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_nr_sectors(bdev, size);
258 if (!set_capacity_revalidate_and_notify(lo->lo_disk, size, false))
259 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
263 lo_do_transfer(struct loop_device *lo, int cmd,
264 struct page *rpage, unsigned roffs,
265 struct page *lpage, unsigned loffs,
266 int size, sector_t rblock)
270 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
274 printk_ratelimited(KERN_ERR
275 "loop: Transfer error at byte offset %llu, length %i.\n",
276 (unsigned long long)rblock << 9, size);
280 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
285 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
287 file_start_write(file);
288 bw = vfs_iter_write(file, &i, ppos, 0);
289 file_end_write(file);
291 if (likely(bw == bvec->bv_len))
294 printk_ratelimited(KERN_ERR
295 "loop: Write error at byte offset %llu, length %i.\n",
296 (unsigned long long)*ppos, bvec->bv_len);
302 static int lo_write_simple(struct loop_device *lo, struct request *rq,
306 struct req_iterator iter;
309 rq_for_each_segment(bvec, rq, iter) {
310 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
320 * This is the slow, transforming version that needs to double buffer the
321 * data as it cannot do the transformations in place without having direct
322 * access to the destination pages of the backing file.
324 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
327 struct bio_vec bvec, b;
328 struct req_iterator iter;
332 page = alloc_page(GFP_NOIO);
336 rq_for_each_segment(bvec, rq, iter) {
337 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
338 bvec.bv_offset, bvec.bv_len, pos >> 9);
344 b.bv_len = bvec.bv_len;
345 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
354 static int lo_read_simple(struct loop_device *lo, struct request *rq,
358 struct req_iterator iter;
362 rq_for_each_segment(bvec, rq, iter) {
363 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
364 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
368 flush_dcache_page(bvec.bv_page);
370 if (len != bvec.bv_len) {
373 __rq_for_each_bio(bio, rq)
383 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
386 struct bio_vec bvec, b;
387 struct req_iterator iter;
393 page = alloc_page(GFP_NOIO);
397 rq_for_each_segment(bvec, rq, iter) {
402 b.bv_len = bvec.bv_len;
404 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
405 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
411 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
412 bvec.bv_offset, len, offset >> 9);
416 flush_dcache_page(bvec.bv_page);
418 if (len != bvec.bv_len) {
421 __rq_for_each_bio(bio, rq)
433 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
437 * We use fallocate to manipulate the space mappings used by the image
438 * a.k.a. discard/zerorange. However we do not support this if
439 * encryption is enabled, because it may give an attacker useful
442 struct file *file = lo->lo_backing_file;
443 struct request_queue *q = lo->lo_queue;
446 mode |= FALLOC_FL_KEEP_SIZE;
448 if (!blk_queue_discard(q)) {
453 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
454 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
460 static int lo_req_flush(struct loop_device *lo, struct request *rq)
462 struct file *file = lo->lo_backing_file;
463 int ret = vfs_fsync(file, 0);
464 if (unlikely(ret && ret != -EINVAL))
470 static void lo_complete_rq(struct request *rq)
472 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
473 blk_status_t ret = BLK_STS_OK;
475 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
476 req_op(rq) != REQ_OP_READ) {
478 ret = errno_to_blk_status(cmd->ret);
483 * Short READ - if we got some data, advance our request and
484 * retry it. If we got no data, end the rest with EIO.
487 blk_update_request(rq, BLK_STS_OK, cmd->ret);
489 blk_mq_requeue_request(rq, true);
492 struct bio *bio = rq->bio;
501 blk_mq_end_request(rq, ret);
505 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
507 struct request *rq = blk_mq_rq_from_pdu(cmd);
509 if (!atomic_dec_and_test(&cmd->ref))
513 if (likely(!blk_should_fake_timeout(rq->q)))
514 blk_mq_complete_request(rq);
517 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
519 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
524 lo_rw_aio_do_completion(cmd);
527 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
530 struct iov_iter iter;
531 struct req_iterator rq_iter;
532 struct bio_vec *bvec;
533 struct request *rq = blk_mq_rq_from_pdu(cmd);
534 struct bio *bio = rq->bio;
535 struct file *file = lo->lo_backing_file;
541 rq_for_each_bvec(tmp, rq, rq_iter)
544 if (rq->bio != rq->biotail) {
546 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
553 * The bios of the request may be started from the middle of
554 * the 'bvec' because of bio splitting, so we can't directly
555 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
556 * API will take care of all details for us.
558 rq_for_each_bvec(tmp, rq, rq_iter) {
566 * Same here, this bio may be started from the middle of the
567 * 'bvec' because of bio splitting, so offset from the bvec
568 * must be passed to iov iterator
570 offset = bio->bi_iter.bi_bvec_done;
571 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
573 atomic_set(&cmd->ref, 2);
575 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
576 iter.iov_offset = offset;
578 cmd->iocb.ki_pos = pos;
579 cmd->iocb.ki_filp = file;
580 cmd->iocb.ki_complete = lo_rw_aio_complete;
581 cmd->iocb.ki_flags = IOCB_DIRECT;
582 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
584 kthread_associate_blkcg(cmd->css);
587 ret = call_write_iter(file, &cmd->iocb, &iter);
589 ret = call_read_iter(file, &cmd->iocb, &iter);
591 lo_rw_aio_do_completion(cmd);
592 kthread_associate_blkcg(NULL);
594 if (ret != -EIOCBQUEUED)
595 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
599 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
601 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
602 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
605 * lo_write_simple and lo_read_simple should have been covered
606 * by io submit style function like lo_rw_aio(), one blocker
607 * is that lo_read_simple() need to call flush_dcache_page after
608 * the page is written from kernel, and it isn't easy to handle
609 * this in io submit style function which submits all segments
610 * of the req at one time. And direct read IO doesn't need to
611 * run flush_dcache_page().
613 switch (req_op(rq)) {
615 return lo_req_flush(lo, rq);
616 case REQ_OP_WRITE_ZEROES:
618 * If the caller doesn't want deallocation, call zeroout to
619 * write zeroes the range. Otherwise, punch them out.
621 return lo_fallocate(lo, rq, pos,
622 (rq->cmd_flags & REQ_NOUNMAP) ?
623 FALLOC_FL_ZERO_RANGE :
624 FALLOC_FL_PUNCH_HOLE);
626 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
629 return lo_write_transfer(lo, rq, pos);
630 else if (cmd->use_aio)
631 return lo_rw_aio(lo, cmd, pos, WRITE);
633 return lo_write_simple(lo, rq, pos);
636 return lo_read_transfer(lo, rq, pos);
637 else if (cmd->use_aio)
638 return lo_rw_aio(lo, cmd, pos, READ);
640 return lo_read_simple(lo, rq, pos);
647 static inline void loop_update_dio(struct loop_device *lo)
649 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
653 static void loop_reread_partitions(struct loop_device *lo,
654 struct block_device *bdev)
658 mutex_lock(&bdev->bd_mutex);
659 rc = bdev_disk_changed(bdev, false);
660 mutex_unlock(&bdev->bd_mutex);
662 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
663 __func__, lo->lo_number, lo->lo_file_name, rc);
666 static inline int is_loop_device(struct file *file)
668 struct inode *i = file->f_mapping->host;
670 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
673 static int loop_validate_file(struct file *file, struct block_device *bdev)
675 struct inode *inode = file->f_mapping->host;
676 struct file *f = file;
678 /* Avoid recursion */
679 while (is_loop_device(f)) {
680 struct loop_device *l;
682 if (f->f_mapping->host->i_bdev == bdev)
685 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
686 if (l->lo_state != Lo_bound) {
689 f = l->lo_backing_file;
691 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
697 * loop_change_fd switched the backing store of a loopback device to
698 * a new file. This is useful for operating system installers to free up
699 * the original file and in High Availability environments to switch to
700 * an alternative location for the content in case of server meltdown.
701 * This can only work if the loop device is used read-only, and if the
702 * new backing store is the same size and type as the old backing store.
704 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
707 struct file *file = NULL, *old_file;
711 error = mutex_lock_killable(&loop_ctl_mutex);
715 if (lo->lo_state != Lo_bound)
718 /* the loop device has to be read-only */
720 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
728 error = loop_validate_file(file, bdev);
732 old_file = lo->lo_backing_file;
736 /* size of the new backing store needs to be the same */
737 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
741 blk_mq_freeze_queue(lo->lo_queue);
742 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
743 lo->lo_backing_file = file;
744 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
745 mapping_set_gfp_mask(file->f_mapping,
746 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
748 blk_mq_unfreeze_queue(lo->lo_queue);
749 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
750 mutex_unlock(&loop_ctl_mutex);
752 * We must drop file reference outside of loop_ctl_mutex as dropping
753 * the file ref can take bd_mutex which creates circular locking
758 loop_reread_partitions(lo, bdev);
762 mutex_unlock(&loop_ctl_mutex);
768 /* loop sysfs attributes */
770 static ssize_t loop_attr_show(struct device *dev, char *page,
771 ssize_t (*callback)(struct loop_device *, char *))
773 struct gendisk *disk = dev_to_disk(dev);
774 struct loop_device *lo = disk->private_data;
776 return callback(lo, page);
779 #define LOOP_ATTR_RO(_name) \
780 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
781 static ssize_t loop_attr_do_show_##_name(struct device *d, \
782 struct device_attribute *attr, char *b) \
784 return loop_attr_show(d, b, loop_attr_##_name##_show); \
786 static struct device_attribute loop_attr_##_name = \
787 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
789 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
794 spin_lock_irq(&lo->lo_lock);
795 if (lo->lo_backing_file)
796 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
797 spin_unlock_irq(&lo->lo_lock);
799 if (IS_ERR_OR_NULL(p))
803 memmove(buf, p, ret);
811 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
813 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
816 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
818 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
821 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
823 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
825 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
828 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
830 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
832 return sprintf(buf, "%s\n", partscan ? "1" : "0");
835 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
837 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
839 return sprintf(buf, "%s\n", dio ? "1" : "0");
842 LOOP_ATTR_RO(backing_file);
843 LOOP_ATTR_RO(offset);
844 LOOP_ATTR_RO(sizelimit);
845 LOOP_ATTR_RO(autoclear);
846 LOOP_ATTR_RO(partscan);
849 static struct attribute *loop_attrs[] = {
850 &loop_attr_backing_file.attr,
851 &loop_attr_offset.attr,
852 &loop_attr_sizelimit.attr,
853 &loop_attr_autoclear.attr,
854 &loop_attr_partscan.attr,
859 static struct attribute_group loop_attribute_group = {
864 static void loop_sysfs_init(struct loop_device *lo)
866 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
867 &loop_attribute_group);
870 static void loop_sysfs_exit(struct loop_device *lo)
872 if (lo->sysfs_inited)
873 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
874 &loop_attribute_group);
877 static void loop_config_discard(struct loop_device *lo)
879 struct file *file = lo->lo_backing_file;
880 struct inode *inode = file->f_mapping->host;
881 struct request_queue *q = lo->lo_queue;
882 u32 granularity, max_discard_sectors;
885 * If the backing device is a block device, mirror its zeroing
886 * capability. Set the discard sectors to the block device's zeroing
887 * capabilities because loop discards result in blkdev_issue_zeroout(),
888 * not blkdev_issue_discard(). This maintains consistent behavior with
889 * file-backed loop devices: discarded regions read back as zero.
891 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
892 struct request_queue *backingq;
894 backingq = bdev_get_queue(inode->i_bdev);
896 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
897 granularity = backingq->limits.discard_granularity ?:
898 queue_physical_block_size(backingq);
901 * We use punch hole to reclaim the free space used by the
902 * image a.k.a. discard. However we do not support discard if
903 * encryption is enabled, because it may give an attacker
904 * useful information.
906 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
907 max_discard_sectors = 0;
911 max_discard_sectors = UINT_MAX >> 9;
912 granularity = inode->i_sb->s_blocksize;
915 if (max_discard_sectors) {
916 q->limits.discard_granularity = granularity;
917 blk_queue_max_discard_sectors(q, max_discard_sectors);
918 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
919 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
921 q->limits.discard_granularity = 0;
922 blk_queue_max_discard_sectors(q, 0);
923 blk_queue_max_write_zeroes_sectors(q, 0);
924 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
926 q->limits.discard_alignment = 0;
929 static void loop_unprepare_queue(struct loop_device *lo)
931 kthread_flush_worker(&lo->worker);
932 kthread_stop(lo->worker_task);
935 static int loop_kthread_worker_fn(void *worker_ptr)
937 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
938 return kthread_worker_fn(worker_ptr);
941 static int loop_prepare_queue(struct loop_device *lo)
943 kthread_init_worker(&lo->worker);
944 lo->worker_task = kthread_run(loop_kthread_worker_fn,
945 &lo->worker, "loop%d", lo->lo_number);
946 if (IS_ERR(lo->worker_task))
948 set_user_nice(lo->worker_task, MIN_NICE);
952 static void loop_update_rotational(struct loop_device *lo)
954 struct file *file = lo->lo_backing_file;
955 struct inode *file_inode = file->f_mapping->host;
956 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
957 struct request_queue *q = lo->lo_queue;
960 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
962 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
965 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
967 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
971 loop_release_xfer(struct loop_device *lo)
974 struct loop_func_table *xfer = lo->lo_encryption;
978 err = xfer->release(lo);
980 lo->lo_encryption = NULL;
981 module_put(xfer->owner);
987 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
988 const struct loop_info64 *i)
993 struct module *owner = xfer->owner;
995 if (!try_module_get(owner))
998 err = xfer->init(lo, i);
1002 lo->lo_encryption = xfer;
1008 * loop_set_status_from_info - configure device from loop_info
1009 * @lo: struct loop_device to configure
1010 * @info: struct loop_info64 to configure the device with
1012 * Configures the loop device parameters according to the passed
1013 * in loop_info64 configuration.
1016 loop_set_status_from_info(struct loop_device *lo,
1017 const struct loop_info64 *info)
1020 struct loop_func_table *xfer;
1021 kuid_t uid = current_uid();
1023 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1026 err = loop_release_xfer(lo);
1030 if (info->lo_encrypt_type) {
1031 unsigned int type = info->lo_encrypt_type;
1033 if (type >= MAX_LO_CRYPT)
1035 xfer = xfer_funcs[type];
1041 err = loop_init_xfer(lo, xfer, info);
1045 lo->lo_offset = info->lo_offset;
1046 lo->lo_sizelimit = info->lo_sizelimit;
1047 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1048 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1049 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1050 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1054 lo->transfer = xfer->transfer;
1055 lo->ioctl = xfer->ioctl;
1057 lo->lo_flags = info->lo_flags;
1059 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1060 lo->lo_init[0] = info->lo_init[0];
1061 lo->lo_init[1] = info->lo_init[1];
1062 if (info->lo_encrypt_key_size) {
1063 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1064 info->lo_encrypt_key_size);
1065 lo->lo_key_owner = uid;
1071 static int loop_configure(struct loop_device *lo, fmode_t mode,
1072 struct block_device *bdev,
1073 const struct loop_config *config)
1076 struct inode *inode;
1077 struct address_space *mapping;
1078 struct block_device *claimed_bdev = NULL;
1082 unsigned short bsize;
1084 /* This is safe, since we have a reference from open(). */
1085 __module_get(THIS_MODULE);
1088 file = fget(config->fd);
1093 * If we don't hold exclusive handle for the device, upgrade to it
1094 * here to avoid changing device under exclusive owner.
1096 if (!(mode & FMODE_EXCL)) {
1097 claimed_bdev = bdev->bd_contains;
1098 error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure);
1103 error = mutex_lock_killable(&loop_ctl_mutex);
1108 if (lo->lo_state != Lo_unbound)
1111 error = loop_validate_file(file, bdev);
1115 mapping = file->f_mapping;
1116 inode = mapping->host;
1118 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1123 if (config->block_size) {
1124 error = loop_validate_block_size(config->block_size);
1129 error = loop_set_status_from_info(lo, &config->info);
1133 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1134 !file->f_op->write_iter)
1135 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1137 error = loop_prepare_queue(lo);
1141 set_device_ro(bdev, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1143 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1144 lo->lo_device = bdev;
1145 lo->lo_backing_file = file;
1146 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1147 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1149 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1150 blk_queue_write_cache(lo->lo_queue, true, false);
1152 if (config->block_size)
1153 bsize = config->block_size;
1154 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1155 /* In case of direct I/O, match underlying block size */
1156 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1160 blk_queue_logical_block_size(lo->lo_queue, bsize);
1161 blk_queue_physical_block_size(lo->lo_queue, bsize);
1162 blk_queue_io_min(lo->lo_queue, bsize);
1164 loop_update_rotational(lo);
1165 loop_update_dio(lo);
1166 loop_sysfs_init(lo);
1168 size = get_loop_size(lo, file);
1169 loop_set_size(lo, size);
1171 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1172 block_size(inode->i_bdev) : PAGE_SIZE);
1174 lo->lo_state = Lo_bound;
1176 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1177 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1179 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1181 /* Grab the block_device to prevent its destruction after we
1182 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1185 mutex_unlock(&loop_ctl_mutex);
1187 loop_reread_partitions(lo, bdev);
1189 bd_abort_claiming(bdev, claimed_bdev, loop_configure);
1193 mutex_unlock(&loop_ctl_mutex);
1196 bd_abort_claiming(bdev, claimed_bdev, loop_configure);
1200 /* This is safe: open() is still holding a reference. */
1201 module_put(THIS_MODULE);
1205 static int __loop_clr_fd(struct loop_device *lo, bool release)
1207 struct file *filp = NULL;
1208 gfp_t gfp = lo->old_gfp_mask;
1209 struct block_device *bdev = lo->lo_device;
1211 bool partscan = false;
1214 mutex_lock(&loop_ctl_mutex);
1215 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1220 filp = lo->lo_backing_file;
1226 /* freeze request queue during the transition */
1227 blk_mq_freeze_queue(lo->lo_queue);
1229 spin_lock_irq(&lo->lo_lock);
1230 lo->lo_backing_file = NULL;
1231 spin_unlock_irq(&lo->lo_lock);
1233 loop_release_xfer(lo);
1234 lo->transfer = NULL;
1236 lo->lo_device = NULL;
1237 lo->lo_encryption = NULL;
1239 lo->lo_sizelimit = 0;
1240 lo->lo_encrypt_key_size = 0;
1241 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1242 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1243 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1244 blk_queue_logical_block_size(lo->lo_queue, 512);
1245 blk_queue_physical_block_size(lo->lo_queue, 512);
1246 blk_queue_io_min(lo->lo_queue, 512);
1249 invalidate_bdev(bdev);
1250 bdev->bd_inode->i_mapping->wb_err = 0;
1252 set_capacity(lo->lo_disk, 0);
1253 loop_sysfs_exit(lo);
1255 bd_set_nr_sectors(bdev, 0);
1256 /* let user-space know about this change */
1257 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1259 mapping_set_gfp_mask(filp->f_mapping, gfp);
1260 /* This is safe: open() is still holding a reference. */
1261 module_put(THIS_MODULE);
1262 blk_mq_unfreeze_queue(lo->lo_queue);
1264 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1265 lo_number = lo->lo_number;
1266 loop_unprepare_queue(lo);
1268 mutex_unlock(&loop_ctl_mutex);
1271 * bd_mutex has been held already in release path, so don't
1272 * acquire it if this function is called in such case.
1274 * If the reread partition isn't from release path, lo_refcnt
1275 * must be at least one and it can only become zero when the
1276 * current holder is released.
1279 mutex_lock(&bdev->bd_mutex);
1280 err = bdev_disk_changed(bdev, false);
1282 mutex_unlock(&bdev->bd_mutex);
1284 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1285 __func__, lo_number, err);
1286 /* Device is gone, no point in returning error */
1291 * lo->lo_state is set to Lo_unbound here after above partscan has
1294 * There cannot be anybody else entering __loop_clr_fd() as
1295 * lo->lo_backing_file is already cleared and Lo_rundown state
1296 * protects us from all the other places trying to change the 'lo'
1299 mutex_lock(&loop_ctl_mutex);
1302 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1303 lo->lo_state = Lo_unbound;
1304 mutex_unlock(&loop_ctl_mutex);
1307 * Need not hold loop_ctl_mutex to fput backing file.
1308 * Calling fput holding loop_ctl_mutex triggers a circular
1309 * lock dependency possibility warning as fput can take
1310 * bd_mutex which is usually taken before loop_ctl_mutex.
1317 static int loop_clr_fd(struct loop_device *lo)
1321 err = mutex_lock_killable(&loop_ctl_mutex);
1324 if (lo->lo_state != Lo_bound) {
1325 mutex_unlock(&loop_ctl_mutex);
1329 * If we've explicitly asked to tear down the loop device,
1330 * and it has an elevated reference count, set it for auto-teardown when
1331 * the last reference goes away. This stops $!~#$@ udev from
1332 * preventing teardown because it decided that it needs to run blkid on
1333 * the loopback device whenever they appear. xfstests is notorious for
1334 * failing tests because blkid via udev races with a losetup
1335 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1336 * command to fail with EBUSY.
1338 if (atomic_read(&lo->lo_refcnt) > 1) {
1339 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1340 mutex_unlock(&loop_ctl_mutex);
1343 lo->lo_state = Lo_rundown;
1344 mutex_unlock(&loop_ctl_mutex);
1346 return __loop_clr_fd(lo, false);
1350 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1353 struct block_device *bdev;
1354 kuid_t uid = current_uid();
1356 bool partscan = false;
1357 bool size_changed = false;
1359 err = mutex_lock_killable(&loop_ctl_mutex);
1362 if (lo->lo_encrypt_key_size &&
1363 !uid_eq(lo->lo_key_owner, uid) &&
1364 !capable(CAP_SYS_ADMIN)) {
1368 if (lo->lo_state != Lo_bound) {
1373 if (lo->lo_offset != info->lo_offset ||
1374 lo->lo_sizelimit != info->lo_sizelimit) {
1375 size_changed = true;
1376 sync_blockdev(lo->lo_device);
1377 invalidate_bdev(lo->lo_device);
1380 /* I/O need to be drained during transfer transition */
1381 blk_mq_freeze_queue(lo->lo_queue);
1383 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1384 /* If any pages were dirtied after invalidate_bdev(), try again */
1386 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1387 __func__, lo->lo_number, lo->lo_file_name,
1388 lo->lo_device->bd_inode->i_mapping->nrpages);
1392 prev_lo_flags = lo->lo_flags;
1394 err = loop_set_status_from_info(lo, info);
1398 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1399 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1400 /* For those flags, use the previous values instead */
1401 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1402 /* For flags that can't be cleared, use previous values too */
1403 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1406 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1407 lo->lo_backing_file);
1408 loop_set_size(lo, new_size);
1411 loop_config_discard(lo);
1413 /* update dio if lo_offset or transfer is changed */
1414 __loop_update_dio(lo, lo->use_dio);
1417 blk_mq_unfreeze_queue(lo->lo_queue);
1419 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1420 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1421 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1422 bdev = lo->lo_device;
1426 mutex_unlock(&loop_ctl_mutex);
1428 loop_reread_partitions(lo, bdev);
1434 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1440 ret = mutex_lock_killable(&loop_ctl_mutex);
1443 if (lo->lo_state != Lo_bound) {
1444 mutex_unlock(&loop_ctl_mutex);
1448 memset(info, 0, sizeof(*info));
1449 info->lo_number = lo->lo_number;
1450 info->lo_offset = lo->lo_offset;
1451 info->lo_sizelimit = lo->lo_sizelimit;
1452 info->lo_flags = lo->lo_flags;
1453 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1454 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1455 info->lo_encrypt_type =
1456 lo->lo_encryption ? lo->lo_encryption->number : 0;
1457 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1458 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1459 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1460 lo->lo_encrypt_key_size);
1463 /* Drop loop_ctl_mutex while we call into the filesystem. */
1464 path = lo->lo_backing_file->f_path;
1466 mutex_unlock(&loop_ctl_mutex);
1467 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1469 info->lo_device = huge_encode_dev(stat.dev);
1470 info->lo_inode = stat.ino;
1471 info->lo_rdevice = huge_encode_dev(stat.rdev);
1478 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1480 memset(info64, 0, sizeof(*info64));
1481 info64->lo_number = info->lo_number;
1482 info64->lo_device = info->lo_device;
1483 info64->lo_inode = info->lo_inode;
1484 info64->lo_rdevice = info->lo_rdevice;
1485 info64->lo_offset = info->lo_offset;
1486 info64->lo_sizelimit = 0;
1487 info64->lo_encrypt_type = info->lo_encrypt_type;
1488 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1489 info64->lo_flags = info->lo_flags;
1490 info64->lo_init[0] = info->lo_init[0];
1491 info64->lo_init[1] = info->lo_init[1];
1492 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1493 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1495 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1496 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1500 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1502 memset(info, 0, sizeof(*info));
1503 info->lo_number = info64->lo_number;
1504 info->lo_device = info64->lo_device;
1505 info->lo_inode = info64->lo_inode;
1506 info->lo_rdevice = info64->lo_rdevice;
1507 info->lo_offset = info64->lo_offset;
1508 info->lo_encrypt_type = info64->lo_encrypt_type;
1509 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1510 info->lo_flags = info64->lo_flags;
1511 info->lo_init[0] = info64->lo_init[0];
1512 info->lo_init[1] = info64->lo_init[1];
1513 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1514 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1516 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1517 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1519 /* error in case values were truncated */
1520 if (info->lo_device != info64->lo_device ||
1521 info->lo_rdevice != info64->lo_rdevice ||
1522 info->lo_inode != info64->lo_inode ||
1523 info->lo_offset != info64->lo_offset)
1530 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1532 struct loop_info info;
1533 struct loop_info64 info64;
1535 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1537 loop_info64_from_old(&info, &info64);
1538 return loop_set_status(lo, &info64);
1542 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1544 struct loop_info64 info64;
1546 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1548 return loop_set_status(lo, &info64);
1552 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1553 struct loop_info info;
1554 struct loop_info64 info64;
1559 err = loop_get_status(lo, &info64);
1561 err = loop_info64_to_old(&info64, &info);
1562 if (!err && copy_to_user(arg, &info, sizeof(info)))
1569 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1570 struct loop_info64 info64;
1575 err = loop_get_status(lo, &info64);
1576 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1582 static int loop_set_capacity(struct loop_device *lo)
1586 if (unlikely(lo->lo_state != Lo_bound))
1589 size = get_loop_size(lo, lo->lo_backing_file);
1590 loop_set_size(lo, size);
1595 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1598 if (lo->lo_state != Lo_bound)
1601 __loop_update_dio(lo, !!arg);
1602 if (lo->use_dio == !!arg)
1609 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1613 if (lo->lo_state != Lo_bound)
1616 err = loop_validate_block_size(arg);
1620 if (lo->lo_queue->limits.logical_block_size == arg)
1623 sync_blockdev(lo->lo_device);
1624 invalidate_bdev(lo->lo_device);
1626 blk_mq_freeze_queue(lo->lo_queue);
1628 /* invalidate_bdev should have truncated all the pages */
1629 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1631 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1632 __func__, lo->lo_number, lo->lo_file_name,
1633 lo->lo_device->bd_inode->i_mapping->nrpages);
1637 blk_queue_logical_block_size(lo->lo_queue, arg);
1638 blk_queue_physical_block_size(lo->lo_queue, arg);
1639 blk_queue_io_min(lo->lo_queue, arg);
1640 loop_update_dio(lo);
1642 blk_mq_unfreeze_queue(lo->lo_queue);
1647 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1652 err = mutex_lock_killable(&loop_ctl_mutex);
1656 case LOOP_SET_CAPACITY:
1657 err = loop_set_capacity(lo);
1659 case LOOP_SET_DIRECT_IO:
1660 err = loop_set_dio(lo, arg);
1662 case LOOP_SET_BLOCK_SIZE:
1663 err = loop_set_block_size(lo, arg);
1666 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1668 mutex_unlock(&loop_ctl_mutex);
1672 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1673 unsigned int cmd, unsigned long arg)
1675 struct loop_device *lo = bdev->bd_disk->private_data;
1676 void __user *argp = (void __user *) arg;
1682 * Legacy case - pass in a zeroed out struct loop_config with
1683 * only the file descriptor set , which corresponds with the
1684 * default parameters we'd have used otherwise.
1686 struct loop_config config;
1688 memset(&config, 0, sizeof(config));
1691 return loop_configure(lo, mode, bdev, &config);
1693 case LOOP_CONFIGURE: {
1694 struct loop_config config;
1696 if (copy_from_user(&config, argp, sizeof(config)))
1699 return loop_configure(lo, mode, bdev, &config);
1701 case LOOP_CHANGE_FD:
1702 return loop_change_fd(lo, bdev, arg);
1704 return loop_clr_fd(lo);
1705 case LOOP_SET_STATUS:
1707 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1708 err = loop_set_status_old(lo, argp);
1711 case LOOP_GET_STATUS:
1712 return loop_get_status_old(lo, argp);
1713 case LOOP_SET_STATUS64:
1715 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1716 err = loop_set_status64(lo, argp);
1719 case LOOP_GET_STATUS64:
1720 return loop_get_status64(lo, argp);
1721 case LOOP_SET_CAPACITY:
1722 case LOOP_SET_DIRECT_IO:
1723 case LOOP_SET_BLOCK_SIZE:
1724 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1728 err = lo_simple_ioctl(lo, cmd, arg);
1735 #ifdef CONFIG_COMPAT
1736 struct compat_loop_info {
1737 compat_int_t lo_number; /* ioctl r/o */
1738 compat_dev_t lo_device; /* ioctl r/o */
1739 compat_ulong_t lo_inode; /* ioctl r/o */
1740 compat_dev_t lo_rdevice; /* ioctl r/o */
1741 compat_int_t lo_offset;
1742 compat_int_t lo_encrypt_type;
1743 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1744 compat_int_t lo_flags; /* ioctl r/o */
1745 char lo_name[LO_NAME_SIZE];
1746 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1747 compat_ulong_t lo_init[2];
1752 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1753 * - noinlined to reduce stack space usage in main part of driver
1756 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1757 struct loop_info64 *info64)
1759 struct compat_loop_info info;
1761 if (copy_from_user(&info, arg, sizeof(info)))
1764 memset(info64, 0, sizeof(*info64));
1765 info64->lo_number = info.lo_number;
1766 info64->lo_device = info.lo_device;
1767 info64->lo_inode = info.lo_inode;
1768 info64->lo_rdevice = info.lo_rdevice;
1769 info64->lo_offset = info.lo_offset;
1770 info64->lo_sizelimit = 0;
1771 info64->lo_encrypt_type = info.lo_encrypt_type;
1772 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1773 info64->lo_flags = info.lo_flags;
1774 info64->lo_init[0] = info.lo_init[0];
1775 info64->lo_init[1] = info.lo_init[1];
1776 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1777 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1779 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1780 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1785 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1786 * - noinlined to reduce stack space usage in main part of driver
1789 loop_info64_to_compat(const struct loop_info64 *info64,
1790 struct compat_loop_info __user *arg)
1792 struct compat_loop_info info;
1794 memset(&info, 0, sizeof(info));
1795 info.lo_number = info64->lo_number;
1796 info.lo_device = info64->lo_device;
1797 info.lo_inode = info64->lo_inode;
1798 info.lo_rdevice = info64->lo_rdevice;
1799 info.lo_offset = info64->lo_offset;
1800 info.lo_encrypt_type = info64->lo_encrypt_type;
1801 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1802 info.lo_flags = info64->lo_flags;
1803 info.lo_init[0] = info64->lo_init[0];
1804 info.lo_init[1] = info64->lo_init[1];
1805 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1806 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1808 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1809 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1811 /* error in case values were truncated */
1812 if (info.lo_device != info64->lo_device ||
1813 info.lo_rdevice != info64->lo_rdevice ||
1814 info.lo_inode != info64->lo_inode ||
1815 info.lo_offset != info64->lo_offset ||
1816 info.lo_init[0] != info64->lo_init[0] ||
1817 info.lo_init[1] != info64->lo_init[1])
1820 if (copy_to_user(arg, &info, sizeof(info)))
1826 loop_set_status_compat(struct loop_device *lo,
1827 const struct compat_loop_info __user *arg)
1829 struct loop_info64 info64;
1832 ret = loop_info64_from_compat(arg, &info64);
1835 return loop_set_status(lo, &info64);
1839 loop_get_status_compat(struct loop_device *lo,
1840 struct compat_loop_info __user *arg)
1842 struct loop_info64 info64;
1847 err = loop_get_status(lo, &info64);
1849 err = loop_info64_to_compat(&info64, arg);
1853 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1854 unsigned int cmd, unsigned long arg)
1856 struct loop_device *lo = bdev->bd_disk->private_data;
1860 case LOOP_SET_STATUS:
1861 err = loop_set_status_compat(lo,
1862 (const struct compat_loop_info __user *)arg);
1864 case LOOP_GET_STATUS:
1865 err = loop_get_status_compat(lo,
1866 (struct compat_loop_info __user *)arg);
1868 case LOOP_SET_CAPACITY:
1870 case LOOP_GET_STATUS64:
1871 case LOOP_SET_STATUS64:
1872 case LOOP_CONFIGURE:
1873 arg = (unsigned long) compat_ptr(arg);
1876 case LOOP_CHANGE_FD:
1877 case LOOP_SET_BLOCK_SIZE:
1878 case LOOP_SET_DIRECT_IO:
1879 err = lo_ioctl(bdev, mode, cmd, arg);
1889 static int lo_open(struct block_device *bdev, fmode_t mode)
1891 struct loop_device *lo;
1894 err = mutex_lock_killable(&loop_ctl_mutex);
1897 lo = bdev->bd_disk->private_data;
1903 atomic_inc(&lo->lo_refcnt);
1905 mutex_unlock(&loop_ctl_mutex);
1909 static void lo_release(struct gendisk *disk, fmode_t mode)
1911 struct loop_device *lo;
1913 mutex_lock(&loop_ctl_mutex);
1914 lo = disk->private_data;
1915 if (atomic_dec_return(&lo->lo_refcnt))
1918 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1919 if (lo->lo_state != Lo_bound)
1921 lo->lo_state = Lo_rundown;
1922 mutex_unlock(&loop_ctl_mutex);
1924 * In autoclear mode, stop the loop thread
1925 * and remove configuration after last close.
1927 __loop_clr_fd(lo, true);
1929 } else if (lo->lo_state == Lo_bound) {
1931 * Otherwise keep thread (if running) and config,
1932 * but flush possible ongoing bios in thread.
1934 blk_mq_freeze_queue(lo->lo_queue);
1935 blk_mq_unfreeze_queue(lo->lo_queue);
1939 mutex_unlock(&loop_ctl_mutex);
1942 static const struct block_device_operations lo_fops = {
1943 .owner = THIS_MODULE,
1945 .release = lo_release,
1947 #ifdef CONFIG_COMPAT
1948 .compat_ioctl = lo_compat_ioctl,
1953 * And now the modules code and kernel interface.
1955 static int max_loop;
1956 module_param(max_loop, int, 0444);
1957 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1958 module_param(max_part, int, 0444);
1959 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1960 MODULE_LICENSE("GPL");
1961 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1963 int loop_register_transfer(struct loop_func_table *funcs)
1965 unsigned int n = funcs->number;
1967 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1969 xfer_funcs[n] = funcs;
1973 static int unregister_transfer_cb(int id, void *ptr, void *data)
1975 struct loop_device *lo = ptr;
1976 struct loop_func_table *xfer = data;
1978 mutex_lock(&loop_ctl_mutex);
1979 if (lo->lo_encryption == xfer)
1980 loop_release_xfer(lo);
1981 mutex_unlock(&loop_ctl_mutex);
1985 int loop_unregister_transfer(int number)
1987 unsigned int n = number;
1988 struct loop_func_table *xfer;
1990 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1993 xfer_funcs[n] = NULL;
1994 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1998 EXPORT_SYMBOL(loop_register_transfer);
1999 EXPORT_SYMBOL(loop_unregister_transfer);
2001 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
2002 const struct blk_mq_queue_data *bd)
2004 struct request *rq = bd->rq;
2005 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2006 struct loop_device *lo = rq->q->queuedata;
2008 blk_mq_start_request(rq);
2010 if (lo->lo_state != Lo_bound)
2011 return BLK_STS_IOERR;
2013 switch (req_op(rq)) {
2015 case REQ_OP_DISCARD:
2016 case REQ_OP_WRITE_ZEROES:
2017 cmd->use_aio = false;
2020 cmd->use_aio = lo->use_dio;
2024 /* always use the first bio's css */
2025 #ifdef CONFIG_BLK_CGROUP
2026 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2027 cmd->css = &bio_blkcg(rq->bio)->css;
2032 kthread_queue_work(&lo->worker, &cmd->work);
2037 static void loop_handle_cmd(struct loop_cmd *cmd)
2039 struct request *rq = blk_mq_rq_from_pdu(cmd);
2040 const bool write = op_is_write(req_op(rq));
2041 struct loop_device *lo = rq->q->queuedata;
2044 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2049 ret = do_req_filebacked(lo, rq);
2051 /* complete non-aio request */
2052 if (!cmd->use_aio || ret) {
2053 if (ret == -EOPNOTSUPP)
2056 cmd->ret = ret ? -EIO : 0;
2057 if (likely(!blk_should_fake_timeout(rq->q)))
2058 blk_mq_complete_request(rq);
2062 static void loop_queue_work(struct kthread_work *work)
2064 struct loop_cmd *cmd =
2065 container_of(work, struct loop_cmd, work);
2067 loop_handle_cmd(cmd);
2070 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2071 unsigned int hctx_idx, unsigned int numa_node)
2073 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2075 kthread_init_work(&cmd->work, loop_queue_work);
2079 static const struct blk_mq_ops loop_mq_ops = {
2080 .queue_rq = loop_queue_rq,
2081 .init_request = loop_init_request,
2082 .complete = lo_complete_rq,
2085 static int loop_add(struct loop_device **l, int i)
2087 struct loop_device *lo;
2088 struct gendisk *disk;
2092 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2096 lo->lo_state = Lo_unbound;
2098 /* allocate id, if @id >= 0, we're requesting that specific id */
2100 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2104 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2111 lo->tag_set.ops = &loop_mq_ops;
2112 lo->tag_set.nr_hw_queues = 1;
2113 lo->tag_set.queue_depth = 128;
2114 lo->tag_set.numa_node = NUMA_NO_NODE;
2115 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2116 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
2117 lo->tag_set.driver_data = lo;
2119 err = blk_mq_alloc_tag_set(&lo->tag_set);
2123 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2124 if (IS_ERR(lo->lo_queue)) {
2125 err = PTR_ERR(lo->lo_queue);
2126 goto out_cleanup_tags;
2128 lo->lo_queue->queuedata = lo;
2130 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2133 * By default, we do buffer IO, so it doesn't make sense to enable
2134 * merge because the I/O submitted to backing file is handled page by
2135 * page. For directio mode, merge does help to dispatch bigger request
2136 * to underlayer disk. We will enable merge once directio is enabled.
2138 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2141 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2143 goto out_free_queue;
2146 * Disable partition scanning by default. The in-kernel partition
2147 * scanning can be requested individually per-device during its
2148 * setup. Userspace can always add and remove partitions from all
2149 * devices. The needed partition minors are allocated from the
2150 * extended minor space, the main loop device numbers will continue
2151 * to match the loop minors, regardless of the number of partitions
2154 * If max_part is given, partition scanning is globally enabled for
2155 * all loop devices. The minors for the main loop devices will be
2156 * multiples of max_part.
2158 * Note: Global-for-all-devices, set-only-at-init, read-only module
2159 * parameteters like 'max_loop' and 'max_part' make things needlessly
2160 * complicated, are too static, inflexible and may surprise
2161 * userspace tools. Parameters like this in general should be avoided.
2164 disk->flags |= GENHD_FL_NO_PART_SCAN;
2165 disk->flags |= GENHD_FL_EXT_DEVT;
2166 atomic_set(&lo->lo_refcnt, 0);
2168 spin_lock_init(&lo->lo_lock);
2169 disk->major = LOOP_MAJOR;
2170 disk->first_minor = i << part_shift;
2171 disk->fops = &lo_fops;
2172 disk->private_data = lo;
2173 disk->queue = lo->lo_queue;
2174 sprintf(disk->disk_name, "loop%d", i);
2177 return lo->lo_number;
2180 blk_cleanup_queue(lo->lo_queue);
2182 blk_mq_free_tag_set(&lo->tag_set);
2184 idr_remove(&loop_index_idr, i);
2191 static void loop_remove(struct loop_device *lo)
2193 del_gendisk(lo->lo_disk);
2194 blk_cleanup_queue(lo->lo_queue);
2195 blk_mq_free_tag_set(&lo->tag_set);
2196 put_disk(lo->lo_disk);
2200 static int find_free_cb(int id, void *ptr, void *data)
2202 struct loop_device *lo = ptr;
2203 struct loop_device **l = data;
2205 if (lo->lo_state == Lo_unbound) {
2212 static int loop_lookup(struct loop_device **l, int i)
2214 struct loop_device *lo;
2220 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2223 ret = lo->lo_number;
2228 /* lookup and return a specific i */
2229 lo = idr_find(&loop_index_idr, i);
2232 ret = lo->lo_number;
2238 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2240 struct loop_device *lo;
2241 struct kobject *kobj;
2244 mutex_lock(&loop_ctl_mutex);
2245 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2247 err = loop_add(&lo, MINOR(dev) >> part_shift);
2251 kobj = get_disk_and_module(lo->lo_disk);
2252 mutex_unlock(&loop_ctl_mutex);
2258 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2261 struct loop_device *lo;
2264 ret = mutex_lock_killable(&loop_ctl_mutex);
2271 ret = loop_lookup(&lo, parm);
2276 ret = loop_add(&lo, parm);
2278 case LOOP_CTL_REMOVE:
2279 ret = loop_lookup(&lo, parm);
2282 if (lo->lo_state != Lo_unbound) {
2286 if (atomic_read(&lo->lo_refcnt) > 0) {
2290 lo->lo_disk->private_data = NULL;
2291 idr_remove(&loop_index_idr, lo->lo_number);
2294 case LOOP_CTL_GET_FREE:
2295 ret = loop_lookup(&lo, -1);
2298 ret = loop_add(&lo, -1);
2300 mutex_unlock(&loop_ctl_mutex);
2305 static const struct file_operations loop_ctl_fops = {
2306 .open = nonseekable_open,
2307 .unlocked_ioctl = loop_control_ioctl,
2308 .compat_ioctl = loop_control_ioctl,
2309 .owner = THIS_MODULE,
2310 .llseek = noop_llseek,
2313 static struct miscdevice loop_misc = {
2314 .minor = LOOP_CTRL_MINOR,
2315 .name = "loop-control",
2316 .fops = &loop_ctl_fops,
2319 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2320 MODULE_ALIAS("devname:loop-control");
2322 static int __init loop_init(void)
2325 unsigned long range;
2326 struct loop_device *lo;
2331 part_shift = fls(max_part);
2334 * Adjust max_part according to part_shift as it is exported
2335 * to user space so that user can decide correct minor number
2336 * if [s]he want to create more devices.
2338 * Note that -1 is required because partition 0 is reserved
2339 * for the whole disk.
2341 max_part = (1UL << part_shift) - 1;
2344 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2349 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2355 * If max_loop is specified, create that many devices upfront.
2356 * This also becomes a hard limit. If max_loop is not specified,
2357 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2358 * init time. Loop devices can be requested on-demand with the
2359 * /dev/loop-control interface, or be instantiated by accessing
2360 * a 'dead' device node.
2364 range = max_loop << part_shift;
2366 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2367 range = 1UL << MINORBITS;
2370 err = misc_register(&loop_misc);
2375 if (register_blkdev(LOOP_MAJOR, "loop")) {
2380 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2381 THIS_MODULE, loop_probe, NULL, NULL);
2383 /* pre-create number of devices given by config or max_loop */
2384 mutex_lock(&loop_ctl_mutex);
2385 for (i = 0; i < nr; i++)
2387 mutex_unlock(&loop_ctl_mutex);
2389 printk(KERN_INFO "loop: module loaded\n");
2393 misc_deregister(&loop_misc);
2398 static int loop_exit_cb(int id, void *ptr, void *data)
2400 struct loop_device *lo = ptr;
2406 static void __exit loop_exit(void)
2408 unsigned long range;
2410 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2412 mutex_lock(&loop_ctl_mutex);
2414 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2415 idr_destroy(&loop_index_idr);
2417 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2418 unregister_blkdev(LOOP_MAJOR, "loop");
2420 misc_deregister(&loop_misc);
2422 mutex_unlock(&loop_ctl_mutex);
2425 module_init(loop_init);
2426 module_exit(loop_exit);
2429 static int __init max_loop_setup(char *str)
2431 max_loop = simple_strtol(str, NULL, 0);
2435 __setup("max_loop=", max_loop_setup);