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;
397 ret = buf->ops->confirm(pipe, buf);
401 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
407 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
408 printk(KERN_ERR "loop: transfer error block %ld\n",
413 flush_dcache_page(p->page);
422 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
424 return __splice_from_pipe(pipe, sd, lo_splice_actor);
428 do_lo_receive(struct loop_device *lo,
429 struct bio_vec *bvec, int bsize, loff_t pos)
431 struct lo_read_data cookie;
432 struct splice_desc sd;
437 cookie.page = bvec->bv_page;
438 cookie.offset = bvec->bv_offset;
439 cookie.bsize = bsize;
442 sd.total_len = bvec->bv_len;
447 file = lo->lo_backing_file;
448 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
457 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
459 struct bio_vec *bvec;
462 bio_for_each_segment(bvec, bio, i) {
463 ret = do_lo_receive(lo, bvec, bsize, pos);
471 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
476 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
478 if (bio_rw(bio) == WRITE) {
479 int barrier = bio_barrier(bio);
480 struct file *file = lo->lo_backing_file;
483 if (unlikely(!file->f_op->fsync)) {
488 ret = vfs_fsync(file, file->f_path.dentry, 0);
495 ret = lo_send(lo, bio, pos);
497 if (barrier && !ret) {
498 ret = vfs_fsync(file, file->f_path.dentry, 0);
503 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
510 * Add bio to back of pending list
512 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
514 if (lo->lo_biotail) {
515 lo->lo_biotail->bi_next = bio;
516 lo->lo_biotail = bio;
518 lo->lo_bio = lo->lo_biotail = bio;
522 * Grab first pending buffer
524 static struct bio *loop_get_bio(struct loop_device *lo)
528 if ((bio = lo->lo_bio)) {
529 if (bio == lo->lo_biotail)
530 lo->lo_biotail = NULL;
531 lo->lo_bio = bio->bi_next;
538 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
540 struct loop_device *lo = q->queuedata;
541 int rw = bio_rw(old_bio);
546 BUG_ON(!lo || (rw != READ && rw != WRITE));
548 spin_lock_irq(&lo->lo_lock);
549 if (lo->lo_state != Lo_bound)
551 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
553 loop_add_bio(lo, old_bio);
554 wake_up(&lo->lo_event);
555 spin_unlock_irq(&lo->lo_lock);
559 spin_unlock_irq(&lo->lo_lock);
560 bio_io_error(old_bio);
565 * kick off io on the underlying address space
567 static void loop_unplug(struct request_queue *q)
569 struct loop_device *lo = q->queuedata;
571 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
572 blk_run_address_space(lo->lo_backing_file->f_mapping);
575 struct switch_request {
577 struct completion wait;
580 static void do_loop_switch(struct loop_device *, struct switch_request *);
582 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
584 if (unlikely(!bio->bi_bdev)) {
585 do_loop_switch(lo, bio->bi_private);
588 int ret = do_bio_filebacked(lo, bio);
594 * worker thread that handles reads/writes to file backed loop devices,
595 * to avoid blocking in our make_request_fn. it also does loop decrypting
596 * on reads for block backed loop, as that is too heavy to do from
597 * b_end_io context where irqs may be disabled.
599 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
600 * calling kthread_stop(). Therefore once kthread_should_stop() is
601 * true, make_request will not place any more requests. Therefore
602 * once kthread_should_stop() is true and lo_bio is NULL, we are
603 * done with the loop.
605 static int loop_thread(void *data)
607 struct loop_device *lo = data;
610 set_user_nice(current, -20);
612 while (!kthread_should_stop() || lo->lo_bio) {
614 wait_event_interruptible(lo->lo_event,
615 lo->lo_bio || kthread_should_stop());
619 spin_lock_irq(&lo->lo_lock);
620 bio = loop_get_bio(lo);
621 spin_unlock_irq(&lo->lo_lock);
624 loop_handle_bio(lo, bio);
631 * loop_switch performs the hard work of switching a backing store.
632 * First it needs to flush existing IO, it does this by sending a magic
633 * BIO down the pipe. The completion of this BIO does the actual switch.
635 static int loop_switch(struct loop_device *lo, struct file *file)
637 struct switch_request w;
638 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
641 init_completion(&w.wait);
643 bio->bi_private = &w;
645 loop_make_request(lo->lo_queue, bio);
646 wait_for_completion(&w.wait);
651 * Helper to flush the IOs in loop, but keeping loop thread running
653 static int loop_flush(struct loop_device *lo)
655 /* loop not yet configured, no running thread, nothing to flush */
659 return loop_switch(lo, NULL);
663 * Do the actual switch; called from the BIO completion routine
665 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
667 struct file *file = p->file;
668 struct file *old_file = lo->lo_backing_file;
669 struct address_space *mapping;
671 /* if no new file, only flush of queued bios requested */
675 mapping = file->f_mapping;
676 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
677 lo->lo_backing_file = file;
678 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
679 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
680 lo->old_gfp_mask = mapping_gfp_mask(mapping);
681 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
688 * loop_change_fd switched the backing store of a loopback device to
689 * a new file. This is useful for operating system installers to free up
690 * the original file and in High Availability environments to switch to
691 * an alternative location for the content in case of server meltdown.
692 * This can only work if the loop device is used read-only, and if the
693 * new backing store is the same size and type as the old backing store.
695 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
698 struct file *file, *old_file;
703 if (lo->lo_state != Lo_bound)
706 /* the loop device has to be read-only */
708 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
716 inode = file->f_mapping->host;
717 old_file = lo->lo_backing_file;
721 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
724 /* new backing store needs to support loop (eg splice_read) */
725 if (!inode->i_fop->splice_read)
728 /* size of the new backing store needs to be the same */
729 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
733 error = loop_switch(lo, file);
739 ioctl_by_bdev(bdev, BLKRRPART, 0);
748 static inline int is_loop_device(struct file *file)
750 struct inode *i = file->f_mapping->host;
752 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
755 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
756 struct block_device *bdev, unsigned int arg)
758 struct file *file, *f;
760 struct address_space *mapping;
761 unsigned lo_blocksize;
766 /* This is safe, since we have a reference from open(). */
767 __module_get(THIS_MODULE);
775 if (lo->lo_state != Lo_unbound)
778 /* Avoid recursion */
780 while (is_loop_device(f)) {
781 struct loop_device *l;
783 if (f->f_mapping->host->i_bdev == bdev)
786 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
787 if (l->lo_state == Lo_unbound) {
791 f = l->lo_backing_file;
794 mapping = file->f_mapping;
795 inode = mapping->host;
797 if (!(file->f_mode & FMODE_WRITE))
798 lo_flags |= LO_FLAGS_READ_ONLY;
801 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
802 const struct address_space_operations *aops = mapping->a_ops;
804 * If we can't read - sorry. If we only can't write - well,
805 * it's going to be read-only.
807 if (!file->f_op->splice_read)
809 if (aops->write_begin)
810 lo_flags |= LO_FLAGS_USE_AOPS;
811 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
812 lo_flags |= LO_FLAGS_READ_ONLY;
814 lo_blocksize = S_ISBLK(inode->i_mode) ?
815 inode->i_bdev->bd_block_size : PAGE_SIZE;
822 size = get_loop_size(lo, file);
824 if ((loff_t)(sector_t)size != size) {
829 if (!(mode & FMODE_WRITE))
830 lo_flags |= LO_FLAGS_READ_ONLY;
832 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
834 lo->lo_blocksize = lo_blocksize;
835 lo->lo_device = bdev;
836 lo->lo_flags = lo_flags;
837 lo->lo_backing_file = file;
838 lo->transfer = transfer_none;
840 lo->lo_sizelimit = 0;
841 lo->old_gfp_mask = mapping_gfp_mask(mapping);
842 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
844 lo->lo_bio = lo->lo_biotail = NULL;
847 * set queue make_request_fn, and add limits based on lower level
850 blk_queue_make_request(lo->lo_queue, loop_make_request);
851 lo->lo_queue->queuedata = lo;
852 lo->lo_queue->unplug_fn = loop_unplug;
854 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
855 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL);
857 set_capacity(lo->lo_disk, size);
858 bd_set_size(bdev, size << 9);
860 set_blocksize(bdev, lo_blocksize);
862 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
864 if (IS_ERR(lo->lo_thread)) {
865 error = PTR_ERR(lo->lo_thread);
868 lo->lo_state = Lo_bound;
869 wake_up_process(lo->lo_thread);
871 ioctl_by_bdev(bdev, BLKRRPART, 0);
875 lo->lo_thread = NULL;
876 lo->lo_device = NULL;
877 lo->lo_backing_file = NULL;
879 set_capacity(lo->lo_disk, 0);
880 invalidate_bdev(bdev);
881 bd_set_size(bdev, 0);
882 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
883 lo->lo_state = Lo_unbound;
887 /* This is safe: open() is still holding a reference. */
888 module_put(THIS_MODULE);
893 loop_release_xfer(struct loop_device *lo)
896 struct loop_func_table *xfer = lo->lo_encryption;
900 err = xfer->release(lo);
902 lo->lo_encryption = NULL;
903 module_put(xfer->owner);
909 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
910 const struct loop_info64 *i)
915 struct module *owner = xfer->owner;
917 if (!try_module_get(owner))
920 err = xfer->init(lo, i);
924 lo->lo_encryption = xfer;
929 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
931 struct file *filp = lo->lo_backing_file;
932 gfp_t gfp = lo->old_gfp_mask;
934 if (lo->lo_state != Lo_bound)
937 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
943 spin_lock_irq(&lo->lo_lock);
944 lo->lo_state = Lo_rundown;
945 spin_unlock_irq(&lo->lo_lock);
947 kthread_stop(lo->lo_thread);
949 lo->lo_queue->unplug_fn = NULL;
950 lo->lo_backing_file = NULL;
952 loop_release_xfer(lo);
955 lo->lo_device = NULL;
956 lo->lo_encryption = NULL;
958 lo->lo_sizelimit = 0;
959 lo->lo_encrypt_key_size = 0;
961 lo->lo_thread = NULL;
962 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
963 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
964 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
966 invalidate_bdev(bdev);
967 set_capacity(lo->lo_disk, 0);
969 bd_set_size(bdev, 0);
970 mapping_set_gfp_mask(filp->f_mapping, gfp);
971 lo->lo_state = Lo_unbound;
972 /* This is safe: open() is still holding a reference. */
973 module_put(THIS_MODULE);
975 ioctl_by_bdev(bdev, BLKRRPART, 0);
976 mutex_unlock(&lo->lo_ctl_mutex);
978 * Need not hold lo_ctl_mutex to fput backing file.
979 * Calling fput holding lo_ctl_mutex triggers a circular
980 * lock dependency possibility warning as fput can take
981 * bd_mutex which is usually taken before lo_ctl_mutex.
988 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
991 struct loop_func_table *xfer;
992 uid_t uid = current_uid();
994 if (lo->lo_encrypt_key_size &&
995 lo->lo_key_owner != uid &&
996 !capable(CAP_SYS_ADMIN))
998 if (lo->lo_state != Lo_bound)
1000 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1003 err = loop_release_xfer(lo);
1007 if (info->lo_encrypt_type) {
1008 unsigned int type = info->lo_encrypt_type;
1010 if (type >= MAX_LO_CRYPT)
1012 xfer = xfer_funcs[type];
1018 err = loop_init_xfer(lo, xfer, info);
1022 if (lo->lo_offset != info->lo_offset ||
1023 lo->lo_sizelimit != info->lo_sizelimit) {
1024 lo->lo_offset = info->lo_offset;
1025 lo->lo_sizelimit = info->lo_sizelimit;
1026 if (figure_loop_size(lo))
1030 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1031 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1032 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1033 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1037 lo->transfer = xfer->transfer;
1038 lo->ioctl = xfer->ioctl;
1040 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1041 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1042 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1044 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1045 lo->lo_init[0] = info->lo_init[0];
1046 lo->lo_init[1] = info->lo_init[1];
1047 if (info->lo_encrypt_key_size) {
1048 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1049 info->lo_encrypt_key_size);
1050 lo->lo_key_owner = uid;
1057 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1059 struct file *file = lo->lo_backing_file;
1063 if (lo->lo_state != Lo_bound)
1065 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1068 memset(info, 0, sizeof(*info));
1069 info->lo_number = lo->lo_number;
1070 info->lo_device = huge_encode_dev(stat.dev);
1071 info->lo_inode = stat.ino;
1072 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1073 info->lo_offset = lo->lo_offset;
1074 info->lo_sizelimit = lo->lo_sizelimit;
1075 info->lo_flags = lo->lo_flags;
1076 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1077 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1078 info->lo_encrypt_type =
1079 lo->lo_encryption ? lo->lo_encryption->number : 0;
1080 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1081 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1082 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1083 lo->lo_encrypt_key_size);
1089 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1091 memset(info64, 0, sizeof(*info64));
1092 info64->lo_number = info->lo_number;
1093 info64->lo_device = info->lo_device;
1094 info64->lo_inode = info->lo_inode;
1095 info64->lo_rdevice = info->lo_rdevice;
1096 info64->lo_offset = info->lo_offset;
1097 info64->lo_sizelimit = 0;
1098 info64->lo_encrypt_type = info->lo_encrypt_type;
1099 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1100 info64->lo_flags = info->lo_flags;
1101 info64->lo_init[0] = info->lo_init[0];
1102 info64->lo_init[1] = info->lo_init[1];
1103 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1104 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1106 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1107 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1111 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1113 memset(info, 0, sizeof(*info));
1114 info->lo_number = info64->lo_number;
1115 info->lo_device = info64->lo_device;
1116 info->lo_inode = info64->lo_inode;
1117 info->lo_rdevice = info64->lo_rdevice;
1118 info->lo_offset = info64->lo_offset;
1119 info->lo_encrypt_type = info64->lo_encrypt_type;
1120 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1121 info->lo_flags = info64->lo_flags;
1122 info->lo_init[0] = info64->lo_init[0];
1123 info->lo_init[1] = info64->lo_init[1];
1124 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1125 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1127 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1128 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1130 /* error in case values were truncated */
1131 if (info->lo_device != info64->lo_device ||
1132 info->lo_rdevice != info64->lo_rdevice ||
1133 info->lo_inode != info64->lo_inode ||
1134 info->lo_offset != info64->lo_offset)
1141 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1143 struct loop_info info;
1144 struct loop_info64 info64;
1146 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1148 loop_info64_from_old(&info, &info64);
1149 return loop_set_status(lo, &info64);
1153 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1155 struct loop_info64 info64;
1157 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1159 return loop_set_status(lo, &info64);
1163 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1164 struct loop_info info;
1165 struct loop_info64 info64;
1171 err = loop_get_status(lo, &info64);
1173 err = loop_info64_to_old(&info64, &info);
1174 if (!err && copy_to_user(arg, &info, sizeof(info)))
1181 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1182 struct loop_info64 info64;
1188 err = loop_get_status(lo, &info64);
1189 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1195 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1202 if (unlikely(lo->lo_state != Lo_bound))
1204 err = figure_loop_size(lo);
1207 sec = get_capacity(lo->lo_disk);
1208 /* the width of sector_t may be narrow for bit-shift */
1211 mutex_lock(&bdev->bd_mutex);
1212 bd_set_size(bdev, sz);
1213 mutex_unlock(&bdev->bd_mutex);
1219 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1220 unsigned int cmd, unsigned long arg)
1222 struct loop_device *lo = bdev->bd_disk->private_data;
1225 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1228 err = loop_set_fd(lo, mode, bdev, arg);
1230 case LOOP_CHANGE_FD:
1231 err = loop_change_fd(lo, bdev, arg);
1234 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1235 err = loop_clr_fd(lo, bdev);
1239 case LOOP_SET_STATUS:
1240 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1242 case LOOP_GET_STATUS:
1243 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1245 case LOOP_SET_STATUS64:
1246 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1248 case LOOP_GET_STATUS64:
1249 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1251 case LOOP_SET_CAPACITY:
1253 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1254 err = loop_set_capacity(lo, bdev);
1257 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1259 mutex_unlock(&lo->lo_ctl_mutex);
1265 #ifdef CONFIG_COMPAT
1266 struct compat_loop_info {
1267 compat_int_t lo_number; /* ioctl r/o */
1268 compat_dev_t lo_device; /* ioctl r/o */
1269 compat_ulong_t lo_inode; /* ioctl r/o */
1270 compat_dev_t lo_rdevice; /* ioctl r/o */
1271 compat_int_t lo_offset;
1272 compat_int_t lo_encrypt_type;
1273 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1274 compat_int_t lo_flags; /* ioctl r/o */
1275 char lo_name[LO_NAME_SIZE];
1276 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1277 compat_ulong_t lo_init[2];
1282 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1283 * - noinlined to reduce stack space usage in main part of driver
1286 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1287 struct loop_info64 *info64)
1289 struct compat_loop_info info;
1291 if (copy_from_user(&info, arg, sizeof(info)))
1294 memset(info64, 0, sizeof(*info64));
1295 info64->lo_number = info.lo_number;
1296 info64->lo_device = info.lo_device;
1297 info64->lo_inode = info.lo_inode;
1298 info64->lo_rdevice = info.lo_rdevice;
1299 info64->lo_offset = info.lo_offset;
1300 info64->lo_sizelimit = 0;
1301 info64->lo_encrypt_type = info.lo_encrypt_type;
1302 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1303 info64->lo_flags = info.lo_flags;
1304 info64->lo_init[0] = info.lo_init[0];
1305 info64->lo_init[1] = info.lo_init[1];
1306 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1307 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1309 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1310 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1315 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1316 * - noinlined to reduce stack space usage in main part of driver
1319 loop_info64_to_compat(const struct loop_info64 *info64,
1320 struct compat_loop_info __user *arg)
1322 struct compat_loop_info info;
1324 memset(&info, 0, sizeof(info));
1325 info.lo_number = info64->lo_number;
1326 info.lo_device = info64->lo_device;
1327 info.lo_inode = info64->lo_inode;
1328 info.lo_rdevice = info64->lo_rdevice;
1329 info.lo_offset = info64->lo_offset;
1330 info.lo_encrypt_type = info64->lo_encrypt_type;
1331 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1332 info.lo_flags = info64->lo_flags;
1333 info.lo_init[0] = info64->lo_init[0];
1334 info.lo_init[1] = info64->lo_init[1];
1335 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1336 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1338 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1339 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1341 /* error in case values were truncated */
1342 if (info.lo_device != info64->lo_device ||
1343 info.lo_rdevice != info64->lo_rdevice ||
1344 info.lo_inode != info64->lo_inode ||
1345 info.lo_offset != info64->lo_offset ||
1346 info.lo_init[0] != info64->lo_init[0] ||
1347 info.lo_init[1] != info64->lo_init[1])
1350 if (copy_to_user(arg, &info, sizeof(info)))
1356 loop_set_status_compat(struct loop_device *lo,
1357 const struct compat_loop_info __user *arg)
1359 struct loop_info64 info64;
1362 ret = loop_info64_from_compat(arg, &info64);
1365 return loop_set_status(lo, &info64);
1369 loop_get_status_compat(struct loop_device *lo,
1370 struct compat_loop_info __user *arg)
1372 struct loop_info64 info64;
1378 err = loop_get_status(lo, &info64);
1380 err = loop_info64_to_compat(&info64, arg);
1384 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1385 unsigned int cmd, unsigned long arg)
1387 struct loop_device *lo = bdev->bd_disk->private_data;
1391 case LOOP_SET_STATUS:
1392 mutex_lock(&lo->lo_ctl_mutex);
1393 err = loop_set_status_compat(
1394 lo, (const struct compat_loop_info __user *) arg);
1395 mutex_unlock(&lo->lo_ctl_mutex);
1397 case LOOP_GET_STATUS:
1398 mutex_lock(&lo->lo_ctl_mutex);
1399 err = loop_get_status_compat(
1400 lo, (struct compat_loop_info __user *) arg);
1401 mutex_unlock(&lo->lo_ctl_mutex);
1403 case LOOP_SET_CAPACITY:
1405 case LOOP_GET_STATUS64:
1406 case LOOP_SET_STATUS64:
1407 arg = (unsigned long) compat_ptr(arg);
1409 case LOOP_CHANGE_FD:
1410 err = lo_ioctl(bdev, mode, cmd, arg);
1420 static int lo_open(struct block_device *bdev, fmode_t mode)
1422 struct loop_device *lo = bdev->bd_disk->private_data;
1424 mutex_lock(&lo->lo_ctl_mutex);
1426 mutex_unlock(&lo->lo_ctl_mutex);
1431 static int lo_release(struct gendisk *disk, fmode_t mode)
1433 struct loop_device *lo = disk->private_data;
1436 mutex_lock(&lo->lo_ctl_mutex);
1438 if (--lo->lo_refcnt)
1441 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1443 * In autoclear mode, stop the loop thread
1444 * and remove configuration after last close.
1446 err = loop_clr_fd(lo, NULL);
1451 * Otherwise keep thread (if running) and config,
1452 * but flush possible ongoing bios in thread.
1458 mutex_unlock(&lo->lo_ctl_mutex);
1463 static struct block_device_operations lo_fops = {
1464 .owner = THIS_MODULE,
1466 .release = lo_release,
1468 #ifdef CONFIG_COMPAT
1469 .compat_ioctl = lo_compat_ioctl,
1474 * And now the modules code and kernel interface.
1476 static int max_loop;
1477 module_param(max_loop, int, 0);
1478 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1479 module_param(max_part, int, 0);
1480 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1481 MODULE_LICENSE("GPL");
1482 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1484 int loop_register_transfer(struct loop_func_table *funcs)
1486 unsigned int n = funcs->number;
1488 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1490 xfer_funcs[n] = funcs;
1494 int loop_unregister_transfer(int number)
1496 unsigned int n = number;
1497 struct loop_device *lo;
1498 struct loop_func_table *xfer;
1500 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1503 xfer_funcs[n] = NULL;
1505 list_for_each_entry(lo, &loop_devices, lo_list) {
1506 mutex_lock(&lo->lo_ctl_mutex);
1508 if (lo->lo_encryption == xfer)
1509 loop_release_xfer(lo);
1511 mutex_unlock(&lo->lo_ctl_mutex);
1517 EXPORT_SYMBOL(loop_register_transfer);
1518 EXPORT_SYMBOL(loop_unregister_transfer);
1520 static struct loop_device *loop_alloc(int i)
1522 struct loop_device *lo;
1523 struct gendisk *disk;
1525 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1529 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1533 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1535 goto out_free_queue;
1537 mutex_init(&lo->lo_ctl_mutex);
1539 lo->lo_thread = NULL;
1540 init_waitqueue_head(&lo->lo_event);
1541 spin_lock_init(&lo->lo_lock);
1542 disk->major = LOOP_MAJOR;
1543 disk->first_minor = i << part_shift;
1544 disk->fops = &lo_fops;
1545 disk->private_data = lo;
1546 disk->queue = lo->lo_queue;
1547 sprintf(disk->disk_name, "loop%d", i);
1551 blk_cleanup_queue(lo->lo_queue);
1558 static void loop_free(struct loop_device *lo)
1560 blk_cleanup_queue(lo->lo_queue);
1561 put_disk(lo->lo_disk);
1562 list_del(&lo->lo_list);
1566 static struct loop_device *loop_init_one(int i)
1568 struct loop_device *lo;
1570 list_for_each_entry(lo, &loop_devices, lo_list) {
1571 if (lo->lo_number == i)
1577 add_disk(lo->lo_disk);
1578 list_add_tail(&lo->lo_list, &loop_devices);
1583 static void loop_del_one(struct loop_device *lo)
1585 del_gendisk(lo->lo_disk);
1589 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1591 struct loop_device *lo;
1592 struct kobject *kobj;
1594 mutex_lock(&loop_devices_mutex);
1595 lo = loop_init_one(dev & MINORMASK);
1596 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1597 mutex_unlock(&loop_devices_mutex);
1603 static int __init loop_init(void)
1606 unsigned long range;
1607 struct loop_device *lo, *next;
1610 * loop module now has a feature to instantiate underlying device
1611 * structure on-demand, provided that there is an access dev node.
1612 * However, this will not work well with user space tool that doesn't
1613 * know about such "feature". In order to not break any existing
1614 * tool, we do the following:
1616 * (1) if max_loop is specified, create that many upfront, and this
1617 * also becomes a hard limit.
1618 * (2) if max_loop is not specified, create 8 loop device on module
1619 * load, user can further extend loop device by create dev node
1620 * themselves and have kernel automatically instantiate actual
1626 part_shift = fls(max_part);
1628 if (max_loop > 1UL << (MINORBITS - part_shift))
1636 range = 1UL << (MINORBITS - part_shift);
1639 if (register_blkdev(LOOP_MAJOR, "loop"))
1642 for (i = 0; i < nr; i++) {
1646 list_add_tail(&lo->lo_list, &loop_devices);
1649 /* point of no return */
1651 list_for_each_entry(lo, &loop_devices, lo_list)
1652 add_disk(lo->lo_disk);
1654 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1655 THIS_MODULE, loop_probe, NULL, NULL);
1657 printk(KERN_INFO "loop: module loaded\n");
1661 printk(KERN_INFO "loop: out of memory\n");
1663 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1666 unregister_blkdev(LOOP_MAJOR, "loop");
1670 static void __exit loop_exit(void)
1672 unsigned long range;
1673 struct loop_device *lo, *next;
1675 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1677 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1680 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1681 unregister_blkdev(LOOP_MAJOR, "loop");
1684 module_init(loop_init);
1685 module_exit(loop_exit);
1688 static int __init max_loop_setup(char *str)
1690 max_loop = simple_strtol(str, NULL, 0);
1694 __setup("max_loop=", max_loop_setup);