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>
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;
430 struct request_queue *q = lo->lo_queue;
433 mode |= FALLOC_FL_KEEP_SIZE;
435 if (!blk_queue_discard(q)) {
440 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
441 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
447 static int lo_req_flush(struct loop_device *lo, struct request *rq)
449 struct file *file = lo->lo_backing_file;
450 int ret = vfs_fsync(file, 0);
451 if (unlikely(ret && ret != -EINVAL))
457 static void lo_complete_rq(struct request *rq)
459 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
460 blk_status_t ret = BLK_STS_OK;
462 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
463 req_op(rq) != REQ_OP_READ) {
470 * Short READ - if we got some data, advance our request and
471 * retry it. If we got no data, end the rest with EIO.
474 blk_update_request(rq, BLK_STS_OK, cmd->ret);
476 blk_mq_requeue_request(rq, true);
479 struct bio *bio = rq->bio;
488 blk_mq_end_request(rq, ret);
492 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
494 struct request *rq = blk_mq_rq_from_pdu(cmd);
496 if (!atomic_dec_and_test(&cmd->ref))
500 blk_mq_complete_request(rq);
503 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
505 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
510 lo_rw_aio_do_completion(cmd);
513 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
516 struct iov_iter iter;
517 struct req_iterator rq_iter;
518 struct bio_vec *bvec;
519 struct request *rq = blk_mq_rq_from_pdu(cmd);
520 struct bio *bio = rq->bio;
521 struct file *file = lo->lo_backing_file;
527 rq_for_each_bvec(tmp, rq, rq_iter)
530 if (rq->bio != rq->biotail) {
532 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
539 * The bios of the request may be started from the middle of
540 * the 'bvec' because of bio splitting, so we can't directly
541 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
542 * API will take care of all details for us.
544 rq_for_each_bvec(tmp, rq, rq_iter) {
552 * Same here, this bio may be started from the middle of the
553 * 'bvec' because of bio splitting, so offset from the bvec
554 * must be passed to iov iterator
556 offset = bio->bi_iter.bi_bvec_done;
557 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
559 atomic_set(&cmd->ref, 2);
561 iov_iter_bvec(&iter, rw, bvec, nr_bvec, 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);
633 static inline void loop_update_dio(struct loop_device *lo)
635 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
639 static void loop_reread_partitions(struct loop_device *lo,
640 struct block_device *bdev)
644 rc = blkdev_reread_part(bdev);
646 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
647 __func__, lo->lo_number, lo->lo_file_name, rc);
650 static inline int is_loop_device(struct file *file)
652 struct inode *i = file->f_mapping->host;
654 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
657 static int loop_validate_file(struct file *file, struct block_device *bdev)
659 struct inode *inode = file->f_mapping->host;
660 struct file *f = file;
662 /* Avoid recursion */
663 while (is_loop_device(f)) {
664 struct loop_device *l;
666 if (f->f_mapping->host->i_bdev == bdev)
669 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
670 if (l->lo_state != Lo_bound) {
673 f = l->lo_backing_file;
675 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
681 * loop_change_fd switched the backing store of a loopback device to
682 * a new file. This is useful for operating system installers to free up
683 * the original file and in High Availability environments to switch to
684 * an alternative location for the content in case of server meltdown.
685 * This can only work if the loop device is used read-only, and if the
686 * new backing store is the same size and type as the old backing store.
688 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
691 struct file *file = NULL, *old_file;
695 error = mutex_lock_killable(&loop_ctl_mutex);
699 if (lo->lo_state != Lo_bound)
702 /* the loop device has to be read-only */
704 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
712 error = loop_validate_file(file, bdev);
716 old_file = lo->lo_backing_file;
720 /* size of the new backing store needs to be the same */
721 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
725 blk_mq_freeze_queue(lo->lo_queue);
726 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
727 lo->lo_backing_file = file;
728 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
729 mapping_set_gfp_mask(file->f_mapping,
730 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
732 blk_mq_unfreeze_queue(lo->lo_queue);
733 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
734 mutex_unlock(&loop_ctl_mutex);
736 * We must drop file reference outside of loop_ctl_mutex as dropping
737 * the file ref can take bd_mutex which creates circular locking
742 loop_reread_partitions(lo, bdev);
746 mutex_unlock(&loop_ctl_mutex);
752 /* loop sysfs attributes */
754 static ssize_t loop_attr_show(struct device *dev, char *page,
755 ssize_t (*callback)(struct loop_device *, char *))
757 struct gendisk *disk = dev_to_disk(dev);
758 struct loop_device *lo = disk->private_data;
760 return callback(lo, page);
763 #define LOOP_ATTR_RO(_name) \
764 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
765 static ssize_t loop_attr_do_show_##_name(struct device *d, \
766 struct device_attribute *attr, char *b) \
768 return loop_attr_show(d, b, loop_attr_##_name##_show); \
770 static struct device_attribute loop_attr_##_name = \
771 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
773 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
778 spin_lock_irq(&lo->lo_lock);
779 if (lo->lo_backing_file)
780 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
781 spin_unlock_irq(&lo->lo_lock);
783 if (IS_ERR_OR_NULL(p))
787 memmove(buf, p, ret);
795 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
797 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
800 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
802 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
805 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
807 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
809 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
812 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
814 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
816 return sprintf(buf, "%s\n", partscan ? "1" : "0");
819 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
821 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
823 return sprintf(buf, "%s\n", dio ? "1" : "0");
826 LOOP_ATTR_RO(backing_file);
827 LOOP_ATTR_RO(offset);
828 LOOP_ATTR_RO(sizelimit);
829 LOOP_ATTR_RO(autoclear);
830 LOOP_ATTR_RO(partscan);
833 static struct attribute *loop_attrs[] = {
834 &loop_attr_backing_file.attr,
835 &loop_attr_offset.attr,
836 &loop_attr_sizelimit.attr,
837 &loop_attr_autoclear.attr,
838 &loop_attr_partscan.attr,
843 static struct attribute_group loop_attribute_group = {
848 static void loop_sysfs_init(struct loop_device *lo)
850 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
851 &loop_attribute_group);
854 static void loop_sysfs_exit(struct loop_device *lo)
856 if (lo->sysfs_inited)
857 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
858 &loop_attribute_group);
861 static void loop_config_discard(struct loop_device *lo)
863 struct file *file = lo->lo_backing_file;
864 struct inode *inode = file->f_mapping->host;
865 struct request_queue *q = lo->lo_queue;
868 * If the backing device is a block device, mirror its zeroing
869 * capability. Set the discard sectors to the block device's zeroing
870 * capabilities because loop discards result in blkdev_issue_zeroout(),
871 * not blkdev_issue_discard(). This maintains consistent behavior with
872 * file-backed loop devices: discarded regions read back as zero.
874 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
875 struct request_queue *backingq;
877 backingq = bdev_get_queue(inode->i_bdev);
878 blk_queue_max_discard_sectors(q,
879 backingq->limits.max_write_zeroes_sectors);
881 blk_queue_max_write_zeroes_sectors(q,
882 backingq->limits.max_write_zeroes_sectors);
885 * We use punch hole to reclaim the free space used by the
886 * image a.k.a. discard. However we do not support discard if
887 * encryption is enabled, because it may give an attacker
888 * useful information.
890 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
891 q->limits.discard_granularity = 0;
892 q->limits.discard_alignment = 0;
893 blk_queue_max_discard_sectors(q, 0);
894 blk_queue_max_write_zeroes_sectors(q, 0);
897 q->limits.discard_granularity = inode->i_sb->s_blocksize;
898 q->limits.discard_alignment = 0;
900 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
901 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
904 if (q->limits.max_write_zeroes_sectors)
905 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
907 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
910 static void loop_unprepare_queue(struct loop_device *lo)
912 kthread_flush_worker(&lo->worker);
913 kthread_stop(lo->worker_task);
916 static int loop_kthread_worker_fn(void *worker_ptr)
918 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
919 return kthread_worker_fn(worker_ptr);
922 static int loop_prepare_queue(struct loop_device *lo)
924 kthread_init_worker(&lo->worker);
925 lo->worker_task = kthread_run(loop_kthread_worker_fn,
926 &lo->worker, "loop%d", lo->lo_number);
927 if (IS_ERR(lo->worker_task))
929 set_user_nice(lo->worker_task, MIN_NICE);
933 static void loop_update_rotational(struct loop_device *lo)
935 struct file *file = lo->lo_backing_file;
936 struct inode *file_inode = file->f_mapping->host;
937 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
938 struct request_queue *q = lo->lo_queue;
941 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
943 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
946 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
948 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
951 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
952 struct block_device *bdev, unsigned int arg)
956 struct address_space *mapping;
957 struct block_device *claimed_bdev = NULL;
963 /* This is safe, since we have a reference from open(). */
964 __module_get(THIS_MODULE);
972 * If we don't hold exclusive handle for the device, upgrade to it
973 * here to avoid changing device under exclusive owner.
975 if (!(mode & FMODE_EXCL)) {
976 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
977 if (IS_ERR(claimed_bdev)) {
978 error = PTR_ERR(claimed_bdev);
983 error = mutex_lock_killable(&loop_ctl_mutex);
988 if (lo->lo_state != Lo_unbound)
991 error = loop_validate_file(file, bdev);
995 mapping = file->f_mapping;
996 inode = mapping->host;
998 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
999 !file->f_op->write_iter)
1000 lo_flags |= LO_FLAGS_READ_ONLY;
1003 size = get_loop_size(lo, file);
1004 if ((loff_t)(sector_t)size != size)
1006 error = loop_prepare_queue(lo);
1012 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
1014 lo->use_dio = false;
1015 lo->lo_device = bdev;
1016 lo->lo_flags = lo_flags;
1017 lo->lo_backing_file = file;
1018 lo->transfer = NULL;
1020 lo->lo_sizelimit = 0;
1021 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1022 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1024 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1025 blk_queue_write_cache(lo->lo_queue, true, false);
1027 if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) {
1028 /* In case of direct I/O, match underlying block size */
1029 unsigned short bsize = bdev_logical_block_size(
1030 inode->i_sb->s_bdev);
1032 blk_queue_logical_block_size(lo->lo_queue, bsize);
1033 blk_queue_physical_block_size(lo->lo_queue, bsize);
1034 blk_queue_io_min(lo->lo_queue, bsize);
1037 loop_update_rotational(lo);
1038 loop_update_dio(lo);
1039 set_capacity(lo->lo_disk, size);
1040 bd_set_size(bdev, size << 9);
1041 loop_sysfs_init(lo);
1042 /* let user-space know about the new size */
1043 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1045 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1046 block_size(inode->i_bdev) : PAGE_SIZE);
1048 lo->lo_state = Lo_bound;
1050 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1051 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1053 /* Grab the block_device to prevent its destruction after we
1054 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1057 mutex_unlock(&loop_ctl_mutex);
1059 loop_reread_partitions(lo, bdev);
1061 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1065 mutex_unlock(&loop_ctl_mutex);
1068 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1072 /* This is safe: open() is still holding a reference. */
1073 module_put(THIS_MODULE);
1078 loop_release_xfer(struct loop_device *lo)
1081 struct loop_func_table *xfer = lo->lo_encryption;
1085 err = xfer->release(lo);
1086 lo->transfer = NULL;
1087 lo->lo_encryption = NULL;
1088 module_put(xfer->owner);
1094 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1095 const struct loop_info64 *i)
1100 struct module *owner = xfer->owner;
1102 if (!try_module_get(owner))
1105 err = xfer->init(lo, i);
1109 lo->lo_encryption = xfer;
1114 static int __loop_clr_fd(struct loop_device *lo, bool release)
1116 struct file *filp = NULL;
1117 gfp_t gfp = lo->old_gfp_mask;
1118 struct block_device *bdev = lo->lo_device;
1120 bool partscan = false;
1123 mutex_lock(&loop_ctl_mutex);
1124 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1129 filp = lo->lo_backing_file;
1135 /* freeze request queue during the transition */
1136 blk_mq_freeze_queue(lo->lo_queue);
1138 spin_lock_irq(&lo->lo_lock);
1139 lo->lo_backing_file = NULL;
1140 spin_unlock_irq(&lo->lo_lock);
1142 loop_release_xfer(lo);
1143 lo->transfer = NULL;
1145 lo->lo_device = NULL;
1146 lo->lo_encryption = NULL;
1148 lo->lo_sizelimit = 0;
1149 lo->lo_encrypt_key_size = 0;
1150 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1151 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1152 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1153 blk_queue_logical_block_size(lo->lo_queue, 512);
1154 blk_queue_physical_block_size(lo->lo_queue, 512);
1155 blk_queue_io_min(lo->lo_queue, 512);
1158 invalidate_bdev(bdev);
1159 bdev->bd_inode->i_mapping->wb_err = 0;
1161 set_capacity(lo->lo_disk, 0);
1162 loop_sysfs_exit(lo);
1164 bd_set_size(bdev, 0);
1165 /* let user-space know about this change */
1166 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1168 mapping_set_gfp_mask(filp->f_mapping, gfp);
1169 /* This is safe: open() is still holding a reference. */
1170 module_put(THIS_MODULE);
1171 blk_mq_unfreeze_queue(lo->lo_queue);
1173 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1174 lo_number = lo->lo_number;
1175 loop_unprepare_queue(lo);
1177 mutex_unlock(&loop_ctl_mutex);
1180 * bd_mutex has been held already in release path, so don't
1181 * acquire it if this function is called in such case.
1183 * If the reread partition isn't from release path, lo_refcnt
1184 * must be at least one and it can only become zero when the
1185 * current holder is released.
1188 err = __blkdev_reread_part(bdev);
1190 err = blkdev_reread_part(bdev);
1192 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1193 __func__, lo_number, err);
1194 /* Device is gone, no point in returning error */
1199 * lo->lo_state is set to Lo_unbound here after above partscan has
1202 * There cannot be anybody else entering __loop_clr_fd() as
1203 * lo->lo_backing_file is already cleared and Lo_rundown state
1204 * protects us from all the other places trying to change the 'lo'
1207 mutex_lock(&loop_ctl_mutex);
1210 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1211 lo->lo_state = Lo_unbound;
1212 mutex_unlock(&loop_ctl_mutex);
1215 * Need not hold loop_ctl_mutex to fput backing file.
1216 * Calling fput holding loop_ctl_mutex triggers a circular
1217 * lock dependency possibility warning as fput can take
1218 * bd_mutex which is usually taken before loop_ctl_mutex.
1225 static int loop_clr_fd(struct loop_device *lo)
1229 err = mutex_lock_killable(&loop_ctl_mutex);
1232 if (lo->lo_state != Lo_bound) {
1233 mutex_unlock(&loop_ctl_mutex);
1237 * If we've explicitly asked to tear down the loop device,
1238 * and it has an elevated reference count, set it for auto-teardown when
1239 * the last reference goes away. This stops $!~#$@ udev from
1240 * preventing teardown because it decided that it needs to run blkid on
1241 * the loopback device whenever they appear. xfstests is notorious for
1242 * failing tests because blkid via udev races with a losetup
1243 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1244 * command to fail with EBUSY.
1246 if (atomic_read(&lo->lo_refcnt) > 1) {
1247 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1248 mutex_unlock(&loop_ctl_mutex);
1251 lo->lo_state = Lo_rundown;
1252 mutex_unlock(&loop_ctl_mutex);
1254 return __loop_clr_fd(lo, false);
1258 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1261 struct loop_func_table *xfer;
1262 kuid_t uid = current_uid();
1263 struct block_device *bdev;
1264 bool partscan = false;
1266 err = mutex_lock_killable(&loop_ctl_mutex);
1269 if (lo->lo_encrypt_key_size &&
1270 !uid_eq(lo->lo_key_owner, uid) &&
1271 !capable(CAP_SYS_ADMIN)) {
1275 if (lo->lo_state != Lo_bound) {
1279 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1284 if (lo->lo_offset != info->lo_offset ||
1285 lo->lo_sizelimit != info->lo_sizelimit) {
1286 sync_blockdev(lo->lo_device);
1287 kill_bdev(lo->lo_device);
1290 /* I/O need to be drained during transfer transition */
1291 blk_mq_freeze_queue(lo->lo_queue);
1293 err = loop_release_xfer(lo);
1297 if (info->lo_encrypt_type) {
1298 unsigned int type = info->lo_encrypt_type;
1300 if (type >= MAX_LO_CRYPT) {
1304 xfer = xfer_funcs[type];
1312 err = loop_init_xfer(lo, xfer, info);
1316 if (lo->lo_offset != info->lo_offset ||
1317 lo->lo_sizelimit != info->lo_sizelimit) {
1318 /* kill_bdev should have truncated all the pages */
1319 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1321 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1322 __func__, lo->lo_number, lo->lo_file_name,
1323 lo->lo_device->bd_inode->i_mapping->nrpages);
1326 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1332 loop_config_discard(lo);
1334 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1335 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1336 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1337 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1341 lo->transfer = xfer->transfer;
1342 lo->ioctl = xfer->ioctl;
1344 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1345 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1346 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1348 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1349 lo->lo_init[0] = info->lo_init[0];
1350 lo->lo_init[1] = info->lo_init[1];
1351 if (info->lo_encrypt_key_size) {
1352 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1353 info->lo_encrypt_key_size);
1354 lo->lo_key_owner = uid;
1357 /* update dio if lo_offset or transfer is changed */
1358 __loop_update_dio(lo, lo->use_dio);
1361 blk_mq_unfreeze_queue(lo->lo_queue);
1363 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1364 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1365 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1366 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1367 bdev = lo->lo_device;
1371 mutex_unlock(&loop_ctl_mutex);
1373 loop_reread_partitions(lo, bdev);
1379 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1385 ret = mutex_lock_killable(&loop_ctl_mutex);
1388 if (lo->lo_state != Lo_bound) {
1389 mutex_unlock(&loop_ctl_mutex);
1393 memset(info, 0, sizeof(*info));
1394 info->lo_number = lo->lo_number;
1395 info->lo_offset = lo->lo_offset;
1396 info->lo_sizelimit = lo->lo_sizelimit;
1397 info->lo_flags = lo->lo_flags;
1398 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1399 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1400 info->lo_encrypt_type =
1401 lo->lo_encryption ? lo->lo_encryption->number : 0;
1402 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1403 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1404 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1405 lo->lo_encrypt_key_size);
1408 /* Drop loop_ctl_mutex while we call into the filesystem. */
1409 path = lo->lo_backing_file->f_path;
1411 mutex_unlock(&loop_ctl_mutex);
1412 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1414 info->lo_device = huge_encode_dev(stat.dev);
1415 info->lo_inode = stat.ino;
1416 info->lo_rdevice = huge_encode_dev(stat.rdev);
1423 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1425 memset(info64, 0, sizeof(*info64));
1426 info64->lo_number = info->lo_number;
1427 info64->lo_device = info->lo_device;
1428 info64->lo_inode = info->lo_inode;
1429 info64->lo_rdevice = info->lo_rdevice;
1430 info64->lo_offset = info->lo_offset;
1431 info64->lo_sizelimit = 0;
1432 info64->lo_encrypt_type = info->lo_encrypt_type;
1433 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1434 info64->lo_flags = info->lo_flags;
1435 info64->lo_init[0] = info->lo_init[0];
1436 info64->lo_init[1] = info->lo_init[1];
1437 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1438 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1440 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1441 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1445 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1447 memset(info, 0, sizeof(*info));
1448 info->lo_number = info64->lo_number;
1449 info->lo_device = info64->lo_device;
1450 info->lo_inode = info64->lo_inode;
1451 info->lo_rdevice = info64->lo_rdevice;
1452 info->lo_offset = info64->lo_offset;
1453 info->lo_encrypt_type = info64->lo_encrypt_type;
1454 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1455 info->lo_flags = info64->lo_flags;
1456 info->lo_init[0] = info64->lo_init[0];
1457 info->lo_init[1] = info64->lo_init[1];
1458 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1459 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1461 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1462 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1464 /* error in case values were truncated */
1465 if (info->lo_device != info64->lo_device ||
1466 info->lo_rdevice != info64->lo_rdevice ||
1467 info->lo_inode != info64->lo_inode ||
1468 info->lo_offset != info64->lo_offset)
1475 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1477 struct loop_info info;
1478 struct loop_info64 info64;
1480 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1482 loop_info64_from_old(&info, &info64);
1483 return loop_set_status(lo, &info64);
1487 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1489 struct loop_info64 info64;
1491 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1493 return loop_set_status(lo, &info64);
1497 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1498 struct loop_info info;
1499 struct loop_info64 info64;
1504 err = loop_get_status(lo, &info64);
1506 err = loop_info64_to_old(&info64, &info);
1507 if (!err && copy_to_user(arg, &info, sizeof(info)))
1514 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1515 struct loop_info64 info64;
1520 err = loop_get_status(lo, &info64);
1521 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1527 static int loop_set_capacity(struct loop_device *lo)
1529 if (unlikely(lo->lo_state != Lo_bound))
1532 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1535 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1538 if (lo->lo_state != Lo_bound)
1541 __loop_update_dio(lo, !!arg);
1542 if (lo->use_dio == !!arg)
1549 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1553 if (lo->lo_state != Lo_bound)
1556 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1559 if (lo->lo_queue->limits.logical_block_size != arg) {
1560 sync_blockdev(lo->lo_device);
1561 kill_bdev(lo->lo_device);
1564 blk_mq_freeze_queue(lo->lo_queue);
1566 /* kill_bdev should have truncated all the pages */
1567 if (lo->lo_queue->limits.logical_block_size != arg &&
1568 lo->lo_device->bd_inode->i_mapping->nrpages) {
1570 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1571 __func__, lo->lo_number, lo->lo_file_name,
1572 lo->lo_device->bd_inode->i_mapping->nrpages);
1576 blk_queue_logical_block_size(lo->lo_queue, arg);
1577 blk_queue_physical_block_size(lo->lo_queue, arg);
1578 blk_queue_io_min(lo->lo_queue, arg);
1579 loop_update_dio(lo);
1581 blk_mq_unfreeze_queue(lo->lo_queue);
1586 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1591 err = mutex_lock_killable(&loop_ctl_mutex);
1595 case LOOP_SET_CAPACITY:
1596 err = loop_set_capacity(lo);
1598 case LOOP_SET_DIRECT_IO:
1599 err = loop_set_dio(lo, arg);
1601 case LOOP_SET_BLOCK_SIZE:
1602 err = loop_set_block_size(lo, arg);
1605 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1607 mutex_unlock(&loop_ctl_mutex);
1611 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1612 unsigned int cmd, unsigned long arg)
1614 struct loop_device *lo = bdev->bd_disk->private_data;
1619 return loop_set_fd(lo, mode, bdev, arg);
1620 case LOOP_CHANGE_FD:
1621 return loop_change_fd(lo, bdev, arg);
1623 return loop_clr_fd(lo);
1624 case LOOP_SET_STATUS:
1626 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1627 err = loop_set_status_old(lo,
1628 (struct loop_info __user *)arg);
1631 case LOOP_GET_STATUS:
1632 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1633 case LOOP_SET_STATUS64:
1635 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1636 err = loop_set_status64(lo,
1637 (struct loop_info64 __user *) arg);
1640 case LOOP_GET_STATUS64:
1641 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1642 case LOOP_SET_CAPACITY:
1643 case LOOP_SET_DIRECT_IO:
1644 case LOOP_SET_BLOCK_SIZE:
1645 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1649 err = lo_simple_ioctl(lo, cmd, arg);
1656 #ifdef CONFIG_COMPAT
1657 struct compat_loop_info {
1658 compat_int_t lo_number; /* ioctl r/o */
1659 compat_dev_t lo_device; /* ioctl r/o */
1660 compat_ulong_t lo_inode; /* ioctl r/o */
1661 compat_dev_t lo_rdevice; /* ioctl r/o */
1662 compat_int_t lo_offset;
1663 compat_int_t lo_encrypt_type;
1664 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1665 compat_int_t lo_flags; /* ioctl r/o */
1666 char lo_name[LO_NAME_SIZE];
1667 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1668 compat_ulong_t lo_init[2];
1673 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1674 * - noinlined to reduce stack space usage in main part of driver
1677 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1678 struct loop_info64 *info64)
1680 struct compat_loop_info info;
1682 if (copy_from_user(&info, arg, sizeof(info)))
1685 memset(info64, 0, sizeof(*info64));
1686 info64->lo_number = info.lo_number;
1687 info64->lo_device = info.lo_device;
1688 info64->lo_inode = info.lo_inode;
1689 info64->lo_rdevice = info.lo_rdevice;
1690 info64->lo_offset = info.lo_offset;
1691 info64->lo_sizelimit = 0;
1692 info64->lo_encrypt_type = info.lo_encrypt_type;
1693 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1694 info64->lo_flags = info.lo_flags;
1695 info64->lo_init[0] = info.lo_init[0];
1696 info64->lo_init[1] = info.lo_init[1];
1697 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1698 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1700 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1701 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1706 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1707 * - noinlined to reduce stack space usage in main part of driver
1710 loop_info64_to_compat(const struct loop_info64 *info64,
1711 struct compat_loop_info __user *arg)
1713 struct compat_loop_info info;
1715 memset(&info, 0, sizeof(info));
1716 info.lo_number = info64->lo_number;
1717 info.lo_device = info64->lo_device;
1718 info.lo_inode = info64->lo_inode;
1719 info.lo_rdevice = info64->lo_rdevice;
1720 info.lo_offset = info64->lo_offset;
1721 info.lo_encrypt_type = info64->lo_encrypt_type;
1722 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1723 info.lo_flags = info64->lo_flags;
1724 info.lo_init[0] = info64->lo_init[0];
1725 info.lo_init[1] = info64->lo_init[1];
1726 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1727 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1729 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1730 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1732 /* error in case values were truncated */
1733 if (info.lo_device != info64->lo_device ||
1734 info.lo_rdevice != info64->lo_rdevice ||
1735 info.lo_inode != info64->lo_inode ||
1736 info.lo_offset != info64->lo_offset ||
1737 info.lo_init[0] != info64->lo_init[0] ||
1738 info.lo_init[1] != info64->lo_init[1])
1741 if (copy_to_user(arg, &info, sizeof(info)))
1747 loop_set_status_compat(struct loop_device *lo,
1748 const struct compat_loop_info __user *arg)
1750 struct loop_info64 info64;
1753 ret = loop_info64_from_compat(arg, &info64);
1756 return loop_set_status(lo, &info64);
1760 loop_get_status_compat(struct loop_device *lo,
1761 struct compat_loop_info __user *arg)
1763 struct loop_info64 info64;
1768 err = loop_get_status(lo, &info64);
1770 err = loop_info64_to_compat(&info64, arg);
1774 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1775 unsigned int cmd, unsigned long arg)
1777 struct loop_device *lo = bdev->bd_disk->private_data;
1781 case LOOP_SET_STATUS:
1782 err = loop_set_status_compat(lo,
1783 (const struct compat_loop_info __user *)arg);
1785 case LOOP_GET_STATUS:
1786 err = loop_get_status_compat(lo,
1787 (struct compat_loop_info __user *)arg);
1789 case LOOP_SET_CAPACITY:
1791 case LOOP_GET_STATUS64:
1792 case LOOP_SET_STATUS64:
1793 arg = (unsigned long) compat_ptr(arg);
1796 case LOOP_CHANGE_FD:
1797 case LOOP_SET_BLOCK_SIZE:
1798 case LOOP_SET_DIRECT_IO:
1799 err = lo_ioctl(bdev, mode, cmd, arg);
1809 static int lo_open(struct block_device *bdev, fmode_t mode)
1811 struct loop_device *lo;
1814 err = mutex_lock_killable(&loop_ctl_mutex);
1817 lo = bdev->bd_disk->private_data;
1823 atomic_inc(&lo->lo_refcnt);
1825 mutex_unlock(&loop_ctl_mutex);
1829 static void lo_release(struct gendisk *disk, fmode_t mode)
1831 struct loop_device *lo;
1833 mutex_lock(&loop_ctl_mutex);
1834 lo = disk->private_data;
1835 if (atomic_dec_return(&lo->lo_refcnt))
1838 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1839 if (lo->lo_state != Lo_bound)
1841 lo->lo_state = Lo_rundown;
1842 mutex_unlock(&loop_ctl_mutex);
1844 * In autoclear mode, stop the loop thread
1845 * and remove configuration after last close.
1847 __loop_clr_fd(lo, true);
1849 } else if (lo->lo_state == Lo_bound) {
1851 * Otherwise keep thread (if running) and config,
1852 * but flush possible ongoing bios in thread.
1854 blk_mq_freeze_queue(lo->lo_queue);
1855 blk_mq_unfreeze_queue(lo->lo_queue);
1859 mutex_unlock(&loop_ctl_mutex);
1862 static const struct block_device_operations lo_fops = {
1863 .owner = THIS_MODULE,
1865 .release = lo_release,
1867 #ifdef CONFIG_COMPAT
1868 .compat_ioctl = lo_compat_ioctl,
1873 * And now the modules code and kernel interface.
1875 static int max_loop;
1876 module_param(max_loop, int, 0444);
1877 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1878 module_param(max_part, int, 0444);
1879 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1880 MODULE_LICENSE("GPL");
1881 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1883 int loop_register_transfer(struct loop_func_table *funcs)
1885 unsigned int n = funcs->number;
1887 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1889 xfer_funcs[n] = funcs;
1893 static int unregister_transfer_cb(int id, void *ptr, void *data)
1895 struct loop_device *lo = ptr;
1896 struct loop_func_table *xfer = data;
1898 mutex_lock(&loop_ctl_mutex);
1899 if (lo->lo_encryption == xfer)
1900 loop_release_xfer(lo);
1901 mutex_unlock(&loop_ctl_mutex);
1905 int loop_unregister_transfer(int number)
1907 unsigned int n = number;
1908 struct loop_func_table *xfer;
1910 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1913 xfer_funcs[n] = NULL;
1914 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1918 EXPORT_SYMBOL(loop_register_transfer);
1919 EXPORT_SYMBOL(loop_unregister_transfer);
1921 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1922 const struct blk_mq_queue_data *bd)
1924 struct request *rq = bd->rq;
1925 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1926 struct loop_device *lo = rq->q->queuedata;
1928 blk_mq_start_request(rq);
1930 if (lo->lo_state != Lo_bound)
1931 return BLK_STS_IOERR;
1933 switch (req_op(rq)) {
1935 case REQ_OP_DISCARD:
1936 case REQ_OP_WRITE_ZEROES:
1937 cmd->use_aio = false;
1940 cmd->use_aio = lo->use_dio;
1944 /* always use the first bio's css */
1945 #ifdef CONFIG_BLK_CGROUP
1946 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1947 cmd->css = &bio_blkcg(rq->bio)->css;
1952 kthread_queue_work(&lo->worker, &cmd->work);
1957 static void loop_handle_cmd(struct loop_cmd *cmd)
1959 struct request *rq = blk_mq_rq_from_pdu(cmd);
1960 const bool write = op_is_write(req_op(rq));
1961 struct loop_device *lo = rq->q->queuedata;
1964 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1969 ret = do_req_filebacked(lo, rq);
1971 /* complete non-aio request */
1972 if (!cmd->use_aio || ret) {
1973 cmd->ret = ret ? -EIO : 0;
1974 blk_mq_complete_request(rq);
1978 static void loop_queue_work(struct kthread_work *work)
1980 struct loop_cmd *cmd =
1981 container_of(work, struct loop_cmd, work);
1983 loop_handle_cmd(cmd);
1986 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1987 unsigned int hctx_idx, unsigned int numa_node)
1989 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1991 kthread_init_work(&cmd->work, loop_queue_work);
1995 static const struct blk_mq_ops loop_mq_ops = {
1996 .queue_rq = loop_queue_rq,
1997 .init_request = loop_init_request,
1998 .complete = lo_complete_rq,
2001 static int loop_add(struct loop_device **l, int i)
2003 struct loop_device *lo;
2004 struct gendisk *disk;
2008 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2012 lo->lo_state = Lo_unbound;
2014 /* allocate id, if @id >= 0, we're requesting that specific id */
2016 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2020 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2027 lo->tag_set.ops = &loop_mq_ops;
2028 lo->tag_set.nr_hw_queues = 1;
2029 lo->tag_set.queue_depth = 128;
2030 lo->tag_set.numa_node = NUMA_NO_NODE;
2031 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2032 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2033 lo->tag_set.driver_data = lo;
2035 err = blk_mq_alloc_tag_set(&lo->tag_set);
2039 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2040 if (IS_ERR(lo->lo_queue)) {
2041 err = PTR_ERR(lo->lo_queue);
2042 goto out_cleanup_tags;
2044 lo->lo_queue->queuedata = lo;
2046 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2049 * By default, we do buffer IO, so it doesn't make sense to enable
2050 * merge because the I/O submitted to backing file is handled page by
2051 * page. For directio mode, merge does help to dispatch bigger request
2052 * to underlayer disk. We will enable merge once directio is enabled.
2054 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2057 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2059 goto out_free_queue;
2062 * Disable partition scanning by default. The in-kernel partition
2063 * scanning can be requested individually per-device during its
2064 * setup. Userspace can always add and remove partitions from all
2065 * devices. The needed partition minors are allocated from the
2066 * extended minor space, the main loop device numbers will continue
2067 * to match the loop minors, regardless of the number of partitions
2070 * If max_part is given, partition scanning is globally enabled for
2071 * all loop devices. The minors for the main loop devices will be
2072 * multiples of max_part.
2074 * Note: Global-for-all-devices, set-only-at-init, read-only module
2075 * parameteters like 'max_loop' and 'max_part' make things needlessly
2076 * complicated, are too static, inflexible and may surprise
2077 * userspace tools. Parameters like this in general should be avoided.
2080 disk->flags |= GENHD_FL_NO_PART_SCAN;
2081 disk->flags |= GENHD_FL_EXT_DEVT;
2082 atomic_set(&lo->lo_refcnt, 0);
2084 spin_lock_init(&lo->lo_lock);
2085 disk->major = LOOP_MAJOR;
2086 disk->first_minor = i << part_shift;
2087 disk->fops = &lo_fops;
2088 disk->private_data = lo;
2089 disk->queue = lo->lo_queue;
2090 sprintf(disk->disk_name, "loop%d", i);
2093 return lo->lo_number;
2096 blk_cleanup_queue(lo->lo_queue);
2098 blk_mq_free_tag_set(&lo->tag_set);
2100 idr_remove(&loop_index_idr, i);
2107 static void loop_remove(struct loop_device *lo)
2109 del_gendisk(lo->lo_disk);
2110 blk_cleanup_queue(lo->lo_queue);
2111 blk_mq_free_tag_set(&lo->tag_set);
2112 put_disk(lo->lo_disk);
2116 static int find_free_cb(int id, void *ptr, void *data)
2118 struct loop_device *lo = ptr;
2119 struct loop_device **l = data;
2121 if (lo->lo_state == Lo_unbound) {
2128 static int loop_lookup(struct loop_device **l, int i)
2130 struct loop_device *lo;
2136 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2139 ret = lo->lo_number;
2144 /* lookup and return a specific i */
2145 lo = idr_find(&loop_index_idr, i);
2148 ret = lo->lo_number;
2154 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2156 struct loop_device *lo;
2157 struct kobject *kobj;
2160 mutex_lock(&loop_ctl_mutex);
2161 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2163 err = loop_add(&lo, MINOR(dev) >> part_shift);
2167 kobj = get_disk_and_module(lo->lo_disk);
2168 mutex_unlock(&loop_ctl_mutex);
2174 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2177 struct loop_device *lo;
2180 ret = mutex_lock_killable(&loop_ctl_mutex);
2187 ret = loop_lookup(&lo, parm);
2192 ret = loop_add(&lo, parm);
2194 case LOOP_CTL_REMOVE:
2195 ret = loop_lookup(&lo, parm);
2198 if (lo->lo_state != Lo_unbound) {
2202 if (atomic_read(&lo->lo_refcnt) > 0) {
2206 lo->lo_disk->private_data = NULL;
2207 idr_remove(&loop_index_idr, lo->lo_number);
2210 case LOOP_CTL_GET_FREE:
2211 ret = loop_lookup(&lo, -1);
2214 ret = loop_add(&lo, -1);
2216 mutex_unlock(&loop_ctl_mutex);
2221 static const struct file_operations loop_ctl_fops = {
2222 .open = nonseekable_open,
2223 .unlocked_ioctl = loop_control_ioctl,
2224 .compat_ioctl = loop_control_ioctl,
2225 .owner = THIS_MODULE,
2226 .llseek = noop_llseek,
2229 static struct miscdevice loop_misc = {
2230 .minor = LOOP_CTRL_MINOR,
2231 .name = "loop-control",
2232 .fops = &loop_ctl_fops,
2235 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2236 MODULE_ALIAS("devname:loop-control");
2238 static int __init loop_init(void)
2241 unsigned long range;
2242 struct loop_device *lo;
2247 part_shift = fls(max_part);
2250 * Adjust max_part according to part_shift as it is exported
2251 * to user space so that user can decide correct minor number
2252 * if [s]he want to create more devices.
2254 * Note that -1 is required because partition 0 is reserved
2255 * for the whole disk.
2257 max_part = (1UL << part_shift) - 1;
2260 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2265 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2271 * If max_loop is specified, create that many devices upfront.
2272 * This also becomes a hard limit. If max_loop is not specified,
2273 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2274 * init time. Loop devices can be requested on-demand with the
2275 * /dev/loop-control interface, or be instantiated by accessing
2276 * a 'dead' device node.
2280 range = max_loop << part_shift;
2282 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2283 range = 1UL << MINORBITS;
2286 err = misc_register(&loop_misc);
2291 if (register_blkdev(LOOP_MAJOR, "loop")) {
2296 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2297 THIS_MODULE, loop_probe, NULL, NULL);
2299 /* pre-create number of devices given by config or max_loop */
2300 mutex_lock(&loop_ctl_mutex);
2301 for (i = 0; i < nr; i++)
2303 mutex_unlock(&loop_ctl_mutex);
2305 printk(KERN_INFO "loop: module loaded\n");
2309 misc_deregister(&loop_misc);
2314 static int loop_exit_cb(int id, void *ptr, void *data)
2316 struct loop_device *lo = ptr;
2322 static void __exit loop_exit(void)
2324 unsigned long range;
2326 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2328 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2329 idr_destroy(&loop_index_idr);
2331 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2332 unregister_blkdev(LOOP_MAJOR, "loop");
2334 misc_deregister(&loop_misc);
2337 module_init(loop_init);
2338 module_exit(loop_exit);
2341 static int __init max_loop_setup(char *str)
2343 max_loop = simple_strtol(str, NULL, 0);
2347 __setup("max_loop=", max_loop_setup);