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/smp_lock.h>
65 #include <linux/swap.h>
66 #include <linux/slab.h>
67 #include <linux/loop.h>
68 #include <linux/compat.h>
69 #include <linux/suspend.h>
70 #include <linux/freezer.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/gfp.h>
76 #include <linux/kthread.h>
77 #include <linux/splice.h>
79 #include <asm/uaccess.h>
81 static LIST_HEAD(loop_devices);
82 static DEFINE_MUTEX(loop_devices_mutex);
85 static int part_shift;
90 static int transfer_none(struct loop_device *lo, int cmd,
91 struct page *raw_page, unsigned raw_off,
92 struct page *loop_page, unsigned loop_off,
93 int size, sector_t real_block)
95 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
96 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
99 memcpy(loop_buf, raw_buf, size);
101 memcpy(raw_buf, loop_buf, size);
103 kunmap_atomic(raw_buf, KM_USER0);
104 kunmap_atomic(loop_buf, KM_USER1);
109 static int transfer_xor(struct loop_device *lo, int cmd,
110 struct page *raw_page, unsigned raw_off,
111 struct page *loop_page, unsigned loop_off,
112 int size, sector_t real_block)
114 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
115 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
116 char *in, *out, *key;
127 key = lo->lo_encrypt_key;
128 keysize = lo->lo_encrypt_key_size;
129 for (i = 0; i < size; i++)
130 *out++ = *in++ ^ key[(i & 511) % keysize];
132 kunmap_atomic(raw_buf, KM_USER0);
133 kunmap_atomic(loop_buf, KM_USER1);
138 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140 if (unlikely(info->lo_encrypt_key_size <= 0))
145 static struct loop_func_table none_funcs = {
146 .number = LO_CRYPT_NONE,
147 .transfer = transfer_none,
150 static struct loop_func_table xor_funcs = {
151 .number = LO_CRYPT_XOR,
152 .transfer = transfer_xor,
156 /* xfer_funcs[0] is special - its release function is never called */
157 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164 loff_t size, offset, loopsize;
166 /* Compute loopsize in bytes */
167 size = i_size_read(file->f_mapping->host);
168 offset = lo->lo_offset;
169 loopsize = size - offset;
170 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
171 loopsize = lo->lo_sizelimit;
174 * Unfortunately, if we want to do I/O on the device,
175 * the number of 512-byte sectors has to fit into a sector_t.
177 return loopsize >> 9;
181 figure_loop_size(struct loop_device *lo)
183 loff_t size = get_loop_size(lo, lo->lo_backing_file);
184 sector_t x = (sector_t)size;
186 if (unlikely((loff_t)x != size))
189 set_capacity(lo->lo_disk, x);
194 lo_do_transfer(struct loop_device *lo, int cmd,
195 struct page *rpage, unsigned roffs,
196 struct page *lpage, unsigned loffs,
197 int size, sector_t rblock)
199 if (unlikely(!lo->transfer))
202 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
206 * do_lo_send_aops - helper for writing data to a loop device
208 * This is the fast version for backing filesystems which implement the address
209 * space operations write_begin and write_end.
211 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
212 loff_t pos, struct page *unused)
214 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
215 struct address_space *mapping = file->f_mapping;
217 unsigned offset, bv_offs;
220 mutex_lock(&mapping->host->i_mutex);
221 index = pos >> PAGE_CACHE_SHIFT;
222 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
223 bv_offs = bvec->bv_offset;
227 unsigned size, copied;
232 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
233 size = PAGE_CACHE_SIZE - offset;
237 ret = pagecache_write_begin(file, mapping, pos, size, 0,
242 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
243 bvec->bv_page, bv_offs, size, IV);
245 if (unlikely(transfer_result))
248 ret = pagecache_write_end(file, mapping, pos, size, copied,
250 if (ret < 0 || ret != copied)
253 if (unlikely(transfer_result))
264 mutex_unlock(&mapping->host->i_mutex);
272 * __do_lo_send_write - helper for writing data to a loop device
274 * This helper just factors out common code between do_lo_send_direct_write()
275 * and do_lo_send_write().
277 static int __do_lo_send_write(struct file *file,
278 u8 *buf, const int len, loff_t pos)
281 mm_segment_t old_fs = get_fs();
284 bw = file->f_op->write(file, buf, len, &pos);
286 if (likely(bw == len))
288 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
289 (unsigned long long)pos, len);
296 * do_lo_send_direct_write - helper for writing data to a loop device
298 * This is the fast, non-transforming version for backing filesystems which do
299 * not implement the address space operations write_begin and write_end.
300 * It uses the write file operation which should be present on all writeable
303 static int do_lo_send_direct_write(struct loop_device *lo,
304 struct bio_vec *bvec, loff_t pos, struct page *page)
306 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
307 kmap(bvec->bv_page) + bvec->bv_offset,
309 kunmap(bvec->bv_page);
315 * do_lo_send_write - helper for writing data to a loop device
317 * This is the slow, transforming version for filesystems which do not
318 * implement the address space operations write_begin and write_end. It
319 * uses the write file operation which should be present on all writeable
322 * Using fops->write is slower than using aops->{prepare,commit}_write in the
323 * transforming case because we need to double buffer the data as we cannot do
324 * the transformations in place as we do not have direct access to the
325 * destination pages of the backing file.
327 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
328 loff_t pos, struct page *page)
330 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
331 bvec->bv_offset, bvec->bv_len, pos >> 9);
333 return __do_lo_send_write(lo->lo_backing_file,
334 page_address(page), bvec->bv_len,
336 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
337 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
343 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
345 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
347 struct bio_vec *bvec;
348 struct page *page = NULL;
351 do_lo_send = do_lo_send_aops;
352 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
353 do_lo_send = do_lo_send_direct_write;
354 if (lo->transfer != transfer_none) {
355 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
359 do_lo_send = do_lo_send_write;
362 bio_for_each_segment(bvec, bio, i) {
363 ret = do_lo_send(lo, bvec, pos, page);
375 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
380 struct lo_read_data {
381 struct loop_device *lo;
388 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
389 struct splice_desc *sd)
391 struct lo_read_data *p = sd->u.data;
392 struct loop_device *lo = p->lo;
393 struct page *page = buf->page;
398 ret = buf->ops->confirm(pipe, buf);
402 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
408 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
409 printk(KERN_ERR "loop: transfer error block %ld\n",
414 flush_dcache_page(p->page);
423 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
425 return __splice_from_pipe(pipe, sd, lo_splice_actor);
429 do_lo_receive(struct loop_device *lo,
430 struct bio_vec *bvec, int bsize, loff_t pos)
432 struct lo_read_data cookie;
433 struct splice_desc sd;
438 cookie.page = bvec->bv_page;
439 cookie.offset = bvec->bv_offset;
440 cookie.bsize = bsize;
443 sd.total_len = bvec->bv_len;
448 file = lo->lo_backing_file;
449 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
458 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
460 struct bio_vec *bvec;
463 bio_for_each_segment(bvec, bio, i) {
464 ret = do_lo_receive(lo, bvec, bsize, pos);
472 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
477 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
478 if (bio_rw(bio) == WRITE)
479 ret = lo_send(lo, bio, pos);
481 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
486 * Add bio to back of pending list
488 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
490 if (lo->lo_biotail) {
491 lo->lo_biotail->bi_next = bio;
492 lo->lo_biotail = bio;
494 lo->lo_bio = lo->lo_biotail = bio;
498 * Grab first pending buffer
500 static struct bio *loop_get_bio(struct loop_device *lo)
504 if ((bio = lo->lo_bio)) {
505 if (bio == lo->lo_biotail)
506 lo->lo_biotail = NULL;
507 lo->lo_bio = bio->bi_next;
514 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
516 struct loop_device *lo = q->queuedata;
517 int rw = bio_rw(old_bio);
522 BUG_ON(!lo || (rw != READ && rw != WRITE));
524 spin_lock_irq(&lo->lo_lock);
525 if (lo->lo_state != Lo_bound)
527 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
529 loop_add_bio(lo, old_bio);
530 wake_up(&lo->lo_event);
531 spin_unlock_irq(&lo->lo_lock);
535 spin_unlock_irq(&lo->lo_lock);
536 bio_io_error(old_bio);
541 * kick off io on the underlying address space
543 static void loop_unplug(struct request_queue *q)
545 struct loop_device *lo = q->queuedata;
547 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
548 blk_run_address_space(lo->lo_backing_file->f_mapping);
551 struct switch_request {
553 struct completion wait;
556 static void do_loop_switch(struct loop_device *, struct switch_request *);
558 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
560 if (unlikely(!bio->bi_bdev)) {
561 do_loop_switch(lo, bio->bi_private);
564 int ret = do_bio_filebacked(lo, bio);
570 * worker thread that handles reads/writes to file backed loop devices,
571 * to avoid blocking in our make_request_fn. it also does loop decrypting
572 * on reads for block backed loop, as that is too heavy to do from
573 * b_end_io context where irqs may be disabled.
575 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
576 * calling kthread_stop(). Therefore once kthread_should_stop() is
577 * true, make_request will not place any more requests. Therefore
578 * once kthread_should_stop() is true and lo_bio is NULL, we are
579 * done with the loop.
581 static int loop_thread(void *data)
583 struct loop_device *lo = data;
586 set_user_nice(current, -20);
588 while (!kthread_should_stop() || lo->lo_bio) {
590 wait_event_interruptible(lo->lo_event,
591 lo->lo_bio || kthread_should_stop());
595 spin_lock_irq(&lo->lo_lock);
596 bio = loop_get_bio(lo);
597 spin_unlock_irq(&lo->lo_lock);
600 loop_handle_bio(lo, bio);
607 * loop_switch performs the hard work of switching a backing store.
608 * First it needs to flush existing IO, it does this by sending a magic
609 * BIO down the pipe. The completion of this BIO does the actual switch.
611 static int loop_switch(struct loop_device *lo, struct file *file)
613 struct switch_request w;
614 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
617 init_completion(&w.wait);
619 bio->bi_private = &w;
621 loop_make_request(lo->lo_queue, bio);
622 wait_for_completion(&w.wait);
627 * Do the actual switch; called from the BIO completion routine
629 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
631 struct file *file = p->file;
632 struct file *old_file = lo->lo_backing_file;
633 struct address_space *mapping = file->f_mapping;
635 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
636 lo->lo_backing_file = file;
637 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
638 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
639 lo->old_gfp_mask = mapping_gfp_mask(mapping);
640 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
646 * loop_change_fd switched the backing store of a loopback device to
647 * a new file. This is useful for operating system installers to free up
648 * the original file and in High Availability environments to switch to
649 * an alternative location for the content in case of server meltdown.
650 * This can only work if the loop device is used read-only, and if the
651 * new backing store is the same size and type as the old backing store.
653 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
656 struct file *file, *old_file;
661 if (lo->lo_state != Lo_bound)
664 /* the loop device has to be read-only */
666 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
674 inode = file->f_mapping->host;
675 old_file = lo->lo_backing_file;
679 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
682 /* new backing store needs to support loop (eg splice_read) */
683 if (!inode->i_fop->splice_read)
686 /* size of the new backing store needs to be the same */
687 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
691 error = loop_switch(lo, file);
697 ioctl_by_bdev(bdev, BLKRRPART, 0);
706 static inline int is_loop_device(struct file *file)
708 struct inode *i = file->f_mapping->host;
710 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
713 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
714 struct block_device *bdev, unsigned int arg)
716 struct file *file, *f;
718 struct address_space *mapping;
719 unsigned lo_blocksize;
724 /* This is safe, since we have a reference from open(). */
725 __module_get(THIS_MODULE);
733 if (lo->lo_state != Lo_unbound)
736 /* Avoid recursion */
738 while (is_loop_device(f)) {
739 struct loop_device *l;
741 if (f->f_mapping->host->i_bdev == bdev)
744 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
745 if (l->lo_state == Lo_unbound) {
749 f = l->lo_backing_file;
752 mapping = file->f_mapping;
753 inode = mapping->host;
755 if (!(file->f_mode & FMODE_WRITE))
756 lo_flags |= LO_FLAGS_READ_ONLY;
759 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
760 const struct address_space_operations *aops = mapping->a_ops;
762 * If we can't read - sorry. If we only can't write - well,
763 * it's going to be read-only.
765 if (!file->f_op->splice_read)
767 if (aops->write_begin)
768 lo_flags |= LO_FLAGS_USE_AOPS;
769 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
770 lo_flags |= LO_FLAGS_READ_ONLY;
772 lo_blocksize = S_ISBLK(inode->i_mode) ?
773 inode->i_bdev->bd_block_size : PAGE_SIZE;
780 size = get_loop_size(lo, file);
782 if ((loff_t)(sector_t)size != size) {
787 if (!(mode & FMODE_WRITE))
788 lo_flags |= LO_FLAGS_READ_ONLY;
790 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
792 lo->lo_blocksize = lo_blocksize;
793 lo->lo_device = bdev;
794 lo->lo_flags = lo_flags;
795 lo->lo_backing_file = file;
796 lo->transfer = transfer_none;
798 lo->lo_sizelimit = 0;
799 lo->old_gfp_mask = mapping_gfp_mask(mapping);
800 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
802 lo->lo_bio = lo->lo_biotail = NULL;
805 * set queue make_request_fn, and add limits based on lower level
808 blk_queue_make_request(lo->lo_queue, loop_make_request);
809 lo->lo_queue->queuedata = lo;
810 lo->lo_queue->unplug_fn = loop_unplug;
812 set_capacity(lo->lo_disk, size);
813 bd_set_size(bdev, size << 9);
815 set_blocksize(bdev, lo_blocksize);
817 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
819 if (IS_ERR(lo->lo_thread)) {
820 error = PTR_ERR(lo->lo_thread);
823 lo->lo_state = Lo_bound;
824 wake_up_process(lo->lo_thread);
826 ioctl_by_bdev(bdev, BLKRRPART, 0);
830 lo->lo_thread = NULL;
831 lo->lo_device = NULL;
832 lo->lo_backing_file = NULL;
834 set_capacity(lo->lo_disk, 0);
835 invalidate_bdev(bdev);
836 bd_set_size(bdev, 0);
837 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
838 lo->lo_state = Lo_unbound;
842 /* This is safe: open() is still holding a reference. */
843 module_put(THIS_MODULE);
848 loop_release_xfer(struct loop_device *lo)
851 struct loop_func_table *xfer = lo->lo_encryption;
855 err = xfer->release(lo);
857 lo->lo_encryption = NULL;
858 module_put(xfer->owner);
864 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
865 const struct loop_info64 *i)
870 struct module *owner = xfer->owner;
872 if (!try_module_get(owner))
875 err = xfer->init(lo, i);
879 lo->lo_encryption = xfer;
884 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
886 struct file *filp = lo->lo_backing_file;
887 gfp_t gfp = lo->old_gfp_mask;
889 if (lo->lo_state != Lo_bound)
892 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
898 spin_lock_irq(&lo->lo_lock);
899 lo->lo_state = Lo_rundown;
900 spin_unlock_irq(&lo->lo_lock);
902 kthread_stop(lo->lo_thread);
904 lo->lo_backing_file = NULL;
906 loop_release_xfer(lo);
909 lo->lo_device = NULL;
910 lo->lo_encryption = NULL;
912 lo->lo_sizelimit = 0;
913 lo->lo_encrypt_key_size = 0;
915 lo->lo_thread = NULL;
916 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
917 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
918 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
920 invalidate_bdev(bdev);
921 set_capacity(lo->lo_disk, 0);
923 bd_set_size(bdev, 0);
924 mapping_set_gfp_mask(filp->f_mapping, gfp);
925 lo->lo_state = Lo_unbound;
927 /* This is safe: open() is still holding a reference. */
928 module_put(THIS_MODULE);
930 ioctl_by_bdev(bdev, BLKRRPART, 0);
935 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
938 struct loop_func_table *xfer;
939 uid_t uid = current_uid();
941 if (lo->lo_encrypt_key_size &&
942 lo->lo_key_owner != uid &&
943 !capable(CAP_SYS_ADMIN))
945 if (lo->lo_state != Lo_bound)
947 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
950 err = loop_release_xfer(lo);
954 if (info->lo_encrypt_type) {
955 unsigned int type = info->lo_encrypt_type;
957 if (type >= MAX_LO_CRYPT)
959 xfer = xfer_funcs[type];
965 err = loop_init_xfer(lo, xfer, info);
969 if (lo->lo_offset != info->lo_offset ||
970 lo->lo_sizelimit != info->lo_sizelimit) {
971 lo->lo_offset = info->lo_offset;
972 lo->lo_sizelimit = info->lo_sizelimit;
973 if (figure_loop_size(lo))
977 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
978 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
979 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
980 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
984 lo->transfer = xfer->transfer;
985 lo->ioctl = xfer->ioctl;
987 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
988 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
989 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
991 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
992 lo->lo_init[0] = info->lo_init[0];
993 lo->lo_init[1] = info->lo_init[1];
994 if (info->lo_encrypt_key_size) {
995 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
996 info->lo_encrypt_key_size);
997 lo->lo_key_owner = uid;
1004 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1006 struct file *file = lo->lo_backing_file;
1010 if (lo->lo_state != Lo_bound)
1012 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1015 memset(info, 0, sizeof(*info));
1016 info->lo_number = lo->lo_number;
1017 info->lo_device = huge_encode_dev(stat.dev);
1018 info->lo_inode = stat.ino;
1019 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1020 info->lo_offset = lo->lo_offset;
1021 info->lo_sizelimit = lo->lo_sizelimit;
1022 info->lo_flags = lo->lo_flags;
1023 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1024 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1025 info->lo_encrypt_type =
1026 lo->lo_encryption ? lo->lo_encryption->number : 0;
1027 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1028 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1029 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1030 lo->lo_encrypt_key_size);
1036 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1038 memset(info64, 0, sizeof(*info64));
1039 info64->lo_number = info->lo_number;
1040 info64->lo_device = info->lo_device;
1041 info64->lo_inode = info->lo_inode;
1042 info64->lo_rdevice = info->lo_rdevice;
1043 info64->lo_offset = info->lo_offset;
1044 info64->lo_sizelimit = 0;
1045 info64->lo_encrypt_type = info->lo_encrypt_type;
1046 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1047 info64->lo_flags = info->lo_flags;
1048 info64->lo_init[0] = info->lo_init[0];
1049 info64->lo_init[1] = info->lo_init[1];
1050 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1051 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1053 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1054 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1058 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1060 memset(info, 0, sizeof(*info));
1061 info->lo_number = info64->lo_number;
1062 info->lo_device = info64->lo_device;
1063 info->lo_inode = info64->lo_inode;
1064 info->lo_rdevice = info64->lo_rdevice;
1065 info->lo_offset = info64->lo_offset;
1066 info->lo_encrypt_type = info64->lo_encrypt_type;
1067 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1068 info->lo_flags = info64->lo_flags;
1069 info->lo_init[0] = info64->lo_init[0];
1070 info->lo_init[1] = info64->lo_init[1];
1071 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1072 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1074 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1075 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1077 /* error in case values were truncated */
1078 if (info->lo_device != info64->lo_device ||
1079 info->lo_rdevice != info64->lo_rdevice ||
1080 info->lo_inode != info64->lo_inode ||
1081 info->lo_offset != info64->lo_offset)
1088 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1090 struct loop_info info;
1091 struct loop_info64 info64;
1093 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1095 loop_info64_from_old(&info, &info64);
1096 return loop_set_status(lo, &info64);
1100 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1102 struct loop_info64 info64;
1104 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1106 return loop_set_status(lo, &info64);
1110 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1111 struct loop_info info;
1112 struct loop_info64 info64;
1118 err = loop_get_status(lo, &info64);
1120 err = loop_info64_to_old(&info64, &info);
1121 if (!err && copy_to_user(arg, &info, sizeof(info)))
1128 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1129 struct loop_info64 info64;
1135 err = loop_get_status(lo, &info64);
1136 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1142 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1143 unsigned int cmd, unsigned long arg)
1145 struct loop_device *lo = bdev->bd_disk->private_data;
1148 mutex_lock(&lo->lo_ctl_mutex);
1151 err = loop_set_fd(lo, mode, bdev, arg);
1153 case LOOP_CHANGE_FD:
1154 err = loop_change_fd(lo, bdev, arg);
1157 err = loop_clr_fd(lo, bdev);
1159 case LOOP_SET_STATUS:
1160 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1162 case LOOP_GET_STATUS:
1163 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1165 case LOOP_SET_STATUS64:
1166 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1168 case LOOP_GET_STATUS64:
1169 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1172 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1174 mutex_unlock(&lo->lo_ctl_mutex);
1178 #ifdef CONFIG_COMPAT
1179 struct compat_loop_info {
1180 compat_int_t lo_number; /* ioctl r/o */
1181 compat_dev_t lo_device; /* ioctl r/o */
1182 compat_ulong_t lo_inode; /* ioctl r/o */
1183 compat_dev_t lo_rdevice; /* ioctl r/o */
1184 compat_int_t lo_offset;
1185 compat_int_t lo_encrypt_type;
1186 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1187 compat_int_t lo_flags; /* ioctl r/o */
1188 char lo_name[LO_NAME_SIZE];
1189 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1190 compat_ulong_t lo_init[2];
1195 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1196 * - noinlined to reduce stack space usage in main part of driver
1199 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1200 struct loop_info64 *info64)
1202 struct compat_loop_info info;
1204 if (copy_from_user(&info, arg, sizeof(info)))
1207 memset(info64, 0, sizeof(*info64));
1208 info64->lo_number = info.lo_number;
1209 info64->lo_device = info.lo_device;
1210 info64->lo_inode = info.lo_inode;
1211 info64->lo_rdevice = info.lo_rdevice;
1212 info64->lo_offset = info.lo_offset;
1213 info64->lo_sizelimit = 0;
1214 info64->lo_encrypt_type = info.lo_encrypt_type;
1215 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1216 info64->lo_flags = info.lo_flags;
1217 info64->lo_init[0] = info.lo_init[0];
1218 info64->lo_init[1] = info.lo_init[1];
1219 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1220 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1222 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1223 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1228 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1229 * - noinlined to reduce stack space usage in main part of driver
1232 loop_info64_to_compat(const struct loop_info64 *info64,
1233 struct compat_loop_info __user *arg)
1235 struct compat_loop_info info;
1237 memset(&info, 0, sizeof(info));
1238 info.lo_number = info64->lo_number;
1239 info.lo_device = info64->lo_device;
1240 info.lo_inode = info64->lo_inode;
1241 info.lo_rdevice = info64->lo_rdevice;
1242 info.lo_offset = info64->lo_offset;
1243 info.lo_encrypt_type = info64->lo_encrypt_type;
1244 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1245 info.lo_flags = info64->lo_flags;
1246 info.lo_init[0] = info64->lo_init[0];
1247 info.lo_init[1] = info64->lo_init[1];
1248 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1249 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1251 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1252 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1254 /* error in case values were truncated */
1255 if (info.lo_device != info64->lo_device ||
1256 info.lo_rdevice != info64->lo_rdevice ||
1257 info.lo_inode != info64->lo_inode ||
1258 info.lo_offset != info64->lo_offset ||
1259 info.lo_init[0] != info64->lo_init[0] ||
1260 info.lo_init[1] != info64->lo_init[1])
1263 if (copy_to_user(arg, &info, sizeof(info)))
1269 loop_set_status_compat(struct loop_device *lo,
1270 const struct compat_loop_info __user *arg)
1272 struct loop_info64 info64;
1275 ret = loop_info64_from_compat(arg, &info64);
1278 return loop_set_status(lo, &info64);
1282 loop_get_status_compat(struct loop_device *lo,
1283 struct compat_loop_info __user *arg)
1285 struct loop_info64 info64;
1291 err = loop_get_status(lo, &info64);
1293 err = loop_info64_to_compat(&info64, arg);
1297 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1298 unsigned int cmd, unsigned long arg)
1300 struct loop_device *lo = bdev->bd_disk->private_data;
1304 case LOOP_SET_STATUS:
1305 mutex_lock(&lo->lo_ctl_mutex);
1306 err = loop_set_status_compat(
1307 lo, (const struct compat_loop_info __user *) arg);
1308 mutex_unlock(&lo->lo_ctl_mutex);
1310 case LOOP_GET_STATUS:
1311 mutex_lock(&lo->lo_ctl_mutex);
1312 err = loop_get_status_compat(
1313 lo, (struct compat_loop_info __user *) arg);
1314 mutex_unlock(&lo->lo_ctl_mutex);
1317 case LOOP_GET_STATUS64:
1318 case LOOP_SET_STATUS64:
1319 arg = (unsigned long) compat_ptr(arg);
1321 case LOOP_CHANGE_FD:
1322 err = lo_ioctl(bdev, mode, cmd, arg);
1332 static int lo_open(struct block_device *bdev, fmode_t mode)
1334 struct loop_device *lo = bdev->bd_disk->private_data;
1336 mutex_lock(&lo->lo_ctl_mutex);
1338 mutex_unlock(&lo->lo_ctl_mutex);
1343 static int lo_release(struct gendisk *disk, fmode_t mode)
1345 struct loop_device *lo = disk->private_data;
1347 mutex_lock(&lo->lo_ctl_mutex);
1350 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) && !lo->lo_refcnt)
1351 loop_clr_fd(lo, NULL);
1353 mutex_unlock(&lo->lo_ctl_mutex);
1358 static struct block_device_operations lo_fops = {
1359 .owner = THIS_MODULE,
1361 .release = lo_release,
1363 #ifdef CONFIG_COMPAT
1364 .compat_ioctl = lo_compat_ioctl,
1369 * And now the modules code and kernel interface.
1371 static int max_loop;
1372 module_param(max_loop, int, 0);
1373 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1374 module_param(max_part, int, 0);
1375 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1376 MODULE_LICENSE("GPL");
1377 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1379 int loop_register_transfer(struct loop_func_table *funcs)
1381 unsigned int n = funcs->number;
1383 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1385 xfer_funcs[n] = funcs;
1389 int loop_unregister_transfer(int number)
1391 unsigned int n = number;
1392 struct loop_device *lo;
1393 struct loop_func_table *xfer;
1395 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1398 xfer_funcs[n] = NULL;
1400 list_for_each_entry(lo, &loop_devices, lo_list) {
1401 mutex_lock(&lo->lo_ctl_mutex);
1403 if (lo->lo_encryption == xfer)
1404 loop_release_xfer(lo);
1406 mutex_unlock(&lo->lo_ctl_mutex);
1412 EXPORT_SYMBOL(loop_register_transfer);
1413 EXPORT_SYMBOL(loop_unregister_transfer);
1415 static struct loop_device *loop_alloc(int i)
1417 struct loop_device *lo;
1418 struct gendisk *disk;
1420 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1424 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1428 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1430 goto out_free_queue;
1432 mutex_init(&lo->lo_ctl_mutex);
1434 lo->lo_thread = NULL;
1435 init_waitqueue_head(&lo->lo_event);
1436 spin_lock_init(&lo->lo_lock);
1437 disk->major = LOOP_MAJOR;
1438 disk->first_minor = i << part_shift;
1439 disk->fops = &lo_fops;
1440 disk->private_data = lo;
1441 disk->queue = lo->lo_queue;
1442 sprintf(disk->disk_name, "loop%d", i);
1446 blk_cleanup_queue(lo->lo_queue);
1453 static void loop_free(struct loop_device *lo)
1455 blk_cleanup_queue(lo->lo_queue);
1456 put_disk(lo->lo_disk);
1457 list_del(&lo->lo_list);
1461 static struct loop_device *loop_init_one(int i)
1463 struct loop_device *lo;
1465 list_for_each_entry(lo, &loop_devices, lo_list) {
1466 if (lo->lo_number == i)
1472 add_disk(lo->lo_disk);
1473 list_add_tail(&lo->lo_list, &loop_devices);
1478 static void loop_del_one(struct loop_device *lo)
1480 del_gendisk(lo->lo_disk);
1484 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1486 struct loop_device *lo;
1487 struct kobject *kobj;
1489 mutex_lock(&loop_devices_mutex);
1490 lo = loop_init_one(dev & MINORMASK);
1491 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1492 mutex_unlock(&loop_devices_mutex);
1498 static int __init loop_init(void)
1501 unsigned long range;
1502 struct loop_device *lo, *next;
1505 * loop module now has a feature to instantiate underlying device
1506 * structure on-demand, provided that there is an access dev node.
1507 * However, this will not work well with user space tool that doesn't
1508 * know about such "feature". In order to not break any existing
1509 * tool, we do the following:
1511 * (1) if max_loop is specified, create that many upfront, and this
1512 * also becomes a hard limit.
1513 * (2) if max_loop is not specified, create 8 loop device on module
1514 * load, user can further extend loop device by create dev node
1515 * themselves and have kernel automatically instantiate actual
1521 part_shift = fls(max_part);
1523 if (max_loop > 1UL << (MINORBITS - part_shift))
1531 range = 1UL << (MINORBITS - part_shift);
1534 if (register_blkdev(LOOP_MAJOR, "loop"))
1537 for (i = 0; i < nr; i++) {
1541 list_add_tail(&lo->lo_list, &loop_devices);
1544 /* point of no return */
1546 list_for_each_entry(lo, &loop_devices, lo_list)
1547 add_disk(lo->lo_disk);
1549 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1550 THIS_MODULE, loop_probe, NULL, NULL);
1552 printk(KERN_INFO "loop: module loaded\n");
1556 printk(KERN_INFO "loop: out of memory\n");
1558 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1561 unregister_blkdev(LOOP_MAJOR, "loop");
1565 static void __exit loop_exit(void)
1567 unsigned long range;
1568 struct loop_device *lo, *next;
1570 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1572 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1575 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1576 unregister_blkdev(LOOP_MAJOR, "loop");
1579 module_init(loop_init);
1580 module_exit(loop_exit);
1583 static int __init max_loop_setup(char *str)
1585 max_loop = simple_strtol(str, NULL, 0);
1589 __setup("max_loop=", max_loop_setup);