2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include <linux/ioprio.h>
80 #include <linux/blk-cgroup.h>
84 #include <linux/uaccess.h>
86 static DEFINE_IDR(loop_index_idr);
87 static DEFINE_MUTEX(loop_ctl_mutex);
90 static int part_shift;
92 static int transfer_xor(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
97 char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page) + loop_off;
110 key = lo->lo_encrypt_key;
111 keysize = lo->lo_encrypt_key_size;
112 for (i = 0; i < size; i++)
113 *out++ = *in++ ^ key[(i & 511) % keysize];
115 kunmap_atomic(loop_buf);
116 kunmap_atomic(raw_buf);
121 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
123 if (unlikely(info->lo_encrypt_key_size <= 0))
128 static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
132 static struct loop_func_table xor_funcs = {
133 .number = LO_CRYPT_XOR,
134 .transfer = transfer_xor,
138 /* xfer_funcs[0] is special - its release function is never called */
139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
144 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
148 /* Compute loopsize in bytes */
149 loopsize = i_size_read(file->f_mapping->host);
152 /* offset is beyond i_size, weird but possible */
156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
162 return loopsize >> 9;
165 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
167 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
170 static void __loop_update_dio(struct loop_device *lo, bool dio)
172 struct file *file = lo->lo_backing_file;
173 struct address_space *mapping = file->f_mapping;
174 struct inode *inode = mapping->host;
175 unsigned short sb_bsize = 0;
176 unsigned dio_align = 0;
179 if (inode->i_sb->s_bdev) {
180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 dio_align = sb_bsize - 1;
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 !(lo->lo_offset & dio_align) &&
197 mapping->a_ops->direct_IO &&
206 if (lo->use_dio == use_dio)
209 /* flush dirty pages before changing direct IO */
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
217 blk_mq_freeze_queue(lo->lo_queue);
218 lo->use_dio = use_dio;
220 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
221 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
223 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
224 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
226 blk_mq_unfreeze_queue(lo->lo_queue);
230 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
232 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
233 sector_t x = (sector_t)size;
234 struct block_device *bdev = lo->lo_device;
236 if (unlikely((loff_t)x != size))
238 if (lo->lo_offset != offset)
239 lo->lo_offset = offset;
240 if (lo->lo_sizelimit != sizelimit)
241 lo->lo_sizelimit = sizelimit;
242 set_capacity(lo->lo_disk, x);
243 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
244 /* let user-space know about the new size */
245 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
250 lo_do_transfer(struct loop_device *lo, int cmd,
251 struct page *rpage, unsigned roffs,
252 struct page *lpage, unsigned loffs,
253 int size, sector_t rblock)
257 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
261 printk_ratelimited(KERN_ERR
262 "loop: Transfer error at byte offset %llu, length %i.\n",
263 (unsigned long long)rblock << 9, size);
267 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
272 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
274 file_start_write(file);
275 bw = vfs_iter_write(file, &i, ppos, 0);
276 file_end_write(file);
278 if (likely(bw == bvec->bv_len))
281 printk_ratelimited(KERN_ERR
282 "loop: Write error at byte offset %llu, length %i.\n",
283 (unsigned long long)*ppos, bvec->bv_len);
289 static int lo_write_simple(struct loop_device *lo, struct request *rq,
293 struct req_iterator iter;
296 rq_for_each_segment(bvec, rq, iter) {
297 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
307 * This is the slow, transforming version that needs to double buffer the
308 * data as it cannot do the transformations in place without having direct
309 * access to the destination pages of the backing file.
311 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
314 struct bio_vec bvec, b;
315 struct req_iterator iter;
319 page = alloc_page(GFP_NOIO);
323 rq_for_each_segment(bvec, rq, iter) {
324 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
325 bvec.bv_offset, bvec.bv_len, pos >> 9);
331 b.bv_len = bvec.bv_len;
332 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
341 static int lo_read_simple(struct loop_device *lo, struct request *rq,
345 struct req_iterator iter;
349 rq_for_each_segment(bvec, rq, iter) {
350 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
351 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
355 flush_dcache_page(bvec.bv_page);
357 if (len != bvec.bv_len) {
360 __rq_for_each_bio(bio, rq)
370 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
373 struct bio_vec bvec, b;
374 struct req_iterator iter;
380 page = alloc_page(GFP_NOIO);
384 rq_for_each_segment(bvec, rq, iter) {
389 b.bv_len = bvec.bv_len;
391 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
392 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
398 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
399 bvec.bv_offset, len, offset >> 9);
403 flush_dcache_page(bvec.bv_page);
405 if (len != bvec.bv_len) {
408 __rq_for_each_bio(bio, rq)
420 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
423 * We use punch hole to reclaim the free space used by the
424 * image a.k.a. discard. However we do not support discard if
425 * encryption is enabled, because it may give an attacker
426 * useful information.
428 struct file *file = lo->lo_backing_file;
429 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
432 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
437 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
438 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
444 static int lo_req_flush(struct loop_device *lo, struct request *rq)
446 struct file *file = lo->lo_backing_file;
447 int ret = vfs_fsync(file, 0);
448 if (unlikely(ret && ret != -EINVAL))
454 static void lo_complete_rq(struct request *rq)
456 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
457 blk_status_t ret = BLK_STS_OK;
459 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
460 req_op(rq) != REQ_OP_READ) {
467 * Short READ - if we got some data, advance our request and
468 * retry it. If we got no data, end the rest with EIO.
471 blk_update_request(rq, BLK_STS_OK, cmd->ret);
473 blk_mq_requeue_request(rq, true);
476 struct bio *bio = rq->bio;
485 blk_mq_end_request(rq, ret);
489 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
491 struct request *rq = blk_mq_rq_from_pdu(cmd);
493 if (!atomic_dec_and_test(&cmd->ref))
497 blk_mq_complete_request(rq);
500 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
502 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
507 lo_rw_aio_do_completion(cmd);
510 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
513 struct iov_iter iter;
514 struct req_iterator rq_iter;
515 struct bio_vec *bvec;
516 struct request *rq = blk_mq_rq_from_pdu(cmd);
517 struct bio *bio = rq->bio;
518 struct file *file = lo->lo_backing_file;
524 rq_for_each_bvec(tmp, rq, rq_iter)
527 if (rq->bio != rq->biotail) {
529 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
536 * The bios of the request may be started from the middle of
537 * the 'bvec' because of bio splitting, so we can't directly
538 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
539 * API will take care of all details for us.
541 rq_for_each_bvec(tmp, rq, rq_iter) {
549 * Same here, this bio may be started from the middle of the
550 * 'bvec' because of bio splitting, so offset from the bvec
551 * must be passed to iov iterator
553 offset = bio->bi_iter.bi_bvec_done;
554 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
556 atomic_set(&cmd->ref, 2);
558 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
559 iter.iov_offset = offset;
561 cmd->iocb.ki_pos = pos;
562 cmd->iocb.ki_filp = file;
563 cmd->iocb.ki_complete = lo_rw_aio_complete;
564 cmd->iocb.ki_flags = IOCB_DIRECT;
565 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
567 kthread_associate_blkcg(cmd->css);
570 ret = call_write_iter(file, &cmd->iocb, &iter);
572 ret = call_read_iter(file, &cmd->iocb, &iter);
574 lo_rw_aio_do_completion(cmd);
575 kthread_associate_blkcg(NULL);
577 if (ret != -EIOCBQUEUED)
578 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
582 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
584 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
585 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
588 * lo_write_simple and lo_read_simple should have been covered
589 * by io submit style function like lo_rw_aio(), one blocker
590 * is that lo_read_simple() need to call flush_dcache_page after
591 * the page is written from kernel, and it isn't easy to handle
592 * this in io submit style function which submits all segments
593 * of the req at one time. And direct read IO doesn't need to
594 * run flush_dcache_page().
596 switch (req_op(rq)) {
598 return lo_req_flush(lo, rq);
600 case REQ_OP_WRITE_ZEROES:
601 return lo_discard(lo, rq, pos);
604 return lo_write_transfer(lo, rq, pos);
605 else if (cmd->use_aio)
606 return lo_rw_aio(lo, cmd, pos, WRITE);
608 return lo_write_simple(lo, rq, pos);
611 return lo_read_transfer(lo, rq, pos);
612 else if (cmd->use_aio)
613 return lo_rw_aio(lo, cmd, pos, READ);
615 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)
633 rc = blkdev_reread_part(bdev);
635 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
636 __func__, lo->lo_number, lo->lo_file_name, rc);
639 static inline int is_loop_device(struct file *file)
641 struct inode *i = file->f_mapping->host;
643 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
646 static int loop_validate_file(struct file *file, struct block_device *bdev)
648 struct inode *inode = file->f_mapping->host;
649 struct file *f = file;
651 /* Avoid recursion */
652 while (is_loop_device(f)) {
653 struct loop_device *l;
655 if (f->f_mapping->host->i_bdev == bdev)
658 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
659 if (l->lo_state != Lo_bound) {
662 f = l->lo_backing_file;
664 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
670 * loop_change_fd switched the backing store of a loopback device to
671 * a new file. This is useful for operating system installers to free up
672 * the original file and in High Availability environments to switch to
673 * an alternative location for the content in case of server meltdown.
674 * This can only work if the loop device is used read-only, and if the
675 * new backing store is the same size and type as the old backing store.
677 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
680 struct file *file = NULL, *old_file;
684 error = mutex_lock_killable(&loop_ctl_mutex);
688 if (lo->lo_state != Lo_bound)
691 /* the loop device has to be read-only */
693 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
701 error = loop_validate_file(file, bdev);
705 old_file = lo->lo_backing_file;
709 /* size of the new backing store needs to be the same */
710 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
714 blk_mq_freeze_queue(lo->lo_queue);
715 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
716 lo->lo_backing_file = file;
717 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
718 mapping_set_gfp_mask(file->f_mapping,
719 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
721 blk_mq_unfreeze_queue(lo->lo_queue);
722 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
723 mutex_unlock(&loop_ctl_mutex);
725 * We must drop file reference outside of loop_ctl_mutex as dropping
726 * the file ref can take bd_mutex which creates circular locking
731 loop_reread_partitions(lo, bdev);
735 mutex_unlock(&loop_ctl_mutex);
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 | PF_MEMALLOC_NOIO;
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 void loop_update_rotational(struct loop_device *lo)
905 struct file *file = lo->lo_backing_file;
906 struct inode *file_inode = file->f_mapping->host;
907 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
908 struct request_queue *q = lo->lo_queue;
911 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
913 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
916 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
918 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
921 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
922 struct block_device *bdev, unsigned int arg)
926 struct address_space *mapping;
927 struct block_device *claimed_bdev = NULL;
933 /* This is safe, since we have a reference from open(). */
934 __module_get(THIS_MODULE);
942 * If we don't hold exclusive handle for the device, upgrade to it
943 * here to avoid changing device under exclusive owner.
945 if (!(mode & FMODE_EXCL)) {
946 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
947 if (IS_ERR(claimed_bdev)) {
948 error = PTR_ERR(claimed_bdev);
953 error = mutex_lock_killable(&loop_ctl_mutex);
958 if (lo->lo_state != Lo_unbound)
961 error = loop_validate_file(file, bdev);
965 mapping = file->f_mapping;
966 inode = mapping->host;
968 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
969 !file->f_op->write_iter)
970 lo_flags |= LO_FLAGS_READ_ONLY;
973 size = get_loop_size(lo, file);
974 if ((loff_t)(sector_t)size != size)
976 error = loop_prepare_queue(lo);
982 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
985 lo->lo_device = bdev;
986 lo->lo_flags = lo_flags;
987 lo->lo_backing_file = file;
990 lo->lo_sizelimit = 0;
991 lo->old_gfp_mask = mapping_gfp_mask(mapping);
992 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
994 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
995 blk_queue_write_cache(lo->lo_queue, true, false);
997 loop_update_rotational(lo);
999 set_capacity(lo->lo_disk, size);
1000 bd_set_size(bdev, size << 9);
1001 loop_sysfs_init(lo);
1002 /* let user-space know about the new size */
1003 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1005 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1006 block_size(inode->i_bdev) : PAGE_SIZE);
1008 lo->lo_state = Lo_bound;
1010 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1011 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1013 /* Grab the block_device to prevent its destruction after we
1014 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1017 mutex_unlock(&loop_ctl_mutex);
1019 loop_reread_partitions(lo, bdev);
1021 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1025 mutex_unlock(&loop_ctl_mutex);
1028 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1032 /* This is safe: open() is still holding a reference. */
1033 module_put(THIS_MODULE);
1038 loop_release_xfer(struct loop_device *lo)
1041 struct loop_func_table *xfer = lo->lo_encryption;
1045 err = xfer->release(lo);
1046 lo->transfer = NULL;
1047 lo->lo_encryption = NULL;
1048 module_put(xfer->owner);
1054 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1055 const struct loop_info64 *i)
1060 struct module *owner = xfer->owner;
1062 if (!try_module_get(owner))
1065 err = xfer->init(lo, i);
1069 lo->lo_encryption = xfer;
1074 static int __loop_clr_fd(struct loop_device *lo, bool release)
1076 struct file *filp = NULL;
1077 gfp_t gfp = lo->old_gfp_mask;
1078 struct block_device *bdev = lo->lo_device;
1080 bool partscan = false;
1083 mutex_lock(&loop_ctl_mutex);
1084 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1089 filp = lo->lo_backing_file;
1095 /* freeze request queue during the transition */
1096 blk_mq_freeze_queue(lo->lo_queue);
1098 spin_lock_irq(&lo->lo_lock);
1099 lo->lo_backing_file = NULL;
1100 spin_unlock_irq(&lo->lo_lock);
1102 loop_release_xfer(lo);
1103 lo->transfer = NULL;
1105 lo->lo_device = NULL;
1106 lo->lo_encryption = NULL;
1108 lo->lo_sizelimit = 0;
1109 lo->lo_encrypt_key_size = 0;
1110 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1111 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1112 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1113 blk_queue_logical_block_size(lo->lo_queue, 512);
1114 blk_queue_physical_block_size(lo->lo_queue, 512);
1115 blk_queue_io_min(lo->lo_queue, 512);
1118 invalidate_bdev(bdev);
1119 bdev->bd_inode->i_mapping->wb_err = 0;
1121 set_capacity(lo->lo_disk, 0);
1122 loop_sysfs_exit(lo);
1124 bd_set_size(bdev, 0);
1125 /* let user-space know about this change */
1126 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1128 mapping_set_gfp_mask(filp->f_mapping, gfp);
1129 /* This is safe: open() is still holding a reference. */
1130 module_put(THIS_MODULE);
1131 blk_mq_unfreeze_queue(lo->lo_queue);
1133 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1134 lo_number = lo->lo_number;
1135 loop_unprepare_queue(lo);
1137 mutex_unlock(&loop_ctl_mutex);
1140 * bd_mutex has been held already in release path, so don't
1141 * acquire it if this function is called in such case.
1143 * If the reread partition isn't from release path, lo_refcnt
1144 * must be at least one and it can only become zero when the
1145 * current holder is released.
1148 err = __blkdev_reread_part(bdev);
1150 err = blkdev_reread_part(bdev);
1152 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1153 __func__, lo_number, err);
1154 /* Device is gone, no point in returning error */
1159 * lo->lo_state is set to Lo_unbound here after above partscan has
1162 * There cannot be anybody else entering __loop_clr_fd() as
1163 * lo->lo_backing_file is already cleared and Lo_rundown state
1164 * protects us from all the other places trying to change the 'lo'
1167 mutex_lock(&loop_ctl_mutex);
1170 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1171 lo->lo_state = Lo_unbound;
1172 mutex_unlock(&loop_ctl_mutex);
1175 * Need not hold loop_ctl_mutex to fput backing file.
1176 * Calling fput holding loop_ctl_mutex triggers a circular
1177 * lock dependency possibility warning as fput can take
1178 * bd_mutex which is usually taken before loop_ctl_mutex.
1185 static int loop_clr_fd(struct loop_device *lo)
1189 err = mutex_lock_killable(&loop_ctl_mutex);
1192 if (lo->lo_state != Lo_bound) {
1193 mutex_unlock(&loop_ctl_mutex);
1197 * If we've explicitly asked to tear down the loop device,
1198 * and it has an elevated reference count, set it for auto-teardown when
1199 * the last reference goes away. This stops $!~#$@ udev from
1200 * preventing teardown because it decided that it needs to run blkid on
1201 * the loopback device whenever they appear. xfstests is notorious for
1202 * failing tests because blkid via udev races with a losetup
1203 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1204 * command to fail with EBUSY.
1206 if (atomic_read(&lo->lo_refcnt) > 1) {
1207 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1208 mutex_unlock(&loop_ctl_mutex);
1211 lo->lo_state = Lo_rundown;
1212 mutex_unlock(&loop_ctl_mutex);
1214 return __loop_clr_fd(lo, false);
1218 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1221 struct loop_func_table *xfer;
1222 kuid_t uid = current_uid();
1223 struct block_device *bdev;
1224 bool partscan = false;
1226 err = mutex_lock_killable(&loop_ctl_mutex);
1229 if (lo->lo_encrypt_key_size &&
1230 !uid_eq(lo->lo_key_owner, uid) &&
1231 !capable(CAP_SYS_ADMIN)) {
1235 if (lo->lo_state != Lo_bound) {
1239 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1244 if (lo->lo_offset != info->lo_offset ||
1245 lo->lo_sizelimit != info->lo_sizelimit) {
1246 sync_blockdev(lo->lo_device);
1247 kill_bdev(lo->lo_device);
1250 /* I/O need to be drained during transfer transition */
1251 blk_mq_freeze_queue(lo->lo_queue);
1253 err = loop_release_xfer(lo);
1257 if (info->lo_encrypt_type) {
1258 unsigned int type = info->lo_encrypt_type;
1260 if (type >= MAX_LO_CRYPT) {
1264 xfer = xfer_funcs[type];
1272 err = loop_init_xfer(lo, xfer, info);
1276 if (lo->lo_offset != info->lo_offset ||
1277 lo->lo_sizelimit != info->lo_sizelimit) {
1278 /* kill_bdev should have truncated all the pages */
1279 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1281 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1282 __func__, lo->lo_number, lo->lo_file_name,
1283 lo->lo_device->bd_inode->i_mapping->nrpages);
1286 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1292 loop_config_discard(lo);
1294 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1295 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1296 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1297 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1301 lo->transfer = xfer->transfer;
1302 lo->ioctl = xfer->ioctl;
1304 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1305 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1306 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1308 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1309 lo->lo_init[0] = info->lo_init[0];
1310 lo->lo_init[1] = info->lo_init[1];
1311 if (info->lo_encrypt_key_size) {
1312 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1313 info->lo_encrypt_key_size);
1314 lo->lo_key_owner = uid;
1317 /* update dio if lo_offset or transfer is changed */
1318 __loop_update_dio(lo, lo->use_dio);
1321 blk_mq_unfreeze_queue(lo->lo_queue);
1323 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1324 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1325 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1326 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1327 bdev = lo->lo_device;
1331 mutex_unlock(&loop_ctl_mutex);
1333 loop_reread_partitions(lo, bdev);
1339 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1345 ret = mutex_lock_killable(&loop_ctl_mutex);
1348 if (lo->lo_state != Lo_bound) {
1349 mutex_unlock(&loop_ctl_mutex);
1353 memset(info, 0, sizeof(*info));
1354 info->lo_number = lo->lo_number;
1355 info->lo_offset = lo->lo_offset;
1356 info->lo_sizelimit = lo->lo_sizelimit;
1357 info->lo_flags = lo->lo_flags;
1358 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1359 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1360 info->lo_encrypt_type =
1361 lo->lo_encryption ? lo->lo_encryption->number : 0;
1362 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1363 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1364 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1365 lo->lo_encrypt_key_size);
1368 /* Drop loop_ctl_mutex while we call into the filesystem. */
1369 path = lo->lo_backing_file->f_path;
1371 mutex_unlock(&loop_ctl_mutex);
1372 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1374 info->lo_device = huge_encode_dev(stat.dev);
1375 info->lo_inode = stat.ino;
1376 info->lo_rdevice = huge_encode_dev(stat.rdev);
1383 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1385 memset(info64, 0, sizeof(*info64));
1386 info64->lo_number = info->lo_number;
1387 info64->lo_device = info->lo_device;
1388 info64->lo_inode = info->lo_inode;
1389 info64->lo_rdevice = info->lo_rdevice;
1390 info64->lo_offset = info->lo_offset;
1391 info64->lo_sizelimit = 0;
1392 info64->lo_encrypt_type = info->lo_encrypt_type;
1393 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1394 info64->lo_flags = info->lo_flags;
1395 info64->lo_init[0] = info->lo_init[0];
1396 info64->lo_init[1] = info->lo_init[1];
1397 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1398 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1400 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1401 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1405 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1407 memset(info, 0, sizeof(*info));
1408 info->lo_number = info64->lo_number;
1409 info->lo_device = info64->lo_device;
1410 info->lo_inode = info64->lo_inode;
1411 info->lo_rdevice = info64->lo_rdevice;
1412 info->lo_offset = info64->lo_offset;
1413 info->lo_encrypt_type = info64->lo_encrypt_type;
1414 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1415 info->lo_flags = info64->lo_flags;
1416 info->lo_init[0] = info64->lo_init[0];
1417 info->lo_init[1] = info64->lo_init[1];
1418 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1419 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1421 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1422 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1424 /* error in case values were truncated */
1425 if (info->lo_device != info64->lo_device ||
1426 info->lo_rdevice != info64->lo_rdevice ||
1427 info->lo_inode != info64->lo_inode ||
1428 info->lo_offset != info64->lo_offset)
1435 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1437 struct loop_info info;
1438 struct loop_info64 info64;
1440 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1442 loop_info64_from_old(&info, &info64);
1443 return loop_set_status(lo, &info64);
1447 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1449 struct loop_info64 info64;
1451 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1453 return loop_set_status(lo, &info64);
1457 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1458 struct loop_info info;
1459 struct loop_info64 info64;
1464 err = loop_get_status(lo, &info64);
1466 err = loop_info64_to_old(&info64, &info);
1467 if (!err && copy_to_user(arg, &info, sizeof(info)))
1474 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1475 struct loop_info64 info64;
1480 err = loop_get_status(lo, &info64);
1481 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1487 static int loop_set_capacity(struct loop_device *lo)
1489 if (unlikely(lo->lo_state != Lo_bound))
1492 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1495 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1498 if (lo->lo_state != Lo_bound)
1501 __loop_update_dio(lo, !!arg);
1502 if (lo->use_dio == !!arg)
1509 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1513 if (lo->lo_state != Lo_bound)
1516 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1519 if (lo->lo_queue->limits.logical_block_size != arg) {
1520 sync_blockdev(lo->lo_device);
1521 kill_bdev(lo->lo_device);
1524 blk_mq_freeze_queue(lo->lo_queue);
1526 /* kill_bdev should have truncated all the pages */
1527 if (lo->lo_queue->limits.logical_block_size != arg &&
1528 lo->lo_device->bd_inode->i_mapping->nrpages) {
1530 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1531 __func__, lo->lo_number, lo->lo_file_name,
1532 lo->lo_device->bd_inode->i_mapping->nrpages);
1536 blk_queue_logical_block_size(lo->lo_queue, arg);
1537 blk_queue_physical_block_size(lo->lo_queue, arg);
1538 blk_queue_io_min(lo->lo_queue, arg);
1539 loop_update_dio(lo);
1541 blk_mq_unfreeze_queue(lo->lo_queue);
1546 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1551 err = mutex_lock_killable(&loop_ctl_mutex);
1555 case LOOP_SET_CAPACITY:
1556 err = loop_set_capacity(lo);
1558 case LOOP_SET_DIRECT_IO:
1559 err = loop_set_dio(lo, arg);
1561 case LOOP_SET_BLOCK_SIZE:
1562 err = loop_set_block_size(lo, arg);
1565 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1567 mutex_unlock(&loop_ctl_mutex);
1571 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1572 unsigned int cmd, unsigned long arg)
1574 struct loop_device *lo = bdev->bd_disk->private_data;
1579 return loop_set_fd(lo, mode, bdev, arg);
1580 case LOOP_CHANGE_FD:
1581 return loop_change_fd(lo, bdev, arg);
1583 return loop_clr_fd(lo);
1584 case LOOP_SET_STATUS:
1586 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1587 err = loop_set_status_old(lo,
1588 (struct loop_info __user *)arg);
1591 case LOOP_GET_STATUS:
1592 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1593 case LOOP_SET_STATUS64:
1595 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1596 err = loop_set_status64(lo,
1597 (struct loop_info64 __user *) arg);
1600 case LOOP_GET_STATUS64:
1601 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1602 case LOOP_SET_CAPACITY:
1603 case LOOP_SET_DIRECT_IO:
1604 case LOOP_SET_BLOCK_SIZE:
1605 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1609 err = lo_simple_ioctl(lo, cmd, arg);
1616 #ifdef CONFIG_COMPAT
1617 struct compat_loop_info {
1618 compat_int_t lo_number; /* ioctl r/o */
1619 compat_dev_t lo_device; /* ioctl r/o */
1620 compat_ulong_t lo_inode; /* ioctl r/o */
1621 compat_dev_t lo_rdevice; /* ioctl r/o */
1622 compat_int_t lo_offset;
1623 compat_int_t lo_encrypt_type;
1624 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1625 compat_int_t lo_flags; /* ioctl r/o */
1626 char lo_name[LO_NAME_SIZE];
1627 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1628 compat_ulong_t lo_init[2];
1633 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1634 * - noinlined to reduce stack space usage in main part of driver
1637 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1638 struct loop_info64 *info64)
1640 struct compat_loop_info info;
1642 if (copy_from_user(&info, arg, sizeof(info)))
1645 memset(info64, 0, sizeof(*info64));
1646 info64->lo_number = info.lo_number;
1647 info64->lo_device = info.lo_device;
1648 info64->lo_inode = info.lo_inode;
1649 info64->lo_rdevice = info.lo_rdevice;
1650 info64->lo_offset = info.lo_offset;
1651 info64->lo_sizelimit = 0;
1652 info64->lo_encrypt_type = info.lo_encrypt_type;
1653 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1654 info64->lo_flags = info.lo_flags;
1655 info64->lo_init[0] = info.lo_init[0];
1656 info64->lo_init[1] = info.lo_init[1];
1657 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1658 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1660 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1661 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1666 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1667 * - noinlined to reduce stack space usage in main part of driver
1670 loop_info64_to_compat(const struct loop_info64 *info64,
1671 struct compat_loop_info __user *arg)
1673 struct compat_loop_info info;
1675 memset(&info, 0, sizeof(info));
1676 info.lo_number = info64->lo_number;
1677 info.lo_device = info64->lo_device;
1678 info.lo_inode = info64->lo_inode;
1679 info.lo_rdevice = info64->lo_rdevice;
1680 info.lo_offset = info64->lo_offset;
1681 info.lo_encrypt_type = info64->lo_encrypt_type;
1682 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1683 info.lo_flags = info64->lo_flags;
1684 info.lo_init[0] = info64->lo_init[0];
1685 info.lo_init[1] = info64->lo_init[1];
1686 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1687 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1689 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1690 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1692 /* error in case values were truncated */
1693 if (info.lo_device != info64->lo_device ||
1694 info.lo_rdevice != info64->lo_rdevice ||
1695 info.lo_inode != info64->lo_inode ||
1696 info.lo_offset != info64->lo_offset ||
1697 info.lo_init[0] != info64->lo_init[0] ||
1698 info.lo_init[1] != info64->lo_init[1])
1701 if (copy_to_user(arg, &info, sizeof(info)))
1707 loop_set_status_compat(struct loop_device *lo,
1708 const struct compat_loop_info __user *arg)
1710 struct loop_info64 info64;
1713 ret = loop_info64_from_compat(arg, &info64);
1716 return loop_set_status(lo, &info64);
1720 loop_get_status_compat(struct loop_device *lo,
1721 struct compat_loop_info __user *arg)
1723 struct loop_info64 info64;
1728 err = loop_get_status(lo, &info64);
1730 err = loop_info64_to_compat(&info64, arg);
1734 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1735 unsigned int cmd, unsigned long arg)
1737 struct loop_device *lo = bdev->bd_disk->private_data;
1741 case LOOP_SET_STATUS:
1742 err = loop_set_status_compat(lo,
1743 (const struct compat_loop_info __user *)arg);
1745 case LOOP_GET_STATUS:
1746 err = loop_get_status_compat(lo,
1747 (struct compat_loop_info __user *)arg);
1749 case LOOP_SET_CAPACITY:
1751 case LOOP_GET_STATUS64:
1752 case LOOP_SET_STATUS64:
1753 arg = (unsigned long) compat_ptr(arg);
1756 case LOOP_CHANGE_FD:
1757 case LOOP_SET_BLOCK_SIZE:
1758 err = lo_ioctl(bdev, mode, cmd, arg);
1768 static int lo_open(struct block_device *bdev, fmode_t mode)
1770 struct loop_device *lo;
1773 err = mutex_lock_killable(&loop_ctl_mutex);
1776 lo = bdev->bd_disk->private_data;
1782 atomic_inc(&lo->lo_refcnt);
1784 mutex_unlock(&loop_ctl_mutex);
1788 static void lo_release(struct gendisk *disk, fmode_t mode)
1790 struct loop_device *lo;
1792 mutex_lock(&loop_ctl_mutex);
1793 lo = disk->private_data;
1794 if (atomic_dec_return(&lo->lo_refcnt))
1797 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1798 if (lo->lo_state != Lo_bound)
1800 lo->lo_state = Lo_rundown;
1801 mutex_unlock(&loop_ctl_mutex);
1803 * In autoclear mode, stop the loop thread
1804 * and remove configuration after last close.
1806 __loop_clr_fd(lo, true);
1808 } else if (lo->lo_state == Lo_bound) {
1810 * Otherwise keep thread (if running) and config,
1811 * but flush possible ongoing bios in thread.
1813 blk_mq_freeze_queue(lo->lo_queue);
1814 blk_mq_unfreeze_queue(lo->lo_queue);
1818 mutex_unlock(&loop_ctl_mutex);
1821 static const struct block_device_operations lo_fops = {
1822 .owner = THIS_MODULE,
1824 .release = lo_release,
1826 #ifdef CONFIG_COMPAT
1827 .compat_ioctl = lo_compat_ioctl,
1832 * And now the modules code and kernel interface.
1834 static int max_loop;
1835 module_param(max_loop, int, 0444);
1836 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1837 module_param(max_part, int, 0444);
1838 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1839 MODULE_LICENSE("GPL");
1840 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1842 int loop_register_transfer(struct loop_func_table *funcs)
1844 unsigned int n = funcs->number;
1846 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1848 xfer_funcs[n] = funcs;
1852 static int unregister_transfer_cb(int id, void *ptr, void *data)
1854 struct loop_device *lo = ptr;
1855 struct loop_func_table *xfer = data;
1857 mutex_lock(&loop_ctl_mutex);
1858 if (lo->lo_encryption == xfer)
1859 loop_release_xfer(lo);
1860 mutex_unlock(&loop_ctl_mutex);
1864 int loop_unregister_transfer(int number)
1866 unsigned int n = number;
1867 struct loop_func_table *xfer;
1869 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1872 xfer_funcs[n] = NULL;
1873 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1877 EXPORT_SYMBOL(loop_register_transfer);
1878 EXPORT_SYMBOL(loop_unregister_transfer);
1880 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1881 const struct blk_mq_queue_data *bd)
1883 struct request *rq = bd->rq;
1884 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1885 struct loop_device *lo = rq->q->queuedata;
1887 blk_mq_start_request(rq);
1889 if (lo->lo_state != Lo_bound)
1890 return BLK_STS_IOERR;
1892 switch (req_op(rq)) {
1894 case REQ_OP_DISCARD:
1895 case REQ_OP_WRITE_ZEROES:
1896 cmd->use_aio = false;
1899 cmd->use_aio = lo->use_dio;
1903 /* always use the first bio's css */
1904 #ifdef CONFIG_BLK_CGROUP
1905 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1906 cmd->css = &bio_blkcg(rq->bio)->css;
1911 kthread_queue_work(&lo->worker, &cmd->work);
1916 static void loop_handle_cmd(struct loop_cmd *cmd)
1918 struct request *rq = blk_mq_rq_from_pdu(cmd);
1919 const bool write = op_is_write(req_op(rq));
1920 struct loop_device *lo = rq->q->queuedata;
1923 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1928 ret = do_req_filebacked(lo, rq);
1930 /* complete non-aio request */
1931 if (!cmd->use_aio || ret) {
1932 cmd->ret = ret ? -EIO : 0;
1933 blk_mq_complete_request(rq);
1937 static void loop_queue_work(struct kthread_work *work)
1939 struct loop_cmd *cmd =
1940 container_of(work, struct loop_cmd, work);
1942 loop_handle_cmd(cmd);
1945 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1946 unsigned int hctx_idx, unsigned int numa_node)
1948 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1950 kthread_init_work(&cmd->work, loop_queue_work);
1954 static const struct blk_mq_ops loop_mq_ops = {
1955 .queue_rq = loop_queue_rq,
1956 .init_request = loop_init_request,
1957 .complete = lo_complete_rq,
1960 static int loop_add(struct loop_device **l, int i)
1962 struct loop_device *lo;
1963 struct gendisk *disk;
1967 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1971 lo->lo_state = Lo_unbound;
1973 /* allocate id, if @id >= 0, we're requesting that specific id */
1975 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1979 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1986 lo->tag_set.ops = &loop_mq_ops;
1987 lo->tag_set.nr_hw_queues = 1;
1988 lo->tag_set.queue_depth = 128;
1989 lo->tag_set.numa_node = NUMA_NO_NODE;
1990 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1991 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1992 lo->tag_set.driver_data = lo;
1994 err = blk_mq_alloc_tag_set(&lo->tag_set);
1998 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1999 if (IS_ERR(lo->lo_queue)) {
2000 err = PTR_ERR(lo->lo_queue);
2001 goto out_cleanup_tags;
2003 lo->lo_queue->queuedata = lo;
2005 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2008 * By default, we do buffer IO, so it doesn't make sense to enable
2009 * merge because the I/O submitted to backing file is handled page by
2010 * page. For directio mode, merge does help to dispatch bigger request
2011 * to underlayer disk. We will enable merge once directio is enabled.
2013 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2016 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2018 goto out_free_queue;
2021 * Disable partition scanning by default. The in-kernel partition
2022 * scanning can be requested individually per-device during its
2023 * setup. Userspace can always add and remove partitions from all
2024 * devices. The needed partition minors are allocated from the
2025 * extended minor space, the main loop device numbers will continue
2026 * to match the loop minors, regardless of the number of partitions
2029 * If max_part is given, partition scanning is globally enabled for
2030 * all loop devices. The minors for the main loop devices will be
2031 * multiples of max_part.
2033 * Note: Global-for-all-devices, set-only-at-init, read-only module
2034 * parameteters like 'max_loop' and 'max_part' make things needlessly
2035 * complicated, are too static, inflexible and may surprise
2036 * userspace tools. Parameters like this in general should be avoided.
2039 disk->flags |= GENHD_FL_NO_PART_SCAN;
2040 disk->flags |= GENHD_FL_EXT_DEVT;
2041 atomic_set(&lo->lo_refcnt, 0);
2043 spin_lock_init(&lo->lo_lock);
2044 disk->major = LOOP_MAJOR;
2045 disk->first_minor = i << part_shift;
2046 disk->fops = &lo_fops;
2047 disk->private_data = lo;
2048 disk->queue = lo->lo_queue;
2049 sprintf(disk->disk_name, "loop%d", i);
2052 return lo->lo_number;
2055 blk_cleanup_queue(lo->lo_queue);
2057 blk_mq_free_tag_set(&lo->tag_set);
2059 idr_remove(&loop_index_idr, i);
2066 static void loop_remove(struct loop_device *lo)
2068 del_gendisk(lo->lo_disk);
2069 blk_cleanup_queue(lo->lo_queue);
2070 blk_mq_free_tag_set(&lo->tag_set);
2071 put_disk(lo->lo_disk);
2075 static int find_free_cb(int id, void *ptr, void *data)
2077 struct loop_device *lo = ptr;
2078 struct loop_device **l = data;
2080 if (lo->lo_state == Lo_unbound) {
2087 static int loop_lookup(struct loop_device **l, int i)
2089 struct loop_device *lo;
2095 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2098 ret = lo->lo_number;
2103 /* lookup and return a specific i */
2104 lo = idr_find(&loop_index_idr, i);
2107 ret = lo->lo_number;
2113 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2115 struct loop_device *lo;
2116 struct kobject *kobj;
2119 mutex_lock(&loop_ctl_mutex);
2120 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2122 err = loop_add(&lo, MINOR(dev) >> part_shift);
2126 kobj = get_disk_and_module(lo->lo_disk);
2127 mutex_unlock(&loop_ctl_mutex);
2133 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2136 struct loop_device *lo;
2139 ret = mutex_lock_killable(&loop_ctl_mutex);
2146 ret = loop_lookup(&lo, parm);
2151 ret = loop_add(&lo, parm);
2153 case LOOP_CTL_REMOVE:
2154 ret = loop_lookup(&lo, parm);
2157 if (lo->lo_state != Lo_unbound) {
2161 if (atomic_read(&lo->lo_refcnt) > 0) {
2165 lo->lo_disk->private_data = NULL;
2166 idr_remove(&loop_index_idr, lo->lo_number);
2169 case LOOP_CTL_GET_FREE:
2170 ret = loop_lookup(&lo, -1);
2173 ret = loop_add(&lo, -1);
2175 mutex_unlock(&loop_ctl_mutex);
2180 static const struct file_operations loop_ctl_fops = {
2181 .open = nonseekable_open,
2182 .unlocked_ioctl = loop_control_ioctl,
2183 .compat_ioctl = loop_control_ioctl,
2184 .owner = THIS_MODULE,
2185 .llseek = noop_llseek,
2188 static struct miscdevice loop_misc = {
2189 .minor = LOOP_CTRL_MINOR,
2190 .name = "loop-control",
2191 .fops = &loop_ctl_fops,
2194 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2195 MODULE_ALIAS("devname:loop-control");
2197 static int __init loop_init(void)
2200 unsigned long range;
2201 struct loop_device *lo;
2206 part_shift = fls(max_part);
2209 * Adjust max_part according to part_shift as it is exported
2210 * to user space so that user can decide correct minor number
2211 * if [s]he want to create more devices.
2213 * Note that -1 is required because partition 0 is reserved
2214 * for the whole disk.
2216 max_part = (1UL << part_shift) - 1;
2219 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2224 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2230 * If max_loop is specified, create that many devices upfront.
2231 * This also becomes a hard limit. If max_loop is not specified,
2232 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2233 * init time. Loop devices can be requested on-demand with the
2234 * /dev/loop-control interface, or be instantiated by accessing
2235 * a 'dead' device node.
2239 range = max_loop << part_shift;
2241 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2242 range = 1UL << MINORBITS;
2245 err = misc_register(&loop_misc);
2250 if (register_blkdev(LOOP_MAJOR, "loop")) {
2255 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2256 THIS_MODULE, loop_probe, NULL, NULL);
2258 /* pre-create number of devices given by config or max_loop */
2259 mutex_lock(&loop_ctl_mutex);
2260 for (i = 0; i < nr; i++)
2262 mutex_unlock(&loop_ctl_mutex);
2264 printk(KERN_INFO "loop: module loaded\n");
2268 misc_deregister(&loop_misc);
2273 static int loop_exit_cb(int id, void *ptr, void *data)
2275 struct loop_device *lo = ptr;
2281 static void __exit loop_exit(void)
2283 unsigned long range;
2285 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2287 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2288 idr_destroy(&loop_index_idr);
2290 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2291 unregister_blkdev(LOOP_MAJOR, "loop");
2293 misc_deregister(&loop_misc);
2296 module_init(loop_init);
2297 module_exit(loop_exit);
2300 static int __init max_loop_setup(char *str)
2302 max_loop = simple_strtol(str, NULL, 0);
2306 __setup("max_loop=", max_loop_setup);