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/pagemap.h>
57 #include <linux/file.h>
58 #include <linux/stat.h>
59 #include <linux/errno.h>
60 #include <linux/major.h>
61 #include <linux/wait.h>
62 #include <linux/blkdev.h>
63 #include <linux/blkpg.h>
64 #include <linux/init.h>
65 #include <linux/swap.h>
66 #include <linux/slab.h>
67 #include <linux/compat.h>
68 #include <linux/suspend.h>
69 #include <linux/freezer.h>
70 #include <linux/mutex.h>
71 #include <linux/writeback.h>
72 #include <linux/completion.h>
73 #include <linux/highmem.h>
74 #include <linux/kthread.h>
75 #include <linux/splice.h>
76 #include <linux/sysfs.h>
77 #include <linux/miscdevice.h>
78 #include <linux/falloc.h>
79 #include <linux/uio.h>
80 #include <linux/ioprio.h>
81 #include <linux/blk-cgroup.h>
85 #include <linux/uaccess.h>
87 static DEFINE_IDR(loop_index_idr);
88 static DEFINE_MUTEX(loop_ctl_mutex);
91 static int part_shift;
93 static int transfer_xor(struct loop_device *lo, int cmd,
94 struct page *raw_page, unsigned raw_off,
95 struct page *loop_page, unsigned loop_off,
96 int size, sector_t real_block)
98 char *raw_buf = kmap_atomic(raw_page) + raw_off;
99 char *loop_buf = kmap_atomic(loop_page) + loop_off;
100 char *in, *out, *key;
111 key = lo->lo_encrypt_key;
112 keysize = lo->lo_encrypt_key_size;
113 for (i = 0; i < size; i++)
114 *out++ = *in++ ^ key[(i & 511) % keysize];
116 kunmap_atomic(loop_buf);
117 kunmap_atomic(raw_buf);
122 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
124 if (unlikely(info->lo_encrypt_key_size <= 0))
129 static struct loop_func_table none_funcs = {
130 .number = LO_CRYPT_NONE,
133 static struct loop_func_table xor_funcs = {
134 .number = LO_CRYPT_XOR,
135 .transfer = transfer_xor,
139 /* xfer_funcs[0] is special - its release function is never called */
140 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
145 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
149 /* Compute loopsize in bytes */
150 loopsize = i_size_read(file->f_mapping->host);
153 /* offset is beyond i_size, weird but possible */
157 if (sizelimit > 0 && sizelimit < loopsize)
158 loopsize = sizelimit;
160 * Unfortunately, if we want to do I/O on the device,
161 * the number of 512-byte sectors has to fit into a sector_t.
163 return loopsize >> 9;
166 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
168 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
171 static void __loop_update_dio(struct loop_device *lo, bool dio)
173 struct file *file = lo->lo_backing_file;
174 struct address_space *mapping = file->f_mapping;
175 struct inode *inode = mapping->host;
176 unsigned short sb_bsize = 0;
177 unsigned dio_align = 0;
180 if (inode->i_sb->s_bdev) {
181 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
182 dio_align = sb_bsize - 1;
186 * We support direct I/O only if lo_offset is aligned with the
187 * logical I/O size of backing device, and the logical block
188 * size of loop is bigger than the backing device's and the loop
189 * needn't transform transfer.
191 * TODO: the above condition may be loosed in the future, and
192 * direct I/O may be switched runtime at that time because most
193 * of requests in sane applications should be PAGE_SIZE aligned
196 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
197 !(lo->lo_offset & dio_align) &&
198 mapping->a_ops->direct_IO &&
207 if (lo->use_dio == use_dio)
210 /* flush dirty pages before changing direct IO */
214 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
215 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
216 * will get updated by ioctl(LOOP_GET_STATUS)
218 if (lo->lo_state == Lo_bound)
219 blk_mq_freeze_queue(lo->lo_queue);
220 lo->use_dio = use_dio;
222 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
223 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
225 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
226 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
228 if (lo->lo_state == Lo_bound)
229 blk_mq_unfreeze_queue(lo->lo_queue);
233 * loop_validate_block_size() - validates the passed in block size
234 * @bsize: size to validate
237 loop_validate_block_size(unsigned short bsize)
239 if (bsize < 512 || bsize > PAGE_SIZE || !is_power_of_2(bsize))
246 * loop_set_size() - sets device size and notifies userspace
247 * @lo: struct loop_device to set the size for
248 * @size: new size of the loop device
250 * Callers must validate that the size passed into this function fits into
251 * a sector_t, eg using loop_validate_size()
253 static void loop_set_size(struct loop_device *lo, loff_t size)
255 if (!set_capacity_and_notify(lo->lo_disk, size))
256 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
260 lo_do_transfer(struct loop_device *lo, int cmd,
261 struct page *rpage, unsigned roffs,
262 struct page *lpage, unsigned loffs,
263 int size, sector_t rblock)
267 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
271 printk_ratelimited(KERN_ERR
272 "loop: Transfer error at byte offset %llu, length %i.\n",
273 (unsigned long long)rblock << 9, size);
277 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
282 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
284 file_start_write(file);
285 bw = vfs_iter_write(file, &i, ppos, 0);
286 file_end_write(file);
288 if (likely(bw == bvec->bv_len))
291 printk_ratelimited(KERN_ERR
292 "loop: Write error at byte offset %llu, length %i.\n",
293 (unsigned long long)*ppos, bvec->bv_len);
299 static int lo_write_simple(struct loop_device *lo, struct request *rq,
303 struct req_iterator iter;
306 rq_for_each_segment(bvec, rq, iter) {
307 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
317 * This is the slow, transforming version that needs to double buffer the
318 * data as it cannot do the transformations in place without having direct
319 * access to the destination pages of the backing file.
321 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
324 struct bio_vec bvec, b;
325 struct req_iterator iter;
329 page = alloc_page(GFP_NOIO);
333 rq_for_each_segment(bvec, rq, iter) {
334 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
335 bvec.bv_offset, bvec.bv_len, pos >> 9);
341 b.bv_len = bvec.bv_len;
342 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
351 static int lo_read_simple(struct loop_device *lo, struct request *rq,
355 struct req_iterator iter;
359 rq_for_each_segment(bvec, rq, iter) {
360 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
361 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
365 flush_dcache_page(bvec.bv_page);
367 if (len != bvec.bv_len) {
370 __rq_for_each_bio(bio, rq)
380 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
383 struct bio_vec bvec, b;
384 struct req_iterator iter;
390 page = alloc_page(GFP_NOIO);
394 rq_for_each_segment(bvec, rq, iter) {
399 b.bv_len = bvec.bv_len;
401 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
402 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
408 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
409 bvec.bv_offset, len, offset >> 9);
413 flush_dcache_page(bvec.bv_page);
415 if (len != bvec.bv_len) {
418 __rq_for_each_bio(bio, rq)
430 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
434 * We use fallocate to manipulate the space mappings used by the image
435 * a.k.a. discard/zerorange. However we do not support this if
436 * encryption is enabled, because it may give an attacker useful
439 struct file *file = lo->lo_backing_file;
440 struct request_queue *q = lo->lo_queue;
443 mode |= FALLOC_FL_KEEP_SIZE;
445 if (!blk_queue_discard(q)) {
450 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
451 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
457 static int lo_req_flush(struct loop_device *lo, struct request *rq)
459 struct file *file = lo->lo_backing_file;
460 int ret = vfs_fsync(file, 0);
461 if (unlikely(ret && ret != -EINVAL))
467 static void lo_complete_rq(struct request *rq)
469 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
470 blk_status_t ret = BLK_STS_OK;
472 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
473 req_op(rq) != REQ_OP_READ) {
475 ret = errno_to_blk_status(cmd->ret);
480 * Short READ - if we got some data, advance our request and
481 * retry it. If we got no data, end the rest with EIO.
484 blk_update_request(rq, BLK_STS_OK, cmd->ret);
486 blk_mq_requeue_request(rq, true);
489 struct bio *bio = rq->bio;
498 blk_mq_end_request(rq, ret);
502 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
504 struct request *rq = blk_mq_rq_from_pdu(cmd);
506 if (!atomic_dec_and_test(&cmd->ref))
510 if (likely(!blk_should_fake_timeout(rq->q)))
511 blk_mq_complete_request(rq);
514 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
516 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
521 lo_rw_aio_do_completion(cmd);
524 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
527 struct iov_iter iter;
528 struct req_iterator rq_iter;
529 struct bio_vec *bvec;
530 struct request *rq = blk_mq_rq_from_pdu(cmd);
531 struct bio *bio = rq->bio;
532 struct file *file = lo->lo_backing_file;
538 rq_for_each_bvec(tmp, rq, rq_iter)
541 if (rq->bio != rq->biotail) {
543 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
550 * The bios of the request may be started from the middle of
551 * the 'bvec' because of bio splitting, so we can't directly
552 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
553 * API will take care of all details for us.
555 rq_for_each_bvec(tmp, rq, rq_iter) {
563 * Same here, this bio may be started from the middle of the
564 * 'bvec' because of bio splitting, so offset from the bvec
565 * must be passed to iov iterator
567 offset = bio->bi_iter.bi_bvec_done;
568 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
570 atomic_set(&cmd->ref, 2);
572 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
573 iter.iov_offset = offset;
575 cmd->iocb.ki_pos = pos;
576 cmd->iocb.ki_filp = file;
577 cmd->iocb.ki_complete = lo_rw_aio_complete;
578 cmd->iocb.ki_flags = IOCB_DIRECT;
579 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
581 kthread_associate_blkcg(cmd->css);
584 ret = call_write_iter(file, &cmd->iocb, &iter);
586 ret = call_read_iter(file, &cmd->iocb, &iter);
588 lo_rw_aio_do_completion(cmd);
589 kthread_associate_blkcg(NULL);
591 if (ret != -EIOCBQUEUED)
592 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
596 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
598 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
599 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
602 * lo_write_simple and lo_read_simple should have been covered
603 * by io submit style function like lo_rw_aio(), one blocker
604 * is that lo_read_simple() need to call flush_dcache_page after
605 * the page is written from kernel, and it isn't easy to handle
606 * this in io submit style function which submits all segments
607 * of the req at one time. And direct read IO doesn't need to
608 * run flush_dcache_page().
610 switch (req_op(rq)) {
612 return lo_req_flush(lo, rq);
613 case REQ_OP_WRITE_ZEROES:
615 * If the caller doesn't want deallocation, call zeroout to
616 * write zeroes the range. Otherwise, punch them out.
618 return lo_fallocate(lo, rq, pos,
619 (rq->cmd_flags & REQ_NOUNMAP) ?
620 FALLOC_FL_ZERO_RANGE :
621 FALLOC_FL_PUNCH_HOLE);
623 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
626 return lo_write_transfer(lo, rq, pos);
627 else if (cmd->use_aio)
628 return lo_rw_aio(lo, cmd, pos, WRITE);
630 return lo_write_simple(lo, rq, pos);
633 return lo_read_transfer(lo, rq, pos);
634 else if (cmd->use_aio)
635 return lo_rw_aio(lo, cmd, pos, READ);
637 return lo_read_simple(lo, rq, pos);
644 static inline void loop_update_dio(struct loop_device *lo)
646 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
650 static void loop_reread_partitions(struct loop_device *lo,
651 struct block_device *bdev)
655 mutex_lock(&bdev->bd_mutex);
656 rc = bdev_disk_changed(bdev, false);
657 mutex_unlock(&bdev->bd_mutex);
659 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
660 __func__, lo->lo_number, lo->lo_file_name, rc);
663 static inline int is_loop_device(struct file *file)
665 struct inode *i = file->f_mapping->host;
667 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
670 static int loop_validate_file(struct file *file, struct block_device *bdev)
672 struct inode *inode = file->f_mapping->host;
673 struct file *f = file;
675 /* Avoid recursion */
676 while (is_loop_device(f)) {
677 struct loop_device *l;
679 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
682 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
683 if (l->lo_state != Lo_bound) {
686 f = l->lo_backing_file;
688 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
694 * loop_change_fd switched the backing store of a loopback device to
695 * a new file. This is useful for operating system installers to free up
696 * the original file and in High Availability environments to switch to
697 * an alternative location for the content in case of server meltdown.
698 * This can only work if the loop device is used read-only, and if the
699 * new backing store is the same size and type as the old backing store.
701 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
704 struct file *file = NULL, *old_file;
708 error = mutex_lock_killable(&lo->lo_mutex);
712 if (lo->lo_state != Lo_bound)
715 /* the loop device has to be read-only */
717 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
725 error = loop_validate_file(file, bdev);
729 old_file = lo->lo_backing_file;
733 /* size of the new backing store needs to be the same */
734 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
738 blk_mq_freeze_queue(lo->lo_queue);
739 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
740 lo->lo_backing_file = file;
741 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
742 mapping_set_gfp_mask(file->f_mapping,
743 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
745 blk_mq_unfreeze_queue(lo->lo_queue);
746 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
747 mutex_unlock(&lo->lo_mutex);
749 * We must drop file reference outside of lo_mutex as dropping
750 * the file ref can take bd_mutex which creates circular locking
755 loop_reread_partitions(lo, bdev);
759 mutex_unlock(&lo->lo_mutex);
765 /* loop sysfs attributes */
767 static ssize_t loop_attr_show(struct device *dev, char *page,
768 ssize_t (*callback)(struct loop_device *, char *))
770 struct gendisk *disk = dev_to_disk(dev);
771 struct loop_device *lo = disk->private_data;
773 return callback(lo, page);
776 #define LOOP_ATTR_RO(_name) \
777 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
778 static ssize_t loop_attr_do_show_##_name(struct device *d, \
779 struct device_attribute *attr, char *b) \
781 return loop_attr_show(d, b, loop_attr_##_name##_show); \
783 static struct device_attribute loop_attr_##_name = \
784 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
786 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
791 spin_lock_irq(&lo->lo_lock);
792 if (lo->lo_backing_file)
793 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
794 spin_unlock_irq(&lo->lo_lock);
796 if (IS_ERR_OR_NULL(p))
800 memmove(buf, p, ret);
808 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
810 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
813 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
815 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
818 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
820 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
822 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
825 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
827 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
829 return sprintf(buf, "%s\n", partscan ? "1" : "0");
832 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
834 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
836 return sprintf(buf, "%s\n", dio ? "1" : "0");
839 LOOP_ATTR_RO(backing_file);
840 LOOP_ATTR_RO(offset);
841 LOOP_ATTR_RO(sizelimit);
842 LOOP_ATTR_RO(autoclear);
843 LOOP_ATTR_RO(partscan);
846 static struct attribute *loop_attrs[] = {
847 &loop_attr_backing_file.attr,
848 &loop_attr_offset.attr,
849 &loop_attr_sizelimit.attr,
850 &loop_attr_autoclear.attr,
851 &loop_attr_partscan.attr,
856 static struct attribute_group loop_attribute_group = {
861 static void loop_sysfs_init(struct loop_device *lo)
863 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
864 &loop_attribute_group);
867 static void loop_sysfs_exit(struct loop_device *lo)
869 if (lo->sysfs_inited)
870 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
871 &loop_attribute_group);
874 static void loop_config_discard(struct loop_device *lo)
876 struct file *file = lo->lo_backing_file;
877 struct inode *inode = file->f_mapping->host;
878 struct request_queue *q = lo->lo_queue;
879 u32 granularity, max_discard_sectors;
882 * If the backing device is a block device, mirror its zeroing
883 * capability. Set the discard sectors to the block device's zeroing
884 * capabilities because loop discards result in blkdev_issue_zeroout(),
885 * not blkdev_issue_discard(). This maintains consistent behavior with
886 * file-backed loop devices: discarded regions read back as zero.
888 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
889 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
891 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
892 granularity = backingq->limits.discard_granularity ?:
893 queue_physical_block_size(backingq);
896 * We use punch hole to reclaim the free space used by the
897 * image a.k.a. discard. However we do not support discard if
898 * encryption is enabled, because it may give an attacker
899 * useful information.
901 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
902 max_discard_sectors = 0;
906 max_discard_sectors = UINT_MAX >> 9;
907 granularity = inode->i_sb->s_blocksize;
910 if (max_discard_sectors) {
911 q->limits.discard_granularity = granularity;
912 blk_queue_max_discard_sectors(q, max_discard_sectors);
913 blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
914 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
916 q->limits.discard_granularity = 0;
917 blk_queue_max_discard_sectors(q, 0);
918 blk_queue_max_write_zeroes_sectors(q, 0);
919 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
921 q->limits.discard_alignment = 0;
924 static void loop_unprepare_queue(struct loop_device *lo)
926 kthread_flush_worker(&lo->worker);
927 kthread_stop(lo->worker_task);
930 static int loop_kthread_worker_fn(void *worker_ptr)
932 current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
933 return kthread_worker_fn(worker_ptr);
936 static int loop_prepare_queue(struct loop_device *lo)
938 kthread_init_worker(&lo->worker);
939 lo->worker_task = kthread_run(loop_kthread_worker_fn,
940 &lo->worker, "loop%d", lo->lo_number);
941 if (IS_ERR(lo->worker_task))
943 set_user_nice(lo->worker_task, MIN_NICE);
947 static void loop_update_rotational(struct loop_device *lo)
949 struct file *file = lo->lo_backing_file;
950 struct inode *file_inode = file->f_mapping->host;
951 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
952 struct request_queue *q = lo->lo_queue;
955 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
957 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
960 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
962 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
966 loop_release_xfer(struct loop_device *lo)
969 struct loop_func_table *xfer = lo->lo_encryption;
973 err = xfer->release(lo);
975 lo->lo_encryption = NULL;
976 module_put(xfer->owner);
982 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
983 const struct loop_info64 *i)
988 struct module *owner = xfer->owner;
990 if (!try_module_get(owner))
993 err = xfer->init(lo, i);
997 lo->lo_encryption = xfer;
1003 * loop_set_status_from_info - configure device from loop_info
1004 * @lo: struct loop_device to configure
1005 * @info: struct loop_info64 to configure the device with
1007 * Configures the loop device parameters according to the passed
1008 * in loop_info64 configuration.
1011 loop_set_status_from_info(struct loop_device *lo,
1012 const struct loop_info64 *info)
1015 struct loop_func_table *xfer;
1016 kuid_t uid = current_uid();
1018 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1021 err = loop_release_xfer(lo);
1025 if (info->lo_encrypt_type) {
1026 unsigned int type = info->lo_encrypt_type;
1028 if (type >= MAX_LO_CRYPT)
1030 xfer = xfer_funcs[type];
1036 err = loop_init_xfer(lo, xfer, info);
1040 lo->lo_offset = info->lo_offset;
1041 lo->lo_sizelimit = info->lo_sizelimit;
1042 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1043 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1044 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1045 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1049 lo->transfer = xfer->transfer;
1050 lo->ioctl = xfer->ioctl;
1052 lo->lo_flags = info->lo_flags;
1054 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1055 lo->lo_init[0] = info->lo_init[0];
1056 lo->lo_init[1] = info->lo_init[1];
1057 if (info->lo_encrypt_key_size) {
1058 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1059 info->lo_encrypt_key_size);
1060 lo->lo_key_owner = uid;
1066 static int loop_configure(struct loop_device *lo, fmode_t mode,
1067 struct block_device *bdev,
1068 const struct loop_config *config)
1071 struct inode *inode;
1072 struct address_space *mapping;
1076 unsigned short bsize;
1078 /* This is safe, since we have a reference from open(). */
1079 __module_get(THIS_MODULE);
1082 file = fget(config->fd);
1087 * If we don't hold exclusive handle for the device, upgrade to it
1088 * here to avoid changing device under exclusive owner.
1090 if (!(mode & FMODE_EXCL)) {
1091 error = bd_prepare_to_claim(bdev, loop_configure);
1096 error = mutex_lock_killable(&lo->lo_mutex);
1101 if (lo->lo_state != Lo_unbound)
1104 error = loop_validate_file(file, bdev);
1108 mapping = file->f_mapping;
1109 inode = mapping->host;
1111 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1116 if (config->block_size) {
1117 error = loop_validate_block_size(config->block_size);
1122 error = loop_set_status_from_info(lo, &config->info);
1126 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1127 !file->f_op->write_iter)
1128 lo->lo_flags |= LO_FLAGS_READ_ONLY;
1130 error = loop_prepare_queue(lo);
1134 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1136 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1137 lo->lo_device = bdev;
1138 lo->lo_backing_file = file;
1139 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1140 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1142 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1143 blk_queue_write_cache(lo->lo_queue, true, false);
1145 if (config->block_size)
1146 bsize = config->block_size;
1147 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1148 /* In case of direct I/O, match underlying block size */
1149 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1153 blk_queue_logical_block_size(lo->lo_queue, bsize);
1154 blk_queue_physical_block_size(lo->lo_queue, bsize);
1155 blk_queue_io_min(lo->lo_queue, bsize);
1157 loop_update_rotational(lo);
1158 loop_update_dio(lo);
1159 loop_sysfs_init(lo);
1161 size = get_loop_size(lo, file);
1162 loop_set_size(lo, size);
1164 lo->lo_state = Lo_bound;
1166 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1167 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1169 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1171 /* Grab the block_device to prevent its destruction after we
1172 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1175 mutex_unlock(&lo->lo_mutex);
1177 loop_reread_partitions(lo, bdev);
1178 if (!(mode & FMODE_EXCL))
1179 bd_abort_claiming(bdev, loop_configure);
1183 mutex_unlock(&lo->lo_mutex);
1185 if (!(mode & FMODE_EXCL))
1186 bd_abort_claiming(bdev, loop_configure);
1190 /* This is safe: open() is still holding a reference. */
1191 module_put(THIS_MODULE);
1195 static int __loop_clr_fd(struct loop_device *lo, bool release)
1197 struct file *filp = NULL;
1198 gfp_t gfp = lo->old_gfp_mask;
1199 struct block_device *bdev = lo->lo_device;
1201 bool partscan = false;
1204 mutex_lock(&lo->lo_mutex);
1205 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1210 filp = lo->lo_backing_file;
1216 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1217 blk_queue_write_cache(lo->lo_queue, false, false);
1219 /* freeze request queue during the transition */
1220 blk_mq_freeze_queue(lo->lo_queue);
1222 spin_lock_irq(&lo->lo_lock);
1223 lo->lo_backing_file = NULL;
1224 spin_unlock_irq(&lo->lo_lock);
1226 loop_release_xfer(lo);
1227 lo->transfer = NULL;
1229 lo->lo_device = NULL;
1230 lo->lo_encryption = NULL;
1232 lo->lo_sizelimit = 0;
1233 lo->lo_encrypt_key_size = 0;
1234 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1235 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1236 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1237 blk_queue_logical_block_size(lo->lo_queue, 512);
1238 blk_queue_physical_block_size(lo->lo_queue, 512);
1239 blk_queue_io_min(lo->lo_queue, 512);
1242 invalidate_bdev(bdev);
1243 bdev->bd_inode->i_mapping->wb_err = 0;
1245 set_capacity(lo->lo_disk, 0);
1246 loop_sysfs_exit(lo);
1248 /* let user-space know about this change */
1249 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1251 mapping_set_gfp_mask(filp->f_mapping, gfp);
1252 /* This is safe: open() is still holding a reference. */
1253 module_put(THIS_MODULE);
1254 blk_mq_unfreeze_queue(lo->lo_queue);
1256 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1257 lo_number = lo->lo_number;
1258 loop_unprepare_queue(lo);
1260 mutex_unlock(&lo->lo_mutex);
1263 * bd_mutex has been held already in release path, so don't
1264 * acquire it if this function is called in such case.
1266 * If the reread partition isn't from release path, lo_refcnt
1267 * must be at least one and it can only become zero when the
1268 * current holder is released.
1271 mutex_lock(&bdev->bd_mutex);
1272 err = bdev_disk_changed(bdev, false);
1274 mutex_unlock(&bdev->bd_mutex);
1276 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1277 __func__, lo_number, err);
1278 /* Device is gone, no point in returning error */
1283 * lo->lo_state is set to Lo_unbound here after above partscan has
1286 * There cannot be anybody else entering __loop_clr_fd() as
1287 * lo->lo_backing_file is already cleared and Lo_rundown state
1288 * protects us from all the other places trying to change the 'lo'
1291 mutex_lock(&lo->lo_mutex);
1294 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1295 lo->lo_state = Lo_unbound;
1296 mutex_unlock(&lo->lo_mutex);
1299 * Need not hold lo_mutex to fput backing file. Calling fput holding
1300 * lo_mutex triggers a circular lock dependency possibility warning as
1301 * fput can take bd_mutex which is usually taken before lo_mutex.
1308 static int loop_clr_fd(struct loop_device *lo)
1312 err = mutex_lock_killable(&lo->lo_mutex);
1315 if (lo->lo_state != Lo_bound) {
1316 mutex_unlock(&lo->lo_mutex);
1320 * If we've explicitly asked to tear down the loop device,
1321 * and it has an elevated reference count, set it for auto-teardown when
1322 * the last reference goes away. This stops $!~#$@ udev from
1323 * preventing teardown because it decided that it needs to run blkid on
1324 * the loopback device whenever they appear. xfstests is notorious for
1325 * failing tests because blkid via udev races with a losetup
1326 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1327 * command to fail with EBUSY.
1329 if (atomic_read(&lo->lo_refcnt) > 1) {
1330 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1331 mutex_unlock(&lo->lo_mutex);
1334 lo->lo_state = Lo_rundown;
1335 mutex_unlock(&lo->lo_mutex);
1337 return __loop_clr_fd(lo, false);
1341 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1344 struct block_device *bdev;
1345 kuid_t uid = current_uid();
1347 bool partscan = false;
1348 bool size_changed = false;
1350 err = mutex_lock_killable(&lo->lo_mutex);
1353 if (lo->lo_encrypt_key_size &&
1354 !uid_eq(lo->lo_key_owner, uid) &&
1355 !capable(CAP_SYS_ADMIN)) {
1359 if (lo->lo_state != Lo_bound) {
1364 if (lo->lo_offset != info->lo_offset ||
1365 lo->lo_sizelimit != info->lo_sizelimit) {
1366 size_changed = true;
1367 sync_blockdev(lo->lo_device);
1368 invalidate_bdev(lo->lo_device);
1371 /* I/O need to be drained during transfer transition */
1372 blk_mq_freeze_queue(lo->lo_queue);
1374 if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) {
1375 /* If any pages were dirtied after invalidate_bdev(), try again */
1377 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1378 __func__, lo->lo_number, lo->lo_file_name,
1379 lo->lo_device->bd_inode->i_mapping->nrpages);
1383 prev_lo_flags = lo->lo_flags;
1385 err = loop_set_status_from_info(lo, info);
1389 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1390 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1391 /* For those flags, use the previous values instead */
1392 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1393 /* For flags that can't be cleared, use previous values too */
1394 lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1397 loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1398 lo->lo_backing_file);
1399 loop_set_size(lo, new_size);
1402 loop_config_discard(lo);
1404 /* update dio if lo_offset or transfer is changed */
1405 __loop_update_dio(lo, lo->use_dio);
1408 blk_mq_unfreeze_queue(lo->lo_queue);
1410 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1411 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1412 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1413 bdev = lo->lo_device;
1417 mutex_unlock(&lo->lo_mutex);
1419 loop_reread_partitions(lo, bdev);
1425 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1431 ret = mutex_lock_killable(&lo->lo_mutex);
1434 if (lo->lo_state != Lo_bound) {
1435 mutex_unlock(&lo->lo_mutex);
1439 memset(info, 0, sizeof(*info));
1440 info->lo_number = lo->lo_number;
1441 info->lo_offset = lo->lo_offset;
1442 info->lo_sizelimit = lo->lo_sizelimit;
1443 info->lo_flags = lo->lo_flags;
1444 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1445 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1446 info->lo_encrypt_type =
1447 lo->lo_encryption ? lo->lo_encryption->number : 0;
1448 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1449 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1450 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1451 lo->lo_encrypt_key_size);
1454 /* Drop lo_mutex while we call into the filesystem. */
1455 path = lo->lo_backing_file->f_path;
1457 mutex_unlock(&lo->lo_mutex);
1458 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1460 info->lo_device = huge_encode_dev(stat.dev);
1461 info->lo_inode = stat.ino;
1462 info->lo_rdevice = huge_encode_dev(stat.rdev);
1469 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1471 memset(info64, 0, sizeof(*info64));
1472 info64->lo_number = info->lo_number;
1473 info64->lo_device = info->lo_device;
1474 info64->lo_inode = info->lo_inode;
1475 info64->lo_rdevice = info->lo_rdevice;
1476 info64->lo_offset = info->lo_offset;
1477 info64->lo_sizelimit = 0;
1478 info64->lo_encrypt_type = info->lo_encrypt_type;
1479 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1480 info64->lo_flags = info->lo_flags;
1481 info64->lo_init[0] = info->lo_init[0];
1482 info64->lo_init[1] = info->lo_init[1];
1483 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1484 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1486 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1487 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1491 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1493 memset(info, 0, sizeof(*info));
1494 info->lo_number = info64->lo_number;
1495 info->lo_device = info64->lo_device;
1496 info->lo_inode = info64->lo_inode;
1497 info->lo_rdevice = info64->lo_rdevice;
1498 info->lo_offset = info64->lo_offset;
1499 info->lo_encrypt_type = info64->lo_encrypt_type;
1500 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1501 info->lo_flags = info64->lo_flags;
1502 info->lo_init[0] = info64->lo_init[0];
1503 info->lo_init[1] = info64->lo_init[1];
1504 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1505 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1507 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1508 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1510 /* error in case values were truncated */
1511 if (info->lo_device != info64->lo_device ||
1512 info->lo_rdevice != info64->lo_rdevice ||
1513 info->lo_inode != info64->lo_inode ||
1514 info->lo_offset != info64->lo_offset)
1521 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1523 struct loop_info info;
1524 struct loop_info64 info64;
1526 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1528 loop_info64_from_old(&info, &info64);
1529 return loop_set_status(lo, &info64);
1533 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1535 struct loop_info64 info64;
1537 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1539 return loop_set_status(lo, &info64);
1543 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1544 struct loop_info info;
1545 struct loop_info64 info64;
1550 err = loop_get_status(lo, &info64);
1552 err = loop_info64_to_old(&info64, &info);
1553 if (!err && copy_to_user(arg, &info, sizeof(info)))
1560 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1561 struct loop_info64 info64;
1566 err = loop_get_status(lo, &info64);
1567 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1573 static int loop_set_capacity(struct loop_device *lo)
1577 if (unlikely(lo->lo_state != Lo_bound))
1580 size = get_loop_size(lo, lo->lo_backing_file);
1581 loop_set_size(lo, size);
1586 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1589 if (lo->lo_state != Lo_bound)
1592 __loop_update_dio(lo, !!arg);
1593 if (lo->use_dio == !!arg)
1600 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1604 if (lo->lo_state != Lo_bound)
1607 err = loop_validate_block_size(arg);
1611 if (lo->lo_queue->limits.logical_block_size == arg)
1614 sync_blockdev(lo->lo_device);
1615 invalidate_bdev(lo->lo_device);
1617 blk_mq_freeze_queue(lo->lo_queue);
1619 /* invalidate_bdev should have truncated all the pages */
1620 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1622 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1623 __func__, lo->lo_number, lo->lo_file_name,
1624 lo->lo_device->bd_inode->i_mapping->nrpages);
1628 blk_queue_logical_block_size(lo->lo_queue, arg);
1629 blk_queue_physical_block_size(lo->lo_queue, arg);
1630 blk_queue_io_min(lo->lo_queue, arg);
1631 loop_update_dio(lo);
1633 blk_mq_unfreeze_queue(lo->lo_queue);
1638 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1643 err = mutex_lock_killable(&lo->lo_mutex);
1647 case LOOP_SET_CAPACITY:
1648 err = loop_set_capacity(lo);
1650 case LOOP_SET_DIRECT_IO:
1651 err = loop_set_dio(lo, arg);
1653 case LOOP_SET_BLOCK_SIZE:
1654 err = loop_set_block_size(lo, arg);
1657 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1659 mutex_unlock(&lo->lo_mutex);
1663 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1664 unsigned int cmd, unsigned long arg)
1666 struct loop_device *lo = bdev->bd_disk->private_data;
1667 void __user *argp = (void __user *) arg;
1673 * Legacy case - pass in a zeroed out struct loop_config with
1674 * only the file descriptor set , which corresponds with the
1675 * default parameters we'd have used otherwise.
1677 struct loop_config config;
1679 memset(&config, 0, sizeof(config));
1682 return loop_configure(lo, mode, bdev, &config);
1684 case LOOP_CONFIGURE: {
1685 struct loop_config config;
1687 if (copy_from_user(&config, argp, sizeof(config)))
1690 return loop_configure(lo, mode, bdev, &config);
1692 case LOOP_CHANGE_FD:
1693 return loop_change_fd(lo, bdev, arg);
1695 return loop_clr_fd(lo);
1696 case LOOP_SET_STATUS:
1698 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1699 err = loop_set_status_old(lo, argp);
1702 case LOOP_GET_STATUS:
1703 return loop_get_status_old(lo, argp);
1704 case LOOP_SET_STATUS64:
1706 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1707 err = loop_set_status64(lo, argp);
1710 case LOOP_GET_STATUS64:
1711 return loop_get_status64(lo, argp);
1712 case LOOP_SET_CAPACITY:
1713 case LOOP_SET_DIRECT_IO:
1714 case LOOP_SET_BLOCK_SIZE:
1715 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1719 err = lo_simple_ioctl(lo, cmd, arg);
1726 #ifdef CONFIG_COMPAT
1727 struct compat_loop_info {
1728 compat_int_t lo_number; /* ioctl r/o */
1729 compat_dev_t lo_device; /* ioctl r/o */
1730 compat_ulong_t lo_inode; /* ioctl r/o */
1731 compat_dev_t lo_rdevice; /* ioctl r/o */
1732 compat_int_t lo_offset;
1733 compat_int_t lo_encrypt_type;
1734 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1735 compat_int_t lo_flags; /* ioctl r/o */
1736 char lo_name[LO_NAME_SIZE];
1737 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1738 compat_ulong_t lo_init[2];
1743 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1744 * - noinlined to reduce stack space usage in main part of driver
1747 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1748 struct loop_info64 *info64)
1750 struct compat_loop_info info;
1752 if (copy_from_user(&info, arg, sizeof(info)))
1755 memset(info64, 0, sizeof(*info64));
1756 info64->lo_number = info.lo_number;
1757 info64->lo_device = info.lo_device;
1758 info64->lo_inode = info.lo_inode;
1759 info64->lo_rdevice = info.lo_rdevice;
1760 info64->lo_offset = info.lo_offset;
1761 info64->lo_sizelimit = 0;
1762 info64->lo_encrypt_type = info.lo_encrypt_type;
1763 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1764 info64->lo_flags = info.lo_flags;
1765 info64->lo_init[0] = info.lo_init[0];
1766 info64->lo_init[1] = info.lo_init[1];
1767 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1768 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1770 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1771 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1776 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1777 * - noinlined to reduce stack space usage in main part of driver
1780 loop_info64_to_compat(const struct loop_info64 *info64,
1781 struct compat_loop_info __user *arg)
1783 struct compat_loop_info info;
1785 memset(&info, 0, sizeof(info));
1786 info.lo_number = info64->lo_number;
1787 info.lo_device = info64->lo_device;
1788 info.lo_inode = info64->lo_inode;
1789 info.lo_rdevice = info64->lo_rdevice;
1790 info.lo_offset = info64->lo_offset;
1791 info.lo_encrypt_type = info64->lo_encrypt_type;
1792 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1793 info.lo_flags = info64->lo_flags;
1794 info.lo_init[0] = info64->lo_init[0];
1795 info.lo_init[1] = info64->lo_init[1];
1796 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1797 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1799 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1800 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1802 /* error in case values were truncated */
1803 if (info.lo_device != info64->lo_device ||
1804 info.lo_rdevice != info64->lo_rdevice ||
1805 info.lo_inode != info64->lo_inode ||
1806 info.lo_offset != info64->lo_offset ||
1807 info.lo_init[0] != info64->lo_init[0] ||
1808 info.lo_init[1] != info64->lo_init[1])
1811 if (copy_to_user(arg, &info, sizeof(info)))
1817 loop_set_status_compat(struct loop_device *lo,
1818 const struct compat_loop_info __user *arg)
1820 struct loop_info64 info64;
1823 ret = loop_info64_from_compat(arg, &info64);
1826 return loop_set_status(lo, &info64);
1830 loop_get_status_compat(struct loop_device *lo,
1831 struct compat_loop_info __user *arg)
1833 struct loop_info64 info64;
1838 err = loop_get_status(lo, &info64);
1840 err = loop_info64_to_compat(&info64, arg);
1844 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1845 unsigned int cmd, unsigned long arg)
1847 struct loop_device *lo = bdev->bd_disk->private_data;
1851 case LOOP_SET_STATUS:
1852 err = loop_set_status_compat(lo,
1853 (const struct compat_loop_info __user *)arg);
1855 case LOOP_GET_STATUS:
1856 err = loop_get_status_compat(lo,
1857 (struct compat_loop_info __user *)arg);
1859 case LOOP_SET_CAPACITY:
1861 case LOOP_GET_STATUS64:
1862 case LOOP_SET_STATUS64:
1863 case LOOP_CONFIGURE:
1864 arg = (unsigned long) compat_ptr(arg);
1867 case LOOP_CHANGE_FD:
1868 case LOOP_SET_BLOCK_SIZE:
1869 case LOOP_SET_DIRECT_IO:
1870 err = lo_ioctl(bdev, mode, cmd, arg);
1880 static int lo_open(struct block_device *bdev, fmode_t mode)
1882 struct loop_device *lo = bdev->bd_disk->private_data;
1885 err = mutex_lock_killable(&lo->lo_mutex);
1888 if (lo->lo_state == Lo_deleting)
1891 atomic_inc(&lo->lo_refcnt);
1892 mutex_unlock(&lo->lo_mutex);
1896 static void lo_release(struct gendisk *disk, fmode_t mode)
1898 struct loop_device *lo = disk->private_data;
1900 mutex_lock(&lo->lo_mutex);
1901 if (atomic_dec_return(&lo->lo_refcnt))
1904 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1905 if (lo->lo_state != Lo_bound)
1907 lo->lo_state = Lo_rundown;
1908 mutex_unlock(&lo->lo_mutex);
1910 * In autoclear mode, stop the loop thread
1911 * and remove configuration after last close.
1913 __loop_clr_fd(lo, true);
1915 } else if (lo->lo_state == Lo_bound) {
1917 * Otherwise keep thread (if running) and config,
1918 * but flush possible ongoing bios in thread.
1920 blk_mq_freeze_queue(lo->lo_queue);
1921 blk_mq_unfreeze_queue(lo->lo_queue);
1925 mutex_unlock(&lo->lo_mutex);
1928 static const struct block_device_operations lo_fops = {
1929 .owner = THIS_MODULE,
1931 .release = lo_release,
1933 #ifdef CONFIG_COMPAT
1934 .compat_ioctl = lo_compat_ioctl,
1939 * And now the modules code and kernel interface.
1941 static int max_loop;
1942 module_param(max_loop, int, 0444);
1943 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1944 module_param(max_part, int, 0444);
1945 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1946 MODULE_LICENSE("GPL");
1947 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1949 int loop_register_transfer(struct loop_func_table *funcs)
1951 unsigned int n = funcs->number;
1953 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1955 xfer_funcs[n] = funcs;
1959 static int unregister_transfer_cb(int id, void *ptr, void *data)
1961 struct loop_device *lo = ptr;
1962 struct loop_func_table *xfer = data;
1964 mutex_lock(&lo->lo_mutex);
1965 if (lo->lo_encryption == xfer)
1966 loop_release_xfer(lo);
1967 mutex_unlock(&lo->lo_mutex);
1971 int loop_unregister_transfer(int number)
1973 unsigned int n = number;
1974 struct loop_func_table *xfer;
1976 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1979 xfer_funcs[n] = NULL;
1980 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1984 EXPORT_SYMBOL(loop_register_transfer);
1985 EXPORT_SYMBOL(loop_unregister_transfer);
1987 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1988 const struct blk_mq_queue_data *bd)
1990 struct request *rq = bd->rq;
1991 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1992 struct loop_device *lo = rq->q->queuedata;
1994 blk_mq_start_request(rq);
1996 if (lo->lo_state != Lo_bound)
1997 return BLK_STS_IOERR;
1999 switch (req_op(rq)) {
2001 case REQ_OP_DISCARD:
2002 case REQ_OP_WRITE_ZEROES:
2003 cmd->use_aio = false;
2006 cmd->use_aio = lo->use_dio;
2010 /* always use the first bio's css */
2011 #ifdef CONFIG_BLK_CGROUP
2012 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
2013 cmd->css = &bio_blkcg(rq->bio)->css;
2018 kthread_queue_work(&lo->worker, &cmd->work);
2023 static void loop_handle_cmd(struct loop_cmd *cmd)
2025 struct request *rq = blk_mq_rq_from_pdu(cmd);
2026 const bool write = op_is_write(req_op(rq));
2027 struct loop_device *lo = rq->q->queuedata;
2030 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
2035 ret = do_req_filebacked(lo, rq);
2037 /* complete non-aio request */
2038 if (!cmd->use_aio || ret) {
2039 if (ret == -EOPNOTSUPP)
2042 cmd->ret = ret ? -EIO : 0;
2043 if (likely(!blk_should_fake_timeout(rq->q)))
2044 blk_mq_complete_request(rq);
2048 static void loop_queue_work(struct kthread_work *work)
2050 struct loop_cmd *cmd =
2051 container_of(work, struct loop_cmd, work);
2053 loop_handle_cmd(cmd);
2056 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
2057 unsigned int hctx_idx, unsigned int numa_node)
2059 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
2061 kthread_init_work(&cmd->work, loop_queue_work);
2065 static const struct blk_mq_ops loop_mq_ops = {
2066 .queue_rq = loop_queue_rq,
2067 .init_request = loop_init_request,
2068 .complete = lo_complete_rq,
2071 static int loop_add(struct loop_device **l, int i)
2073 struct loop_device *lo;
2074 struct gendisk *disk;
2078 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2082 lo->lo_state = Lo_unbound;
2084 /* allocate id, if @id >= 0, we're requesting that specific id */
2086 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2090 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2097 lo->tag_set.ops = &loop_mq_ops;
2098 lo->tag_set.nr_hw_queues = 1;
2099 lo->tag_set.queue_depth = 128;
2100 lo->tag_set.numa_node = NUMA_NO_NODE;
2101 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2102 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING;
2103 lo->tag_set.driver_data = lo;
2105 err = blk_mq_alloc_tag_set(&lo->tag_set);
2109 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2110 if (IS_ERR(lo->lo_queue)) {
2111 err = PTR_ERR(lo->lo_queue);
2112 goto out_cleanup_tags;
2114 lo->lo_queue->queuedata = lo;
2116 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2119 * By default, we do buffer IO, so it doesn't make sense to enable
2120 * merge because the I/O submitted to backing file is handled page by
2121 * page. For directio mode, merge does help to dispatch bigger request
2122 * to underlayer disk. We will enable merge once directio is enabled.
2124 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2127 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2129 goto out_free_queue;
2132 * Disable partition scanning by default. The in-kernel partition
2133 * scanning can be requested individually per-device during its
2134 * setup. Userspace can always add and remove partitions from all
2135 * devices. The needed partition minors are allocated from the
2136 * extended minor space, the main loop device numbers will continue
2137 * to match the loop minors, regardless of the number of partitions
2140 * If max_part is given, partition scanning is globally enabled for
2141 * all loop devices. The minors for the main loop devices will be
2142 * multiples of max_part.
2144 * Note: Global-for-all-devices, set-only-at-init, read-only module
2145 * parameteters like 'max_loop' and 'max_part' make things needlessly
2146 * complicated, are too static, inflexible and may surprise
2147 * userspace tools. Parameters like this in general should be avoided.
2150 disk->flags |= GENHD_FL_NO_PART_SCAN;
2151 disk->flags |= GENHD_FL_EXT_DEVT;
2152 atomic_set(&lo->lo_refcnt, 0);
2153 mutex_init(&lo->lo_mutex);
2155 spin_lock_init(&lo->lo_lock);
2156 disk->major = LOOP_MAJOR;
2157 disk->first_minor = i << part_shift;
2158 disk->fops = &lo_fops;
2159 disk->private_data = lo;
2160 disk->queue = lo->lo_queue;
2161 sprintf(disk->disk_name, "loop%d", i);
2164 return lo->lo_number;
2167 blk_cleanup_queue(lo->lo_queue);
2169 blk_mq_free_tag_set(&lo->tag_set);
2171 idr_remove(&loop_index_idr, i);
2178 static void loop_remove(struct loop_device *lo)
2180 del_gendisk(lo->lo_disk);
2181 blk_cleanup_queue(lo->lo_queue);
2182 blk_mq_free_tag_set(&lo->tag_set);
2183 put_disk(lo->lo_disk);
2184 mutex_destroy(&lo->lo_mutex);
2188 static int find_free_cb(int id, void *ptr, void *data)
2190 struct loop_device *lo = ptr;
2191 struct loop_device **l = data;
2193 if (lo->lo_state == Lo_unbound) {
2200 static int loop_lookup(struct loop_device **l, int i)
2202 struct loop_device *lo;
2208 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2211 ret = lo->lo_number;
2216 /* lookup and return a specific i */
2217 lo = idr_find(&loop_index_idr, i);
2220 ret = lo->lo_number;
2226 static void loop_probe(dev_t dev)
2228 int idx = MINOR(dev) >> part_shift;
2229 struct loop_device *lo;
2231 if (max_loop && idx >= max_loop)
2234 mutex_lock(&loop_ctl_mutex);
2235 if (loop_lookup(&lo, idx) < 0)
2237 mutex_unlock(&loop_ctl_mutex);
2240 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2243 struct loop_device *lo;
2246 ret = mutex_lock_killable(&loop_ctl_mutex);
2253 ret = loop_lookup(&lo, parm);
2258 ret = loop_add(&lo, parm);
2260 case LOOP_CTL_REMOVE:
2261 ret = loop_lookup(&lo, parm);
2264 ret = mutex_lock_killable(&lo->lo_mutex);
2267 if (lo->lo_state != Lo_unbound) {
2269 mutex_unlock(&lo->lo_mutex);
2272 if (atomic_read(&lo->lo_refcnt) > 0) {
2274 mutex_unlock(&lo->lo_mutex);
2277 lo->lo_state = Lo_deleting;
2278 mutex_unlock(&lo->lo_mutex);
2279 idr_remove(&loop_index_idr, lo->lo_number);
2282 case LOOP_CTL_GET_FREE:
2283 ret = loop_lookup(&lo, -1);
2286 ret = loop_add(&lo, -1);
2288 mutex_unlock(&loop_ctl_mutex);
2293 static const struct file_operations loop_ctl_fops = {
2294 .open = nonseekable_open,
2295 .unlocked_ioctl = loop_control_ioctl,
2296 .compat_ioctl = loop_control_ioctl,
2297 .owner = THIS_MODULE,
2298 .llseek = noop_llseek,
2301 static struct miscdevice loop_misc = {
2302 .minor = LOOP_CTRL_MINOR,
2303 .name = "loop-control",
2304 .fops = &loop_ctl_fops,
2307 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2308 MODULE_ALIAS("devname:loop-control");
2310 static int __init loop_init(void)
2313 struct loop_device *lo;
2318 part_shift = fls(max_part);
2321 * Adjust max_part according to part_shift as it is exported
2322 * to user space so that user can decide correct minor number
2323 * if [s]he want to create more devices.
2325 * Note that -1 is required because partition 0 is reserved
2326 * for the whole disk.
2328 max_part = (1UL << part_shift) - 1;
2331 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2336 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2342 * If max_loop is specified, create that many devices upfront.
2343 * This also becomes a hard limit. If max_loop is not specified,
2344 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2345 * init time. Loop devices can be requested on-demand with the
2346 * /dev/loop-control interface, or be instantiated by accessing
2347 * a 'dead' device node.
2352 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2354 err = misc_register(&loop_misc);
2359 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2364 /* pre-create number of devices given by config or max_loop */
2365 mutex_lock(&loop_ctl_mutex);
2366 for (i = 0; i < nr; i++)
2368 mutex_unlock(&loop_ctl_mutex);
2370 printk(KERN_INFO "loop: module loaded\n");
2374 misc_deregister(&loop_misc);
2379 static int loop_exit_cb(int id, void *ptr, void *data)
2381 struct loop_device *lo = ptr;
2387 static void __exit loop_exit(void)
2389 mutex_lock(&loop_ctl_mutex);
2391 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2392 idr_destroy(&loop_index_idr);
2394 unregister_blkdev(LOOP_MAJOR, "loop");
2396 misc_deregister(&loop_misc);
2398 mutex_unlock(&loop_ctl_mutex);
2401 module_init(loop_init);
2402 module_exit(loop_exit);
2405 static int __init max_loop_setup(char *str)
2407 max_loop = simple_strtol(str, NULL, 0);
2411 __setup("max_loop=", max_loop_setup);