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);
88 static int transfer_none(struct loop_device *lo, int cmd,
89 struct page *raw_page, unsigned raw_off,
90 struct page *loop_page, unsigned loop_off,
91 int size, sector_t real_block)
93 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
94 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
97 memcpy(loop_buf, raw_buf, size);
99 memcpy(raw_buf, loop_buf, size);
101 kunmap_atomic(raw_buf, KM_USER0);
102 kunmap_atomic(loop_buf, KM_USER1);
107 static int transfer_xor(struct loop_device *lo, int cmd,
108 struct page *raw_page, unsigned raw_off,
109 struct page *loop_page, unsigned loop_off,
110 int size, sector_t real_block)
112 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
113 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
114 char *in, *out, *key;
125 key = lo->lo_encrypt_key;
126 keysize = lo->lo_encrypt_key_size;
127 for (i = 0; i < size; i++)
128 *out++ = *in++ ^ key[(i & 511) % keysize];
130 kunmap_atomic(raw_buf, KM_USER0);
131 kunmap_atomic(loop_buf, KM_USER1);
136 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
138 if (unlikely(info->lo_encrypt_key_size <= 0))
143 static struct loop_func_table none_funcs = {
144 .number = LO_CRYPT_NONE,
145 .transfer = transfer_none,
148 static struct loop_func_table xor_funcs = {
149 .number = LO_CRYPT_XOR,
150 .transfer = transfer_xor,
154 /* xfer_funcs[0] is special - its release function is never called */
155 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
160 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
162 loff_t size, offset, loopsize;
164 /* Compute loopsize in bytes */
165 size = i_size_read(file->f_mapping->host);
166 offset = lo->lo_offset;
167 loopsize = size - offset;
168 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
169 loopsize = lo->lo_sizelimit;
172 * Unfortunately, if we want to do I/O on the device,
173 * the number of 512-byte sectors has to fit into a sector_t.
175 return loopsize >> 9;
179 figure_loop_size(struct loop_device *lo)
181 loff_t size = get_loop_size(lo, lo->lo_backing_file);
182 sector_t x = (sector_t)size;
184 if (unlikely((loff_t)x != size))
187 set_capacity(lo->lo_disk, x);
192 lo_do_transfer(struct loop_device *lo, int cmd,
193 struct page *rpage, unsigned roffs,
194 struct page *lpage, unsigned loffs,
195 int size, sector_t rblock)
197 if (unlikely(!lo->transfer))
200 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
204 * This is best effort. We really wouldn't know what to do with a returned
205 * error. This code is taken from the implementation of fsync.
207 static int sync_file(struct file * file, int full_sync)
209 struct address_space *mapping;
212 if (!file->f_op || !file->f_op->fsync)
215 mapping = file->f_mapping;
217 ret = filemap_fdatawrite(mapping);
221 * We need to protect against concurrent writers,
222 * which could cause livelocks in fsync_buffers_list
225 ret = file->f_op->fsync(file, file->f_dentry, 1);
227 ret2 = filemap_fdatawait(mapping);
236 * do_lo_send_aops - helper for writing data to a loop device
238 * This is the fast version for backing filesystems which implement the address
239 * space operations write_begin and write_end.
241 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
242 int bsize, loff_t pos, struct page *unused)
244 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
245 struct address_space *mapping = file->f_mapping;
246 struct inode *inode = file->f_dentry->d_inode;
248 unsigned offset, bv_offs;
250 unsigned long old_blocks;
252 mutex_lock(&mapping->host->i_mutex);
254 spin_lock(&inode->i_lock);
255 old_blocks = inode->i_blocks;
256 spin_unlock(&inode->i_lock);
258 index = pos >> PAGE_CACHE_SHIFT;
259 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
260 bv_offs = bvec->bv_offset;
264 unsigned size, copied;
269 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
270 size = PAGE_CACHE_SIZE - offset;
274 ret = pagecache_write_begin(file, mapping, pos, size, 0,
279 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
280 bvec->bv_page, bv_offs, size, IV);
282 if (unlikely(transfer_result))
285 ret = pagecache_write_end(file, mapping, pos, size, copied,
287 if (ret < 0 || ret != copied)
290 if (unlikely(transfer_result))
301 if (file->f_flags & O_SYNC) {
303 spin_lock(&inode->i_lock);
304 if (inode->i_blocks > old_blocks)
306 spin_unlock(&inode->i_lock);
307 ret = sync_file(file, full_sync);
310 mutex_unlock(&mapping->host->i_mutex);
318 * __do_lo_send_write - helper for writing data to a loop device
320 * This helper just factors out common code between do_lo_send_direct_write()
321 * and do_lo_send_write().
323 static int __do_lo_send_write(struct file *file,
324 u8 *buf, const int len, loff_t pos)
327 mm_segment_t old_fs = get_fs();
330 bw = file->f_op->write(file, buf, len, &pos);
332 if (likely(bw == len))
334 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
335 (unsigned long long)pos, len);
342 * do_lo_send_direct_write - helper for writing data to a loop device
344 * This is the fast, non-transforming version for backing filesystems which do
345 * not implement the address space operations write_begin and write_end.
346 * It uses the write file operation which should be present on all writeable
349 static int do_lo_send_direct_write(struct loop_device *lo,
350 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
352 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
353 kmap(bvec->bv_page) + bvec->bv_offset,
355 kunmap(bvec->bv_page);
361 * do_lo_send_write - helper for writing data to a loop device
363 * This is the slow, transforming version for filesystems which do not
364 * implement the address space operations write_begin and write_end. It
365 * uses the write file operation which should be present on all writeable
368 * Using fops->write is slower than using aops->{prepare,commit}_write in the
369 * transforming case because we need to double buffer the data as we cannot do
370 * the transformations in place as we do not have direct access to the
371 * destination pages of the backing file.
373 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
374 int bsize, loff_t pos, struct page *page)
376 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
377 bvec->bv_offset, bvec->bv_len, pos >> 9);
379 return __do_lo_send_write(lo->lo_backing_file,
380 page_address(page), bvec->bv_len,
382 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
383 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
389 static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
392 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
394 struct bio_vec *bvec;
395 struct page *page = NULL;
398 do_lo_send = do_lo_send_aops;
399 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
400 do_lo_send = do_lo_send_direct_write;
401 if (lo->transfer != transfer_none) {
402 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
406 do_lo_send = do_lo_send_write;
409 bio_for_each_segment(bvec, bio, i) {
410 ret = do_lo_send(lo, bvec, bsize, pos, page);
422 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
427 struct lo_read_data {
428 struct loop_device *lo;
435 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
436 struct splice_desc *sd)
438 struct lo_read_data *p = sd->u.data;
439 struct loop_device *lo = p->lo;
440 struct page *page = buf->page;
445 ret = buf->ops->confirm(pipe, buf);
449 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
455 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
456 printk(KERN_ERR "loop: transfer error block %ld\n",
461 flush_dcache_page(p->page);
470 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
472 return __splice_from_pipe(pipe, sd, lo_splice_actor);
476 do_lo_receive(struct loop_device *lo,
477 struct bio_vec *bvec, int bsize, loff_t pos)
479 struct lo_read_data cookie;
480 struct splice_desc sd;
485 cookie.page = bvec->bv_page;
486 cookie.offset = bvec->bv_offset;
487 cookie.bsize = bsize;
490 sd.total_len = bvec->bv_len;
495 file = lo->lo_backing_file;
496 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
505 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
507 struct bio_vec *bvec;
510 bio_for_each_segment(bvec, bio, i) {
511 ret = do_lo_receive(lo, bvec, bsize, pos);
519 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
523 int sync = bio_sync(bio);
524 int barrier = bio_barrier(bio);
527 ret = sync_file(lo->lo_backing_file, 1);
532 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
533 if (bio_rw(bio) == WRITE)
534 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
536 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
538 if ((barrier || sync) && !ret)
539 ret = sync_file(lo->lo_backing_file, 1);
545 * Add bio to back of pending list
547 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
549 if (lo->lo_biotail) {
550 lo->lo_biotail->bi_next = bio;
551 lo->lo_biotail = bio;
553 lo->lo_bio = lo->lo_biotail = bio;
557 * Grab first pending buffer
559 static struct bio *loop_get_bio(struct loop_device *lo)
563 if ((bio = lo->lo_bio)) {
564 if (bio == lo->lo_biotail)
565 lo->lo_biotail = NULL;
566 lo->lo_bio = bio->bi_next;
573 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
575 struct loop_device *lo = q->queuedata;
576 int rw = bio_rw(old_bio);
581 BUG_ON(!lo || (rw != READ && rw != WRITE));
583 spin_lock_irq(&lo->lo_lock);
584 if (lo->lo_state != Lo_bound)
586 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
588 loop_add_bio(lo, old_bio);
589 wake_up(&lo->lo_event);
590 spin_unlock_irq(&lo->lo_lock);
594 spin_unlock_irq(&lo->lo_lock);
595 bio_io_error(old_bio);
600 * kick off io on the underlying address space
602 static void loop_unplug(struct request_queue *q)
604 struct loop_device *lo = q->queuedata;
606 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
607 blk_run_address_space(lo->lo_backing_file->f_mapping);
610 struct switch_request {
612 struct completion wait;
615 static void do_loop_switch(struct loop_device *, struct switch_request *);
617 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
619 if (unlikely(!bio->bi_bdev)) {
620 do_loop_switch(lo, bio->bi_private);
623 int ret = do_bio_filebacked(lo, bio);
629 * worker thread that handles reads/writes to file backed loop devices,
630 * to avoid blocking in our make_request_fn. it also does loop decrypting
631 * on reads for block backed loop, as that is too heavy to do from
632 * b_end_io context where irqs may be disabled.
634 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
635 * calling kthread_stop(). Therefore once kthread_should_stop() is
636 * true, make_request will not place any more requests. Therefore
637 * once kthread_should_stop() is true and lo_bio is NULL, we are
638 * done with the loop.
640 static int loop_thread(void *data)
642 struct loop_device *lo = data;
645 set_user_nice(current, -20);
647 while (!kthread_should_stop() || lo->lo_bio) {
649 wait_event_interruptible(lo->lo_event,
650 lo->lo_bio || kthread_should_stop());
654 spin_lock_irq(&lo->lo_lock);
655 bio = loop_get_bio(lo);
656 spin_unlock_irq(&lo->lo_lock);
659 loop_handle_bio(lo, bio);
666 * loop_switch performs the hard work of switching a backing store.
667 * First it needs to flush existing IO, it does this by sending a magic
668 * BIO down the pipe. The completion of this BIO does the actual switch.
670 static int loop_switch(struct loop_device *lo, struct file *file)
672 struct switch_request w;
673 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
676 init_completion(&w.wait);
678 bio->bi_private = &w;
680 loop_make_request(lo->lo_queue, bio);
681 wait_for_completion(&w.wait);
686 * Do the actual switch; called from the BIO completion routine
688 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
690 struct file *file = p->file;
691 struct file *old_file = lo->lo_backing_file;
692 struct address_space *mapping = file->f_mapping;
694 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
695 lo->lo_backing_file = file;
696 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
697 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
698 lo->old_gfp_mask = mapping_gfp_mask(mapping);
699 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
705 * loop_change_fd switched the backing store of a loopback device to
706 * a new file. This is useful for operating system installers to free up
707 * the original file and in High Availability environments to switch to
708 * an alternative location for the content in case of server meltdown.
709 * This can only work if the loop device is used read-only, and if the
710 * new backing store is the same size and type as the old backing store.
712 static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
713 struct block_device *bdev, unsigned int arg)
715 struct file *file, *old_file;
720 if (lo->lo_state != Lo_bound)
723 /* the loop device has to be read-only */
725 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
733 inode = file->f_mapping->host;
734 old_file = lo->lo_backing_file;
738 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
741 /* new backing store needs to support loop (eg splice_read) */
742 if (!inode->i_fop->splice_read)
745 /* size of the new backing store needs to be the same */
746 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
750 error = loop_switch(lo, file);
763 static inline int is_loop_device(struct file *file)
765 struct inode *i = file->f_mapping->host;
767 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
770 static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
771 struct block_device *bdev, unsigned int arg)
773 struct file *file, *f;
775 struct address_space *mapping;
776 unsigned lo_blocksize;
781 /* This is safe, since we have a reference from open(). */
782 __module_get(THIS_MODULE);
790 if (lo->lo_state != Lo_unbound)
793 /* Avoid recursion */
795 while (is_loop_device(f)) {
796 struct loop_device *l;
798 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
801 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
802 if (l->lo_state == Lo_unbound) {
806 f = l->lo_backing_file;
809 mapping = file->f_mapping;
810 inode = mapping->host;
812 if (!(file->f_mode & FMODE_WRITE))
813 lo_flags |= LO_FLAGS_READ_ONLY;
815 if ((file->f_flags & O_SYNC) && (!file->f_op || !file->f_op->fsync))
819 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
820 const struct address_space_operations *aops = mapping->a_ops;
822 * If we can't read - sorry. If we only can't write - well,
823 * it's going to be read-only.
825 if (!file->f_op->splice_read)
827 if (aops->prepare_write || aops->write_begin)
828 lo_flags |= LO_FLAGS_USE_AOPS;
829 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
830 lo_flags |= LO_FLAGS_READ_ONLY;
832 lo_blocksize = S_ISBLK(inode->i_mode) ?
833 inode->i_bdev->bd_block_size : PAGE_SIZE;
840 size = get_loop_size(lo, file);
842 if ((loff_t)(sector_t)size != size) {
847 if (!(lo_file->f_mode & FMODE_WRITE))
848 lo_flags |= LO_FLAGS_READ_ONLY;
850 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
852 lo->lo_blocksize = lo_blocksize;
853 lo->lo_device = bdev;
854 lo->lo_flags = lo_flags;
855 lo->lo_backing_file = file;
856 lo->transfer = transfer_none;
858 lo->lo_sizelimit = 0;
859 lo->old_gfp_mask = mapping_gfp_mask(mapping);
860 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
862 lo->lo_bio = lo->lo_biotail = NULL;
865 * set queue make_request_fn, and add limits based on lower level
868 blk_queue_make_request(lo->lo_queue, loop_make_request);
869 lo->lo_queue->queuedata = lo;
870 lo->lo_queue->unplug_fn = loop_unplug;
872 set_capacity(lo->lo_disk, size);
873 bd_set_size(bdev, size << 9);
875 set_blocksize(bdev, lo_blocksize);
877 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
879 if (IS_ERR(lo->lo_thread)) {
880 error = PTR_ERR(lo->lo_thread);
883 lo->lo_state = Lo_bound;
884 wake_up_process(lo->lo_thread);
888 lo->lo_thread = NULL;
889 lo->lo_device = NULL;
890 lo->lo_backing_file = NULL;
892 set_capacity(lo->lo_disk, 0);
893 invalidate_bdev(bdev);
894 bd_set_size(bdev, 0);
895 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
896 lo->lo_state = Lo_unbound;
900 /* This is safe: open() is still holding a reference. */
901 module_put(THIS_MODULE);
906 loop_release_xfer(struct loop_device *lo)
909 struct loop_func_table *xfer = lo->lo_encryption;
913 err = xfer->release(lo);
915 lo->lo_encryption = NULL;
916 module_put(xfer->owner);
922 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
923 const struct loop_info64 *i)
928 struct module *owner = xfer->owner;
930 if (!try_module_get(owner))
933 err = xfer->init(lo, i);
937 lo->lo_encryption = xfer;
942 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
944 struct file *filp = lo->lo_backing_file;
945 gfp_t gfp = lo->old_gfp_mask;
947 if (lo->lo_state != Lo_bound)
950 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
956 spin_lock_irq(&lo->lo_lock);
957 lo->lo_state = Lo_rundown;
958 spin_unlock_irq(&lo->lo_lock);
960 kthread_stop(lo->lo_thread);
962 lo->lo_backing_file = NULL;
964 loop_release_xfer(lo);
967 lo->lo_device = NULL;
968 lo->lo_encryption = NULL;
970 lo->lo_sizelimit = 0;
971 lo->lo_encrypt_key_size = 0;
973 lo->lo_thread = NULL;
974 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
975 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
976 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
977 invalidate_bdev(bdev);
978 set_capacity(lo->lo_disk, 0);
979 bd_set_size(bdev, 0);
980 mapping_set_gfp_mask(filp->f_mapping, gfp);
981 lo->lo_state = Lo_unbound;
983 /* This is safe: open() is still holding a reference. */
984 module_put(THIS_MODULE);
989 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
992 struct loop_func_table *xfer;
994 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
995 !capable(CAP_SYS_ADMIN))
997 if (lo->lo_state != Lo_bound)
999 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1002 err = loop_release_xfer(lo);
1006 if (info->lo_encrypt_type) {
1007 unsigned int type = info->lo_encrypt_type;
1009 if (type >= MAX_LO_CRYPT)
1011 xfer = xfer_funcs[type];
1017 err = loop_init_xfer(lo, xfer, info);
1021 if (lo->lo_offset != info->lo_offset ||
1022 lo->lo_sizelimit != info->lo_sizelimit) {
1023 lo->lo_offset = info->lo_offset;
1024 lo->lo_sizelimit = info->lo_sizelimit;
1025 if (figure_loop_size(lo))
1029 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1030 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1031 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1032 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1036 lo->transfer = xfer->transfer;
1037 lo->ioctl = xfer->ioctl;
1039 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1040 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1041 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1043 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1044 lo->lo_init[0] = info->lo_init[0];
1045 lo->lo_init[1] = info->lo_init[1];
1046 if (info->lo_encrypt_key_size) {
1047 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1048 info->lo_encrypt_key_size);
1049 lo->lo_key_owner = current->uid;
1056 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1058 struct file *file = lo->lo_backing_file;
1062 if (lo->lo_state != Lo_bound)
1064 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1067 memset(info, 0, sizeof(*info));
1068 info->lo_number = lo->lo_number;
1069 info->lo_device = huge_encode_dev(stat.dev);
1070 info->lo_inode = stat.ino;
1071 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1072 info->lo_offset = lo->lo_offset;
1073 info->lo_sizelimit = lo->lo_sizelimit;
1074 info->lo_flags = lo->lo_flags;
1075 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1076 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1077 info->lo_encrypt_type =
1078 lo->lo_encryption ? lo->lo_encryption->number : 0;
1079 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1080 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1081 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1082 lo->lo_encrypt_key_size);
1088 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1090 memset(info64, 0, sizeof(*info64));
1091 info64->lo_number = info->lo_number;
1092 info64->lo_device = info->lo_device;
1093 info64->lo_inode = info->lo_inode;
1094 info64->lo_rdevice = info->lo_rdevice;
1095 info64->lo_offset = info->lo_offset;
1096 info64->lo_sizelimit = 0;
1097 info64->lo_encrypt_type = info->lo_encrypt_type;
1098 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1099 info64->lo_flags = info->lo_flags;
1100 info64->lo_init[0] = info->lo_init[0];
1101 info64->lo_init[1] = info->lo_init[1];
1102 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1103 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1105 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1106 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1110 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1112 memset(info, 0, sizeof(*info));
1113 info->lo_number = info64->lo_number;
1114 info->lo_device = info64->lo_device;
1115 info->lo_inode = info64->lo_inode;
1116 info->lo_rdevice = info64->lo_rdevice;
1117 info->lo_offset = info64->lo_offset;
1118 info->lo_encrypt_type = info64->lo_encrypt_type;
1119 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1120 info->lo_flags = info64->lo_flags;
1121 info->lo_init[0] = info64->lo_init[0];
1122 info->lo_init[1] = info64->lo_init[1];
1123 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1124 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1126 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1127 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1129 /* error in case values were truncated */
1130 if (info->lo_device != info64->lo_device ||
1131 info->lo_rdevice != info64->lo_rdevice ||
1132 info->lo_inode != info64->lo_inode ||
1133 info->lo_offset != info64->lo_offset)
1140 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1142 struct loop_info info;
1143 struct loop_info64 info64;
1145 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1147 loop_info64_from_old(&info, &info64);
1148 return loop_set_status(lo, &info64);
1152 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1154 struct loop_info64 info64;
1156 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1158 return loop_set_status(lo, &info64);
1162 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1163 struct loop_info info;
1164 struct loop_info64 info64;
1170 err = loop_get_status(lo, &info64);
1172 err = loop_info64_to_old(&info64, &info);
1173 if (!err && copy_to_user(arg, &info, sizeof(info)))
1180 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1181 struct loop_info64 info64;
1187 err = loop_get_status(lo, &info64);
1188 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1194 static int lo_ioctl(struct inode * inode, struct file * file,
1195 unsigned int cmd, unsigned long arg)
1197 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1200 mutex_lock(&lo->lo_ctl_mutex);
1203 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1205 case LOOP_CHANGE_FD:
1206 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1209 err = loop_clr_fd(lo, inode->i_bdev);
1211 case LOOP_SET_STATUS:
1212 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1214 case LOOP_GET_STATUS:
1215 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1217 case LOOP_SET_STATUS64:
1218 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1220 case LOOP_GET_STATUS64:
1221 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1224 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1226 mutex_unlock(&lo->lo_ctl_mutex);
1230 #ifdef CONFIG_COMPAT
1231 struct compat_loop_info {
1232 compat_int_t lo_number; /* ioctl r/o */
1233 compat_dev_t lo_device; /* ioctl r/o */
1234 compat_ulong_t lo_inode; /* ioctl r/o */
1235 compat_dev_t lo_rdevice; /* ioctl r/o */
1236 compat_int_t lo_offset;
1237 compat_int_t lo_encrypt_type;
1238 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1239 compat_int_t lo_flags; /* ioctl r/o */
1240 char lo_name[LO_NAME_SIZE];
1241 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1242 compat_ulong_t lo_init[2];
1247 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1248 * - noinlined to reduce stack space usage in main part of driver
1251 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1252 struct loop_info64 *info64)
1254 struct compat_loop_info info;
1256 if (copy_from_user(&info, arg, sizeof(info)))
1259 memset(info64, 0, sizeof(*info64));
1260 info64->lo_number = info.lo_number;
1261 info64->lo_device = info.lo_device;
1262 info64->lo_inode = info.lo_inode;
1263 info64->lo_rdevice = info.lo_rdevice;
1264 info64->lo_offset = info.lo_offset;
1265 info64->lo_sizelimit = 0;
1266 info64->lo_encrypt_type = info.lo_encrypt_type;
1267 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1268 info64->lo_flags = info.lo_flags;
1269 info64->lo_init[0] = info.lo_init[0];
1270 info64->lo_init[1] = info.lo_init[1];
1271 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1272 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1274 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1275 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1280 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1281 * - noinlined to reduce stack space usage in main part of driver
1284 loop_info64_to_compat(const struct loop_info64 *info64,
1285 struct compat_loop_info __user *arg)
1287 struct compat_loop_info info;
1289 memset(&info, 0, sizeof(info));
1290 info.lo_number = info64->lo_number;
1291 info.lo_device = info64->lo_device;
1292 info.lo_inode = info64->lo_inode;
1293 info.lo_rdevice = info64->lo_rdevice;
1294 info.lo_offset = info64->lo_offset;
1295 info.lo_encrypt_type = info64->lo_encrypt_type;
1296 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1297 info.lo_flags = info64->lo_flags;
1298 info.lo_init[0] = info64->lo_init[0];
1299 info.lo_init[1] = info64->lo_init[1];
1300 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1301 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1303 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1304 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1306 /* error in case values were truncated */
1307 if (info.lo_device != info64->lo_device ||
1308 info.lo_rdevice != info64->lo_rdevice ||
1309 info.lo_inode != info64->lo_inode ||
1310 info.lo_offset != info64->lo_offset ||
1311 info.lo_init[0] != info64->lo_init[0] ||
1312 info.lo_init[1] != info64->lo_init[1])
1315 if (copy_to_user(arg, &info, sizeof(info)))
1321 loop_set_status_compat(struct loop_device *lo,
1322 const struct compat_loop_info __user *arg)
1324 struct loop_info64 info64;
1327 ret = loop_info64_from_compat(arg, &info64);
1330 return loop_set_status(lo, &info64);
1334 loop_get_status_compat(struct loop_device *lo,
1335 struct compat_loop_info __user *arg)
1337 struct loop_info64 info64;
1343 err = loop_get_status(lo, &info64);
1345 err = loop_info64_to_compat(&info64, arg);
1349 static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1351 struct inode *inode = file->f_path.dentry->d_inode;
1352 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1356 case LOOP_SET_STATUS:
1357 mutex_lock(&lo->lo_ctl_mutex);
1358 err = loop_set_status_compat(
1359 lo, (const struct compat_loop_info __user *) arg);
1360 mutex_unlock(&lo->lo_ctl_mutex);
1362 case LOOP_GET_STATUS:
1363 mutex_lock(&lo->lo_ctl_mutex);
1364 err = loop_get_status_compat(
1365 lo, (struct compat_loop_info __user *) arg);
1366 mutex_unlock(&lo->lo_ctl_mutex);
1369 case LOOP_GET_STATUS64:
1370 case LOOP_SET_STATUS64:
1371 arg = (unsigned long) compat_ptr(arg);
1373 case LOOP_CHANGE_FD:
1374 err = lo_ioctl(inode, file, cmd, arg);
1384 static int lo_open(struct inode *inode, struct file *file)
1386 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1388 mutex_lock(&lo->lo_ctl_mutex);
1390 mutex_unlock(&lo->lo_ctl_mutex);
1395 static int lo_release(struct inode *inode, struct file *file)
1397 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1399 mutex_lock(&lo->lo_ctl_mutex);
1402 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) && !lo->lo_refcnt)
1403 loop_clr_fd(lo, inode->i_bdev);
1405 mutex_unlock(&lo->lo_ctl_mutex);
1410 static struct block_device_operations lo_fops = {
1411 .owner = THIS_MODULE,
1413 .release = lo_release,
1415 #ifdef CONFIG_COMPAT
1416 .compat_ioctl = lo_compat_ioctl,
1421 * And now the modules code and kernel interface.
1423 static int max_loop;
1424 module_param(max_loop, int, 0);
1425 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1426 MODULE_LICENSE("GPL");
1427 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1429 int loop_register_transfer(struct loop_func_table *funcs)
1431 unsigned int n = funcs->number;
1433 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1435 xfer_funcs[n] = funcs;
1439 int loop_unregister_transfer(int number)
1441 unsigned int n = number;
1442 struct loop_device *lo;
1443 struct loop_func_table *xfer;
1445 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1448 xfer_funcs[n] = NULL;
1450 list_for_each_entry(lo, &loop_devices, lo_list) {
1451 mutex_lock(&lo->lo_ctl_mutex);
1453 if (lo->lo_encryption == xfer)
1454 loop_release_xfer(lo);
1456 mutex_unlock(&lo->lo_ctl_mutex);
1462 EXPORT_SYMBOL(loop_register_transfer);
1463 EXPORT_SYMBOL(loop_unregister_transfer);
1465 static struct loop_device *loop_alloc(int i)
1467 struct loop_device *lo;
1468 struct gendisk *disk;
1470 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1474 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1478 disk = lo->lo_disk = alloc_disk(1);
1480 goto out_free_queue;
1482 mutex_init(&lo->lo_ctl_mutex);
1484 lo->lo_thread = NULL;
1485 init_waitqueue_head(&lo->lo_event);
1486 spin_lock_init(&lo->lo_lock);
1487 disk->major = LOOP_MAJOR;
1488 disk->first_minor = i;
1489 disk->fops = &lo_fops;
1490 disk->private_data = lo;
1491 disk->queue = lo->lo_queue;
1492 sprintf(disk->disk_name, "loop%d", i);
1496 blk_cleanup_queue(lo->lo_queue);
1503 static void loop_free(struct loop_device *lo)
1505 blk_cleanup_queue(lo->lo_queue);
1506 put_disk(lo->lo_disk);
1507 list_del(&lo->lo_list);
1511 static struct loop_device *loop_init_one(int i)
1513 struct loop_device *lo;
1515 list_for_each_entry(lo, &loop_devices, lo_list) {
1516 if (lo->lo_number == i)
1522 add_disk(lo->lo_disk);
1523 list_add_tail(&lo->lo_list, &loop_devices);
1528 static void loop_del_one(struct loop_device *lo)
1530 del_gendisk(lo->lo_disk);
1534 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1536 struct loop_device *lo;
1537 struct kobject *kobj;
1539 mutex_lock(&loop_devices_mutex);
1540 lo = loop_init_one(dev & MINORMASK);
1541 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1542 mutex_unlock(&loop_devices_mutex);
1548 static int __init loop_init(void)
1551 unsigned long range;
1552 struct loop_device *lo, *next;
1555 * loop module now has a feature to instantiate underlying device
1556 * structure on-demand, provided that there is an access dev node.
1557 * However, this will not work well with user space tool that doesn't
1558 * know about such "feature". In order to not break any existing
1559 * tool, we do the following:
1561 * (1) if max_loop is specified, create that many upfront, and this
1562 * also becomes a hard limit.
1563 * (2) if max_loop is not specified, create 8 loop device on module
1564 * load, user can further extend loop device by create dev node
1565 * themselves and have kernel automatically instantiate actual
1568 if (max_loop > 1UL << MINORBITS)
1576 range = 1UL << MINORBITS;
1579 if (register_blkdev(LOOP_MAJOR, "loop"))
1582 for (i = 0; i < nr; i++) {
1586 list_add_tail(&lo->lo_list, &loop_devices);
1589 /* point of no return */
1591 list_for_each_entry(lo, &loop_devices, lo_list)
1592 add_disk(lo->lo_disk);
1594 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1595 THIS_MODULE, loop_probe, NULL, NULL);
1597 printk(KERN_INFO "loop: module loaded\n");
1601 printk(KERN_INFO "loop: out of memory\n");
1603 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1606 unregister_blkdev(LOOP_MAJOR, "loop");
1610 static void __exit loop_exit(void)
1612 unsigned long range;
1613 struct loop_device *lo, *next;
1615 range = max_loop ? max_loop : 1UL << MINORBITS;
1617 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1620 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1621 unregister_blkdev(LOOP_MAJOR, "loop");
1624 module_init(loop_init);
1625 module_exit(loop_exit);
1628 static int __init max_loop_setup(char *str)
1630 max_loop = simple_strtol(str, NULL, 0);
1634 __setup("max_loop=", max_loop_setup);