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 if (lo->lo_state == Lo_bound)
218 blk_mq_freeze_queue(lo->lo_queue);
219 lo->use_dio = use_dio;
221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
222 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
225 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
227 if (lo->lo_state == Lo_bound)
228 blk_mq_unfreeze_queue(lo->lo_queue);
232 * loop_validate_block_size() - validates the passed in block size
233 * @bsize: size to validate
236 loop_validate_block_size(unsigned short bsize)
238 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
245 * loop_set_size() - sets device size and notifies userspace
246 * @lo: struct loop_device to set the size for
247 * @size: new size of the loop device
249 * Callers must validate that the size passed into this function fits into
250 * a sector_t, eg using loop_validate_size()
252 static void loop_set_size(struct loop_device *lo, loff_t size)
254 struct block_device *bdev = lo->lo_device;
256 bd_set_size(bdev, size << SECTOR_SHIFT);
258 set_capacity_revalidate_and_notify(lo->lo_disk, size, false);
262 lo_do_transfer(struct loop_device *lo, int cmd,
263 struct page *rpage, unsigned roffs,
264 struct page *lpage, unsigned loffs,
265 int size, sector_t rblock)
269 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
273 printk_ratelimited(KERN_ERR
274 "loop: Transfer error at byte offset %llu, length %i.\n",
275 (unsigned long long)rblock << 9, size);
279 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
284 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
286 file_start_write(file);
287 bw = vfs_iter_write(file, &i, ppos, 0);
288 file_end_write(file);
290 if (likely(bw == bvec->bv_len))
293 printk_ratelimited(KERN_ERR
294 "loop: Write error at byte offset %llu, length %i.\n",
295 (unsigned long long)*ppos, bvec->bv_len);
301 static int lo_write_simple(struct loop_device *lo, struct request *rq,
305 struct req_iterator iter;
308 rq_for_each_segment(bvec, rq, iter) {
309 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
319 * This is the slow, transforming version that needs to double buffer the
320 * data as it cannot do the transformations in place without having direct
321 * access to the destination pages of the backing file.
323 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
326 struct bio_vec bvec, b;
327 struct req_iterator iter;
331 page = alloc_page(GFP_NOIO);
335 rq_for_each_segment(bvec, rq, iter) {
336 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
337 bvec.bv_offset, bvec.bv_len, pos >> 9);
343 b.bv_len = bvec.bv_len;
344 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
353 static int lo_read_simple(struct loop_device *lo, struct request *rq,
357 struct req_iterator iter;
361 rq_for_each_segment(bvec, rq, iter) {
362 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
363 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
367 flush_dcache_page(bvec.bv_page);
369 if (len != bvec.bv_len) {
372 __rq_for_each_bio(bio, rq)
382 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
385 struct bio_vec bvec, b;
386 struct req_iterator iter;
392 page = alloc_page(GFP_NOIO);
396 rq_for_each_segment(bvec, rq, iter) {
401 b.bv_len = bvec.bv_len;
403 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
404 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
410 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
411 bvec.bv_offset, len, offset >> 9);
415 flush_dcache_page(bvec.bv_page);
417 if (len != bvec.bv_len) {
420 __rq_for_each_bio(bio, rq)
432 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
436 * We use fallocate to manipulate the space mappings used by the image
437 * a.k.a. discard/zerorange. However we do not support this if
438 * encryption is enabled, because it may give an attacker useful
441 struct file *file = lo->lo_backing_file;
442 struct request_queue *q = lo->lo_queue;
445 mode |= FALLOC_FL_KEEP_SIZE;
447 if (!blk_queue_discard(q)) {
452 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
453 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
459 static int lo_req_flush(struct loop_device *lo, struct request *rq)
461 struct file *file = lo->lo_backing_file;
462 int ret = vfs_fsync(file, 0);
463 if (unlikely(ret && ret != -EINVAL))
469 static void lo_complete_rq(struct request *rq)
471 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
472 blk_status_t ret = BLK_STS_OK;
474 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
475 req_op(rq) != REQ_OP_READ) {
477 ret = errno_to_blk_status(cmd->ret);
482 * Short READ - if we got some data, advance our request and
483 * retry it. If we got no data, end the rest with EIO.
486 blk_update_request(rq, BLK_STS_OK, cmd->ret);
488 blk_mq_requeue_request(rq, true);
491 struct bio *bio = rq->bio;
500 blk_mq_end_request(rq, ret);
504 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
506 struct request *rq = blk_mq_rq_from_pdu(cmd);
508 if (!atomic_dec_and_test(&cmd->ref))
512 if (likely(!blk_should_fake_timeout(rq->q)))
513 blk_mq_complete_request(rq);
516 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
518 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
523 lo_rw_aio_do_completion(cmd);
526 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
529 struct iov_iter iter;
530 struct req_iterator rq_iter;
531 struct bio_vec *bvec;
532 struct request *rq = blk_mq_rq_from_pdu(cmd);
533 struct bio *bio = rq->bio;
534 struct file *file = lo->lo_backing_file;
540 rq_for_each_bvec(tmp, rq, rq_iter)
543 if (rq->bio != rq->biotail) {
545 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
552 * The bios of the request may be started from the middle of
553 * the 'bvec' because of bio splitting, so we can't directly
554 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
555 * API will take care of all details for us.
557 rq_for_each_bvec(tmp, rq, rq_iter) {
565 * Same here, this bio may be started from the middle of the
566 * 'bvec' because of bio splitting, so offset from the bvec
567 * must be passed to iov iterator
569 offset = bio->bi_iter.bi_bvec_done;
570 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
572 atomic_set(&cmd->ref, 2);
574 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
575 iter.iov_offset = offset;
577 cmd->iocb.ki_pos = pos;
578 cmd->iocb.ki_filp = file;
579 cmd->iocb.ki_complete = lo_rw_aio_complete;
580 cmd->iocb.ki_flags = IOCB_DIRECT;
581 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
583 kthread_associate_blkcg(cmd->css);
586 ret = call_write_iter(file, &cmd->iocb, &iter);
588 ret = call_read_iter(file, &cmd->iocb, &iter);
590 lo_rw_aio_do_completion(cmd);
591 kthread_associate_blkcg(NULL);
593 if (ret != -EIOCBQUEUED)
594 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
598 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
600 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
601 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
604 * lo_write_simple and lo_read_simple should have been covered
605 * by io submit style function like lo_rw_aio(), one blocker
606 * is that lo_read_simple() need to call flush_dcache_page after
607 * the page is written from kernel, and it isn't easy to handle
608 * this in io submit style function which submits all segments
609 * of the req at one time. And direct read IO doesn't need to
610 * run flush_dcache_page().
612 switch (req_op(rq)) {
614 return lo_req_flush(lo, rq);
615 case REQ_OP_WRITE_ZEROES:
617 * If the caller doesn't want deallocation, call zeroout to
618 * write zeroes the range. Otherwise, punch them out.
620 return lo_fallocate(lo, rq, pos,
621 (rq->cmd_flags & REQ_NOUNMAP) ?
622 FALLOC_FL_ZERO_RANGE :
623 FALLOC_FL_PUNCH_HOLE);
625 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
628 return lo_write_transfer(lo, rq, pos);
629 else if (cmd->use_aio)
630 return lo_rw_aio(lo, cmd, pos, WRITE);
632 return lo_write_simple(lo, rq, pos);
635 return lo_read_transfer(lo, rq, pos);
636 else if (cmd->use_aio)
637 return lo_rw_aio(lo, cmd, pos, READ);
639 return lo_read_simple(lo, rq, pos);
646 static inline void loop_update_dio(struct loop_device *lo)
648 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
652 static void loop_reread_partitions(struct loop_device *lo,
653 struct block_device *bdev)
657 mutex_lock(&bdev->bd_mutex);
658 rc = bdev_disk_changed(bdev, false);
659 mutex_unlock(&bdev->bd_mutex);
661 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
662 __func__, lo->lo_number, lo->lo_file_name, rc);
665 static inline int is_loop_device(struct file *file)
667 struct inode *i = file->f_mapping->host;
669 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
672 static int loop_validate_file(struct file *file, struct block_device *bdev)
674 struct inode *inode = file->f_mapping->host;
675 struct file *f = file;
677 /* Avoid recursion */
678 while (is_loop_device(f)) {
679 struct loop_device *l;
681 if (f->f_mapping->host->i_bdev == bdev)
684 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
685 if (l->lo_state != Lo_bound) {
688 f = l->lo_backing_file;
690 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
696 * loop_change_fd switched the backing store of a loopback device to
697 * a new file. This is useful for operating system installers to free up
698 * the original file and in High Availability environments to switch to
699 * an alternative location for the content in case of server meltdown.
700 * This can only work if the loop device is used read-only, and if the
701 * new backing store is the same size and type as the old backing store.
703 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
706 struct file *file = NULL, *old_file;
710 error = mutex_lock_killable(&loop_ctl_mutex);
714 if (lo->lo_state != Lo_bound)
717 /* the loop device has to be read-only */
719 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
727 error = loop_validate_file(file, bdev);
731 old_file = lo->lo_backing_file;
735 /* size of the new backing store needs to be the same */
736 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
740 blk_mq_freeze_queue(lo->lo_queue);
741 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
742 lo->lo_backing_file = file;
743 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
744 mapping_set_gfp_mask(file->f_mapping,
745 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
747 blk_mq_unfreeze_queue(lo->lo_queue);
748 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
749 mutex_unlock(&loop_ctl_mutex);
751 * We must drop file reference outside of loop_ctl_mutex as dropping
752 * the file ref can take bd_mutex which creates circular locking
757 loop_reread_partitions(lo, bdev);
761 mutex_unlock(&loop_ctl_mutex);
767 /* loop sysfs attributes */
769 static ssize_t loop_attr_show(struct device *dev, char *page,
770 ssize_t (*callback)(struct loop_device *, char *))
772 struct gendisk *disk = dev_to_disk(dev);
773 struct loop_device *lo = disk->private_data;
775 return callback(lo, page);
778 #define LOOP_ATTR_RO(_name) \
779 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
780 static ssize_t loop_attr_do_show_##_name(struct device *d, \
781 struct device_attribute *attr, char *b) \
783 return loop_attr_show(d, b, loop_attr_##_name##_show); \
785 static struct device_attribute loop_attr_##_name = \
786 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
788 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
793 spin_lock_irq(&lo->lo_lock);
794 if (lo->lo_backing_file)
795 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
796 spin_unlock_irq(&lo->lo_lock);
798 if (IS_ERR_OR_NULL(p))
802 memmove(buf, p, ret);
810 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
812 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
815 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
817 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
820 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
822 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
824 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
827 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
829 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
831 return sprintf(buf, "%s\n", partscan ? "1" : "0");
834 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
836 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
838 return sprintf(buf, "%s\n", dio ? "1" : "0");
841 LOOP_ATTR_RO(backing_file);
842 LOOP_ATTR_RO(offset);
843 LOOP_ATTR_RO(sizelimit);
844 LOOP_ATTR_RO(autoclear);
845 LOOP_ATTR_RO(partscan);
848 static struct attribute *loop_attrs[] = {
849 &loop_attr_backing_file.attr,
850 &loop_attr_offset.attr,
851 &loop_attr_sizelimit.attr,
852 &loop_attr_autoclear.attr,
853 &loop_attr_partscan.attr,
858 static struct attribute_group loop_attribute_group = {
863 static void loop_sysfs_init(struct loop_device *lo)
865 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
866 &loop_attribute_group);
869 static void loop_sysfs_exit(struct loop_device *lo)
871 if (lo->sysfs_inited)
872 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
873 &loop_attribute_group);
876 static void loop_config_discard(struct loop_device *lo)
878 struct file *file = lo->lo_backing_file;
879 struct inode *inode = file->f_mapping->host;
880 struct request_queue *q = lo->lo_queue;
881 u32 granularity, max_discard_sectors;
884 * If the backing device is a block device, mirror its zeroing
885 * capability. Set the discard sectors to the block device's zeroing
886 * capabilities because loop discards result in blkdev_issue_zeroout(),
887 * not blkdev_issue_discard(). This maintains consistent behavior with
888 * file-backed loop devices: discarded regions read back as zero.
890 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
891 struct request_queue *backingq;
893 backingq = bdev_get_queue(inode->i_bdev);
895 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
896 granularity = backingq->limits.discard_granularity ?:
897 queue_physical_block_size(backingq);
900 * We use punch hole to reclaim the free space used by the
901 * image a.k.a. discard. However we do not support discard if
902 * encryption is enabled, because it may give an attacker
903 * useful information.
905 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
906 max_discard_sectors = 0;
910 max_discard_sectors = UINT_MAX >> 9;
911 granularity = inode->i_sb->s_blocksize;
914 if (max_discard_sectors) {
915 q->limits.discard_granularity = granularity;
916 blk_queue_max_discard_sectors(q, max_discard_sectors);
917 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
918 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
920 q->limits.discard_granularity = 0;
921 blk_queue_max_discard_sectors(q, 0);
922 blk_queue_max_write_zeroes_sectors(q, 0);
923 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
925 q->limits.discard_alignment = 0;
928 static void loop_unprepare_queue(struct loop_device *lo)
930 kthread_flush_worker(&lo->worker);
931 kthread_stop(lo->worker_task);
934 static int loop_kthread_worker_fn(void *worker_ptr)
936 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
937 return kthread_worker_fn(worker_ptr);
940 static int loop_prepare_queue(struct loop_device *lo)
942 kthread_init_worker(&lo->worker);
943 lo->worker_task = kthread_run(loop_kthread_worker_fn,
944 &lo->worker, "loop%d", lo->lo_number);
945 if (IS_ERR(lo->worker_task))
947 set_user_nice(lo->worker_task, MIN_NICE);
951 static void loop_update_rotational(struct loop_device *lo)
953 struct file *file = lo->lo_backing_file;
954 struct inode *file_inode = file->f_mapping->host;
955 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
956 struct request_queue *q = lo->lo_queue;
959 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
961 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
964 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
966 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
970 loop_release_xfer(struct loop_device *lo)
973 struct loop_func_table *xfer = lo->lo_encryption;
977 err = xfer->release(lo);
979 lo->lo_encryption = NULL;
980 module_put(xfer->owner);
986 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
987 const struct loop_info64 *i)
992 struct module *owner = xfer->owner;
994 if (!try_module_get(owner))
997 err = xfer->init(lo, i);
1001 lo->lo_encryption = xfer;
1007 * loop_set_status_from_info - configure device from loop_info
1008 * @lo: struct loop_device to configure
1009 * @info: struct loop_info64 to configure the device with
1011 * Configures the loop device parameters according to the passed
1012 * in loop_info64 configuration.
1015 loop_set_status_from_info(struct loop_device *lo,
1016 const struct loop_info64 *info)
1019 struct loop_func_table *xfer;
1020 kuid_t uid = current_uid();
1022 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1025 err = loop_release_xfer(lo);
1029 if (info->lo_encrypt_type) {
1030 unsigned int type = info->lo_encrypt_type;
1032 if (type >= MAX_LO_CRYPT)
1034 xfer = xfer_funcs[type];
1040 err = loop_init_xfer(lo, xfer, info);
1044 lo->lo_offset = info->lo_offset;
1045 lo->lo_sizelimit = info->lo_sizelimit;
1046 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1047 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1048 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1049 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1053 lo->transfer = xfer->transfer;
1054 lo->ioctl = xfer->ioctl;
1056 lo->lo_flags = info->lo_flags;
1058 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1059 lo->lo_init[0] = info->lo_init[0];
1060 lo->lo_init[1] = info->lo_init[1];
1061 if (info->lo_encrypt_key_size) {
1062 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1063 info->lo_encrypt_key_size);
1064 lo->lo_key_owner = uid;
1070 static int loop_configure(struct loop_device *lo, fmode_t mode,
1071 struct block_device *bdev,
1072 const struct loop_config *config)
1075 struct inode *inode;
1076 struct address_space *mapping;
1077 struct block_device *claimed_bdev = NULL;
1081 unsigned short bsize;
1083 /* This is safe, since we have a reference from open(). */
1084 __module_get(THIS_MODULE);
1087 file = fget(config->fd);
1092 * If we don't hold exclusive handle for the device, upgrade to it
1093 * here to avoid changing device under exclusive owner.
1095 if (!(mode & FMODE_EXCL)) {
1096 claimed_bdev = bdev->bd_contains;
1097 error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure);
1102 error = mutex_lock_killable(&loop_ctl_mutex);
1107 if (lo->lo_state != Lo_unbound)
1110 error = loop_validate_file(file, bdev);
1114 mapping = file->f_mapping;
1115 inode = mapping->host;
1117 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1122 if (config->block_size) {
1123 error = loop_validate_block_size(config->block_size);
1128 error = loop_set_status_from_info(lo, &config->info);
1132 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1133 !file->f_op->write_iter)
1134 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1136 error = loop_prepare_queue(lo);
1140 set_device_ro(bdev, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1142 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1143 lo->lo_device = bdev;
1144 lo->lo_backing_file = file;
1145 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1146 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1148 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1149 blk_queue_write_cache(lo->lo_queue, true, false);
1151 if (config->block_size)
1152 bsize = config->block_size;
1153 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1154 /* In case of direct I/O, match underlying block size */
1155 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1159 blk_queue_logical_block_size(lo->lo_queue, bsize);
1160 blk_queue_physical_block_size(lo->lo_queue, bsize);
1161 blk_queue_io_min(lo->lo_queue, bsize);
1163 loop_update_rotational(lo);
1164 loop_update_dio(lo);
1165 loop_sysfs_init(lo);
1167 size = get_loop_size(lo, file);
1168 loop_set_size(lo, size);
1170 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1171 block_size(inode->i_bdev) : PAGE_SIZE);
1173 lo->lo_state = Lo_bound;
1175 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1176 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1178 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1180 /* Grab the block_device to prevent its destruction after we
1181 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1184 mutex_unlock(&loop_ctl_mutex);
1186 loop_reread_partitions(lo, bdev);
1188 bd_abort_claiming(bdev, claimed_bdev, loop_configure);
1192 mutex_unlock(&loop_ctl_mutex);
1195 bd_abort_claiming(bdev, claimed_bdev, loop_configure);
1199 /* This is safe: open() is still holding a reference. */
1200 module_put(THIS_MODULE);
1204 static int __loop_clr_fd(struct loop_device *lo, bool release)
1206 struct file *filp = NULL;
1207 gfp_t gfp = lo->old_gfp_mask;
1208 struct block_device *bdev = lo->lo_device;
1210 bool partscan = false;
1213 mutex_lock(&loop_ctl_mutex);
1214 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1219 filp = lo->lo_backing_file;
1225 /* freeze request queue during the transition */
1226 blk_mq_freeze_queue(lo->lo_queue);
1228 spin_lock_irq(&lo->lo_lock);
1229 lo->lo_backing_file = NULL;
1230 spin_unlock_irq(&lo->lo_lock);
1232 loop_release_xfer(lo);
1233 lo->transfer = NULL;
1235 lo->lo_device = NULL;
1236 lo->lo_encryption = NULL;
1238 lo->lo_sizelimit = 0;
1239 lo->lo_encrypt_key_size = 0;
1240 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1241 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1242 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1243 blk_queue_logical_block_size(lo->lo_queue, 512);
1244 blk_queue_physical_block_size(lo->lo_queue, 512);
1245 blk_queue_io_min(lo->lo_queue, 512);
1248 invalidate_bdev(bdev);
1249 bdev->bd_inode->i_mapping->wb_err = 0;
1251 set_capacity(lo->lo_disk, 0);
1252 loop_sysfs_exit(lo);
1254 bd_set_size(bdev, 0);
1255 /* let user-space know about this change */
1256 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1258 mapping_set_gfp_mask(filp->f_mapping, gfp);
1259 /* This is safe: open() is still holding a reference. */
1260 module_put(THIS_MODULE);
1261 blk_mq_unfreeze_queue(lo->lo_queue);
1263 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1264 lo_number = lo->lo_number;
1265 loop_unprepare_queue(lo);
1267 mutex_unlock(&loop_ctl_mutex);
1270 * bd_mutex has been held already in release path, so don't
1271 * acquire it if this function is called in such case.
1273 * If the reread partition isn't from release path, lo_refcnt
1274 * must be at least one and it can only become zero when the
1275 * current holder is released.
1278 mutex_lock(&bdev->bd_mutex);
1279 err = bdev_disk_changed(bdev, false);
1281 mutex_unlock(&bdev->bd_mutex);
1283 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1284 __func__, lo_number, err);
1285 /* Device is gone, no point in returning error */
1290 * lo->lo_state is set to Lo_unbound here after above partscan has
1293 * There cannot be anybody else entering __loop_clr_fd() as
1294 * lo->lo_backing_file is already cleared and Lo_rundown state
1295 * protects us from all the other places trying to change the 'lo'
1298 mutex_lock(&loop_ctl_mutex);
1301 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1302 lo->lo_state = Lo_unbound;
1303 mutex_unlock(&loop_ctl_mutex);
1306 * Need not hold loop_ctl_mutex to fput backing file.
1307 * Calling fput holding loop_ctl_mutex triggers a circular
1308 * lock dependency possibility warning as fput can take
1309 * bd_mutex which is usually taken before loop_ctl_mutex.
1316 static int loop_clr_fd(struct loop_device *lo)
1320 err = mutex_lock_killable(&loop_ctl_mutex);
1323 if (lo->lo_state != Lo_bound) {
1324 mutex_unlock(&loop_ctl_mutex);
1328 * If we've explicitly asked to tear down the loop device,
1329 * and it has an elevated reference count, set it for auto-teardown when
1330 * the last reference goes away. This stops $!~#$@ udev from
1331 * preventing teardown because it decided that it needs to run blkid on
1332 * the loopback device whenever they appear. xfstests is notorious for
1333 * failing tests because blkid via udev races with a losetup
1334 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1335 * command to fail with EBUSY.
1337 if (atomic_read(&lo->lo_refcnt) > 1) {
1338 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1339 mutex_unlock(&loop_ctl_mutex);
1342 lo->lo_state = Lo_rundown;
1343 mutex_unlock(&loop_ctl_mutex);
1345 return __loop_clr_fd(lo, false);
1349 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1352 struct block_device *bdev;
1353 kuid_t uid = current_uid();
1355 bool partscan = false;
1356 bool size_changed = false;
1358 err = mutex_lock_killable(&loop_ctl_mutex);
1361 if (lo->lo_encrypt_key_size &&
1362 !uid_eq(lo->lo_key_owner, uid) &&
1363 !capable(CAP_SYS_ADMIN)) {
1367 if (lo->lo_state != Lo_bound) {
1372 if (lo->lo_offset != info->lo_offset ||
1373 lo->lo_sizelimit != info->lo_sizelimit) {
1374 size_changed = true;
1375 sync_blockdev(lo->lo_device);
1376 invalidate_bdev(lo->lo_device);
1379 /* I/O need to be drained during transfer transition */
1380 blk_mq_freeze_queue(lo->lo_queue);
1382 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1383 /* If any pages were dirtied after invalidate_bdev(), try again */
1385 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1386 __func__, lo->lo_number, lo->lo_file_name,
1387 lo->lo_device->bd_inode->i_mapping->nrpages);
1391 prev_lo_flags = lo->lo_flags;
1393 err = loop_set_status_from_info(lo, info);
1397 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1398 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1399 /* For those flags, use the previous values instead */
1400 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1401 /* For flags that can't be cleared, use previous values too */
1402 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1405 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1406 lo->lo_backing_file);
1407 loop_set_size(lo, new_size);
1410 loop_config_discard(lo);
1412 /* update dio if lo_offset or transfer is changed */
1413 __loop_update_dio(lo, lo->use_dio);
1416 blk_mq_unfreeze_queue(lo->lo_queue);
1418 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1419 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1420 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1421 bdev = lo->lo_device;
1425 mutex_unlock(&loop_ctl_mutex);
1427 loop_reread_partitions(lo, bdev);
1433 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1439 ret = mutex_lock_killable(&loop_ctl_mutex);
1442 if (lo->lo_state != Lo_bound) {
1443 mutex_unlock(&loop_ctl_mutex);
1447 memset(info, 0, sizeof(*info));
1448 info->lo_number = lo->lo_number;
1449 info->lo_offset = lo->lo_offset;
1450 info->lo_sizelimit = lo->lo_sizelimit;
1451 info->lo_flags = lo->lo_flags;
1452 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1453 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1454 info->lo_encrypt_type =
1455 lo->lo_encryption ? lo->lo_encryption->number : 0;
1456 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1457 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1458 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1459 lo->lo_encrypt_key_size);
1462 /* Drop loop_ctl_mutex while we call into the filesystem. */
1463 path = lo->lo_backing_file->f_path;
1465 mutex_unlock(&loop_ctl_mutex);
1466 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1468 info->lo_device = huge_encode_dev(stat.dev);
1469 info->lo_inode = stat.ino;
1470 info->lo_rdevice = huge_encode_dev(stat.rdev);
1477 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1479 memset(info64, 0, sizeof(*info64));
1480 info64->lo_number = info->lo_number;
1481 info64->lo_device = info->lo_device;
1482 info64->lo_inode = info->lo_inode;
1483 info64->lo_rdevice = info->lo_rdevice;
1484 info64->lo_offset = info->lo_offset;
1485 info64->lo_sizelimit = 0;
1486 info64->lo_encrypt_type = info->lo_encrypt_type;
1487 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1488 info64->lo_flags = info->lo_flags;
1489 info64->lo_init[0] = info->lo_init[0];
1490 info64->lo_init[1] = info->lo_init[1];
1491 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1492 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1494 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1495 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1499 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1501 memset(info, 0, sizeof(*info));
1502 info->lo_number = info64->lo_number;
1503 info->lo_device = info64->lo_device;
1504 info->lo_inode = info64->lo_inode;
1505 info->lo_rdevice = info64->lo_rdevice;
1506 info->lo_offset = info64->lo_offset;
1507 info->lo_encrypt_type = info64->lo_encrypt_type;
1508 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1509 info->lo_flags = info64->lo_flags;
1510 info->lo_init[0] = info64->lo_init[0];
1511 info->lo_init[1] = info64->lo_init[1];
1512 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1513 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1515 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1516 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1518 /* error in case values were truncated */
1519 if (info->lo_device != info64->lo_device ||
1520 info->lo_rdevice != info64->lo_rdevice ||
1521 info->lo_inode != info64->lo_inode ||
1522 info->lo_offset != info64->lo_offset)
1529 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1531 struct loop_info info;
1532 struct loop_info64 info64;
1534 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1536 loop_info64_from_old(&info, &info64);
1537 return loop_set_status(lo, &info64);
1541 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1543 struct loop_info64 info64;
1545 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1547 return loop_set_status(lo, &info64);
1551 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1552 struct loop_info info;
1553 struct loop_info64 info64;
1558 err = loop_get_status(lo, &info64);
1560 err = loop_info64_to_old(&info64, &info);
1561 if (!err && copy_to_user(arg, &info, sizeof(info)))
1568 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1569 struct loop_info64 info64;
1574 err = loop_get_status(lo, &info64);
1575 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1581 static int loop_set_capacity(struct loop_device *lo)
1585 if (unlikely(lo->lo_state != Lo_bound))
1588 size = get_loop_size(lo, lo->lo_backing_file);
1589 loop_set_size(lo, size);
1594 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1597 if (lo->lo_state != Lo_bound)
1600 __loop_update_dio(lo, !!arg);
1601 if (lo->use_dio == !!arg)
1608 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1612 if (lo->lo_state != Lo_bound)
1615 err = loop_validate_block_size(arg);
1619 if (lo->lo_queue->limits.logical_block_size == arg)
1622 sync_blockdev(lo->lo_device);
1623 invalidate_bdev(lo->lo_device);
1625 blk_mq_freeze_queue(lo->lo_queue);
1627 /* invalidate_bdev should have truncated all the pages */
1628 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1630 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1631 __func__, lo->lo_number, lo->lo_file_name,
1632 lo->lo_device->bd_inode->i_mapping->nrpages);
1636 blk_queue_logical_block_size(lo->lo_queue, arg);
1637 blk_queue_physical_block_size(lo->lo_queue, arg);
1638 blk_queue_io_min(lo->lo_queue, arg);
1639 loop_update_dio(lo);
1641 blk_mq_unfreeze_queue(lo->lo_queue);
1646 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1651 err = mutex_lock_killable(&loop_ctl_mutex);
1655 case LOOP_SET_CAPACITY:
1656 err = loop_set_capacity(lo);
1658 case LOOP_SET_DIRECT_IO:
1659 err = loop_set_dio(lo, arg);
1661 case LOOP_SET_BLOCK_SIZE:
1662 err = loop_set_block_size(lo, arg);
1665 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1667 mutex_unlock(&loop_ctl_mutex);
1671 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1672 unsigned int cmd, unsigned long arg)
1674 struct loop_device *lo = bdev->bd_disk->private_data;
1675 void __user *argp = (void __user *) arg;
1681 * Legacy case - pass in a zeroed out struct loop_config with
1682 * only the file descriptor set , which corresponds with the
1683 * default parameters we'd have used otherwise.
1685 struct loop_config config;
1687 memset(&config, 0, sizeof(config));
1690 return loop_configure(lo, mode, bdev, &config);
1692 case LOOP_CONFIGURE: {
1693 struct loop_config config;
1695 if (copy_from_user(&config, argp, sizeof(config)))
1698 return loop_configure(lo, mode, bdev, &config);
1700 case LOOP_CHANGE_FD:
1701 return loop_change_fd(lo, bdev, arg);
1703 return loop_clr_fd(lo);
1704 case LOOP_SET_STATUS:
1706 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1707 err = loop_set_status_old(lo, argp);
1710 case LOOP_GET_STATUS:
1711 return loop_get_status_old(lo, argp);
1712 case LOOP_SET_STATUS64:
1714 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1715 err = loop_set_status64(lo, argp);
1718 case LOOP_GET_STATUS64:
1719 return loop_get_status64(lo, argp);
1720 case LOOP_SET_CAPACITY:
1721 case LOOP_SET_DIRECT_IO:
1722 case LOOP_SET_BLOCK_SIZE:
1723 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1727 err = lo_simple_ioctl(lo, cmd, arg);
1734 #ifdef CONFIG_COMPAT
1735 struct compat_loop_info {
1736 compat_int_t lo_number; /* ioctl r/o */
1737 compat_dev_t lo_device; /* ioctl r/o */
1738 compat_ulong_t lo_inode; /* ioctl r/o */
1739 compat_dev_t lo_rdevice; /* ioctl r/o */
1740 compat_int_t lo_offset;
1741 compat_int_t lo_encrypt_type;
1742 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1743 compat_int_t lo_flags; /* ioctl r/o */
1744 char lo_name[LO_NAME_SIZE];
1745 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1746 compat_ulong_t lo_init[2];
1751 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1752 * - noinlined to reduce stack space usage in main part of driver
1755 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1756 struct loop_info64 *info64)
1758 struct compat_loop_info info;
1760 if (copy_from_user(&info, arg, sizeof(info)))
1763 memset(info64, 0, sizeof(*info64));
1764 info64->lo_number = info.lo_number;
1765 info64->lo_device = info.lo_device;
1766 info64->lo_inode = info.lo_inode;
1767 info64->lo_rdevice = info.lo_rdevice;
1768 info64->lo_offset = info.lo_offset;
1769 info64->lo_sizelimit = 0;
1770 info64->lo_encrypt_type = info.lo_encrypt_type;
1771 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1772 info64->lo_flags = info.lo_flags;
1773 info64->lo_init[0] = info.lo_init[0];
1774 info64->lo_init[1] = info.lo_init[1];
1775 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1776 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1778 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1779 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1784 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1785 * - noinlined to reduce stack space usage in main part of driver
1788 loop_info64_to_compat(const struct loop_info64 *info64,
1789 struct compat_loop_info __user *arg)
1791 struct compat_loop_info info;
1793 memset(&info, 0, sizeof(info));
1794 info.lo_number = info64->lo_number;
1795 info.lo_device = info64->lo_device;
1796 info.lo_inode = info64->lo_inode;
1797 info.lo_rdevice = info64->lo_rdevice;
1798 info.lo_offset = info64->lo_offset;
1799 info.lo_encrypt_type = info64->lo_encrypt_type;
1800 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1801 info.lo_flags = info64->lo_flags;
1802 info.lo_init[0] = info64->lo_init[0];
1803 info.lo_init[1] = info64->lo_init[1];
1804 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1805 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1807 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1808 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1810 /* error in case values were truncated */
1811 if (info.lo_device != info64->lo_device ||
1812 info.lo_rdevice != info64->lo_rdevice ||
1813 info.lo_inode != info64->lo_inode ||
1814 info.lo_offset != info64->lo_offset ||
1815 info.lo_init[0] != info64->lo_init[0] ||
1816 info.lo_init[1] != info64->lo_init[1])
1819 if (copy_to_user(arg, &info, sizeof(info)))
1825 loop_set_status_compat(struct loop_device *lo,
1826 const struct compat_loop_info __user *arg)
1828 struct loop_info64 info64;
1831 ret = loop_info64_from_compat(arg, &info64);
1834 return loop_set_status(lo, &info64);
1838 loop_get_status_compat(struct loop_device *lo,
1839 struct compat_loop_info __user *arg)
1841 struct loop_info64 info64;
1846 err = loop_get_status(lo, &info64);
1848 err = loop_info64_to_compat(&info64, arg);
1852 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1853 unsigned int cmd, unsigned long arg)
1855 struct loop_device *lo = bdev->bd_disk->private_data;
1859 case LOOP_SET_STATUS:
1860 err = loop_set_status_compat(lo,
1861 (const struct compat_loop_info __user *)arg);
1863 case LOOP_GET_STATUS:
1864 err = loop_get_status_compat(lo,
1865 (struct compat_loop_info __user *)arg);
1867 case LOOP_SET_CAPACITY:
1869 case LOOP_GET_STATUS64:
1870 case LOOP_SET_STATUS64:
1871 case LOOP_CONFIGURE:
1872 arg = (unsigned long) compat_ptr(arg);
1875 case LOOP_CHANGE_FD:
1876 case LOOP_SET_BLOCK_SIZE:
1877 case LOOP_SET_DIRECT_IO:
1878 err = lo_ioctl(bdev, mode, cmd, arg);
1888 static int lo_open(struct block_device *bdev, fmode_t mode)
1890 struct loop_device *lo;
1893 err = mutex_lock_killable(&loop_ctl_mutex);
1896 lo = bdev->bd_disk->private_data;
1902 atomic_inc(&lo->lo_refcnt);
1904 mutex_unlock(&loop_ctl_mutex);
1908 static void lo_release(struct gendisk *disk, fmode_t mode)
1910 struct loop_device *lo;
1912 mutex_lock(&loop_ctl_mutex);
1913 lo = disk->private_data;
1914 if (atomic_dec_return(&lo->lo_refcnt))
1917 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1918 if (lo->lo_state != Lo_bound)
1920 lo->lo_state = Lo_rundown;
1921 mutex_unlock(&loop_ctl_mutex);
1923 * In autoclear mode, stop the loop thread
1924 * and remove configuration after last close.
1926 __loop_clr_fd(lo, true);
1928 } else if (lo->lo_state == Lo_bound) {
1930 * Otherwise keep thread (if running) and config,
1931 * but flush possible ongoing bios in thread.
1933 blk_mq_freeze_queue(lo->lo_queue);
1934 blk_mq_unfreeze_queue(lo->lo_queue);
1938 mutex_unlock(&loop_ctl_mutex);
1941 static const struct block_device_operations lo_fops = {
1942 .owner = THIS_MODULE,
1944 .release = lo_release,
1946 #ifdef CONFIG_COMPAT
1947 .compat_ioctl = lo_compat_ioctl,
1952 * And now the modules code and kernel interface.
1954 static int max_loop;
1955 module_param(max_loop, int, 0444);
1956 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1957 module_param(max_part, int, 0444);
1958 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1959 MODULE_LICENSE("GPL");
1960 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1962 int loop_register_transfer(struct loop_func_table *funcs)
1964 unsigned int n = funcs->number;
1966 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1968 xfer_funcs[n] = funcs;
1972 static int unregister_transfer_cb(int id, void *ptr, void *data)
1974 struct loop_device *lo = ptr;
1975 struct loop_func_table *xfer = data;
1977 mutex_lock(&loop_ctl_mutex);
1978 if (lo->lo_encryption == xfer)
1979 loop_release_xfer(lo);
1980 mutex_unlock(&loop_ctl_mutex);
1984 int loop_unregister_transfer(int number)
1986 unsigned int n = number;
1987 struct loop_func_table *xfer;
1989 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1992 xfer_funcs[n] = NULL;
1993 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1997 EXPORT_SYMBOL(loop_register_transfer);
1998 EXPORT_SYMBOL(loop_unregister_transfer);
2000 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
2001 const struct blk_mq_queue_data *bd)
2003 struct request *rq = bd->rq;
2004 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2005 struct loop_device *lo = rq->q->queuedata;
2007 blk_mq_start_request(rq);
2009 if (lo->lo_state != Lo_bound)
2010 return BLK_STS_IOERR;
2012 switch (req_op(rq)) {
2014 case REQ_OP_DISCARD:
2015 case REQ_OP_WRITE_ZEROES:
2016 cmd->use_aio = false;
2019 cmd->use_aio = lo->use_dio;
2023 /* always use the first bio's css */
2024 #ifdef CONFIG_BLK_CGROUP
2025 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2026 cmd->css = &bio_blkcg(rq->bio)->css;
2031 kthread_queue_work(&lo->worker, &cmd->work);
2036 static void loop_handle_cmd(struct loop_cmd *cmd)
2038 struct request *rq = blk_mq_rq_from_pdu(cmd);
2039 const bool write = op_is_write(req_op(rq));
2040 struct loop_device *lo = rq->q->queuedata;
2043 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2048 ret = do_req_filebacked(lo, rq);
2050 /* complete non-aio request */
2051 if (!cmd->use_aio || ret) {
2052 if (ret == -EOPNOTSUPP)
2055 cmd->ret = ret ? -EIO : 0;
2056 if (likely(!blk_should_fake_timeout(rq->q)))
2057 blk_mq_complete_request(rq);
2061 static void loop_queue_work(struct kthread_work *work)
2063 struct loop_cmd *cmd =
2064 container_of(work, struct loop_cmd, work);
2066 loop_handle_cmd(cmd);
2069 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2070 unsigned int hctx_idx, unsigned int numa_node)
2072 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2074 kthread_init_work(&cmd->work, loop_queue_work);
2078 static const struct blk_mq_ops loop_mq_ops = {
2079 .queue_rq = loop_queue_rq,
2080 .init_request = loop_init_request,
2081 .complete = lo_complete_rq,
2084 static int loop_add(struct loop_device **l, int i)
2086 struct loop_device *lo;
2087 struct gendisk *disk;
2091 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2095 lo->lo_state = Lo_unbound;
2097 /* allocate id, if @id >= 0, we're requesting that specific id */
2099 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2103 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2110 lo->tag_set.ops = &loop_mq_ops;
2111 lo->tag_set.nr_hw_queues = 1;
2112 lo->tag_set.queue_depth = 128;
2113 lo->tag_set.numa_node = NUMA_NO_NODE;
2114 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2115 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
2116 lo->tag_set.driver_data = lo;
2118 err = blk_mq_alloc_tag_set(&lo->tag_set);
2122 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2123 if (IS_ERR(lo->lo_queue)) {
2124 err = PTR_ERR(lo->lo_queue);
2125 goto out_cleanup_tags;
2127 lo->lo_queue->queuedata = lo;
2129 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2132 * By default, we do buffer IO, so it doesn't make sense to enable
2133 * merge because the I/O submitted to backing file is handled page by
2134 * page. For directio mode, merge does help to dispatch bigger request
2135 * to underlayer disk. We will enable merge once directio is enabled.
2137 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2140 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2142 goto out_free_queue;
2145 * Disable partition scanning by default. The in-kernel partition
2146 * scanning can be requested individually per-device during its
2147 * setup. Userspace can always add and remove partitions from all
2148 * devices. The needed partition minors are allocated from the
2149 * extended minor space, the main loop device numbers will continue
2150 * to match the loop minors, regardless of the number of partitions
2153 * If max_part is given, partition scanning is globally enabled for
2154 * all loop devices. The minors for the main loop devices will be
2155 * multiples of max_part.
2157 * Note: Global-for-all-devices, set-only-at-init, read-only module
2158 * parameteters like 'max_loop' and 'max_part' make things needlessly
2159 * complicated, are too static, inflexible and may surprise
2160 * userspace tools. Parameters like this in general should be avoided.
2163 disk->flags |= GENHD_FL_NO_PART_SCAN;
2164 disk->flags |= GENHD_FL_EXT_DEVT;
2165 atomic_set(&lo->lo_refcnt, 0);
2167 spin_lock_init(&lo->lo_lock);
2168 disk->major = LOOP_MAJOR;
2169 disk->first_minor = i << part_shift;
2170 disk->fops = &lo_fops;
2171 disk->private_data = lo;
2172 disk->queue = lo->lo_queue;
2173 sprintf(disk->disk_name, "loop%d", i);
2176 return lo->lo_number;
2179 blk_cleanup_queue(lo->lo_queue);
2181 blk_mq_free_tag_set(&lo->tag_set);
2183 idr_remove(&loop_index_idr, i);
2190 static void loop_remove(struct loop_device *lo)
2192 del_gendisk(lo->lo_disk);
2193 blk_cleanup_queue(lo->lo_queue);
2194 blk_mq_free_tag_set(&lo->tag_set);
2195 put_disk(lo->lo_disk);
2199 static int find_free_cb(int id, void *ptr, void *data)
2201 struct loop_device *lo = ptr;
2202 struct loop_device **l = data;
2204 if (lo->lo_state == Lo_unbound) {
2211 static int loop_lookup(struct loop_device **l, int i)
2213 struct loop_device *lo;
2219 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2222 ret = lo->lo_number;
2227 /* lookup and return a specific i */
2228 lo = idr_find(&loop_index_idr, i);
2231 ret = lo->lo_number;
2237 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2239 struct loop_device *lo;
2240 struct kobject *kobj;
2243 mutex_lock(&loop_ctl_mutex);
2244 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2246 err = loop_add(&lo, MINOR(dev) >> part_shift);
2250 kobj = get_disk_and_module(lo->lo_disk);
2251 mutex_unlock(&loop_ctl_mutex);
2257 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2260 struct loop_device *lo;
2263 ret = mutex_lock_killable(&loop_ctl_mutex);
2270 ret = loop_lookup(&lo, parm);
2275 ret = loop_add(&lo, parm);
2277 case LOOP_CTL_REMOVE:
2278 ret = loop_lookup(&lo, parm);
2281 if (lo->lo_state != Lo_unbound) {
2285 if (atomic_read(&lo->lo_refcnt) > 0) {
2289 lo->lo_disk->private_data = NULL;
2290 idr_remove(&loop_index_idr, lo->lo_number);
2293 case LOOP_CTL_GET_FREE:
2294 ret = loop_lookup(&lo, -1);
2297 ret = loop_add(&lo, -1);
2299 mutex_unlock(&loop_ctl_mutex);
2304 static const struct file_operations loop_ctl_fops = {
2305 .open = nonseekable_open,
2306 .unlocked_ioctl = loop_control_ioctl,
2307 .compat_ioctl = loop_control_ioctl,
2308 .owner = THIS_MODULE,
2309 .llseek = noop_llseek,
2312 static struct miscdevice loop_misc = {
2313 .minor = LOOP_CTRL_MINOR,
2314 .name = "loop-control",
2315 .fops = &loop_ctl_fops,
2318 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2319 MODULE_ALIAS("devname:loop-control");
2321 static int __init loop_init(void)
2324 unsigned long range;
2325 struct loop_device *lo;
2330 part_shift = fls(max_part);
2333 * Adjust max_part according to part_shift as it is exported
2334 * to user space so that user can decide correct minor number
2335 * if [s]he want to create more devices.
2337 * Note that -1 is required because partition 0 is reserved
2338 * for the whole disk.
2340 max_part = (1UL << part_shift) - 1;
2343 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2348 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2354 * If max_loop is specified, create that many devices upfront.
2355 * This also becomes a hard limit. If max_loop is not specified,
2356 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2357 * init time. Loop devices can be requested on-demand with the
2358 * /dev/loop-control interface, or be instantiated by accessing
2359 * a 'dead' device node.
2363 range = max_loop << part_shift;
2365 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2366 range = 1UL << MINORBITS;
2369 err = misc_register(&loop_misc);
2374 if (register_blkdev(LOOP_MAJOR, "loop")) {
2379 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2380 THIS_MODULE, loop_probe, NULL, NULL);
2382 /* pre-create number of devices given by config or max_loop */
2383 mutex_lock(&loop_ctl_mutex);
2384 for (i = 0; i < nr; i++)
2386 mutex_unlock(&loop_ctl_mutex);
2388 printk(KERN_INFO "loop: module loaded\n");
2392 misc_deregister(&loop_misc);
2397 static int loop_exit_cb(int id, void *ptr, void *data)
2399 struct loop_device *lo = ptr;
2405 static void __exit loop_exit(void)
2407 unsigned long range;
2409 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2411 mutex_lock(&loop_ctl_mutex);
2413 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2414 idr_destroy(&loop_index_idr);
2416 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2417 unregister_blkdev(LOOP_MAJOR, "loop");
2419 misc_deregister(&loop_misc);
2421 mutex_unlock(&loop_ctl_mutex);
2424 module_init(loop_init);
2425 module_exit(loop_exit);
2428 static int __init max_loop_setup(char *str)
2430 max_loop = simple_strtol(str, NULL, 0);
2434 __setup("max_loop=", max_loop_setup);