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/blk-mq.h>
81 #include <asm/uaccess.h>
83 static DEFINE_IDR(loop_index_idr);
84 static DEFINE_MUTEX(loop_index_mutex);
87 static int part_shift;
89 static struct workqueue_struct *loop_wq;
94 static int transfer_none(struct loop_device *lo, int cmd,
95 struct page *raw_page, unsigned raw_off,
96 struct page *loop_page, unsigned loop_off,
97 int size, sector_t real_block)
99 char *raw_buf = kmap_atomic(raw_page) + raw_off;
100 char *loop_buf = kmap_atomic(loop_page) + loop_off;
103 memcpy(loop_buf, raw_buf, size);
105 memcpy(raw_buf, loop_buf, size);
107 kunmap_atomic(loop_buf);
108 kunmap_atomic(raw_buf);
113 static int transfer_xor(struct loop_device *lo, int cmd,
114 struct page *raw_page, unsigned raw_off,
115 struct page *loop_page, unsigned loop_off,
116 int size, sector_t real_block)
118 char *raw_buf = kmap_atomic(raw_page) + raw_off;
119 char *loop_buf = kmap_atomic(loop_page) + loop_off;
120 char *in, *out, *key;
131 key = lo->lo_encrypt_key;
132 keysize = lo->lo_encrypt_key_size;
133 for (i = 0; i < size; i++)
134 *out++ = *in++ ^ key[(i & 511) % keysize];
136 kunmap_atomic(loop_buf);
137 kunmap_atomic(raw_buf);
142 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
144 if (unlikely(info->lo_encrypt_key_size <= 0))
149 static struct loop_func_table none_funcs = {
150 .number = LO_CRYPT_NONE,
151 .transfer = transfer_none,
154 static struct loop_func_table xor_funcs = {
155 .number = LO_CRYPT_XOR,
156 .transfer = transfer_xor,
160 /* xfer_funcs[0] is special - its release function is never called */
161 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
166 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
170 /* Compute loopsize in bytes */
171 loopsize = i_size_read(file->f_mapping->host);
174 /* offset is beyond i_size, weird but possible */
178 if (sizelimit > 0 && sizelimit < loopsize)
179 loopsize = sizelimit;
181 * Unfortunately, if we want to do I/O on the device,
182 * the number of 512-byte sectors has to fit into a sector_t.
184 return loopsize >> 9;
187 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
189 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
193 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
195 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
196 sector_t x = (sector_t)size;
197 struct block_device *bdev = lo->lo_device;
199 if (unlikely((loff_t)x != size))
201 if (lo->lo_offset != offset)
202 lo->lo_offset = offset;
203 if (lo->lo_sizelimit != sizelimit)
204 lo->lo_sizelimit = sizelimit;
205 set_capacity(lo->lo_disk, x);
206 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
207 /* let user-space know about the new size */
208 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
213 lo_do_transfer(struct loop_device *lo, int cmd,
214 struct page *rpage, unsigned roffs,
215 struct page *lpage, unsigned loffs,
216 int size, sector_t rblock)
218 if (unlikely(!lo->transfer))
221 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
225 * __do_lo_send_write - helper for writing data to a loop device
227 * This helper just factors out common code between do_lo_send_direct_write()
228 * and do_lo_send_write().
230 static int __do_lo_send_write(struct file *file,
231 u8 *buf, const int len, loff_t pos)
234 mm_segment_t old_fs = get_fs();
236 file_start_write(file);
238 bw = file->f_op->write(file, buf, len, &pos);
240 file_end_write(file);
241 if (likely(bw == len))
243 printk_ratelimited(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
244 (unsigned long long)pos, len);
251 * do_lo_send_direct_write - helper for writing data to a loop device
253 * This is the fast, non-transforming version that does not need double
256 static int do_lo_send_direct_write(struct loop_device *lo,
257 struct bio_vec *bvec, loff_t pos, struct page *page)
259 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
260 kmap(bvec->bv_page) + bvec->bv_offset,
262 kunmap(bvec->bv_page);
268 * do_lo_send_write - helper for writing data to a loop device
270 * This is the slow, transforming version that needs to double buffer the
271 * data as it cannot do the transformations in place without having direct
272 * access to the destination pages of the backing file.
274 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
275 loff_t pos, struct page *page)
277 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
278 bvec->bv_offset, bvec->bv_len, pos >> 9);
280 return __do_lo_send_write(lo->lo_backing_file,
281 page_address(page), bvec->bv_len,
283 printk_ratelimited(KERN_ERR "loop: Transfer error at byte offset %llu, "
284 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
290 static int lo_send(struct loop_device *lo, struct request *rq, loff_t pos)
292 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
295 struct req_iterator iter;
296 struct page *page = NULL;
299 if (lo->transfer != transfer_none) {
300 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
304 do_lo_send = do_lo_send_write;
306 do_lo_send = do_lo_send_direct_write;
309 rq_for_each_segment(bvec, rq, iter) {
310 ret = do_lo_send(lo, &bvec, pos, page);
322 printk_ratelimited(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
327 struct lo_read_data {
328 struct loop_device *lo;
335 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
336 struct splice_desc *sd)
338 struct lo_read_data *p = sd->u.data;
339 struct loop_device *lo = p->lo;
340 struct page *page = buf->page;
344 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
350 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
351 printk_ratelimited(KERN_ERR "loop: transfer error block %ld\n",
356 flush_dcache_page(p->page);
365 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
367 return __splice_from_pipe(pipe, sd, lo_splice_actor);
371 do_lo_receive(struct loop_device *lo,
372 struct bio_vec *bvec, int bsize, loff_t pos)
374 struct lo_read_data cookie;
375 struct splice_desc sd;
380 cookie.page = bvec->bv_page;
381 cookie.offset = bvec->bv_offset;
382 cookie.bsize = bsize;
385 sd.total_len = bvec->bv_len;
390 file = lo->lo_backing_file;
391 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
397 lo_receive(struct loop_device *lo, struct request *rq, int bsize, loff_t pos)
400 struct req_iterator iter;
403 rq_for_each_segment(bvec, rq, iter) {
404 s = do_lo_receive(lo, &bvec, bsize, pos);
408 if (s != bvec.bv_len) {
411 __rq_for_each_bio(bio, rq)
420 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
423 * We use punch hole to reclaim the free space used by the
424 * image a.k.a. discard. However we do not support discard if
425 * encryption is enabled, because it may give an attacker
426 * useful information.
428 struct file *file = lo->lo_backing_file;
429 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
432 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
437 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
438 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
444 static int lo_req_flush(struct loop_device *lo, struct request *rq)
446 struct file *file = lo->lo_backing_file;
447 int ret = vfs_fsync(file, 0);
448 if (unlikely(ret && ret != -EINVAL))
454 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
459 pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
461 if (rq->cmd_flags & REQ_WRITE) {
463 if (rq->cmd_flags & REQ_FLUSH)
464 ret = lo_req_flush(lo, rq);
466 if (rq->cmd_flags & REQ_DISCARD) {
467 ret = lo_discard(lo, rq, pos);
471 ret = lo_send(lo, rq, pos);
473 if ((rq->cmd_flags & REQ_FUA) && !ret)
474 ret = lo_req_flush(lo, rq);
476 ret = lo_receive(lo, rq, lo->lo_blocksize, pos);
482 struct switch_request {
484 struct completion wait;
488 * Do the actual switch; called from the BIO completion routine
490 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
492 struct file *file = p->file;
493 struct file *old_file = lo->lo_backing_file;
494 struct address_space *mapping;
496 /* if no new file, only flush of queued bios requested */
500 mapping = file->f_mapping;
501 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
502 lo->lo_backing_file = file;
503 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
504 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
505 lo->old_gfp_mask = mapping_gfp_mask(mapping);
506 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
510 * loop_switch performs the hard work of switching a backing store.
511 * First it needs to flush existing IO, it does this by sending a magic
512 * BIO down the pipe. The completion of this BIO does the actual switch.
514 static int loop_switch(struct loop_device *lo, struct file *file)
516 struct switch_request w;
520 /* freeze queue and wait for completion of scheduled requests */
521 blk_mq_freeze_queue(lo->lo_queue);
523 /* do the switch action */
524 do_loop_switch(lo, &w);
527 blk_mq_unfreeze_queue(lo->lo_queue);
533 * Helper to flush the IOs in loop, but keeping loop thread running
535 static int loop_flush(struct loop_device *lo)
537 return loop_switch(lo, NULL);
541 * loop_change_fd switched the backing store of a loopback device to
542 * a new file. This is useful for operating system installers to free up
543 * the original file and in High Availability environments to switch to
544 * an alternative location for the content in case of server meltdown.
545 * This can only work if the loop device is used read-only, and if the
546 * new backing store is the same size and type as the old backing store.
548 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
551 struct file *file, *old_file;
556 if (lo->lo_state != Lo_bound)
559 /* the loop device has to be read-only */
561 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
569 inode = file->f_mapping->host;
570 old_file = lo->lo_backing_file;
574 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
577 /* size of the new backing store needs to be the same */
578 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
582 error = loop_switch(lo, file);
587 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
588 ioctl_by_bdev(bdev, BLKRRPART, 0);
597 static inline int is_loop_device(struct file *file)
599 struct inode *i = file->f_mapping->host;
601 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
604 /* loop sysfs attributes */
606 static ssize_t loop_attr_show(struct device *dev, char *page,
607 ssize_t (*callback)(struct loop_device *, char *))
609 struct gendisk *disk = dev_to_disk(dev);
610 struct loop_device *lo = disk->private_data;
612 return callback(lo, page);
615 #define LOOP_ATTR_RO(_name) \
616 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
617 static ssize_t loop_attr_do_show_##_name(struct device *d, \
618 struct device_attribute *attr, char *b) \
620 return loop_attr_show(d, b, loop_attr_##_name##_show); \
622 static struct device_attribute loop_attr_##_name = \
623 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
625 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
630 spin_lock_irq(&lo->lo_lock);
631 if (lo->lo_backing_file)
632 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
633 spin_unlock_irq(&lo->lo_lock);
635 if (IS_ERR_OR_NULL(p))
639 memmove(buf, p, ret);
647 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
649 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
652 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
654 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
657 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
659 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
661 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
664 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
666 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
668 return sprintf(buf, "%s\n", partscan ? "1" : "0");
671 LOOP_ATTR_RO(backing_file);
672 LOOP_ATTR_RO(offset);
673 LOOP_ATTR_RO(sizelimit);
674 LOOP_ATTR_RO(autoclear);
675 LOOP_ATTR_RO(partscan);
677 static struct attribute *loop_attrs[] = {
678 &loop_attr_backing_file.attr,
679 &loop_attr_offset.attr,
680 &loop_attr_sizelimit.attr,
681 &loop_attr_autoclear.attr,
682 &loop_attr_partscan.attr,
686 static struct attribute_group loop_attribute_group = {
691 static int loop_sysfs_init(struct loop_device *lo)
693 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
694 &loop_attribute_group);
697 static void loop_sysfs_exit(struct loop_device *lo)
699 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
700 &loop_attribute_group);
703 static void loop_config_discard(struct loop_device *lo)
705 struct file *file = lo->lo_backing_file;
706 struct inode *inode = file->f_mapping->host;
707 struct request_queue *q = lo->lo_queue;
710 * We use punch hole to reclaim the free space used by the
711 * image a.k.a. discard. However we do not support discard if
712 * encryption is enabled, because it may give an attacker
713 * useful information.
715 if ((!file->f_op->fallocate) ||
716 lo->lo_encrypt_key_size) {
717 q->limits.discard_granularity = 0;
718 q->limits.discard_alignment = 0;
719 q->limits.max_discard_sectors = 0;
720 q->limits.discard_zeroes_data = 0;
721 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
725 q->limits.discard_granularity = inode->i_sb->s_blocksize;
726 q->limits.discard_alignment = 0;
727 q->limits.max_discard_sectors = UINT_MAX >> 9;
728 q->limits.discard_zeroes_data = 1;
729 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
732 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
733 struct block_device *bdev, unsigned int arg)
735 struct file *file, *f;
737 struct address_space *mapping;
738 unsigned lo_blocksize;
743 /* This is safe, since we have a reference from open(). */
744 __module_get(THIS_MODULE);
752 if (lo->lo_state != Lo_unbound)
755 /* Avoid recursion */
757 while (is_loop_device(f)) {
758 struct loop_device *l;
760 if (f->f_mapping->host->i_bdev == bdev)
763 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
764 if (l->lo_state == Lo_unbound) {
768 f = l->lo_backing_file;
771 mapping = file->f_mapping;
772 inode = mapping->host;
775 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
778 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
780 lo_flags |= LO_FLAGS_READ_ONLY;
782 lo_blocksize = S_ISBLK(inode->i_mode) ?
783 inode->i_bdev->bd_block_size : PAGE_SIZE;
786 size = get_loop_size(lo, file);
787 if ((loff_t)(sector_t)size != size)
792 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
794 lo->lo_blocksize = lo_blocksize;
795 lo->lo_device = bdev;
796 lo->lo_flags = lo_flags;
797 lo->lo_backing_file = file;
798 lo->transfer = transfer_none;
800 lo->lo_sizelimit = 0;
801 lo->old_gfp_mask = mapping_gfp_mask(mapping);
802 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
804 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
805 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
807 set_capacity(lo->lo_disk, size);
808 bd_set_size(bdev, size << 9);
810 /* let user-space know about the new size */
811 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
813 set_blocksize(bdev, lo_blocksize);
815 lo->lo_state = Lo_bound;
817 lo->lo_flags |= LO_FLAGS_PARTSCAN;
818 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
819 ioctl_by_bdev(bdev, BLKRRPART, 0);
821 /* Grab the block_device to prevent its destruction after we
822 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
830 /* This is safe: open() is still holding a reference. */
831 module_put(THIS_MODULE);
836 loop_release_xfer(struct loop_device *lo)
839 struct loop_func_table *xfer = lo->lo_encryption;
843 err = xfer->release(lo);
845 lo->lo_encryption = NULL;
846 module_put(xfer->owner);
852 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
853 const struct loop_info64 *i)
858 struct module *owner = xfer->owner;
860 if (!try_module_get(owner))
863 err = xfer->init(lo, i);
867 lo->lo_encryption = xfer;
872 static int loop_clr_fd(struct loop_device *lo)
874 struct file *filp = lo->lo_backing_file;
875 gfp_t gfp = lo->old_gfp_mask;
876 struct block_device *bdev = lo->lo_device;
878 if (lo->lo_state != Lo_bound)
882 * If we've explicitly asked to tear down the loop device,
883 * and it has an elevated reference count, set it for auto-teardown when
884 * the last reference goes away. This stops $!~#$@ udev from
885 * preventing teardown because it decided that it needs to run blkid on
886 * the loopback device whenever they appear. xfstests is notorious for
887 * failing tests because blkid via udev races with a losetup
888 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
889 * command to fail with EBUSY.
891 if (lo->lo_refcnt > 1) {
892 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
893 mutex_unlock(&lo->lo_ctl_mutex);
900 spin_lock_irq(&lo->lo_lock);
901 lo->lo_state = Lo_rundown;
902 lo->lo_backing_file = NULL;
903 spin_unlock_irq(&lo->lo_lock);
905 loop_release_xfer(lo);
908 lo->lo_device = NULL;
909 lo->lo_encryption = NULL;
911 lo->lo_sizelimit = 0;
912 lo->lo_encrypt_key_size = 0;
913 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
914 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
915 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
918 invalidate_bdev(bdev);
920 set_capacity(lo->lo_disk, 0);
923 bd_set_size(bdev, 0);
924 /* let user-space know about this change */
925 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
927 mapping_set_gfp_mask(filp->f_mapping, gfp);
928 lo->lo_state = Lo_unbound;
929 /* This is safe: open() is still holding a reference. */
930 module_put(THIS_MODULE);
931 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
932 ioctl_by_bdev(bdev, BLKRRPART, 0);
935 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
936 mutex_unlock(&lo->lo_ctl_mutex);
938 * Need not hold lo_ctl_mutex to fput backing file.
939 * Calling fput holding lo_ctl_mutex triggers a circular
940 * lock dependency possibility warning as fput can take
941 * bd_mutex which is usually taken before lo_ctl_mutex.
948 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
951 struct loop_func_table *xfer;
952 kuid_t uid = current_uid();
954 if (lo->lo_encrypt_key_size &&
955 !uid_eq(lo->lo_key_owner, uid) &&
956 !capable(CAP_SYS_ADMIN))
958 if (lo->lo_state != Lo_bound)
960 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
963 err = loop_release_xfer(lo);
967 if (info->lo_encrypt_type) {
968 unsigned int type = info->lo_encrypt_type;
970 if (type >= MAX_LO_CRYPT)
972 xfer = xfer_funcs[type];
978 err = loop_init_xfer(lo, xfer, info);
982 if (lo->lo_offset != info->lo_offset ||
983 lo->lo_sizelimit != info->lo_sizelimit)
984 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
987 loop_config_discard(lo);
989 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
990 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
991 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
992 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
996 lo->transfer = xfer->transfer;
997 lo->ioctl = xfer->ioctl;
999 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1000 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1001 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1003 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1004 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1005 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1006 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1007 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1010 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1011 lo->lo_init[0] = info->lo_init[0];
1012 lo->lo_init[1] = info->lo_init[1];
1013 if (info->lo_encrypt_key_size) {
1014 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1015 info->lo_encrypt_key_size);
1016 lo->lo_key_owner = uid;
1023 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1025 struct file *file = lo->lo_backing_file;
1029 if (lo->lo_state != Lo_bound)
1031 error = vfs_getattr(&file->f_path, &stat);
1034 memset(info, 0, sizeof(*info));
1035 info->lo_number = lo->lo_number;
1036 info->lo_device = huge_encode_dev(stat.dev);
1037 info->lo_inode = stat.ino;
1038 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1039 info->lo_offset = lo->lo_offset;
1040 info->lo_sizelimit = lo->lo_sizelimit;
1041 info->lo_flags = lo->lo_flags;
1042 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1043 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1044 info->lo_encrypt_type =
1045 lo->lo_encryption ? lo->lo_encryption->number : 0;
1046 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1047 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1048 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1049 lo->lo_encrypt_key_size);
1055 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1057 memset(info64, 0, sizeof(*info64));
1058 info64->lo_number = info->lo_number;
1059 info64->lo_device = info->lo_device;
1060 info64->lo_inode = info->lo_inode;
1061 info64->lo_rdevice = info->lo_rdevice;
1062 info64->lo_offset = info->lo_offset;
1063 info64->lo_sizelimit = 0;
1064 info64->lo_encrypt_type = info->lo_encrypt_type;
1065 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1066 info64->lo_flags = info->lo_flags;
1067 info64->lo_init[0] = info->lo_init[0];
1068 info64->lo_init[1] = info->lo_init[1];
1069 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1070 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1072 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1073 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1077 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1079 memset(info, 0, sizeof(*info));
1080 info->lo_number = info64->lo_number;
1081 info->lo_device = info64->lo_device;
1082 info->lo_inode = info64->lo_inode;
1083 info->lo_rdevice = info64->lo_rdevice;
1084 info->lo_offset = info64->lo_offset;
1085 info->lo_encrypt_type = info64->lo_encrypt_type;
1086 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1087 info->lo_flags = info64->lo_flags;
1088 info->lo_init[0] = info64->lo_init[0];
1089 info->lo_init[1] = info64->lo_init[1];
1090 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1091 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1093 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1094 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1096 /* error in case values were truncated */
1097 if (info->lo_device != info64->lo_device ||
1098 info->lo_rdevice != info64->lo_rdevice ||
1099 info->lo_inode != info64->lo_inode ||
1100 info->lo_offset != info64->lo_offset)
1107 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1109 struct loop_info info;
1110 struct loop_info64 info64;
1112 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1114 loop_info64_from_old(&info, &info64);
1115 return loop_set_status(lo, &info64);
1119 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1121 struct loop_info64 info64;
1123 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1125 return loop_set_status(lo, &info64);
1129 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1130 struct loop_info info;
1131 struct loop_info64 info64;
1137 err = loop_get_status(lo, &info64);
1139 err = loop_info64_to_old(&info64, &info);
1140 if (!err && copy_to_user(arg, &info, sizeof(info)))
1147 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1148 struct loop_info64 info64;
1154 err = loop_get_status(lo, &info64);
1155 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1161 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1163 if (unlikely(lo->lo_state != Lo_bound))
1166 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1169 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1170 unsigned int cmd, unsigned long arg)
1172 struct loop_device *lo = bdev->bd_disk->private_data;
1175 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1178 err = loop_set_fd(lo, mode, bdev, arg);
1180 case LOOP_CHANGE_FD:
1181 err = loop_change_fd(lo, bdev, arg);
1184 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1185 err = loop_clr_fd(lo);
1189 case LOOP_SET_STATUS:
1191 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1192 err = loop_set_status_old(lo,
1193 (struct loop_info __user *)arg);
1195 case LOOP_GET_STATUS:
1196 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1198 case LOOP_SET_STATUS64:
1200 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1201 err = loop_set_status64(lo,
1202 (struct loop_info64 __user *) arg);
1204 case LOOP_GET_STATUS64:
1205 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1207 case LOOP_SET_CAPACITY:
1209 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1210 err = loop_set_capacity(lo, bdev);
1213 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1215 mutex_unlock(&lo->lo_ctl_mutex);
1221 #ifdef CONFIG_COMPAT
1222 struct compat_loop_info {
1223 compat_int_t lo_number; /* ioctl r/o */
1224 compat_dev_t lo_device; /* ioctl r/o */
1225 compat_ulong_t lo_inode; /* ioctl r/o */
1226 compat_dev_t lo_rdevice; /* ioctl r/o */
1227 compat_int_t lo_offset;
1228 compat_int_t lo_encrypt_type;
1229 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1230 compat_int_t lo_flags; /* ioctl r/o */
1231 char lo_name[LO_NAME_SIZE];
1232 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1233 compat_ulong_t lo_init[2];
1238 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1239 * - noinlined to reduce stack space usage in main part of driver
1242 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1243 struct loop_info64 *info64)
1245 struct compat_loop_info info;
1247 if (copy_from_user(&info, arg, sizeof(info)))
1250 memset(info64, 0, sizeof(*info64));
1251 info64->lo_number = info.lo_number;
1252 info64->lo_device = info.lo_device;
1253 info64->lo_inode = info.lo_inode;
1254 info64->lo_rdevice = info.lo_rdevice;
1255 info64->lo_offset = info.lo_offset;
1256 info64->lo_sizelimit = 0;
1257 info64->lo_encrypt_type = info.lo_encrypt_type;
1258 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1259 info64->lo_flags = info.lo_flags;
1260 info64->lo_init[0] = info.lo_init[0];
1261 info64->lo_init[1] = info.lo_init[1];
1262 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1263 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1265 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1266 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1271 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1272 * - noinlined to reduce stack space usage in main part of driver
1275 loop_info64_to_compat(const struct loop_info64 *info64,
1276 struct compat_loop_info __user *arg)
1278 struct compat_loop_info info;
1280 memset(&info, 0, sizeof(info));
1281 info.lo_number = info64->lo_number;
1282 info.lo_device = info64->lo_device;
1283 info.lo_inode = info64->lo_inode;
1284 info.lo_rdevice = info64->lo_rdevice;
1285 info.lo_offset = info64->lo_offset;
1286 info.lo_encrypt_type = info64->lo_encrypt_type;
1287 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1288 info.lo_flags = info64->lo_flags;
1289 info.lo_init[0] = info64->lo_init[0];
1290 info.lo_init[1] = info64->lo_init[1];
1291 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1292 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1294 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1295 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1297 /* error in case values were truncated */
1298 if (info.lo_device != info64->lo_device ||
1299 info.lo_rdevice != info64->lo_rdevice ||
1300 info.lo_inode != info64->lo_inode ||
1301 info.lo_offset != info64->lo_offset ||
1302 info.lo_init[0] != info64->lo_init[0] ||
1303 info.lo_init[1] != info64->lo_init[1])
1306 if (copy_to_user(arg, &info, sizeof(info)))
1312 loop_set_status_compat(struct loop_device *lo,
1313 const struct compat_loop_info __user *arg)
1315 struct loop_info64 info64;
1318 ret = loop_info64_from_compat(arg, &info64);
1321 return loop_set_status(lo, &info64);
1325 loop_get_status_compat(struct loop_device *lo,
1326 struct compat_loop_info __user *arg)
1328 struct loop_info64 info64;
1334 err = loop_get_status(lo, &info64);
1336 err = loop_info64_to_compat(&info64, arg);
1340 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1341 unsigned int cmd, unsigned long arg)
1343 struct loop_device *lo = bdev->bd_disk->private_data;
1347 case LOOP_SET_STATUS:
1348 mutex_lock(&lo->lo_ctl_mutex);
1349 err = loop_set_status_compat(
1350 lo, (const struct compat_loop_info __user *) arg);
1351 mutex_unlock(&lo->lo_ctl_mutex);
1353 case LOOP_GET_STATUS:
1354 mutex_lock(&lo->lo_ctl_mutex);
1355 err = loop_get_status_compat(
1356 lo, (struct compat_loop_info __user *) arg);
1357 mutex_unlock(&lo->lo_ctl_mutex);
1359 case LOOP_SET_CAPACITY:
1361 case LOOP_GET_STATUS64:
1362 case LOOP_SET_STATUS64:
1363 arg = (unsigned long) compat_ptr(arg);
1365 case LOOP_CHANGE_FD:
1366 err = lo_ioctl(bdev, mode, cmd, arg);
1376 static int lo_open(struct block_device *bdev, fmode_t mode)
1378 struct loop_device *lo;
1381 mutex_lock(&loop_index_mutex);
1382 lo = bdev->bd_disk->private_data;
1388 mutex_lock(&lo->lo_ctl_mutex);
1390 mutex_unlock(&lo->lo_ctl_mutex);
1392 mutex_unlock(&loop_index_mutex);
1396 static void lo_release(struct gendisk *disk, fmode_t mode)
1398 struct loop_device *lo = disk->private_data;
1401 mutex_lock(&lo->lo_ctl_mutex);
1403 if (--lo->lo_refcnt)
1406 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1408 * In autoclear mode, stop the loop thread
1409 * and remove configuration after last close.
1411 err = loop_clr_fd(lo);
1416 * Otherwise keep thread (if running) and config,
1417 * but flush possible ongoing bios in thread.
1423 mutex_unlock(&lo->lo_ctl_mutex);
1426 static const struct block_device_operations lo_fops = {
1427 .owner = THIS_MODULE,
1429 .release = lo_release,
1431 #ifdef CONFIG_COMPAT
1432 .compat_ioctl = lo_compat_ioctl,
1437 * And now the modules code and kernel interface.
1439 static int max_loop;
1440 module_param(max_loop, int, S_IRUGO);
1441 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1442 module_param(max_part, int, S_IRUGO);
1443 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1444 MODULE_LICENSE("GPL");
1445 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1447 int loop_register_transfer(struct loop_func_table *funcs)
1449 unsigned int n = funcs->number;
1451 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1453 xfer_funcs[n] = funcs;
1457 static int unregister_transfer_cb(int id, void *ptr, void *data)
1459 struct loop_device *lo = ptr;
1460 struct loop_func_table *xfer = data;
1462 mutex_lock(&lo->lo_ctl_mutex);
1463 if (lo->lo_encryption == xfer)
1464 loop_release_xfer(lo);
1465 mutex_unlock(&lo->lo_ctl_mutex);
1469 int loop_unregister_transfer(int number)
1471 unsigned int n = number;
1472 struct loop_func_table *xfer;
1474 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1477 xfer_funcs[n] = NULL;
1478 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1482 EXPORT_SYMBOL(loop_register_transfer);
1483 EXPORT_SYMBOL(loop_unregister_transfer);
1485 static int loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1486 const struct blk_mq_queue_data *bd)
1488 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1490 blk_mq_start_request(bd->rq);
1492 if (cmd->rq->cmd_flags & REQ_WRITE) {
1493 struct loop_device *lo = cmd->rq->q->queuedata;
1494 bool need_sched = true;
1496 spin_lock_irq(&lo->lo_lock);
1497 if (lo->write_started)
1500 lo->write_started = true;
1501 list_add_tail(&cmd->list, &lo->write_cmd_head);
1502 spin_unlock_irq(&lo->lo_lock);
1505 queue_work(loop_wq, &lo->write_work);
1507 queue_work(loop_wq, &cmd->read_work);
1510 return BLK_MQ_RQ_QUEUE_OK;
1513 static void loop_handle_cmd(struct loop_cmd *cmd)
1515 const bool write = cmd->rq->cmd_flags & REQ_WRITE;
1516 struct loop_device *lo = cmd->rq->q->queuedata;
1519 if (lo->lo_state != Lo_bound)
1522 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY))
1525 ret = do_req_filebacked(lo, cmd->rq);
1529 cmd->rq->errors = -EIO;
1530 blk_mq_complete_request(cmd->rq);
1533 static void loop_queue_write_work(struct work_struct *work)
1535 struct loop_device *lo =
1536 container_of(work, struct loop_device, write_work);
1537 LIST_HEAD(cmd_list);
1539 spin_lock_irq(&lo->lo_lock);
1541 list_splice_init(&lo->write_cmd_head, &cmd_list);
1542 spin_unlock_irq(&lo->lo_lock);
1544 while (!list_empty(&cmd_list)) {
1545 struct loop_cmd *cmd = list_first_entry(&cmd_list,
1546 struct loop_cmd, list);
1547 list_del_init(&cmd->list);
1548 loop_handle_cmd(cmd);
1551 spin_lock_irq(&lo->lo_lock);
1552 if (!list_empty(&lo->write_cmd_head))
1554 lo->write_started = false;
1555 spin_unlock_irq(&lo->lo_lock);
1558 static void loop_queue_read_work(struct work_struct *work)
1560 struct loop_cmd *cmd =
1561 container_of(work, struct loop_cmd, read_work);
1563 loop_handle_cmd(cmd);
1566 static int loop_init_request(void *data, struct request *rq,
1567 unsigned int hctx_idx, unsigned int request_idx,
1568 unsigned int numa_node)
1570 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1573 INIT_WORK(&cmd->read_work, loop_queue_read_work);
1578 static struct blk_mq_ops loop_mq_ops = {
1579 .queue_rq = loop_queue_rq,
1580 .map_queue = blk_mq_map_queue,
1581 .init_request = loop_init_request,
1584 static int loop_add(struct loop_device **l, int i)
1586 struct loop_device *lo;
1587 struct gendisk *disk;
1591 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1595 lo->lo_state = Lo_unbound;
1597 /* allocate id, if @id >= 0, we're requesting that specific id */
1599 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1603 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1610 lo->tag_set.ops = &loop_mq_ops;
1611 lo->tag_set.nr_hw_queues = 1;
1612 lo->tag_set.queue_depth = 128;
1613 lo->tag_set.numa_node = NUMA_NO_NODE;
1614 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1615 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1616 lo->tag_set.driver_data = lo;
1618 err = blk_mq_alloc_tag_set(&lo->tag_set);
1622 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1623 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1624 err = PTR_ERR(lo->lo_queue);
1625 goto out_cleanup_tags;
1627 lo->lo_queue->queuedata = lo;
1629 INIT_LIST_HEAD(&lo->write_cmd_head);
1630 INIT_WORK(&lo->write_work, loop_queue_write_work);
1632 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1634 goto out_free_queue;
1637 * Disable partition scanning by default. The in-kernel partition
1638 * scanning can be requested individually per-device during its
1639 * setup. Userspace can always add and remove partitions from all
1640 * devices. The needed partition minors are allocated from the
1641 * extended minor space, the main loop device numbers will continue
1642 * to match the loop minors, regardless of the number of partitions
1645 * If max_part is given, partition scanning is globally enabled for
1646 * all loop devices. The minors for the main loop devices will be
1647 * multiples of max_part.
1649 * Note: Global-for-all-devices, set-only-at-init, read-only module
1650 * parameteters like 'max_loop' and 'max_part' make things needlessly
1651 * complicated, are too static, inflexible and may surprise
1652 * userspace tools. Parameters like this in general should be avoided.
1655 disk->flags |= GENHD_FL_NO_PART_SCAN;
1656 disk->flags |= GENHD_FL_EXT_DEVT;
1657 mutex_init(&lo->lo_ctl_mutex);
1659 spin_lock_init(&lo->lo_lock);
1660 disk->major = LOOP_MAJOR;
1661 disk->first_minor = i << part_shift;
1662 disk->fops = &lo_fops;
1663 disk->private_data = lo;
1664 disk->queue = lo->lo_queue;
1665 sprintf(disk->disk_name, "loop%d", i);
1668 return lo->lo_number;
1671 blk_cleanup_queue(lo->lo_queue);
1673 blk_mq_free_tag_set(&lo->tag_set);
1675 idr_remove(&loop_index_idr, i);
1682 static void loop_remove(struct loop_device *lo)
1684 del_gendisk(lo->lo_disk);
1685 blk_cleanup_queue(lo->lo_queue);
1686 blk_mq_free_tag_set(&lo->tag_set);
1687 put_disk(lo->lo_disk);
1691 static int find_free_cb(int id, void *ptr, void *data)
1693 struct loop_device *lo = ptr;
1694 struct loop_device **l = data;
1696 if (lo->lo_state == Lo_unbound) {
1703 static int loop_lookup(struct loop_device **l, int i)
1705 struct loop_device *lo;
1711 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1714 ret = lo->lo_number;
1719 /* lookup and return a specific i */
1720 lo = idr_find(&loop_index_idr, i);
1723 ret = lo->lo_number;
1729 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1731 struct loop_device *lo;
1732 struct kobject *kobj;
1735 mutex_lock(&loop_index_mutex);
1736 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1738 err = loop_add(&lo, MINOR(dev) >> part_shift);
1742 kobj = get_disk(lo->lo_disk);
1743 mutex_unlock(&loop_index_mutex);
1749 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1752 struct loop_device *lo;
1755 mutex_lock(&loop_index_mutex);
1758 ret = loop_lookup(&lo, parm);
1763 ret = loop_add(&lo, parm);
1765 case LOOP_CTL_REMOVE:
1766 ret = loop_lookup(&lo, parm);
1769 mutex_lock(&lo->lo_ctl_mutex);
1770 if (lo->lo_state != Lo_unbound) {
1772 mutex_unlock(&lo->lo_ctl_mutex);
1775 if (lo->lo_refcnt > 0) {
1777 mutex_unlock(&lo->lo_ctl_mutex);
1780 lo->lo_disk->private_data = NULL;
1781 mutex_unlock(&lo->lo_ctl_mutex);
1782 idr_remove(&loop_index_idr, lo->lo_number);
1785 case LOOP_CTL_GET_FREE:
1786 ret = loop_lookup(&lo, -1);
1789 ret = loop_add(&lo, -1);
1791 mutex_unlock(&loop_index_mutex);
1796 static const struct file_operations loop_ctl_fops = {
1797 .open = nonseekable_open,
1798 .unlocked_ioctl = loop_control_ioctl,
1799 .compat_ioctl = loop_control_ioctl,
1800 .owner = THIS_MODULE,
1801 .llseek = noop_llseek,
1804 static struct miscdevice loop_misc = {
1805 .minor = LOOP_CTRL_MINOR,
1806 .name = "loop-control",
1807 .fops = &loop_ctl_fops,
1810 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1811 MODULE_ALIAS("devname:loop-control");
1813 static int __init loop_init(void)
1816 unsigned long range;
1817 struct loop_device *lo;
1820 err = misc_register(&loop_misc);
1826 part_shift = fls(max_part);
1829 * Adjust max_part according to part_shift as it is exported
1830 * to user space so that user can decide correct minor number
1831 * if [s]he want to create more devices.
1833 * Note that -1 is required because partition 0 is reserved
1834 * for the whole disk.
1836 max_part = (1UL << part_shift) - 1;
1839 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1844 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1850 * If max_loop is specified, create that many devices upfront.
1851 * This also becomes a hard limit. If max_loop is not specified,
1852 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1853 * init time. Loop devices can be requested on-demand with the
1854 * /dev/loop-control interface, or be instantiated by accessing
1855 * a 'dead' device node.
1859 range = max_loop << part_shift;
1861 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1862 range = 1UL << MINORBITS;
1865 if (register_blkdev(LOOP_MAJOR, "loop")) {
1870 loop_wq = alloc_workqueue("kloopd",
1871 WQ_MEM_RECLAIM | WQ_HIGHPRI | WQ_UNBOUND, 0);
1877 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1878 THIS_MODULE, loop_probe, NULL, NULL);
1880 /* pre-create number of devices given by config or max_loop */
1881 mutex_lock(&loop_index_mutex);
1882 for (i = 0; i < nr; i++)
1884 mutex_unlock(&loop_index_mutex);
1886 printk(KERN_INFO "loop: module loaded\n");
1890 misc_deregister(&loop_misc);
1894 static int loop_exit_cb(int id, void *ptr, void *data)
1896 struct loop_device *lo = ptr;
1902 static void __exit loop_exit(void)
1904 unsigned long range;
1906 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1908 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1909 idr_destroy(&loop_index_idr);
1911 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1912 unregister_blkdev(LOOP_MAJOR, "loop");
1914 destroy_workqueue(loop_wq);
1916 misc_deregister(&loop_misc);
1919 module_init(loop_init);
1920 module_exit(loop_exit);
1923 static int __init max_loop_setup(char *str)
1925 max_loop = simple_strtol(str, NULL, 0);
1929 __setup("max_loop=", max_loop_setup);