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>
81 #include <linux/uaccess.h>
83 static DEFINE_IDR(loop_index_idr);
84 static DEFINE_MUTEX(loop_index_mutex);
87 static int part_shift;
89 static int transfer_xor(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
94 char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page) + loop_off;
107 key = lo->lo_encrypt_key;
108 keysize = lo->lo_encrypt_key_size;
109 for (i = 0; i < size; i++)
110 *out++ = *in++ ^ key[(i & 511) % keysize];
112 kunmap_atomic(loop_buf);
113 kunmap_atomic(raw_buf);
118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
120 if (unlikely(info->lo_encrypt_key_size <= 0))
125 static struct loop_func_table none_funcs = {
126 .number = LO_CRYPT_NONE,
129 static struct loop_func_table xor_funcs = {
130 .number = LO_CRYPT_XOR,
131 .transfer = transfer_xor,
135 /* xfer_funcs[0] is special - its release function is never called */
136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
145 /* Compute loopsize in bytes */
146 loopsize = i_size_read(file->f_mapping->host);
149 /* offset is beyond i_size, weird but possible */
153 if (sizelimit > 0 && sizelimit < loopsize)
154 loopsize = sizelimit;
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
159 return loopsize >> 9;
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
167 static void __loop_update_dio(struct loop_device *lo, bool dio)
169 struct file *file = lo->lo_backing_file;
170 struct address_space *mapping = file->f_mapping;
171 struct inode *inode = mapping->host;
172 unsigned short sb_bsize = 0;
173 unsigned dio_align = 0;
176 if (inode->i_sb->s_bdev) {
177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 dio_align = sb_bsize - 1;
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane applications should be PAGE_SIZE aligned
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 !(lo->lo_offset & dio_align) &&
194 mapping->a_ops->direct_IO &&
203 if (lo->use_dio == use_dio)
206 /* flush dirty pages before changing direct IO */
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
214 blk_mq_freeze_queue(lo->lo_queue);
215 lo->use_dio = use_dio;
217 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
218 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
220 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
221 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
223 blk_mq_unfreeze_queue(lo->lo_queue);
227 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
229 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
230 sector_t x = (sector_t)size;
231 struct block_device *bdev = lo->lo_device;
233 if (unlikely((loff_t)x != size))
235 if (lo->lo_offset != offset)
236 lo->lo_offset = offset;
237 if (lo->lo_sizelimit != sizelimit)
238 lo->lo_sizelimit = sizelimit;
239 set_capacity(lo->lo_disk, x);
240 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
241 /* let user-space know about the new size */
242 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
247 lo_do_transfer(struct loop_device *lo, int cmd,
248 struct page *rpage, unsigned roffs,
249 struct page *lpage, unsigned loffs,
250 int size, sector_t rblock)
254 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
258 printk_ratelimited(KERN_ERR
259 "loop: Transfer error at byte offset %llu, length %i.\n",
260 (unsigned long long)rblock << 9, size);
264 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
269 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len);
271 file_start_write(file);
272 bw = vfs_iter_write(file, &i, ppos, 0);
273 file_end_write(file);
275 if (likely(bw == bvec->bv_len))
278 printk_ratelimited(KERN_ERR
279 "loop: Write error at byte offset %llu, length %i.\n",
280 (unsigned long long)*ppos, bvec->bv_len);
286 static int lo_write_simple(struct loop_device *lo, struct request *rq,
290 struct req_iterator iter;
293 rq_for_each_segment(bvec, rq, iter) {
294 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
304 * This is the slow, transforming version that needs to double buffer the
305 * data as it cannot do the transformations in place without having direct
306 * access to the destination pages of the backing file.
308 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
311 struct bio_vec bvec, b;
312 struct req_iterator iter;
316 page = alloc_page(GFP_NOIO);
320 rq_for_each_segment(bvec, rq, iter) {
321 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
322 bvec.bv_offset, bvec.bv_len, pos >> 9);
328 b.bv_len = bvec.bv_len;
329 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
338 static int lo_read_simple(struct loop_device *lo, struct request *rq,
342 struct req_iterator iter;
346 rq_for_each_segment(bvec, rq, iter) {
347 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
348 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
352 flush_dcache_page(bvec.bv_page);
354 if (len != bvec.bv_len) {
357 __rq_for_each_bio(bio, rq)
367 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
370 struct bio_vec bvec, b;
371 struct req_iterator iter;
377 page = alloc_page(GFP_NOIO);
381 rq_for_each_segment(bvec, rq, iter) {
386 b.bv_len = bvec.bv_len;
388 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
389 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
395 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
396 bvec.bv_offset, len, offset >> 9);
400 flush_dcache_page(bvec.bv_page);
402 if (len != bvec.bv_len) {
405 __rq_for_each_bio(bio, rq)
417 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
420 * We use punch hole to reclaim the free space used by the
421 * image a.k.a. discard. However we do not support discard if
422 * encryption is enabled, because it may give an attacker
423 * useful information.
425 struct file *file = lo->lo_backing_file;
426 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
429 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
434 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
435 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
441 static int lo_req_flush(struct loop_device *lo, struct request *rq)
443 struct file *file = lo->lo_backing_file;
444 int ret = vfs_fsync(file, 0);
445 if (unlikely(ret && ret != -EINVAL))
451 static void lo_complete_rq(struct request *rq)
453 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
455 if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio &&
456 cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) {
457 struct bio *bio = cmd->rq->bio;
459 bio_advance(bio, cmd->ret);
463 blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK);
466 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
468 if (!atomic_dec_and_test(&cmd->ref))
472 blk_mq_complete_request(cmd->rq);
475 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
477 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
482 lo_rw_aio_do_completion(cmd);
485 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
488 struct iov_iter iter;
489 struct bio_vec *bvec;
490 struct request *rq = cmd->rq;
491 struct bio *bio = rq->bio;
492 struct file *file = lo->lo_backing_file;
497 if (rq->bio != rq->biotail) {
498 struct req_iterator iter;
501 __rq_for_each_bio(bio, rq)
502 segments += bio_segments(bio);
503 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO);
509 * The bios of the request may be started from the middle of
510 * the 'bvec' because of bio splitting, so we can't directly
511 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
512 * API will take care of all details for us.
514 rq_for_each_segment(tmp, rq, iter) {
522 * Same here, this bio may be started from the middle of the
523 * 'bvec' because of bio splitting, so offset from the bvec
524 * must be passed to iov iterator
526 offset = bio->bi_iter.bi_bvec_done;
527 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
528 segments = bio_segments(bio);
530 atomic_set(&cmd->ref, 2);
532 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
533 segments, blk_rq_bytes(rq));
534 iter.iov_offset = offset;
536 cmd->iocb.ki_pos = pos;
537 cmd->iocb.ki_filp = file;
538 cmd->iocb.ki_complete = lo_rw_aio_complete;
539 cmd->iocb.ki_flags = IOCB_DIRECT;
541 kthread_associate_blkcg(cmd->css);
544 ret = call_write_iter(file, &cmd->iocb, &iter);
546 ret = call_read_iter(file, &cmd->iocb, &iter);
548 lo_rw_aio_do_completion(cmd);
549 kthread_associate_blkcg(NULL);
551 if (ret != -EIOCBQUEUED)
552 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
556 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
558 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
559 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
562 * lo_write_simple and lo_read_simple should have been covered
563 * by io submit style function like lo_rw_aio(), one blocker
564 * is that lo_read_simple() need to call flush_dcache_page after
565 * the page is written from kernel, and it isn't easy to handle
566 * this in io submit style function which submits all segments
567 * of the req at one time. And direct read IO doesn't need to
568 * run flush_dcache_page().
570 switch (req_op(rq)) {
572 return lo_req_flush(lo, rq);
574 case REQ_OP_WRITE_ZEROES:
575 return lo_discard(lo, rq, pos);
578 return lo_write_transfer(lo, rq, pos);
579 else if (cmd->use_aio)
580 return lo_rw_aio(lo, cmd, pos, WRITE);
582 return lo_write_simple(lo, rq, pos);
585 return lo_read_transfer(lo, rq, pos);
586 else if (cmd->use_aio)
587 return lo_rw_aio(lo, cmd, pos, READ);
589 return lo_read_simple(lo, rq, pos);
597 static inline void loop_update_dio(struct loop_device *lo)
599 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
603 static void loop_reread_partitions(struct loop_device *lo,
604 struct block_device *bdev)
609 * bd_mutex has been held already in release path, so don't
610 * acquire it if this function is called in such case.
612 * If the reread partition isn't from release path, lo_refcnt
613 * must be at least one and it can only become zero when the
614 * current holder is released.
616 if (!atomic_read(&lo->lo_refcnt))
617 rc = __blkdev_reread_part(bdev);
619 rc = blkdev_reread_part(bdev);
621 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
622 __func__, lo->lo_number, lo->lo_file_name, rc);
626 * loop_change_fd switched the backing store of a loopback device to
627 * a new file. This is useful for operating system installers to free up
628 * the original file and in High Availability environments to switch to
629 * an alternative location for the content in case of server meltdown.
630 * This can only work if the loop device is used read-only, and if the
631 * new backing store is the same size and type as the old backing store.
633 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
636 struct file *file, *old_file;
641 if (lo->lo_state != Lo_bound)
644 /* the loop device has to be read-only */
646 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
654 inode = file->f_mapping->host;
655 old_file = lo->lo_backing_file;
659 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
662 /* size of the new backing store needs to be the same */
663 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
667 blk_mq_freeze_queue(lo->lo_queue);
668 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
669 lo->lo_backing_file = file;
670 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
671 mapping_set_gfp_mask(file->f_mapping,
672 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
674 blk_mq_unfreeze_queue(lo->lo_queue);
677 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
678 loop_reread_partitions(lo, bdev);
687 static inline int is_loop_device(struct file *file)
689 struct inode *i = file->f_mapping->host;
691 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
694 /* loop sysfs attributes */
696 static ssize_t loop_attr_show(struct device *dev, char *page,
697 ssize_t (*callback)(struct loop_device *, char *))
699 struct gendisk *disk = dev_to_disk(dev);
700 struct loop_device *lo = disk->private_data;
702 return callback(lo, page);
705 #define LOOP_ATTR_RO(_name) \
706 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
707 static ssize_t loop_attr_do_show_##_name(struct device *d, \
708 struct device_attribute *attr, char *b) \
710 return loop_attr_show(d, b, loop_attr_##_name##_show); \
712 static struct device_attribute loop_attr_##_name = \
713 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
715 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
720 spin_lock_irq(&lo->lo_lock);
721 if (lo->lo_backing_file)
722 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
723 spin_unlock_irq(&lo->lo_lock);
725 if (IS_ERR_OR_NULL(p))
729 memmove(buf, p, ret);
737 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
739 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
742 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
744 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
747 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
749 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
751 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
754 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
756 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
758 return sprintf(buf, "%s\n", partscan ? "1" : "0");
761 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
763 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
765 return sprintf(buf, "%s\n", dio ? "1" : "0");
768 LOOP_ATTR_RO(backing_file);
769 LOOP_ATTR_RO(offset);
770 LOOP_ATTR_RO(sizelimit);
771 LOOP_ATTR_RO(autoclear);
772 LOOP_ATTR_RO(partscan);
775 static struct attribute *loop_attrs[] = {
776 &loop_attr_backing_file.attr,
777 &loop_attr_offset.attr,
778 &loop_attr_sizelimit.attr,
779 &loop_attr_autoclear.attr,
780 &loop_attr_partscan.attr,
785 static struct attribute_group loop_attribute_group = {
790 static int loop_sysfs_init(struct loop_device *lo)
792 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
793 &loop_attribute_group);
796 static void loop_sysfs_exit(struct loop_device *lo)
798 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
799 &loop_attribute_group);
802 static void loop_config_discard(struct loop_device *lo)
804 struct file *file = lo->lo_backing_file;
805 struct inode *inode = file->f_mapping->host;
806 struct request_queue *q = lo->lo_queue;
809 * We use punch hole to reclaim the free space used by the
810 * image a.k.a. discard. However we do not support discard if
811 * encryption is enabled, because it may give an attacker
812 * useful information.
814 if ((!file->f_op->fallocate) ||
815 lo->lo_encrypt_key_size) {
816 q->limits.discard_granularity = 0;
817 q->limits.discard_alignment = 0;
818 blk_queue_max_discard_sectors(q, 0);
819 blk_queue_max_write_zeroes_sectors(q, 0);
820 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
824 q->limits.discard_granularity = inode->i_sb->s_blocksize;
825 q->limits.discard_alignment = 0;
827 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
828 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
829 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
832 static void loop_unprepare_queue(struct loop_device *lo)
834 kthread_flush_worker(&lo->worker);
835 kthread_stop(lo->worker_task);
838 static int loop_kthread_worker_fn(void *worker_ptr)
840 current->flags |= PF_LESS_THROTTLE;
841 return kthread_worker_fn(worker_ptr);
844 static int loop_prepare_queue(struct loop_device *lo)
846 kthread_init_worker(&lo->worker);
847 lo->worker_task = kthread_run(loop_kthread_worker_fn,
848 &lo->worker, "loop%d", lo->lo_number);
849 if (IS_ERR(lo->worker_task))
851 set_user_nice(lo->worker_task, MIN_NICE);
855 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
856 struct block_device *bdev, unsigned int arg)
858 struct file *file, *f;
860 struct address_space *mapping;
865 /* This is safe, since we have a reference from open(). */
866 __module_get(THIS_MODULE);
874 if (lo->lo_state != Lo_unbound)
877 /* Avoid recursion */
879 while (is_loop_device(f)) {
880 struct loop_device *l;
882 if (f->f_mapping->host->i_bdev == bdev)
885 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
886 if (l->lo_state == Lo_unbound) {
890 f = l->lo_backing_file;
893 mapping = file->f_mapping;
894 inode = mapping->host;
897 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
900 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
901 !file->f_op->write_iter)
902 lo_flags |= LO_FLAGS_READ_ONLY;
905 size = get_loop_size(lo, file);
906 if ((loff_t)(sector_t)size != size)
908 error = loop_prepare_queue(lo);
914 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
917 lo->lo_device = bdev;
918 lo->lo_flags = lo_flags;
919 lo->lo_backing_file = file;
922 lo->lo_sizelimit = 0;
923 lo->old_gfp_mask = mapping_gfp_mask(mapping);
924 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
926 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
927 blk_queue_write_cache(lo->lo_queue, true, false);
930 set_capacity(lo->lo_disk, size);
931 bd_set_size(bdev, size << 9);
933 /* let user-space know about the new size */
934 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
936 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
937 block_size(inode->i_bdev) : PAGE_SIZE);
939 lo->lo_state = Lo_bound;
941 lo->lo_flags |= LO_FLAGS_PARTSCAN;
942 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
943 loop_reread_partitions(lo, bdev);
945 /* Grab the block_device to prevent its destruction after we
946 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
954 /* This is safe: open() is still holding a reference. */
955 module_put(THIS_MODULE);
960 loop_release_xfer(struct loop_device *lo)
963 struct loop_func_table *xfer = lo->lo_encryption;
967 err = xfer->release(lo);
969 lo->lo_encryption = NULL;
970 module_put(xfer->owner);
976 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
977 const struct loop_info64 *i)
982 struct module *owner = xfer->owner;
984 if (!try_module_get(owner))
987 err = xfer->init(lo, i);
991 lo->lo_encryption = xfer;
996 static int loop_clr_fd(struct loop_device *lo)
998 struct file *filp = lo->lo_backing_file;
999 gfp_t gfp = lo->old_gfp_mask;
1000 struct block_device *bdev = lo->lo_device;
1002 if (lo->lo_state != Lo_bound)
1006 * If we've explicitly asked to tear down the loop device,
1007 * and it has an elevated reference count, set it for auto-teardown when
1008 * the last reference goes away. This stops $!~#$@ udev from
1009 * preventing teardown because it decided that it needs to run blkid on
1010 * the loopback device whenever they appear. xfstests is notorious for
1011 * failing tests because blkid via udev races with a losetup
1012 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1013 * command to fail with EBUSY.
1015 if (atomic_read(&lo->lo_refcnt) > 1) {
1016 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1017 mutex_unlock(&lo->lo_ctl_mutex);
1024 /* freeze request queue during the transition */
1025 blk_mq_freeze_queue(lo->lo_queue);
1027 spin_lock_irq(&lo->lo_lock);
1028 lo->lo_state = Lo_rundown;
1029 lo->lo_backing_file = NULL;
1030 spin_unlock_irq(&lo->lo_lock);
1032 loop_release_xfer(lo);
1033 lo->transfer = NULL;
1035 lo->lo_device = NULL;
1036 lo->lo_encryption = NULL;
1038 lo->lo_sizelimit = 0;
1039 lo->lo_encrypt_key_size = 0;
1040 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1041 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1042 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1043 blk_queue_logical_block_size(lo->lo_queue, 512);
1044 blk_queue_physical_block_size(lo->lo_queue, 512);
1045 blk_queue_io_min(lo->lo_queue, 512);
1048 invalidate_bdev(bdev);
1050 set_capacity(lo->lo_disk, 0);
1051 loop_sysfs_exit(lo);
1053 bd_set_size(bdev, 0);
1054 /* let user-space know about this change */
1055 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1057 mapping_set_gfp_mask(filp->f_mapping, gfp);
1058 lo->lo_state = Lo_unbound;
1059 /* This is safe: open() is still holding a reference. */
1060 module_put(THIS_MODULE);
1061 blk_mq_unfreeze_queue(lo->lo_queue);
1063 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1064 loop_reread_partitions(lo, bdev);
1067 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1068 loop_unprepare_queue(lo);
1069 mutex_unlock(&lo->lo_ctl_mutex);
1071 * Need not hold lo_ctl_mutex to fput backing file.
1072 * Calling fput holding lo_ctl_mutex triggers a circular
1073 * lock dependency possibility warning as fput can take
1074 * bd_mutex which is usually taken before lo_ctl_mutex.
1081 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1084 struct loop_func_table *xfer;
1085 kuid_t uid = current_uid();
1087 if (lo->lo_encrypt_key_size &&
1088 !uid_eq(lo->lo_key_owner, uid) &&
1089 !capable(CAP_SYS_ADMIN))
1091 if (lo->lo_state != Lo_bound)
1093 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1096 /* I/O need to be drained during transfer transition */
1097 blk_mq_freeze_queue(lo->lo_queue);
1099 err = loop_release_xfer(lo);
1103 if (info->lo_encrypt_type) {
1104 unsigned int type = info->lo_encrypt_type;
1106 if (type >= MAX_LO_CRYPT)
1108 xfer = xfer_funcs[type];
1114 err = loop_init_xfer(lo, xfer, info);
1118 if (lo->lo_offset != info->lo_offset ||
1119 lo->lo_sizelimit != info->lo_sizelimit) {
1120 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1126 loop_config_discard(lo);
1128 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1129 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1130 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1131 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1135 lo->transfer = xfer->transfer;
1136 lo->ioctl = xfer->ioctl;
1138 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1139 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1140 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1142 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1143 lo->lo_init[0] = info->lo_init[0];
1144 lo->lo_init[1] = info->lo_init[1];
1145 if (info->lo_encrypt_key_size) {
1146 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1147 info->lo_encrypt_key_size);
1148 lo->lo_key_owner = uid;
1151 /* update dio if lo_offset or transfer is changed */
1152 __loop_update_dio(lo, lo->use_dio);
1155 blk_mq_unfreeze_queue(lo->lo_queue);
1157 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1158 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1159 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1160 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1161 loop_reread_partitions(lo, lo->lo_device);
1168 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1174 if (lo->lo_state != Lo_bound) {
1175 mutex_unlock(&lo->lo_ctl_mutex);
1179 memset(info, 0, sizeof(*info));
1180 info->lo_number = lo->lo_number;
1181 info->lo_offset = lo->lo_offset;
1182 info->lo_sizelimit = lo->lo_sizelimit;
1183 info->lo_flags = lo->lo_flags;
1184 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1185 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1186 info->lo_encrypt_type =
1187 lo->lo_encryption ? lo->lo_encryption->number : 0;
1188 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1189 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1190 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1191 lo->lo_encrypt_key_size);
1194 /* Drop lo_ctl_mutex while we call into the filesystem. */
1195 file = get_file(lo->lo_backing_file);
1196 mutex_unlock(&lo->lo_ctl_mutex);
1197 ret = vfs_getattr(&file->f_path, &stat, STATX_INO,
1198 AT_STATX_SYNC_AS_STAT);
1200 info->lo_device = huge_encode_dev(stat.dev);
1201 info->lo_inode = stat.ino;
1202 info->lo_rdevice = huge_encode_dev(stat.rdev);
1209 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1211 memset(info64, 0, sizeof(*info64));
1212 info64->lo_number = info->lo_number;
1213 info64->lo_device = info->lo_device;
1214 info64->lo_inode = info->lo_inode;
1215 info64->lo_rdevice = info->lo_rdevice;
1216 info64->lo_offset = info->lo_offset;
1217 info64->lo_sizelimit = 0;
1218 info64->lo_encrypt_type = info->lo_encrypt_type;
1219 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1220 info64->lo_flags = info->lo_flags;
1221 info64->lo_init[0] = info->lo_init[0];
1222 info64->lo_init[1] = info->lo_init[1];
1223 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1224 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1226 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1227 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1231 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1233 memset(info, 0, sizeof(*info));
1234 info->lo_number = info64->lo_number;
1235 info->lo_device = info64->lo_device;
1236 info->lo_inode = info64->lo_inode;
1237 info->lo_rdevice = info64->lo_rdevice;
1238 info->lo_offset = info64->lo_offset;
1239 info->lo_encrypt_type = info64->lo_encrypt_type;
1240 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1241 info->lo_flags = info64->lo_flags;
1242 info->lo_init[0] = info64->lo_init[0];
1243 info->lo_init[1] = info64->lo_init[1];
1244 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1245 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1247 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1248 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1250 /* error in case values were truncated */
1251 if (info->lo_device != info64->lo_device ||
1252 info->lo_rdevice != info64->lo_rdevice ||
1253 info->lo_inode != info64->lo_inode ||
1254 info->lo_offset != info64->lo_offset)
1261 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1263 struct loop_info info;
1264 struct loop_info64 info64;
1266 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1268 loop_info64_from_old(&info, &info64);
1269 return loop_set_status(lo, &info64);
1273 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1275 struct loop_info64 info64;
1277 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1279 return loop_set_status(lo, &info64);
1283 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1284 struct loop_info info;
1285 struct loop_info64 info64;
1291 err = loop_get_status(lo, &info64);
1293 err = loop_info64_to_old(&info64, &info);
1294 if (!err && copy_to_user(arg, &info, sizeof(info)))
1301 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1302 struct loop_info64 info64;
1308 err = loop_get_status(lo, &info64);
1309 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1315 static int loop_set_capacity(struct loop_device *lo)
1317 if (unlikely(lo->lo_state != Lo_bound))
1320 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1323 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1326 if (lo->lo_state != Lo_bound)
1329 __loop_update_dio(lo, !!arg);
1330 if (lo->use_dio == !!arg)
1337 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1339 if (lo->lo_state != Lo_bound)
1342 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1345 blk_mq_freeze_queue(lo->lo_queue);
1347 blk_queue_logical_block_size(lo->lo_queue, arg);
1348 blk_queue_physical_block_size(lo->lo_queue, arg);
1349 blk_queue_io_min(lo->lo_queue, arg);
1350 loop_update_dio(lo);
1352 blk_mq_unfreeze_queue(lo->lo_queue);
1357 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1358 unsigned int cmd, unsigned long arg)
1360 struct loop_device *lo = bdev->bd_disk->private_data;
1363 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1);
1369 err = loop_set_fd(lo, mode, bdev, arg);
1371 case LOOP_CHANGE_FD:
1372 err = loop_change_fd(lo, bdev, arg);
1375 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1376 err = loop_clr_fd(lo);
1380 case LOOP_SET_STATUS:
1382 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1383 err = loop_set_status_old(lo,
1384 (struct loop_info __user *)arg);
1386 case LOOP_GET_STATUS:
1387 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1388 /* loop_get_status() unlocks lo_ctl_mutex */
1390 case LOOP_SET_STATUS64:
1392 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1393 err = loop_set_status64(lo,
1394 (struct loop_info64 __user *) arg);
1396 case LOOP_GET_STATUS64:
1397 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1398 /* loop_get_status() unlocks lo_ctl_mutex */
1400 case LOOP_SET_CAPACITY:
1402 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1403 err = loop_set_capacity(lo);
1405 case LOOP_SET_DIRECT_IO:
1407 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1408 err = loop_set_dio(lo, arg);
1410 case LOOP_SET_BLOCK_SIZE:
1412 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1413 err = loop_set_block_size(lo, arg);
1416 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1418 mutex_unlock(&lo->lo_ctl_mutex);
1424 #ifdef CONFIG_COMPAT
1425 struct compat_loop_info {
1426 compat_int_t lo_number; /* ioctl r/o */
1427 compat_dev_t lo_device; /* ioctl r/o */
1428 compat_ulong_t lo_inode; /* ioctl r/o */
1429 compat_dev_t lo_rdevice; /* ioctl r/o */
1430 compat_int_t lo_offset;
1431 compat_int_t lo_encrypt_type;
1432 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1433 compat_int_t lo_flags; /* ioctl r/o */
1434 char lo_name[LO_NAME_SIZE];
1435 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1436 compat_ulong_t lo_init[2];
1441 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1442 * - noinlined to reduce stack space usage in main part of driver
1445 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1446 struct loop_info64 *info64)
1448 struct compat_loop_info info;
1450 if (copy_from_user(&info, arg, sizeof(info)))
1453 memset(info64, 0, sizeof(*info64));
1454 info64->lo_number = info.lo_number;
1455 info64->lo_device = info.lo_device;
1456 info64->lo_inode = info.lo_inode;
1457 info64->lo_rdevice = info.lo_rdevice;
1458 info64->lo_offset = info.lo_offset;
1459 info64->lo_sizelimit = 0;
1460 info64->lo_encrypt_type = info.lo_encrypt_type;
1461 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1462 info64->lo_flags = info.lo_flags;
1463 info64->lo_init[0] = info.lo_init[0];
1464 info64->lo_init[1] = info.lo_init[1];
1465 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1466 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1468 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1469 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1474 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1475 * - noinlined to reduce stack space usage in main part of driver
1478 loop_info64_to_compat(const struct loop_info64 *info64,
1479 struct compat_loop_info __user *arg)
1481 struct compat_loop_info info;
1483 memset(&info, 0, sizeof(info));
1484 info.lo_number = info64->lo_number;
1485 info.lo_device = info64->lo_device;
1486 info.lo_inode = info64->lo_inode;
1487 info.lo_rdevice = info64->lo_rdevice;
1488 info.lo_offset = info64->lo_offset;
1489 info.lo_encrypt_type = info64->lo_encrypt_type;
1490 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1491 info.lo_flags = info64->lo_flags;
1492 info.lo_init[0] = info64->lo_init[0];
1493 info.lo_init[1] = info64->lo_init[1];
1494 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1495 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1497 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1498 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1500 /* error in case values were truncated */
1501 if (info.lo_device != info64->lo_device ||
1502 info.lo_rdevice != info64->lo_rdevice ||
1503 info.lo_inode != info64->lo_inode ||
1504 info.lo_offset != info64->lo_offset ||
1505 info.lo_init[0] != info64->lo_init[0] ||
1506 info.lo_init[1] != info64->lo_init[1])
1509 if (copy_to_user(arg, &info, sizeof(info)))
1515 loop_set_status_compat(struct loop_device *lo,
1516 const struct compat_loop_info __user *arg)
1518 struct loop_info64 info64;
1521 ret = loop_info64_from_compat(arg, &info64);
1524 return loop_set_status(lo, &info64);
1528 loop_get_status_compat(struct loop_device *lo,
1529 struct compat_loop_info __user *arg)
1531 struct loop_info64 info64;
1537 err = loop_get_status(lo, &info64);
1539 err = loop_info64_to_compat(&info64, arg);
1543 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1544 unsigned int cmd, unsigned long arg)
1546 struct loop_device *lo = bdev->bd_disk->private_data;
1550 case LOOP_SET_STATUS:
1551 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1553 err = loop_set_status_compat(lo,
1554 (const struct compat_loop_info __user *)arg);
1555 mutex_unlock(&lo->lo_ctl_mutex);
1558 case LOOP_GET_STATUS:
1559 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1561 err = loop_get_status_compat(lo,
1562 (struct compat_loop_info __user *)arg);
1563 /* loop_get_status() unlocks lo_ctl_mutex */
1566 case LOOP_SET_CAPACITY:
1568 case LOOP_GET_STATUS64:
1569 case LOOP_SET_STATUS64:
1570 arg = (unsigned long) compat_ptr(arg);
1572 case LOOP_CHANGE_FD:
1573 err = lo_ioctl(bdev, mode, cmd, arg);
1583 static int lo_open(struct block_device *bdev, fmode_t mode)
1585 struct loop_device *lo;
1588 mutex_lock(&loop_index_mutex);
1589 lo = bdev->bd_disk->private_data;
1595 atomic_inc(&lo->lo_refcnt);
1597 mutex_unlock(&loop_index_mutex);
1601 static void __lo_release(struct loop_device *lo)
1605 if (atomic_dec_return(&lo->lo_refcnt))
1608 mutex_lock(&lo->lo_ctl_mutex);
1609 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1611 * In autoclear mode, stop the loop thread
1612 * and remove configuration after last close.
1614 err = loop_clr_fd(lo);
1617 } else if (lo->lo_state == Lo_bound) {
1619 * Otherwise keep thread (if running) and config,
1620 * but flush possible ongoing bios in thread.
1622 blk_mq_freeze_queue(lo->lo_queue);
1623 blk_mq_unfreeze_queue(lo->lo_queue);
1626 mutex_unlock(&lo->lo_ctl_mutex);
1629 static void lo_release(struct gendisk *disk, fmode_t mode)
1631 mutex_lock(&loop_index_mutex);
1632 __lo_release(disk->private_data);
1633 mutex_unlock(&loop_index_mutex);
1636 static const struct block_device_operations lo_fops = {
1637 .owner = THIS_MODULE,
1639 .release = lo_release,
1641 #ifdef CONFIG_COMPAT
1642 .compat_ioctl = lo_compat_ioctl,
1647 * And now the modules code and kernel interface.
1649 static int max_loop;
1650 module_param(max_loop, int, S_IRUGO);
1651 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1652 module_param(max_part, int, S_IRUGO);
1653 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1654 MODULE_LICENSE("GPL");
1655 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1657 int loop_register_transfer(struct loop_func_table *funcs)
1659 unsigned int n = funcs->number;
1661 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1663 xfer_funcs[n] = funcs;
1667 static int unregister_transfer_cb(int id, void *ptr, void *data)
1669 struct loop_device *lo = ptr;
1670 struct loop_func_table *xfer = data;
1672 mutex_lock(&lo->lo_ctl_mutex);
1673 if (lo->lo_encryption == xfer)
1674 loop_release_xfer(lo);
1675 mutex_unlock(&lo->lo_ctl_mutex);
1679 int loop_unregister_transfer(int number)
1681 unsigned int n = number;
1682 struct loop_func_table *xfer;
1684 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1687 xfer_funcs[n] = NULL;
1688 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1692 EXPORT_SYMBOL(loop_register_transfer);
1693 EXPORT_SYMBOL(loop_unregister_transfer);
1695 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1696 const struct blk_mq_queue_data *bd)
1698 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1699 struct loop_device *lo = cmd->rq->q->queuedata;
1701 blk_mq_start_request(bd->rq);
1703 if (lo->lo_state != Lo_bound)
1704 return BLK_STS_IOERR;
1706 switch (req_op(cmd->rq)) {
1708 case REQ_OP_DISCARD:
1709 case REQ_OP_WRITE_ZEROES:
1710 cmd->use_aio = false;
1713 cmd->use_aio = lo->use_dio;
1717 /* always use the first bio's css */
1718 #ifdef CONFIG_BLK_CGROUP
1719 if (cmd->use_aio && cmd->rq->bio && cmd->rq->bio->bi_css) {
1720 cmd->css = cmd->rq->bio->bi_css;
1725 kthread_queue_work(&lo->worker, &cmd->work);
1730 static void loop_handle_cmd(struct loop_cmd *cmd)
1732 const bool write = op_is_write(req_op(cmd->rq));
1733 struct loop_device *lo = cmd->rq->q->queuedata;
1736 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1741 ret = do_req_filebacked(lo, cmd->rq);
1743 /* complete non-aio request */
1744 if (!cmd->use_aio || ret) {
1745 cmd->ret = ret ? -EIO : 0;
1746 blk_mq_complete_request(cmd->rq);
1750 static void loop_queue_work(struct kthread_work *work)
1752 struct loop_cmd *cmd =
1753 container_of(work, struct loop_cmd, work);
1755 loop_handle_cmd(cmd);
1758 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1759 unsigned int hctx_idx, unsigned int numa_node)
1761 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1764 kthread_init_work(&cmd->work, loop_queue_work);
1769 static const struct blk_mq_ops loop_mq_ops = {
1770 .queue_rq = loop_queue_rq,
1771 .init_request = loop_init_request,
1772 .complete = lo_complete_rq,
1775 static int loop_add(struct loop_device **l, int i)
1777 struct loop_device *lo;
1778 struct gendisk *disk;
1782 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1786 lo->lo_state = Lo_unbound;
1788 /* allocate id, if @id >= 0, we're requesting that specific id */
1790 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1794 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1801 lo->tag_set.ops = &loop_mq_ops;
1802 lo->tag_set.nr_hw_queues = 1;
1803 lo->tag_set.queue_depth = 128;
1804 lo->tag_set.numa_node = NUMA_NO_NODE;
1805 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1806 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1807 lo->tag_set.driver_data = lo;
1809 err = blk_mq_alloc_tag_set(&lo->tag_set);
1813 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1814 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1815 err = PTR_ERR(lo->lo_queue);
1816 goto out_cleanup_tags;
1818 lo->lo_queue->queuedata = lo;
1820 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1823 * By default, we do buffer IO, so it doesn't make sense to enable
1824 * merge because the I/O submitted to backing file is handled page by
1825 * page. For directio mode, merge does help to dispatch bigger request
1826 * to underlayer disk. We will enable merge once directio is enabled.
1828 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1831 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1833 goto out_free_queue;
1836 * Disable partition scanning by default. The in-kernel partition
1837 * scanning can be requested individually per-device during its
1838 * setup. Userspace can always add and remove partitions from all
1839 * devices. The needed partition minors are allocated from the
1840 * extended minor space, the main loop device numbers will continue
1841 * to match the loop minors, regardless of the number of partitions
1844 * If max_part is given, partition scanning is globally enabled for
1845 * all loop devices. The minors for the main loop devices will be
1846 * multiples of max_part.
1848 * Note: Global-for-all-devices, set-only-at-init, read-only module
1849 * parameteters like 'max_loop' and 'max_part' make things needlessly
1850 * complicated, are too static, inflexible and may surprise
1851 * userspace tools. Parameters like this in general should be avoided.
1854 disk->flags |= GENHD_FL_NO_PART_SCAN;
1855 disk->flags |= GENHD_FL_EXT_DEVT;
1856 mutex_init(&lo->lo_ctl_mutex);
1857 atomic_set(&lo->lo_refcnt, 0);
1859 spin_lock_init(&lo->lo_lock);
1860 disk->major = LOOP_MAJOR;
1861 disk->first_minor = i << part_shift;
1862 disk->fops = &lo_fops;
1863 disk->private_data = lo;
1864 disk->queue = lo->lo_queue;
1865 sprintf(disk->disk_name, "loop%d", i);
1868 return lo->lo_number;
1871 blk_cleanup_queue(lo->lo_queue);
1873 blk_mq_free_tag_set(&lo->tag_set);
1875 idr_remove(&loop_index_idr, i);
1882 static void loop_remove(struct loop_device *lo)
1884 del_gendisk(lo->lo_disk);
1885 blk_cleanup_queue(lo->lo_queue);
1886 blk_mq_free_tag_set(&lo->tag_set);
1887 put_disk(lo->lo_disk);
1891 static int find_free_cb(int id, void *ptr, void *data)
1893 struct loop_device *lo = ptr;
1894 struct loop_device **l = data;
1896 if (lo->lo_state == Lo_unbound) {
1903 static int loop_lookup(struct loop_device **l, int i)
1905 struct loop_device *lo;
1911 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1914 ret = lo->lo_number;
1919 /* lookup and return a specific i */
1920 lo = idr_find(&loop_index_idr, i);
1923 ret = lo->lo_number;
1929 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1931 struct loop_device *lo;
1932 struct kobject *kobj;
1935 mutex_lock(&loop_index_mutex);
1936 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1938 err = loop_add(&lo, MINOR(dev) >> part_shift);
1942 kobj = get_disk_and_module(lo->lo_disk);
1943 mutex_unlock(&loop_index_mutex);
1949 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1952 struct loop_device *lo;
1955 mutex_lock(&loop_index_mutex);
1958 ret = loop_lookup(&lo, parm);
1963 ret = loop_add(&lo, parm);
1965 case LOOP_CTL_REMOVE:
1966 ret = loop_lookup(&lo, parm);
1969 ret = mutex_lock_killable(&lo->lo_ctl_mutex);
1972 if (lo->lo_state != Lo_unbound) {
1974 mutex_unlock(&lo->lo_ctl_mutex);
1977 if (atomic_read(&lo->lo_refcnt) > 0) {
1979 mutex_unlock(&lo->lo_ctl_mutex);
1982 lo->lo_disk->private_data = NULL;
1983 mutex_unlock(&lo->lo_ctl_mutex);
1984 idr_remove(&loop_index_idr, lo->lo_number);
1987 case LOOP_CTL_GET_FREE:
1988 ret = loop_lookup(&lo, -1);
1991 ret = loop_add(&lo, -1);
1993 mutex_unlock(&loop_index_mutex);
1998 static const struct file_operations loop_ctl_fops = {
1999 .open = nonseekable_open,
2000 .unlocked_ioctl = loop_control_ioctl,
2001 .compat_ioctl = loop_control_ioctl,
2002 .owner = THIS_MODULE,
2003 .llseek = noop_llseek,
2006 static struct miscdevice loop_misc = {
2007 .minor = LOOP_CTRL_MINOR,
2008 .name = "loop-control",
2009 .fops = &loop_ctl_fops,
2012 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2013 MODULE_ALIAS("devname:loop-control");
2015 static int __init loop_init(void)
2018 unsigned long range;
2019 struct loop_device *lo;
2024 part_shift = fls(max_part);
2027 * Adjust max_part according to part_shift as it is exported
2028 * to user space so that user can decide correct minor number
2029 * if [s]he want to create more devices.
2031 * Note that -1 is required because partition 0 is reserved
2032 * for the whole disk.
2034 max_part = (1UL << part_shift) - 1;
2037 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2042 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2048 * If max_loop is specified, create that many devices upfront.
2049 * This also becomes a hard limit. If max_loop is not specified,
2050 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2051 * init time. Loop devices can be requested on-demand with the
2052 * /dev/loop-control interface, or be instantiated by accessing
2053 * a 'dead' device node.
2057 range = max_loop << part_shift;
2059 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2060 range = 1UL << MINORBITS;
2063 err = misc_register(&loop_misc);
2068 if (register_blkdev(LOOP_MAJOR, "loop")) {
2073 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2074 THIS_MODULE, loop_probe, NULL, NULL);
2076 /* pre-create number of devices given by config or max_loop */
2077 mutex_lock(&loop_index_mutex);
2078 for (i = 0; i < nr; i++)
2080 mutex_unlock(&loop_index_mutex);
2082 printk(KERN_INFO "loop: module loaded\n");
2086 misc_deregister(&loop_misc);
2091 static int loop_exit_cb(int id, void *ptr, void *data)
2093 struct loop_device *lo = ptr;
2099 static void __exit loop_exit(void)
2101 unsigned long range;
2103 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2105 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2106 idr_destroy(&loop_index_idr);
2108 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2109 unregister_blkdev(LOOP_MAJOR, "loop");
2111 misc_deregister(&loop_misc);
2114 module_init(loop_init);
2115 module_exit(loop_exit);
2118 static int __init max_loop_setup(char *str)
2120 max_loop = simple_strtol(str, NULL, 0);
2124 __setup("max_loop=", max_loop_setup);