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/loop.h>
67 #include <linux/compat.h>
68 #include <linux/suspend.h>
69 #include <linux/freezer.h>
70 #include <linux/mutex.h>
71 #include <linux/writeback.h>
72 #include <linux/buffer_head.h> /* for invalidate_bdev() */
73 #include <linux/completion.h>
74 #include <linux/highmem.h>
75 #include <linux/kthread.h>
76 #include <linux/splice.h>
77 #include <linux/sysfs.h>
78 #include <linux/miscdevice.h>
79 #include <linux/falloc.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;
92 static int transfer_none(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
97 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
101 memcpy(loop_buf, raw_buf, size);
103 memcpy(raw_buf, loop_buf, size);
105 kunmap_atomic(loop_buf, KM_USER1);
106 kunmap_atomic(raw_buf, KM_USER0);
111 static int transfer_xor(struct loop_device *lo, int cmd,
112 struct page *raw_page, unsigned raw_off,
113 struct page *loop_page, unsigned loop_off,
114 int size, sector_t real_block)
116 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
117 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
118 char *in, *out, *key;
129 key = lo->lo_encrypt_key;
130 keysize = lo->lo_encrypt_key_size;
131 for (i = 0; i < size; i++)
132 *out++ = *in++ ^ key[(i & 511) % keysize];
134 kunmap_atomic(loop_buf, KM_USER1);
135 kunmap_atomic(raw_buf, KM_USER0);
140 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
142 if (unlikely(info->lo_encrypt_key_size <= 0))
147 static struct loop_func_table none_funcs = {
148 .number = LO_CRYPT_NONE,
149 .transfer = transfer_none,
152 static struct loop_func_table xor_funcs = {
153 .number = LO_CRYPT_XOR,
154 .transfer = transfer_xor,
158 /* xfer_funcs[0] is special - its release function is never called */
159 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
164 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
166 loff_t size, loopsize;
168 /* Compute loopsize in bytes */
169 size = i_size_read(file->f_mapping->host);
170 loopsize = size - offset;
171 /* offset is beyond i_size, wierd but possible */
175 if (sizelimit > 0 && sizelimit < loopsize)
176 loopsize = sizelimit;
178 * Unfortunately, if we want to do I/O on the device,
179 * the number of 512-byte sectors has to fit into a sector_t.
181 return loopsize >> 9;
184 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
186 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
190 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
192 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
193 sector_t x = (sector_t)size;
195 if (unlikely((loff_t)x != size))
197 if (lo->lo_offset != offset)
198 lo->lo_offset = offset;
199 if (lo->lo_sizelimit != sizelimit)
200 lo->lo_sizelimit = sizelimit;
201 set_capacity(lo->lo_disk, x);
206 lo_do_transfer(struct loop_device *lo, int cmd,
207 struct page *rpage, unsigned roffs,
208 struct page *lpage, unsigned loffs,
209 int size, sector_t rblock)
211 if (unlikely(!lo->transfer))
214 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
218 * __do_lo_send_write - helper for writing data to a loop device
220 * This helper just factors out common code between do_lo_send_direct_write()
221 * and do_lo_send_write().
223 static int __do_lo_send_write(struct file *file,
224 u8 *buf, const int len, loff_t pos)
227 mm_segment_t old_fs = get_fs();
230 bw = file->f_op->write(file, buf, len, &pos);
232 if (likely(bw == len))
234 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
235 (unsigned long long)pos, len);
242 * do_lo_send_direct_write - helper for writing data to a loop device
244 * This is the fast, non-transforming version that does not need double
247 static int do_lo_send_direct_write(struct loop_device *lo,
248 struct bio_vec *bvec, loff_t pos, struct page *page)
250 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
251 kmap(bvec->bv_page) + bvec->bv_offset,
253 kunmap(bvec->bv_page);
259 * do_lo_send_write - helper for writing data to a loop device
261 * This is the slow, transforming version that needs to double buffer the
262 * data as it cannot do the transformations in place without having direct
263 * access to the destination pages of the backing file.
265 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
266 loff_t pos, struct page *page)
268 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
269 bvec->bv_offset, bvec->bv_len, pos >> 9);
271 return __do_lo_send_write(lo->lo_backing_file,
272 page_address(page), bvec->bv_len,
274 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
275 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
281 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
283 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
285 struct bio_vec *bvec;
286 struct page *page = NULL;
289 if (lo->transfer != transfer_none) {
290 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
294 do_lo_send = do_lo_send_write;
296 do_lo_send = do_lo_send_direct_write;
299 bio_for_each_segment(bvec, bio, i) {
300 ret = do_lo_send(lo, bvec, pos, page);
312 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
317 struct lo_read_data {
318 struct loop_device *lo;
325 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
326 struct splice_desc *sd)
328 struct lo_read_data *p = sd->u.data;
329 struct loop_device *lo = p->lo;
330 struct page *page = buf->page;
334 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
340 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
341 printk(KERN_ERR "loop: transfer error block %ld\n",
346 flush_dcache_page(p->page);
355 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
357 return __splice_from_pipe(pipe, sd, lo_splice_actor);
361 do_lo_receive(struct loop_device *lo,
362 struct bio_vec *bvec, int bsize, loff_t pos)
364 struct lo_read_data cookie;
365 struct splice_desc sd;
370 cookie.page = bvec->bv_page;
371 cookie.offset = bvec->bv_offset;
372 cookie.bsize = bsize;
375 sd.total_len = bvec->bv_len;
380 file = lo->lo_backing_file;
381 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
385 if (retval != bvec->bv_len)
391 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
393 struct bio_vec *bvec;
396 bio_for_each_segment(bvec, bio, i) {
397 ret = do_lo_receive(lo, bvec, bsize, pos);
405 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
410 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
412 if (bio_rw(bio) == WRITE) {
413 struct file *file = lo->lo_backing_file;
415 if (bio->bi_rw & REQ_FLUSH) {
416 ret = vfs_fsync(file, 0);
417 if (unlikely(ret && ret != -EINVAL)) {
424 * We use punch hole to reclaim the free space used by the
425 * image a.k.a. discard. However we do not support discard if
426 * encryption is enabled, because it may give an attacker
427 * useful information.
429 if (bio->bi_rw & REQ_DISCARD) {
430 struct file *file = lo->lo_backing_file;
431 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
433 if ((!file->f_op->fallocate) ||
434 lo->lo_encrypt_key_size) {
438 ret = file->f_op->fallocate(file, mode, pos,
440 if (unlikely(ret && ret != -EINVAL &&
446 ret = lo_send(lo, bio, pos);
448 if ((bio->bi_rw & REQ_FUA) && !ret) {
449 ret = vfs_fsync(file, 0);
450 if (unlikely(ret && ret != -EINVAL))
454 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
461 * Add bio to back of pending list
463 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
465 bio_list_add(&lo->lo_bio_list, bio);
469 * Grab first pending buffer
471 static struct bio *loop_get_bio(struct loop_device *lo)
473 return bio_list_pop(&lo->lo_bio_list);
476 static void loop_make_request(struct request_queue *q, struct bio *old_bio)
478 struct loop_device *lo = q->queuedata;
479 int rw = bio_rw(old_bio);
484 BUG_ON(!lo || (rw != READ && rw != WRITE));
486 spin_lock_irq(&lo->lo_lock);
487 if (lo->lo_state != Lo_bound)
489 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
491 loop_add_bio(lo, old_bio);
492 wake_up(&lo->lo_event);
493 spin_unlock_irq(&lo->lo_lock);
497 spin_unlock_irq(&lo->lo_lock);
498 bio_io_error(old_bio);
501 struct switch_request {
503 struct completion wait;
506 static void do_loop_switch(struct loop_device *, struct switch_request *);
508 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
510 if (unlikely(!bio->bi_bdev)) {
511 do_loop_switch(lo, bio->bi_private);
514 int ret = do_bio_filebacked(lo, bio);
520 * worker thread that handles reads/writes to file backed loop devices,
521 * to avoid blocking in our make_request_fn. it also does loop decrypting
522 * on reads for block backed loop, as that is too heavy to do from
523 * b_end_io context where irqs may be disabled.
525 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
526 * calling kthread_stop(). Therefore once kthread_should_stop() is
527 * true, make_request will not place any more requests. Therefore
528 * once kthread_should_stop() is true and lo_bio is NULL, we are
529 * done with the loop.
531 static int loop_thread(void *data)
533 struct loop_device *lo = data;
536 set_user_nice(current, -20);
538 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
540 wait_event_interruptible(lo->lo_event,
541 !bio_list_empty(&lo->lo_bio_list) ||
542 kthread_should_stop());
544 if (bio_list_empty(&lo->lo_bio_list))
546 spin_lock_irq(&lo->lo_lock);
547 bio = loop_get_bio(lo);
548 spin_unlock_irq(&lo->lo_lock);
551 loop_handle_bio(lo, bio);
558 * loop_switch performs the hard work of switching a backing store.
559 * First it needs to flush existing IO, it does this by sending a magic
560 * BIO down the pipe. The completion of this BIO does the actual switch.
562 static int loop_switch(struct loop_device *lo, struct file *file)
564 struct switch_request w;
565 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
568 init_completion(&w.wait);
570 bio->bi_private = &w;
572 loop_make_request(lo->lo_queue, bio);
573 wait_for_completion(&w.wait);
578 * Helper to flush the IOs in loop, but keeping loop thread running
580 static int loop_flush(struct loop_device *lo)
582 /* loop not yet configured, no running thread, nothing to flush */
586 return loop_switch(lo, NULL);
590 * Do the actual switch; called from the BIO completion routine
592 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
594 struct file *file = p->file;
595 struct file *old_file = lo->lo_backing_file;
596 struct address_space *mapping;
598 /* if no new file, only flush of queued bios requested */
602 mapping = file->f_mapping;
603 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
604 lo->lo_backing_file = file;
605 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
606 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
607 lo->old_gfp_mask = mapping_gfp_mask(mapping);
608 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
615 * loop_change_fd switched the backing store of a loopback device to
616 * a new file. This is useful for operating system installers to free up
617 * the original file and in High Availability environments to switch to
618 * an alternative location for the content in case of server meltdown.
619 * This can only work if the loop device is used read-only, and if the
620 * new backing store is the same size and type as the old backing store.
622 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
625 struct file *file, *old_file;
630 if (lo->lo_state != Lo_bound)
633 /* the loop device has to be read-only */
635 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
643 inode = file->f_mapping->host;
644 old_file = lo->lo_backing_file;
648 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
651 /* size of the new backing store needs to be the same */
652 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
656 error = loop_switch(lo, file);
661 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
662 ioctl_by_bdev(bdev, BLKRRPART, 0);
671 static inline int is_loop_device(struct file *file)
673 struct inode *i = file->f_mapping->host;
675 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
678 /* loop sysfs attributes */
680 static ssize_t loop_attr_show(struct device *dev, char *page,
681 ssize_t (*callback)(struct loop_device *, char *))
683 struct gendisk *disk = dev_to_disk(dev);
684 struct loop_device *lo = disk->private_data;
686 return callback(lo, page);
689 #define LOOP_ATTR_RO(_name) \
690 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
691 static ssize_t loop_attr_do_show_##_name(struct device *d, \
692 struct device_attribute *attr, char *b) \
694 return loop_attr_show(d, b, loop_attr_##_name##_show); \
696 static struct device_attribute loop_attr_##_name = \
697 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
699 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
704 spin_lock_irq(&lo->lo_lock);
705 if (lo->lo_backing_file)
706 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
707 spin_unlock_irq(&lo->lo_lock);
709 if (IS_ERR_OR_NULL(p))
713 memmove(buf, p, ret);
721 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
723 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
726 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
728 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
731 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
733 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
735 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
738 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
740 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
742 return sprintf(buf, "%s\n", partscan ? "1" : "0");
745 LOOP_ATTR_RO(backing_file);
746 LOOP_ATTR_RO(offset);
747 LOOP_ATTR_RO(sizelimit);
748 LOOP_ATTR_RO(autoclear);
749 LOOP_ATTR_RO(partscan);
751 static struct attribute *loop_attrs[] = {
752 &loop_attr_backing_file.attr,
753 &loop_attr_offset.attr,
754 &loop_attr_sizelimit.attr,
755 &loop_attr_autoclear.attr,
756 &loop_attr_partscan.attr,
760 static struct attribute_group loop_attribute_group = {
765 static int loop_sysfs_init(struct loop_device *lo)
767 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
768 &loop_attribute_group);
771 static void loop_sysfs_exit(struct loop_device *lo)
773 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
774 &loop_attribute_group);
777 static void loop_config_discard(struct loop_device *lo)
779 struct file *file = lo->lo_backing_file;
780 struct inode *inode = file->f_mapping->host;
781 struct request_queue *q = lo->lo_queue;
784 * We use punch hole to reclaim the free space used by the
785 * image a.k.a. discard. However we do support discard if
786 * encryption is enabled, because it may give an attacker
787 * useful information.
789 if ((!file->f_op->fallocate) ||
790 lo->lo_encrypt_key_size) {
791 q->limits.discard_granularity = 0;
792 q->limits.discard_alignment = 0;
793 q->limits.max_discard_sectors = 0;
794 q->limits.discard_zeroes_data = 0;
795 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
799 q->limits.discard_granularity = inode->i_sb->s_blocksize;
800 q->limits.discard_alignment = 0;
801 q->limits.max_discard_sectors = UINT_MAX >> 9;
802 q->limits.discard_zeroes_data = 1;
803 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
806 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
807 struct block_device *bdev, unsigned int arg)
809 struct file *file, *f;
811 struct address_space *mapping;
812 unsigned lo_blocksize;
817 /* This is safe, since we have a reference from open(). */
818 __module_get(THIS_MODULE);
826 if (lo->lo_state != Lo_unbound)
829 /* Avoid recursion */
831 while (is_loop_device(f)) {
832 struct loop_device *l;
834 if (f->f_mapping->host->i_bdev == bdev)
837 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
838 if (l->lo_state == Lo_unbound) {
842 f = l->lo_backing_file;
845 mapping = file->f_mapping;
846 inode = mapping->host;
849 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
852 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
854 lo_flags |= LO_FLAGS_READ_ONLY;
856 lo_blocksize = S_ISBLK(inode->i_mode) ?
857 inode->i_bdev->bd_block_size : PAGE_SIZE;
860 size = get_loop_size(lo, file);
861 if ((loff_t)(sector_t)size != size)
866 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
868 lo->lo_blocksize = lo_blocksize;
869 lo->lo_device = bdev;
870 lo->lo_flags = lo_flags;
871 lo->lo_backing_file = file;
872 lo->transfer = transfer_none;
874 lo->lo_sizelimit = 0;
875 lo->old_gfp_mask = mapping_gfp_mask(mapping);
876 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
878 bio_list_init(&lo->lo_bio_list);
881 * set queue make_request_fn, and add limits based on lower level
884 blk_queue_make_request(lo->lo_queue, loop_make_request);
885 lo->lo_queue->queuedata = lo;
887 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
888 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
890 set_capacity(lo->lo_disk, size);
891 bd_set_size(bdev, size << 9);
893 /* let user-space know about the new size */
894 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
896 set_blocksize(bdev, lo_blocksize);
898 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
900 if (IS_ERR(lo->lo_thread)) {
901 error = PTR_ERR(lo->lo_thread);
904 lo->lo_state = Lo_bound;
905 wake_up_process(lo->lo_thread);
907 lo->lo_flags |= LO_FLAGS_PARTSCAN;
908 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
909 ioctl_by_bdev(bdev, BLKRRPART, 0);
911 /* Grab the block_device to prevent its destruction after we
912 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
919 lo->lo_thread = NULL;
920 lo->lo_device = NULL;
921 lo->lo_backing_file = NULL;
923 set_capacity(lo->lo_disk, 0);
924 invalidate_bdev(bdev);
925 bd_set_size(bdev, 0);
926 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
927 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
928 lo->lo_state = Lo_unbound;
932 /* This is safe: open() is still holding a reference. */
933 module_put(THIS_MODULE);
938 loop_release_xfer(struct loop_device *lo)
941 struct loop_func_table *xfer = lo->lo_encryption;
945 err = xfer->release(lo);
947 lo->lo_encryption = NULL;
948 module_put(xfer->owner);
954 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
955 const struct loop_info64 *i)
960 struct module *owner = xfer->owner;
962 if (!try_module_get(owner))
965 err = xfer->init(lo, i);
969 lo->lo_encryption = xfer;
974 static int loop_clr_fd(struct loop_device *lo)
976 struct file *filp = lo->lo_backing_file;
977 gfp_t gfp = lo->old_gfp_mask;
978 struct block_device *bdev = lo->lo_device;
980 if (lo->lo_state != Lo_bound)
983 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
989 spin_lock_irq(&lo->lo_lock);
990 lo->lo_state = Lo_rundown;
991 spin_unlock_irq(&lo->lo_lock);
993 kthread_stop(lo->lo_thread);
995 spin_lock_irq(&lo->lo_lock);
996 lo->lo_backing_file = NULL;
997 spin_unlock_irq(&lo->lo_lock);
999 loop_release_xfer(lo);
1000 lo->transfer = NULL;
1002 lo->lo_device = NULL;
1003 lo->lo_encryption = NULL;
1005 lo->lo_sizelimit = 0;
1006 lo->lo_encrypt_key_size = 0;
1007 lo->lo_thread = NULL;
1008 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1009 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1010 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1013 invalidate_bdev(bdev);
1015 set_capacity(lo->lo_disk, 0);
1016 loop_sysfs_exit(lo);
1018 bd_set_size(bdev, 0);
1019 /* let user-space know about this change */
1020 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1022 mapping_set_gfp_mask(filp->f_mapping, gfp);
1023 lo->lo_state = Lo_unbound;
1024 /* This is safe: open() is still holding a reference. */
1025 module_put(THIS_MODULE);
1026 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1027 ioctl_by_bdev(bdev, BLKRRPART, 0);
1030 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1031 mutex_unlock(&lo->lo_ctl_mutex);
1033 * Need not hold lo_ctl_mutex to fput backing file.
1034 * Calling fput holding lo_ctl_mutex triggers a circular
1035 * lock dependency possibility warning as fput can take
1036 * bd_mutex which is usually taken before lo_ctl_mutex.
1043 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1046 struct loop_func_table *xfer;
1047 uid_t uid = current_uid();
1049 if (lo->lo_encrypt_key_size &&
1050 lo->lo_key_owner != uid &&
1051 !capable(CAP_SYS_ADMIN))
1053 if (lo->lo_state != Lo_bound)
1055 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1058 err = loop_release_xfer(lo);
1062 if (info->lo_encrypt_type) {
1063 unsigned int type = info->lo_encrypt_type;
1065 if (type >= MAX_LO_CRYPT)
1067 xfer = xfer_funcs[type];
1073 err = loop_init_xfer(lo, xfer, info);
1077 if (lo->lo_offset != info->lo_offset ||
1078 lo->lo_sizelimit != info->lo_sizelimit) {
1079 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1082 loop_config_discard(lo);
1084 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1085 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1086 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1087 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1091 lo->transfer = xfer->transfer;
1092 lo->ioctl = xfer->ioctl;
1094 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1095 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1096 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1098 if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1099 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1100 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1101 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1102 ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
1105 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1106 lo->lo_init[0] = info->lo_init[0];
1107 lo->lo_init[1] = info->lo_init[1];
1108 if (info->lo_encrypt_key_size) {
1109 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1110 info->lo_encrypt_key_size);
1111 lo->lo_key_owner = uid;
1118 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1120 struct file *file = lo->lo_backing_file;
1124 if (lo->lo_state != Lo_bound)
1126 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1129 memset(info, 0, sizeof(*info));
1130 info->lo_number = lo->lo_number;
1131 info->lo_device = huge_encode_dev(stat.dev);
1132 info->lo_inode = stat.ino;
1133 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1134 info->lo_offset = lo->lo_offset;
1135 info->lo_sizelimit = lo->lo_sizelimit;
1136 info->lo_flags = lo->lo_flags;
1137 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1138 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1139 info->lo_encrypt_type =
1140 lo->lo_encryption ? lo->lo_encryption->number : 0;
1141 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1142 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1143 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1144 lo->lo_encrypt_key_size);
1150 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1152 memset(info64, 0, sizeof(*info64));
1153 info64->lo_number = info->lo_number;
1154 info64->lo_device = info->lo_device;
1155 info64->lo_inode = info->lo_inode;
1156 info64->lo_rdevice = info->lo_rdevice;
1157 info64->lo_offset = info->lo_offset;
1158 info64->lo_sizelimit = 0;
1159 info64->lo_encrypt_type = info->lo_encrypt_type;
1160 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1161 info64->lo_flags = info->lo_flags;
1162 info64->lo_init[0] = info->lo_init[0];
1163 info64->lo_init[1] = info->lo_init[1];
1164 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1165 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1167 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1168 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1172 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1174 memset(info, 0, sizeof(*info));
1175 info->lo_number = info64->lo_number;
1176 info->lo_device = info64->lo_device;
1177 info->lo_inode = info64->lo_inode;
1178 info->lo_rdevice = info64->lo_rdevice;
1179 info->lo_offset = info64->lo_offset;
1180 info->lo_encrypt_type = info64->lo_encrypt_type;
1181 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1182 info->lo_flags = info64->lo_flags;
1183 info->lo_init[0] = info64->lo_init[0];
1184 info->lo_init[1] = info64->lo_init[1];
1185 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1186 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1188 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1189 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1191 /* error in case values were truncated */
1192 if (info->lo_device != info64->lo_device ||
1193 info->lo_rdevice != info64->lo_rdevice ||
1194 info->lo_inode != info64->lo_inode ||
1195 info->lo_offset != info64->lo_offset)
1202 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1204 struct loop_info info;
1205 struct loop_info64 info64;
1207 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1209 loop_info64_from_old(&info, &info64);
1210 return loop_set_status(lo, &info64);
1214 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1216 struct loop_info64 info64;
1218 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1220 return loop_set_status(lo, &info64);
1224 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1225 struct loop_info info;
1226 struct loop_info64 info64;
1232 err = loop_get_status(lo, &info64);
1234 err = loop_info64_to_old(&info64, &info);
1235 if (!err && copy_to_user(arg, &info, sizeof(info)))
1242 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1243 struct loop_info64 info64;
1249 err = loop_get_status(lo, &info64);
1250 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1256 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1263 if (unlikely(lo->lo_state != Lo_bound))
1265 err = figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1268 sec = get_capacity(lo->lo_disk);
1269 /* the width of sector_t may be narrow for bit-shift */
1272 bd_set_size(bdev, sz);
1273 /* let user-space know about the new size */
1274 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1280 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1281 unsigned int cmd, unsigned long arg)
1283 struct loop_device *lo = bdev->bd_disk->private_data;
1286 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1289 err = loop_set_fd(lo, mode, bdev, arg);
1291 case LOOP_CHANGE_FD:
1292 err = loop_change_fd(lo, bdev, arg);
1295 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1296 err = loop_clr_fd(lo);
1300 case LOOP_SET_STATUS:
1302 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1303 err = loop_set_status_old(lo,
1304 (struct loop_info __user *)arg);
1306 case LOOP_GET_STATUS:
1307 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1309 case LOOP_SET_STATUS64:
1311 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1312 err = loop_set_status64(lo,
1313 (struct loop_info64 __user *) arg);
1315 case LOOP_GET_STATUS64:
1316 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1318 case LOOP_SET_CAPACITY:
1320 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1321 err = loop_set_capacity(lo, bdev);
1324 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1326 mutex_unlock(&lo->lo_ctl_mutex);
1332 #ifdef CONFIG_COMPAT
1333 struct compat_loop_info {
1334 compat_int_t lo_number; /* ioctl r/o */
1335 compat_dev_t lo_device; /* ioctl r/o */
1336 compat_ulong_t lo_inode; /* ioctl r/o */
1337 compat_dev_t lo_rdevice; /* ioctl r/o */
1338 compat_int_t lo_offset;
1339 compat_int_t lo_encrypt_type;
1340 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1341 compat_int_t lo_flags; /* ioctl r/o */
1342 char lo_name[LO_NAME_SIZE];
1343 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1344 compat_ulong_t lo_init[2];
1349 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1350 * - noinlined to reduce stack space usage in main part of driver
1353 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1354 struct loop_info64 *info64)
1356 struct compat_loop_info info;
1358 if (copy_from_user(&info, arg, sizeof(info)))
1361 memset(info64, 0, sizeof(*info64));
1362 info64->lo_number = info.lo_number;
1363 info64->lo_device = info.lo_device;
1364 info64->lo_inode = info.lo_inode;
1365 info64->lo_rdevice = info.lo_rdevice;
1366 info64->lo_offset = info.lo_offset;
1367 info64->lo_sizelimit = 0;
1368 info64->lo_encrypt_type = info.lo_encrypt_type;
1369 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1370 info64->lo_flags = info.lo_flags;
1371 info64->lo_init[0] = info.lo_init[0];
1372 info64->lo_init[1] = info.lo_init[1];
1373 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1374 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1376 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1377 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1382 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1383 * - noinlined to reduce stack space usage in main part of driver
1386 loop_info64_to_compat(const struct loop_info64 *info64,
1387 struct compat_loop_info __user *arg)
1389 struct compat_loop_info info;
1391 memset(&info, 0, sizeof(info));
1392 info.lo_number = info64->lo_number;
1393 info.lo_device = info64->lo_device;
1394 info.lo_inode = info64->lo_inode;
1395 info.lo_rdevice = info64->lo_rdevice;
1396 info.lo_offset = info64->lo_offset;
1397 info.lo_encrypt_type = info64->lo_encrypt_type;
1398 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1399 info.lo_flags = info64->lo_flags;
1400 info.lo_init[0] = info64->lo_init[0];
1401 info.lo_init[1] = info64->lo_init[1];
1402 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1403 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1405 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1406 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1408 /* error in case values were truncated */
1409 if (info.lo_device != info64->lo_device ||
1410 info.lo_rdevice != info64->lo_rdevice ||
1411 info.lo_inode != info64->lo_inode ||
1412 info.lo_offset != info64->lo_offset ||
1413 info.lo_init[0] != info64->lo_init[0] ||
1414 info.lo_init[1] != info64->lo_init[1])
1417 if (copy_to_user(arg, &info, sizeof(info)))
1423 loop_set_status_compat(struct loop_device *lo,
1424 const struct compat_loop_info __user *arg)
1426 struct loop_info64 info64;
1429 ret = loop_info64_from_compat(arg, &info64);
1432 return loop_set_status(lo, &info64);
1436 loop_get_status_compat(struct loop_device *lo,
1437 struct compat_loop_info __user *arg)
1439 struct loop_info64 info64;
1445 err = loop_get_status(lo, &info64);
1447 err = loop_info64_to_compat(&info64, arg);
1451 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1452 unsigned int cmd, unsigned long arg)
1454 struct loop_device *lo = bdev->bd_disk->private_data;
1458 case LOOP_SET_STATUS:
1459 mutex_lock(&lo->lo_ctl_mutex);
1460 err = loop_set_status_compat(
1461 lo, (const struct compat_loop_info __user *) arg);
1462 mutex_unlock(&lo->lo_ctl_mutex);
1464 case LOOP_GET_STATUS:
1465 mutex_lock(&lo->lo_ctl_mutex);
1466 err = loop_get_status_compat(
1467 lo, (struct compat_loop_info __user *) arg);
1468 mutex_unlock(&lo->lo_ctl_mutex);
1470 case LOOP_SET_CAPACITY:
1472 case LOOP_GET_STATUS64:
1473 case LOOP_SET_STATUS64:
1474 arg = (unsigned long) compat_ptr(arg);
1476 case LOOP_CHANGE_FD:
1477 err = lo_ioctl(bdev, mode, cmd, arg);
1487 static int lo_open(struct block_device *bdev, fmode_t mode)
1489 struct loop_device *lo;
1492 mutex_lock(&loop_index_mutex);
1493 lo = bdev->bd_disk->private_data;
1499 mutex_lock(&lo->lo_ctl_mutex);
1501 mutex_unlock(&lo->lo_ctl_mutex);
1503 mutex_unlock(&loop_index_mutex);
1507 static int lo_release(struct gendisk *disk, fmode_t mode)
1509 struct loop_device *lo = disk->private_data;
1512 mutex_lock(&lo->lo_ctl_mutex);
1514 if (--lo->lo_refcnt)
1517 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1519 * In autoclear mode, stop the loop thread
1520 * and remove configuration after last close.
1522 err = loop_clr_fd(lo);
1527 * Otherwise keep thread (if running) and config,
1528 * but flush possible ongoing bios in thread.
1534 mutex_unlock(&lo->lo_ctl_mutex);
1539 static const struct block_device_operations lo_fops = {
1540 .owner = THIS_MODULE,
1542 .release = lo_release,
1544 #ifdef CONFIG_COMPAT
1545 .compat_ioctl = lo_compat_ioctl,
1550 * And now the modules code and kernel interface.
1552 static int max_loop;
1553 module_param(max_loop, int, S_IRUGO);
1554 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1555 module_param(max_part, int, S_IRUGO);
1556 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1557 MODULE_LICENSE("GPL");
1558 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1560 int loop_register_transfer(struct loop_func_table *funcs)
1562 unsigned int n = funcs->number;
1564 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1566 xfer_funcs[n] = funcs;
1570 static int unregister_transfer_cb(int id, void *ptr, void *data)
1572 struct loop_device *lo = ptr;
1573 struct loop_func_table *xfer = data;
1575 mutex_lock(&lo->lo_ctl_mutex);
1576 if (lo->lo_encryption == xfer)
1577 loop_release_xfer(lo);
1578 mutex_unlock(&lo->lo_ctl_mutex);
1582 int loop_unregister_transfer(int number)
1584 unsigned int n = number;
1585 struct loop_func_table *xfer;
1587 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1590 xfer_funcs[n] = NULL;
1591 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1595 EXPORT_SYMBOL(loop_register_transfer);
1596 EXPORT_SYMBOL(loop_unregister_transfer);
1598 static int loop_add(struct loop_device **l, int i)
1600 struct loop_device *lo;
1601 struct gendisk *disk;
1604 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1610 err = idr_pre_get(&loop_index_idr, GFP_KERNEL);
1617 /* create specific i in the index */
1618 err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1619 if (err >= 0 && i != m) {
1620 idr_remove(&loop_index_idr, m);
1623 } else if (i == -1) {
1626 /* get next free nr */
1627 err = idr_get_new(&loop_index_idr, lo, &m);
1636 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1640 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1642 goto out_free_queue;
1645 * Disable partition scanning by default. The in-kernel partition
1646 * scanning can be requested individually per-device during its
1647 * setup. Userspace can always add and remove partitions from all
1648 * devices. The needed partition minors are allocated from the
1649 * extended minor space, the main loop device numbers will continue
1650 * to match the loop minors, regardless of the number of partitions
1653 * If max_part is given, partition scanning is globally enabled for
1654 * all loop devices. The minors for the main loop devices will be
1655 * multiples of max_part.
1657 * Note: Global-for-all-devices, set-only-at-init, read-only module
1658 * parameteters like 'max_loop' and 'max_part' make things needlessly
1659 * complicated, are too static, inflexible and may surprise
1660 * userspace tools. Parameters like this in general should be avoided.
1663 disk->flags |= GENHD_FL_NO_PART_SCAN;
1664 disk->flags |= GENHD_FL_EXT_DEVT;
1665 mutex_init(&lo->lo_ctl_mutex);
1667 lo->lo_thread = NULL;
1668 init_waitqueue_head(&lo->lo_event);
1669 spin_lock_init(&lo->lo_lock);
1670 disk->major = LOOP_MAJOR;
1671 disk->first_minor = i << part_shift;
1672 disk->fops = &lo_fops;
1673 disk->private_data = lo;
1674 disk->queue = lo->lo_queue;
1675 sprintf(disk->disk_name, "loop%d", i);
1678 return lo->lo_number;
1681 blk_cleanup_queue(lo->lo_queue);
1683 idr_remove(&loop_index_idr, i);
1690 static void loop_remove(struct loop_device *lo)
1692 del_gendisk(lo->lo_disk);
1693 blk_cleanup_queue(lo->lo_queue);
1694 put_disk(lo->lo_disk);
1698 static int find_free_cb(int id, void *ptr, void *data)
1700 struct loop_device *lo = ptr;
1701 struct loop_device **l = data;
1703 if (lo->lo_state == Lo_unbound) {
1710 static int loop_lookup(struct loop_device **l, int i)
1712 struct loop_device *lo;
1718 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1721 ret = lo->lo_number;
1726 /* lookup and return a specific i */
1727 lo = idr_find(&loop_index_idr, i);
1730 ret = lo->lo_number;
1736 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1738 struct loop_device *lo;
1739 struct kobject *kobj;
1742 mutex_lock(&loop_index_mutex);
1743 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1745 err = loop_add(&lo, MINOR(dev) >> part_shift);
1749 kobj = get_disk(lo->lo_disk);
1750 mutex_unlock(&loop_index_mutex);
1756 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1759 struct loop_device *lo;
1762 mutex_lock(&loop_index_mutex);
1765 ret = loop_lookup(&lo, parm);
1770 ret = loop_add(&lo, parm);
1772 case LOOP_CTL_REMOVE:
1773 ret = loop_lookup(&lo, parm);
1776 mutex_lock(&lo->lo_ctl_mutex);
1777 if (lo->lo_state != Lo_unbound) {
1779 mutex_unlock(&lo->lo_ctl_mutex);
1782 if (lo->lo_refcnt > 0) {
1784 mutex_unlock(&lo->lo_ctl_mutex);
1787 lo->lo_disk->private_data = NULL;
1788 mutex_unlock(&lo->lo_ctl_mutex);
1789 idr_remove(&loop_index_idr, lo->lo_number);
1792 case LOOP_CTL_GET_FREE:
1793 ret = loop_lookup(&lo, -1);
1796 ret = loop_add(&lo, -1);
1798 mutex_unlock(&loop_index_mutex);
1803 static const struct file_operations loop_ctl_fops = {
1804 .open = nonseekable_open,
1805 .unlocked_ioctl = loop_control_ioctl,
1806 .compat_ioctl = loop_control_ioctl,
1807 .owner = THIS_MODULE,
1808 .llseek = noop_llseek,
1811 static struct miscdevice loop_misc = {
1812 .minor = LOOP_CTRL_MINOR,
1813 .name = "loop-control",
1814 .fops = &loop_ctl_fops,
1817 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1818 MODULE_ALIAS("devname:loop-control");
1820 static int __init loop_init(void)
1823 unsigned long range;
1824 struct loop_device *lo;
1827 err = misc_register(&loop_misc);
1833 part_shift = fls(max_part);
1836 * Adjust max_part according to part_shift as it is exported
1837 * to user space so that user can decide correct minor number
1838 * if [s]he want to create more devices.
1840 * Note that -1 is required because partition 0 is reserved
1841 * for the whole disk.
1843 max_part = (1UL << part_shift) - 1;
1846 if ((1UL << part_shift) > DISK_MAX_PARTS) {
1851 if (max_loop > 1UL << (MINORBITS - part_shift)) {
1857 * If max_loop is specified, create that many devices upfront.
1858 * This also becomes a hard limit. If max_loop is not specified,
1859 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1860 * init time. Loop devices can be requested on-demand with the
1861 * /dev/loop-control interface, or be instantiated by accessing
1862 * a 'dead' device node.
1866 range = max_loop << part_shift;
1868 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1869 range = 1UL << MINORBITS;
1872 if (register_blkdev(LOOP_MAJOR, "loop")) {
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_remove_all(&loop_index_idr);
1910 idr_destroy(&loop_index_idr);
1912 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1913 unregister_blkdev(LOOP_MAJOR, "loop");
1915 misc_deregister(&loop_misc);
1918 module_init(loop_init);
1919 module_exit(loop_exit);
1922 static int __init max_loop_setup(char *str)
1924 max_loop = simple_strtol(str, NULL, 0);
1928 __setup("max_loop=", max_loop_setup);