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
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
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.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
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
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
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>
83 #include <linux/uaccess.h>
85 static DEFINE_IDR(loop_index_idr);
86 static DEFINE_MUTEX(loop_ctl_mutex);
89 static int part_shift;
91 static int transfer_xor(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
96 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
109 key = lo->lo_encrypt_key;
110 keysize = lo->lo_encrypt_key_size;
111 for (i = 0; i < size; i++)
112 *out++ = *in++ ^ key[(i & 511) % keysize];
114 kunmap_atomic(loop_buf);
115 kunmap_atomic(raw_buf);
120 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122 if (unlikely(info->lo_encrypt_key_size <= 0))
127 static struct loop_func_table none_funcs = {
128 .number = LO_CRYPT_NONE,
131 static struct loop_func_table xor_funcs = {
132 .number = LO_CRYPT_XOR,
133 .transfer = transfer_xor,
137 /* xfer_funcs[0] is special - its release function is never called */
138 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
143 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
147 /* Compute loopsize in bytes */
148 loopsize = i_size_read(file->f_mapping->host);
151 /* offset is beyond i_size, weird but possible */
155 if (sizelimit > 0 && sizelimit < loopsize)
156 loopsize = sizelimit;
158 * Unfortunately, if we want to do I/O on the device,
159 * the number of 512-byte sectors has to fit into a sector_t.
161 return loopsize >> 9;
164 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
169 static void __loop_update_dio(struct loop_device *lo, bool dio)
171 struct file *file = lo->lo_backing_file;
172 struct address_space *mapping = file->f_mapping;
173 struct inode *inode = mapping->host;
174 unsigned short sb_bsize = 0;
175 unsigned dio_align = 0;
178 if (inode->i_sb->s_bdev) {
179 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
180 dio_align = sb_bsize - 1;
184 * We support direct I/O only if lo_offset is aligned with the
185 * logical I/O size of backing device, and the logical block
186 * size of loop is bigger than the backing device's and the loop
187 * needn't transform transfer.
189 * TODO: the above condition may be loosed in the future, and
190 * direct I/O may be switched runtime at that time because most
191 * of requests in sane applications should be PAGE_SIZE aligned
194 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
195 !(lo->lo_offset & dio_align) &&
196 mapping->a_ops->direct_IO &&
205 if (lo->use_dio == use_dio)
208 /* flush dirty pages before changing direct IO */
212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
214 * will get updated by ioctl(LOOP_GET_STATUS)
216 blk_mq_freeze_queue(lo->lo_queue);
217 lo->use_dio = use_dio;
219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
220 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
223 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
225 blk_mq_unfreeze_queue(lo->lo_queue);
229 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
231 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
232 sector_t x = (sector_t)size;
233 struct block_device *bdev = lo->lo_device;
235 if (unlikely((loff_t)x != size))
237 if (lo->lo_offset != offset)
238 lo->lo_offset = offset;
239 if (lo->lo_sizelimit != sizelimit)
240 lo->lo_sizelimit = sizelimit;
241 set_capacity(lo->lo_disk, x);
242 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
243 /* let user-space know about the new size */
244 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
249 lo_do_transfer(struct loop_device *lo, int cmd,
250 struct page *rpage, unsigned roffs,
251 struct page *lpage, unsigned loffs,
252 int size, sector_t rblock)
256 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
260 printk_ratelimited(KERN_ERR
261 "loop: Transfer error at byte offset %llu, length %i.\n",
262 (unsigned long long)rblock << 9, size);
266 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
271 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len);
273 file_start_write(file);
274 bw = vfs_iter_write(file, &i, ppos, 0);
275 file_end_write(file);
277 if (likely(bw == bvec->bv_len))
280 printk_ratelimited(KERN_ERR
281 "loop: Write error at byte offset %llu, length %i.\n",
282 (unsigned long long)*ppos, bvec->bv_len);
288 static int lo_write_simple(struct loop_device *lo, struct request *rq,
292 struct req_iterator iter;
295 rq_for_each_segment(bvec, rq, iter) {
296 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
306 * This is the slow, transforming version that needs to double buffer the
307 * data as it cannot do the transformations in place without having direct
308 * access to the destination pages of the backing file.
310 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
313 struct bio_vec bvec, b;
314 struct req_iterator iter;
318 page = alloc_page(GFP_NOIO);
322 rq_for_each_segment(bvec, rq, iter) {
323 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
324 bvec.bv_offset, bvec.bv_len, pos >> 9);
330 b.bv_len = bvec.bv_len;
331 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
340 static int lo_read_simple(struct loop_device *lo, struct request *rq,
344 struct req_iterator iter;
348 rq_for_each_segment(bvec, rq, iter) {
349 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
350 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
354 flush_dcache_page(bvec.bv_page);
356 if (len != bvec.bv_len) {
359 __rq_for_each_bio(bio, rq)
369 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
372 struct bio_vec bvec, b;
373 struct req_iterator iter;
379 page = alloc_page(GFP_NOIO);
383 rq_for_each_segment(bvec, rq, iter) {
388 b.bv_len = bvec.bv_len;
390 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
397 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
398 bvec.bv_offset, len, offset >> 9);
402 flush_dcache_page(bvec.bv_page);
404 if (len != bvec.bv_len) {
407 __rq_for_each_bio(bio, rq)
419 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
423 * We use fallocate to manipulate the space mappings used by the image
424 * a.k.a. discard/zerorange. However we do not support this if
425 * encryption is enabled, because it may give an attacker useful
428 struct file *file = lo->lo_backing_file;
429 struct request_queue *q = lo->lo_queue;
432 mode |= FALLOC_FL_KEEP_SIZE;
434 if (!blk_queue_discard(q)) {
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 bio_vec *bvec;
517 struct request *rq = blk_mq_rq_from_pdu(cmd);
518 struct bio *bio = rq->bio;
519 struct file *file = lo->lo_backing_file;
524 if (rq->bio != rq->biotail) {
525 struct req_iterator iter;
528 __rq_for_each_bio(bio, rq)
529 segments += bio_segments(bio);
530 bvec = kmalloc_array(segments, sizeof(struct bio_vec),
537 * The bios of the request may be started from the middle of
538 * the 'bvec' because of bio splitting, so we can't directly
539 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
540 * API will take care of all details for us.
542 rq_for_each_segment(tmp, rq, iter) {
550 * Same here, this bio may be started from the middle of the
551 * 'bvec' because of bio splitting, so offset from the bvec
552 * must be passed to iov iterator
554 offset = bio->bi_iter.bi_bvec_done;
555 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
556 segments = bio_segments(bio);
558 atomic_set(&cmd->ref, 2);
560 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
561 segments, blk_rq_bytes(rq));
562 iter.iov_offset = offset;
564 cmd->iocb.ki_pos = pos;
565 cmd->iocb.ki_filp = file;
566 cmd->iocb.ki_complete = lo_rw_aio_complete;
567 cmd->iocb.ki_flags = IOCB_DIRECT;
568 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
570 kthread_associate_blkcg(cmd->css);
573 ret = call_write_iter(file, &cmd->iocb, &iter);
575 ret = call_read_iter(file, &cmd->iocb, &iter);
577 lo_rw_aio_do_completion(cmd);
578 kthread_associate_blkcg(NULL);
580 if (ret != -EIOCBQUEUED)
581 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
585 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
587 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
588 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
591 * lo_write_simple and lo_read_simple should have been covered
592 * by io submit style function like lo_rw_aio(), one blocker
593 * is that lo_read_simple() need to call flush_dcache_page after
594 * the page is written from kernel, and it isn't easy to handle
595 * this in io submit style function which submits all segments
596 * of the req at one time. And direct read IO doesn't need to
597 * run flush_dcache_page().
599 switch (req_op(rq)) {
601 return lo_req_flush(lo, rq);
602 case REQ_OP_WRITE_ZEROES:
604 * If the caller doesn't want deallocation, call zeroout to
605 * write zeroes the range. Otherwise, punch them out.
607 return lo_fallocate(lo, rq, pos,
608 (rq->cmd_flags & REQ_NOUNMAP) ?
609 FALLOC_FL_ZERO_RANGE :
610 FALLOC_FL_PUNCH_HOLE);
612 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
615 return lo_write_transfer(lo, rq, pos);
616 else if (cmd->use_aio)
617 return lo_rw_aio(lo, cmd, pos, WRITE);
619 return lo_write_simple(lo, rq, pos);
622 return lo_read_transfer(lo, rq, pos);
623 else if (cmd->use_aio)
624 return lo_rw_aio(lo, cmd, pos, READ);
626 return lo_read_simple(lo, rq, pos);
634 static inline void loop_update_dio(struct loop_device *lo)
636 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
640 static void loop_reread_partitions(struct loop_device *lo,
641 struct block_device *bdev)
645 rc = blkdev_reread_part(bdev);
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 sysfs_emit(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 sysfs_emit(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 sysfs_emit(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 sysfs_emit(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 sysfs_emit(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;
867 u32 granularity, max_discard_sectors;
870 * If the backing device is a block device, mirror its zeroing
871 * capability. Set the discard sectors to the block device's zeroing
872 * capabilities because loop discards result in blkdev_issue_zeroout(),
873 * not blkdev_issue_discard(). This maintains consistent behavior with
874 * file-backed loop devices: discarded regions read back as zero.
876 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
877 struct request_queue *backingq;
879 backingq = bdev_get_queue(inode->i_bdev);
881 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
882 granularity = backingq->limits.discard_granularity ?:
883 queue_physical_block_size(backingq);
886 * We use punch hole to reclaim the free space used by the
887 * image a.k.a. discard. However we do not support discard if
888 * encryption is enabled, because it may give an attacker
889 * useful information.
891 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
892 max_discard_sectors = 0;
896 max_discard_sectors = UINT_MAX >> 9;
897 granularity = inode->i_sb->s_blocksize;
900 if (max_discard_sectors) {
901 q->limits.discard_granularity = granularity;
902 blk_queue_max_discard_sectors(q, max_discard_sectors);
903 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
904 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
906 q->limits.discard_granularity = 0;
907 blk_queue_max_discard_sectors(q, 0);
908 blk_queue_max_write_zeroes_sectors(q, 0);
909 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
911 q->limits.discard_alignment = 0;
914 static void loop_unprepare_queue(struct loop_device *lo)
916 kthread_flush_worker(&lo->worker);
917 kthread_stop(lo->worker_task);
920 static int loop_kthread_worker_fn(void *worker_ptr)
922 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
923 return kthread_worker_fn(worker_ptr);
926 static int loop_prepare_queue(struct loop_device *lo)
928 kthread_init_worker(&lo->worker);
929 lo->worker_task = kthread_run(loop_kthread_worker_fn,
930 &lo->worker, "loop%d", lo->lo_number);
931 if (IS_ERR(lo->worker_task))
933 set_user_nice(lo->worker_task, MIN_NICE);
937 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
938 struct block_device *bdev, unsigned int arg)
942 struct address_space *mapping;
948 /* This is safe, since we have a reference from open(). */
949 __module_get(THIS_MODULE);
956 error = mutex_lock_killable(&loop_ctl_mutex);
961 if (lo->lo_state != Lo_unbound)
964 error = loop_validate_file(file, bdev);
968 mapping = file->f_mapping;
969 inode = mapping->host;
971 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
972 !file->f_op->write_iter)
973 lo_flags |= LO_FLAGS_READ_ONLY;
976 size = get_loop_size(lo, file);
977 if ((loff_t)(sector_t)size != size)
979 error = loop_prepare_queue(lo);
985 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
988 lo->lo_device = bdev;
989 lo->lo_flags = lo_flags;
990 lo->lo_backing_file = file;
993 lo->lo_sizelimit = 0;
994 lo->old_gfp_mask = mapping_gfp_mask(mapping);
995 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
997 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
998 blk_queue_write_cache(lo->lo_queue, true, false);
1000 loop_update_dio(lo);
1001 set_capacity(lo->lo_disk, size);
1002 bd_set_size(bdev, size << 9);
1003 loop_sysfs_init(lo);
1004 /* let user-space know about the new size */
1005 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1007 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1008 block_size(inode->i_bdev) : PAGE_SIZE);
1010 lo->lo_state = Lo_bound;
1012 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1013 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1015 /* Grab the block_device to prevent its destruction after we
1016 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1019 mutex_unlock(&loop_ctl_mutex);
1021 loop_reread_partitions(lo, bdev);
1025 mutex_unlock(&loop_ctl_mutex);
1029 /* This is safe: open() is still holding a reference. */
1030 module_put(THIS_MODULE);
1035 loop_release_xfer(struct loop_device *lo)
1038 struct loop_func_table *xfer = lo->lo_encryption;
1042 err = xfer->release(lo);
1043 lo->transfer = NULL;
1044 lo->lo_encryption = NULL;
1045 module_put(xfer->owner);
1051 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1052 const struct loop_info64 *i)
1057 struct module *owner = xfer->owner;
1059 if (!try_module_get(owner))
1062 err = xfer->init(lo, i);
1066 lo->lo_encryption = xfer;
1071 static int __loop_clr_fd(struct loop_device *lo, bool release)
1073 struct file *filp = NULL;
1074 gfp_t gfp = lo->old_gfp_mask;
1075 struct block_device *bdev = lo->lo_device;
1077 bool partscan = false;
1080 mutex_lock(&loop_ctl_mutex);
1081 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1086 filp = lo->lo_backing_file;
1092 /* freeze request queue during the transition */
1093 blk_mq_freeze_queue(lo->lo_queue);
1095 spin_lock_irq(&lo->lo_lock);
1096 lo->lo_backing_file = NULL;
1097 spin_unlock_irq(&lo->lo_lock);
1099 loop_release_xfer(lo);
1100 lo->transfer = NULL;
1102 lo->lo_device = NULL;
1103 lo->lo_encryption = NULL;
1105 lo->lo_sizelimit = 0;
1106 lo->lo_encrypt_key_size = 0;
1107 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1108 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1109 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1110 blk_queue_logical_block_size(lo->lo_queue, 512);
1111 blk_queue_physical_block_size(lo->lo_queue, 512);
1112 blk_queue_io_min(lo->lo_queue, 512);
1115 invalidate_bdev(bdev);
1116 bdev->bd_inode->i_mapping->wb_err = 0;
1118 set_capacity(lo->lo_disk, 0);
1119 loop_sysfs_exit(lo);
1121 bd_set_size(bdev, 0);
1122 /* let user-space know about this change */
1123 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1125 mapping_set_gfp_mask(filp->f_mapping, gfp);
1126 /* This is safe: open() is still holding a reference. */
1127 module_put(THIS_MODULE);
1128 blk_mq_unfreeze_queue(lo->lo_queue);
1130 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1131 lo_number = lo->lo_number;
1132 loop_unprepare_queue(lo);
1134 mutex_unlock(&loop_ctl_mutex);
1137 * bd_mutex has been held already in release path, so don't
1138 * acquire it if this function is called in such case.
1140 * If the reread partition isn't from release path, lo_refcnt
1141 * must be at least one and it can only become zero when the
1142 * current holder is released.
1145 err = __blkdev_reread_part(bdev);
1147 err = blkdev_reread_part(bdev);
1149 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1150 __func__, lo_number, err);
1151 /* Device is gone, no point in returning error */
1156 * lo->lo_state is set to Lo_unbound here after above partscan has
1159 * There cannot be anybody else entering __loop_clr_fd() as
1160 * lo->lo_backing_file is already cleared and Lo_rundown state
1161 * protects us from all the other places trying to change the 'lo'
1164 mutex_lock(&loop_ctl_mutex);
1167 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1168 lo->lo_state = Lo_unbound;
1169 mutex_unlock(&loop_ctl_mutex);
1172 * Need not hold loop_ctl_mutex to fput backing file.
1173 * Calling fput holding loop_ctl_mutex triggers a circular
1174 * lock dependency possibility warning as fput can take
1175 * bd_mutex which is usually taken before loop_ctl_mutex.
1182 static int loop_clr_fd(struct loop_device *lo)
1186 err = mutex_lock_killable(&loop_ctl_mutex);
1189 if (lo->lo_state != Lo_bound) {
1190 mutex_unlock(&loop_ctl_mutex);
1194 * If we've explicitly asked to tear down the loop device,
1195 * and it has an elevated reference count, set it for auto-teardown when
1196 * the last reference goes away. This stops $!~#$@ udev from
1197 * preventing teardown because it decided that it needs to run blkid on
1198 * the loopback device whenever they appear. xfstests is notorious for
1199 * failing tests because blkid via udev races with a losetup
1200 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1201 * command to fail with EBUSY.
1203 if (atomic_read(&lo->lo_refcnt) > 1) {
1204 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1205 mutex_unlock(&loop_ctl_mutex);
1208 lo->lo_state = Lo_rundown;
1209 mutex_unlock(&loop_ctl_mutex);
1211 return __loop_clr_fd(lo, false);
1215 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1218 struct loop_func_table *xfer;
1219 kuid_t uid = current_uid();
1220 struct block_device *bdev;
1221 bool partscan = false;
1223 err = mutex_lock_killable(&loop_ctl_mutex);
1226 if (lo->lo_encrypt_key_size &&
1227 !uid_eq(lo->lo_key_owner, uid) &&
1228 !capable(CAP_SYS_ADMIN)) {
1232 if (lo->lo_state != Lo_bound) {
1236 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1241 if (lo->lo_offset != info->lo_offset ||
1242 lo->lo_sizelimit != info->lo_sizelimit) {
1243 sync_blockdev(lo->lo_device);
1244 invalidate_bdev(lo->lo_device);
1247 /* I/O need to be drained during transfer transition */
1248 blk_mq_freeze_queue(lo->lo_queue);
1250 err = loop_release_xfer(lo);
1254 if (info->lo_encrypt_type) {
1255 unsigned int type = info->lo_encrypt_type;
1257 if (type >= MAX_LO_CRYPT) {
1261 xfer = xfer_funcs[type];
1269 err = loop_init_xfer(lo, xfer, info);
1273 if (lo->lo_offset != info->lo_offset ||
1274 lo->lo_sizelimit != info->lo_sizelimit) {
1275 /* kill_bdev should have truncated all the pages */
1276 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1278 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1279 __func__, lo->lo_number, lo->lo_file_name,
1280 lo->lo_device->bd_inode->i_mapping->nrpages);
1283 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1289 loop_config_discard(lo);
1291 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1292 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1293 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1294 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1298 lo->transfer = xfer->transfer;
1299 lo->ioctl = xfer->ioctl;
1301 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1302 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1303 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1305 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1306 lo->lo_init[0] = info->lo_init[0];
1307 lo->lo_init[1] = info->lo_init[1];
1308 if (info->lo_encrypt_key_size) {
1309 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1310 info->lo_encrypt_key_size);
1311 lo->lo_key_owner = uid;
1314 /* update dio if lo_offset or transfer is changed */
1315 __loop_update_dio(lo, lo->use_dio);
1318 blk_mq_unfreeze_queue(lo->lo_queue);
1320 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1321 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1322 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1323 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1324 bdev = lo->lo_device;
1328 mutex_unlock(&loop_ctl_mutex);
1330 loop_reread_partitions(lo, bdev);
1336 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1342 ret = mutex_lock_killable(&loop_ctl_mutex);
1345 if (lo->lo_state != Lo_bound) {
1346 mutex_unlock(&loop_ctl_mutex);
1350 memset(info, 0, sizeof(*info));
1351 info->lo_number = lo->lo_number;
1352 info->lo_offset = lo->lo_offset;
1353 info->lo_sizelimit = lo->lo_sizelimit;
1354 info->lo_flags = lo->lo_flags;
1355 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1356 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1357 info->lo_encrypt_type =
1358 lo->lo_encryption ? lo->lo_encryption->number : 0;
1359 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1360 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1361 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1362 lo->lo_encrypt_key_size);
1365 /* Drop loop_ctl_mutex while we call into the filesystem. */
1366 path = lo->lo_backing_file->f_path;
1368 mutex_unlock(&loop_ctl_mutex);
1369 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1371 info->lo_device = huge_encode_dev(stat.dev);
1372 info->lo_inode = stat.ino;
1373 info->lo_rdevice = huge_encode_dev(stat.rdev);
1380 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1382 memset(info64, 0, sizeof(*info64));
1383 info64->lo_number = info->lo_number;
1384 info64->lo_device = info->lo_device;
1385 info64->lo_inode = info->lo_inode;
1386 info64->lo_rdevice = info->lo_rdevice;
1387 info64->lo_offset = info->lo_offset;
1388 info64->lo_sizelimit = 0;
1389 info64->lo_encrypt_type = info->lo_encrypt_type;
1390 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1391 info64->lo_flags = info->lo_flags;
1392 info64->lo_init[0] = info->lo_init[0];
1393 info64->lo_init[1] = info->lo_init[1];
1394 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1395 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1397 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1398 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1402 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1404 memset(info, 0, sizeof(*info));
1405 info->lo_number = info64->lo_number;
1406 info->lo_device = info64->lo_device;
1407 info->lo_inode = info64->lo_inode;
1408 info->lo_rdevice = info64->lo_rdevice;
1409 info->lo_offset = info64->lo_offset;
1410 info->lo_encrypt_type = info64->lo_encrypt_type;
1411 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1412 info->lo_flags = info64->lo_flags;
1413 info->lo_init[0] = info64->lo_init[0];
1414 info->lo_init[1] = info64->lo_init[1];
1415 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1416 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1418 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1419 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1421 /* error in case values were truncated */
1422 if (info->lo_device != info64->lo_device ||
1423 info->lo_rdevice != info64->lo_rdevice ||
1424 info->lo_inode != info64->lo_inode ||
1425 info->lo_offset != info64->lo_offset)
1432 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1434 struct loop_info info;
1435 struct loop_info64 info64;
1437 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1439 loop_info64_from_old(&info, &info64);
1440 return loop_set_status(lo, &info64);
1444 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1446 struct loop_info64 info64;
1448 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1450 return loop_set_status(lo, &info64);
1454 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1455 struct loop_info info;
1456 struct loop_info64 info64;
1461 err = loop_get_status(lo, &info64);
1463 err = loop_info64_to_old(&info64, &info);
1464 if (!err && copy_to_user(arg, &info, sizeof(info)))
1471 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1472 struct loop_info64 info64;
1477 err = loop_get_status(lo, &info64);
1478 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1484 static int loop_set_capacity(struct loop_device *lo)
1486 if (unlikely(lo->lo_state != Lo_bound))
1489 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1492 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1495 if (lo->lo_state != Lo_bound)
1498 __loop_update_dio(lo, !!arg);
1499 if (lo->use_dio == !!arg)
1506 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1510 if (lo->lo_state != Lo_bound)
1513 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1516 if (lo->lo_queue->limits.logical_block_size != arg) {
1517 sync_blockdev(lo->lo_device);
1518 invalidate_bdev(lo->lo_device);
1521 blk_mq_freeze_queue(lo->lo_queue);
1523 /* invalidate_bdev should have truncated all the pages */
1524 if (lo->lo_queue->limits.logical_block_size != arg &&
1525 lo->lo_device->bd_inode->i_mapping->nrpages) {
1527 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1528 __func__, lo->lo_number, lo->lo_file_name,
1529 lo->lo_device->bd_inode->i_mapping->nrpages);
1533 blk_queue_logical_block_size(lo->lo_queue, arg);
1534 blk_queue_physical_block_size(lo->lo_queue, arg);
1535 blk_queue_io_min(lo->lo_queue, arg);
1536 loop_update_dio(lo);
1538 blk_mq_unfreeze_queue(lo->lo_queue);
1543 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1548 err = mutex_lock_killable(&loop_ctl_mutex);
1552 case LOOP_SET_CAPACITY:
1553 err = loop_set_capacity(lo);
1555 case LOOP_SET_DIRECT_IO:
1556 err = loop_set_dio(lo, arg);
1558 case LOOP_SET_BLOCK_SIZE:
1559 err = loop_set_block_size(lo, arg);
1562 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1564 mutex_unlock(&loop_ctl_mutex);
1568 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1569 unsigned int cmd, unsigned long arg)
1571 struct loop_device *lo = bdev->bd_disk->private_data;
1576 return loop_set_fd(lo, mode, bdev, arg);
1577 case LOOP_CHANGE_FD:
1578 return loop_change_fd(lo, bdev, arg);
1580 return loop_clr_fd(lo);
1581 case LOOP_SET_STATUS:
1583 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1584 err = loop_set_status_old(lo,
1585 (struct loop_info __user *)arg);
1588 case LOOP_GET_STATUS:
1589 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1590 case LOOP_SET_STATUS64:
1592 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1593 err = loop_set_status64(lo,
1594 (struct loop_info64 __user *) arg);
1597 case LOOP_GET_STATUS64:
1598 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1599 case LOOP_SET_CAPACITY:
1600 case LOOP_SET_DIRECT_IO:
1601 case LOOP_SET_BLOCK_SIZE:
1602 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1606 err = lo_simple_ioctl(lo, cmd, arg);
1613 #ifdef CONFIG_COMPAT
1614 struct compat_loop_info {
1615 compat_int_t lo_number; /* ioctl r/o */
1616 compat_dev_t lo_device; /* ioctl r/o */
1617 compat_ulong_t lo_inode; /* ioctl r/o */
1618 compat_dev_t lo_rdevice; /* ioctl r/o */
1619 compat_int_t lo_offset;
1620 compat_int_t lo_encrypt_type;
1621 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1622 compat_int_t lo_flags; /* ioctl r/o */
1623 char lo_name[LO_NAME_SIZE];
1624 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1625 compat_ulong_t lo_init[2];
1630 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1631 * - noinlined to reduce stack space usage in main part of driver
1634 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1635 struct loop_info64 *info64)
1637 struct compat_loop_info info;
1639 if (copy_from_user(&info, arg, sizeof(info)))
1642 memset(info64, 0, sizeof(*info64));
1643 info64->lo_number = info.lo_number;
1644 info64->lo_device = info.lo_device;
1645 info64->lo_inode = info.lo_inode;
1646 info64->lo_rdevice = info.lo_rdevice;
1647 info64->lo_offset = info.lo_offset;
1648 info64->lo_sizelimit = 0;
1649 info64->lo_encrypt_type = info.lo_encrypt_type;
1650 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1651 info64->lo_flags = info.lo_flags;
1652 info64->lo_init[0] = info.lo_init[0];
1653 info64->lo_init[1] = info.lo_init[1];
1654 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1655 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1657 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1658 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1663 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1664 * - noinlined to reduce stack space usage in main part of driver
1667 loop_info64_to_compat(const struct loop_info64 *info64,
1668 struct compat_loop_info __user *arg)
1670 struct compat_loop_info info;
1672 memset(&info, 0, sizeof(info));
1673 info.lo_number = info64->lo_number;
1674 info.lo_device = info64->lo_device;
1675 info.lo_inode = info64->lo_inode;
1676 info.lo_rdevice = info64->lo_rdevice;
1677 info.lo_offset = info64->lo_offset;
1678 info.lo_encrypt_type = info64->lo_encrypt_type;
1679 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1680 info.lo_flags = info64->lo_flags;
1681 info.lo_init[0] = info64->lo_init[0];
1682 info.lo_init[1] = info64->lo_init[1];
1683 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1684 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1686 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1687 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1689 /* error in case values were truncated */
1690 if (info.lo_device != info64->lo_device ||
1691 info.lo_rdevice != info64->lo_rdevice ||
1692 info.lo_inode != info64->lo_inode ||
1693 info.lo_offset != info64->lo_offset ||
1694 info.lo_init[0] != info64->lo_init[0] ||
1695 info.lo_init[1] != info64->lo_init[1])
1698 if (copy_to_user(arg, &info, sizeof(info)))
1704 loop_set_status_compat(struct loop_device *lo,
1705 const struct compat_loop_info __user *arg)
1707 struct loop_info64 info64;
1710 ret = loop_info64_from_compat(arg, &info64);
1713 return loop_set_status(lo, &info64);
1717 loop_get_status_compat(struct loop_device *lo,
1718 struct compat_loop_info __user *arg)
1720 struct loop_info64 info64;
1725 err = loop_get_status(lo, &info64);
1727 err = loop_info64_to_compat(&info64, arg);
1731 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1732 unsigned int cmd, unsigned long arg)
1734 struct loop_device *lo = bdev->bd_disk->private_data;
1738 case LOOP_SET_STATUS:
1739 err = loop_set_status_compat(lo,
1740 (const struct compat_loop_info __user *)arg);
1742 case LOOP_GET_STATUS:
1743 err = loop_get_status_compat(lo,
1744 (struct compat_loop_info __user *)arg);
1746 case LOOP_SET_CAPACITY:
1748 case LOOP_GET_STATUS64:
1749 case LOOP_SET_STATUS64:
1750 arg = (unsigned long) compat_ptr(arg);
1753 case LOOP_CHANGE_FD:
1754 case LOOP_SET_BLOCK_SIZE:
1755 case LOOP_SET_DIRECT_IO:
1756 err = lo_ioctl(bdev, mode, cmd, arg);
1766 static int lo_open(struct block_device *bdev, fmode_t mode)
1768 struct loop_device *lo;
1771 err = mutex_lock_killable(&loop_ctl_mutex);
1774 lo = bdev->bd_disk->private_data;
1780 atomic_inc(&lo->lo_refcnt);
1782 mutex_unlock(&loop_ctl_mutex);
1786 static void lo_release(struct gendisk *disk, fmode_t mode)
1788 struct loop_device *lo;
1790 mutex_lock(&loop_ctl_mutex);
1791 lo = disk->private_data;
1792 if (atomic_dec_return(&lo->lo_refcnt))
1795 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1796 if (lo->lo_state != Lo_bound)
1798 lo->lo_state = Lo_rundown;
1799 mutex_unlock(&loop_ctl_mutex);
1801 * In autoclear mode, stop the loop thread
1802 * and remove configuration after last close.
1804 __loop_clr_fd(lo, true);
1806 } else if (lo->lo_state == Lo_bound) {
1808 * Otherwise keep thread (if running) and config,
1809 * but flush possible ongoing bios in thread.
1811 blk_mq_freeze_queue(lo->lo_queue);
1812 blk_mq_unfreeze_queue(lo->lo_queue);
1816 mutex_unlock(&loop_ctl_mutex);
1819 static const struct block_device_operations lo_fops = {
1820 .owner = THIS_MODULE,
1822 .release = lo_release,
1824 #ifdef CONFIG_COMPAT
1825 .compat_ioctl = lo_compat_ioctl,
1830 * And now the modules code and kernel interface.
1832 static int max_loop;
1833 module_param(max_loop, int, 0444);
1834 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1835 module_param(max_part, int, 0444);
1836 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1837 MODULE_LICENSE("GPL");
1838 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1840 int loop_register_transfer(struct loop_func_table *funcs)
1842 unsigned int n = funcs->number;
1844 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1846 xfer_funcs[n] = funcs;
1850 static int unregister_transfer_cb(int id, void *ptr, void *data)
1852 struct loop_device *lo = ptr;
1853 struct loop_func_table *xfer = data;
1855 mutex_lock(&loop_ctl_mutex);
1856 if (lo->lo_encryption == xfer)
1857 loop_release_xfer(lo);
1858 mutex_unlock(&loop_ctl_mutex);
1862 int loop_unregister_transfer(int number)
1864 unsigned int n = number;
1865 struct loop_func_table *xfer;
1867 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1870 xfer_funcs[n] = NULL;
1871 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1875 EXPORT_SYMBOL(loop_register_transfer);
1876 EXPORT_SYMBOL(loop_unregister_transfer);
1878 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1879 const struct blk_mq_queue_data *bd)
1881 struct request *rq = bd->rq;
1882 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1883 struct loop_device *lo = rq->q->queuedata;
1885 blk_mq_start_request(rq);
1887 if (lo->lo_state != Lo_bound)
1888 return BLK_STS_IOERR;
1890 switch (req_op(rq)) {
1892 case REQ_OP_DISCARD:
1893 case REQ_OP_WRITE_ZEROES:
1894 cmd->use_aio = false;
1897 cmd->use_aio = lo->use_dio;
1901 /* always use the first bio's css */
1902 #ifdef CONFIG_BLK_CGROUP
1903 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1904 cmd->css = rq->bio->bi_css;
1909 kthread_queue_work(&lo->worker, &cmd->work);
1914 static void loop_handle_cmd(struct loop_cmd *cmd)
1916 struct request *rq = blk_mq_rq_from_pdu(cmd);
1917 const bool write = op_is_write(req_op(rq));
1918 struct loop_device *lo = rq->q->queuedata;
1921 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1926 ret = do_req_filebacked(lo, rq);
1928 /* complete non-aio request */
1929 if (!cmd->use_aio || ret) {
1930 cmd->ret = ret ? -EIO : 0;
1931 blk_mq_complete_request(rq);
1935 static void loop_queue_work(struct kthread_work *work)
1937 struct loop_cmd *cmd =
1938 container_of(work, struct loop_cmd, work);
1940 loop_handle_cmd(cmd);
1943 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1944 unsigned int hctx_idx, unsigned int numa_node)
1946 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1948 kthread_init_work(&cmd->work, loop_queue_work);
1952 static const struct blk_mq_ops loop_mq_ops = {
1953 .queue_rq = loop_queue_rq,
1954 .init_request = loop_init_request,
1955 .complete = lo_complete_rq,
1958 static int loop_add(struct loop_device **l, int i)
1960 struct loop_device *lo;
1961 struct gendisk *disk;
1965 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1969 lo->lo_state = Lo_unbound;
1971 /* allocate id, if @id >= 0, we're requesting that specific id */
1973 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1977 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1984 lo->tag_set.ops = &loop_mq_ops;
1985 lo->tag_set.nr_hw_queues = 1;
1986 lo->tag_set.queue_depth = 128;
1987 lo->tag_set.numa_node = NUMA_NO_NODE;
1988 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1989 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1990 lo->tag_set.driver_data = lo;
1992 err = blk_mq_alloc_tag_set(&lo->tag_set);
1996 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1997 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1998 err = PTR_ERR(lo->lo_queue);
1999 goto out_cleanup_tags;
2001 lo->lo_queue->queuedata = lo;
2003 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2006 * By default, we do buffer IO, so it doesn't make sense to enable
2007 * merge because the I/O submitted to backing file is handled page by
2008 * page. For directio mode, merge does help to dispatch bigger request
2009 * to underlayer disk. We will enable merge once directio is enabled.
2011 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2014 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2016 goto out_free_queue;
2019 * Disable partition scanning by default. The in-kernel partition
2020 * scanning can be requested individually per-device during its
2021 * setup. Userspace can always add and remove partitions from all
2022 * devices. The needed partition minors are allocated from the
2023 * extended minor space, the main loop device numbers will continue
2024 * to match the loop minors, regardless of the number of partitions
2027 * If max_part is given, partition scanning is globally enabled for
2028 * all loop devices. The minors for the main loop devices will be
2029 * multiples of max_part.
2031 * Note: Global-for-all-devices, set-only-at-init, read-only module
2032 * parameteters like 'max_loop' and 'max_part' make things needlessly
2033 * complicated, are too static, inflexible and may surprise
2034 * userspace tools. Parameters like this in general should be avoided.
2037 disk->flags |= GENHD_FL_NO_PART_SCAN;
2038 disk->flags |= GENHD_FL_EXT_DEVT;
2039 atomic_set(&lo->lo_refcnt, 0);
2041 spin_lock_init(&lo->lo_lock);
2042 disk->major = LOOP_MAJOR;
2043 disk->first_minor = i << part_shift;
2044 disk->fops = &lo_fops;
2045 disk->private_data = lo;
2046 disk->queue = lo->lo_queue;
2047 sprintf(disk->disk_name, "loop%d", i);
2050 return lo->lo_number;
2053 blk_cleanup_queue(lo->lo_queue);
2055 blk_mq_free_tag_set(&lo->tag_set);
2057 idr_remove(&loop_index_idr, i);
2064 static void loop_remove(struct loop_device *lo)
2066 del_gendisk(lo->lo_disk);
2067 blk_cleanup_queue(lo->lo_queue);
2068 blk_mq_free_tag_set(&lo->tag_set);
2069 put_disk(lo->lo_disk);
2073 static int find_free_cb(int id, void *ptr, void *data)
2075 struct loop_device *lo = ptr;
2076 struct loop_device **l = data;
2078 if (lo->lo_state == Lo_unbound) {
2085 static int loop_lookup(struct loop_device **l, int i)
2087 struct loop_device *lo;
2093 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2096 ret = lo->lo_number;
2101 /* lookup and return a specific i */
2102 lo = idr_find(&loop_index_idr, i);
2105 ret = lo->lo_number;
2111 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2113 struct loop_device *lo;
2114 struct kobject *kobj;
2117 mutex_lock(&loop_ctl_mutex);
2118 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2120 err = loop_add(&lo, MINOR(dev) >> part_shift);
2124 kobj = get_disk_and_module(lo->lo_disk);
2125 mutex_unlock(&loop_ctl_mutex);
2131 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2134 struct loop_device *lo;
2137 ret = mutex_lock_killable(&loop_ctl_mutex);
2144 ret = loop_lookup(&lo, parm);
2149 ret = loop_add(&lo, parm);
2151 case LOOP_CTL_REMOVE:
2152 ret = loop_lookup(&lo, parm);
2155 if (lo->lo_state != Lo_unbound) {
2159 if (atomic_read(&lo->lo_refcnt) > 0) {
2163 lo->lo_disk->private_data = NULL;
2164 idr_remove(&loop_index_idr, lo->lo_number);
2167 case LOOP_CTL_GET_FREE:
2168 ret = loop_lookup(&lo, -1);
2171 ret = loop_add(&lo, -1);
2173 mutex_unlock(&loop_ctl_mutex);
2178 static const struct file_operations loop_ctl_fops = {
2179 .open = nonseekable_open,
2180 .unlocked_ioctl = loop_control_ioctl,
2181 .compat_ioctl = loop_control_ioctl,
2182 .owner = THIS_MODULE,
2183 .llseek = noop_llseek,
2186 static struct miscdevice loop_misc = {
2187 .minor = LOOP_CTRL_MINOR,
2188 .name = "loop-control",
2189 .fops = &loop_ctl_fops,
2192 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2193 MODULE_ALIAS("devname:loop-control");
2195 static int __init loop_init(void)
2198 unsigned long range;
2199 struct loop_device *lo;
2204 part_shift = fls(max_part);
2207 * Adjust max_part according to part_shift as it is exported
2208 * to user space so that user can decide correct minor number
2209 * if [s]he want to create more devices.
2211 * Note that -1 is required because partition 0 is reserved
2212 * for the whole disk.
2214 max_part = (1UL << part_shift) - 1;
2217 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2222 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2228 * If max_loop is specified, create that many devices upfront.
2229 * This also becomes a hard limit. If max_loop is not specified,
2230 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2231 * init time. Loop devices can be requested on-demand with the
2232 * /dev/loop-control interface, or be instantiated by accessing
2233 * a 'dead' device node.
2237 range = max_loop << part_shift;
2239 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2240 range = 1UL << MINORBITS;
2243 err = misc_register(&loop_misc);
2248 if (register_blkdev(LOOP_MAJOR, "loop")) {
2253 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2254 THIS_MODULE, loop_probe, NULL, NULL);
2256 /* pre-create number of devices given by config or max_loop */
2257 mutex_lock(&loop_ctl_mutex);
2258 for (i = 0; i < nr; i++)
2260 mutex_unlock(&loop_ctl_mutex);
2262 printk(KERN_INFO "loop: module loaded\n");
2266 misc_deregister(&loop_misc);
2271 static int loop_exit_cb(int id, void *ptr, void *data)
2273 struct loop_device *lo = ptr;
2279 static void __exit loop_exit(void)
2281 unsigned long range;
2283 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2285 mutex_lock(&loop_ctl_mutex);
2287 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2288 idr_destroy(&loop_index_idr);
2290 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2291 unregister_blkdev(LOOP_MAJOR, "loop");
2293 misc_deregister(&loop_misc);
2295 mutex_unlock(&loop_ctl_mutex);
2298 module_init(loop_init);
2299 module_exit(loop_exit);
2302 static int __init max_loop_setup(char *str)
2304 max_loop = simple_strtol(str, NULL, 0);
2308 __setup("max_loop=", max_loop_setup);