
1920 if (!use_aio || ret) {
1921 if (ret == -EOPNOTSUPP)
1924 cmd->ret = ret ? -EIO : 0;
1925 if (likely(!blk_should_fake_timeout(rq->q)))
1926 blk_mq_complete_request(rq);
1930 static void loop_process_work(struct loop_worker *worker,
1931 struct list_head *cmd_list, struct loop_device *lo)
1933 int orig_flags = current->flags;
1934 struct loop_cmd *cmd;
1936 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1937 spin_lock_irq(&lo->lo_work_lock);
1938 while (!list_empty(cmd_list)) {
1940 cmd_list->next, struct loop_cmd, list_entry);
1941 list_del(cmd_list->next);
1942 spin_unlock_irq(&lo->lo_work_lock);
1944 loop_handle_cmd(cmd);
1947 spin_lock_irq(&lo->lo_work_lock);
1951 * We only add to the idle list if there are no pending cmds
1952 * *and* the worker will not run again which ensures that it
1953 * is safe to free any worker on the idle list
1955 if (worker && !work_pending(&worker->work)) {
1956 worker->last_ran_at = jiffies;
1957 list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1960 spin_unlock_irq(&lo->lo_work_lock);
1961 current->flags = orig_flags;
1964 static void loop_workfn(struct work_struct *work)
1966 struct loop_worker *worker =
1967 container_of(work, struct loop_worker, work);
1968 loop_process_work(worker, &worker->cmd_list, worker->lo);
1971 static void loop_rootcg_workfn(struct work_struct *work)
1973 struct loop_device *lo =
1974 container_of(work, struct loop_device, rootcg_work);
1975 loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1978 static const struct blk_mq_ops loop_mq_ops = {
1979 .queue_rq = loop_queue_rq,
1980 .complete = lo_complete_rq,
1983 static int loop_add(int i)
1985 struct loop_device *lo;
1986 struct gendisk *disk;
1990 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1993 lo->worker_tree = RB_ROOT;
1994 INIT_LIST_HEAD(&lo->idle_worker_list);
1995 timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
1996 lo->lo_state = Lo_unbound;
1998 err = mutex_lock_killable(&loop_ctl_mutex);
2002 /* allocate id, if @id >= 0, we're requesting that specific id */
2004 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2008 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2010 mutex_unlock(&loop_ctl_mutex);
2015 lo->tag_set.ops = &loop_mq_ops;
2016 lo->tag_set.nr_hw_queues = 1;
2017 lo->tag_set.queue_depth = hw_queue_depth;
2018 lo->tag_set.numa_node = NUMA_NO_NODE;
2019 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2020 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
2021 BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2022 lo->tag_set.driver_data = lo;
2024 err = blk_mq_alloc_tag_set(&lo->tag_set);
2028 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
2030 err = PTR_ERR(disk);
2031 goto out_cleanup_tags;
2033 lo->lo_queue = lo->lo_disk->queue;
2035 /* random number picked from the history block max_sectors cap */
2036 blk_queue_max_hw_sectors(lo->lo_queue, 2560u);
2039 * By default, we do buffer IO, so it doesn't make sense to enable
2040 * merge because the I/O submitted to backing file is handled page by
2041 * page. For directio mode, merge does help to dispatch bigger request
2042 * to underlayer disk. We will enable merge once directio is enabled.
2044 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2047 * Disable partition scanning by default. The in-kernel partition
2048 * scanning can be requested individually per-device during its
2049 * setup. Userspace can always add and remove partitions from all
2050 * devices. The needed partition minors are allocated from the
2051 * extended minor space, the main loop device numbers will continue
2052 * to match the loop minors, regardless of the number of partitions
2055 * If max_part is given, partition scanning is globally enabled for
2056 * all loop devices. The minors for the main loop devices will be
2057 * multiples of max_part.
2059 * Note: Global-for-all-devices, set-only-at-init, read-only module
2060 * parameteters like 'max_loop' and 'max_part' make things needlessly
2061 * complicated, are too static, inflexible and may surprise
2062 * userspace tools. Parameters like this in general should be avoided.
2065 set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2066 mutex_init(&lo->lo_mutex);
2068 spin_lock_init(&lo->lo_lock);
2069 spin_lock_init(&lo->lo_work_lock);
2070 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2071 INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2072 disk->major = LOOP_MAJOR;
2073 disk->first_minor = i << part_shift;
2074 disk->minors = 1 << part_shift;
2075 disk->fops = &lo_fops;
2076 disk->private_data = lo;
2077 disk->queue = lo->lo_queue;
2078 disk->events = DISK_EVENT_MEDIA_CHANGE;
2079 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
2080 sprintf(disk->disk_name, "loop%d", i);
2081 /* Make this loop device reachable from pathname. */
2082 err = add_disk(disk);
2084 goto out_cleanup_disk;
2086 /* Show this loop device. */
2087 mutex_lock(&loop_ctl_mutex);
2088 lo->idr_visible = true;
2089 mutex_unlock(&loop_ctl_mutex);
2096 blk_mq_free_tag_set(&lo->tag_set);
2098 mutex_lock(&loop_ctl_mutex);
2099 idr_remove(&loop_index_idr, i);
2100 mutex_unlock(&loop_ctl_mutex);
2107 static void loop_remove(struct loop_device *lo)
2109 /* Make this loop device unreachable from pathname. */
2110 del_gendisk(lo->lo_disk);
2111 blk_mq_free_tag_set(&lo->tag_set);
2113 mutex_lock(&loop_ctl_mutex);
2114 idr_remove(&loop_index_idr, lo->lo_number);
2115 mutex_unlock(&loop_ctl_mutex);
2117 put_disk(lo->lo_disk);
2120 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2121 static void loop_probe(dev_t dev)
2123 int idx = MINOR(dev) >> part_shift;
2125 if (max_loop_specified && max_loop && idx >= max_loop)
2130 #define loop_probe NULL
2131 #endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2133 static int loop_control_remove(int idx)
2135 struct loop_device *lo;
2139 pr_warn_once("deleting an unspecified loop device is not supported.\n");
2143 /* Hide this loop device for serialization. */
2144 ret = mutex_lock_killable(&loop_ctl_mutex);
2147 lo = idr_find(&loop_index_idr, idx);
2148 if (!lo || !lo->idr_visible)
2151 lo->idr_visible = false;
2152 mutex_unlock(&loop_ctl_mutex);
2156 /* Check whether this loop device can be removed. */
2157 ret = mutex_lock_killable(&lo->lo_mutex);
2160 if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2161 mutex_unlock(&lo->lo_mutex);
2165 /* Mark this loop device as no more bound, but not quite unbound yet */
2166 lo->lo_state = Lo_deleting;
2167 mutex_unlock(&lo->lo_mutex);
2173 /* Show this loop device again. */
2174 mutex_lock(&loop_ctl_mutex);
2175 lo->idr_visible = true;
2176 mutex_unlock(&loop_ctl_mutex);
2180 static int loop_control_get_free(int idx)
2182 struct loop_device *lo;
2185 ret = mutex_lock_killable(&loop_ctl_mutex);
2188 idr_for_each_entry(&loop_index_idr, lo, id) {
2189 /* Hitting a race results in creating a new loop device which is harmless. */
2190 if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2193 mutex_unlock(&loop_ctl_mutex);
2194 return loop_add(-1);
2196 mutex_unlock(&loop_ctl_mutex);
2200 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2205 return loop_add(parm);
2206 case LOOP_CTL_REMOVE:
2207 return loop_control_remove(parm);
2208 case LOOP_CTL_GET_FREE:
2209 return loop_control_get_free(parm);
2215 static const struct file_operations loop_ctl_fops = {
2216 .open = nonseekable_open,
2217 .unlocked_ioctl = loop_control_ioctl,
2218 .compat_ioctl = loop_control_ioctl,
2219 .owner = THIS_MODULE,
2220 .llseek = noop_llseek,
2223 static struct miscdevice loop_misc = {
2224 .minor = LOOP_CTRL_MINOR,
2225 .name = "loop-control",
2226 .fops = &loop_ctl_fops,
2229 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2230 MODULE_ALIAS("devname:loop-control");
2232 static int __init loop_init(void)
2239 part_shift = fls(max_part);
2242 * Adjust max_part according to part_shift as it is exported
2243 * to user space so that user can decide correct minor number
2244 * if [s]he want to create more devices.
2246 * Note that -1 is required because partition 0 is reserved
2247 * for the whole disk.
2249 max_part = (1UL << part_shift) - 1;
2252 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2257 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2262 err = misc_register(&loop_misc);
2267 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2272 /* pre-create number of devices given by config or max_loop */
2273 for (i = 0; i < max_loop; i++)
2276 printk(KERN_INFO "loop: module loaded\n");
2280 misc_deregister(&loop_misc);
2285 static void __exit loop_exit(void)
2287 struct loop_device *lo;
2290 unregister_blkdev(LOOP_MAJOR, "loop");
2291 misc_deregister(&loop_misc);
2294 * There is no need to use loop_ctl_mutex here, for nobody else can
2295 * access loop_index_idr when this module is unloading (unless forced
2296 * module unloading is requested). If this is not a clean unloading,
2297 * we have no means to avoid kernel crash.
2299 idr_for_each_entry(&loop_index_idr, lo, id)
2302 idr_destroy(&loop_index_idr);
2305 module_init(loop_init);
2306 module_exit(loop_exit);
2309 static int __init max_loop_setup(char *str)
2311 max_loop = simple_strtol(str, NULL, 0);
2312 #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2313 max_loop_specified = true;
2318 __setup("max_loop=", max_loop_setup);