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_unbound) {
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;
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;
914 /* This is safe, since we have a reference from open(). */
915 __module_get(THIS_MODULE);
922 error = mutex_lock_killable(&loop_ctl_mutex);
927 if (lo->lo_state != Lo_unbound)
930 error = loop_validate_file(file, bdev);
934 mapping = file->f_mapping;
935 inode = mapping->host;
937 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
938 !file->f_op->write_iter)
939 lo_flags |= LO_FLAGS_READ_ONLY;
942 size = get_loop_size(lo, file);
943 if ((loff_t)(sector_t)size != size)
945 error = loop_prepare_queue(lo);
951 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
954 lo->lo_device = bdev;
955 lo->lo_flags = lo_flags;
956 lo->lo_backing_file = file;
959 lo->lo_sizelimit = 0;
960 lo->old_gfp_mask = mapping_gfp_mask(mapping);
961 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
963 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
964 blk_queue_write_cache(lo->lo_queue, true, false);
967 set_capacity(lo->lo_disk, size);
968 bd_set_size(bdev, size << 9);
970 /* let user-space know about the new size */
971 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
973 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
974 block_size(inode->i_bdev) : PAGE_SIZE);
976 lo->lo_state = Lo_bound;
978 lo->lo_flags |= LO_FLAGS_PARTSCAN;
979 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
981 /* Grab the block_device to prevent its destruction after we
982 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
985 mutex_unlock(&loop_ctl_mutex);
987 loop_reread_partitions(lo, bdev);
991 mutex_unlock(&loop_ctl_mutex);
995 /* This is safe: open() is still holding a reference. */
996 module_put(THIS_MODULE);
1001 loop_release_xfer(struct loop_device *lo)
1004 struct loop_func_table *xfer = lo->lo_encryption;
1008 err = xfer->release(lo);
1009 lo->transfer = NULL;
1010 lo->lo_encryption = NULL;
1011 module_put(xfer->owner);
1017 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1018 const struct loop_info64 *i)
1023 struct module *owner = xfer->owner;
1025 if (!try_module_get(owner))
1028 err = xfer->init(lo, i);
1032 lo->lo_encryption = xfer;
1037 static int __loop_clr_fd(struct loop_device *lo, bool release)
1039 struct file *filp = NULL;
1040 gfp_t gfp = lo->old_gfp_mask;
1041 struct block_device *bdev = lo->lo_device;
1043 bool partscan = false;
1046 mutex_lock(&loop_ctl_mutex);
1047 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1052 filp = lo->lo_backing_file;
1058 /* freeze request queue during the transition */
1059 blk_mq_freeze_queue(lo->lo_queue);
1061 spin_lock_irq(&lo->lo_lock);
1062 lo->lo_backing_file = NULL;
1063 spin_unlock_irq(&lo->lo_lock);
1065 loop_release_xfer(lo);
1066 lo->transfer = NULL;
1068 lo->lo_device = NULL;
1069 lo->lo_encryption = NULL;
1071 lo->lo_sizelimit = 0;
1072 lo->lo_encrypt_key_size = 0;
1073 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1074 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1075 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1076 blk_queue_logical_block_size(lo->lo_queue, 512);
1077 blk_queue_physical_block_size(lo->lo_queue, 512);
1078 blk_queue_io_min(lo->lo_queue, 512);
1081 invalidate_bdev(bdev);
1082 bdev->bd_inode->i_mapping->wb_err = 0;
1084 set_capacity(lo->lo_disk, 0);
1085 loop_sysfs_exit(lo);
1087 bd_set_size(bdev, 0);
1088 /* let user-space know about this change */
1089 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1091 mapping_set_gfp_mask(filp->f_mapping, gfp);
1092 lo->lo_state = Lo_unbound;
1093 /* This is safe: open() is still holding a reference. */
1094 module_put(THIS_MODULE);
1095 blk_mq_unfreeze_queue(lo->lo_queue);
1097 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1098 lo_number = lo->lo_number;
1101 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1102 loop_unprepare_queue(lo);
1104 mutex_unlock(&loop_ctl_mutex);
1107 * bd_mutex has been held already in release path, so don't
1108 * acquire it if this function is called in such case.
1110 * If the reread partition isn't from release path, lo_refcnt
1111 * must be at least one and it can only become zero when the
1112 * current holder is released.
1115 err = __blkdev_reread_part(bdev);
1117 err = blkdev_reread_part(bdev);
1119 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1120 __func__, lo_number, err);
1121 /* Device is gone, no point in returning error */
1125 * Need not hold loop_ctl_mutex to fput backing file.
1126 * Calling fput holding loop_ctl_mutex triggers a circular
1127 * lock dependency possibility warning as fput can take
1128 * bd_mutex which is usually taken before loop_ctl_mutex.
1135 static int loop_clr_fd(struct loop_device *lo)
1139 err = mutex_lock_killable(&loop_ctl_mutex);
1142 if (lo->lo_state != Lo_bound) {
1143 mutex_unlock(&loop_ctl_mutex);
1147 * If we've explicitly asked to tear down the loop device,
1148 * and it has an elevated reference count, set it for auto-teardown when
1149 * the last reference goes away. This stops $!~#$@ udev from
1150 * preventing teardown because it decided that it needs to run blkid on
1151 * the loopback device whenever they appear. xfstests is notorious for
1152 * failing tests because blkid via udev races with a losetup
1153 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1154 * command to fail with EBUSY.
1156 if (atomic_read(&lo->lo_refcnt) > 1) {
1157 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1158 mutex_unlock(&loop_ctl_mutex);
1161 lo->lo_state = Lo_rundown;
1162 mutex_unlock(&loop_ctl_mutex);
1164 return __loop_clr_fd(lo, false);
1168 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1171 struct loop_func_table *xfer;
1172 kuid_t uid = current_uid();
1173 struct block_device *bdev;
1174 bool partscan = false;
1176 err = mutex_lock_killable(&loop_ctl_mutex);
1179 if (lo->lo_encrypt_key_size &&
1180 !uid_eq(lo->lo_key_owner, uid) &&
1181 !capable(CAP_SYS_ADMIN)) {
1185 if (lo->lo_state != Lo_bound) {
1189 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1194 if (lo->lo_offset != info->lo_offset ||
1195 lo->lo_sizelimit != info->lo_sizelimit) {
1196 sync_blockdev(lo->lo_device);
1197 kill_bdev(lo->lo_device);
1200 /* I/O need to be drained during transfer transition */
1201 blk_mq_freeze_queue(lo->lo_queue);
1203 err = loop_release_xfer(lo);
1207 if (info->lo_encrypt_type) {
1208 unsigned int type = info->lo_encrypt_type;
1210 if (type >= MAX_LO_CRYPT) {
1214 xfer = xfer_funcs[type];
1222 err = loop_init_xfer(lo, xfer, info);
1226 if (lo->lo_offset != info->lo_offset ||
1227 lo->lo_sizelimit != info->lo_sizelimit) {
1228 /* kill_bdev should have truncated all the pages */
1229 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1231 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1232 __func__, lo->lo_number, lo->lo_file_name,
1233 lo->lo_device->bd_inode->i_mapping->nrpages);
1236 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1242 loop_config_discard(lo);
1244 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1245 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1246 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1247 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1251 lo->transfer = xfer->transfer;
1252 lo->ioctl = xfer->ioctl;
1254 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1255 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1256 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1258 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1259 lo->lo_init[0] = info->lo_init[0];
1260 lo->lo_init[1] = info->lo_init[1];
1261 if (info->lo_encrypt_key_size) {
1262 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1263 info->lo_encrypt_key_size);
1264 lo->lo_key_owner = uid;
1267 /* update dio if lo_offset or transfer is changed */
1268 __loop_update_dio(lo, lo->use_dio);
1271 blk_mq_unfreeze_queue(lo->lo_queue);
1273 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1274 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1275 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1276 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1277 bdev = lo->lo_device;
1281 mutex_unlock(&loop_ctl_mutex);
1283 loop_reread_partitions(lo, bdev);
1289 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1295 ret = mutex_lock_killable(&loop_ctl_mutex);
1298 if (lo->lo_state != Lo_bound) {
1299 mutex_unlock(&loop_ctl_mutex);
1303 memset(info, 0, sizeof(*info));
1304 info->lo_number = lo->lo_number;
1305 info->lo_offset = lo->lo_offset;
1306 info->lo_sizelimit = lo->lo_sizelimit;
1307 info->lo_flags = lo->lo_flags;
1308 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1309 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1310 info->lo_encrypt_type =
1311 lo->lo_encryption ? lo->lo_encryption->number : 0;
1312 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1313 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1314 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1315 lo->lo_encrypt_key_size);
1318 /* Drop loop_ctl_mutex while we call into the filesystem. */
1319 path = lo->lo_backing_file->f_path;
1321 mutex_unlock(&loop_ctl_mutex);
1322 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1324 info->lo_device = huge_encode_dev(stat.dev);
1325 info->lo_inode = stat.ino;
1326 info->lo_rdevice = huge_encode_dev(stat.rdev);
1333 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1335 memset(info64, 0, sizeof(*info64));
1336 info64->lo_number = info->lo_number;
1337 info64->lo_device = info->lo_device;
1338 info64->lo_inode = info->lo_inode;
1339 info64->lo_rdevice = info->lo_rdevice;
1340 info64->lo_offset = info->lo_offset;
1341 info64->lo_sizelimit = 0;
1342 info64->lo_encrypt_type = info->lo_encrypt_type;
1343 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1344 info64->lo_flags = info->lo_flags;
1345 info64->lo_init[0] = info->lo_init[0];
1346 info64->lo_init[1] = info->lo_init[1];
1347 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1348 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1350 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1351 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1355 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1357 memset(info, 0, sizeof(*info));
1358 info->lo_number = info64->lo_number;
1359 info->lo_device = info64->lo_device;
1360 info->lo_inode = info64->lo_inode;
1361 info->lo_rdevice = info64->lo_rdevice;
1362 info->lo_offset = info64->lo_offset;
1363 info->lo_encrypt_type = info64->lo_encrypt_type;
1364 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1365 info->lo_flags = info64->lo_flags;
1366 info->lo_init[0] = info64->lo_init[0];
1367 info->lo_init[1] = info64->lo_init[1];
1368 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1369 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1371 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1372 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1374 /* error in case values were truncated */
1375 if (info->lo_device != info64->lo_device ||
1376 info->lo_rdevice != info64->lo_rdevice ||
1377 info->lo_inode != info64->lo_inode ||
1378 info->lo_offset != info64->lo_offset)
1385 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1387 struct loop_info info;
1388 struct loop_info64 info64;
1390 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1392 loop_info64_from_old(&info, &info64);
1393 return loop_set_status(lo, &info64);
1397 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1399 struct loop_info64 info64;
1401 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1403 return loop_set_status(lo, &info64);
1407 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1408 struct loop_info info;
1409 struct loop_info64 info64;
1414 err = loop_get_status(lo, &info64);
1416 err = loop_info64_to_old(&info64, &info);
1417 if (!err && copy_to_user(arg, &info, sizeof(info)))
1424 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1425 struct loop_info64 info64;
1430 err = loop_get_status(lo, &info64);
1431 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1437 static int loop_set_capacity(struct loop_device *lo)
1439 if (unlikely(lo->lo_state != Lo_bound))
1442 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1445 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1448 if (lo->lo_state != Lo_bound)
1451 __loop_update_dio(lo, !!arg);
1452 if (lo->use_dio == !!arg)
1459 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1463 if (lo->lo_state != Lo_bound)
1466 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1469 if (lo->lo_queue->limits.logical_block_size != arg) {
1470 sync_blockdev(lo->lo_device);
1471 kill_bdev(lo->lo_device);
1474 blk_mq_freeze_queue(lo->lo_queue);
1476 /* kill_bdev should have truncated all the pages */
1477 if (lo->lo_queue->limits.logical_block_size != arg &&
1478 lo->lo_device->bd_inode->i_mapping->nrpages) {
1480 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1481 __func__, lo->lo_number, lo->lo_file_name,
1482 lo->lo_device->bd_inode->i_mapping->nrpages);
1486 blk_queue_logical_block_size(lo->lo_queue, arg);
1487 blk_queue_physical_block_size(lo->lo_queue, arg);
1488 blk_queue_io_min(lo->lo_queue, arg);
1489 loop_update_dio(lo);
1491 blk_mq_unfreeze_queue(lo->lo_queue);
1496 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1501 err = mutex_lock_killable(&loop_ctl_mutex);
1505 case LOOP_SET_CAPACITY:
1506 err = loop_set_capacity(lo);
1508 case LOOP_SET_DIRECT_IO:
1509 err = loop_set_dio(lo, arg);
1511 case LOOP_SET_BLOCK_SIZE:
1512 err = loop_set_block_size(lo, arg);
1515 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1517 mutex_unlock(&loop_ctl_mutex);
1521 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1522 unsigned int cmd, unsigned long arg)
1524 struct loop_device *lo = bdev->bd_disk->private_data;
1529 return loop_set_fd(lo, mode, bdev, arg);
1530 case LOOP_CHANGE_FD:
1531 return loop_change_fd(lo, bdev, arg);
1533 return loop_clr_fd(lo);
1534 case LOOP_SET_STATUS:
1536 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1537 err = loop_set_status_old(lo,
1538 (struct loop_info __user *)arg);
1541 case LOOP_GET_STATUS:
1542 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1543 case LOOP_SET_STATUS64:
1545 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1546 err = loop_set_status64(lo,
1547 (struct loop_info64 __user *) arg);
1550 case LOOP_GET_STATUS64:
1551 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1552 case LOOP_SET_CAPACITY:
1553 case LOOP_SET_DIRECT_IO:
1554 case LOOP_SET_BLOCK_SIZE:
1555 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1559 err = lo_simple_ioctl(lo, cmd, arg);
1566 #ifdef CONFIG_COMPAT
1567 struct compat_loop_info {
1568 compat_int_t lo_number; /* ioctl r/o */
1569 compat_dev_t lo_device; /* ioctl r/o */
1570 compat_ulong_t lo_inode; /* ioctl r/o */
1571 compat_dev_t lo_rdevice; /* ioctl r/o */
1572 compat_int_t lo_offset;
1573 compat_int_t lo_encrypt_type;
1574 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1575 compat_int_t lo_flags; /* ioctl r/o */
1576 char lo_name[LO_NAME_SIZE];
1577 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1578 compat_ulong_t lo_init[2];
1583 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1584 * - noinlined to reduce stack space usage in main part of driver
1587 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1588 struct loop_info64 *info64)
1590 struct compat_loop_info info;
1592 if (copy_from_user(&info, arg, sizeof(info)))
1595 memset(info64, 0, sizeof(*info64));
1596 info64->lo_number = info.lo_number;
1597 info64->lo_device = info.lo_device;
1598 info64->lo_inode = info.lo_inode;
1599 info64->lo_rdevice = info.lo_rdevice;
1600 info64->lo_offset = info.lo_offset;
1601 info64->lo_sizelimit = 0;
1602 info64->lo_encrypt_type = info.lo_encrypt_type;
1603 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1604 info64->lo_flags = info.lo_flags;
1605 info64->lo_init[0] = info.lo_init[0];
1606 info64->lo_init[1] = info.lo_init[1];
1607 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1608 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1610 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1611 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1616 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1617 * - noinlined to reduce stack space usage in main part of driver
1620 loop_info64_to_compat(const struct loop_info64 *info64,
1621 struct compat_loop_info __user *arg)
1623 struct compat_loop_info info;
1625 memset(&info, 0, sizeof(info));
1626 info.lo_number = info64->lo_number;
1627 info.lo_device = info64->lo_device;
1628 info.lo_inode = info64->lo_inode;
1629 info.lo_rdevice = info64->lo_rdevice;
1630 info.lo_offset = info64->lo_offset;
1631 info.lo_encrypt_type = info64->lo_encrypt_type;
1632 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1633 info.lo_flags = info64->lo_flags;
1634 info.lo_init[0] = info64->lo_init[0];
1635 info.lo_init[1] = info64->lo_init[1];
1636 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1637 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1639 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1640 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1642 /* error in case values were truncated */
1643 if (info.lo_device != info64->lo_device ||
1644 info.lo_rdevice != info64->lo_rdevice ||
1645 info.lo_inode != info64->lo_inode ||
1646 info.lo_offset != info64->lo_offset ||
1647 info.lo_init[0] != info64->lo_init[0] ||
1648 info.lo_init[1] != info64->lo_init[1])
1651 if (copy_to_user(arg, &info, sizeof(info)))
1657 loop_set_status_compat(struct loop_device *lo,
1658 const struct compat_loop_info __user *arg)
1660 struct loop_info64 info64;
1663 ret = loop_info64_from_compat(arg, &info64);
1666 return loop_set_status(lo, &info64);
1670 loop_get_status_compat(struct loop_device *lo,
1671 struct compat_loop_info __user *arg)
1673 struct loop_info64 info64;
1678 err = loop_get_status(lo, &info64);
1680 err = loop_info64_to_compat(&info64, arg);
1684 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1685 unsigned int cmd, unsigned long arg)
1687 struct loop_device *lo = bdev->bd_disk->private_data;
1691 case LOOP_SET_STATUS:
1692 err = loop_set_status_compat(lo,
1693 (const struct compat_loop_info __user *)arg);
1695 case LOOP_GET_STATUS:
1696 err = loop_get_status_compat(lo,
1697 (struct compat_loop_info __user *)arg);
1699 case LOOP_SET_CAPACITY:
1701 case LOOP_GET_STATUS64:
1702 case LOOP_SET_STATUS64:
1703 arg = (unsigned long) compat_ptr(arg);
1706 case LOOP_CHANGE_FD:
1707 case LOOP_SET_BLOCK_SIZE:
1708 err = lo_ioctl(bdev, mode, cmd, arg);
1718 static int lo_open(struct block_device *bdev, fmode_t mode)
1720 struct loop_device *lo;
1723 err = mutex_lock_killable(&loop_ctl_mutex);
1726 lo = bdev->bd_disk->private_data;
1732 atomic_inc(&lo->lo_refcnt);
1734 mutex_unlock(&loop_ctl_mutex);
1738 static void lo_release(struct gendisk *disk, fmode_t mode)
1740 struct loop_device *lo;
1742 mutex_lock(&loop_ctl_mutex);
1743 lo = disk->private_data;
1744 if (atomic_dec_return(&lo->lo_refcnt))
1747 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1748 if (lo->lo_state != Lo_bound)
1750 lo->lo_state = Lo_rundown;
1751 mutex_unlock(&loop_ctl_mutex);
1753 * In autoclear mode, stop the loop thread
1754 * and remove configuration after last close.
1756 __loop_clr_fd(lo, true);
1758 } else if (lo->lo_state == Lo_bound) {
1760 * Otherwise keep thread (if running) and config,
1761 * but flush possible ongoing bios in thread.
1763 blk_mq_freeze_queue(lo->lo_queue);
1764 blk_mq_unfreeze_queue(lo->lo_queue);
1768 mutex_unlock(&loop_ctl_mutex);
1771 static const struct block_device_operations lo_fops = {
1772 .owner = THIS_MODULE,
1774 .release = lo_release,
1776 #ifdef CONFIG_COMPAT
1777 .compat_ioctl = lo_compat_ioctl,
1782 * And now the modules code and kernel interface.
1784 static int max_loop;
1785 module_param(max_loop, int, 0444);
1786 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1787 module_param(max_part, int, 0444);
1788 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1789 MODULE_LICENSE("GPL");
1790 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1792 int loop_register_transfer(struct loop_func_table *funcs)
1794 unsigned int n = funcs->number;
1796 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1798 xfer_funcs[n] = funcs;
1802 static int unregister_transfer_cb(int id, void *ptr, void *data)
1804 struct loop_device *lo = ptr;
1805 struct loop_func_table *xfer = data;
1807 mutex_lock(&loop_ctl_mutex);
1808 if (lo->lo_encryption == xfer)
1809 loop_release_xfer(lo);
1810 mutex_unlock(&loop_ctl_mutex);
1814 int loop_unregister_transfer(int number)
1816 unsigned int n = number;
1817 struct loop_func_table *xfer;
1819 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1822 xfer_funcs[n] = NULL;
1823 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1827 EXPORT_SYMBOL(loop_register_transfer);
1828 EXPORT_SYMBOL(loop_unregister_transfer);
1830 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1831 const struct blk_mq_queue_data *bd)
1833 struct request *rq = bd->rq;
1834 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1835 struct loop_device *lo = rq->q->queuedata;
1837 blk_mq_start_request(rq);
1839 if (lo->lo_state != Lo_bound)
1840 return BLK_STS_IOERR;
1842 switch (req_op(rq)) {
1844 case REQ_OP_DISCARD:
1845 case REQ_OP_WRITE_ZEROES:
1846 cmd->use_aio = false;
1849 cmd->use_aio = lo->use_dio;
1853 /* always use the first bio's css */
1854 #ifdef CONFIG_BLK_CGROUP
1855 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1856 cmd->css = &bio_blkcg(rq->bio)->css;
1861 kthread_queue_work(&lo->worker, &cmd->work);
1866 static void loop_handle_cmd(struct loop_cmd *cmd)
1868 struct request *rq = blk_mq_rq_from_pdu(cmd);
1869 const bool write = op_is_write(req_op(rq));
1870 struct loop_device *lo = rq->q->queuedata;
1873 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1878 ret = do_req_filebacked(lo, rq);
1880 /* complete non-aio request */
1881 if (!cmd->use_aio || ret) {
1882 cmd->ret = ret ? -EIO : 0;
1883 blk_mq_complete_request(rq);
1887 static void loop_queue_work(struct kthread_work *work)
1889 struct loop_cmd *cmd =
1890 container_of(work, struct loop_cmd, work);
1892 loop_handle_cmd(cmd);
1895 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1896 unsigned int hctx_idx, unsigned int numa_node)
1898 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1900 kthread_init_work(&cmd->work, loop_queue_work);
1904 static const struct blk_mq_ops loop_mq_ops = {
1905 .queue_rq = loop_queue_rq,
1906 .init_request = loop_init_request,
1907 .complete = lo_complete_rq,
1910 static int loop_add(struct loop_device **l, int i)
1912 struct loop_device *lo;
1913 struct gendisk *disk;
1917 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1921 lo->lo_state = Lo_unbound;
1923 /* allocate id, if @id >= 0, we're requesting that specific id */
1925 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1929 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1936 lo->tag_set.ops = &loop_mq_ops;
1937 lo->tag_set.nr_hw_queues = 1;
1938 lo->tag_set.queue_depth = 128;
1939 lo->tag_set.numa_node = NUMA_NO_NODE;
1940 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1941 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1942 lo->tag_set.driver_data = lo;
1944 err = blk_mq_alloc_tag_set(&lo->tag_set);
1948 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1949 if (IS_ERR(lo->lo_queue)) {
1950 err = PTR_ERR(lo->lo_queue);
1951 goto out_cleanup_tags;
1953 lo->lo_queue->queuedata = lo;
1955 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1958 * By default, we do buffer IO, so it doesn't make sense to enable
1959 * merge because the I/O submitted to backing file is handled page by
1960 * page. For directio mode, merge does help to dispatch bigger request
1961 * to underlayer disk. We will enable merge once directio is enabled.
1963 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1966 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1968 goto out_free_queue;
1971 * Disable partition scanning by default. The in-kernel partition
1972 * scanning can be requested individually per-device during its
1973 * setup. Userspace can always add and remove partitions from all
1974 * devices. The needed partition minors are allocated from the
1975 * extended minor space, the main loop device numbers will continue
1976 * to match the loop minors, regardless of the number of partitions
1979 * If max_part is given, partition scanning is globally enabled for
1980 * all loop devices. The minors for the main loop devices will be
1981 * multiples of max_part.
1983 * Note: Global-for-all-devices, set-only-at-init, read-only module
1984 * parameteters like 'max_loop' and 'max_part' make things needlessly
1985 * complicated, are too static, inflexible and may surprise
1986 * userspace tools. Parameters like this in general should be avoided.
1989 disk->flags |= GENHD_FL_NO_PART_SCAN;
1990 disk->flags |= GENHD_FL_EXT_DEVT;
1991 atomic_set(&lo->lo_refcnt, 0);
1993 spin_lock_init(&lo->lo_lock);
1994 disk->major = LOOP_MAJOR;
1995 disk->first_minor = i << part_shift;
1996 disk->fops = &lo_fops;
1997 disk->private_data = lo;
1998 disk->queue = lo->lo_queue;
1999 sprintf(disk->disk_name, "loop%d", i);
2002 return lo->lo_number;
2005 blk_cleanup_queue(lo->lo_queue);
2007 blk_mq_free_tag_set(&lo->tag_set);
2009 idr_remove(&loop_index_idr, i);
2016 static void loop_remove(struct loop_device *lo)
2018 del_gendisk(lo->lo_disk);
2019 blk_cleanup_queue(lo->lo_queue);
2020 blk_mq_free_tag_set(&lo->tag_set);
2021 put_disk(lo->lo_disk);
2025 static int find_free_cb(int id, void *ptr, void *data)
2027 struct loop_device *lo = ptr;
2028 struct loop_device **l = data;
2030 if (lo->lo_state == Lo_unbound) {
2037 static int loop_lookup(struct loop_device **l, int i)
2039 struct loop_device *lo;
2045 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2048 ret = lo->lo_number;
2053 /* lookup and return a specific i */
2054 lo = idr_find(&loop_index_idr, i);
2057 ret = lo->lo_number;
2063 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2065 struct loop_device *lo;
2066 struct kobject *kobj;
2069 mutex_lock(&loop_ctl_mutex);
2070 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2072 err = loop_add(&lo, MINOR(dev) >> part_shift);
2076 kobj = get_disk_and_module(lo->lo_disk);
2077 mutex_unlock(&loop_ctl_mutex);
2083 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2086 struct loop_device *lo;
2089 ret = mutex_lock_killable(&loop_ctl_mutex);
2096 ret = loop_lookup(&lo, parm);
2101 ret = loop_add(&lo, parm);
2103 case LOOP_CTL_REMOVE:
2104 ret = loop_lookup(&lo, parm);
2107 if (lo->lo_state != Lo_unbound) {
2111 if (atomic_read(&lo->lo_refcnt) > 0) {
2115 lo->lo_disk->private_data = NULL;
2116 idr_remove(&loop_index_idr, lo->lo_number);
2119 case LOOP_CTL_GET_FREE:
2120 ret = loop_lookup(&lo, -1);
2123 ret = loop_add(&lo, -1);
2125 mutex_unlock(&loop_ctl_mutex);
2130 static const struct file_operations loop_ctl_fops = {
2131 .open = nonseekable_open,
2132 .unlocked_ioctl = loop_control_ioctl,
2133 .compat_ioctl = loop_control_ioctl,
2134 .owner = THIS_MODULE,
2135 .llseek = noop_llseek,
2138 static struct miscdevice loop_misc = {
2139 .minor = LOOP_CTRL_MINOR,
2140 .name = "loop-control",
2141 .fops = &loop_ctl_fops,
2144 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2145 MODULE_ALIAS("devname:loop-control");
2147 static int __init loop_init(void)
2150 unsigned long range;
2151 struct loop_device *lo;
2156 part_shift = fls(max_part);
2159 * Adjust max_part according to part_shift as it is exported
2160 * to user space so that user can decide correct minor number
2161 * if [s]he want to create more devices.
2163 * Note that -1 is required because partition 0 is reserved
2164 * for the whole disk.
2166 max_part = (1UL << part_shift) - 1;
2169 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2174 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2180 * If max_loop is specified, create that many devices upfront.
2181 * This also becomes a hard limit. If max_loop is not specified,
2182 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2183 * init time. Loop devices can be requested on-demand with the
2184 * /dev/loop-control interface, or be instantiated by accessing
2185 * a 'dead' device node.
2189 range = max_loop << part_shift;
2191 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2192 range = 1UL << MINORBITS;
2195 err = misc_register(&loop_misc);
2200 if (register_blkdev(LOOP_MAJOR, "loop")) {
2205 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2206 THIS_MODULE, loop_probe, NULL, NULL);
2208 /* pre-create number of devices given by config or max_loop */
2209 mutex_lock(&loop_ctl_mutex);
2210 for (i = 0; i < nr; i++)
2212 mutex_unlock(&loop_ctl_mutex);
2214 printk(KERN_INFO "loop: module loaded\n");
2218 misc_deregister(&loop_misc);
2223 static int loop_exit_cb(int id, void *ptr, void *data)
2225 struct loop_device *lo = ptr;
2231 static void __exit loop_exit(void)
2233 unsigned long range;
2235 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2237 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2238 idr_destroy(&loop_index_idr);
2240 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2241 unregister_blkdev(LOOP_MAJOR, "loop");
2243 misc_deregister(&loop_misc);
2246 module_init(loop_init);
2247 module_exit(loop_exit);
2250 static int __init max_loop_setup(char *str)
2252 max_loop = simple_strtol(str, NULL, 0);
2256 __setup("max_loop=", max_loop_setup);