2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
39 * Support up to 256 loop devices
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include <linux/ioprio.h>
80 #include <linux/blk-cgroup.h>
84 #include <linux/uaccess.h>
86 static DEFINE_IDR(loop_index_idr);
87 static DEFINE_MUTEX(loop_ctl_mutex);
90 static int part_shift;
92 static int transfer_xor(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
97 char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page) + loop_off;
110 key = lo->lo_encrypt_key;
111 keysize = lo->lo_encrypt_key_size;
112 for (i = 0; i < size; i++)
113 *out++ = *in++ ^ key[(i & 511) % keysize];
115 kunmap_atomic(loop_buf);
116 kunmap_atomic(raw_buf);
121 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
123 if (unlikely(info->lo_encrypt_key_size <= 0))
128 static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
132 static struct loop_func_table xor_funcs = {
133 .number = LO_CRYPT_XOR,
134 .transfer = transfer_xor,
138 /* xfer_funcs[0] is special - its release function is never called */
139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
144 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
148 /* Compute loopsize in bytes */
149 loopsize = i_size_read(file->f_mapping->host);
152 /* offset is beyond i_size, weird but possible */
156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
162 return loopsize >> 9;
165 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
167 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
170 static void __loop_update_dio(struct loop_device *lo, bool dio)
172 struct file *file = lo->lo_backing_file;
173 struct address_space *mapping = file->f_mapping;
174 struct inode *inode = mapping->host;
175 unsigned short sb_bsize = 0;
176 unsigned dio_align = 0;
179 if (inode->i_sb->s_bdev) {
180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 dio_align = sb_bsize - 1;
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 !(lo->lo_offset & dio_align) &&
197 mapping->a_ops->direct_IO &&
206 if (lo->use_dio == use_dio)
209 /* flush dirty pages before changing direct IO */
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
217 if (lo->lo_state == Lo_bound)
218 blk_mq_freeze_queue(lo->lo_queue);
219 lo->use_dio = use_dio;
221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
222 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
225 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
227 if (lo->lo_state == Lo_bound)
228 blk_mq_unfreeze_queue(lo->lo_queue);
232 * loop_validate_block_size() - validates the passed in block size
233 * @bsize: size to validate
236 loop_validate_block_size(unsigned short bsize)
238 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
245 * loop_set_size() - sets device size and notifies userspace
246 * @lo: struct loop_device to set the size for
247 * @size: new size of the loop device
249 * Callers must validate that the size passed into this function fits into
250 * a sector_t, eg using loop_validate_size()
252 static void loop_set_size(struct loop_device *lo, loff_t size)
254 if (!set_capacity_and_notify(lo->lo_disk, size))
255 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
259 lo_do_transfer(struct loop_device *lo, int cmd,
260 struct page *rpage, unsigned roffs,
261 struct page *lpage, unsigned loffs,
262 int size, sector_t rblock)
266 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
270 printk_ratelimited(KERN_ERR
271 "loop: Transfer error at byte offset %llu, length %i.\n",
272 (unsigned long long)rblock << 9, size);
276 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
281 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
283 file_start_write(file);
284 bw = vfs_iter_write(file, &i, ppos, 0);
285 file_end_write(file);
287 if (likely(bw == bvec->bv_len))
290 printk_ratelimited(KERN_ERR
291 "loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)*ppos, bvec->bv_len);
298 static int lo_write_simple(struct loop_device *lo, struct request *rq,
302 struct req_iterator iter;
305 rq_for_each_segment(bvec, rq, iter) {
306 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
316 * This is the slow, transforming version that needs to double buffer the
317 * data as it cannot do the transformations in place without having direct
318 * access to the destination pages of the backing file.
320 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
323 struct bio_vec bvec, b;
324 struct req_iterator iter;
328 page = alloc_page(GFP_NOIO);
332 rq_for_each_segment(bvec, rq, iter) {
333 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
334 bvec.bv_offset, bvec.bv_len, pos >> 9);
340 b.bv_len = bvec.bv_len;
341 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
350 static int lo_read_simple(struct loop_device *lo, struct request *rq,
354 struct req_iterator iter;
358 rq_for_each_segment(bvec, rq, iter) {
359 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
360 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
364 flush_dcache_page(bvec.bv_page);
366 if (len != bvec.bv_len) {
369 __rq_for_each_bio(bio, rq)
379 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
382 struct bio_vec bvec, b;
383 struct req_iterator iter;
389 page = alloc_page(GFP_NOIO);
393 rq_for_each_segment(bvec, rq, iter) {
398 b.bv_len = bvec.bv_len;
400 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
401 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
407 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
408 bvec.bv_offset, len, offset >> 9);
412 flush_dcache_page(bvec.bv_page);
414 if (len != bvec.bv_len) {
417 __rq_for_each_bio(bio, rq)
429 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
433 * We use fallocate to manipulate the space mappings used by the image
434 * a.k.a. discard/zerorange. However we do not support this if
435 * encryption is enabled, because it may give an attacker useful
438 struct file *file = lo->lo_backing_file;
439 struct request_queue *q = lo->lo_queue;
442 mode |= FALLOC_FL_KEEP_SIZE;
444 if (!blk_queue_discard(q)) {
449 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
450 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
456 static int lo_req_flush(struct loop_device *lo, struct request *rq)
458 struct file *file = lo->lo_backing_file;
459 int ret = vfs_fsync(file, 0);
460 if (unlikely(ret && ret != -EINVAL))
466 static void lo_complete_rq(struct request *rq)
468 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
469 blk_status_t ret = BLK_STS_OK;
471 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
472 req_op(rq) != REQ_OP_READ) {
474 ret = errno_to_blk_status(cmd->ret);
479 * Short READ - if we got some data, advance our request and
480 * retry it. If we got no data, end the rest with EIO.
483 blk_update_request(rq, BLK_STS_OK, cmd->ret);
485 blk_mq_requeue_request(rq, true);
488 struct bio *bio = rq->bio;
497 blk_mq_end_request(rq, ret);
501 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
503 struct request *rq = blk_mq_rq_from_pdu(cmd);
505 if (!atomic_dec_and_test(&cmd->ref))
509 if (likely(!blk_should_fake_timeout(rq->q)))
510 blk_mq_complete_request(rq);
513 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
515 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
520 lo_rw_aio_do_completion(cmd);
523 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
526 struct iov_iter iter;
527 struct req_iterator rq_iter;
528 struct bio_vec *bvec;
529 struct request *rq = blk_mq_rq_from_pdu(cmd);
530 struct bio *bio = rq->bio;
531 struct file *file = lo->lo_backing_file;
537 rq_for_each_bvec(tmp, rq, rq_iter)
540 if (rq->bio != rq->biotail) {
542 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
549 * The bios of the request may be started from the middle of
550 * the 'bvec' because of bio splitting, so we can't directly
551 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
552 * API will take care of all details for us.
554 rq_for_each_bvec(tmp, rq, rq_iter) {
562 * Same here, this bio may be started from the middle of the
563 * 'bvec' because of bio splitting, so offset from the bvec
564 * must be passed to iov iterator
566 offset = bio->bi_iter.bi_bvec_done;
567 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
569 atomic_set(&cmd->ref, 2);
571 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
572 iter.iov_offset = offset;
574 cmd->iocb.ki_pos = pos;
575 cmd->iocb.ki_filp = file;
576 cmd->iocb.ki_complete = lo_rw_aio_complete;
577 cmd->iocb.ki_flags = IOCB_DIRECT;
578 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
580 kthread_associate_blkcg(cmd->css);
583 ret = call_write_iter(file, &cmd->iocb, &iter);
585 ret = call_read_iter(file, &cmd->iocb, &iter);
587 lo_rw_aio_do_completion(cmd);
588 kthread_associate_blkcg(NULL);
590 if (ret != -EIOCBQUEUED)
591 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
595 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
597 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
598 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
601 * lo_write_simple and lo_read_simple should have been covered
602 * by io submit style function like lo_rw_aio(), one blocker
603 * is that lo_read_simple() need to call flush_dcache_page after
604 * the page is written from kernel, and it isn't easy to handle
605 * this in io submit style function which submits all segments
606 * of the req at one time. And direct read IO doesn't need to
607 * run flush_dcache_page().
609 switch (req_op(rq)) {
611 return lo_req_flush(lo, rq);
612 case REQ_OP_WRITE_ZEROES:
614 * If the caller doesn't want deallocation, call zeroout to
615 * write zeroes the range. Otherwise, punch them out.
617 return lo_fallocate(lo, rq, pos,
618 (rq->cmd_flags & REQ_NOUNMAP) ?
619 FALLOC_FL_ZERO_RANGE :
620 FALLOC_FL_PUNCH_HOLE);
622 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
625 return lo_write_transfer(lo, rq, pos);
626 else if (cmd->use_aio)
627 return lo_rw_aio(lo, cmd, pos, WRITE);
629 return lo_write_simple(lo, rq, pos);
632 return lo_read_transfer(lo, rq, pos);
633 else if (cmd->use_aio)
634 return lo_rw_aio(lo, cmd, pos, READ);
636 return lo_read_simple(lo, rq, pos);
643 static inline void loop_update_dio(struct loop_device *lo)
645 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
649 static void loop_reread_partitions(struct loop_device *lo,
650 struct block_device *bdev)
654 mutex_lock(&bdev->bd_mutex);
655 rc = bdev_disk_changed(bdev, false);
656 mutex_unlock(&bdev->bd_mutex);
658 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
659 __func__, lo->lo_number, lo->lo_file_name, rc);
662 static inline int is_loop_device(struct file *file)
664 struct inode *i = file->f_mapping->host;
666 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
669 static int loop_validate_file(struct file *file, struct block_device *bdev)
671 struct inode *inode = file->f_mapping->host;
672 struct file *f = file;
674 /* Avoid recursion */
675 while (is_loop_device(f)) {
676 struct loop_device *l;
678 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
681 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
682 if (l->lo_state != Lo_bound) {
685 f = l->lo_backing_file;
687 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
693 * loop_change_fd switched the backing store of a loopback device to
694 * a new file. This is useful for operating system installers to free up
695 * the original file and in High Availability environments to switch to
696 * an alternative location for the content in case of server meltdown.
697 * This can only work if the loop device is used read-only, and if the
698 * new backing store is the same size and type as the old backing store.
700 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
703 struct file *file = NULL, *old_file;
707 error = mutex_lock_killable(&lo->lo_mutex);
711 if (lo->lo_state != Lo_bound)
714 /* the loop device has to be read-only */
716 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
724 error = loop_validate_file(file, bdev);
728 old_file = lo->lo_backing_file;
732 /* size of the new backing store needs to be the same */
733 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
737 blk_mq_freeze_queue(lo->lo_queue);
738 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
739 lo->lo_backing_file = file;
740 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
741 mapping_set_gfp_mask(file->f_mapping,
742 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
744 blk_mq_unfreeze_queue(lo->lo_queue);
745 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
746 mutex_unlock(&lo->lo_mutex);
748 * We must drop file reference outside of lo_mutex as dropping
749 * the file ref can take bd_mutex which creates circular locking
754 loop_reread_partitions(lo, bdev);
758 mutex_unlock(&lo->lo_mutex);
764 /* loop sysfs attributes */
766 static ssize_t loop_attr_show(struct device *dev, char *page,
767 ssize_t (*callback)(struct loop_device *, char *))
769 struct gendisk *disk = dev_to_disk(dev);
770 struct loop_device *lo = disk->private_data;
772 return callback(lo, page);
775 #define LOOP_ATTR_RO(_name) \
776 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
777 static ssize_t loop_attr_do_show_##_name(struct device *d, \
778 struct device_attribute *attr, char *b) \
780 return loop_attr_show(d, b, loop_attr_##_name##_show); \
782 static struct device_attribute loop_attr_##_name = \
783 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
785 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
790 spin_lock_irq(&lo->lo_lock);
791 if (lo->lo_backing_file)
792 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
793 spin_unlock_irq(&lo->lo_lock);
795 if (IS_ERR_OR_NULL(p))
799 memmove(buf, p, ret);
807 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
809 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
812 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
814 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
817 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
819 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
821 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
824 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
826 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
828 return sprintf(buf, "%s\n", partscan ? "1" : "0");
831 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
833 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
835 return sprintf(buf, "%s\n", dio ? "1" : "0");
838 LOOP_ATTR_RO(backing_file);
839 LOOP_ATTR_RO(offset);
840 LOOP_ATTR_RO(sizelimit);
841 LOOP_ATTR_RO(autoclear);
842 LOOP_ATTR_RO(partscan);
845 static struct attribute *loop_attrs[] = {
846 &loop_attr_backing_file.attr,
847 &loop_attr_offset.attr,
848 &loop_attr_sizelimit.attr,
849 &loop_attr_autoclear.attr,
850 &loop_attr_partscan.attr,
855 static struct attribute_group loop_attribute_group = {
860 static void loop_sysfs_init(struct loop_device *lo)
862 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
863 &loop_attribute_group);
866 static void loop_sysfs_exit(struct loop_device *lo)
868 if (lo->sysfs_inited)
869 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
870 &loop_attribute_group);
873 static void loop_config_discard(struct loop_device *lo)
875 struct file *file = lo->lo_backing_file;
876 struct inode *inode = file->f_mapping->host;
877 struct request_queue *q = lo->lo_queue;
878 u32 granularity, max_discard_sectors;
881 * If the backing device is a block device, mirror its zeroing
882 * capability. Set the discard sectors to the block device's zeroing
883 * capabilities because loop discards result in blkdev_issue_zeroout(),
884 * not blkdev_issue_discard(). This maintains consistent behavior with
885 * file-backed loop devices: discarded regions read back as zero.
887 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
888 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
890 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
891 granularity = backingq->limits.discard_granularity ?:
892 queue_physical_block_size(backingq);
895 * We use punch hole to reclaim the free space used by the
896 * image a.k.a. discard. However we do not support discard if
897 * encryption is enabled, because it may give an attacker
898 * useful information.
900 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
901 max_discard_sectors = 0;
905 max_discard_sectors = UINT_MAX >> 9;
906 granularity = inode->i_sb->s_blocksize;
909 if (max_discard_sectors) {
910 q->limits.discard_granularity = granularity;
911 blk_queue_max_discard_sectors(q, max_discard_sectors);
912 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
913 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
915 q->limits.discard_granularity = 0;
916 blk_queue_max_discard_sectors(q, 0);
917 blk_queue_max_write_zeroes_sectors(q, 0);
918 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
920 q->limits.discard_alignment = 0;
923 static void loop_unprepare_queue(struct loop_device *lo)
925 kthread_flush_worker(&lo->worker);
926 kthread_stop(lo->worker_task);
929 static int loop_kthread_worker_fn(void *worker_ptr)
931 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
932 return kthread_worker_fn(worker_ptr);
935 static int loop_prepare_queue(struct loop_device *lo)
937 kthread_init_worker(&lo->worker);
938 lo->worker_task = kthread_run(loop_kthread_worker_fn,
939 &lo->worker, "loop%d", lo->lo_number);
940 if (IS_ERR(lo->worker_task))
942 set_user_nice(lo->worker_task, MIN_NICE);
946 static void loop_update_rotational(struct loop_device *lo)
948 struct file *file = lo->lo_backing_file;
949 struct inode *file_inode = file->f_mapping->host;
950 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
951 struct request_queue *q = lo->lo_queue;
954 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
956 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
959 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
961 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
965 loop_release_xfer(struct loop_device *lo)
968 struct loop_func_table *xfer = lo->lo_encryption;
972 err = xfer->release(lo);
974 lo->lo_encryption = NULL;
975 module_put(xfer->owner);
981 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
982 const struct loop_info64 *i)
987 struct module *owner = xfer->owner;
989 if (!try_module_get(owner))
992 err = xfer->init(lo, i);
996 lo->lo_encryption = xfer;
1002 * loop_set_status_from_info - configure device from loop_info
1003 * @lo: struct loop_device to configure
1004 * @info: struct loop_info64 to configure the device with
1006 * Configures the loop device parameters according to the passed
1007 * in loop_info64 configuration.
1010 loop_set_status_from_info(struct loop_device *lo,
1011 const struct loop_info64 *info)
1014 struct loop_func_table *xfer;
1015 kuid_t uid = current_uid();
1017 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1020 err = loop_release_xfer(lo);
1024 if (info->lo_encrypt_type) {
1025 unsigned int type = info->lo_encrypt_type;
1027 if (type >= MAX_LO_CRYPT)
1029 xfer = xfer_funcs[type];
1035 err = loop_init_xfer(lo, xfer, info);
1039 lo->lo_offset = info->lo_offset;
1040 lo->lo_sizelimit = info->lo_sizelimit;
1041 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1042 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1043 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1044 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1048 lo->transfer = xfer->transfer;
1049 lo->ioctl = xfer->ioctl;
1051 lo->lo_flags = info->lo_flags;
1053 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1054 lo->lo_init[0] = info->lo_init[0];
1055 lo->lo_init[1] = info->lo_init[1];
1056 if (info->lo_encrypt_key_size) {
1057 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1058 info->lo_encrypt_key_size);
1059 lo->lo_key_owner = uid;
1065 static int loop_configure(struct loop_device *lo, fmode_t mode,
1066 struct block_device *bdev,
1067 const struct loop_config *config)
1070 struct inode *inode;
1071 struct address_space *mapping;
1075 unsigned short bsize;
1077 /* This is safe, since we have a reference from open(). */
1078 __module_get(THIS_MODULE);
1081 file = fget(config->fd);
1086 * If we don't hold exclusive handle for the device, upgrade to it
1087 * here to avoid changing device under exclusive owner.
1089 if (!(mode & FMODE_EXCL)) {
1090 error = bd_prepare_to_claim(bdev, loop_configure);
1095 error = mutex_lock_killable(&lo->lo_mutex);
1100 if (lo->lo_state != Lo_unbound)
1103 error = loop_validate_file(file, bdev);
1107 mapping = file->f_mapping;
1108 inode = mapping->host;
1110 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1115 if (config->block_size) {
1116 error = loop_validate_block_size(config->block_size);
1121 error = loop_set_status_from_info(lo, &config->info);
1125 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1126 !file->f_op->write_iter)
1127 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1129 error = loop_prepare_queue(lo);
1133 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1135 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1136 lo->lo_device = bdev;
1137 lo->lo_backing_file = file;
1138 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1139 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1141 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1142 blk_queue_write_cache(lo->lo_queue, true, false);
1144 if (config->block_size)
1145 bsize = config->block_size;
1146 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1147 /* In case of direct I/O, match underlying block size */
1148 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1152 blk_queue_logical_block_size(lo->lo_queue, bsize);
1153 blk_queue_physical_block_size(lo->lo_queue, bsize);
1154 blk_queue_io_min(lo->lo_queue, bsize);
1156 loop_update_rotational(lo);
1157 loop_update_dio(lo);
1158 loop_sysfs_init(lo);
1160 size = get_loop_size(lo, file);
1161 loop_set_size(lo, size);
1163 lo->lo_state = Lo_bound;
1165 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1166 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1168 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1170 /* Grab the block_device to prevent its destruction after we
1171 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1174 mutex_unlock(&lo->lo_mutex);
1176 loop_reread_partitions(lo, bdev);
1177 if (!(mode & FMODE_EXCL))
1178 bd_abort_claiming(bdev, loop_configure);
1182 mutex_unlock(&lo->lo_mutex);
1184 if (!(mode & FMODE_EXCL))
1185 bd_abort_claiming(bdev, loop_configure);
1189 /* This is safe: open() is still holding a reference. */
1190 module_put(THIS_MODULE);
1194 static int __loop_clr_fd(struct loop_device *lo, bool release)
1196 struct file *filp = NULL;
1197 gfp_t gfp = lo->old_gfp_mask;
1198 struct block_device *bdev = lo->lo_device;
1200 bool partscan = false;
1203 mutex_lock(&lo->lo_mutex);
1204 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1209 filp = lo->lo_backing_file;
1215 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1216 blk_queue_write_cache(lo->lo_queue, false, false);
1218 /* freeze request queue during the transition */
1219 blk_mq_freeze_queue(lo->lo_queue);
1221 spin_lock_irq(&lo->lo_lock);
1222 lo->lo_backing_file = NULL;
1223 spin_unlock_irq(&lo->lo_lock);
1225 loop_release_xfer(lo);
1226 lo->transfer = NULL;
1228 lo->lo_device = NULL;
1229 lo->lo_encryption = NULL;
1231 lo->lo_sizelimit = 0;
1232 lo->lo_encrypt_key_size = 0;
1233 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1234 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1235 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1236 blk_queue_logical_block_size(lo->lo_queue, 512);
1237 blk_queue_physical_block_size(lo->lo_queue, 512);
1238 blk_queue_io_min(lo->lo_queue, 512);
1241 invalidate_bdev(bdev);
1242 bdev->bd_inode->i_mapping->wb_err = 0;
1244 set_capacity(lo->lo_disk, 0);
1245 loop_sysfs_exit(lo);
1247 /* let user-space know about this change */
1248 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1250 mapping_set_gfp_mask(filp->f_mapping, gfp);
1251 /* This is safe: open() is still holding a reference. */
1252 module_put(THIS_MODULE);
1253 blk_mq_unfreeze_queue(lo->lo_queue);
1255 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1256 lo_number = lo->lo_number;
1257 loop_unprepare_queue(lo);
1259 mutex_unlock(&lo->lo_mutex);
1262 * bd_mutex has been held already in release path, so don't
1263 * acquire it if this function is called in such case.
1265 * If the reread partition isn't from release path, lo_refcnt
1266 * must be at least one and it can only become zero when the
1267 * current holder is released.
1270 mutex_lock(&bdev->bd_mutex);
1271 err = bdev_disk_changed(bdev, false);
1273 mutex_unlock(&bdev->bd_mutex);
1275 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1276 __func__, lo_number, err);
1277 /* Device is gone, no point in returning error */
1282 * lo->lo_state is set to Lo_unbound here after above partscan has
1285 * There cannot be anybody else entering __loop_clr_fd() as
1286 * lo->lo_backing_file is already cleared and Lo_rundown state
1287 * protects us from all the other places trying to change the 'lo'
1290 mutex_lock(&lo->lo_mutex);
1293 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1294 lo->lo_state = Lo_unbound;
1295 mutex_unlock(&lo->lo_mutex);
1298 * Need not hold lo_mutex to fput backing file. Calling fput holding
1299 * lo_mutex triggers a circular lock dependency possibility warning as
1300 * fput can take bd_mutex which is usually taken before lo_mutex.
1307 static int loop_clr_fd(struct loop_device *lo)
1311 err = mutex_lock_killable(&lo->lo_mutex);
1314 if (lo->lo_state != Lo_bound) {
1315 mutex_unlock(&lo->lo_mutex);
1319 * If we've explicitly asked to tear down the loop device,
1320 * and it has an elevated reference count, set it for auto-teardown when
1321 * the last reference goes away. This stops $!~#$@ udev from
1322 * preventing teardown because it decided that it needs to run blkid on
1323 * the loopback device whenever they appear. xfstests is notorious for
1324 * failing tests because blkid via udev races with a losetup
1325 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1326 * command to fail with EBUSY.
1328 if (atomic_read(&lo->lo_refcnt) > 1) {
1329 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1330 mutex_unlock(&lo->lo_mutex);
1333 lo->lo_state = Lo_rundown;
1334 mutex_unlock(&lo->lo_mutex);
1336 return __loop_clr_fd(lo, false);
1340 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1343 struct block_device *bdev;
1344 kuid_t uid = current_uid();
1346 bool partscan = false;
1347 bool size_changed = false;
1349 err = mutex_lock_killable(&lo->lo_mutex);
1352 if (lo->lo_encrypt_key_size &&
1353 !uid_eq(lo->lo_key_owner, uid) &&
1354 !capable(CAP_SYS_ADMIN)) {
1358 if (lo->lo_state != Lo_bound) {
1363 if (lo->lo_offset != info->lo_offset ||
1364 lo->lo_sizelimit != info->lo_sizelimit) {
1365 size_changed = true;
1366 sync_blockdev(lo->lo_device);
1367 invalidate_bdev(lo->lo_device);
1370 /* I/O need to be drained during transfer transition */
1371 blk_mq_freeze_queue(lo->lo_queue);
1373 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1374 /* If any pages were dirtied after invalidate_bdev(), try again */
1376 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1377 __func__, lo->lo_number, lo->lo_file_name,
1378 lo->lo_device->bd_inode->i_mapping->nrpages);
1382 prev_lo_flags = lo->lo_flags;
1384 err = loop_set_status_from_info(lo, info);
1388 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1389 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1390 /* For those flags, use the previous values instead */
1391 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1392 /* For flags that can't be cleared, use previous values too */
1393 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1396 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1397 lo->lo_backing_file);
1398 loop_set_size(lo, new_size);
1401 loop_config_discard(lo);
1403 /* update dio if lo_offset or transfer is changed */
1404 __loop_update_dio(lo, lo->use_dio);
1407 blk_mq_unfreeze_queue(lo->lo_queue);
1409 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1410 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1411 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1412 bdev = lo->lo_device;
1416 mutex_unlock(&lo->lo_mutex);
1418 loop_reread_partitions(lo, bdev);
1424 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1430 ret = mutex_lock_killable(&lo->lo_mutex);
1433 if (lo->lo_state != Lo_bound) {
1434 mutex_unlock(&lo->lo_mutex);
1438 memset(info, 0, sizeof(*info));
1439 info->lo_number = lo->lo_number;
1440 info->lo_offset = lo->lo_offset;
1441 info->lo_sizelimit = lo->lo_sizelimit;
1442 info->lo_flags = lo->lo_flags;
1443 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1444 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1445 info->lo_encrypt_type =
1446 lo->lo_encryption ? lo->lo_encryption->number : 0;
1447 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1448 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1449 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1450 lo->lo_encrypt_key_size);
1453 /* Drop lo_mutex while we call into the filesystem. */
1454 path = lo->lo_backing_file->f_path;
1456 mutex_unlock(&lo->lo_mutex);
1457 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1459 info->lo_device = huge_encode_dev(stat.dev);
1460 info->lo_inode = stat.ino;
1461 info->lo_rdevice = huge_encode_dev(stat.rdev);
1468 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1470 memset(info64, 0, sizeof(*info64));
1471 info64->lo_number = info->lo_number;
1472 info64->lo_device = info->lo_device;
1473 info64->lo_inode = info->lo_inode;
1474 info64->lo_rdevice = info->lo_rdevice;
1475 info64->lo_offset = info->lo_offset;
1476 info64->lo_sizelimit = 0;
1477 info64->lo_encrypt_type = info->lo_encrypt_type;
1478 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1479 info64->lo_flags = info->lo_flags;
1480 info64->lo_init[0] = info->lo_init[0];
1481 info64->lo_init[1] = info->lo_init[1];
1482 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1483 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1485 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1486 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1490 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1492 memset(info, 0, sizeof(*info));
1493 info->lo_number = info64->lo_number;
1494 info->lo_device = info64->lo_device;
1495 info->lo_inode = info64->lo_inode;
1496 info->lo_rdevice = info64->lo_rdevice;
1497 info->lo_offset = info64->lo_offset;
1498 info->lo_encrypt_type = info64->lo_encrypt_type;
1499 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1500 info->lo_flags = info64->lo_flags;
1501 info->lo_init[0] = info64->lo_init[0];
1502 info->lo_init[1] = info64->lo_init[1];
1503 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1504 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1506 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1507 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1509 /* error in case values were truncated */
1510 if (info->lo_device != info64->lo_device ||
1511 info->lo_rdevice != info64->lo_rdevice ||
1512 info->lo_inode != info64->lo_inode ||
1513 info->lo_offset != info64->lo_offset)
1520 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1522 struct loop_info info;
1523 struct loop_info64 info64;
1525 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1527 loop_info64_from_old(&info, &info64);
1528 return loop_set_status(lo, &info64);
1532 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1534 struct loop_info64 info64;
1536 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1538 return loop_set_status(lo, &info64);
1542 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1543 struct loop_info info;
1544 struct loop_info64 info64;
1549 err = loop_get_status(lo, &info64);
1551 err = loop_info64_to_old(&info64, &info);
1552 if (!err && copy_to_user(arg, &info, sizeof(info)))
1559 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1560 struct loop_info64 info64;
1565 err = loop_get_status(lo, &info64);
1566 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1572 static int loop_set_capacity(struct loop_device *lo)
1576 if (unlikely(lo->lo_state != Lo_bound))
1579 size = get_loop_size(lo, lo->lo_backing_file);
1580 loop_set_size(lo, size);
1585 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1588 if (lo->lo_state != Lo_bound)
1591 __loop_update_dio(lo, !!arg);
1592 if (lo->use_dio == !!arg)
1599 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1603 if (lo->lo_state != Lo_bound)
1606 err = loop_validate_block_size(arg);
1610 if (lo->lo_queue->limits.logical_block_size == arg)
1613 sync_blockdev(lo->lo_device);
1614 invalidate_bdev(lo->lo_device);
1616 blk_mq_freeze_queue(lo->lo_queue);
1618 /* invalidate_bdev should have truncated all the pages */
1619 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1621 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1622 __func__, lo->lo_number, lo->lo_file_name,
1623 lo->lo_device->bd_inode->i_mapping->nrpages);
1627 blk_queue_logical_block_size(lo->lo_queue, arg);
1628 blk_queue_physical_block_size(lo->lo_queue, arg);
1629 blk_queue_io_min(lo->lo_queue, arg);
1630 loop_update_dio(lo);
1632 blk_mq_unfreeze_queue(lo->lo_queue);
1637 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1642 err = mutex_lock_killable(&lo->lo_mutex);
1646 case LOOP_SET_CAPACITY:
1647 err = loop_set_capacity(lo);
1649 case LOOP_SET_DIRECT_IO:
1650 err = loop_set_dio(lo, arg);
1652 case LOOP_SET_BLOCK_SIZE:
1653 err = loop_set_block_size(lo, arg);
1656 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1658 mutex_unlock(&lo->lo_mutex);
1662 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1663 unsigned int cmd, unsigned long arg)
1665 struct loop_device *lo = bdev->bd_disk->private_data;
1666 void __user *argp = (void __user *) arg;
1672 * Legacy case - pass in a zeroed out struct loop_config with
1673 * only the file descriptor set , which corresponds with the
1674 * default parameters we'd have used otherwise.
1676 struct loop_config config;
1678 memset(&config, 0, sizeof(config));
1681 return loop_configure(lo, mode, bdev, &config);
1683 case LOOP_CONFIGURE: {
1684 struct loop_config config;
1686 if (copy_from_user(&config, argp, sizeof(config)))
1689 return loop_configure(lo, mode, bdev, &config);
1691 case LOOP_CHANGE_FD:
1692 return loop_change_fd(lo, bdev, arg);
1694 return loop_clr_fd(lo);
1695 case LOOP_SET_STATUS:
1697 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1698 err = loop_set_status_old(lo, argp);
1701 case LOOP_GET_STATUS:
1702 return loop_get_status_old(lo, argp);
1703 case LOOP_SET_STATUS64:
1705 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1706 err = loop_set_status64(lo, argp);
1709 case LOOP_GET_STATUS64:
1710 return loop_get_status64(lo, argp);
1711 case LOOP_SET_CAPACITY:
1712 case LOOP_SET_DIRECT_IO:
1713 case LOOP_SET_BLOCK_SIZE:
1714 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1718 err = lo_simple_ioctl(lo, cmd, arg);
1725 #ifdef CONFIG_COMPAT
1726 struct compat_loop_info {
1727 compat_int_t lo_number; /* ioctl r/o */
1728 compat_dev_t lo_device; /* ioctl r/o */
1729 compat_ulong_t lo_inode; /* ioctl r/o */
1730 compat_dev_t lo_rdevice; /* ioctl r/o */
1731 compat_int_t lo_offset;
1732 compat_int_t lo_encrypt_type;
1733 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1734 compat_int_t lo_flags; /* ioctl r/o */
1735 char lo_name[LO_NAME_SIZE];
1736 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1737 compat_ulong_t lo_init[2];
1742 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1743 * - noinlined to reduce stack space usage in main part of driver
1746 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1747 struct loop_info64 *info64)
1749 struct compat_loop_info info;
1751 if (copy_from_user(&info, arg, sizeof(info)))
1754 memset(info64, 0, sizeof(*info64));
1755 info64->lo_number = info.lo_number;
1756 info64->lo_device = info.lo_device;
1757 info64->lo_inode = info.lo_inode;
1758 info64->lo_rdevice = info.lo_rdevice;
1759 info64->lo_offset = info.lo_offset;
1760 info64->lo_sizelimit = 0;
1761 info64->lo_encrypt_type = info.lo_encrypt_type;
1762 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1763 info64->lo_flags = info.lo_flags;
1764 info64->lo_init[0] = info.lo_init[0];
1765 info64->lo_init[1] = info.lo_init[1];
1766 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1767 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1769 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1770 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1775 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1776 * - noinlined to reduce stack space usage in main part of driver
1779 loop_info64_to_compat(const struct loop_info64 *info64,
1780 struct compat_loop_info __user *arg)
1782 struct compat_loop_info info;
1784 memset(&info, 0, sizeof(info));
1785 info.lo_number = info64->lo_number;
1786 info.lo_device = info64->lo_device;
1787 info.lo_inode = info64->lo_inode;
1788 info.lo_rdevice = info64->lo_rdevice;
1789 info.lo_offset = info64->lo_offset;
1790 info.lo_encrypt_type = info64->lo_encrypt_type;
1791 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1792 info.lo_flags = info64->lo_flags;
1793 info.lo_init[0] = info64->lo_init[0];
1794 info.lo_init[1] = info64->lo_init[1];
1795 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1796 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1798 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1799 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1801 /* error in case values were truncated */
1802 if (info.lo_device != info64->lo_device ||
1803 info.lo_rdevice != info64->lo_rdevice ||
1804 info.lo_inode != info64->lo_inode ||
1805 info.lo_offset != info64->lo_offset ||
1806 info.lo_init[0] != info64->lo_init[0] ||
1807 info.lo_init[1] != info64->lo_init[1])
1810 if (copy_to_user(arg, &info, sizeof(info)))
1816 loop_set_status_compat(struct loop_device *lo,
1817 const struct compat_loop_info __user *arg)
1819 struct loop_info64 info64;
1822 ret = loop_info64_from_compat(arg, &info64);
1825 return loop_set_status(lo, &info64);
1829 loop_get_status_compat(struct loop_device *lo,
1830 struct compat_loop_info __user *arg)
1832 struct loop_info64 info64;
1837 err = loop_get_status(lo, &info64);
1839 err = loop_info64_to_compat(&info64, arg);
1843 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1844 unsigned int cmd, unsigned long arg)
1846 struct loop_device *lo = bdev->bd_disk->private_data;
1850 case LOOP_SET_STATUS:
1851 err = loop_set_status_compat(lo,
1852 (const struct compat_loop_info __user *)arg);
1854 case LOOP_GET_STATUS:
1855 err = loop_get_status_compat(lo,
1856 (struct compat_loop_info __user *)arg);
1858 case LOOP_SET_CAPACITY:
1860 case LOOP_GET_STATUS64:
1861 case LOOP_SET_STATUS64:
1862 case LOOP_CONFIGURE:
1863 arg = (unsigned long) compat_ptr(arg);
1866 case LOOP_CHANGE_FD:
1867 case LOOP_SET_BLOCK_SIZE:
1868 case LOOP_SET_DIRECT_IO:
1869 err = lo_ioctl(bdev, mode, cmd, arg);
1879 static int lo_open(struct block_device *bdev, fmode_t mode)
1881 struct loop_device *lo;
1885 * take loop_ctl_mutex to protect lo pointer from race with
1886 * loop_control_ioctl(LOOP_CTL_REMOVE), however, to reduce contention
1887 * release it prior to updating lo->lo_refcnt.
1889 err = mutex_lock_killable(&loop_ctl_mutex);
1892 lo = bdev->bd_disk->private_data;
1894 mutex_unlock(&loop_ctl_mutex);
1897 err = mutex_lock_killable(&lo->lo_mutex);
1898 mutex_unlock(&loop_ctl_mutex);
1901 atomic_inc(&lo->lo_refcnt);
1902 mutex_unlock(&lo->lo_mutex);
1906 static void lo_release(struct gendisk *disk, fmode_t mode)
1908 struct loop_device *lo = disk->private_data;
1910 mutex_lock(&lo->lo_mutex);
1911 if (atomic_dec_return(&lo->lo_refcnt))
1914 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1915 if (lo->lo_state != Lo_bound)
1917 lo->lo_state = Lo_rundown;
1918 mutex_unlock(&lo->lo_mutex);
1920 * In autoclear mode, stop the loop thread
1921 * and remove configuration after last close.
1923 __loop_clr_fd(lo, true);
1925 } else if (lo->lo_state == Lo_bound) {
1927 * Otherwise keep thread (if running) and config,
1928 * but flush possible ongoing bios in thread.
1930 blk_mq_freeze_queue(lo->lo_queue);
1931 blk_mq_unfreeze_queue(lo->lo_queue);
1935 mutex_unlock(&lo->lo_mutex);
1938 static const struct block_device_operations lo_fops = {
1939 .owner = THIS_MODULE,
1941 .release = lo_release,
1943 #ifdef CONFIG_COMPAT
1944 .compat_ioctl = lo_compat_ioctl,
1949 * And now the modules code and kernel interface.
1951 static int max_loop;
1952 module_param(max_loop, int, 0444);
1953 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1954 module_param(max_part, int, 0444);
1955 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1956 MODULE_LICENSE("GPL");
1957 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1959 int loop_register_transfer(struct loop_func_table *funcs)
1961 unsigned int n = funcs->number;
1963 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1965 xfer_funcs[n] = funcs;
1969 static int unregister_transfer_cb(int id, void *ptr, void *data)
1971 struct loop_device *lo = ptr;
1972 struct loop_func_table *xfer = data;
1974 mutex_lock(&lo->lo_mutex);
1975 if (lo->lo_encryption == xfer)
1976 loop_release_xfer(lo);
1977 mutex_unlock(&lo->lo_mutex);
1981 int loop_unregister_transfer(int number)
1983 unsigned int n = number;
1984 struct loop_func_table *xfer;
1986 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1989 xfer_funcs[n] = NULL;
1990 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1994 EXPORT_SYMBOL(loop_register_transfer);
1995 EXPORT_SYMBOL(loop_unregister_transfer);
1997 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1998 const struct blk_mq_queue_data *bd)
2000 struct request *rq = bd->rq;
2001 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2002 struct loop_device *lo = rq->q->queuedata;
2004 blk_mq_start_request(rq);
2006 if (lo->lo_state != Lo_bound)
2007 return BLK_STS_IOERR;
2009 switch (req_op(rq)) {
2011 case REQ_OP_DISCARD:
2012 case REQ_OP_WRITE_ZEROES:
2013 cmd->use_aio = false;
2016 cmd->use_aio = lo->use_dio;
2020 /* always use the first bio's css */
2021 #ifdef CONFIG_BLK_CGROUP
2022 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2023 cmd->css = &bio_blkcg(rq->bio)->css;
2028 kthread_queue_work(&lo->worker, &cmd->work);
2033 static void loop_handle_cmd(struct loop_cmd *cmd)
2035 struct request *rq = blk_mq_rq_from_pdu(cmd);
2036 const bool write = op_is_write(req_op(rq));
2037 struct loop_device *lo = rq->q->queuedata;
2040 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2045 ret = do_req_filebacked(lo, rq);
2047 /* complete non-aio request */
2048 if (!cmd->use_aio || ret) {
2049 if (ret == -EOPNOTSUPP)
2052 cmd->ret = ret ? -EIO : 0;
2053 if (likely(!blk_should_fake_timeout(rq->q)))
2054 blk_mq_complete_request(rq);
2058 static void loop_queue_work(struct kthread_work *work)
2060 struct loop_cmd *cmd =
2061 container_of(work, struct loop_cmd, work);
2063 loop_handle_cmd(cmd);
2066 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2067 unsigned int hctx_idx, unsigned int numa_node)
2069 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2071 kthread_init_work(&cmd->work, loop_queue_work);
2075 static const struct blk_mq_ops loop_mq_ops = {
2076 .queue_rq = loop_queue_rq,
2077 .init_request = loop_init_request,
2078 .complete = lo_complete_rq,
2081 static int loop_add(struct loop_device **l, int i)
2083 struct loop_device *lo;
2084 struct gendisk *disk;
2088 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2092 lo->lo_state = Lo_unbound;
2094 /* allocate id, if @id >= 0, we're requesting that specific id */
2096 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2100 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2107 lo->tag_set.ops = &loop_mq_ops;
2108 lo->tag_set.nr_hw_queues = 1;
2109 lo->tag_set.queue_depth = 128;
2110 lo->tag_set.numa_node = NUMA_NO_NODE;
2111 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2112 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
2113 lo->tag_set.driver_data = lo;
2115 err = blk_mq_alloc_tag_set(&lo->tag_set);
2119 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2120 if (IS_ERR(lo->lo_queue)) {
2121 err = PTR_ERR(lo->lo_queue);
2122 goto out_cleanup_tags;
2124 lo->lo_queue->queuedata = lo;
2126 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2129 * By default, we do buffer IO, so it doesn't make sense to enable
2130 * merge because the I/O submitted to backing file is handled page by
2131 * page. For directio mode, merge does help to dispatch bigger request
2132 * to underlayer disk. We will enable merge once directio is enabled.
2134 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2137 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2139 goto out_free_queue;
2142 * Disable partition scanning by default. The in-kernel partition
2143 * scanning can be requested individually per-device during its
2144 * setup. Userspace can always add and remove partitions from all
2145 * devices. The needed partition minors are allocated from the
2146 * extended minor space, the main loop device numbers will continue
2147 * to match the loop minors, regardless of the number of partitions
2150 * If max_part is given, partition scanning is globally enabled for
2151 * all loop devices. The minors for the main loop devices will be
2152 * multiples of max_part.
2154 * Note: Global-for-all-devices, set-only-at-init, read-only module
2155 * parameteters like 'max_loop' and 'max_part' make things needlessly
2156 * complicated, are too static, inflexible and may surprise
2157 * userspace tools. Parameters like this in general should be avoided.
2160 disk->flags |= GENHD_FL_NO_PART_SCAN;
2161 disk->flags |= GENHD_FL_EXT_DEVT;
2162 atomic_set(&lo->lo_refcnt, 0);
2163 mutex_init(&lo->lo_mutex);
2165 spin_lock_init(&lo->lo_lock);
2166 disk->major = LOOP_MAJOR;
2167 disk->first_minor = i << part_shift;
2168 disk->fops = &lo_fops;
2169 disk->private_data = lo;
2170 disk->queue = lo->lo_queue;
2171 sprintf(disk->disk_name, "loop%d", i);
2174 return lo->lo_number;
2177 blk_cleanup_queue(lo->lo_queue);
2179 blk_mq_free_tag_set(&lo->tag_set);
2181 idr_remove(&loop_index_idr, i);
2188 static void loop_remove(struct loop_device *lo)
2190 del_gendisk(lo->lo_disk);
2191 blk_cleanup_queue(lo->lo_queue);
2192 blk_mq_free_tag_set(&lo->tag_set);
2193 put_disk(lo->lo_disk);
2194 mutex_destroy(&lo->lo_mutex);
2198 static int find_free_cb(int id, void *ptr, void *data)
2200 struct loop_device *lo = ptr;
2201 struct loop_device **l = data;
2203 if (lo->lo_state == Lo_unbound) {
2210 static int loop_lookup(struct loop_device **l, int i)
2212 struct loop_device *lo;
2218 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2221 ret = lo->lo_number;
2226 /* lookup and return a specific i */
2227 lo = idr_find(&loop_index_idr, i);
2230 ret = lo->lo_number;
2236 static void loop_probe(dev_t dev)
2238 int idx = MINOR(dev) >> part_shift;
2239 struct loop_device *lo;
2241 if (max_loop && idx >= max_loop)
2244 mutex_lock(&loop_ctl_mutex);
2245 if (loop_lookup(&lo, idx) < 0)
2247 mutex_unlock(&loop_ctl_mutex);
2250 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2253 struct loop_device *lo;
2256 ret = mutex_lock_killable(&loop_ctl_mutex);
2263 ret = loop_lookup(&lo, parm);
2268 ret = loop_add(&lo, parm);
2270 case LOOP_CTL_REMOVE:
2271 ret = loop_lookup(&lo, parm);
2274 ret = mutex_lock_killable(&lo->lo_mutex);
2277 if (lo->lo_state != Lo_unbound) {
2279 mutex_unlock(&lo->lo_mutex);
2282 if (atomic_read(&lo->lo_refcnt) > 0) {
2284 mutex_unlock(&lo->lo_mutex);
2287 lo->lo_disk->private_data = NULL;
2288 mutex_unlock(&lo->lo_mutex);
2289 idr_remove(&loop_index_idr, lo->lo_number);
2292 case LOOP_CTL_GET_FREE:
2293 ret = loop_lookup(&lo, -1);
2296 ret = loop_add(&lo, -1);
2298 mutex_unlock(&loop_ctl_mutex);
2303 static const struct file_operations loop_ctl_fops = {
2304 .open = nonseekable_open,
2305 .unlocked_ioctl = loop_control_ioctl,
2306 .compat_ioctl = loop_control_ioctl,
2307 .owner = THIS_MODULE,
2308 .llseek = noop_llseek,
2311 static struct miscdevice loop_misc = {
2312 .minor = LOOP_CTRL_MINOR,
2313 .name = "loop-control",
2314 .fops = &loop_ctl_fops,
2317 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2318 MODULE_ALIAS("devname:loop-control");
2320 static int __init loop_init(void)
2323 struct loop_device *lo;
2328 part_shift = fls(max_part);
2331 * Adjust max_part according to part_shift as it is exported
2332 * to user space so that user can decide correct minor number
2333 * if [s]he want to create more devices.
2335 * Note that -1 is required because partition 0 is reserved
2336 * for the whole disk.
2338 max_part = (1UL << part_shift) - 1;
2341 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2346 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2352 * If max_loop is specified, create that many devices upfront.
2353 * This also becomes a hard limit. If max_loop is not specified,
2354 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2355 * init time. Loop devices can be requested on-demand with the
2356 * /dev/loop-control interface, or be instantiated by accessing
2357 * a 'dead' device node.
2362 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2364 err = misc_register(&loop_misc);
2369 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2374 /* pre-create number of devices given by config or max_loop */
2375 mutex_lock(&loop_ctl_mutex);
2376 for (i = 0; i < nr; i++)
2378 mutex_unlock(&loop_ctl_mutex);
2380 printk(KERN_INFO "loop: module loaded\n");
2384 misc_deregister(&loop_misc);
2389 static int loop_exit_cb(int id, void *ptr, void *data)
2391 struct loop_device *lo = ptr;
2397 static void __exit loop_exit(void)
2399 mutex_lock(&loop_ctl_mutex);
2401 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2402 idr_destroy(&loop_index_idr);
2404 unregister_blkdev(LOOP_MAJOR, "loop");
2406 misc_deregister(&loop_misc);
2408 mutex_unlock(&loop_ctl_mutex);
2411 module_init(loop_init);
2412 module_exit(loop_exit);
2415 static int __init max_loop_setup(char *str)
2417 max_loop = simple_strtol(str, NULL, 0);
2421 __setup("max_loop=", max_loop_setup);