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 queue_flag_clear_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
218 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
220 queue_flag_set_unlocked(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);
480 lo_rw_aio_do_completion(cmd);
483 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
486 struct iov_iter iter;
487 struct bio_vec *bvec;
488 struct request *rq = cmd->rq;
489 struct bio *bio = rq->bio;
490 struct file *file = lo->lo_backing_file;
495 if (rq->bio != rq->biotail) {
496 struct req_iterator iter;
499 __rq_for_each_bio(bio, rq)
500 segments += bio_segments(bio);
501 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO);
507 * The bios of the request may be started from the middle of
508 * the 'bvec' because of bio splitting, so we can't directly
509 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
510 * API will take care of all details for us.
512 rq_for_each_segment(tmp, rq, iter) {
520 * Same here, this bio may be started from the middle of the
521 * 'bvec' because of bio splitting, so offset from the bvec
522 * must be passed to iov iterator
524 offset = bio->bi_iter.bi_bvec_done;
525 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
526 segments = bio_segments(bio);
528 atomic_set(&cmd->ref, 2);
530 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
531 segments, blk_rq_bytes(rq));
532 iter.iov_offset = offset;
534 cmd->iocb.ki_pos = pos;
535 cmd->iocb.ki_filp = file;
536 cmd->iocb.ki_complete = lo_rw_aio_complete;
537 cmd->iocb.ki_flags = IOCB_DIRECT;
540 ret = call_write_iter(file, &cmd->iocb, &iter);
542 ret = call_read_iter(file, &cmd->iocb, &iter);
544 lo_rw_aio_do_completion(cmd);
546 if (ret != -EIOCBQUEUED)
547 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
551 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
553 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
554 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
557 * lo_write_simple and lo_read_simple should have been covered
558 * by io submit style function like lo_rw_aio(), one blocker
559 * is that lo_read_simple() need to call flush_dcache_page after
560 * the page is written from kernel, and it isn't easy to handle
561 * this in io submit style function which submits all segments
562 * of the req at one time. And direct read IO doesn't need to
563 * run flush_dcache_page().
565 switch (req_op(rq)) {
567 return lo_req_flush(lo, rq);
569 case REQ_OP_WRITE_ZEROES:
570 return lo_discard(lo, rq, pos);
573 return lo_write_transfer(lo, rq, pos);
574 else if (cmd->use_aio)
575 return lo_rw_aio(lo, cmd, pos, WRITE);
577 return lo_write_simple(lo, rq, pos);
580 return lo_read_transfer(lo, rq, pos);
581 else if (cmd->use_aio)
582 return lo_rw_aio(lo, cmd, pos, READ);
584 return lo_read_simple(lo, rq, pos);
592 static inline void loop_update_dio(struct loop_device *lo)
594 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
598 static void loop_reread_partitions(struct loop_device *lo,
599 struct block_device *bdev)
604 * bd_mutex has been held already in release path, so don't
605 * acquire it if this function is called in such case.
607 * If the reread partition isn't from release path, lo_refcnt
608 * must be at least one and it can only become zero when the
609 * current holder is released.
611 if (!atomic_read(&lo->lo_refcnt))
612 rc = __blkdev_reread_part(bdev);
614 rc = blkdev_reread_part(bdev);
616 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
617 __func__, lo->lo_number, lo->lo_file_name, rc);
620 static inline int is_loop_device(struct file *file)
622 struct inode *i = file->f_mapping->host;
624 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
627 static int loop_validate_file(struct file *file, struct block_device *bdev)
629 struct inode *inode = file->f_mapping->host;
630 struct file *f = file;
632 /* Avoid recursion */
633 while (is_loop_device(f)) {
634 struct loop_device *l;
636 if (f->f_mapping->host->i_bdev == bdev)
639 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
640 if (l->lo_state == Lo_unbound) {
643 f = l->lo_backing_file;
645 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
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 error = loop_validate_file(file, bdev);
683 inode = file->f_mapping->host;
684 old_file = lo->lo_backing_file;
688 /* size of the new backing store needs to be the same */
689 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
693 blk_mq_freeze_queue(lo->lo_queue);
694 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
695 lo->lo_backing_file = file;
696 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
697 mapping_set_gfp_mask(file->f_mapping,
698 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
700 blk_mq_unfreeze_queue(lo->lo_queue);
703 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
704 loop_reread_partitions(lo, bdev);
713 /* loop sysfs attributes */
715 static ssize_t loop_attr_show(struct device *dev, char *page,
716 ssize_t (*callback)(struct loop_device *, char *))
718 struct gendisk *disk = dev_to_disk(dev);
719 struct loop_device *lo = disk->private_data;
721 return callback(lo, page);
724 #define LOOP_ATTR_RO(_name) \
725 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
726 static ssize_t loop_attr_do_show_##_name(struct device *d, \
727 struct device_attribute *attr, char *b) \
729 return loop_attr_show(d, b, loop_attr_##_name##_show); \
731 static struct device_attribute loop_attr_##_name = \
732 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
734 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
739 spin_lock_irq(&lo->lo_lock);
740 if (lo->lo_backing_file)
741 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
742 spin_unlock_irq(&lo->lo_lock);
744 if (IS_ERR_OR_NULL(p))
748 memmove(buf, p, ret);
756 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
758 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
761 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
763 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
766 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
768 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
770 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
773 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
775 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
777 return sprintf(buf, "%s\n", partscan ? "1" : "0");
780 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
782 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
784 return sprintf(buf, "%s\n", dio ? "1" : "0");
787 LOOP_ATTR_RO(backing_file);
788 LOOP_ATTR_RO(offset);
789 LOOP_ATTR_RO(sizelimit);
790 LOOP_ATTR_RO(autoclear);
791 LOOP_ATTR_RO(partscan);
794 static struct attribute *loop_attrs[] = {
795 &loop_attr_backing_file.attr,
796 &loop_attr_offset.attr,
797 &loop_attr_sizelimit.attr,
798 &loop_attr_autoclear.attr,
799 &loop_attr_partscan.attr,
804 static struct attribute_group loop_attribute_group = {
809 static void loop_sysfs_init(struct loop_device *lo)
811 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
812 &loop_attribute_group);
815 static void loop_sysfs_exit(struct loop_device *lo)
817 if (lo->sysfs_inited)
818 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
819 &loop_attribute_group);
822 static void loop_config_discard(struct loop_device *lo)
824 struct file *file = lo->lo_backing_file;
825 struct inode *inode = file->f_mapping->host;
826 struct request_queue *q = lo->lo_queue;
829 * We use punch hole to reclaim the free space used by the
830 * image a.k.a. discard. However we do not support discard if
831 * encryption is enabled, because it may give an attacker
832 * useful information.
834 if ((!file->f_op->fallocate) ||
835 lo->lo_encrypt_key_size) {
836 q->limits.discard_granularity = 0;
837 q->limits.discard_alignment = 0;
838 blk_queue_max_discard_sectors(q, 0);
839 blk_queue_max_write_zeroes_sectors(q, 0);
840 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
844 q->limits.discard_granularity = inode->i_sb->s_blocksize;
845 q->limits.discard_alignment = 0;
847 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
848 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
849 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
852 static void loop_unprepare_queue(struct loop_device *lo)
854 kthread_flush_worker(&lo->worker);
855 kthread_stop(lo->worker_task);
858 static int loop_kthread_worker_fn(void *worker_ptr)
860 current->flags |= PF_LESS_THROTTLE;
861 return kthread_worker_fn(worker_ptr);
864 static int loop_prepare_queue(struct loop_device *lo)
866 kthread_init_worker(&lo->worker);
867 lo->worker_task = kthread_run(loop_kthread_worker_fn,
868 &lo->worker, "loop%d", lo->lo_number);
869 if (IS_ERR(lo->worker_task))
871 set_user_nice(lo->worker_task, MIN_NICE);
875 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
876 struct block_device *bdev, unsigned int arg)
880 struct address_space *mapping;
885 /* This is safe, since we have a reference from open(). */
886 __module_get(THIS_MODULE);
894 if (lo->lo_state != Lo_unbound)
897 error = loop_validate_file(file, bdev);
901 mapping = file->f_mapping;
902 inode = mapping->host;
904 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
905 !file->f_op->write_iter)
906 lo_flags |= LO_FLAGS_READ_ONLY;
909 size = get_loop_size(lo, file);
910 if ((loff_t)(sector_t)size != size)
912 error = loop_prepare_queue(lo);
918 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
921 lo->lo_device = bdev;
922 lo->lo_flags = lo_flags;
923 lo->lo_backing_file = file;
926 lo->lo_sizelimit = 0;
927 lo->old_gfp_mask = mapping_gfp_mask(mapping);
928 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
930 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
931 blk_queue_write_cache(lo->lo_queue, true, false);
934 set_capacity(lo->lo_disk, size);
935 bd_set_size(bdev, size << 9);
937 /* let user-space know about the new size */
938 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
940 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
941 block_size(inode->i_bdev) : PAGE_SIZE);
943 lo->lo_state = Lo_bound;
945 lo->lo_flags |= LO_FLAGS_PARTSCAN;
946 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
947 loop_reread_partitions(lo, bdev);
949 /* Grab the block_device to prevent its destruction after we
950 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
958 /* This is safe: open() is still holding a reference. */
959 module_put(THIS_MODULE);
964 loop_release_xfer(struct loop_device *lo)
967 struct loop_func_table *xfer = lo->lo_encryption;
971 err = xfer->release(lo);
973 lo->lo_encryption = NULL;
974 module_put(xfer->owner);
980 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
981 const struct loop_info64 *i)
986 struct module *owner = xfer->owner;
988 if (!try_module_get(owner))
991 err = xfer->init(lo, i);
995 lo->lo_encryption = xfer;
1000 static int loop_clr_fd(struct loop_device *lo)
1002 struct file *filp = lo->lo_backing_file;
1003 gfp_t gfp = lo->old_gfp_mask;
1004 struct block_device *bdev = lo->lo_device;
1006 if (lo->lo_state != Lo_bound)
1010 * If we've explicitly asked to tear down the loop device,
1011 * and it has an elevated reference count, set it for auto-teardown when
1012 * the last reference goes away. This stops $!~#$@ udev from
1013 * preventing teardown because it decided that it needs to run blkid on
1014 * the loopback device whenever they appear. xfstests is notorious for
1015 * failing tests because blkid via udev races with a losetup
1016 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1017 * command to fail with EBUSY.
1019 if (atomic_read(&lo->lo_refcnt) > 1) {
1020 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1021 mutex_unlock(&lo->lo_ctl_mutex);
1028 /* freeze request queue during the transition */
1029 blk_mq_freeze_queue(lo->lo_queue);
1031 spin_lock_irq(&lo->lo_lock);
1032 lo->lo_state = Lo_rundown;
1033 lo->lo_backing_file = NULL;
1034 spin_unlock_irq(&lo->lo_lock);
1036 loop_release_xfer(lo);
1037 lo->transfer = NULL;
1039 lo->lo_device = NULL;
1040 lo->lo_encryption = NULL;
1042 lo->lo_sizelimit = 0;
1043 lo->lo_encrypt_key_size = 0;
1044 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1045 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1046 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1047 blk_queue_logical_block_size(lo->lo_queue, 512);
1048 blk_queue_physical_block_size(lo->lo_queue, 512);
1049 blk_queue_io_min(lo->lo_queue, 512);
1052 invalidate_bdev(bdev);
1054 set_capacity(lo->lo_disk, 0);
1055 loop_sysfs_exit(lo);
1057 bd_set_size(bdev, 0);
1058 /* let user-space know about this change */
1059 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1061 mapping_set_gfp_mask(filp->f_mapping, gfp);
1062 lo->lo_state = Lo_unbound;
1063 /* This is safe: open() is still holding a reference. */
1064 module_put(THIS_MODULE);
1065 blk_mq_unfreeze_queue(lo->lo_queue);
1067 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1068 loop_reread_partitions(lo, bdev);
1071 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1072 loop_unprepare_queue(lo);
1073 mutex_unlock(&lo->lo_ctl_mutex);
1075 * Need not hold lo_ctl_mutex to fput backing file.
1076 * Calling fput holding lo_ctl_mutex triggers a circular
1077 * lock dependency possibility warning as fput can take
1078 * bd_mutex which is usually taken before lo_ctl_mutex.
1085 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1088 struct loop_func_table *xfer;
1089 kuid_t uid = current_uid();
1091 if (lo->lo_encrypt_key_size &&
1092 !uid_eq(lo->lo_key_owner, uid) &&
1093 !capable(CAP_SYS_ADMIN))
1095 if (lo->lo_state != Lo_bound)
1097 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1100 /* I/O need to be drained during transfer transition */
1101 blk_mq_freeze_queue(lo->lo_queue);
1103 err = loop_release_xfer(lo);
1107 if (info->lo_encrypt_type) {
1108 unsigned int type = info->lo_encrypt_type;
1110 if (type >= MAX_LO_CRYPT) {
1114 xfer = xfer_funcs[type];
1122 err = loop_init_xfer(lo, xfer, info);
1126 if (lo->lo_offset != info->lo_offset ||
1127 lo->lo_sizelimit != info->lo_sizelimit) {
1128 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1134 loop_config_discard(lo);
1136 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1137 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1138 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1139 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1143 lo->transfer = xfer->transfer;
1144 lo->ioctl = xfer->ioctl;
1146 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1147 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1148 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1150 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1151 lo->lo_init[0] = info->lo_init[0];
1152 lo->lo_init[1] = info->lo_init[1];
1153 if (info->lo_encrypt_key_size) {
1154 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1155 info->lo_encrypt_key_size);
1156 lo->lo_key_owner = uid;
1159 /* update dio if lo_offset or transfer is changed */
1160 __loop_update_dio(lo, lo->use_dio);
1163 blk_mq_unfreeze_queue(lo->lo_queue);
1165 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1166 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1167 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1168 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1169 loop_reread_partitions(lo, lo->lo_device);
1176 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1182 if (lo->lo_state != Lo_bound) {
1183 mutex_unlock(&lo->lo_ctl_mutex);
1187 memset(info, 0, sizeof(*info));
1188 info->lo_number = lo->lo_number;
1189 info->lo_offset = lo->lo_offset;
1190 info->lo_sizelimit = lo->lo_sizelimit;
1191 info->lo_flags = lo->lo_flags;
1192 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1193 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1194 info->lo_encrypt_type =
1195 lo->lo_encryption ? lo->lo_encryption->number : 0;
1196 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1197 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1198 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1199 lo->lo_encrypt_key_size);
1202 /* Drop lo_ctl_mutex while we call into the filesystem. */
1203 file = get_file(lo->lo_backing_file);
1204 mutex_unlock(&lo->lo_ctl_mutex);
1205 ret = vfs_getattr(&file->f_path, &stat, STATX_INO,
1206 AT_STATX_SYNC_AS_STAT);
1208 info->lo_device = huge_encode_dev(stat.dev);
1209 info->lo_inode = stat.ino;
1210 info->lo_rdevice = huge_encode_dev(stat.rdev);
1217 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1219 memset(info64, 0, sizeof(*info64));
1220 info64->lo_number = info->lo_number;
1221 info64->lo_device = info->lo_device;
1222 info64->lo_inode = info->lo_inode;
1223 info64->lo_rdevice = info->lo_rdevice;
1224 info64->lo_offset = info->lo_offset;
1225 info64->lo_sizelimit = 0;
1226 info64->lo_encrypt_type = info->lo_encrypt_type;
1227 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1228 info64->lo_flags = info->lo_flags;
1229 info64->lo_init[0] = info->lo_init[0];
1230 info64->lo_init[1] = info->lo_init[1];
1231 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1232 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1234 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1235 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1239 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1241 memset(info, 0, sizeof(*info));
1242 info->lo_number = info64->lo_number;
1243 info->lo_device = info64->lo_device;
1244 info->lo_inode = info64->lo_inode;
1245 info->lo_rdevice = info64->lo_rdevice;
1246 info->lo_offset = info64->lo_offset;
1247 info->lo_encrypt_type = info64->lo_encrypt_type;
1248 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1249 info->lo_flags = info64->lo_flags;
1250 info->lo_init[0] = info64->lo_init[0];
1251 info->lo_init[1] = info64->lo_init[1];
1252 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1253 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1255 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1256 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1258 /* error in case values were truncated */
1259 if (info->lo_device != info64->lo_device ||
1260 info->lo_rdevice != info64->lo_rdevice ||
1261 info->lo_inode != info64->lo_inode ||
1262 info->lo_offset != info64->lo_offset)
1269 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1271 struct loop_info info;
1272 struct loop_info64 info64;
1274 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1276 loop_info64_from_old(&info, &info64);
1277 return loop_set_status(lo, &info64);
1281 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1283 struct loop_info64 info64;
1285 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1287 return loop_set_status(lo, &info64);
1291 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1292 struct loop_info info;
1293 struct loop_info64 info64;
1297 mutex_unlock(&lo->lo_ctl_mutex);
1300 err = loop_get_status(lo, &info64);
1302 err = loop_info64_to_old(&info64, &info);
1303 if (!err && copy_to_user(arg, &info, sizeof(info)))
1310 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1311 struct loop_info64 info64;
1315 mutex_unlock(&lo->lo_ctl_mutex);
1318 err = loop_get_status(lo, &info64);
1319 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1325 static int loop_set_capacity(struct loop_device *lo)
1327 if (unlikely(lo->lo_state != Lo_bound))
1330 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1333 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1336 if (lo->lo_state != Lo_bound)
1339 __loop_update_dio(lo, !!arg);
1340 if (lo->use_dio == !!arg)
1347 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1349 if (lo->lo_state != Lo_bound)
1352 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1355 blk_mq_freeze_queue(lo->lo_queue);
1357 blk_queue_logical_block_size(lo->lo_queue, arg);
1358 blk_queue_physical_block_size(lo->lo_queue, arg);
1359 blk_queue_io_min(lo->lo_queue, arg);
1360 loop_update_dio(lo);
1362 blk_mq_unfreeze_queue(lo->lo_queue);
1367 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1368 unsigned int cmd, unsigned long arg)
1370 struct loop_device *lo = bdev->bd_disk->private_data;
1373 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1376 err = loop_set_fd(lo, mode, bdev, arg);
1378 case LOOP_CHANGE_FD:
1379 err = loop_change_fd(lo, bdev, arg);
1382 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1383 err = loop_clr_fd(lo);
1387 case LOOP_SET_STATUS:
1389 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1390 err = loop_set_status_old(lo,
1391 (struct loop_info __user *)arg);
1393 case LOOP_GET_STATUS:
1394 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1395 /* loop_get_status() unlocks lo_ctl_mutex */
1397 case LOOP_SET_STATUS64:
1399 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1400 err = loop_set_status64(lo,
1401 (struct loop_info64 __user *) arg);
1403 case LOOP_GET_STATUS64:
1404 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1405 /* loop_get_status() unlocks lo_ctl_mutex */
1407 case LOOP_SET_CAPACITY:
1409 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1410 err = loop_set_capacity(lo);
1412 case LOOP_SET_DIRECT_IO:
1414 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1415 err = loop_set_dio(lo, arg);
1417 case LOOP_SET_BLOCK_SIZE:
1419 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1420 err = loop_set_block_size(lo, arg);
1423 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1425 mutex_unlock(&lo->lo_ctl_mutex);
1431 #ifdef CONFIG_COMPAT
1432 struct compat_loop_info {
1433 compat_int_t lo_number; /* ioctl r/o */
1434 compat_dev_t lo_device; /* ioctl r/o */
1435 compat_ulong_t lo_inode; /* ioctl r/o */
1436 compat_dev_t lo_rdevice; /* ioctl r/o */
1437 compat_int_t lo_offset;
1438 compat_int_t lo_encrypt_type;
1439 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1440 compat_int_t lo_flags; /* ioctl r/o */
1441 char lo_name[LO_NAME_SIZE];
1442 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1443 compat_ulong_t lo_init[2];
1448 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1449 * - noinlined to reduce stack space usage in main part of driver
1452 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1453 struct loop_info64 *info64)
1455 struct compat_loop_info info;
1457 if (copy_from_user(&info, arg, sizeof(info)))
1460 memset(info64, 0, sizeof(*info64));
1461 info64->lo_number = info.lo_number;
1462 info64->lo_device = info.lo_device;
1463 info64->lo_inode = info.lo_inode;
1464 info64->lo_rdevice = info.lo_rdevice;
1465 info64->lo_offset = info.lo_offset;
1466 info64->lo_sizelimit = 0;
1467 info64->lo_encrypt_type = info.lo_encrypt_type;
1468 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1469 info64->lo_flags = info.lo_flags;
1470 info64->lo_init[0] = info.lo_init[0];
1471 info64->lo_init[1] = info.lo_init[1];
1472 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1473 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1475 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1476 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1481 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1482 * - noinlined to reduce stack space usage in main part of driver
1485 loop_info64_to_compat(const struct loop_info64 *info64,
1486 struct compat_loop_info __user *arg)
1488 struct compat_loop_info info;
1490 memset(&info, 0, sizeof(info));
1491 info.lo_number = info64->lo_number;
1492 info.lo_device = info64->lo_device;
1493 info.lo_inode = info64->lo_inode;
1494 info.lo_rdevice = info64->lo_rdevice;
1495 info.lo_offset = info64->lo_offset;
1496 info.lo_encrypt_type = info64->lo_encrypt_type;
1497 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1498 info.lo_flags = info64->lo_flags;
1499 info.lo_init[0] = info64->lo_init[0];
1500 info.lo_init[1] = info64->lo_init[1];
1501 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1502 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1504 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1505 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1507 /* error in case values were truncated */
1508 if (info.lo_device != info64->lo_device ||
1509 info.lo_rdevice != info64->lo_rdevice ||
1510 info.lo_inode != info64->lo_inode ||
1511 info.lo_offset != info64->lo_offset ||
1512 info.lo_init[0] != info64->lo_init[0] ||
1513 info.lo_init[1] != info64->lo_init[1])
1516 if (copy_to_user(arg, &info, sizeof(info)))
1522 loop_set_status_compat(struct loop_device *lo,
1523 const struct compat_loop_info __user *arg)
1525 struct loop_info64 info64;
1528 ret = loop_info64_from_compat(arg, &info64);
1531 return loop_set_status(lo, &info64);
1535 loop_get_status_compat(struct loop_device *lo,
1536 struct compat_loop_info __user *arg)
1538 struct loop_info64 info64;
1542 mutex_unlock(&lo->lo_ctl_mutex);
1545 err = loop_get_status(lo, &info64);
1547 err = loop_info64_to_compat(&info64, arg);
1551 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1552 unsigned int cmd, unsigned long arg)
1554 struct loop_device *lo = bdev->bd_disk->private_data;
1558 case LOOP_SET_STATUS:
1559 mutex_lock(&lo->lo_ctl_mutex);
1560 err = loop_set_status_compat(
1561 lo, (const struct compat_loop_info __user *) arg);
1562 mutex_unlock(&lo->lo_ctl_mutex);
1564 case LOOP_GET_STATUS:
1565 mutex_lock(&lo->lo_ctl_mutex);
1566 err = loop_get_status_compat(
1567 lo, (struct compat_loop_info __user *) arg);
1568 /* loop_get_status() unlocks lo_ctl_mutex */
1570 case LOOP_SET_CAPACITY:
1572 case LOOP_GET_STATUS64:
1573 case LOOP_SET_STATUS64:
1574 arg = (unsigned long) compat_ptr(arg);
1576 case LOOP_CHANGE_FD:
1577 err = lo_ioctl(bdev, mode, cmd, arg);
1587 static int lo_open(struct block_device *bdev, fmode_t mode)
1589 struct loop_device *lo;
1592 mutex_lock(&loop_index_mutex);
1593 lo = bdev->bd_disk->private_data;
1599 atomic_inc(&lo->lo_refcnt);
1601 mutex_unlock(&loop_index_mutex);
1605 static void __lo_release(struct loop_device *lo)
1609 if (atomic_dec_return(&lo->lo_refcnt))
1612 mutex_lock(&lo->lo_ctl_mutex);
1613 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1615 * In autoclear mode, stop the loop thread
1616 * and remove configuration after last close.
1618 err = loop_clr_fd(lo);
1621 } else if (lo->lo_state == Lo_bound) {
1623 * Otherwise keep thread (if running) and config,
1624 * but flush possible ongoing bios in thread.
1626 blk_mq_freeze_queue(lo->lo_queue);
1627 blk_mq_unfreeze_queue(lo->lo_queue);
1630 mutex_unlock(&lo->lo_ctl_mutex);
1633 static void lo_release(struct gendisk *disk, fmode_t mode)
1635 mutex_lock(&loop_index_mutex);
1636 __lo_release(disk->private_data);
1637 mutex_unlock(&loop_index_mutex);
1640 static const struct block_device_operations lo_fops = {
1641 .owner = THIS_MODULE,
1643 .release = lo_release,
1645 #ifdef CONFIG_COMPAT
1646 .compat_ioctl = lo_compat_ioctl,
1651 * And now the modules code and kernel interface.
1653 static int max_loop;
1654 module_param(max_loop, int, S_IRUGO);
1655 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1656 module_param(max_part, int, S_IRUGO);
1657 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1658 MODULE_LICENSE("GPL");
1659 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1661 int loop_register_transfer(struct loop_func_table *funcs)
1663 unsigned int n = funcs->number;
1665 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1667 xfer_funcs[n] = funcs;
1671 static int unregister_transfer_cb(int id, void *ptr, void *data)
1673 struct loop_device *lo = ptr;
1674 struct loop_func_table *xfer = data;
1676 mutex_lock(&lo->lo_ctl_mutex);
1677 if (lo->lo_encryption == xfer)
1678 loop_release_xfer(lo);
1679 mutex_unlock(&lo->lo_ctl_mutex);
1683 int loop_unregister_transfer(int number)
1685 unsigned int n = number;
1686 struct loop_func_table *xfer;
1688 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1691 xfer_funcs[n] = NULL;
1692 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1696 EXPORT_SYMBOL(loop_register_transfer);
1697 EXPORT_SYMBOL(loop_unregister_transfer);
1699 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1700 const struct blk_mq_queue_data *bd)
1702 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1703 struct loop_device *lo = cmd->rq->q->queuedata;
1705 blk_mq_start_request(bd->rq);
1707 if (lo->lo_state != Lo_bound)
1708 return BLK_STS_IOERR;
1710 switch (req_op(cmd->rq)) {
1712 case REQ_OP_DISCARD:
1713 case REQ_OP_WRITE_ZEROES:
1714 cmd->use_aio = false;
1717 cmd->use_aio = lo->use_dio;
1721 kthread_queue_work(&lo->worker, &cmd->work);
1726 static void loop_handle_cmd(struct loop_cmd *cmd)
1728 const bool write = op_is_write(req_op(cmd->rq));
1729 struct loop_device *lo = cmd->rq->q->queuedata;
1732 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1737 ret = do_req_filebacked(lo, cmd->rq);
1739 /* complete non-aio request */
1740 if (!cmd->use_aio || ret) {
1741 cmd->ret = ret ? -EIO : 0;
1742 blk_mq_complete_request(cmd->rq);
1746 static void loop_queue_work(struct kthread_work *work)
1748 struct loop_cmd *cmd =
1749 container_of(work, struct loop_cmd, work);
1751 loop_handle_cmd(cmd);
1754 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1755 unsigned int hctx_idx, unsigned int numa_node)
1757 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1760 kthread_init_work(&cmd->work, loop_queue_work);
1765 static const struct blk_mq_ops loop_mq_ops = {
1766 .queue_rq = loop_queue_rq,
1767 .init_request = loop_init_request,
1768 .complete = lo_complete_rq,
1771 static int loop_add(struct loop_device **l, int i)
1773 struct loop_device *lo;
1774 struct gendisk *disk;
1778 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1782 lo->lo_state = Lo_unbound;
1784 /* allocate id, if @id >= 0, we're requesting that specific id */
1786 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1790 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1797 lo->tag_set.ops = &loop_mq_ops;
1798 lo->tag_set.nr_hw_queues = 1;
1799 lo->tag_set.queue_depth = 128;
1800 lo->tag_set.numa_node = NUMA_NO_NODE;
1801 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1802 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1803 lo->tag_set.driver_data = lo;
1805 err = blk_mq_alloc_tag_set(&lo->tag_set);
1809 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1810 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1811 err = PTR_ERR(lo->lo_queue);
1812 goto out_cleanup_tags;
1814 lo->lo_queue->queuedata = lo;
1816 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1819 * By default, we do buffer IO, so it doesn't make sense to enable
1820 * merge because the I/O submitted to backing file is handled page by
1821 * page. For directio mode, merge does help to dispatch bigger request
1822 * to underlayer disk. We will enable merge once directio is enabled.
1824 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1827 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1829 goto out_free_queue;
1832 * Disable partition scanning by default. The in-kernel partition
1833 * scanning can be requested individually per-device during its
1834 * setup. Userspace can always add and remove partitions from all
1835 * devices. The needed partition minors are allocated from the
1836 * extended minor space, the main loop device numbers will continue
1837 * to match the loop minors, regardless of the number of partitions
1840 * If max_part is given, partition scanning is globally enabled for
1841 * all loop devices. The minors for the main loop devices will be
1842 * multiples of max_part.
1844 * Note: Global-for-all-devices, set-only-at-init, read-only module
1845 * parameteters like 'max_loop' and 'max_part' make things needlessly
1846 * complicated, are too static, inflexible and may surprise
1847 * userspace tools. Parameters like this in general should be avoided.
1850 disk->flags |= GENHD_FL_NO_PART_SCAN;
1851 disk->flags |= GENHD_FL_EXT_DEVT;
1852 mutex_init(&lo->lo_ctl_mutex);
1853 atomic_set(&lo->lo_refcnt, 0);
1855 spin_lock_init(&lo->lo_lock);
1856 disk->major = LOOP_MAJOR;
1857 disk->first_minor = i << part_shift;
1858 disk->fops = &lo_fops;
1859 disk->private_data = lo;
1860 disk->queue = lo->lo_queue;
1861 sprintf(disk->disk_name, "loop%d", i);
1864 return lo->lo_number;
1867 blk_cleanup_queue(lo->lo_queue);
1869 blk_mq_free_tag_set(&lo->tag_set);
1871 idr_remove(&loop_index_idr, i);
1878 static void loop_remove(struct loop_device *lo)
1880 blk_cleanup_queue(lo->lo_queue);
1881 del_gendisk(lo->lo_disk);
1882 blk_mq_free_tag_set(&lo->tag_set);
1883 put_disk(lo->lo_disk);
1887 static int find_free_cb(int id, void *ptr, void *data)
1889 struct loop_device *lo = ptr;
1890 struct loop_device **l = data;
1892 if (lo->lo_state == Lo_unbound) {
1899 static int loop_lookup(struct loop_device **l, int i)
1901 struct loop_device *lo;
1907 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1910 ret = lo->lo_number;
1915 /* lookup and return a specific i */
1916 lo = idr_find(&loop_index_idr, i);
1919 ret = lo->lo_number;
1925 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1927 struct loop_device *lo;
1928 struct kobject *kobj;
1931 mutex_lock(&loop_index_mutex);
1932 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1934 err = loop_add(&lo, MINOR(dev) >> part_shift);
1938 kobj = get_disk(lo->lo_disk);
1939 mutex_unlock(&loop_index_mutex);
1945 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1948 struct loop_device *lo;
1951 mutex_lock(&loop_index_mutex);
1954 ret = loop_lookup(&lo, parm);
1959 ret = loop_add(&lo, parm);
1961 case LOOP_CTL_REMOVE:
1962 ret = loop_lookup(&lo, parm);
1965 mutex_lock(&lo->lo_ctl_mutex);
1966 if (lo->lo_state != Lo_unbound) {
1968 mutex_unlock(&lo->lo_ctl_mutex);
1971 if (atomic_read(&lo->lo_refcnt) > 0) {
1973 mutex_unlock(&lo->lo_ctl_mutex);
1976 lo->lo_disk->private_data = NULL;
1977 mutex_unlock(&lo->lo_ctl_mutex);
1978 idr_remove(&loop_index_idr, lo->lo_number);
1981 case LOOP_CTL_GET_FREE:
1982 ret = loop_lookup(&lo, -1);
1985 ret = loop_add(&lo, -1);
1987 mutex_unlock(&loop_index_mutex);
1992 static const struct file_operations loop_ctl_fops = {
1993 .open = nonseekable_open,
1994 .unlocked_ioctl = loop_control_ioctl,
1995 .compat_ioctl = loop_control_ioctl,
1996 .owner = THIS_MODULE,
1997 .llseek = noop_llseek,
2000 static struct miscdevice loop_misc = {
2001 .minor = LOOP_CTRL_MINOR,
2002 .name = "loop-control",
2003 .fops = &loop_ctl_fops,
2006 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2007 MODULE_ALIAS("devname:loop-control");
2009 static int __init loop_init(void)
2012 unsigned long range;
2013 struct loop_device *lo;
2018 part_shift = fls(max_part);
2021 * Adjust max_part according to part_shift as it is exported
2022 * to user space so that user can decide correct minor number
2023 * if [s]he want to create more devices.
2025 * Note that -1 is required because partition 0 is reserved
2026 * for the whole disk.
2028 max_part = (1UL << part_shift) - 1;
2031 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2036 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2042 * If max_loop is specified, create that many devices upfront.
2043 * This also becomes a hard limit. If max_loop is not specified,
2044 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2045 * init time. Loop devices can be requested on-demand with the
2046 * /dev/loop-control interface, or be instantiated by accessing
2047 * a 'dead' device node.
2051 range = max_loop << part_shift;
2053 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2054 range = 1UL << MINORBITS;
2057 err = misc_register(&loop_misc);
2062 if (register_blkdev(LOOP_MAJOR, "loop")) {
2067 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2068 THIS_MODULE, loop_probe, NULL, NULL);
2070 /* pre-create number of devices given by config or max_loop */
2071 mutex_lock(&loop_index_mutex);
2072 for (i = 0; i < nr; i++)
2074 mutex_unlock(&loop_index_mutex);
2076 printk(KERN_INFO "loop: module loaded\n");
2080 misc_deregister(&loop_misc);
2085 static int loop_exit_cb(int id, void *ptr, void *data)
2087 struct loop_device *lo = ptr;
2093 static void __exit loop_exit(void)
2095 unsigned long range;
2097 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2099 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2100 idr_destroy(&loop_index_idr);
2102 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2103 unregister_blkdev(LOOP_MAJOR, "loop");
2105 misc_deregister(&loop_misc);
2108 module_init(loop_init);
2109 module_exit(loop_exit);
2112 static int __init max_loop_setup(char *str)
2114 max_loop = simple_strtol(str, NULL, 0);
2118 __setup("max_loop=", max_loop_setup);