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 prepare_write and/or commit_write are not available on the
45 * Anton Altaparmakov, 16 Feb 2005
48 * - Advisory locking is ignored here.
49 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
53 #include <linux/module.h>
54 #include <linux/moduleparam.h>
55 #include <linux/sched.h>
57 #include <linux/file.h>
58 #include <linux/stat.h>
59 #include <linux/errno.h>
60 #include <linux/major.h>
61 #include <linux/wait.h>
62 #include <linux/blkdev.h>
63 #include <linux/blkpg.h>
64 #include <linux/init.h>
65 #include <linux/smp_lock.h>
66 #include <linux/swap.h>
67 #include <linux/slab.h>
68 #include <linux/loop.h>
69 #include <linux/compat.h>
70 #include <linux/suspend.h>
71 #include <linux/freezer.h>
72 #include <linux/writeback.h>
73 #include <linux/buffer_head.h> /* for invalidate_bdev() */
74 #include <linux/completion.h>
75 #include <linux/highmem.h>
76 #include <linux/gfp.h>
77 #include <linux/kthread.h>
78 #include <linux/splice.h>
80 #include <asm/uaccess.h>
82 static LIST_HEAD(loop_devices);
83 static DEFINE_MUTEX(loop_devices_mutex);
86 static int part_shift;
91 static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
96 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
100 memcpy(loop_buf, raw_buf, size);
102 memcpy(raw_buf, loop_buf, size);
104 kunmap_atomic(raw_buf, KM_USER0);
105 kunmap_atomic(loop_buf, KM_USER1);
110 static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
115 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
117 char *in, *out, *key;
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
133 kunmap_atomic(raw_buf, KM_USER0);
134 kunmap_atomic(loop_buf, KM_USER1);
139 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
141 if (unlikely(info->lo_encrypt_key_size <= 0))
146 static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
151 static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
157 /* xfer_funcs[0] is special - its release function is never called */
158 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
163 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
165 loff_t size, offset, loopsize;
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
169 offset = lo->lo_offset;
170 loopsize = size - offset;
171 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
172 loopsize = lo->lo_sizelimit;
175 * Unfortunately, if we want to do I/O on the device,
176 * the number of 512-byte sectors has to fit into a sector_t.
178 return loopsize >> 9;
182 figure_loop_size(struct loop_device *lo)
184 loff_t size = get_loop_size(lo, lo->lo_backing_file);
185 sector_t x = (sector_t)size;
187 if (unlikely((loff_t)x != size))
190 set_capacity(lo->lo_disk, x);
195 lo_do_transfer(struct loop_device *lo, int cmd,
196 struct page *rpage, unsigned roffs,
197 struct page *lpage, unsigned loffs,
198 int size, sector_t rblock)
200 if (unlikely(!lo->transfer))
203 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
207 * do_lo_send_aops - helper for writing data to a loop device
209 * This is the fast version for backing filesystems which implement the address
210 * space operations write_begin and write_end.
212 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
213 loff_t pos, struct page *unused)
215 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
216 struct address_space *mapping = file->f_mapping;
218 unsigned offset, bv_offs;
221 mutex_lock(&mapping->host->i_mutex);
222 index = pos >> PAGE_CACHE_SHIFT;
223 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
224 bv_offs = bvec->bv_offset;
228 unsigned size, copied;
233 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
234 size = PAGE_CACHE_SIZE - offset;
238 ret = pagecache_write_begin(file, mapping, pos, size, 0,
243 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
244 bvec->bv_page, bv_offs, size, IV);
246 if (unlikely(transfer_result))
249 ret = pagecache_write_end(file, mapping, pos, size, copied,
251 if (ret < 0 || ret != copied)
254 if (unlikely(transfer_result))
265 mutex_unlock(&mapping->host->i_mutex);
273 * __do_lo_send_write - helper for writing data to a loop device
275 * This helper just factors out common code between do_lo_send_direct_write()
276 * and do_lo_send_write().
278 static int __do_lo_send_write(struct file *file,
279 u8 *buf, const int len, loff_t pos)
282 mm_segment_t old_fs = get_fs();
285 bw = file->f_op->write(file, buf, len, &pos);
287 if (likely(bw == len))
289 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
290 (unsigned long long)pos, len);
297 * do_lo_send_direct_write - helper for writing data to a loop device
299 * This is the fast, non-transforming version for backing filesystems which do
300 * not implement the address space operations write_begin and write_end.
301 * It uses the write file operation which should be present on all writeable
304 static int do_lo_send_direct_write(struct loop_device *lo,
305 struct bio_vec *bvec, loff_t pos, struct page *page)
307 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
308 kmap(bvec->bv_page) + bvec->bv_offset,
310 kunmap(bvec->bv_page);
316 * do_lo_send_write - helper for writing data to a loop device
318 * This is the slow, transforming version for filesystems which do not
319 * implement the address space operations write_begin and write_end. It
320 * uses the write file operation which should be present on all writeable
323 * Using fops->write is slower than using aops->{prepare,commit}_write in the
324 * transforming case because we need to double buffer the data as we cannot do
325 * the transformations in place as we do not have direct access to the
326 * destination pages of the backing file.
328 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
329 loff_t pos, struct page *page)
331 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
332 bvec->bv_offset, bvec->bv_len, pos >> 9);
334 return __do_lo_send_write(lo->lo_backing_file,
335 page_address(page), bvec->bv_len,
337 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
338 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
344 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
346 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
348 struct bio_vec *bvec;
349 struct page *page = NULL;
352 do_lo_send = do_lo_send_aops;
353 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
354 do_lo_send = do_lo_send_direct_write;
355 if (lo->transfer != transfer_none) {
356 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
360 do_lo_send = do_lo_send_write;
363 bio_for_each_segment(bvec, bio, i) {
364 ret = do_lo_send(lo, bvec, pos, page);
376 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
381 struct lo_read_data {
382 struct loop_device *lo;
389 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
390 struct splice_desc *sd)
392 struct lo_read_data *p = sd->u.data;
393 struct loop_device *lo = p->lo;
394 struct page *page = buf->page;
399 ret = buf->ops->confirm(pipe, buf);
403 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
409 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
410 printk(KERN_ERR "loop: transfer error block %ld\n",
415 flush_dcache_page(p->page);
424 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
426 return __splice_from_pipe(pipe, sd, lo_splice_actor);
430 do_lo_receive(struct loop_device *lo,
431 struct bio_vec *bvec, int bsize, loff_t pos)
433 struct lo_read_data cookie;
434 struct splice_desc sd;
439 cookie.page = bvec->bv_page;
440 cookie.offset = bvec->bv_offset;
441 cookie.bsize = bsize;
444 sd.total_len = bvec->bv_len;
449 file = lo->lo_backing_file;
450 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
459 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
461 struct bio_vec *bvec;
464 bio_for_each_segment(bvec, bio, i) {
465 ret = do_lo_receive(lo, bvec, bsize, pos);
473 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
478 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
479 if (bio_rw(bio) == WRITE)
480 ret = lo_send(lo, bio, pos);
482 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
487 * Add bio to back of pending list
489 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
491 if (lo->lo_biotail) {
492 lo->lo_biotail->bi_next = bio;
493 lo->lo_biotail = bio;
495 lo->lo_bio = lo->lo_biotail = bio;
499 * Grab first pending buffer
501 static struct bio *loop_get_bio(struct loop_device *lo)
505 if ((bio = lo->lo_bio)) {
506 if (bio == lo->lo_biotail)
507 lo->lo_biotail = NULL;
508 lo->lo_bio = bio->bi_next;
515 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
517 struct loop_device *lo = q->queuedata;
518 int rw = bio_rw(old_bio);
523 BUG_ON(!lo || (rw != READ && rw != WRITE));
525 spin_lock_irq(&lo->lo_lock);
526 if (lo->lo_state != Lo_bound)
528 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
530 loop_add_bio(lo, old_bio);
531 wake_up(&lo->lo_event);
532 spin_unlock_irq(&lo->lo_lock);
536 spin_unlock_irq(&lo->lo_lock);
537 bio_io_error(old_bio);
542 * kick off io on the underlying address space
544 static void loop_unplug(struct request_queue *q)
546 struct loop_device *lo = q->queuedata;
548 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
549 blk_run_address_space(lo->lo_backing_file->f_mapping);
552 struct switch_request {
554 struct completion wait;
557 static void do_loop_switch(struct loop_device *, struct switch_request *);
559 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
561 if (unlikely(!bio->bi_bdev)) {
562 do_loop_switch(lo, bio->bi_private);
565 int ret = do_bio_filebacked(lo, bio);
571 * worker thread that handles reads/writes to file backed loop devices,
572 * to avoid blocking in our make_request_fn. it also does loop decrypting
573 * on reads for block backed loop, as that is too heavy to do from
574 * b_end_io context where irqs may be disabled.
576 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
577 * calling kthread_stop(). Therefore once kthread_should_stop() is
578 * true, make_request will not place any more requests. Therefore
579 * once kthread_should_stop() is true and lo_bio is NULL, we are
580 * done with the loop.
582 static int loop_thread(void *data)
584 struct loop_device *lo = data;
587 set_user_nice(current, -20);
589 while (!kthread_should_stop() || lo->lo_bio) {
591 wait_event_interruptible(lo->lo_event,
592 lo->lo_bio || kthread_should_stop());
596 spin_lock_irq(&lo->lo_lock);
597 bio = loop_get_bio(lo);
598 spin_unlock_irq(&lo->lo_lock);
601 loop_handle_bio(lo, bio);
608 * loop_switch performs the hard work of switching a backing store.
609 * First it needs to flush existing IO, it does this by sending a magic
610 * BIO down the pipe. The completion of this BIO does the actual switch.
612 static int loop_switch(struct loop_device *lo, struct file *file)
614 struct switch_request w;
615 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
618 init_completion(&w.wait);
620 bio->bi_private = &w;
622 loop_make_request(lo->lo_queue, bio);
623 wait_for_completion(&w.wait);
628 * Do the actual switch; called from the BIO completion routine
630 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
632 struct file *file = p->file;
633 struct file *old_file = lo->lo_backing_file;
634 struct address_space *mapping = file->f_mapping;
636 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
637 lo->lo_backing_file = file;
638 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
639 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
640 lo->old_gfp_mask = mapping_gfp_mask(mapping);
641 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
647 * loop_change_fd switched the backing store of a loopback device to
648 * a new file. This is useful for operating system installers to free up
649 * the original file and in High Availability environments to switch to
650 * an alternative location for the content in case of server meltdown.
651 * This can only work if the loop device is used read-only, and if the
652 * new backing store is the same size and type as the old backing store.
654 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
657 struct file *file, *old_file;
662 if (lo->lo_state != Lo_bound)
665 /* the loop device has to be read-only */
667 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
675 inode = file->f_mapping->host;
676 old_file = lo->lo_backing_file;
680 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
683 /* new backing store needs to support loop (eg splice_read) */
684 if (!inode->i_fop->splice_read)
687 /* size of the new backing store needs to be the same */
688 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
692 error = loop_switch(lo, file);
698 ioctl_by_bdev(bdev, BLKRRPART, 0);
707 static inline int is_loop_device(struct file *file)
709 struct inode *i = file->f_mapping->host;
711 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
714 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
715 struct block_device *bdev, unsigned int arg)
717 struct file *file, *f;
719 struct address_space *mapping;
720 unsigned lo_blocksize;
725 /* This is safe, since we have a reference from open(). */
726 __module_get(THIS_MODULE);
734 if (lo->lo_state != Lo_unbound)
737 /* Avoid recursion */
739 while (is_loop_device(f)) {
740 struct loop_device *l;
742 if (f->f_mapping->host->i_bdev == bdev)
745 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
746 if (l->lo_state == Lo_unbound) {
750 f = l->lo_backing_file;
753 mapping = file->f_mapping;
754 inode = mapping->host;
756 if (!(file->f_mode & FMODE_WRITE))
757 lo_flags |= LO_FLAGS_READ_ONLY;
760 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
761 const struct address_space_operations *aops = mapping->a_ops;
763 * If we can't read - sorry. If we only can't write - well,
764 * it's going to be read-only.
766 if (!file->f_op->splice_read)
768 if (aops->prepare_write || aops->write_begin)
769 lo_flags |= LO_FLAGS_USE_AOPS;
770 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
771 lo_flags |= LO_FLAGS_READ_ONLY;
773 lo_blocksize = S_ISBLK(inode->i_mode) ?
774 inode->i_bdev->bd_block_size : PAGE_SIZE;
781 size = get_loop_size(lo, file);
783 if ((loff_t)(sector_t)size != size) {
788 if (!(mode & FMODE_WRITE))
789 lo_flags |= LO_FLAGS_READ_ONLY;
791 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
793 lo->lo_blocksize = lo_blocksize;
794 lo->lo_device = bdev;
795 lo->lo_flags = lo_flags;
796 lo->lo_backing_file = file;
797 lo->transfer = transfer_none;
799 lo->lo_sizelimit = 0;
800 lo->old_gfp_mask = mapping_gfp_mask(mapping);
801 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
803 lo->lo_bio = lo->lo_biotail = NULL;
806 * set queue make_request_fn, and add limits based on lower level
809 blk_queue_make_request(lo->lo_queue, loop_make_request);
810 lo->lo_queue->queuedata = lo;
811 lo->lo_queue->unplug_fn = loop_unplug;
813 set_capacity(lo->lo_disk, size);
814 bd_set_size(bdev, size << 9);
816 set_blocksize(bdev, lo_blocksize);
818 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
820 if (IS_ERR(lo->lo_thread)) {
821 error = PTR_ERR(lo->lo_thread);
824 lo->lo_state = Lo_bound;
825 wake_up_process(lo->lo_thread);
827 ioctl_by_bdev(bdev, BLKRRPART, 0);
831 lo->lo_thread = NULL;
832 lo->lo_device = NULL;
833 lo->lo_backing_file = NULL;
835 set_capacity(lo->lo_disk, 0);
836 invalidate_bdev(bdev);
837 bd_set_size(bdev, 0);
838 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
839 lo->lo_state = Lo_unbound;
843 /* This is safe: open() is still holding a reference. */
844 module_put(THIS_MODULE);
849 loop_release_xfer(struct loop_device *lo)
852 struct loop_func_table *xfer = lo->lo_encryption;
856 err = xfer->release(lo);
858 lo->lo_encryption = NULL;
859 module_put(xfer->owner);
865 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
866 const struct loop_info64 *i)
871 struct module *owner = xfer->owner;
873 if (!try_module_get(owner))
876 err = xfer->init(lo, i);
880 lo->lo_encryption = xfer;
885 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
887 struct file *filp = lo->lo_backing_file;
888 gfp_t gfp = lo->old_gfp_mask;
890 if (lo->lo_state != Lo_bound)
893 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
899 spin_lock_irq(&lo->lo_lock);
900 lo->lo_state = Lo_rundown;
901 spin_unlock_irq(&lo->lo_lock);
903 kthread_stop(lo->lo_thread);
905 lo->lo_backing_file = NULL;
907 loop_release_xfer(lo);
910 lo->lo_device = NULL;
911 lo->lo_encryption = NULL;
913 lo->lo_sizelimit = 0;
914 lo->lo_encrypt_key_size = 0;
916 lo->lo_thread = NULL;
917 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
918 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
919 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
921 invalidate_bdev(bdev);
922 set_capacity(lo->lo_disk, 0);
924 bd_set_size(bdev, 0);
925 mapping_set_gfp_mask(filp->f_mapping, gfp);
926 lo->lo_state = Lo_unbound;
928 /* This is safe: open() is still holding a reference. */
929 module_put(THIS_MODULE);
931 ioctl_by_bdev(bdev, BLKRRPART, 0);
936 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
939 struct loop_func_table *xfer;
941 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
942 !capable(CAP_SYS_ADMIN))
944 if (lo->lo_state != Lo_bound)
946 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
949 err = loop_release_xfer(lo);
953 if (info->lo_encrypt_type) {
954 unsigned int type = info->lo_encrypt_type;
956 if (type >= MAX_LO_CRYPT)
958 xfer = xfer_funcs[type];
964 err = loop_init_xfer(lo, xfer, info);
968 if (lo->lo_offset != info->lo_offset ||
969 lo->lo_sizelimit != info->lo_sizelimit) {
970 lo->lo_offset = info->lo_offset;
971 lo->lo_sizelimit = info->lo_sizelimit;
972 if (figure_loop_size(lo))
976 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
977 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
978 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
979 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
983 lo->transfer = xfer->transfer;
984 lo->ioctl = xfer->ioctl;
986 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
987 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
988 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
990 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
991 lo->lo_init[0] = info->lo_init[0];
992 lo->lo_init[1] = info->lo_init[1];
993 if (info->lo_encrypt_key_size) {
994 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
995 info->lo_encrypt_key_size);
996 lo->lo_key_owner = current->uid;
1003 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1005 struct file *file = lo->lo_backing_file;
1009 if (lo->lo_state != Lo_bound)
1011 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1014 memset(info, 0, sizeof(*info));
1015 info->lo_number = lo->lo_number;
1016 info->lo_device = huge_encode_dev(stat.dev);
1017 info->lo_inode = stat.ino;
1018 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1019 info->lo_offset = lo->lo_offset;
1020 info->lo_sizelimit = lo->lo_sizelimit;
1021 info->lo_flags = lo->lo_flags;
1022 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1023 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1024 info->lo_encrypt_type =
1025 lo->lo_encryption ? lo->lo_encryption->number : 0;
1026 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1027 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1028 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1029 lo->lo_encrypt_key_size);
1035 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1037 memset(info64, 0, sizeof(*info64));
1038 info64->lo_number = info->lo_number;
1039 info64->lo_device = info->lo_device;
1040 info64->lo_inode = info->lo_inode;
1041 info64->lo_rdevice = info->lo_rdevice;
1042 info64->lo_offset = info->lo_offset;
1043 info64->lo_sizelimit = 0;
1044 info64->lo_encrypt_type = info->lo_encrypt_type;
1045 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1046 info64->lo_flags = info->lo_flags;
1047 info64->lo_init[0] = info->lo_init[0];
1048 info64->lo_init[1] = info->lo_init[1];
1049 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1050 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1052 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1053 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1057 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1059 memset(info, 0, sizeof(*info));
1060 info->lo_number = info64->lo_number;
1061 info->lo_device = info64->lo_device;
1062 info->lo_inode = info64->lo_inode;
1063 info->lo_rdevice = info64->lo_rdevice;
1064 info->lo_offset = info64->lo_offset;
1065 info->lo_encrypt_type = info64->lo_encrypt_type;
1066 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1067 info->lo_flags = info64->lo_flags;
1068 info->lo_init[0] = info64->lo_init[0];
1069 info->lo_init[1] = info64->lo_init[1];
1070 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1071 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1073 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1074 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1076 /* error in case values were truncated */
1077 if (info->lo_device != info64->lo_device ||
1078 info->lo_rdevice != info64->lo_rdevice ||
1079 info->lo_inode != info64->lo_inode ||
1080 info->lo_offset != info64->lo_offset)
1087 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1089 struct loop_info info;
1090 struct loop_info64 info64;
1092 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1094 loop_info64_from_old(&info, &info64);
1095 return loop_set_status(lo, &info64);
1099 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1101 struct loop_info64 info64;
1103 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1105 return loop_set_status(lo, &info64);
1109 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1110 struct loop_info info;
1111 struct loop_info64 info64;
1117 err = loop_get_status(lo, &info64);
1119 err = loop_info64_to_old(&info64, &info);
1120 if (!err && copy_to_user(arg, &info, sizeof(info)))
1127 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1128 struct loop_info64 info64;
1134 err = loop_get_status(lo, &info64);
1135 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1141 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1142 unsigned int cmd, unsigned long arg)
1144 struct loop_device *lo = bdev->bd_disk->private_data;
1147 mutex_lock(&lo->lo_ctl_mutex);
1150 err = loop_set_fd(lo, mode, bdev, arg);
1152 case LOOP_CHANGE_FD:
1153 err = loop_change_fd(lo, bdev, arg);
1156 err = loop_clr_fd(lo, bdev);
1158 case LOOP_SET_STATUS:
1159 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1161 case LOOP_GET_STATUS:
1162 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1164 case LOOP_SET_STATUS64:
1165 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1167 case LOOP_GET_STATUS64:
1168 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1171 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1173 mutex_unlock(&lo->lo_ctl_mutex);
1177 #ifdef CONFIG_COMPAT
1178 struct compat_loop_info {
1179 compat_int_t lo_number; /* ioctl r/o */
1180 compat_dev_t lo_device; /* ioctl r/o */
1181 compat_ulong_t lo_inode; /* ioctl r/o */
1182 compat_dev_t lo_rdevice; /* ioctl r/o */
1183 compat_int_t lo_offset;
1184 compat_int_t lo_encrypt_type;
1185 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1186 compat_int_t lo_flags; /* ioctl r/o */
1187 char lo_name[LO_NAME_SIZE];
1188 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1189 compat_ulong_t lo_init[2];
1194 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1195 * - noinlined to reduce stack space usage in main part of driver
1198 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1199 struct loop_info64 *info64)
1201 struct compat_loop_info info;
1203 if (copy_from_user(&info, arg, sizeof(info)))
1206 memset(info64, 0, sizeof(*info64));
1207 info64->lo_number = info.lo_number;
1208 info64->lo_device = info.lo_device;
1209 info64->lo_inode = info.lo_inode;
1210 info64->lo_rdevice = info.lo_rdevice;
1211 info64->lo_offset = info.lo_offset;
1212 info64->lo_sizelimit = 0;
1213 info64->lo_encrypt_type = info.lo_encrypt_type;
1214 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1215 info64->lo_flags = info.lo_flags;
1216 info64->lo_init[0] = info.lo_init[0];
1217 info64->lo_init[1] = info.lo_init[1];
1218 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1219 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1221 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1222 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1227 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1228 * - noinlined to reduce stack space usage in main part of driver
1231 loop_info64_to_compat(const struct loop_info64 *info64,
1232 struct compat_loop_info __user *arg)
1234 struct compat_loop_info info;
1236 memset(&info, 0, sizeof(info));
1237 info.lo_number = info64->lo_number;
1238 info.lo_device = info64->lo_device;
1239 info.lo_inode = info64->lo_inode;
1240 info.lo_rdevice = info64->lo_rdevice;
1241 info.lo_offset = info64->lo_offset;
1242 info.lo_encrypt_type = info64->lo_encrypt_type;
1243 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1244 info.lo_flags = info64->lo_flags;
1245 info.lo_init[0] = info64->lo_init[0];
1246 info.lo_init[1] = info64->lo_init[1];
1247 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1248 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1250 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1251 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1253 /* error in case values were truncated */
1254 if (info.lo_device != info64->lo_device ||
1255 info.lo_rdevice != info64->lo_rdevice ||
1256 info.lo_inode != info64->lo_inode ||
1257 info.lo_offset != info64->lo_offset ||
1258 info.lo_init[0] != info64->lo_init[0] ||
1259 info.lo_init[1] != info64->lo_init[1])
1262 if (copy_to_user(arg, &info, sizeof(info)))
1268 loop_set_status_compat(struct loop_device *lo,
1269 const struct compat_loop_info __user *arg)
1271 struct loop_info64 info64;
1274 ret = loop_info64_from_compat(arg, &info64);
1277 return loop_set_status(lo, &info64);
1281 loop_get_status_compat(struct loop_device *lo,
1282 struct compat_loop_info __user *arg)
1284 struct loop_info64 info64;
1290 err = loop_get_status(lo, &info64);
1292 err = loop_info64_to_compat(&info64, arg);
1296 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1297 unsigned int cmd, unsigned long arg)
1299 struct loop_device *lo = bdev->bd_disk->private_data;
1303 case LOOP_SET_STATUS:
1304 mutex_lock(&lo->lo_ctl_mutex);
1305 err = loop_set_status_compat(
1306 lo, (const struct compat_loop_info __user *) arg);
1307 mutex_unlock(&lo->lo_ctl_mutex);
1309 case LOOP_GET_STATUS:
1310 mutex_lock(&lo->lo_ctl_mutex);
1311 err = loop_get_status_compat(
1312 lo, (struct compat_loop_info __user *) arg);
1313 mutex_unlock(&lo->lo_ctl_mutex);
1316 case LOOP_GET_STATUS64:
1317 case LOOP_SET_STATUS64:
1318 arg = (unsigned long) compat_ptr(arg);
1320 case LOOP_CHANGE_FD:
1321 err = lo_ioctl(bdev, mode, cmd, arg);
1331 static int lo_open(struct block_device *bdev, fmode_t mode)
1333 struct loop_device *lo = bdev->bd_disk->private_data;
1335 mutex_lock(&lo->lo_ctl_mutex);
1337 mutex_unlock(&lo->lo_ctl_mutex);
1342 static int lo_release(struct gendisk *disk, fmode_t mode)
1344 struct loop_device *lo = disk->private_data;
1346 mutex_lock(&lo->lo_ctl_mutex);
1349 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) && !lo->lo_refcnt)
1350 loop_clr_fd(lo, NULL);
1352 mutex_unlock(&lo->lo_ctl_mutex);
1357 static struct block_device_operations lo_fops = {
1358 .owner = THIS_MODULE,
1360 .release = lo_release,
1362 #ifdef CONFIG_COMPAT
1363 .compat_ioctl = lo_compat_ioctl,
1368 * And now the modules code and kernel interface.
1370 static int max_loop;
1371 module_param(max_loop, int, 0);
1372 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1373 module_param(max_part, int, 0);
1374 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1375 MODULE_LICENSE("GPL");
1376 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1378 int loop_register_transfer(struct loop_func_table *funcs)
1380 unsigned int n = funcs->number;
1382 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1384 xfer_funcs[n] = funcs;
1388 int loop_unregister_transfer(int number)
1390 unsigned int n = number;
1391 struct loop_device *lo;
1392 struct loop_func_table *xfer;
1394 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1397 xfer_funcs[n] = NULL;
1399 list_for_each_entry(lo, &loop_devices, lo_list) {
1400 mutex_lock(&lo->lo_ctl_mutex);
1402 if (lo->lo_encryption == xfer)
1403 loop_release_xfer(lo);
1405 mutex_unlock(&lo->lo_ctl_mutex);
1411 EXPORT_SYMBOL(loop_register_transfer);
1412 EXPORT_SYMBOL(loop_unregister_transfer);
1414 static struct loop_device *loop_alloc(int i)
1416 struct loop_device *lo;
1417 struct gendisk *disk;
1419 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1423 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1427 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1429 goto out_free_queue;
1431 mutex_init(&lo->lo_ctl_mutex);
1433 lo->lo_thread = NULL;
1434 init_waitqueue_head(&lo->lo_event);
1435 spin_lock_init(&lo->lo_lock);
1436 disk->major = LOOP_MAJOR;
1437 disk->first_minor = i << part_shift;
1438 disk->fops = &lo_fops;
1439 disk->private_data = lo;
1440 disk->queue = lo->lo_queue;
1441 sprintf(disk->disk_name, "loop%d", i);
1445 blk_cleanup_queue(lo->lo_queue);
1452 static void loop_free(struct loop_device *lo)
1454 blk_cleanup_queue(lo->lo_queue);
1455 put_disk(lo->lo_disk);
1456 list_del(&lo->lo_list);
1460 static struct loop_device *loop_init_one(int i)
1462 struct loop_device *lo;
1464 list_for_each_entry(lo, &loop_devices, lo_list) {
1465 if (lo->lo_number == i)
1471 add_disk(lo->lo_disk);
1472 list_add_tail(&lo->lo_list, &loop_devices);
1477 static void loop_del_one(struct loop_device *lo)
1479 del_gendisk(lo->lo_disk);
1483 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1485 struct loop_device *lo;
1486 struct kobject *kobj;
1488 mutex_lock(&loop_devices_mutex);
1489 lo = loop_init_one(dev & MINORMASK);
1490 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1491 mutex_unlock(&loop_devices_mutex);
1497 static int __init loop_init(void)
1500 unsigned long range;
1501 struct loop_device *lo, *next;
1504 * loop module now has a feature to instantiate underlying device
1505 * structure on-demand, provided that there is an access dev node.
1506 * However, this will not work well with user space tool that doesn't
1507 * know about such "feature". In order to not break any existing
1508 * tool, we do the following:
1510 * (1) if max_loop is specified, create that many upfront, and this
1511 * also becomes a hard limit.
1512 * (2) if max_loop is not specified, create 8 loop device on module
1513 * load, user can further extend loop device by create dev node
1514 * themselves and have kernel automatically instantiate actual
1520 part_shift = fls(max_part);
1522 if (max_loop > 1UL << (MINORBITS - part_shift))
1530 range = 1UL << (MINORBITS - part_shift);
1533 if (register_blkdev(LOOP_MAJOR, "loop"))
1536 for (i = 0; i < nr; i++) {
1540 list_add_tail(&lo->lo_list, &loop_devices);
1543 /* point of no return */
1545 list_for_each_entry(lo, &loop_devices, lo_list)
1546 add_disk(lo->lo_disk);
1548 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1549 THIS_MODULE, loop_probe, NULL, NULL);
1551 printk(KERN_INFO "loop: module loaded\n");
1555 printk(KERN_INFO "loop: out of memory\n");
1557 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1560 unregister_blkdev(LOOP_MAJOR, "loop");
1564 static void __exit loop_exit(void)
1566 unsigned long range;
1567 struct loop_device *lo, *next;
1569 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1571 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1574 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1575 unregister_blkdev(LOOP_MAJOR, "loop");
1578 module_init(loop_init);
1579 module_exit(loop_exit);
1582 static int __init max_loop_setup(char *str)
1584 max_loop = simple_strtol(str, NULL, 0);
1588 __setup("max_loop=", max_loop_setup);