2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
79 #include <linux/ioprio.h>
83 #include <linux/uaccess.h>
85 static DEFINE_IDR(loop_index_idr);
86 static DEFINE_MUTEX(loop_index_mutex);
89 static int part_shift;
91 static int transfer_xor(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
96 char *raw_buf = kmap_atomic(raw_page) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page) + loop_off;
109 key = lo->lo_encrypt_key;
110 keysize = lo->lo_encrypt_key_size;
111 for (i = 0; i < size; i++)
112 *out++ = *in++ ^ key[(i & 511) % keysize];
114 kunmap_atomic(loop_buf);
115 kunmap_atomic(raw_buf);
120 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
122 if (unlikely(info->lo_encrypt_key_size <= 0))
127 static struct loop_func_table none_funcs = {
128 .number = LO_CRYPT_NONE,
131 static struct loop_func_table xor_funcs = {
132 .number = LO_CRYPT_XOR,
133 .transfer = transfer_xor,
137 /* xfer_funcs[0] is special - its release function is never called */
138 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
143 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
147 /* Compute loopsize in bytes */
148 loopsize = i_size_read(file->f_mapping->host);
151 /* offset is beyond i_size, weird but possible */
155 if (sizelimit > 0 && sizelimit < loopsize)
156 loopsize = sizelimit;
158 * Unfortunately, if we want to do I/O on the device,
159 * the number of 512-byte sectors has to fit into a sector_t.
161 return loopsize >> 9;
164 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
166 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
169 static void __loop_update_dio(struct loop_device *lo, bool dio)
171 struct file *file = lo->lo_backing_file;
172 struct address_space *mapping = file->f_mapping;
173 struct inode *inode = mapping->host;
174 unsigned short sb_bsize = 0;
175 unsigned dio_align = 0;
178 if (inode->i_sb->s_bdev) {
179 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
180 dio_align = sb_bsize - 1;
184 * We support direct I/O only if lo_offset is aligned with the
185 * logical I/O size of backing device, and the logical block
186 * size of loop is bigger than the backing device's and the loop
187 * needn't transform transfer.
189 * TODO: the above condition may be loosed in the future, and
190 * direct I/O may be switched runtime at that time because most
191 * of requests in sane applications should be PAGE_SIZE aligned
194 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
195 !(lo->lo_offset & dio_align) &&
196 mapping->a_ops->direct_IO &&
205 if (lo->use_dio == use_dio)
208 /* flush dirty pages before changing direct IO */
212 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
213 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
214 * will get updated by ioctl(LOOP_GET_STATUS)
216 blk_mq_freeze_queue(lo->lo_queue);
217 lo->use_dio = use_dio;
219 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
220 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
222 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
223 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
225 blk_mq_unfreeze_queue(lo->lo_queue);
229 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
231 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
232 sector_t x = (sector_t)size;
233 struct block_device *bdev = lo->lo_device;
235 if (unlikely((loff_t)x != size))
237 if (lo->lo_offset != offset)
238 lo->lo_offset = offset;
239 if (lo->lo_sizelimit != sizelimit)
240 lo->lo_sizelimit = sizelimit;
241 set_capacity(lo->lo_disk, x);
242 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
243 /* let user-space know about the new size */
244 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
249 lo_do_transfer(struct loop_device *lo, int cmd,
250 struct page *rpage, unsigned roffs,
251 struct page *lpage, unsigned loffs,
252 int size, sector_t rblock)
256 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
260 printk_ratelimited(KERN_ERR
261 "loop: Transfer error at byte offset %llu, length %i.\n",
262 (unsigned long long)rblock << 9, size);
266 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
271 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len);
273 file_start_write(file);
274 bw = vfs_iter_write(file, &i, ppos, 0);
275 file_end_write(file);
277 if (likely(bw == bvec->bv_len))
280 printk_ratelimited(KERN_ERR
281 "loop: Write error at byte offset %llu, length %i.\n",
282 (unsigned long long)*ppos, bvec->bv_len);
288 static int lo_write_simple(struct loop_device *lo, struct request *rq,
292 struct req_iterator iter;
295 rq_for_each_segment(bvec, rq, iter) {
296 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
306 * This is the slow, transforming version that needs to double buffer the
307 * data as it cannot do the transformations in place without having direct
308 * access to the destination pages of the backing file.
310 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
313 struct bio_vec bvec, b;
314 struct req_iterator iter;
318 page = alloc_page(GFP_NOIO);
322 rq_for_each_segment(bvec, rq, iter) {
323 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
324 bvec.bv_offset, bvec.bv_len, pos >> 9);
330 b.bv_len = bvec.bv_len;
331 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
340 static int lo_read_simple(struct loop_device *lo, struct request *rq,
344 struct req_iterator iter;
348 rq_for_each_segment(bvec, rq, iter) {
349 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
350 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
354 flush_dcache_page(bvec.bv_page);
356 if (len != bvec.bv_len) {
359 __rq_for_each_bio(bio, rq)
369 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
372 struct bio_vec bvec, b;
373 struct req_iterator iter;
379 page = alloc_page(GFP_NOIO);
383 rq_for_each_segment(bvec, rq, iter) {
388 b.bv_len = bvec.bv_len;
390 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
391 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
397 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
398 bvec.bv_offset, len, offset >> 9);
402 flush_dcache_page(bvec.bv_page);
404 if (len != bvec.bv_len) {
407 __rq_for_each_bio(bio, rq)
419 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
422 * We use punch hole to reclaim the free space used by the
423 * image a.k.a. discard. However we do not support discard if
424 * encryption is enabled, because it may give an attacker
425 * useful information.
427 struct file *file = lo->lo_backing_file;
428 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
431 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
436 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
437 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
443 static int lo_req_flush(struct loop_device *lo, struct request *rq)
445 struct file *file = lo->lo_backing_file;
446 int ret = vfs_fsync(file, 0);
447 if (unlikely(ret && ret != -EINVAL))
453 static void lo_complete_rq(struct request *rq)
455 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
456 blk_status_t ret = BLK_STS_OK;
458 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
459 req_op(rq) != REQ_OP_READ) {
466 * Short READ - if we got some data, advance our request and
467 * retry it. If we got no data, end the rest with EIO.
470 blk_update_request(rq, BLK_STS_OK, cmd->ret);
472 blk_mq_requeue_request(rq, true);
475 struct bio *bio = rq->bio;
484 blk_mq_end_request(rq, ret);
488 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
490 struct request *rq = blk_mq_rq_from_pdu(cmd);
492 if (!atomic_dec_and_test(&cmd->ref))
496 blk_mq_complete_request(rq);
499 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
501 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
506 lo_rw_aio_do_completion(cmd);
509 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
512 struct iov_iter iter;
513 struct bio_vec *bvec;
514 struct request *rq = blk_mq_rq_from_pdu(cmd);
515 struct bio *bio = rq->bio;
516 struct file *file = lo->lo_backing_file;
521 if (rq->bio != rq->biotail) {
522 struct req_iterator iter;
525 __rq_for_each_bio(bio, rq)
526 segments += bio_segments(bio);
527 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO);
533 * The bios of the request may be started from the middle of
534 * the 'bvec' because of bio splitting, so we can't directly
535 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
536 * API will take care of all details for us.
538 rq_for_each_segment(tmp, rq, iter) {
546 * Same here, this bio may be started from the middle of the
547 * 'bvec' because of bio splitting, so offset from the bvec
548 * must be passed to iov iterator
550 offset = bio->bi_iter.bi_bvec_done;
551 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
552 segments = bio_segments(bio);
554 atomic_set(&cmd->ref, 2);
556 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
557 segments, blk_rq_bytes(rq));
558 iter.iov_offset = offset;
560 cmd->iocb.ki_pos = pos;
561 cmd->iocb.ki_filp = file;
562 cmd->iocb.ki_complete = lo_rw_aio_complete;
563 cmd->iocb.ki_flags = IOCB_DIRECT;
564 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
566 kthread_associate_blkcg(cmd->css);
569 ret = call_write_iter(file, &cmd->iocb, &iter);
571 ret = call_read_iter(file, &cmd->iocb, &iter);
573 lo_rw_aio_do_completion(cmd);
574 kthread_associate_blkcg(NULL);
576 if (ret != -EIOCBQUEUED)
577 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
581 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
583 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
584 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
587 * lo_write_simple and lo_read_simple should have been covered
588 * by io submit style function like lo_rw_aio(), one blocker
589 * is that lo_read_simple() need to call flush_dcache_page after
590 * the page is written from kernel, and it isn't easy to handle
591 * this in io submit style function which submits all segments
592 * of the req at one time. And direct read IO doesn't need to
593 * run flush_dcache_page().
595 switch (req_op(rq)) {
597 return lo_req_flush(lo, rq);
599 case REQ_OP_WRITE_ZEROES:
600 return lo_discard(lo, rq, pos);
603 return lo_write_transfer(lo, rq, pos);
604 else if (cmd->use_aio)
605 return lo_rw_aio(lo, cmd, pos, WRITE);
607 return lo_write_simple(lo, rq, pos);
610 return lo_read_transfer(lo, rq, pos);
611 else if (cmd->use_aio)
612 return lo_rw_aio(lo, cmd, pos, READ);
614 return lo_read_simple(lo, rq, pos);
622 static inline void loop_update_dio(struct loop_device *lo)
624 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
628 static void loop_reread_partitions(struct loop_device *lo,
629 struct block_device *bdev)
634 * bd_mutex has been held already in release path, so don't
635 * acquire it if this function is called in such case.
637 * If the reread partition isn't from release path, lo_refcnt
638 * must be at least one and it can only become zero when the
639 * current holder is released.
641 if (!atomic_read(&lo->lo_refcnt))
642 rc = __blkdev_reread_part(bdev);
644 rc = blkdev_reread_part(bdev);
646 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
647 __func__, lo->lo_number, lo->lo_file_name, rc);
651 * loop_change_fd switched the backing store of a loopback device to
652 * a new file. This is useful for operating system installers to free up
653 * the original file and in High Availability environments to switch to
654 * an alternative location for the content in case of server meltdown.
655 * This can only work if the loop device is used read-only, and if the
656 * new backing store is the same size and type as the old backing store.
658 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
661 struct file *file, *old_file;
666 if (lo->lo_state != Lo_bound)
669 /* the loop device has to be read-only */
671 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
679 inode = file->f_mapping->host;
680 old_file = lo->lo_backing_file;
684 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
687 /* size of the new backing store needs to be the same */
688 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
692 blk_mq_freeze_queue(lo->lo_queue);
693 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
694 lo->lo_backing_file = file;
695 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
696 mapping_set_gfp_mask(file->f_mapping,
697 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
699 blk_mq_unfreeze_queue(lo->lo_queue);
702 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
703 loop_reread_partitions(lo, bdev);
712 static inline int is_loop_device(struct file *file)
714 struct inode *i = file->f_mapping->host;
716 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
719 /* loop sysfs attributes */
721 static ssize_t loop_attr_show(struct device *dev, char *page,
722 ssize_t (*callback)(struct loop_device *, char *))
724 struct gendisk *disk = dev_to_disk(dev);
725 struct loop_device *lo = disk->private_data;
727 return callback(lo, page);
730 #define LOOP_ATTR_RO(_name) \
731 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
732 static ssize_t loop_attr_do_show_##_name(struct device *d, \
733 struct device_attribute *attr, char *b) \
735 return loop_attr_show(d, b, loop_attr_##_name##_show); \
737 static struct device_attribute loop_attr_##_name = \
738 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
740 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
745 spin_lock_irq(&lo->lo_lock);
746 if (lo->lo_backing_file)
747 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
748 spin_unlock_irq(&lo->lo_lock);
750 if (IS_ERR_OR_NULL(p))
754 memmove(buf, p, ret);
762 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
764 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
767 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
769 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
772 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
774 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
776 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
779 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
781 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
783 return sprintf(buf, "%s\n", partscan ? "1" : "0");
786 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
788 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
790 return sprintf(buf, "%s\n", dio ? "1" : "0");
793 LOOP_ATTR_RO(backing_file);
794 LOOP_ATTR_RO(offset);
795 LOOP_ATTR_RO(sizelimit);
796 LOOP_ATTR_RO(autoclear);
797 LOOP_ATTR_RO(partscan);
800 static struct attribute *loop_attrs[] = {
801 &loop_attr_backing_file.attr,
802 &loop_attr_offset.attr,
803 &loop_attr_sizelimit.attr,
804 &loop_attr_autoclear.attr,
805 &loop_attr_partscan.attr,
810 static struct attribute_group loop_attribute_group = {
815 static void loop_sysfs_init(struct loop_device *lo)
817 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
818 &loop_attribute_group);
821 static void loop_sysfs_exit(struct loop_device *lo)
823 if (lo->sysfs_inited)
824 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
825 &loop_attribute_group);
828 static void loop_config_discard(struct loop_device *lo)
830 struct file *file = lo->lo_backing_file;
831 struct inode *inode = file->f_mapping->host;
832 struct request_queue *q = lo->lo_queue;
835 * We use punch hole to reclaim the free space used by the
836 * image a.k.a. discard. However we do not support discard if
837 * encryption is enabled, because it may give an attacker
838 * useful information.
840 if ((!file->f_op->fallocate) ||
841 lo->lo_encrypt_key_size) {
842 q->limits.discard_granularity = 0;
843 q->limits.discard_alignment = 0;
844 blk_queue_max_discard_sectors(q, 0);
845 blk_queue_max_write_zeroes_sectors(q, 0);
846 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
850 q->limits.discard_granularity = inode->i_sb->s_blocksize;
851 q->limits.discard_alignment = 0;
853 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
854 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
855 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
858 static void loop_unprepare_queue(struct loop_device *lo)
860 kthread_flush_worker(&lo->worker);
861 kthread_stop(lo->worker_task);
864 static int loop_kthread_worker_fn(void *worker_ptr)
866 current->flags |= PF_LESS_THROTTLE;
867 return kthread_worker_fn(worker_ptr);
870 static int loop_prepare_queue(struct loop_device *lo)
872 kthread_init_worker(&lo->worker);
873 lo->worker_task = kthread_run(loop_kthread_worker_fn,
874 &lo->worker, "loop%d", lo->lo_number);
875 if (IS_ERR(lo->worker_task))
877 set_user_nice(lo->worker_task, MIN_NICE);
881 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
882 struct block_device *bdev, unsigned int arg)
884 struct file *file, *f;
886 struct address_space *mapping;
891 /* This is safe, since we have a reference from open(). */
892 __module_get(THIS_MODULE);
900 if (lo->lo_state != Lo_unbound)
903 /* Avoid recursion */
905 while (is_loop_device(f)) {
906 struct loop_device *l;
908 if (f->f_mapping->host->i_bdev == bdev)
911 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
912 if (l->lo_state == Lo_unbound) {
916 f = l->lo_backing_file;
919 mapping = file->f_mapping;
920 inode = mapping->host;
923 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
926 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
927 !file->f_op->write_iter)
928 lo_flags |= LO_FLAGS_READ_ONLY;
931 size = get_loop_size(lo, file);
932 if ((loff_t)(sector_t)size != size)
934 error = loop_prepare_queue(lo);
940 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
943 lo->lo_device = bdev;
944 lo->lo_flags = lo_flags;
945 lo->lo_backing_file = file;
948 lo->lo_sizelimit = 0;
949 lo->old_gfp_mask = mapping_gfp_mask(mapping);
950 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
952 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
953 blk_queue_write_cache(lo->lo_queue, true, false);
956 set_capacity(lo->lo_disk, size);
957 bd_set_size(bdev, size << 9);
959 /* let user-space know about the new size */
960 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
962 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
963 block_size(inode->i_bdev) : PAGE_SIZE);
965 lo->lo_state = Lo_bound;
967 lo->lo_flags |= LO_FLAGS_PARTSCAN;
968 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
969 loop_reread_partitions(lo, bdev);
971 /* Grab the block_device to prevent its destruction after we
972 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
980 /* This is safe: open() is still holding a reference. */
981 module_put(THIS_MODULE);
986 loop_release_xfer(struct loop_device *lo)
989 struct loop_func_table *xfer = lo->lo_encryption;
993 err = xfer->release(lo);
995 lo->lo_encryption = NULL;
996 module_put(xfer->owner);
1002 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1003 const struct loop_info64 *i)
1008 struct module *owner = xfer->owner;
1010 if (!try_module_get(owner))
1013 err = xfer->init(lo, i);
1017 lo->lo_encryption = xfer;
1022 static int loop_clr_fd(struct loop_device *lo)
1024 struct file *filp = lo->lo_backing_file;
1025 gfp_t gfp = lo->old_gfp_mask;
1026 struct block_device *bdev = lo->lo_device;
1028 if (lo->lo_state != Lo_bound)
1032 * If we've explicitly asked to tear down the loop device,
1033 * and it has an elevated reference count, set it for auto-teardown when
1034 * the last reference goes away. This stops $!~#$@ udev from
1035 * preventing teardown because it decided that it needs to run blkid on
1036 * the loopback device whenever they appear. xfstests is notorious for
1037 * failing tests because blkid via udev races with a losetup
1038 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1039 * command to fail with EBUSY.
1041 if (atomic_read(&lo->lo_refcnt) > 1) {
1042 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1043 mutex_unlock(&lo->lo_ctl_mutex);
1050 /* freeze request queue during the transition */
1051 blk_mq_freeze_queue(lo->lo_queue);
1053 spin_lock_irq(&lo->lo_lock);
1054 lo->lo_state = Lo_rundown;
1055 lo->lo_backing_file = NULL;
1056 spin_unlock_irq(&lo->lo_lock);
1058 loop_release_xfer(lo);
1059 lo->transfer = NULL;
1061 lo->lo_device = NULL;
1062 lo->lo_encryption = NULL;
1064 lo->lo_sizelimit = 0;
1065 lo->lo_encrypt_key_size = 0;
1066 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1067 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1068 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1069 blk_queue_logical_block_size(lo->lo_queue, 512);
1070 blk_queue_physical_block_size(lo->lo_queue, 512);
1071 blk_queue_io_min(lo->lo_queue, 512);
1074 invalidate_bdev(bdev);
1075 bdev->bd_inode->i_mapping->wb_err = 0;
1077 set_capacity(lo->lo_disk, 0);
1078 loop_sysfs_exit(lo);
1080 bd_set_size(bdev, 0);
1081 /* let user-space know about this change */
1082 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1084 mapping_set_gfp_mask(filp->f_mapping, gfp);
1085 lo->lo_state = Lo_unbound;
1086 /* This is safe: open() is still holding a reference. */
1087 module_put(THIS_MODULE);
1088 blk_mq_unfreeze_queue(lo->lo_queue);
1090 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1091 loop_reread_partitions(lo, bdev);
1094 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1095 loop_unprepare_queue(lo);
1096 mutex_unlock(&lo->lo_ctl_mutex);
1098 * Need not hold lo_ctl_mutex to fput backing file.
1099 * Calling fput holding lo_ctl_mutex triggers a circular
1100 * lock dependency possibility warning as fput can take
1101 * bd_mutex which is usually taken before lo_ctl_mutex.
1108 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1111 struct loop_func_table *xfer;
1112 kuid_t uid = current_uid();
1114 if (lo->lo_encrypt_key_size &&
1115 !uid_eq(lo->lo_key_owner, uid) &&
1116 !capable(CAP_SYS_ADMIN))
1118 if (lo->lo_state != Lo_bound)
1120 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1123 /* I/O need to be drained during transfer transition */
1124 blk_mq_freeze_queue(lo->lo_queue);
1126 err = loop_release_xfer(lo);
1130 if (info->lo_encrypt_type) {
1131 unsigned int type = info->lo_encrypt_type;
1133 if (type >= MAX_LO_CRYPT) {
1137 xfer = xfer_funcs[type];
1145 err = loop_init_xfer(lo, xfer, info);
1149 if (lo->lo_offset != info->lo_offset ||
1150 lo->lo_sizelimit != info->lo_sizelimit) {
1151 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1157 loop_config_discard(lo);
1159 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1160 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1161 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1162 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1166 lo->transfer = xfer->transfer;
1167 lo->ioctl = xfer->ioctl;
1169 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1170 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1171 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1173 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1174 lo->lo_init[0] = info->lo_init[0];
1175 lo->lo_init[1] = info->lo_init[1];
1176 if (info->lo_encrypt_key_size) {
1177 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1178 info->lo_encrypt_key_size);
1179 lo->lo_key_owner = uid;
1182 /* update dio if lo_offset or transfer is changed */
1183 __loop_update_dio(lo, lo->use_dio);
1186 blk_mq_unfreeze_queue(lo->lo_queue);
1188 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1189 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1190 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1191 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1192 loop_reread_partitions(lo, lo->lo_device);
1199 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1205 if (lo->lo_state != Lo_bound) {
1206 mutex_unlock(&lo->lo_ctl_mutex);
1210 memset(info, 0, sizeof(*info));
1211 info->lo_number = lo->lo_number;
1212 info->lo_offset = lo->lo_offset;
1213 info->lo_sizelimit = lo->lo_sizelimit;
1214 info->lo_flags = lo->lo_flags;
1215 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1216 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1217 info->lo_encrypt_type =
1218 lo->lo_encryption ? lo->lo_encryption->number : 0;
1219 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1220 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1221 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1222 lo->lo_encrypt_key_size);
1225 /* Drop lo_ctl_mutex while we call into the filesystem. */
1226 file = get_file(lo->lo_backing_file);
1227 mutex_unlock(&lo->lo_ctl_mutex);
1228 ret = vfs_getattr(&file->f_path, &stat, STATX_INO,
1229 AT_STATX_SYNC_AS_STAT);
1231 info->lo_device = huge_encode_dev(stat.dev);
1232 info->lo_inode = stat.ino;
1233 info->lo_rdevice = huge_encode_dev(stat.rdev);
1240 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1242 memset(info64, 0, sizeof(*info64));
1243 info64->lo_number = info->lo_number;
1244 info64->lo_device = info->lo_device;
1245 info64->lo_inode = info->lo_inode;
1246 info64->lo_rdevice = info->lo_rdevice;
1247 info64->lo_offset = info->lo_offset;
1248 info64->lo_sizelimit = 0;
1249 info64->lo_encrypt_type = info->lo_encrypt_type;
1250 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1251 info64->lo_flags = info->lo_flags;
1252 info64->lo_init[0] = info->lo_init[0];
1253 info64->lo_init[1] = info->lo_init[1];
1254 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1255 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1257 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1258 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1262 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1264 memset(info, 0, sizeof(*info));
1265 info->lo_number = info64->lo_number;
1266 info->lo_device = info64->lo_device;
1267 info->lo_inode = info64->lo_inode;
1268 info->lo_rdevice = info64->lo_rdevice;
1269 info->lo_offset = info64->lo_offset;
1270 info->lo_encrypt_type = info64->lo_encrypt_type;
1271 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1272 info->lo_flags = info64->lo_flags;
1273 info->lo_init[0] = info64->lo_init[0];
1274 info->lo_init[1] = info64->lo_init[1];
1275 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1276 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1278 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1279 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1281 /* error in case values were truncated */
1282 if (info->lo_device != info64->lo_device ||
1283 info->lo_rdevice != info64->lo_rdevice ||
1284 info->lo_inode != info64->lo_inode ||
1285 info->lo_offset != info64->lo_offset)
1292 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1294 struct loop_info info;
1295 struct loop_info64 info64;
1297 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1299 loop_info64_from_old(&info, &info64);
1300 return loop_set_status(lo, &info64);
1304 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1306 struct loop_info64 info64;
1308 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1310 return loop_set_status(lo, &info64);
1314 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1315 struct loop_info info;
1316 struct loop_info64 info64;
1320 mutex_unlock(&lo->lo_ctl_mutex);
1323 err = loop_get_status(lo, &info64);
1325 err = loop_info64_to_old(&info64, &info);
1326 if (!err && copy_to_user(arg, &info, sizeof(info)))
1333 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1334 struct loop_info64 info64;
1338 mutex_unlock(&lo->lo_ctl_mutex);
1341 err = loop_get_status(lo, &info64);
1342 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1348 static int loop_set_capacity(struct loop_device *lo)
1350 if (unlikely(lo->lo_state != Lo_bound))
1353 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1356 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1359 if (lo->lo_state != Lo_bound)
1362 __loop_update_dio(lo, !!arg);
1363 if (lo->use_dio == !!arg)
1370 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1372 if (lo->lo_state != Lo_bound)
1375 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1378 blk_mq_freeze_queue(lo->lo_queue);
1380 blk_queue_logical_block_size(lo->lo_queue, arg);
1381 blk_queue_physical_block_size(lo->lo_queue, arg);
1382 blk_queue_io_min(lo->lo_queue, arg);
1383 loop_update_dio(lo);
1385 blk_mq_unfreeze_queue(lo->lo_queue);
1390 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1391 unsigned int cmd, unsigned long arg)
1393 struct loop_device *lo = bdev->bd_disk->private_data;
1396 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1);
1402 err = loop_set_fd(lo, mode, bdev, arg);
1404 case LOOP_CHANGE_FD:
1405 err = loop_change_fd(lo, bdev, arg);
1408 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1409 err = loop_clr_fd(lo);
1413 case LOOP_SET_STATUS:
1415 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1416 err = loop_set_status_old(lo,
1417 (struct loop_info __user *)arg);
1419 case LOOP_GET_STATUS:
1420 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1421 /* loop_get_status() unlocks lo_ctl_mutex */
1423 case LOOP_SET_STATUS64:
1425 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1426 err = loop_set_status64(lo,
1427 (struct loop_info64 __user *) arg);
1429 case LOOP_GET_STATUS64:
1430 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1431 /* loop_get_status() unlocks lo_ctl_mutex */
1433 case LOOP_SET_CAPACITY:
1435 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1436 err = loop_set_capacity(lo);
1438 case LOOP_SET_DIRECT_IO:
1440 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1441 err = loop_set_dio(lo, arg);
1443 case LOOP_SET_BLOCK_SIZE:
1445 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1446 err = loop_set_block_size(lo, arg);
1449 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1451 mutex_unlock(&lo->lo_ctl_mutex);
1457 #ifdef CONFIG_COMPAT
1458 struct compat_loop_info {
1459 compat_int_t lo_number; /* ioctl r/o */
1460 compat_dev_t lo_device; /* ioctl r/o */
1461 compat_ulong_t lo_inode; /* ioctl r/o */
1462 compat_dev_t lo_rdevice; /* ioctl r/o */
1463 compat_int_t lo_offset;
1464 compat_int_t lo_encrypt_type;
1465 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1466 compat_int_t lo_flags; /* ioctl r/o */
1467 char lo_name[LO_NAME_SIZE];
1468 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1469 compat_ulong_t lo_init[2];
1474 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1475 * - noinlined to reduce stack space usage in main part of driver
1478 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1479 struct loop_info64 *info64)
1481 struct compat_loop_info info;
1483 if (copy_from_user(&info, arg, sizeof(info)))
1486 memset(info64, 0, sizeof(*info64));
1487 info64->lo_number = info.lo_number;
1488 info64->lo_device = info.lo_device;
1489 info64->lo_inode = info.lo_inode;
1490 info64->lo_rdevice = info.lo_rdevice;
1491 info64->lo_offset = info.lo_offset;
1492 info64->lo_sizelimit = 0;
1493 info64->lo_encrypt_type = info.lo_encrypt_type;
1494 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1495 info64->lo_flags = info.lo_flags;
1496 info64->lo_init[0] = info.lo_init[0];
1497 info64->lo_init[1] = info.lo_init[1];
1498 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1499 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1501 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1502 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1507 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1508 * - noinlined to reduce stack space usage in main part of driver
1511 loop_info64_to_compat(const struct loop_info64 *info64,
1512 struct compat_loop_info __user *arg)
1514 struct compat_loop_info info;
1516 memset(&info, 0, sizeof(info));
1517 info.lo_number = info64->lo_number;
1518 info.lo_device = info64->lo_device;
1519 info.lo_inode = info64->lo_inode;
1520 info.lo_rdevice = info64->lo_rdevice;
1521 info.lo_offset = info64->lo_offset;
1522 info.lo_encrypt_type = info64->lo_encrypt_type;
1523 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1524 info.lo_flags = info64->lo_flags;
1525 info.lo_init[0] = info64->lo_init[0];
1526 info.lo_init[1] = info64->lo_init[1];
1527 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1528 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1530 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1531 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1533 /* error in case values were truncated */
1534 if (info.lo_device != info64->lo_device ||
1535 info.lo_rdevice != info64->lo_rdevice ||
1536 info.lo_inode != info64->lo_inode ||
1537 info.lo_offset != info64->lo_offset ||
1538 info.lo_init[0] != info64->lo_init[0] ||
1539 info.lo_init[1] != info64->lo_init[1])
1542 if (copy_to_user(arg, &info, sizeof(info)))
1548 loop_set_status_compat(struct loop_device *lo,
1549 const struct compat_loop_info __user *arg)
1551 struct loop_info64 info64;
1554 ret = loop_info64_from_compat(arg, &info64);
1557 return loop_set_status(lo, &info64);
1561 loop_get_status_compat(struct loop_device *lo,
1562 struct compat_loop_info __user *arg)
1564 struct loop_info64 info64;
1568 mutex_unlock(&lo->lo_ctl_mutex);
1571 err = loop_get_status(lo, &info64);
1573 err = loop_info64_to_compat(&info64, arg);
1577 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1578 unsigned int cmd, unsigned long arg)
1580 struct loop_device *lo = bdev->bd_disk->private_data;
1584 case LOOP_SET_STATUS:
1585 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1587 err = loop_set_status_compat(lo,
1588 (const struct compat_loop_info __user *)arg);
1589 mutex_unlock(&lo->lo_ctl_mutex);
1592 case LOOP_GET_STATUS:
1593 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1595 err = loop_get_status_compat(lo,
1596 (struct compat_loop_info __user *)arg);
1597 /* loop_get_status() unlocks lo_ctl_mutex */
1600 case LOOP_SET_CAPACITY:
1602 case LOOP_GET_STATUS64:
1603 case LOOP_SET_STATUS64:
1604 arg = (unsigned long) compat_ptr(arg);
1606 case LOOP_CHANGE_FD:
1607 err = lo_ioctl(bdev, mode, cmd, arg);
1617 static int lo_open(struct block_device *bdev, fmode_t mode)
1619 struct loop_device *lo;
1622 mutex_lock(&loop_index_mutex);
1623 lo = bdev->bd_disk->private_data;
1629 atomic_inc(&lo->lo_refcnt);
1631 mutex_unlock(&loop_index_mutex);
1635 static void __lo_release(struct loop_device *lo)
1639 if (atomic_dec_return(&lo->lo_refcnt))
1642 mutex_lock(&lo->lo_ctl_mutex);
1643 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1645 * In autoclear mode, stop the loop thread
1646 * and remove configuration after last close.
1648 err = loop_clr_fd(lo);
1651 } else if (lo->lo_state == Lo_bound) {
1653 * Otherwise keep thread (if running) and config,
1654 * but flush possible ongoing bios in thread.
1656 blk_mq_freeze_queue(lo->lo_queue);
1657 blk_mq_unfreeze_queue(lo->lo_queue);
1660 mutex_unlock(&lo->lo_ctl_mutex);
1663 static void lo_release(struct gendisk *disk, fmode_t mode)
1665 mutex_lock(&loop_index_mutex);
1666 __lo_release(disk->private_data);
1667 mutex_unlock(&loop_index_mutex);
1670 static const struct block_device_operations lo_fops = {
1671 .owner = THIS_MODULE,
1673 .release = lo_release,
1675 #ifdef CONFIG_COMPAT
1676 .compat_ioctl = lo_compat_ioctl,
1681 * And now the modules code and kernel interface.
1683 static int max_loop;
1684 module_param(max_loop, int, 0444);
1685 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1686 module_param(max_part, int, 0444);
1687 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1688 MODULE_LICENSE("GPL");
1689 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1691 int loop_register_transfer(struct loop_func_table *funcs)
1693 unsigned int n = funcs->number;
1695 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1697 xfer_funcs[n] = funcs;
1701 static int unregister_transfer_cb(int id, void *ptr, void *data)
1703 struct loop_device *lo = ptr;
1704 struct loop_func_table *xfer = data;
1706 mutex_lock(&lo->lo_ctl_mutex);
1707 if (lo->lo_encryption == xfer)
1708 loop_release_xfer(lo);
1709 mutex_unlock(&lo->lo_ctl_mutex);
1713 int loop_unregister_transfer(int number)
1715 unsigned int n = number;
1716 struct loop_func_table *xfer;
1718 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1721 xfer_funcs[n] = NULL;
1722 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1726 EXPORT_SYMBOL(loop_register_transfer);
1727 EXPORT_SYMBOL(loop_unregister_transfer);
1729 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1730 const struct blk_mq_queue_data *bd)
1732 struct request *rq = bd->rq;
1733 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1734 struct loop_device *lo = rq->q->queuedata;
1736 blk_mq_start_request(rq);
1738 if (lo->lo_state != Lo_bound)
1739 return BLK_STS_IOERR;
1741 switch (req_op(rq)) {
1743 case REQ_OP_DISCARD:
1744 case REQ_OP_WRITE_ZEROES:
1745 cmd->use_aio = false;
1748 cmd->use_aio = lo->use_dio;
1752 /* always use the first bio's css */
1753 #ifdef CONFIG_BLK_CGROUP
1754 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1755 cmd->css = rq->bio->bi_css;
1760 kthread_queue_work(&lo->worker, &cmd->work);
1765 static void loop_handle_cmd(struct loop_cmd *cmd)
1767 struct request *rq = blk_mq_rq_from_pdu(cmd);
1768 const bool write = op_is_write(req_op(rq));
1769 struct loop_device *lo = rq->q->queuedata;
1772 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1777 ret = do_req_filebacked(lo, rq);
1779 /* complete non-aio request */
1780 if (!cmd->use_aio || ret) {
1781 cmd->ret = ret ? -EIO : 0;
1782 blk_mq_complete_request(rq);
1786 static void loop_queue_work(struct kthread_work *work)
1788 struct loop_cmd *cmd =
1789 container_of(work, struct loop_cmd, work);
1791 loop_handle_cmd(cmd);
1794 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1795 unsigned int hctx_idx, unsigned int numa_node)
1797 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1799 kthread_init_work(&cmd->work, loop_queue_work);
1803 static const struct blk_mq_ops loop_mq_ops = {
1804 .queue_rq = loop_queue_rq,
1805 .init_request = loop_init_request,
1806 .complete = lo_complete_rq,
1809 static int loop_add(struct loop_device **l, int i)
1811 struct loop_device *lo;
1812 struct gendisk *disk;
1816 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1820 lo->lo_state = Lo_unbound;
1822 /* allocate id, if @id >= 0, we're requesting that specific id */
1824 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1828 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1835 lo->tag_set.ops = &loop_mq_ops;
1836 lo->tag_set.nr_hw_queues = 1;
1837 lo->tag_set.queue_depth = 128;
1838 lo->tag_set.numa_node = NUMA_NO_NODE;
1839 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1840 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1841 lo->tag_set.driver_data = lo;
1843 err = blk_mq_alloc_tag_set(&lo->tag_set);
1847 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1848 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1849 err = PTR_ERR(lo->lo_queue);
1850 goto out_cleanup_tags;
1852 lo->lo_queue->queuedata = lo;
1854 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1857 * By default, we do buffer IO, so it doesn't make sense to enable
1858 * merge because the I/O submitted to backing file is handled page by
1859 * page. For directio mode, merge does help to dispatch bigger request
1860 * to underlayer disk. We will enable merge once directio is enabled.
1862 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1865 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1867 goto out_free_queue;
1870 * Disable partition scanning by default. The in-kernel partition
1871 * scanning can be requested individually per-device during its
1872 * setup. Userspace can always add and remove partitions from all
1873 * devices. The needed partition minors are allocated from the
1874 * extended minor space, the main loop device numbers will continue
1875 * to match the loop minors, regardless of the number of partitions
1878 * If max_part is given, partition scanning is globally enabled for
1879 * all loop devices. The minors for the main loop devices will be
1880 * multiples of max_part.
1882 * Note: Global-for-all-devices, set-only-at-init, read-only module
1883 * parameteters like 'max_loop' and 'max_part' make things needlessly
1884 * complicated, are too static, inflexible and may surprise
1885 * userspace tools. Parameters like this in general should be avoided.
1888 disk->flags |= GENHD_FL_NO_PART_SCAN;
1889 disk->flags |= GENHD_FL_EXT_DEVT;
1890 mutex_init(&lo->lo_ctl_mutex);
1891 atomic_set(&lo->lo_refcnt, 0);
1893 spin_lock_init(&lo->lo_lock);
1894 disk->major = LOOP_MAJOR;
1895 disk->first_minor = i << part_shift;
1896 disk->fops = &lo_fops;
1897 disk->private_data = lo;
1898 disk->queue = lo->lo_queue;
1899 sprintf(disk->disk_name, "loop%d", i);
1902 return lo->lo_number;
1905 blk_cleanup_queue(lo->lo_queue);
1907 blk_mq_free_tag_set(&lo->tag_set);
1909 idr_remove(&loop_index_idr, i);
1916 static void loop_remove(struct loop_device *lo)
1918 del_gendisk(lo->lo_disk);
1919 blk_cleanup_queue(lo->lo_queue);
1920 blk_mq_free_tag_set(&lo->tag_set);
1921 put_disk(lo->lo_disk);
1925 static int find_free_cb(int id, void *ptr, void *data)
1927 struct loop_device *lo = ptr;
1928 struct loop_device **l = data;
1930 if (lo->lo_state == Lo_unbound) {
1937 static int loop_lookup(struct loop_device **l, int i)
1939 struct loop_device *lo;
1945 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1948 ret = lo->lo_number;
1953 /* lookup and return a specific i */
1954 lo = idr_find(&loop_index_idr, i);
1957 ret = lo->lo_number;
1963 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1965 struct loop_device *lo;
1966 struct kobject *kobj;
1969 mutex_lock(&loop_index_mutex);
1970 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1972 err = loop_add(&lo, MINOR(dev) >> part_shift);
1976 kobj = get_disk_and_module(lo->lo_disk);
1977 mutex_unlock(&loop_index_mutex);
1983 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1986 struct loop_device *lo;
1989 mutex_lock(&loop_index_mutex);
1992 ret = loop_lookup(&lo, parm);
1997 ret = loop_add(&lo, parm);
1999 case LOOP_CTL_REMOVE:
2000 ret = loop_lookup(&lo, parm);
2003 ret = mutex_lock_killable(&lo->lo_ctl_mutex);
2006 if (lo->lo_state != Lo_unbound) {
2008 mutex_unlock(&lo->lo_ctl_mutex);
2011 if (atomic_read(&lo->lo_refcnt) > 0) {
2013 mutex_unlock(&lo->lo_ctl_mutex);
2016 lo->lo_disk->private_data = NULL;
2017 mutex_unlock(&lo->lo_ctl_mutex);
2018 idr_remove(&loop_index_idr, lo->lo_number);
2021 case LOOP_CTL_GET_FREE:
2022 ret = loop_lookup(&lo, -1);
2025 ret = loop_add(&lo, -1);
2027 mutex_unlock(&loop_index_mutex);
2032 static const struct file_operations loop_ctl_fops = {
2033 .open = nonseekable_open,
2034 .unlocked_ioctl = loop_control_ioctl,
2035 .compat_ioctl = loop_control_ioctl,
2036 .owner = THIS_MODULE,
2037 .llseek = noop_llseek,
2040 static struct miscdevice loop_misc = {
2041 .minor = LOOP_CTRL_MINOR,
2042 .name = "loop-control",
2043 .fops = &loop_ctl_fops,
2046 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2047 MODULE_ALIAS("devname:loop-control");
2049 static int __init loop_init(void)
2052 unsigned long range;
2053 struct loop_device *lo;
2058 part_shift = fls(max_part);
2061 * Adjust max_part according to part_shift as it is exported
2062 * to user space so that user can decide correct minor number
2063 * if [s]he want to create more devices.
2065 * Note that -1 is required because partition 0 is reserved
2066 * for the whole disk.
2068 max_part = (1UL << part_shift) - 1;
2071 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2076 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2082 * If max_loop is specified, create that many devices upfront.
2083 * This also becomes a hard limit. If max_loop is not specified,
2084 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2085 * init time. Loop devices can be requested on-demand with the
2086 * /dev/loop-control interface, or be instantiated by accessing
2087 * a 'dead' device node.
2091 range = max_loop << part_shift;
2093 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2094 range = 1UL << MINORBITS;
2097 err = misc_register(&loop_misc);
2102 if (register_blkdev(LOOP_MAJOR, "loop")) {
2107 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2108 THIS_MODULE, loop_probe, NULL, NULL);
2110 /* pre-create number of devices given by config or max_loop */
2111 mutex_lock(&loop_index_mutex);
2112 for (i = 0; i < nr; i++)
2114 mutex_unlock(&loop_index_mutex);
2116 printk(KERN_INFO "loop: module loaded\n");
2120 misc_deregister(&loop_misc);
2125 static int loop_exit_cb(int id, void *ptr, void *data)
2127 struct loop_device *lo = ptr;
2133 static void __exit loop_exit(void)
2135 unsigned long range;
2137 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2139 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2140 idr_destroy(&loop_index_idr);
2142 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2143 unregister_blkdev(LOOP_MAJOR, "loop");
2145 misc_deregister(&loop_misc);
2148 module_init(loop_init);
2149 module_exit(loop_exit);
2152 static int __init max_loop_setup(char *str)
2154 max_loop = simple_strtol(str, NULL, 0);
2158 __setup("max_loop=", max_loop_setup);