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 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
219 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
220 blk_mq_unfreeze_queue(lo->lo_queue);
224 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
226 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
227 sector_t x = (sector_t)size;
228 struct block_device *bdev = lo->lo_device;
230 if (unlikely((loff_t)x != size))
232 if (lo->lo_offset != offset)
233 lo->lo_offset = offset;
234 if (lo->lo_sizelimit != sizelimit)
235 lo->lo_sizelimit = sizelimit;
236 set_capacity(lo->lo_disk, x);
237 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
238 /* let user-space know about the new size */
239 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
244 lo_do_transfer(struct loop_device *lo, int cmd,
245 struct page *rpage, unsigned roffs,
246 struct page *lpage, unsigned loffs,
247 int size, sector_t rblock)
251 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
255 printk_ratelimited(KERN_ERR
256 "loop: Transfer error at byte offset %llu, length %i.\n",
257 (unsigned long long)rblock << 9, size);
261 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
266 iov_iter_bvec(&i, ITER_BVEC, bvec, 1, bvec->bv_len);
268 file_start_write(file);
269 bw = vfs_iter_write(file, &i, ppos, 0);
270 file_end_write(file);
272 if (likely(bw == bvec->bv_len))
275 printk_ratelimited(KERN_ERR
276 "loop: Write error at byte offset %llu, length %i.\n",
277 (unsigned long long)*ppos, bvec->bv_len);
283 static int lo_write_simple(struct loop_device *lo, struct request *rq,
287 struct req_iterator iter;
290 rq_for_each_segment(bvec, rq, iter) {
291 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
301 * This is the slow, transforming version that needs to double buffer the
302 * data as it cannot do the transformations in place without having direct
303 * access to the destination pages of the backing file.
305 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
308 struct bio_vec bvec, b;
309 struct req_iterator iter;
313 page = alloc_page(GFP_NOIO);
317 rq_for_each_segment(bvec, rq, iter) {
318 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
319 bvec.bv_offset, bvec.bv_len, pos >> 9);
325 b.bv_len = bvec.bv_len;
326 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
335 static int lo_read_simple(struct loop_device *lo, struct request *rq,
339 struct req_iterator iter;
343 rq_for_each_segment(bvec, rq, iter) {
344 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
345 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
349 flush_dcache_page(bvec.bv_page);
351 if (len != bvec.bv_len) {
354 __rq_for_each_bio(bio, rq)
364 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
367 struct bio_vec bvec, b;
368 struct req_iterator iter;
374 page = alloc_page(GFP_NOIO);
378 rq_for_each_segment(bvec, rq, iter) {
383 b.bv_len = bvec.bv_len;
385 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
386 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
392 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
393 bvec.bv_offset, len, offset >> 9);
397 flush_dcache_page(bvec.bv_page);
399 if (len != bvec.bv_len) {
402 __rq_for_each_bio(bio, rq)
414 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
417 * We use punch hole to reclaim the free space used by the
418 * image a.k.a. discard. However we do not support discard if
419 * encryption is enabled, because it may give an attacker
420 * useful information.
422 struct file *file = lo->lo_backing_file;
423 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
426 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
431 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
432 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
438 static int lo_req_flush(struct loop_device *lo, struct request *rq)
440 struct file *file = lo->lo_backing_file;
441 int ret = vfs_fsync(file, 0);
442 if (unlikely(ret && ret != -EINVAL))
448 static void lo_complete_rq(struct request *rq)
450 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
452 if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio &&
453 cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) {
454 struct bio *bio = cmd->rq->bio;
456 bio_advance(bio, cmd->ret);
460 blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK);
463 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
465 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
468 blk_mq_complete_request(cmd->rq);
471 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
474 struct iov_iter iter;
475 struct bio_vec *bvec;
476 struct bio *bio = cmd->rq->bio;
477 struct file *file = lo->lo_backing_file;
480 /* nomerge for loop request queue */
481 WARN_ON(cmd->rq->bio != cmd->rq->biotail);
483 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
484 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
485 bio_segments(bio), blk_rq_bytes(cmd->rq));
487 * This bio may be started from the middle of the 'bvec'
488 * because of bio splitting, so offset from the bvec must
489 * be passed to iov iterator
491 iter.iov_offset = bio->bi_iter.bi_bvec_done;
493 cmd->iocb.ki_pos = pos;
494 cmd->iocb.ki_filp = file;
495 cmd->iocb.ki_complete = lo_rw_aio_complete;
496 cmd->iocb.ki_flags = IOCB_DIRECT;
499 ret = call_write_iter(file, &cmd->iocb, &iter);
501 ret = call_read_iter(file, &cmd->iocb, &iter);
503 if (ret != -EIOCBQUEUED)
504 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
508 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
510 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
511 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
514 * lo_write_simple and lo_read_simple should have been covered
515 * by io submit style function like lo_rw_aio(), one blocker
516 * is that lo_read_simple() need to call flush_dcache_page after
517 * the page is written from kernel, and it isn't easy to handle
518 * this in io submit style function which submits all segments
519 * of the req at one time. And direct read IO doesn't need to
520 * run flush_dcache_page().
522 switch (req_op(rq)) {
524 return lo_req_flush(lo, rq);
526 case REQ_OP_WRITE_ZEROES:
527 return lo_discard(lo, rq, pos);
530 return lo_write_transfer(lo, rq, pos);
531 else if (cmd->use_aio)
532 return lo_rw_aio(lo, cmd, pos, WRITE);
534 return lo_write_simple(lo, rq, pos);
537 return lo_read_transfer(lo, rq, pos);
538 else if (cmd->use_aio)
539 return lo_rw_aio(lo, cmd, pos, READ);
541 return lo_read_simple(lo, rq, pos);
549 struct switch_request {
551 struct completion wait;
554 static inline void loop_update_dio(struct loop_device *lo)
556 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
561 * Do the actual switch; called from the BIO completion routine
563 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
565 struct file *file = p->file;
566 struct file *old_file = lo->lo_backing_file;
567 struct address_space *mapping;
569 /* if no new file, only flush of queued bios requested */
573 mapping = file->f_mapping;
574 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
575 lo->lo_backing_file = file;
576 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
577 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
578 lo->old_gfp_mask = mapping_gfp_mask(mapping);
579 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
584 * loop_switch performs the hard work of switching a backing store.
585 * First it needs to flush existing IO, it does this by sending a magic
586 * BIO down the pipe. The completion of this BIO does the actual switch.
588 static int loop_switch(struct loop_device *lo, struct file *file)
590 struct switch_request w;
594 /* freeze queue and wait for completion of scheduled requests */
595 blk_mq_freeze_queue(lo->lo_queue);
597 /* do the switch action */
598 do_loop_switch(lo, &w);
601 blk_mq_unfreeze_queue(lo->lo_queue);
607 * Helper to flush the IOs in loop, but keeping loop thread running
609 static int loop_flush(struct loop_device *lo)
611 /* loop not yet configured, no running thread, nothing to flush */
612 if (lo->lo_state != Lo_bound)
614 return loop_switch(lo, NULL);
617 static void loop_reread_partitions(struct loop_device *lo,
618 struct block_device *bdev)
623 * bd_mutex has been held already in release path, so don't
624 * acquire it if this function is called in such case.
626 * If the reread partition isn't from release path, lo_refcnt
627 * must be at least one and it can only become zero when the
628 * current holder is released.
630 if (!atomic_read(&lo->lo_refcnt))
631 rc = __blkdev_reread_part(bdev);
633 rc = blkdev_reread_part(bdev);
635 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
636 __func__, lo->lo_number, lo->lo_file_name, rc);
640 * loop_change_fd switched the backing store of a loopback device to
641 * a new file. This is useful for operating system installers to free up
642 * the original file and in High Availability environments to switch to
643 * an alternative location for the content in case of server meltdown.
644 * This can only work if the loop device is used read-only, and if the
645 * new backing store is the same size and type as the old backing store.
647 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
650 struct file *file, *old_file;
655 if (lo->lo_state != Lo_bound)
658 /* the loop device has to be read-only */
660 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
668 inode = file->f_mapping->host;
669 old_file = lo->lo_backing_file;
673 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
676 /* size of the new backing store needs to be the same */
677 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
681 error = loop_switch(lo, file);
686 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
687 loop_reread_partitions(lo, bdev);
696 static inline int is_loop_device(struct file *file)
698 struct inode *i = file->f_mapping->host;
700 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
703 /* loop sysfs attributes */
705 static ssize_t loop_attr_show(struct device *dev, char *page,
706 ssize_t (*callback)(struct loop_device *, char *))
708 struct gendisk *disk = dev_to_disk(dev);
709 struct loop_device *lo = disk->private_data;
711 return callback(lo, page);
714 #define LOOP_ATTR_RO(_name) \
715 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
716 static ssize_t loop_attr_do_show_##_name(struct device *d, \
717 struct device_attribute *attr, char *b) \
719 return loop_attr_show(d, b, loop_attr_##_name##_show); \
721 static struct device_attribute loop_attr_##_name = \
722 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
724 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
729 spin_lock_irq(&lo->lo_lock);
730 if (lo->lo_backing_file)
731 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
732 spin_unlock_irq(&lo->lo_lock);
734 if (IS_ERR_OR_NULL(p))
738 memmove(buf, p, ret);
746 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
748 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
751 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
753 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
756 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
758 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
760 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
763 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
765 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
767 return sprintf(buf, "%s\n", partscan ? "1" : "0");
770 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
772 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
774 return sprintf(buf, "%s\n", dio ? "1" : "0");
777 LOOP_ATTR_RO(backing_file);
778 LOOP_ATTR_RO(offset);
779 LOOP_ATTR_RO(sizelimit);
780 LOOP_ATTR_RO(autoclear);
781 LOOP_ATTR_RO(partscan);
784 static struct attribute *loop_attrs[] = {
785 &loop_attr_backing_file.attr,
786 &loop_attr_offset.attr,
787 &loop_attr_sizelimit.attr,
788 &loop_attr_autoclear.attr,
789 &loop_attr_partscan.attr,
794 static struct attribute_group loop_attribute_group = {
799 static int loop_sysfs_init(struct loop_device *lo)
801 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
802 &loop_attribute_group);
805 static void loop_sysfs_exit(struct loop_device *lo)
807 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
808 &loop_attribute_group);
811 static void loop_config_discard(struct loop_device *lo)
813 struct file *file = lo->lo_backing_file;
814 struct inode *inode = file->f_mapping->host;
815 struct request_queue *q = lo->lo_queue;
818 * We use punch hole to reclaim the free space used by the
819 * image a.k.a. discard. However we do not support discard if
820 * encryption is enabled, because it may give an attacker
821 * useful information.
823 if ((!file->f_op->fallocate) ||
824 lo->lo_encrypt_key_size) {
825 q->limits.discard_granularity = 0;
826 q->limits.discard_alignment = 0;
827 blk_queue_max_discard_sectors(q, 0);
828 blk_queue_max_write_zeroes_sectors(q, 0);
829 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
833 q->limits.discard_granularity = inode->i_sb->s_blocksize;
834 q->limits.discard_alignment = 0;
836 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
837 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
838 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
841 static void loop_unprepare_queue(struct loop_device *lo)
843 kthread_flush_worker(&lo->worker);
844 kthread_stop(lo->worker_task);
847 static int loop_kthread_worker_fn(void *worker_ptr)
849 current->flags |= PF_LESS_THROTTLE;
850 return kthread_worker_fn(worker_ptr);
853 static int loop_prepare_queue(struct loop_device *lo)
855 kthread_init_worker(&lo->worker);
856 lo->worker_task = kthread_run(loop_kthread_worker_fn,
857 &lo->worker, "loop%d", lo->lo_number);
858 if (IS_ERR(lo->worker_task))
860 set_user_nice(lo->worker_task, MIN_NICE);
864 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
865 struct block_device *bdev, unsigned int arg)
867 struct file *file, *f;
869 struct address_space *mapping;
870 unsigned lo_blocksize;
875 /* This is safe, since we have a reference from open(). */
876 __module_get(THIS_MODULE);
884 if (lo->lo_state != Lo_unbound)
887 /* Avoid recursion */
889 while (is_loop_device(f)) {
890 struct loop_device *l;
892 if (f->f_mapping->host->i_bdev == bdev)
895 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
896 if (l->lo_state == Lo_unbound) {
900 f = l->lo_backing_file;
903 mapping = file->f_mapping;
904 inode = mapping->host;
907 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
910 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
911 !file->f_op->write_iter)
912 lo_flags |= LO_FLAGS_READ_ONLY;
914 lo_blocksize = S_ISBLK(inode->i_mode) ?
915 inode->i_bdev->bd_block_size : PAGE_SIZE;
918 size = get_loop_size(lo, file);
919 if ((loff_t)(sector_t)size != size)
921 error = loop_prepare_queue(lo);
927 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
930 lo->lo_blocksize = lo_blocksize;
931 lo->lo_device = bdev;
932 lo->lo_flags = lo_flags;
933 lo->lo_backing_file = file;
936 lo->lo_sizelimit = 0;
937 lo->old_gfp_mask = mapping_gfp_mask(mapping);
938 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
940 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
941 blk_queue_write_cache(lo->lo_queue, true, false);
944 set_capacity(lo->lo_disk, size);
945 bd_set_size(bdev, size << 9);
947 /* let user-space know about the new size */
948 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
950 set_blocksize(bdev, lo_blocksize);
952 lo->lo_state = Lo_bound;
954 lo->lo_flags |= LO_FLAGS_PARTSCAN;
955 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
956 loop_reread_partitions(lo, bdev);
958 /* Grab the block_device to prevent its destruction after we
959 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
967 /* This is safe: open() is still holding a reference. */
968 module_put(THIS_MODULE);
973 loop_release_xfer(struct loop_device *lo)
976 struct loop_func_table *xfer = lo->lo_encryption;
980 err = xfer->release(lo);
982 lo->lo_encryption = NULL;
983 module_put(xfer->owner);
989 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
990 const struct loop_info64 *i)
995 struct module *owner = xfer->owner;
997 if (!try_module_get(owner))
1000 err = xfer->init(lo, i);
1004 lo->lo_encryption = xfer;
1009 static int loop_clr_fd(struct loop_device *lo)
1011 struct file *filp = lo->lo_backing_file;
1012 gfp_t gfp = lo->old_gfp_mask;
1013 struct block_device *bdev = lo->lo_device;
1015 if (lo->lo_state != Lo_bound)
1019 * If we've explicitly asked to tear down the loop device,
1020 * and it has an elevated reference count, set it for auto-teardown when
1021 * the last reference goes away. This stops $!~#$@ udev from
1022 * preventing teardown because it decided that it needs to run blkid on
1023 * the loopback device whenever they appear. xfstests is notorious for
1024 * failing tests because blkid via udev races with a losetup
1025 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1026 * command to fail with EBUSY.
1028 if (atomic_read(&lo->lo_refcnt) > 1) {
1029 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1030 mutex_unlock(&lo->lo_ctl_mutex);
1037 /* freeze request queue during the transition */
1038 blk_mq_freeze_queue(lo->lo_queue);
1040 spin_lock_irq(&lo->lo_lock);
1041 lo->lo_state = Lo_rundown;
1042 lo->lo_backing_file = NULL;
1043 spin_unlock_irq(&lo->lo_lock);
1045 loop_release_xfer(lo);
1046 lo->transfer = NULL;
1048 lo->lo_device = NULL;
1049 lo->lo_encryption = NULL;
1051 lo->lo_sizelimit = 0;
1052 lo->lo_encrypt_key_size = 0;
1053 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1054 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1055 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1058 invalidate_bdev(bdev);
1060 set_capacity(lo->lo_disk, 0);
1061 loop_sysfs_exit(lo);
1063 bd_set_size(bdev, 0);
1064 /* let user-space know about this change */
1065 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1067 mapping_set_gfp_mask(filp->f_mapping, gfp);
1068 lo->lo_state = Lo_unbound;
1069 /* This is safe: open() is still holding a reference. */
1070 module_put(THIS_MODULE);
1071 blk_mq_unfreeze_queue(lo->lo_queue);
1073 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1074 loop_reread_partitions(lo, bdev);
1077 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1078 loop_unprepare_queue(lo);
1079 mutex_unlock(&lo->lo_ctl_mutex);
1081 * Need not hold lo_ctl_mutex to fput backing file.
1082 * Calling fput holding lo_ctl_mutex triggers a circular
1083 * lock dependency possibility warning as fput can take
1084 * bd_mutex which is usually taken before lo_ctl_mutex.
1091 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1094 struct loop_func_table *xfer;
1095 kuid_t uid = current_uid();
1097 if (lo->lo_encrypt_key_size &&
1098 !uid_eq(lo->lo_key_owner, uid) &&
1099 !capable(CAP_SYS_ADMIN))
1101 if (lo->lo_state != Lo_bound)
1103 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1106 /* I/O need to be drained during transfer transition */
1107 blk_mq_freeze_queue(lo->lo_queue);
1109 err = loop_release_xfer(lo);
1113 if (info->lo_encrypt_type) {
1114 unsigned int type = info->lo_encrypt_type;
1116 if (type >= MAX_LO_CRYPT)
1118 xfer = xfer_funcs[type];
1124 err = loop_init_xfer(lo, xfer, info);
1128 if (lo->lo_offset != info->lo_offset ||
1129 lo->lo_sizelimit != info->lo_sizelimit) {
1130 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1136 loop_config_discard(lo);
1138 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1139 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1140 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1141 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1145 lo->transfer = xfer->transfer;
1146 lo->ioctl = xfer->ioctl;
1148 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1149 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1150 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1152 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1153 lo->lo_init[0] = info->lo_init[0];
1154 lo->lo_init[1] = info->lo_init[1];
1155 if (info->lo_encrypt_key_size) {
1156 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1157 info->lo_encrypt_key_size);
1158 lo->lo_key_owner = uid;
1161 /* update dio if lo_offset or transfer is changed */
1162 __loop_update_dio(lo, lo->use_dio);
1165 blk_mq_unfreeze_queue(lo->lo_queue);
1167 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1168 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1169 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1170 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1171 loop_reread_partitions(lo, lo->lo_device);
1178 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1180 struct file *file = lo->lo_backing_file;
1184 if (lo->lo_state != Lo_bound)
1186 error = vfs_getattr(&file->f_path, &stat,
1187 STATX_INO, AT_STATX_SYNC_AS_STAT);
1190 memset(info, 0, sizeof(*info));
1191 info->lo_number = lo->lo_number;
1192 info->lo_device = huge_encode_dev(stat.dev);
1193 info->lo_inode = stat.ino;
1194 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1195 info->lo_offset = lo->lo_offset;
1196 info->lo_sizelimit = lo->lo_sizelimit;
1197 info->lo_flags = lo->lo_flags;
1198 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1199 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1200 info->lo_encrypt_type =
1201 lo->lo_encryption ? lo->lo_encryption->number : 0;
1202 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1203 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1204 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1205 lo->lo_encrypt_key_size);
1211 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1213 memset(info64, 0, sizeof(*info64));
1214 info64->lo_number = info->lo_number;
1215 info64->lo_device = info->lo_device;
1216 info64->lo_inode = info->lo_inode;
1217 info64->lo_rdevice = info->lo_rdevice;
1218 info64->lo_offset = info->lo_offset;
1219 info64->lo_sizelimit = 0;
1220 info64->lo_encrypt_type = info->lo_encrypt_type;
1221 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1222 info64->lo_flags = info->lo_flags;
1223 info64->lo_init[0] = info->lo_init[0];
1224 info64->lo_init[1] = info->lo_init[1];
1225 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1226 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1228 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1229 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1233 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1235 memset(info, 0, sizeof(*info));
1236 info->lo_number = info64->lo_number;
1237 info->lo_device = info64->lo_device;
1238 info->lo_inode = info64->lo_inode;
1239 info->lo_rdevice = info64->lo_rdevice;
1240 info->lo_offset = info64->lo_offset;
1241 info->lo_encrypt_type = info64->lo_encrypt_type;
1242 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1243 info->lo_flags = info64->lo_flags;
1244 info->lo_init[0] = info64->lo_init[0];
1245 info->lo_init[1] = info64->lo_init[1];
1246 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1247 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1249 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1250 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1252 /* error in case values were truncated */
1253 if (info->lo_device != info64->lo_device ||
1254 info->lo_rdevice != info64->lo_rdevice ||
1255 info->lo_inode != info64->lo_inode ||
1256 info->lo_offset != info64->lo_offset)
1263 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1265 struct loop_info info;
1266 struct loop_info64 info64;
1268 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1270 loop_info64_from_old(&info, &info64);
1271 return loop_set_status(lo, &info64);
1275 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1277 struct loop_info64 info64;
1279 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1281 return loop_set_status(lo, &info64);
1285 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1286 struct loop_info info;
1287 struct loop_info64 info64;
1293 err = loop_get_status(lo, &info64);
1295 err = loop_info64_to_old(&info64, &info);
1296 if (!err && copy_to_user(arg, &info, sizeof(info)))
1303 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1304 struct loop_info64 info64;
1310 err = loop_get_status(lo, &info64);
1311 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1317 static int loop_set_capacity(struct loop_device *lo)
1319 if (unlikely(lo->lo_state != Lo_bound))
1322 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1325 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1328 if (lo->lo_state != Lo_bound)
1331 __loop_update_dio(lo, !!arg);
1332 if (lo->use_dio == !!arg)
1339 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1340 unsigned int cmd, unsigned long arg)
1342 struct loop_device *lo = bdev->bd_disk->private_data;
1345 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1348 err = loop_set_fd(lo, mode, bdev, arg);
1350 case LOOP_CHANGE_FD:
1351 err = loop_change_fd(lo, bdev, arg);
1354 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1355 err = loop_clr_fd(lo);
1359 case LOOP_SET_STATUS:
1361 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1362 err = loop_set_status_old(lo,
1363 (struct loop_info __user *)arg);
1365 case LOOP_GET_STATUS:
1366 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1368 case LOOP_SET_STATUS64:
1370 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1371 err = loop_set_status64(lo,
1372 (struct loop_info64 __user *) arg);
1374 case LOOP_GET_STATUS64:
1375 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1377 case LOOP_SET_CAPACITY:
1379 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1380 err = loop_set_capacity(lo);
1382 case LOOP_SET_DIRECT_IO:
1384 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1385 err = loop_set_dio(lo, arg);
1388 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1390 mutex_unlock(&lo->lo_ctl_mutex);
1396 #ifdef CONFIG_COMPAT
1397 struct compat_loop_info {
1398 compat_int_t lo_number; /* ioctl r/o */
1399 compat_dev_t lo_device; /* ioctl r/o */
1400 compat_ulong_t lo_inode; /* ioctl r/o */
1401 compat_dev_t lo_rdevice; /* ioctl r/o */
1402 compat_int_t lo_offset;
1403 compat_int_t lo_encrypt_type;
1404 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1405 compat_int_t lo_flags; /* ioctl r/o */
1406 char lo_name[LO_NAME_SIZE];
1407 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1408 compat_ulong_t lo_init[2];
1413 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1414 * - noinlined to reduce stack space usage in main part of driver
1417 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1418 struct loop_info64 *info64)
1420 struct compat_loop_info info;
1422 if (copy_from_user(&info, arg, sizeof(info)))
1425 memset(info64, 0, sizeof(*info64));
1426 info64->lo_number = info.lo_number;
1427 info64->lo_device = info.lo_device;
1428 info64->lo_inode = info.lo_inode;
1429 info64->lo_rdevice = info.lo_rdevice;
1430 info64->lo_offset = info.lo_offset;
1431 info64->lo_sizelimit = 0;
1432 info64->lo_encrypt_type = info.lo_encrypt_type;
1433 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1434 info64->lo_flags = info.lo_flags;
1435 info64->lo_init[0] = info.lo_init[0];
1436 info64->lo_init[1] = info.lo_init[1];
1437 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1438 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1440 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1441 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1446 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1447 * - noinlined to reduce stack space usage in main part of driver
1450 loop_info64_to_compat(const struct loop_info64 *info64,
1451 struct compat_loop_info __user *arg)
1453 struct compat_loop_info info;
1455 memset(&info, 0, sizeof(info));
1456 info.lo_number = info64->lo_number;
1457 info.lo_device = info64->lo_device;
1458 info.lo_inode = info64->lo_inode;
1459 info.lo_rdevice = info64->lo_rdevice;
1460 info.lo_offset = info64->lo_offset;
1461 info.lo_encrypt_type = info64->lo_encrypt_type;
1462 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1463 info.lo_flags = info64->lo_flags;
1464 info.lo_init[0] = info64->lo_init[0];
1465 info.lo_init[1] = info64->lo_init[1];
1466 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1467 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1469 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1470 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1472 /* error in case values were truncated */
1473 if (info.lo_device != info64->lo_device ||
1474 info.lo_rdevice != info64->lo_rdevice ||
1475 info.lo_inode != info64->lo_inode ||
1476 info.lo_offset != info64->lo_offset ||
1477 info.lo_init[0] != info64->lo_init[0] ||
1478 info.lo_init[1] != info64->lo_init[1])
1481 if (copy_to_user(arg, &info, sizeof(info)))
1487 loop_set_status_compat(struct loop_device *lo,
1488 const struct compat_loop_info __user *arg)
1490 struct loop_info64 info64;
1493 ret = loop_info64_from_compat(arg, &info64);
1496 return loop_set_status(lo, &info64);
1500 loop_get_status_compat(struct loop_device *lo,
1501 struct compat_loop_info __user *arg)
1503 struct loop_info64 info64;
1509 err = loop_get_status(lo, &info64);
1511 err = loop_info64_to_compat(&info64, arg);
1515 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1516 unsigned int cmd, unsigned long arg)
1518 struct loop_device *lo = bdev->bd_disk->private_data;
1522 case LOOP_SET_STATUS:
1523 mutex_lock(&lo->lo_ctl_mutex);
1524 err = loop_set_status_compat(
1525 lo, (const struct compat_loop_info __user *) arg);
1526 mutex_unlock(&lo->lo_ctl_mutex);
1528 case LOOP_GET_STATUS:
1529 mutex_lock(&lo->lo_ctl_mutex);
1530 err = loop_get_status_compat(
1531 lo, (struct compat_loop_info __user *) arg);
1532 mutex_unlock(&lo->lo_ctl_mutex);
1534 case LOOP_SET_CAPACITY:
1536 case LOOP_GET_STATUS64:
1537 case LOOP_SET_STATUS64:
1538 arg = (unsigned long) compat_ptr(arg);
1540 case LOOP_CHANGE_FD:
1541 err = lo_ioctl(bdev, mode, cmd, arg);
1551 static int lo_open(struct block_device *bdev, fmode_t mode)
1553 struct loop_device *lo;
1556 mutex_lock(&loop_index_mutex);
1557 lo = bdev->bd_disk->private_data;
1563 atomic_inc(&lo->lo_refcnt);
1565 mutex_unlock(&loop_index_mutex);
1569 static void lo_release(struct gendisk *disk, fmode_t mode)
1571 struct loop_device *lo = disk->private_data;
1574 if (atomic_dec_return(&lo->lo_refcnt))
1577 mutex_lock(&lo->lo_ctl_mutex);
1578 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1580 * In autoclear mode, stop the loop thread
1581 * and remove configuration after last close.
1583 err = loop_clr_fd(lo);
1588 * Otherwise keep thread (if running) and config,
1589 * but flush possible ongoing bios in thread.
1594 mutex_unlock(&lo->lo_ctl_mutex);
1597 static const struct block_device_operations lo_fops = {
1598 .owner = THIS_MODULE,
1600 .release = lo_release,
1602 #ifdef CONFIG_COMPAT
1603 .compat_ioctl = lo_compat_ioctl,
1608 * And now the modules code and kernel interface.
1610 static int max_loop;
1611 module_param(max_loop, int, S_IRUGO);
1612 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1613 module_param(max_part, int, S_IRUGO);
1614 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1615 MODULE_LICENSE("GPL");
1616 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1618 int loop_register_transfer(struct loop_func_table *funcs)
1620 unsigned int n = funcs->number;
1622 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1624 xfer_funcs[n] = funcs;
1628 static int unregister_transfer_cb(int id, void *ptr, void *data)
1630 struct loop_device *lo = ptr;
1631 struct loop_func_table *xfer = data;
1633 mutex_lock(&lo->lo_ctl_mutex);
1634 if (lo->lo_encryption == xfer)
1635 loop_release_xfer(lo);
1636 mutex_unlock(&lo->lo_ctl_mutex);
1640 int loop_unregister_transfer(int number)
1642 unsigned int n = number;
1643 struct loop_func_table *xfer;
1645 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1648 xfer_funcs[n] = NULL;
1649 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1653 EXPORT_SYMBOL(loop_register_transfer);
1654 EXPORT_SYMBOL(loop_unregister_transfer);
1656 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1657 const struct blk_mq_queue_data *bd)
1659 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1660 struct loop_device *lo = cmd->rq->q->queuedata;
1662 blk_mq_start_request(bd->rq);
1664 if (lo->lo_state != Lo_bound)
1665 return BLK_STS_IOERR;
1667 switch (req_op(cmd->rq)) {
1669 case REQ_OP_DISCARD:
1670 case REQ_OP_WRITE_ZEROES:
1671 cmd->use_aio = false;
1674 cmd->use_aio = lo->use_dio;
1678 kthread_queue_work(&lo->worker, &cmd->work);
1683 static void loop_handle_cmd(struct loop_cmd *cmd)
1685 const bool write = op_is_write(req_op(cmd->rq));
1686 struct loop_device *lo = cmd->rq->q->queuedata;
1689 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1694 ret = do_req_filebacked(lo, cmd->rq);
1696 /* complete non-aio request */
1697 if (!cmd->use_aio || ret) {
1698 cmd->ret = ret ? -EIO : 0;
1699 blk_mq_complete_request(cmd->rq);
1703 static void loop_queue_work(struct kthread_work *work)
1705 struct loop_cmd *cmd =
1706 container_of(work, struct loop_cmd, work);
1708 loop_handle_cmd(cmd);
1711 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1712 unsigned int hctx_idx, unsigned int numa_node)
1714 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1717 kthread_init_work(&cmd->work, loop_queue_work);
1722 static const struct blk_mq_ops loop_mq_ops = {
1723 .queue_rq = loop_queue_rq,
1724 .init_request = loop_init_request,
1725 .complete = lo_complete_rq,
1728 static int loop_add(struct loop_device **l, int i)
1730 struct loop_device *lo;
1731 struct gendisk *disk;
1735 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1739 lo->lo_state = Lo_unbound;
1741 /* allocate id, if @id >= 0, we're requesting that specific id */
1743 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1747 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1754 lo->tag_set.ops = &loop_mq_ops;
1755 lo->tag_set.nr_hw_queues = 1;
1756 lo->tag_set.queue_depth = 128;
1757 lo->tag_set.numa_node = NUMA_NO_NODE;
1758 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1759 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1760 lo->tag_set.driver_data = lo;
1762 err = blk_mq_alloc_tag_set(&lo->tag_set);
1766 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1767 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1768 err = PTR_ERR(lo->lo_queue);
1769 goto out_cleanup_tags;
1771 lo->lo_queue->queuedata = lo;
1774 * It doesn't make sense to enable merge because the I/O
1775 * submitted to backing file is handled page by page.
1777 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1780 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1782 goto out_free_queue;
1785 * Disable partition scanning by default. The in-kernel partition
1786 * scanning can be requested individually per-device during its
1787 * setup. Userspace can always add and remove partitions from all
1788 * devices. The needed partition minors are allocated from the
1789 * extended minor space, the main loop device numbers will continue
1790 * to match the loop minors, regardless of the number of partitions
1793 * If max_part is given, partition scanning is globally enabled for
1794 * all loop devices. The minors for the main loop devices will be
1795 * multiples of max_part.
1797 * Note: Global-for-all-devices, set-only-at-init, read-only module
1798 * parameteters like 'max_loop' and 'max_part' make things needlessly
1799 * complicated, are too static, inflexible and may surprise
1800 * userspace tools. Parameters like this in general should be avoided.
1803 disk->flags |= GENHD_FL_NO_PART_SCAN;
1804 disk->flags |= GENHD_FL_EXT_DEVT;
1805 mutex_init(&lo->lo_ctl_mutex);
1806 atomic_set(&lo->lo_refcnt, 0);
1808 spin_lock_init(&lo->lo_lock);
1809 disk->major = LOOP_MAJOR;
1810 disk->first_minor = i << part_shift;
1811 disk->fops = &lo_fops;
1812 disk->private_data = lo;
1813 disk->queue = lo->lo_queue;
1814 sprintf(disk->disk_name, "loop%d", i);
1817 return lo->lo_number;
1820 blk_cleanup_queue(lo->lo_queue);
1822 blk_mq_free_tag_set(&lo->tag_set);
1824 idr_remove(&loop_index_idr, i);
1831 static void loop_remove(struct loop_device *lo)
1833 blk_cleanup_queue(lo->lo_queue);
1834 del_gendisk(lo->lo_disk);
1835 blk_mq_free_tag_set(&lo->tag_set);
1836 put_disk(lo->lo_disk);
1840 static int find_free_cb(int id, void *ptr, void *data)
1842 struct loop_device *lo = ptr;
1843 struct loop_device **l = data;
1845 if (lo->lo_state == Lo_unbound) {
1852 static int loop_lookup(struct loop_device **l, int i)
1854 struct loop_device *lo;
1860 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1863 ret = lo->lo_number;
1868 /* lookup and return a specific i */
1869 lo = idr_find(&loop_index_idr, i);
1872 ret = lo->lo_number;
1878 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1880 struct loop_device *lo;
1881 struct kobject *kobj;
1884 mutex_lock(&loop_index_mutex);
1885 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1887 err = loop_add(&lo, MINOR(dev) >> part_shift);
1891 kobj = get_disk(lo->lo_disk);
1892 mutex_unlock(&loop_index_mutex);
1898 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1901 struct loop_device *lo;
1904 mutex_lock(&loop_index_mutex);
1907 ret = loop_lookup(&lo, parm);
1912 ret = loop_add(&lo, parm);
1914 case LOOP_CTL_REMOVE:
1915 ret = loop_lookup(&lo, parm);
1918 mutex_lock(&lo->lo_ctl_mutex);
1919 if (lo->lo_state != Lo_unbound) {
1921 mutex_unlock(&lo->lo_ctl_mutex);
1924 if (atomic_read(&lo->lo_refcnt) > 0) {
1926 mutex_unlock(&lo->lo_ctl_mutex);
1929 lo->lo_disk->private_data = NULL;
1930 mutex_unlock(&lo->lo_ctl_mutex);
1931 idr_remove(&loop_index_idr, lo->lo_number);
1934 case LOOP_CTL_GET_FREE:
1935 ret = loop_lookup(&lo, -1);
1938 ret = loop_add(&lo, -1);
1940 mutex_unlock(&loop_index_mutex);
1945 static const struct file_operations loop_ctl_fops = {
1946 .open = nonseekable_open,
1947 .unlocked_ioctl = loop_control_ioctl,
1948 .compat_ioctl = loop_control_ioctl,
1949 .owner = THIS_MODULE,
1950 .llseek = noop_llseek,
1953 static struct miscdevice loop_misc = {
1954 .minor = LOOP_CTRL_MINOR,
1955 .name = "loop-control",
1956 .fops = &loop_ctl_fops,
1959 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1960 MODULE_ALIAS("devname:loop-control");
1962 static int __init loop_init(void)
1965 unsigned long range;
1966 struct loop_device *lo;
1969 err = misc_register(&loop_misc);
1975 part_shift = fls(max_part);
1978 * Adjust max_part according to part_shift as it is exported
1979 * to user space so that user can decide correct minor number
1980 * if [s]he want to create more devices.
1982 * Note that -1 is required because partition 0 is reserved
1983 * for the whole disk.
1985 max_part = (1UL << part_shift) - 1;
1988 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1993 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1999 * If max_loop is specified, create that many devices upfront.
2000 * This also becomes a hard limit. If max_loop is not specified,
2001 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2002 * init time. Loop devices can be requested on-demand with the
2003 * /dev/loop-control interface, or be instantiated by accessing
2004 * a 'dead' device node.
2008 range = max_loop << part_shift;
2010 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2011 range = 1UL << MINORBITS;
2014 if (register_blkdev(LOOP_MAJOR, "loop")) {
2019 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2020 THIS_MODULE, loop_probe, NULL, NULL);
2022 /* pre-create number of devices given by config or max_loop */
2023 mutex_lock(&loop_index_mutex);
2024 for (i = 0; i < nr; i++)
2026 mutex_unlock(&loop_index_mutex);
2028 printk(KERN_INFO "loop: module loaded\n");
2032 misc_deregister(&loop_misc);
2036 static int loop_exit_cb(int id, void *ptr, void *data)
2038 struct loop_device *lo = ptr;
2044 static void __exit loop_exit(void)
2046 unsigned long range;
2048 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2050 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2051 idr_destroy(&loop_index_idr);
2053 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2054 unregister_blkdev(LOOP_MAJOR, "loop");
2056 misc_deregister(&loop_misc);
2059 module_init(loop_init);
2060 module_exit(loop_exit);
2063 static int __init max_loop_setup(char *str)
2065 max_loop = simple_strtol(str, NULL, 0);
2069 __setup("max_loop=", max_loop_setup);