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 blk_mq_freeze_queue(lo->lo_queue);
218 lo->use_dio = use_dio;
220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
221 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
224 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
226 blk_mq_unfreeze_queue(lo->lo_queue);
230 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
232 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
233 sector_t x = (sector_t)size;
234 struct block_device *bdev = lo->lo_device;
236 if (unlikely((loff_t)x != size))
238 if (lo->lo_offset != offset)
239 lo->lo_offset = offset;
240 if (lo->lo_sizelimit != sizelimit)
241 lo->lo_sizelimit = sizelimit;
242 set_capacity(lo->lo_disk, x);
243 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
244 /* let user-space know about the new size */
245 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
250 lo_do_transfer(struct loop_device *lo, int cmd,
251 struct page *rpage, unsigned roffs,
252 struct page *lpage, unsigned loffs,
253 int size, sector_t rblock)
257 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
261 printk_ratelimited(KERN_ERR
262 "loop: Transfer error at byte offset %llu, length %i.\n",
263 (unsigned long long)rblock << 9, size);
267 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
272 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
274 file_start_write(file);
275 bw = vfs_iter_write(file, &i, ppos, 0);
276 file_end_write(file);
278 if (likely(bw == bvec->bv_len))
281 printk_ratelimited(KERN_ERR
282 "loop: Write error at byte offset %llu, length %i.\n",
283 (unsigned long long)*ppos, bvec->bv_len);
289 static int lo_write_simple(struct loop_device *lo, struct request *rq,
293 struct req_iterator iter;
296 rq_for_each_segment(bvec, rq, iter) {
297 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
307 * This is the slow, transforming version that needs to double buffer the
308 * data as it cannot do the transformations in place without having direct
309 * access to the destination pages of the backing file.
311 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
314 struct bio_vec bvec, b;
315 struct req_iterator iter;
319 page = alloc_page(GFP_NOIO);
323 rq_for_each_segment(bvec, rq, iter) {
324 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
325 bvec.bv_offset, bvec.bv_len, pos >> 9);
331 b.bv_len = bvec.bv_len;
332 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
341 static int lo_read_simple(struct loop_device *lo, struct request *rq,
345 struct req_iterator iter;
349 rq_for_each_segment(bvec, rq, iter) {
350 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
351 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
355 flush_dcache_page(bvec.bv_page);
357 if (len != bvec.bv_len) {
360 __rq_for_each_bio(bio, rq)
370 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
373 struct bio_vec bvec, b;
374 struct req_iterator iter;
380 page = alloc_page(GFP_NOIO);
384 rq_for_each_segment(bvec, rq, iter) {
389 b.bv_len = bvec.bv_len;
391 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
398 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
399 bvec.bv_offset, len, offset >> 9);
403 flush_dcache_page(bvec.bv_page);
405 if (len != bvec.bv_len) {
408 __rq_for_each_bio(bio, rq)
420 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
424 * We use fallocate to manipulate the space mappings used by the image
425 * a.k.a. discard/zerorange. However we do not support this if
426 * encryption is enabled, because it may give an attacker useful
429 struct file *file = lo->lo_backing_file;
432 mode |= FALLOC_FL_KEEP_SIZE;
434 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
439 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
440 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
446 static int lo_req_flush(struct loop_device *lo, struct request *rq)
448 struct file *file = lo->lo_backing_file;
449 int ret = vfs_fsync(file, 0);
450 if (unlikely(ret && ret != -EINVAL))
456 static void lo_complete_rq(struct request *rq)
458 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
459 blk_status_t ret = BLK_STS_OK;
461 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
462 req_op(rq) != REQ_OP_READ) {
469 * Short READ - if we got some data, advance our request and
470 * retry it. If we got no data, end the rest with EIO.
473 blk_update_request(rq, BLK_STS_OK, cmd->ret);
475 blk_mq_requeue_request(rq, true);
478 struct bio *bio = rq->bio;
487 blk_mq_end_request(rq, ret);
491 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
493 struct request *rq = blk_mq_rq_from_pdu(cmd);
495 if (!atomic_dec_and_test(&cmd->ref))
499 blk_mq_complete_request(rq);
502 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
504 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
509 lo_rw_aio_do_completion(cmd);
512 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
515 struct iov_iter iter;
516 struct req_iterator rq_iter;
517 struct bio_vec *bvec;
518 struct request *rq = blk_mq_rq_from_pdu(cmd);
519 struct bio *bio = rq->bio;
520 struct file *file = lo->lo_backing_file;
526 rq_for_each_bvec(tmp, rq, rq_iter)
529 if (rq->bio != rq->biotail) {
531 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
538 * The bios of the request may be started from the middle of
539 * the 'bvec' because of bio splitting, so we can't directly
540 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
541 * API will take care of all details for us.
543 rq_for_each_bvec(tmp, rq, rq_iter) {
551 * Same here, this bio may be started from the middle of the
552 * 'bvec' because of bio splitting, so offset from the bvec
553 * must be passed to iov iterator
555 offset = bio->bi_iter.bi_bvec_done;
556 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
558 atomic_set(&cmd->ref, 2);
560 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
561 iter.iov_offset = offset;
563 cmd->iocb.ki_pos = pos;
564 cmd->iocb.ki_filp = file;
565 cmd->iocb.ki_complete = lo_rw_aio_complete;
566 cmd->iocb.ki_flags = IOCB_DIRECT;
567 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
569 kthread_associate_blkcg(cmd->css);
572 ret = call_write_iter(file, &cmd->iocb, &iter);
574 ret = call_read_iter(file, &cmd->iocb, &iter);
576 lo_rw_aio_do_completion(cmd);
577 kthread_associate_blkcg(NULL);
579 if (ret != -EIOCBQUEUED)
580 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
584 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
586 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
587 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
590 * lo_write_simple and lo_read_simple should have been covered
591 * by io submit style function like lo_rw_aio(), one blocker
592 * is that lo_read_simple() need to call flush_dcache_page after
593 * the page is written from kernel, and it isn't easy to handle
594 * this in io submit style function which submits all segments
595 * of the req at one time. And direct read IO doesn't need to
596 * run flush_dcache_page().
598 switch (req_op(rq)) {
600 return lo_req_flush(lo, rq);
601 case REQ_OP_WRITE_ZEROES:
603 * If the caller doesn't want deallocation, call zeroout to
604 * write zeroes the range. Otherwise, punch them out.
606 return lo_fallocate(lo, rq, pos,
607 (rq->cmd_flags & REQ_NOUNMAP) ?
608 FALLOC_FL_ZERO_RANGE :
609 FALLOC_FL_PUNCH_HOLE);
611 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
614 return lo_write_transfer(lo, rq, pos);
615 else if (cmd->use_aio)
616 return lo_rw_aio(lo, cmd, pos, WRITE);
618 return lo_write_simple(lo, rq, pos);
621 return lo_read_transfer(lo, rq, pos);
622 else if (cmd->use_aio)
623 return lo_rw_aio(lo, cmd, pos, READ);
625 return lo_read_simple(lo, rq, pos);
632 static inline void loop_update_dio(struct loop_device *lo)
634 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
638 static void loop_reread_partitions(struct loop_device *lo,
639 struct block_device *bdev)
643 mutex_lock(&bdev->bd_mutex);
644 rc = bdev_disk_changed(bdev, false);
645 mutex_unlock(&bdev->bd_mutex);
647 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
648 __func__, lo->lo_number, lo->lo_file_name, rc);
651 static inline int is_loop_device(struct file *file)
653 struct inode *i = file->f_mapping->host;
655 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
658 static int loop_validate_file(struct file *file, struct block_device *bdev)
660 struct inode *inode = file->f_mapping->host;
661 struct file *f = file;
663 /* Avoid recursion */
664 while (is_loop_device(f)) {
665 struct loop_device *l;
667 if (f->f_mapping->host->i_bdev == bdev)
670 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
671 if (l->lo_state != Lo_bound) {
674 f = l->lo_backing_file;
676 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
682 * loop_change_fd switched the backing store of a loopback device to
683 * a new file. This is useful for operating system installers to free up
684 * the original file and in High Availability environments to switch to
685 * an alternative location for the content in case of server meltdown.
686 * This can only work if the loop device is used read-only, and if the
687 * new backing store is the same size and type as the old backing store.
689 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
692 struct file *file = NULL, *old_file;
696 error = mutex_lock_killable(&loop_ctl_mutex);
700 if (lo->lo_state != Lo_bound)
703 /* the loop device has to be read-only */
705 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
713 error = loop_validate_file(file, bdev);
717 old_file = lo->lo_backing_file;
721 /* size of the new backing store needs to be the same */
722 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
726 blk_mq_freeze_queue(lo->lo_queue);
727 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
728 lo->lo_backing_file = file;
729 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
730 mapping_set_gfp_mask(file->f_mapping,
731 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
733 blk_mq_unfreeze_queue(lo->lo_queue);
734 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
735 mutex_unlock(&loop_ctl_mutex);
737 * We must drop file reference outside of loop_ctl_mutex as dropping
738 * the file ref can take bd_mutex which creates circular locking
743 loop_reread_partitions(lo, bdev);
747 mutex_unlock(&loop_ctl_mutex);
753 /* loop sysfs attributes */
755 static ssize_t loop_attr_show(struct device *dev, char *page,
756 ssize_t (*callback)(struct loop_device *, char *))
758 struct gendisk *disk = dev_to_disk(dev);
759 struct loop_device *lo = disk->private_data;
761 return callback(lo, page);
764 #define LOOP_ATTR_RO(_name) \
765 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
766 static ssize_t loop_attr_do_show_##_name(struct device *d, \
767 struct device_attribute *attr, char *b) \
769 return loop_attr_show(d, b, loop_attr_##_name##_show); \
771 static struct device_attribute loop_attr_##_name = \
772 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
774 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
779 spin_lock_irq(&lo->lo_lock);
780 if (lo->lo_backing_file)
781 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
782 spin_unlock_irq(&lo->lo_lock);
784 if (IS_ERR_OR_NULL(p))
788 memmove(buf, p, ret);
796 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
798 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
801 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
803 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
806 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
808 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
810 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
813 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
815 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
817 return sprintf(buf, "%s\n", partscan ? "1" : "0");
820 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
822 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
824 return sprintf(buf, "%s\n", dio ? "1" : "0");
827 LOOP_ATTR_RO(backing_file);
828 LOOP_ATTR_RO(offset);
829 LOOP_ATTR_RO(sizelimit);
830 LOOP_ATTR_RO(autoclear);
831 LOOP_ATTR_RO(partscan);
834 static struct attribute *loop_attrs[] = {
835 &loop_attr_backing_file.attr,
836 &loop_attr_offset.attr,
837 &loop_attr_sizelimit.attr,
838 &loop_attr_autoclear.attr,
839 &loop_attr_partscan.attr,
844 static struct attribute_group loop_attribute_group = {
849 static void loop_sysfs_init(struct loop_device *lo)
851 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
852 &loop_attribute_group);
855 static void loop_sysfs_exit(struct loop_device *lo)
857 if (lo->sysfs_inited)
858 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
859 &loop_attribute_group);
862 static void loop_config_discard(struct loop_device *lo)
864 struct file *file = lo->lo_backing_file;
865 struct inode *inode = file->f_mapping->host;
866 struct request_queue *q = lo->lo_queue;
869 * We use punch hole to reclaim the free space used by the
870 * image a.k.a. discard. However we do not support discard if
871 * encryption is enabled, because it may give an attacker
872 * useful information.
874 if ((!file->f_op->fallocate) ||
875 lo->lo_encrypt_key_size) {
876 q->limits.discard_granularity = 0;
877 q->limits.discard_alignment = 0;
878 blk_queue_max_discard_sectors(q, 0);
879 blk_queue_max_write_zeroes_sectors(q, 0);
880 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
884 q->limits.discard_granularity = inode->i_sb->s_blocksize;
885 q->limits.discard_alignment = 0;
887 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
888 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
889 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
892 static void loop_unprepare_queue(struct loop_device *lo)
894 kthread_flush_worker(&lo->worker);
895 kthread_stop(lo->worker_task);
898 static int loop_kthread_worker_fn(void *worker_ptr)
900 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
901 return kthread_worker_fn(worker_ptr);
904 static int loop_prepare_queue(struct loop_device *lo)
906 kthread_init_worker(&lo->worker);
907 lo->worker_task = kthread_run(loop_kthread_worker_fn,
908 &lo->worker, "loop%d", lo->lo_number);
909 if (IS_ERR(lo->worker_task))
911 set_user_nice(lo->worker_task, MIN_NICE);
915 static void loop_update_rotational(struct loop_device *lo)
917 struct file *file = lo->lo_backing_file;
918 struct inode *file_inode = file->f_mapping->host;
919 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
920 struct request_queue *q = lo->lo_queue;
923 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
925 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
928 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
930 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
933 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
934 struct block_device *bdev, unsigned int arg)
938 struct address_space *mapping;
939 struct block_device *claimed_bdev = NULL;
945 /* This is safe, since we have a reference from open(). */
946 __module_get(THIS_MODULE);
954 * If we don't hold exclusive handle for the device, upgrade to it
955 * here to avoid changing device under exclusive owner.
957 if (!(mode & FMODE_EXCL)) {
958 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
959 if (IS_ERR(claimed_bdev)) {
960 error = PTR_ERR(claimed_bdev);
965 error = mutex_lock_killable(&loop_ctl_mutex);
970 if (lo->lo_state != Lo_unbound)
973 error = loop_validate_file(file, bdev);
977 mapping = file->f_mapping;
978 inode = mapping->host;
980 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
981 !file->f_op->write_iter)
982 lo_flags |= LO_FLAGS_READ_ONLY;
985 size = get_loop_size(lo, file);
986 if ((loff_t)(sector_t)size != size)
988 error = loop_prepare_queue(lo);
994 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
997 lo->lo_device = bdev;
998 lo->lo_flags = lo_flags;
999 lo->lo_backing_file = file;
1000 lo->transfer = NULL;
1002 lo->lo_sizelimit = 0;
1003 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1004 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1006 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1007 blk_queue_write_cache(lo->lo_queue, true, false);
1009 if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) {
1010 /* In case of direct I/O, match underlying block size */
1011 unsigned short bsize = bdev_logical_block_size(
1012 inode->i_sb->s_bdev);
1014 blk_queue_logical_block_size(lo->lo_queue, bsize);
1015 blk_queue_physical_block_size(lo->lo_queue, bsize);
1016 blk_queue_io_min(lo->lo_queue, bsize);
1019 loop_update_rotational(lo);
1020 loop_update_dio(lo);
1021 set_capacity(lo->lo_disk, size);
1022 bd_set_size(bdev, size << 9);
1023 loop_sysfs_init(lo);
1024 /* let user-space know about the new size */
1025 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1027 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1028 block_size(inode->i_bdev) : PAGE_SIZE);
1030 lo->lo_state = Lo_bound;
1032 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1033 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1035 /* Grab the block_device to prevent its destruction after we
1036 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1039 mutex_unlock(&loop_ctl_mutex);
1041 loop_reread_partitions(lo, bdev);
1043 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1047 mutex_unlock(&loop_ctl_mutex);
1050 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1054 /* This is safe: open() is still holding a reference. */
1055 module_put(THIS_MODULE);
1060 loop_release_xfer(struct loop_device *lo)
1063 struct loop_func_table *xfer = lo->lo_encryption;
1067 err = xfer->release(lo);
1068 lo->transfer = NULL;
1069 lo->lo_encryption = NULL;
1070 module_put(xfer->owner);
1076 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1077 const struct loop_info64 *i)
1082 struct module *owner = xfer->owner;
1084 if (!try_module_get(owner))
1087 err = xfer->init(lo, i);
1091 lo->lo_encryption = xfer;
1096 static int __loop_clr_fd(struct loop_device *lo, bool release)
1098 struct file *filp = NULL;
1099 gfp_t gfp = lo->old_gfp_mask;
1100 struct block_device *bdev = lo->lo_device;
1102 bool partscan = false;
1105 mutex_lock(&loop_ctl_mutex);
1106 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1111 filp = lo->lo_backing_file;
1117 /* freeze request queue during the transition */
1118 blk_mq_freeze_queue(lo->lo_queue);
1120 spin_lock_irq(&lo->lo_lock);
1121 lo->lo_backing_file = NULL;
1122 spin_unlock_irq(&lo->lo_lock);
1124 loop_release_xfer(lo);
1125 lo->transfer = NULL;
1127 lo->lo_device = NULL;
1128 lo->lo_encryption = NULL;
1130 lo->lo_sizelimit = 0;
1131 lo->lo_encrypt_key_size = 0;
1132 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1133 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1134 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1135 blk_queue_logical_block_size(lo->lo_queue, 512);
1136 blk_queue_physical_block_size(lo->lo_queue, 512);
1137 blk_queue_io_min(lo->lo_queue, 512);
1140 invalidate_bdev(bdev);
1141 bdev->bd_inode->i_mapping->wb_err = 0;
1143 set_capacity(lo->lo_disk, 0);
1144 loop_sysfs_exit(lo);
1146 bd_set_size(bdev, 0);
1147 /* let user-space know about this change */
1148 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1150 mapping_set_gfp_mask(filp->f_mapping, gfp);
1151 /* This is safe: open() is still holding a reference. */
1152 module_put(THIS_MODULE);
1153 blk_mq_unfreeze_queue(lo->lo_queue);
1155 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1156 lo_number = lo->lo_number;
1157 loop_unprepare_queue(lo);
1159 mutex_unlock(&loop_ctl_mutex);
1162 * bd_mutex has been held already in release path, so don't
1163 * acquire it if this function is called in such case.
1165 * If the reread partition isn't from release path, lo_refcnt
1166 * must be at least one and it can only become zero when the
1167 * current holder is released.
1170 mutex_lock(&bdev->bd_mutex);
1171 err = bdev_disk_changed(bdev, false);
1173 mutex_unlock(&bdev->bd_mutex);
1175 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1176 __func__, lo_number, err);
1177 /* Device is gone, no point in returning error */
1182 * lo->lo_state is set to Lo_unbound here after above partscan has
1185 * There cannot be anybody else entering __loop_clr_fd() as
1186 * lo->lo_backing_file is already cleared and Lo_rundown state
1187 * protects us from all the other places trying to change the 'lo'
1190 mutex_lock(&loop_ctl_mutex);
1193 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1194 lo->lo_state = Lo_unbound;
1195 mutex_unlock(&loop_ctl_mutex);
1198 * Need not hold loop_ctl_mutex to fput backing file.
1199 * Calling fput holding loop_ctl_mutex triggers a circular
1200 * lock dependency possibility warning as fput can take
1201 * bd_mutex which is usually taken before loop_ctl_mutex.
1208 static int loop_clr_fd(struct loop_device *lo)
1212 err = mutex_lock_killable(&loop_ctl_mutex);
1215 if (lo->lo_state != Lo_bound) {
1216 mutex_unlock(&loop_ctl_mutex);
1220 * If we've explicitly asked to tear down the loop device,
1221 * and it has an elevated reference count, set it for auto-teardown when
1222 * the last reference goes away. This stops $!~#$@ udev from
1223 * preventing teardown because it decided that it needs to run blkid on
1224 * the loopback device whenever they appear. xfstests is notorious for
1225 * failing tests because blkid via udev races with a losetup
1226 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1227 * command to fail with EBUSY.
1229 if (atomic_read(&lo->lo_refcnt) > 1) {
1230 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1231 mutex_unlock(&loop_ctl_mutex);
1234 lo->lo_state = Lo_rundown;
1235 mutex_unlock(&loop_ctl_mutex);
1237 return __loop_clr_fd(lo, false);
1241 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1244 struct loop_func_table *xfer;
1245 kuid_t uid = current_uid();
1246 struct block_device *bdev;
1247 bool partscan = false;
1249 err = mutex_lock_killable(&loop_ctl_mutex);
1252 if (lo->lo_encrypt_key_size &&
1253 !uid_eq(lo->lo_key_owner, uid) &&
1254 !capable(CAP_SYS_ADMIN)) {
1258 if (lo->lo_state != Lo_bound) {
1262 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1267 if (lo->lo_offset != info->lo_offset ||
1268 lo->lo_sizelimit != info->lo_sizelimit) {
1269 sync_blockdev(lo->lo_device);
1270 kill_bdev(lo->lo_device);
1273 /* I/O need to be drained during transfer transition */
1274 blk_mq_freeze_queue(lo->lo_queue);
1276 err = loop_release_xfer(lo);
1280 if (info->lo_encrypt_type) {
1281 unsigned int type = info->lo_encrypt_type;
1283 if (type >= MAX_LO_CRYPT) {
1287 xfer = xfer_funcs[type];
1295 err = loop_init_xfer(lo, xfer, info);
1299 if (lo->lo_offset != info->lo_offset ||
1300 lo->lo_sizelimit != info->lo_sizelimit) {
1301 /* kill_bdev should have truncated all the pages */
1302 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1304 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1305 __func__, lo->lo_number, lo->lo_file_name,
1306 lo->lo_device->bd_inode->i_mapping->nrpages);
1309 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1315 loop_config_discard(lo);
1317 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1318 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1319 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1320 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1324 lo->transfer = xfer->transfer;
1325 lo->ioctl = xfer->ioctl;
1327 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1328 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1329 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1331 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1332 lo->lo_init[0] = info->lo_init[0];
1333 lo->lo_init[1] = info->lo_init[1];
1334 if (info->lo_encrypt_key_size) {
1335 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1336 info->lo_encrypt_key_size);
1337 lo->lo_key_owner = uid;
1340 /* update dio if lo_offset or transfer is changed */
1341 __loop_update_dio(lo, lo->use_dio);
1344 blk_mq_unfreeze_queue(lo->lo_queue);
1346 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1347 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1348 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1349 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1350 bdev = lo->lo_device;
1354 mutex_unlock(&loop_ctl_mutex);
1356 loop_reread_partitions(lo, bdev);
1362 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1368 ret = mutex_lock_killable(&loop_ctl_mutex);
1371 if (lo->lo_state != Lo_bound) {
1372 mutex_unlock(&loop_ctl_mutex);
1376 memset(info, 0, sizeof(*info));
1377 info->lo_number = lo->lo_number;
1378 info->lo_offset = lo->lo_offset;
1379 info->lo_sizelimit = lo->lo_sizelimit;
1380 info->lo_flags = lo->lo_flags;
1381 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1382 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1383 info->lo_encrypt_type =
1384 lo->lo_encryption ? lo->lo_encryption->number : 0;
1385 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1386 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1387 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1388 lo->lo_encrypt_key_size);
1391 /* Drop loop_ctl_mutex while we call into the filesystem. */
1392 path = lo->lo_backing_file->f_path;
1394 mutex_unlock(&loop_ctl_mutex);
1395 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1397 info->lo_device = huge_encode_dev(stat.dev);
1398 info->lo_inode = stat.ino;
1399 info->lo_rdevice = huge_encode_dev(stat.rdev);
1406 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1408 memset(info64, 0, sizeof(*info64));
1409 info64->lo_number = info->lo_number;
1410 info64->lo_device = info->lo_device;
1411 info64->lo_inode = info->lo_inode;
1412 info64->lo_rdevice = info->lo_rdevice;
1413 info64->lo_offset = info->lo_offset;
1414 info64->lo_sizelimit = 0;
1415 info64->lo_encrypt_type = info->lo_encrypt_type;
1416 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1417 info64->lo_flags = info->lo_flags;
1418 info64->lo_init[0] = info->lo_init[0];
1419 info64->lo_init[1] = info->lo_init[1];
1420 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1421 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1423 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1424 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1428 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1430 memset(info, 0, sizeof(*info));
1431 info->lo_number = info64->lo_number;
1432 info->lo_device = info64->lo_device;
1433 info->lo_inode = info64->lo_inode;
1434 info->lo_rdevice = info64->lo_rdevice;
1435 info->lo_offset = info64->lo_offset;
1436 info->lo_encrypt_type = info64->lo_encrypt_type;
1437 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1438 info->lo_flags = info64->lo_flags;
1439 info->lo_init[0] = info64->lo_init[0];
1440 info->lo_init[1] = info64->lo_init[1];
1441 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1442 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1444 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1445 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1447 /* error in case values were truncated */
1448 if (info->lo_device != info64->lo_device ||
1449 info->lo_rdevice != info64->lo_rdevice ||
1450 info->lo_inode != info64->lo_inode ||
1451 info->lo_offset != info64->lo_offset)
1458 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1460 struct loop_info info;
1461 struct loop_info64 info64;
1463 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1465 loop_info64_from_old(&info, &info64);
1466 return loop_set_status(lo, &info64);
1470 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1472 struct loop_info64 info64;
1474 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1476 return loop_set_status(lo, &info64);
1480 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1481 struct loop_info info;
1482 struct loop_info64 info64;
1487 err = loop_get_status(lo, &info64);
1489 err = loop_info64_to_old(&info64, &info);
1490 if (!err && copy_to_user(arg, &info, sizeof(info)))
1497 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1498 struct loop_info64 info64;
1503 err = loop_get_status(lo, &info64);
1504 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1510 static int loop_set_capacity(struct loop_device *lo)
1512 if (unlikely(lo->lo_state != Lo_bound))
1515 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1518 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1521 if (lo->lo_state != Lo_bound)
1524 __loop_update_dio(lo, !!arg);
1525 if (lo->use_dio == !!arg)
1532 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1536 if (lo->lo_state != Lo_bound)
1539 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1542 if (lo->lo_queue->limits.logical_block_size != arg) {
1543 sync_blockdev(lo->lo_device);
1544 kill_bdev(lo->lo_device);
1547 blk_mq_freeze_queue(lo->lo_queue);
1549 /* kill_bdev should have truncated all the pages */
1550 if (lo->lo_queue->limits.logical_block_size != arg &&
1551 lo->lo_device->bd_inode->i_mapping->nrpages) {
1553 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1554 __func__, lo->lo_number, lo->lo_file_name,
1555 lo->lo_device->bd_inode->i_mapping->nrpages);
1559 blk_queue_logical_block_size(lo->lo_queue, arg);
1560 blk_queue_physical_block_size(lo->lo_queue, arg);
1561 blk_queue_io_min(lo->lo_queue, arg);
1562 loop_update_dio(lo);
1564 blk_mq_unfreeze_queue(lo->lo_queue);
1569 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1574 err = mutex_lock_killable(&loop_ctl_mutex);
1578 case LOOP_SET_CAPACITY:
1579 err = loop_set_capacity(lo);
1581 case LOOP_SET_DIRECT_IO:
1582 err = loop_set_dio(lo, arg);
1584 case LOOP_SET_BLOCK_SIZE:
1585 err = loop_set_block_size(lo, arg);
1588 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1590 mutex_unlock(&loop_ctl_mutex);
1594 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1595 unsigned int cmd, unsigned long arg)
1597 struct loop_device *lo = bdev->bd_disk->private_data;
1602 return loop_set_fd(lo, mode, bdev, arg);
1603 case LOOP_CHANGE_FD:
1604 return loop_change_fd(lo, bdev, arg);
1606 return loop_clr_fd(lo);
1607 case LOOP_SET_STATUS:
1609 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1610 err = loop_set_status_old(lo,
1611 (struct loop_info __user *)arg);
1614 case LOOP_GET_STATUS:
1615 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1616 case LOOP_SET_STATUS64:
1618 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1619 err = loop_set_status64(lo,
1620 (struct loop_info64 __user *) arg);
1623 case LOOP_GET_STATUS64:
1624 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1625 case LOOP_SET_CAPACITY:
1626 case LOOP_SET_DIRECT_IO:
1627 case LOOP_SET_BLOCK_SIZE:
1628 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1632 err = lo_simple_ioctl(lo, cmd, arg);
1639 #ifdef CONFIG_COMPAT
1640 struct compat_loop_info {
1641 compat_int_t lo_number; /* ioctl r/o */
1642 compat_dev_t lo_device; /* ioctl r/o */
1643 compat_ulong_t lo_inode; /* ioctl r/o */
1644 compat_dev_t lo_rdevice; /* ioctl r/o */
1645 compat_int_t lo_offset;
1646 compat_int_t lo_encrypt_type;
1647 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1648 compat_int_t lo_flags; /* ioctl r/o */
1649 char lo_name[LO_NAME_SIZE];
1650 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1651 compat_ulong_t lo_init[2];
1656 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1657 * - noinlined to reduce stack space usage in main part of driver
1660 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1661 struct loop_info64 *info64)
1663 struct compat_loop_info info;
1665 if (copy_from_user(&info, arg, sizeof(info)))
1668 memset(info64, 0, sizeof(*info64));
1669 info64->lo_number = info.lo_number;
1670 info64->lo_device = info.lo_device;
1671 info64->lo_inode = info.lo_inode;
1672 info64->lo_rdevice = info.lo_rdevice;
1673 info64->lo_offset = info.lo_offset;
1674 info64->lo_sizelimit = 0;
1675 info64->lo_encrypt_type = info.lo_encrypt_type;
1676 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1677 info64->lo_flags = info.lo_flags;
1678 info64->lo_init[0] = info.lo_init[0];
1679 info64->lo_init[1] = info.lo_init[1];
1680 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1681 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1683 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1684 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1689 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1690 * - noinlined to reduce stack space usage in main part of driver
1693 loop_info64_to_compat(const struct loop_info64 *info64,
1694 struct compat_loop_info __user *arg)
1696 struct compat_loop_info info;
1698 memset(&info, 0, sizeof(info));
1699 info.lo_number = info64->lo_number;
1700 info.lo_device = info64->lo_device;
1701 info.lo_inode = info64->lo_inode;
1702 info.lo_rdevice = info64->lo_rdevice;
1703 info.lo_offset = info64->lo_offset;
1704 info.lo_encrypt_type = info64->lo_encrypt_type;
1705 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1706 info.lo_flags = info64->lo_flags;
1707 info.lo_init[0] = info64->lo_init[0];
1708 info.lo_init[1] = info64->lo_init[1];
1709 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1710 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1712 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1713 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1715 /* error in case values were truncated */
1716 if (info.lo_device != info64->lo_device ||
1717 info.lo_rdevice != info64->lo_rdevice ||
1718 info.lo_inode != info64->lo_inode ||
1719 info.lo_offset != info64->lo_offset ||
1720 info.lo_init[0] != info64->lo_init[0] ||
1721 info.lo_init[1] != info64->lo_init[1])
1724 if (copy_to_user(arg, &info, sizeof(info)))
1730 loop_set_status_compat(struct loop_device *lo,
1731 const struct compat_loop_info __user *arg)
1733 struct loop_info64 info64;
1736 ret = loop_info64_from_compat(arg, &info64);
1739 return loop_set_status(lo, &info64);
1743 loop_get_status_compat(struct loop_device *lo,
1744 struct compat_loop_info __user *arg)
1746 struct loop_info64 info64;
1751 err = loop_get_status(lo, &info64);
1753 err = loop_info64_to_compat(&info64, arg);
1757 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1758 unsigned int cmd, unsigned long arg)
1760 struct loop_device *lo = bdev->bd_disk->private_data;
1764 case LOOP_SET_STATUS:
1765 err = loop_set_status_compat(lo,
1766 (const struct compat_loop_info __user *)arg);
1768 case LOOP_GET_STATUS:
1769 err = loop_get_status_compat(lo,
1770 (struct compat_loop_info __user *)arg);
1772 case LOOP_SET_CAPACITY:
1774 case LOOP_GET_STATUS64:
1775 case LOOP_SET_STATUS64:
1776 arg = (unsigned long) compat_ptr(arg);
1779 case LOOP_CHANGE_FD:
1780 case LOOP_SET_BLOCK_SIZE:
1781 case LOOP_SET_DIRECT_IO:
1782 err = lo_ioctl(bdev, mode, cmd, arg);
1792 static int lo_open(struct block_device *bdev, fmode_t mode)
1794 struct loop_device *lo;
1797 err = mutex_lock_killable(&loop_ctl_mutex);
1800 lo = bdev->bd_disk->private_data;
1806 atomic_inc(&lo->lo_refcnt);
1808 mutex_unlock(&loop_ctl_mutex);
1812 static void lo_release(struct gendisk *disk, fmode_t mode)
1814 struct loop_device *lo;
1816 mutex_lock(&loop_ctl_mutex);
1817 lo = disk->private_data;
1818 if (atomic_dec_return(&lo->lo_refcnt))
1821 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1822 if (lo->lo_state != Lo_bound)
1824 lo->lo_state = Lo_rundown;
1825 mutex_unlock(&loop_ctl_mutex);
1827 * In autoclear mode, stop the loop thread
1828 * and remove configuration after last close.
1830 __loop_clr_fd(lo, true);
1832 } else if (lo->lo_state == Lo_bound) {
1834 * Otherwise keep thread (if running) and config,
1835 * but flush possible ongoing bios in thread.
1837 blk_mq_freeze_queue(lo->lo_queue);
1838 blk_mq_unfreeze_queue(lo->lo_queue);
1842 mutex_unlock(&loop_ctl_mutex);
1845 static const struct block_device_operations lo_fops = {
1846 .owner = THIS_MODULE,
1848 .release = lo_release,
1850 #ifdef CONFIG_COMPAT
1851 .compat_ioctl = lo_compat_ioctl,
1856 * And now the modules code and kernel interface.
1858 static int max_loop;
1859 module_param(max_loop, int, 0444);
1860 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1861 module_param(max_part, int, 0444);
1862 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1863 MODULE_LICENSE("GPL");
1864 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1866 int loop_register_transfer(struct loop_func_table *funcs)
1868 unsigned int n = funcs->number;
1870 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1872 xfer_funcs[n] = funcs;
1876 static int unregister_transfer_cb(int id, void *ptr, void *data)
1878 struct loop_device *lo = ptr;
1879 struct loop_func_table *xfer = data;
1881 mutex_lock(&loop_ctl_mutex);
1882 if (lo->lo_encryption == xfer)
1883 loop_release_xfer(lo);
1884 mutex_unlock(&loop_ctl_mutex);
1888 int loop_unregister_transfer(int number)
1890 unsigned int n = number;
1891 struct loop_func_table *xfer;
1893 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1896 xfer_funcs[n] = NULL;
1897 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1901 EXPORT_SYMBOL(loop_register_transfer);
1902 EXPORT_SYMBOL(loop_unregister_transfer);
1904 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1905 const struct blk_mq_queue_data *bd)
1907 struct request *rq = bd->rq;
1908 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1909 struct loop_device *lo = rq->q->queuedata;
1911 blk_mq_start_request(rq);
1913 if (lo->lo_state != Lo_bound)
1914 return BLK_STS_IOERR;
1916 switch (req_op(rq)) {
1918 case REQ_OP_DISCARD:
1919 case REQ_OP_WRITE_ZEROES:
1920 cmd->use_aio = false;
1923 cmd->use_aio = lo->use_dio;
1927 /* always use the first bio's css */
1928 #ifdef CONFIG_BLK_CGROUP
1929 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1930 cmd->css = &bio_blkcg(rq->bio)->css;
1935 kthread_queue_work(&lo->worker, &cmd->work);
1940 static void loop_handle_cmd(struct loop_cmd *cmd)
1942 struct request *rq = blk_mq_rq_from_pdu(cmd);
1943 const bool write = op_is_write(req_op(rq));
1944 struct loop_device *lo = rq->q->queuedata;
1947 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1952 ret = do_req_filebacked(lo, rq);
1954 /* complete non-aio request */
1955 if (!cmd->use_aio || ret) {
1956 cmd->ret = ret ? -EIO : 0;
1957 blk_mq_complete_request(rq);
1961 static void loop_queue_work(struct kthread_work *work)
1963 struct loop_cmd *cmd =
1964 container_of(work, struct loop_cmd, work);
1966 loop_handle_cmd(cmd);
1969 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1970 unsigned int hctx_idx, unsigned int numa_node)
1972 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1974 kthread_init_work(&cmd->work, loop_queue_work);
1978 static const struct blk_mq_ops loop_mq_ops = {
1979 .queue_rq = loop_queue_rq,
1980 .init_request = loop_init_request,
1981 .complete = lo_complete_rq,
1984 static int loop_add(struct loop_device **l, int i)
1986 struct loop_device *lo;
1987 struct gendisk *disk;
1991 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1995 lo->lo_state = Lo_unbound;
1997 /* allocate id, if @id >= 0, we're requesting that specific id */
1999 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2003 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2010 lo->tag_set.ops = &loop_mq_ops;
2011 lo->tag_set.nr_hw_queues = 1;
2012 lo->tag_set.queue_depth = 128;
2013 lo->tag_set.numa_node = NUMA_NO_NODE;
2014 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2015 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2016 lo->tag_set.driver_data = lo;
2018 err = blk_mq_alloc_tag_set(&lo->tag_set);
2022 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2023 if (IS_ERR(lo->lo_queue)) {
2024 err = PTR_ERR(lo->lo_queue);
2025 goto out_cleanup_tags;
2027 lo->lo_queue->queuedata = lo;
2029 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2032 * By default, we do buffer IO, so it doesn't make sense to enable
2033 * merge because the I/O submitted to backing file is handled page by
2034 * page. For directio mode, merge does help to dispatch bigger request
2035 * to underlayer disk. We will enable merge once directio is enabled.
2037 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2040 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2042 goto out_free_queue;
2045 * Disable partition scanning by default. The in-kernel partition
2046 * scanning can be requested individually per-device during its
2047 * setup. Userspace can always add and remove partitions from all
2048 * devices. The needed partition minors are allocated from the
2049 * extended minor space, the main loop device numbers will continue
2050 * to match the loop minors, regardless of the number of partitions
2053 * If max_part is given, partition scanning is globally enabled for
2054 * all loop devices. The minors for the main loop devices will be
2055 * multiples of max_part.
2057 * Note: Global-for-all-devices, set-only-at-init, read-only module
2058 * parameteters like 'max_loop' and 'max_part' make things needlessly
2059 * complicated, are too static, inflexible and may surprise
2060 * userspace tools. Parameters like this in general should be avoided.
2063 disk->flags |= GENHD_FL_NO_PART_SCAN;
2064 disk->flags |= GENHD_FL_EXT_DEVT;
2065 atomic_set(&lo->lo_refcnt, 0);
2067 spin_lock_init(&lo->lo_lock);
2068 disk->major = LOOP_MAJOR;
2069 disk->first_minor = i << part_shift;
2070 disk->fops = &lo_fops;
2071 disk->private_data = lo;
2072 disk->queue = lo->lo_queue;
2073 sprintf(disk->disk_name, "loop%d", i);
2076 return lo->lo_number;
2079 blk_cleanup_queue(lo->lo_queue);
2081 blk_mq_free_tag_set(&lo->tag_set);
2083 idr_remove(&loop_index_idr, i);
2090 static void loop_remove(struct loop_device *lo)
2092 del_gendisk(lo->lo_disk);
2093 blk_cleanup_queue(lo->lo_queue);
2094 blk_mq_free_tag_set(&lo->tag_set);
2095 put_disk(lo->lo_disk);
2099 static int find_free_cb(int id, void *ptr, void *data)
2101 struct loop_device *lo = ptr;
2102 struct loop_device **l = data;
2104 if (lo->lo_state == Lo_unbound) {
2111 static int loop_lookup(struct loop_device **l, int i)
2113 struct loop_device *lo;
2119 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2122 ret = lo->lo_number;
2127 /* lookup and return a specific i */
2128 lo = idr_find(&loop_index_idr, i);
2131 ret = lo->lo_number;
2137 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2139 struct loop_device *lo;
2140 struct kobject *kobj;
2143 mutex_lock(&loop_ctl_mutex);
2144 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2146 err = loop_add(&lo, MINOR(dev) >> part_shift);
2150 kobj = get_disk_and_module(lo->lo_disk);
2151 mutex_unlock(&loop_ctl_mutex);
2157 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2160 struct loop_device *lo;
2163 ret = mutex_lock_killable(&loop_ctl_mutex);
2170 ret = loop_lookup(&lo, parm);
2175 ret = loop_add(&lo, parm);
2177 case LOOP_CTL_REMOVE:
2178 ret = loop_lookup(&lo, parm);
2181 if (lo->lo_state != Lo_unbound) {
2185 if (atomic_read(&lo->lo_refcnt) > 0) {
2189 lo->lo_disk->private_data = NULL;
2190 idr_remove(&loop_index_idr, lo->lo_number);
2193 case LOOP_CTL_GET_FREE:
2194 ret = loop_lookup(&lo, -1);
2197 ret = loop_add(&lo, -1);
2199 mutex_unlock(&loop_ctl_mutex);
2204 static const struct file_operations loop_ctl_fops = {
2205 .open = nonseekable_open,
2206 .unlocked_ioctl = loop_control_ioctl,
2207 .compat_ioctl = loop_control_ioctl,
2208 .owner = THIS_MODULE,
2209 .llseek = noop_llseek,
2212 static struct miscdevice loop_misc = {
2213 .minor = LOOP_CTRL_MINOR,
2214 .name = "loop-control",
2215 .fops = &loop_ctl_fops,
2218 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2219 MODULE_ALIAS("devname:loop-control");
2221 static int __init loop_init(void)
2224 unsigned long range;
2225 struct loop_device *lo;
2230 part_shift = fls(max_part);
2233 * Adjust max_part according to part_shift as it is exported
2234 * to user space so that user can decide correct minor number
2235 * if [s]he want to create more devices.
2237 * Note that -1 is required because partition 0 is reserved
2238 * for the whole disk.
2240 max_part = (1UL << part_shift) - 1;
2243 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2248 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2254 * If max_loop is specified, create that many devices upfront.
2255 * This also becomes a hard limit. If max_loop is not specified,
2256 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2257 * init time. Loop devices can be requested on-demand with the
2258 * /dev/loop-control interface, or be instantiated by accessing
2259 * a 'dead' device node.
2263 range = max_loop << part_shift;
2265 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2266 range = 1UL << MINORBITS;
2269 err = misc_register(&loop_misc);
2274 if (register_blkdev(LOOP_MAJOR, "loop")) {
2279 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2280 THIS_MODULE, loop_probe, NULL, NULL);
2282 /* pre-create number of devices given by config or max_loop */
2283 mutex_lock(&loop_ctl_mutex);
2284 for (i = 0; i < nr; i++)
2286 mutex_unlock(&loop_ctl_mutex);
2288 printk(KERN_INFO "loop: module loaded\n");
2292 misc_deregister(&loop_misc);
2297 static int loop_exit_cb(int id, void *ptr, void *data)
2299 struct loop_device *lo = ptr;
2305 static void __exit loop_exit(void)
2307 unsigned long range;
2309 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2311 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2312 idr_destroy(&loop_index_idr);
2314 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2315 unregister_blkdev(LOOP_MAJOR, "loop");
2317 misc_deregister(&loop_misc);
2320 module_init(loop_init);
2321 module_exit(loop_exit);
2324 static int __init max_loop_setup(char *str)
2326 max_loop = simple_strtol(str, NULL, 0);
2330 __setup("max_loop=", max_loop_setup);