2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include <linux/ioprio.h>
83 #include <linux/uaccess.h>
85 static DEFINE_IDR(loop_index_idr);
86 static DEFINE_MUTEX(loop_ctl_mutex);
89 static int part_shift;
91 static int transfer_xor(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
96 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
109 key = lo->lo_encrypt_key;
110 keysize = lo->lo_encrypt_key_size;
111 for (i = 0; i < size; i++)
112 *out++ = *in++ ^ key[(i & 511) % keysize];
114 kunmap_atomic(loop_buf);
115 kunmap_atomic(raw_buf);
120 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122 if (unlikely(info->lo_encrypt_key_size <= 0))
127 static struct loop_func_table none_funcs = {
128 .number = LO_CRYPT_NONE,
131 static struct loop_func_table xor_funcs = {
132 .number = LO_CRYPT_XOR,
133 .transfer = transfer_xor,
137 /* xfer_funcs[0] is special - its release function is never called */
138 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
143 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
147 /* Compute loopsize in bytes */
148 loopsize = i_size_read(file->f_mapping->host);
151 /* offset is beyond i_size, weird but possible */
155 if (sizelimit > 0 && sizelimit < loopsize)
156 loopsize = sizelimit;
158 * Unfortunately, if we want to do I/O on the device,
159 * the number of 512-byte sectors has to fit into a sector_t.
161 return loopsize >> 9;
164 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
169 static void __loop_update_dio(struct loop_device *lo, bool dio)
171 struct file *file = lo->lo_backing_file;
172 struct address_space *mapping = file->f_mapping;
173 struct inode *inode = mapping->host;
174 unsigned short sb_bsize = 0;
175 unsigned dio_align = 0;
178 if (inode->i_sb->s_bdev) {
179 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
180 dio_align = sb_bsize - 1;
184 * We support direct I/O only if lo_offset is aligned with the
185 * logical I/O size of backing device, and the logical block
186 * size of loop is bigger than the backing device's and the loop
187 * needn't transform transfer.
189 * TODO: the above condition may be loosed in the future, and
190 * direct I/O may be switched runtime at that time because most
191 * of requests in sane applications should be PAGE_SIZE aligned
194 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
195 !(lo->lo_offset & dio_align) &&
196 mapping->a_ops->direct_IO &&
205 if (lo->use_dio == use_dio)
208 /* flush dirty pages before changing direct IO */
212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
214 * will get updated by ioctl(LOOP_GET_STATUS)
216 blk_mq_freeze_queue(lo->lo_queue);
217 lo->use_dio = use_dio;
219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
220 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
223 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
225 blk_mq_unfreeze_queue(lo->lo_queue);
229 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
231 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
232 sector_t x = (sector_t)size;
233 struct block_device *bdev = lo->lo_device;
235 if (unlikely((loff_t)x != size))
237 if (lo->lo_offset != offset)
238 lo->lo_offset = offset;
239 if (lo->lo_sizelimit != sizelimit)
240 lo->lo_sizelimit = sizelimit;
241 set_capacity(lo->lo_disk, x);
242 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
243 /* let user-space know about the new size */
244 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
249 lo_do_transfer(struct loop_device *lo, int cmd,
250 struct page *rpage, unsigned roffs,
251 struct page *lpage, unsigned loffs,
252 int size, sector_t rblock)
256 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
260 printk_ratelimited(KERN_ERR
261 "loop: Transfer error at byte offset %llu, length %i.\n",
262 (unsigned long long)rblock << 9, size);
266 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
271 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
273 file_start_write(file);
274 bw = vfs_iter_write(file, &i, ppos, 0);
275 file_end_write(file);
277 if (likely(bw == bvec->bv_len))
280 printk_ratelimited(KERN_ERR
281 "loop: Write error at byte offset %llu, length %i.\n",
282 (unsigned long long)*ppos, bvec->bv_len);
288 static int lo_write_simple(struct loop_device *lo, struct request *rq,
292 struct req_iterator iter;
295 rq_for_each_segment(bvec, rq, iter) {
296 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
306 * This is the slow, transforming version that needs to double buffer the
307 * data as it cannot do the transformations in place without having direct
308 * access to the destination pages of the backing file.
310 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
313 struct bio_vec bvec, b;
314 struct req_iterator iter;
318 page = alloc_page(GFP_NOIO);
322 rq_for_each_segment(bvec, rq, iter) {
323 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
324 bvec.bv_offset, bvec.bv_len, pos >> 9);
330 b.bv_len = bvec.bv_len;
331 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
340 static int lo_read_simple(struct loop_device *lo, struct request *rq,
344 struct req_iterator iter;
348 rq_for_each_segment(bvec, rq, iter) {
349 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
350 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
354 flush_dcache_page(bvec.bv_page);
356 if (len != bvec.bv_len) {
359 __rq_for_each_bio(bio, rq)
369 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
372 struct bio_vec bvec, b;
373 struct req_iterator iter;
379 page = alloc_page(GFP_NOIO);
383 rq_for_each_segment(bvec, rq, iter) {
388 b.bv_len = bvec.bv_len;
390 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
397 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
398 bvec.bv_offset, len, offset >> 9);
402 flush_dcache_page(bvec.bv_page);
404 if (len != bvec.bv_len) {
407 __rq_for_each_bio(bio, rq)
419 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
422 * We use punch hole to reclaim the free space used by the
423 * image a.k.a. discard. However we do not support discard if
424 * encryption is enabled, because it may give an attacker
425 * useful information.
427 struct file *file = lo->lo_backing_file;
428 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
431 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
436 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
437 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
443 static int lo_req_flush(struct loop_device *lo, struct request *rq)
445 struct file *file = lo->lo_backing_file;
446 int ret = vfs_fsync(file, 0);
447 if (unlikely(ret && ret != -EINVAL))
453 static void lo_complete_rq(struct request *rq)
455 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
456 blk_status_t ret = BLK_STS_OK;
458 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
459 req_op(rq) != REQ_OP_READ) {
466 * Short READ - if we got some data, advance our request and
467 * retry it. If we got no data, end the rest with EIO.
470 blk_update_request(rq, BLK_STS_OK, cmd->ret);
472 blk_mq_requeue_request(rq, true);
475 struct bio *bio = rq->bio;
484 blk_mq_end_request(rq, ret);
488 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
490 struct request *rq = blk_mq_rq_from_pdu(cmd);
492 if (!atomic_dec_and_test(&cmd->ref))
496 blk_mq_complete_request(rq);
499 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
501 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
506 lo_rw_aio_do_completion(cmd);
509 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
512 struct iov_iter iter;
513 struct bio_vec *bvec;
514 struct request *rq = blk_mq_rq_from_pdu(cmd);
515 struct bio *bio = rq->bio;
516 struct file *file = lo->lo_backing_file;
521 if (rq->bio != rq->biotail) {
522 struct req_iterator iter;
525 __rq_for_each_bio(bio, rq)
526 segments += bio_segments(bio);
527 bvec = kmalloc_array(segments, sizeof(struct bio_vec),
534 * The bios of the request may be started from the middle of
535 * the 'bvec' because of bio splitting, so we can't directly
536 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
537 * API will take care of all details for us.
539 rq_for_each_segment(tmp, rq, iter) {
547 * Same here, this bio may be started from the middle of the
548 * 'bvec' because of bio splitting, so offset from the bvec
549 * must be passed to iov iterator
551 offset = bio->bi_iter.bi_bvec_done;
552 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
553 segments = bio_segments(bio);
555 atomic_set(&cmd->ref, 2);
557 iov_iter_bvec(&iter, rw, bvec, segments, blk_rq_bytes(rq));
558 iter.iov_offset = offset;
560 cmd->iocb.ki_pos = pos;
561 cmd->iocb.ki_filp = file;
562 cmd->iocb.ki_complete = lo_rw_aio_complete;
563 cmd->iocb.ki_flags = IOCB_DIRECT;
564 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
566 kthread_associate_blkcg(cmd->css);
569 ret = call_write_iter(file, &cmd->iocb, &iter);
571 ret = call_read_iter(file, &cmd->iocb, &iter);
573 lo_rw_aio_do_completion(cmd);
574 kthread_associate_blkcg(NULL);
576 if (ret != -EIOCBQUEUED)
577 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
581 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
583 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
584 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
587 * lo_write_simple and lo_read_simple should have been covered
588 * by io submit style function like lo_rw_aio(), one blocker
589 * is that lo_read_simple() need to call flush_dcache_page after
590 * the page is written from kernel, and it isn't easy to handle
591 * this in io submit style function which submits all segments
592 * of the req at one time. And direct read IO doesn't need to
593 * run flush_dcache_page().
595 switch (req_op(rq)) {
597 return lo_req_flush(lo, rq);
599 case REQ_OP_WRITE_ZEROES:
600 return lo_discard(lo, rq, pos);
603 return lo_write_transfer(lo, rq, pos);
604 else if (cmd->use_aio)
605 return lo_rw_aio(lo, cmd, pos, WRITE);
607 return lo_write_simple(lo, rq, pos);
610 return lo_read_transfer(lo, rq, pos);
611 else if (cmd->use_aio)
612 return lo_rw_aio(lo, cmd, pos, READ);
614 return lo_read_simple(lo, rq, pos);
622 static inline void loop_update_dio(struct loop_device *lo)
624 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
628 static void loop_reread_partitions(struct loop_device *lo,
629 struct block_device *bdev)
634 * bd_mutex has been held already in release path, so don't
635 * acquire it if this function is called in such case.
637 * If the reread partition isn't from release path, lo_refcnt
638 * must be at least one and it can only become zero when the
639 * current holder is released.
641 if (!atomic_read(&lo->lo_refcnt))
642 rc = __blkdev_reread_part(bdev);
644 rc = blkdev_reread_part(bdev);
646 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
647 __func__, lo->lo_number, lo->lo_file_name, rc);
650 static inline int is_loop_device(struct file *file)
652 struct inode *i = file->f_mapping->host;
654 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
657 static int loop_validate_file(struct file *file, struct block_device *bdev)
659 struct inode *inode = file->f_mapping->host;
660 struct file *f = file;
662 /* Avoid recursion */
663 while (is_loop_device(f)) {
664 struct loop_device *l;
666 if (f->f_mapping->host->i_bdev == bdev)
669 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
670 if (l->lo_state == Lo_unbound) {
673 f = l->lo_backing_file;
675 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
681 * loop_change_fd switched the backing store of a loopback device to
682 * a new file. This is useful for operating system installers to free up
683 * the original file and in High Availability environments to switch to
684 * an alternative location for the content in case of server meltdown.
685 * This can only work if the loop device is used read-only, and if the
686 * new backing store is the same size and type as the old backing store.
688 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
691 struct file *file, *old_file;
695 if (lo->lo_state != Lo_bound)
698 /* the loop device has to be read-only */
700 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
708 error = loop_validate_file(file, bdev);
712 old_file = lo->lo_backing_file;
716 /* size of the new backing store needs to be the same */
717 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
721 blk_mq_freeze_queue(lo->lo_queue);
722 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
723 lo->lo_backing_file = file;
724 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
725 mapping_set_gfp_mask(file->f_mapping,
726 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
728 blk_mq_unfreeze_queue(lo->lo_queue);
731 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
732 loop_reread_partitions(lo, bdev);
741 /* loop sysfs attributes */
743 static ssize_t loop_attr_show(struct device *dev, char *page,
744 ssize_t (*callback)(struct loop_device *, char *))
746 struct gendisk *disk = dev_to_disk(dev);
747 struct loop_device *lo = disk->private_data;
749 return callback(lo, page);
752 #define LOOP_ATTR_RO(_name) \
753 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
754 static ssize_t loop_attr_do_show_##_name(struct device *d, \
755 struct device_attribute *attr, char *b) \
757 return loop_attr_show(d, b, loop_attr_##_name##_show); \
759 static struct device_attribute loop_attr_##_name = \
760 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
762 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
767 spin_lock_irq(&lo->lo_lock);
768 if (lo->lo_backing_file)
769 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
770 spin_unlock_irq(&lo->lo_lock);
772 if (IS_ERR_OR_NULL(p))
776 memmove(buf, p, ret);
784 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
786 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
789 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
791 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
794 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
796 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
798 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
801 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
803 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
805 return sprintf(buf, "%s\n", partscan ? "1" : "0");
808 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
810 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
812 return sprintf(buf, "%s\n", dio ? "1" : "0");
815 LOOP_ATTR_RO(backing_file);
816 LOOP_ATTR_RO(offset);
817 LOOP_ATTR_RO(sizelimit);
818 LOOP_ATTR_RO(autoclear);
819 LOOP_ATTR_RO(partscan);
822 static struct attribute *loop_attrs[] = {
823 &loop_attr_backing_file.attr,
824 &loop_attr_offset.attr,
825 &loop_attr_sizelimit.attr,
826 &loop_attr_autoclear.attr,
827 &loop_attr_partscan.attr,
832 static struct attribute_group loop_attribute_group = {
837 static void loop_sysfs_init(struct loop_device *lo)
839 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
840 &loop_attribute_group);
843 static void loop_sysfs_exit(struct loop_device *lo)
845 if (lo->sysfs_inited)
846 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
847 &loop_attribute_group);
850 static void loop_config_discard(struct loop_device *lo)
852 struct file *file = lo->lo_backing_file;
853 struct inode *inode = file->f_mapping->host;
854 struct request_queue *q = lo->lo_queue;
857 * We use punch hole to reclaim the free space used by the
858 * image a.k.a. discard. However we do not support discard if
859 * encryption is enabled, because it may give an attacker
860 * useful information.
862 if ((!file->f_op->fallocate) ||
863 lo->lo_encrypt_key_size) {
864 q->limits.discard_granularity = 0;
865 q->limits.discard_alignment = 0;
866 blk_queue_max_discard_sectors(q, 0);
867 blk_queue_max_write_zeroes_sectors(q, 0);
868 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
872 q->limits.discard_granularity = inode->i_sb->s_blocksize;
873 q->limits.discard_alignment = 0;
875 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
876 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
877 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
880 static void loop_unprepare_queue(struct loop_device *lo)
882 kthread_flush_worker(&lo->worker);
883 kthread_stop(lo->worker_task);
886 static int loop_kthread_worker_fn(void *worker_ptr)
888 current->flags |= PF_LESS_THROTTLE;
889 return kthread_worker_fn(worker_ptr);
892 static int loop_prepare_queue(struct loop_device *lo)
894 kthread_init_worker(&lo->worker);
895 lo->worker_task = kthread_run(loop_kthread_worker_fn,
896 &lo->worker, "loop%d", lo->lo_number);
897 if (IS_ERR(lo->worker_task))
899 set_user_nice(lo->worker_task, MIN_NICE);
903 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
904 struct block_device *bdev, unsigned int arg)
908 struct address_space *mapping;
913 /* This is safe, since we have a reference from open(). */
914 __module_get(THIS_MODULE);
922 if (lo->lo_state != Lo_unbound)
925 error = loop_validate_file(file, bdev);
929 mapping = file->f_mapping;
930 inode = mapping->host;
932 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
933 !file->f_op->write_iter)
934 lo_flags |= LO_FLAGS_READ_ONLY;
937 size = get_loop_size(lo, file);
938 if ((loff_t)(sector_t)size != size)
940 error = loop_prepare_queue(lo);
946 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
949 lo->lo_device = bdev;
950 lo->lo_flags = lo_flags;
951 lo->lo_backing_file = file;
954 lo->lo_sizelimit = 0;
955 lo->old_gfp_mask = mapping_gfp_mask(mapping);
956 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
958 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
959 blk_queue_write_cache(lo->lo_queue, true, false);
962 set_capacity(lo->lo_disk, size);
963 bd_set_size(bdev, size << 9);
965 /* let user-space know about the new size */
966 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
968 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
969 block_size(inode->i_bdev) : PAGE_SIZE);
971 lo->lo_state = Lo_bound;
973 lo->lo_flags |= LO_FLAGS_PARTSCAN;
974 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
975 loop_reread_partitions(lo, bdev);
977 /* Grab the block_device to prevent its destruction after we
978 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
986 /* This is safe: open() is still holding a reference. */
987 module_put(THIS_MODULE);
992 loop_release_xfer(struct loop_device *lo)
995 struct loop_func_table *xfer = lo->lo_encryption;
999 err = xfer->release(lo);
1000 lo->transfer = NULL;
1001 lo->lo_encryption = NULL;
1002 module_put(xfer->owner);
1008 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1009 const struct loop_info64 *i)
1014 struct module *owner = xfer->owner;
1016 if (!try_module_get(owner))
1019 err = xfer->init(lo, i);
1023 lo->lo_encryption = xfer;
1028 static int __loop_clr_fd(struct loop_device *lo)
1030 struct file *filp = NULL;
1031 gfp_t gfp = lo->old_gfp_mask;
1032 struct block_device *bdev = lo->lo_device;
1035 mutex_lock(&loop_ctl_mutex);
1036 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1041 filp = lo->lo_backing_file;
1047 /* freeze request queue during the transition */
1048 blk_mq_freeze_queue(lo->lo_queue);
1050 spin_lock_irq(&lo->lo_lock);
1051 lo->lo_backing_file = NULL;
1052 spin_unlock_irq(&lo->lo_lock);
1054 loop_release_xfer(lo);
1055 lo->transfer = NULL;
1057 lo->lo_device = NULL;
1058 lo->lo_encryption = NULL;
1060 lo->lo_sizelimit = 0;
1061 lo->lo_encrypt_key_size = 0;
1062 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1063 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1064 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1065 blk_queue_logical_block_size(lo->lo_queue, 512);
1066 blk_queue_physical_block_size(lo->lo_queue, 512);
1067 blk_queue_io_min(lo->lo_queue, 512);
1070 invalidate_bdev(bdev);
1071 bdev->bd_inode->i_mapping->wb_err = 0;
1073 set_capacity(lo->lo_disk, 0);
1074 loop_sysfs_exit(lo);
1076 bd_set_size(bdev, 0);
1077 /* let user-space know about this change */
1078 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1080 mapping_set_gfp_mask(filp->f_mapping, gfp);
1081 lo->lo_state = Lo_unbound;
1082 /* This is safe: open() is still holding a reference. */
1083 module_put(THIS_MODULE);
1084 blk_mq_unfreeze_queue(lo->lo_queue);
1086 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1087 loop_reread_partitions(lo, bdev);
1090 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1091 loop_unprepare_queue(lo);
1093 mutex_unlock(&loop_ctl_mutex);
1095 * Need not hold loop_ctl_mutex to fput backing file.
1096 * Calling fput holding loop_ctl_mutex triggers a circular
1097 * lock dependency possibility warning as fput can take
1098 * bd_mutex which is usually taken before loop_ctl_mutex.
1105 static int loop_clr_fd(struct loop_device *lo)
1109 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1112 if (lo->lo_state != Lo_bound) {
1113 mutex_unlock(&loop_ctl_mutex);
1117 * If we've explicitly asked to tear down the loop device,
1118 * and it has an elevated reference count, set it for auto-teardown when
1119 * the last reference goes away. This stops $!~#$@ udev from
1120 * preventing teardown because it decided that it needs to run blkid on
1121 * the loopback device whenever they appear. xfstests is notorious for
1122 * failing tests because blkid via udev races with a losetup
1123 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1124 * command to fail with EBUSY.
1126 if (atomic_read(&lo->lo_refcnt) > 1) {
1127 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1128 mutex_unlock(&loop_ctl_mutex);
1131 lo->lo_state = Lo_rundown;
1132 mutex_unlock(&loop_ctl_mutex);
1134 return __loop_clr_fd(lo);
1138 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1141 struct loop_func_table *xfer;
1142 kuid_t uid = current_uid();
1144 if (lo->lo_encrypt_key_size &&
1145 !uid_eq(lo->lo_key_owner, uid) &&
1146 !capable(CAP_SYS_ADMIN))
1148 if (lo->lo_state != Lo_bound)
1150 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1153 /* I/O need to be drained during transfer transition */
1154 blk_mq_freeze_queue(lo->lo_queue);
1156 err = loop_release_xfer(lo);
1160 if (info->lo_encrypt_type) {
1161 unsigned int type = info->lo_encrypt_type;
1163 if (type >= MAX_LO_CRYPT) {
1167 xfer = xfer_funcs[type];
1175 err = loop_init_xfer(lo, xfer, info);
1179 if (lo->lo_offset != info->lo_offset ||
1180 lo->lo_sizelimit != info->lo_sizelimit) {
1181 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1187 loop_config_discard(lo);
1189 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1190 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1191 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1192 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1196 lo->transfer = xfer->transfer;
1197 lo->ioctl = xfer->ioctl;
1199 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1200 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1201 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1203 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1204 lo->lo_init[0] = info->lo_init[0];
1205 lo->lo_init[1] = info->lo_init[1];
1206 if (info->lo_encrypt_key_size) {
1207 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1208 info->lo_encrypt_key_size);
1209 lo->lo_key_owner = uid;
1212 /* update dio if lo_offset or transfer is changed */
1213 __loop_update_dio(lo, lo->use_dio);
1216 blk_mq_unfreeze_queue(lo->lo_queue);
1218 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1219 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1220 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1221 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1222 loop_reread_partitions(lo, lo->lo_device);
1229 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1235 if (lo->lo_state != Lo_bound) {
1236 mutex_unlock(&loop_ctl_mutex);
1240 memset(info, 0, sizeof(*info));
1241 info->lo_number = lo->lo_number;
1242 info->lo_offset = lo->lo_offset;
1243 info->lo_sizelimit = lo->lo_sizelimit;
1244 info->lo_flags = lo->lo_flags;
1245 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1246 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1247 info->lo_encrypt_type =
1248 lo->lo_encryption ? lo->lo_encryption->number : 0;
1249 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1250 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1251 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1252 lo->lo_encrypt_key_size);
1255 /* Drop loop_ctl_mutex while we call into the filesystem. */
1256 path = lo->lo_backing_file->f_path;
1258 mutex_unlock(&loop_ctl_mutex);
1259 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1261 info->lo_device = huge_encode_dev(stat.dev);
1262 info->lo_inode = stat.ino;
1263 info->lo_rdevice = huge_encode_dev(stat.rdev);
1270 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1272 memset(info64, 0, sizeof(*info64));
1273 info64->lo_number = info->lo_number;
1274 info64->lo_device = info->lo_device;
1275 info64->lo_inode = info->lo_inode;
1276 info64->lo_rdevice = info->lo_rdevice;
1277 info64->lo_offset = info->lo_offset;
1278 info64->lo_sizelimit = 0;
1279 info64->lo_encrypt_type = info->lo_encrypt_type;
1280 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1281 info64->lo_flags = info->lo_flags;
1282 info64->lo_init[0] = info->lo_init[0];
1283 info64->lo_init[1] = info->lo_init[1];
1284 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1285 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1287 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1288 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1292 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1294 memset(info, 0, sizeof(*info));
1295 info->lo_number = info64->lo_number;
1296 info->lo_device = info64->lo_device;
1297 info->lo_inode = info64->lo_inode;
1298 info->lo_rdevice = info64->lo_rdevice;
1299 info->lo_offset = info64->lo_offset;
1300 info->lo_encrypt_type = info64->lo_encrypt_type;
1301 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1302 info->lo_flags = info64->lo_flags;
1303 info->lo_init[0] = info64->lo_init[0];
1304 info->lo_init[1] = info64->lo_init[1];
1305 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1306 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1308 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1309 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1311 /* error in case values were truncated */
1312 if (info->lo_device != info64->lo_device ||
1313 info->lo_rdevice != info64->lo_rdevice ||
1314 info->lo_inode != info64->lo_inode ||
1315 info->lo_offset != info64->lo_offset)
1322 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1324 struct loop_info info;
1325 struct loop_info64 info64;
1327 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1329 loop_info64_from_old(&info, &info64);
1330 return loop_set_status(lo, &info64);
1334 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1336 struct loop_info64 info64;
1338 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1340 return loop_set_status(lo, &info64);
1344 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1345 struct loop_info info;
1346 struct loop_info64 info64;
1350 mutex_unlock(&loop_ctl_mutex);
1353 err = loop_get_status(lo, &info64);
1355 err = loop_info64_to_old(&info64, &info);
1356 if (!err && copy_to_user(arg, &info, sizeof(info)))
1363 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1364 struct loop_info64 info64;
1368 mutex_unlock(&loop_ctl_mutex);
1371 err = loop_get_status(lo, &info64);
1372 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1378 static int loop_set_capacity(struct loop_device *lo)
1380 if (unlikely(lo->lo_state != Lo_bound))
1383 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1386 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1389 if (lo->lo_state != Lo_bound)
1392 __loop_update_dio(lo, !!arg);
1393 if (lo->use_dio == !!arg)
1400 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1402 if (lo->lo_state != Lo_bound)
1405 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1408 blk_mq_freeze_queue(lo->lo_queue);
1410 blk_queue_logical_block_size(lo->lo_queue, arg);
1411 blk_queue_physical_block_size(lo->lo_queue, arg);
1412 blk_queue_io_min(lo->lo_queue, arg);
1413 loop_update_dio(lo);
1415 blk_mq_unfreeze_queue(lo->lo_queue);
1420 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1425 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1429 case LOOP_SET_CAPACITY:
1430 err = loop_set_capacity(lo);
1432 case LOOP_SET_DIRECT_IO:
1433 err = loop_set_dio(lo, arg);
1435 case LOOP_SET_BLOCK_SIZE:
1436 err = loop_set_block_size(lo, arg);
1439 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1441 mutex_unlock(&loop_ctl_mutex);
1445 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1446 unsigned int cmd, unsigned long arg)
1448 struct loop_device *lo = bdev->bd_disk->private_data;
1453 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1456 err = loop_set_fd(lo, mode, bdev, arg);
1457 mutex_unlock(&loop_ctl_mutex);
1459 case LOOP_CHANGE_FD:
1460 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1463 err = loop_change_fd(lo, bdev, arg);
1464 mutex_unlock(&loop_ctl_mutex);
1467 return loop_clr_fd(lo);
1468 case LOOP_SET_STATUS:
1470 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1471 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1474 err = loop_set_status_old(lo,
1475 (struct loop_info __user *)arg);
1476 mutex_unlock(&loop_ctl_mutex);
1479 case LOOP_GET_STATUS:
1480 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1483 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1484 /* loop_get_status() unlocks loop_ctl_mutex */
1486 case LOOP_SET_STATUS64:
1488 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1489 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1492 err = loop_set_status64(lo,
1493 (struct loop_info64 __user *) arg);
1494 mutex_unlock(&loop_ctl_mutex);
1497 case LOOP_GET_STATUS64:
1498 err = mutex_lock_killable_nested(&loop_ctl_mutex, 1);
1501 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1502 /* loop_get_status() unlocks loop_ctl_mutex */
1504 case LOOP_SET_CAPACITY:
1505 case LOOP_SET_DIRECT_IO:
1506 case LOOP_SET_BLOCK_SIZE:
1507 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1511 err = lo_simple_ioctl(lo, cmd, arg);
1518 #ifdef CONFIG_COMPAT
1519 struct compat_loop_info {
1520 compat_int_t lo_number; /* ioctl r/o */
1521 compat_dev_t lo_device; /* ioctl r/o */
1522 compat_ulong_t lo_inode; /* ioctl r/o */
1523 compat_dev_t lo_rdevice; /* ioctl r/o */
1524 compat_int_t lo_offset;
1525 compat_int_t lo_encrypt_type;
1526 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1527 compat_int_t lo_flags; /* ioctl r/o */
1528 char lo_name[LO_NAME_SIZE];
1529 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1530 compat_ulong_t lo_init[2];
1535 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1536 * - noinlined to reduce stack space usage in main part of driver
1539 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1540 struct loop_info64 *info64)
1542 struct compat_loop_info info;
1544 if (copy_from_user(&info, arg, sizeof(info)))
1547 memset(info64, 0, sizeof(*info64));
1548 info64->lo_number = info.lo_number;
1549 info64->lo_device = info.lo_device;
1550 info64->lo_inode = info.lo_inode;
1551 info64->lo_rdevice = info.lo_rdevice;
1552 info64->lo_offset = info.lo_offset;
1553 info64->lo_sizelimit = 0;
1554 info64->lo_encrypt_type = info.lo_encrypt_type;
1555 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1556 info64->lo_flags = info.lo_flags;
1557 info64->lo_init[0] = info.lo_init[0];
1558 info64->lo_init[1] = info.lo_init[1];
1559 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1560 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1562 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1563 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1568 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1569 * - noinlined to reduce stack space usage in main part of driver
1572 loop_info64_to_compat(const struct loop_info64 *info64,
1573 struct compat_loop_info __user *arg)
1575 struct compat_loop_info info;
1577 memset(&info, 0, sizeof(info));
1578 info.lo_number = info64->lo_number;
1579 info.lo_device = info64->lo_device;
1580 info.lo_inode = info64->lo_inode;
1581 info.lo_rdevice = info64->lo_rdevice;
1582 info.lo_offset = info64->lo_offset;
1583 info.lo_encrypt_type = info64->lo_encrypt_type;
1584 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1585 info.lo_flags = info64->lo_flags;
1586 info.lo_init[0] = info64->lo_init[0];
1587 info.lo_init[1] = info64->lo_init[1];
1588 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1589 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1591 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1592 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1594 /* error in case values were truncated */
1595 if (info.lo_device != info64->lo_device ||
1596 info.lo_rdevice != info64->lo_rdevice ||
1597 info.lo_inode != info64->lo_inode ||
1598 info.lo_offset != info64->lo_offset ||
1599 info.lo_init[0] != info64->lo_init[0] ||
1600 info.lo_init[1] != info64->lo_init[1])
1603 if (copy_to_user(arg, &info, sizeof(info)))
1609 loop_set_status_compat(struct loop_device *lo,
1610 const struct compat_loop_info __user *arg)
1612 struct loop_info64 info64;
1615 ret = loop_info64_from_compat(arg, &info64);
1618 return loop_set_status(lo, &info64);
1622 loop_get_status_compat(struct loop_device *lo,
1623 struct compat_loop_info __user *arg)
1625 struct loop_info64 info64;
1629 mutex_unlock(&loop_ctl_mutex);
1632 err = loop_get_status(lo, &info64);
1634 err = loop_info64_to_compat(&info64, arg);
1638 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1639 unsigned int cmd, unsigned long arg)
1641 struct loop_device *lo = bdev->bd_disk->private_data;
1645 case LOOP_SET_STATUS:
1646 err = mutex_lock_killable(&loop_ctl_mutex);
1648 err = loop_set_status_compat(lo,
1649 (const struct compat_loop_info __user *)arg);
1650 mutex_unlock(&loop_ctl_mutex);
1653 case LOOP_GET_STATUS:
1654 err = mutex_lock_killable(&loop_ctl_mutex);
1656 err = loop_get_status_compat(lo,
1657 (struct compat_loop_info __user *)arg);
1658 /* loop_get_status() unlocks loop_ctl_mutex */
1661 case LOOP_SET_CAPACITY:
1663 case LOOP_GET_STATUS64:
1664 case LOOP_SET_STATUS64:
1665 arg = (unsigned long) compat_ptr(arg);
1668 case LOOP_CHANGE_FD:
1669 case LOOP_SET_BLOCK_SIZE:
1670 err = lo_ioctl(bdev, mode, cmd, arg);
1680 static int lo_open(struct block_device *bdev, fmode_t mode)
1682 struct loop_device *lo;
1685 err = mutex_lock_killable(&loop_ctl_mutex);
1688 lo = bdev->bd_disk->private_data;
1694 atomic_inc(&lo->lo_refcnt);
1696 mutex_unlock(&loop_ctl_mutex);
1700 static void lo_release(struct gendisk *disk, fmode_t mode)
1702 struct loop_device *lo;
1704 mutex_lock(&loop_ctl_mutex);
1705 lo = disk->private_data;
1706 if (atomic_dec_return(&lo->lo_refcnt))
1709 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1710 if (lo->lo_state != Lo_bound)
1712 lo->lo_state = Lo_rundown;
1713 mutex_unlock(&loop_ctl_mutex);
1715 * In autoclear mode, stop the loop thread
1716 * and remove configuration after last close.
1720 } else if (lo->lo_state == Lo_bound) {
1722 * Otherwise keep thread (if running) and config,
1723 * but flush possible ongoing bios in thread.
1725 blk_mq_freeze_queue(lo->lo_queue);
1726 blk_mq_unfreeze_queue(lo->lo_queue);
1730 mutex_unlock(&loop_ctl_mutex);
1733 static const struct block_device_operations lo_fops = {
1734 .owner = THIS_MODULE,
1736 .release = lo_release,
1738 #ifdef CONFIG_COMPAT
1739 .compat_ioctl = lo_compat_ioctl,
1744 * And now the modules code and kernel interface.
1746 static int max_loop;
1747 module_param(max_loop, int, 0444);
1748 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1749 module_param(max_part, int, 0444);
1750 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1751 MODULE_LICENSE("GPL");
1752 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1754 int loop_register_transfer(struct loop_func_table *funcs)
1756 unsigned int n = funcs->number;
1758 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1760 xfer_funcs[n] = funcs;
1764 static int unregister_transfer_cb(int id, void *ptr, void *data)
1766 struct loop_device *lo = ptr;
1767 struct loop_func_table *xfer = data;
1769 mutex_lock(&loop_ctl_mutex);
1770 if (lo->lo_encryption == xfer)
1771 loop_release_xfer(lo);
1772 mutex_unlock(&loop_ctl_mutex);
1776 int loop_unregister_transfer(int number)
1778 unsigned int n = number;
1779 struct loop_func_table *xfer;
1781 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1784 xfer_funcs[n] = NULL;
1785 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1789 EXPORT_SYMBOL(loop_register_transfer);
1790 EXPORT_SYMBOL(loop_unregister_transfer);
1792 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1793 const struct blk_mq_queue_data *bd)
1795 struct request *rq = bd->rq;
1796 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1797 struct loop_device *lo = rq->q->queuedata;
1799 blk_mq_start_request(rq);
1801 if (lo->lo_state != Lo_bound)
1802 return BLK_STS_IOERR;
1804 switch (req_op(rq)) {
1806 case REQ_OP_DISCARD:
1807 case REQ_OP_WRITE_ZEROES:
1808 cmd->use_aio = false;
1811 cmd->use_aio = lo->use_dio;
1815 /* always use the first bio's css */
1816 #ifdef CONFIG_BLK_CGROUP
1817 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1818 cmd->css = rq->bio->bi_css;
1823 kthread_queue_work(&lo->worker, &cmd->work);
1828 static void loop_handle_cmd(struct loop_cmd *cmd)
1830 struct request *rq = blk_mq_rq_from_pdu(cmd);
1831 const bool write = op_is_write(req_op(rq));
1832 struct loop_device *lo = rq->q->queuedata;
1835 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1840 ret = do_req_filebacked(lo, rq);
1842 /* complete non-aio request */
1843 if (!cmd->use_aio || ret) {
1844 cmd->ret = ret ? -EIO : 0;
1845 blk_mq_complete_request(rq);
1849 static void loop_queue_work(struct kthread_work *work)
1851 struct loop_cmd *cmd =
1852 container_of(work, struct loop_cmd, work);
1854 loop_handle_cmd(cmd);
1857 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1858 unsigned int hctx_idx, unsigned int numa_node)
1860 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1862 kthread_init_work(&cmd->work, loop_queue_work);
1866 static const struct blk_mq_ops loop_mq_ops = {
1867 .queue_rq = loop_queue_rq,
1868 .init_request = loop_init_request,
1869 .complete = lo_complete_rq,
1872 static int loop_add(struct loop_device **l, int i)
1874 struct loop_device *lo;
1875 struct gendisk *disk;
1879 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1883 lo->lo_state = Lo_unbound;
1885 /* allocate id, if @id >= 0, we're requesting that specific id */
1887 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1891 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1898 lo->tag_set.ops = &loop_mq_ops;
1899 lo->tag_set.nr_hw_queues = 1;
1900 lo->tag_set.queue_depth = 128;
1901 lo->tag_set.numa_node = NUMA_NO_NODE;
1902 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1903 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1904 lo->tag_set.driver_data = lo;
1906 err = blk_mq_alloc_tag_set(&lo->tag_set);
1910 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1911 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1912 err = PTR_ERR(lo->lo_queue);
1913 goto out_cleanup_tags;
1915 lo->lo_queue->queuedata = lo;
1917 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1920 * By default, we do buffer IO, so it doesn't make sense to enable
1921 * merge because the I/O submitted to backing file is handled page by
1922 * page. For directio mode, merge does help to dispatch bigger request
1923 * to underlayer disk. We will enable merge once directio is enabled.
1925 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1928 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1930 goto out_free_queue;
1933 * Disable partition scanning by default. The in-kernel partition
1934 * scanning can be requested individually per-device during its
1935 * setup. Userspace can always add and remove partitions from all
1936 * devices. The needed partition minors are allocated from the
1937 * extended minor space, the main loop device numbers will continue
1938 * to match the loop minors, regardless of the number of partitions
1941 * If max_part is given, partition scanning is globally enabled for
1942 * all loop devices. The minors for the main loop devices will be
1943 * multiples of max_part.
1945 * Note: Global-for-all-devices, set-only-at-init, read-only module
1946 * parameteters like 'max_loop' and 'max_part' make things needlessly
1947 * complicated, are too static, inflexible and may surprise
1948 * userspace tools. Parameters like this in general should be avoided.
1951 disk->flags |= GENHD_FL_NO_PART_SCAN;
1952 disk->flags |= GENHD_FL_EXT_DEVT;
1953 atomic_set(&lo->lo_refcnt, 0);
1955 spin_lock_init(&lo->lo_lock);
1956 disk->major = LOOP_MAJOR;
1957 disk->first_minor = i << part_shift;
1958 disk->fops = &lo_fops;
1959 disk->private_data = lo;
1960 disk->queue = lo->lo_queue;
1961 sprintf(disk->disk_name, "loop%d", i);
1964 return lo->lo_number;
1967 blk_cleanup_queue(lo->lo_queue);
1969 blk_mq_free_tag_set(&lo->tag_set);
1971 idr_remove(&loop_index_idr, i);
1978 static void loop_remove(struct loop_device *lo)
1980 del_gendisk(lo->lo_disk);
1981 blk_cleanup_queue(lo->lo_queue);
1982 blk_mq_free_tag_set(&lo->tag_set);
1983 put_disk(lo->lo_disk);
1987 static int find_free_cb(int id, void *ptr, void *data)
1989 struct loop_device *lo = ptr;
1990 struct loop_device **l = data;
1992 if (lo->lo_state == Lo_unbound) {
1999 static int loop_lookup(struct loop_device **l, int i)
2001 struct loop_device *lo;
2007 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2010 ret = lo->lo_number;
2015 /* lookup and return a specific i */
2016 lo = idr_find(&loop_index_idr, i);
2019 ret = lo->lo_number;
2025 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2027 struct loop_device *lo;
2028 struct kobject *kobj;
2031 mutex_lock(&loop_ctl_mutex);
2032 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2034 err = loop_add(&lo, MINOR(dev) >> part_shift);
2038 kobj = get_disk_and_module(lo->lo_disk);
2039 mutex_unlock(&loop_ctl_mutex);
2045 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2048 struct loop_device *lo;
2051 ret = mutex_lock_killable(&loop_ctl_mutex);
2058 ret = loop_lookup(&lo, parm);
2063 ret = loop_add(&lo, parm);
2065 case LOOP_CTL_REMOVE:
2066 ret = loop_lookup(&lo, parm);
2069 if (lo->lo_state != Lo_unbound) {
2071 mutex_unlock(&loop_ctl_mutex);
2074 if (atomic_read(&lo->lo_refcnt) > 0) {
2076 mutex_unlock(&loop_ctl_mutex);
2079 lo->lo_disk->private_data = NULL;
2080 idr_remove(&loop_index_idr, lo->lo_number);
2083 case LOOP_CTL_GET_FREE:
2084 ret = loop_lookup(&lo, -1);
2087 ret = loop_add(&lo, -1);
2089 mutex_unlock(&loop_ctl_mutex);
2094 static const struct file_operations loop_ctl_fops = {
2095 .open = nonseekable_open,
2096 .unlocked_ioctl = loop_control_ioctl,
2097 .compat_ioctl = loop_control_ioctl,
2098 .owner = THIS_MODULE,
2099 .llseek = noop_llseek,
2102 static struct miscdevice loop_misc = {
2103 .minor = LOOP_CTRL_MINOR,
2104 .name = "loop-control",
2105 .fops = &loop_ctl_fops,
2108 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2109 MODULE_ALIAS("devname:loop-control");
2111 static int __init loop_init(void)
2114 unsigned long range;
2115 struct loop_device *lo;
2120 part_shift = fls(max_part);
2123 * Adjust max_part according to part_shift as it is exported
2124 * to user space so that user can decide correct minor number
2125 * if [s]he want to create more devices.
2127 * Note that -1 is required because partition 0 is reserved
2128 * for the whole disk.
2130 max_part = (1UL << part_shift) - 1;
2133 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2138 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2144 * If max_loop is specified, create that many devices upfront.
2145 * This also becomes a hard limit. If max_loop is not specified,
2146 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2147 * init time. Loop devices can be requested on-demand with the
2148 * /dev/loop-control interface, or be instantiated by accessing
2149 * a 'dead' device node.
2153 range = max_loop << part_shift;
2155 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2156 range = 1UL << MINORBITS;
2159 err = misc_register(&loop_misc);
2164 if (register_blkdev(LOOP_MAJOR, "loop")) {
2169 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2170 THIS_MODULE, loop_probe, NULL, NULL);
2172 /* pre-create number of devices given by config or max_loop */
2173 mutex_lock(&loop_ctl_mutex);
2174 for (i = 0; i < nr; i++)
2176 mutex_unlock(&loop_ctl_mutex);
2178 printk(KERN_INFO "loop: module loaded\n");
2182 misc_deregister(&loop_misc);
2187 static int loop_exit_cb(int id, void *ptr, void *data)
2189 struct loop_device *lo = ptr;
2195 static void __exit loop_exit(void)
2197 unsigned long range;
2199 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2201 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2202 idr_destroy(&loop_index_idr);
2204 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2205 unregister_blkdev(LOOP_MAJOR, "loop");
2207 misc_deregister(&loop_misc);
2210 module_init(loop_init);
2211 module_exit(loop_exit);
2214 static int __init max_loop_setup(char *str)
2216 max_loop = simple_strtol(str, NULL, 0);
2220 __setup("max_loop=", max_loop_setup);