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 bio_list_add(&lo->lo_bio_list, bio);
518 * Grab first pending buffer
520 static struct bio *loop_get_bio(struct loop_device *lo)
522 return bio_list_pop(&lo->lo_bio_list);
525 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
527 struct loop_device *lo = q->queuedata;
528 int rw = bio_rw(old_bio);
533 BUG_ON(!lo || (rw != READ && rw != WRITE));
535 spin_lock_irq(&lo->lo_lock);
536 if (lo->lo_state != Lo_bound)
538 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
540 loop_add_bio(lo, old_bio);
541 wake_up(&lo->lo_event);
542 spin_unlock_irq(&lo->lo_lock);
546 spin_unlock_irq(&lo->lo_lock);
547 bio_io_error(old_bio);
552 * kick off io on the underlying address space
554 static void loop_unplug(struct request_queue *q)
556 struct loop_device *lo = q->queuedata;
558 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
559 blk_run_address_space(lo->lo_backing_file->f_mapping);
562 struct switch_request {
564 struct completion wait;
567 static void do_loop_switch(struct loop_device *, struct switch_request *);
569 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
571 if (unlikely(!bio->bi_bdev)) {
572 do_loop_switch(lo, bio->bi_private);
575 int ret = do_bio_filebacked(lo, bio);
581 * worker thread that handles reads/writes to file backed loop devices,
582 * to avoid blocking in our make_request_fn. it also does loop decrypting
583 * on reads for block backed loop, as that is too heavy to do from
584 * b_end_io context where irqs may be disabled.
586 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
587 * calling kthread_stop(). Therefore once kthread_should_stop() is
588 * true, make_request will not place any more requests. Therefore
589 * once kthread_should_stop() is true and lo_bio is NULL, we are
590 * done with the loop.
592 static int loop_thread(void *data)
594 struct loop_device *lo = data;
597 set_user_nice(current, -20);
599 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
601 wait_event_interruptible(lo->lo_event,
602 !bio_list_empty(&lo->lo_bio_list) ||
603 kthread_should_stop());
605 if (bio_list_empty(&lo->lo_bio_list))
607 spin_lock_irq(&lo->lo_lock);
608 bio = loop_get_bio(lo);
609 spin_unlock_irq(&lo->lo_lock);
612 loop_handle_bio(lo, bio);
619 * loop_switch performs the hard work of switching a backing store.
620 * First it needs to flush existing IO, it does this by sending a magic
621 * BIO down the pipe. The completion of this BIO does the actual switch.
623 static int loop_switch(struct loop_device *lo, struct file *file)
625 struct switch_request w;
626 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
629 init_completion(&w.wait);
631 bio->bi_private = &w;
633 loop_make_request(lo->lo_queue, bio);
634 wait_for_completion(&w.wait);
639 * Helper to flush the IOs in loop, but keeping loop thread running
641 static int loop_flush(struct loop_device *lo)
643 /* loop not yet configured, no running thread, nothing to flush */
647 return loop_switch(lo, NULL);
651 * Do the actual switch; called from the BIO completion routine
653 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
655 struct file *file = p->file;
656 struct file *old_file = lo->lo_backing_file;
657 struct address_space *mapping;
659 /* if no new file, only flush of queued bios requested */
663 mapping = file->f_mapping;
664 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
665 lo->lo_backing_file = file;
666 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
667 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
668 lo->old_gfp_mask = mapping_gfp_mask(mapping);
669 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
676 * loop_change_fd switched the backing store of a loopback device to
677 * a new file. This is useful for operating system installers to free up
678 * the original file and in High Availability environments to switch to
679 * an alternative location for the content in case of server meltdown.
680 * This can only work if the loop device is used read-only, and if the
681 * new backing store is the same size and type as the old backing store.
683 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
686 struct file *file, *old_file;
691 if (lo->lo_state != Lo_bound)
694 /* the loop device has to be read-only */
696 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
704 inode = file->f_mapping->host;
705 old_file = lo->lo_backing_file;
709 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
712 /* size of the new backing store needs to be the same */
713 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
717 error = loop_switch(lo, file);
723 ioctl_by_bdev(bdev, BLKRRPART, 0);
732 static inline int is_loop_device(struct file *file)
734 struct inode *i = file->f_mapping->host;
736 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
739 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
740 struct block_device *bdev, unsigned int arg)
742 struct file *file, *f;
744 struct address_space *mapping;
745 unsigned lo_blocksize;
750 /* This is safe, since we have a reference from open(). */
751 __module_get(THIS_MODULE);
759 if (lo->lo_state != Lo_unbound)
762 /* Avoid recursion */
764 while (is_loop_device(f)) {
765 struct loop_device *l;
767 if (f->f_mapping->host->i_bdev == bdev)
770 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
771 if (l->lo_state == Lo_unbound) {
775 f = l->lo_backing_file;
778 mapping = file->f_mapping;
779 inode = mapping->host;
781 if (!(file->f_mode & FMODE_WRITE))
782 lo_flags |= LO_FLAGS_READ_ONLY;
785 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
786 const struct address_space_operations *aops = mapping->a_ops;
788 if (aops->write_begin)
789 lo_flags |= LO_FLAGS_USE_AOPS;
790 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
791 lo_flags |= LO_FLAGS_READ_ONLY;
793 lo_blocksize = S_ISBLK(inode->i_mode) ?
794 inode->i_bdev->bd_block_size : PAGE_SIZE;
801 size = get_loop_size(lo, file);
803 if ((loff_t)(sector_t)size != size) {
808 if (!(mode & FMODE_WRITE))
809 lo_flags |= LO_FLAGS_READ_ONLY;
811 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
813 lo->lo_blocksize = lo_blocksize;
814 lo->lo_device = bdev;
815 lo->lo_flags = lo_flags;
816 lo->lo_backing_file = file;
817 lo->transfer = transfer_none;
819 lo->lo_sizelimit = 0;
820 lo->old_gfp_mask = mapping_gfp_mask(mapping);
821 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
823 bio_list_init(&lo->lo_bio_list);
826 * set queue make_request_fn, and add limits based on lower level
829 blk_queue_make_request(lo->lo_queue, loop_make_request);
830 lo->lo_queue->queuedata = lo;
831 lo->lo_queue->unplug_fn = loop_unplug;
833 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
834 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL);
836 set_capacity(lo->lo_disk, size);
837 bd_set_size(bdev, size << 9);
839 set_blocksize(bdev, lo_blocksize);
841 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
843 if (IS_ERR(lo->lo_thread)) {
844 error = PTR_ERR(lo->lo_thread);
847 lo->lo_state = Lo_bound;
848 wake_up_process(lo->lo_thread);
850 ioctl_by_bdev(bdev, BLKRRPART, 0);
854 lo->lo_thread = NULL;
855 lo->lo_device = NULL;
856 lo->lo_backing_file = NULL;
858 set_capacity(lo->lo_disk, 0);
859 invalidate_bdev(bdev);
860 bd_set_size(bdev, 0);
861 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
862 lo->lo_state = Lo_unbound;
866 /* This is safe: open() is still holding a reference. */
867 module_put(THIS_MODULE);
872 loop_release_xfer(struct loop_device *lo)
875 struct loop_func_table *xfer = lo->lo_encryption;
879 err = xfer->release(lo);
881 lo->lo_encryption = NULL;
882 module_put(xfer->owner);
888 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
889 const struct loop_info64 *i)
894 struct module *owner = xfer->owner;
896 if (!try_module_get(owner))
899 err = xfer->init(lo, i);
903 lo->lo_encryption = xfer;
908 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
910 struct file *filp = lo->lo_backing_file;
911 gfp_t gfp = lo->old_gfp_mask;
913 if (lo->lo_state != Lo_bound)
916 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
922 spin_lock_irq(&lo->lo_lock);
923 lo->lo_state = Lo_rundown;
924 spin_unlock_irq(&lo->lo_lock);
926 kthread_stop(lo->lo_thread);
928 lo->lo_queue->unplug_fn = NULL;
929 lo->lo_backing_file = NULL;
931 loop_release_xfer(lo);
934 lo->lo_device = NULL;
935 lo->lo_encryption = NULL;
937 lo->lo_sizelimit = 0;
938 lo->lo_encrypt_key_size = 0;
940 lo->lo_thread = NULL;
941 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
942 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
943 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
945 invalidate_bdev(bdev);
946 set_capacity(lo->lo_disk, 0);
948 bd_set_size(bdev, 0);
949 mapping_set_gfp_mask(filp->f_mapping, gfp);
950 lo->lo_state = Lo_unbound;
951 /* This is safe: open() is still holding a reference. */
952 module_put(THIS_MODULE);
954 ioctl_by_bdev(bdev, BLKRRPART, 0);
955 mutex_unlock(&lo->lo_ctl_mutex);
957 * Need not hold lo_ctl_mutex to fput backing file.
958 * Calling fput holding lo_ctl_mutex triggers a circular
959 * lock dependency possibility warning as fput can take
960 * bd_mutex which is usually taken before lo_ctl_mutex.
967 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
970 struct loop_func_table *xfer;
971 uid_t uid = current_uid();
973 if (lo->lo_encrypt_key_size &&
974 lo->lo_key_owner != uid &&
975 !capable(CAP_SYS_ADMIN))
977 if (lo->lo_state != Lo_bound)
979 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
982 err = loop_release_xfer(lo);
986 if (info->lo_encrypt_type) {
987 unsigned int type = info->lo_encrypt_type;
989 if (type >= MAX_LO_CRYPT)
991 xfer = xfer_funcs[type];
997 err = loop_init_xfer(lo, xfer, info);
1001 if (lo->lo_offset != info->lo_offset ||
1002 lo->lo_sizelimit != info->lo_sizelimit) {
1003 lo->lo_offset = info->lo_offset;
1004 lo->lo_sizelimit = info->lo_sizelimit;
1005 if (figure_loop_size(lo))
1009 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1010 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1011 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1012 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1016 lo->transfer = xfer->transfer;
1017 lo->ioctl = xfer->ioctl;
1019 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1020 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1021 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1023 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1024 lo->lo_init[0] = info->lo_init[0];
1025 lo->lo_init[1] = info->lo_init[1];
1026 if (info->lo_encrypt_key_size) {
1027 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1028 info->lo_encrypt_key_size);
1029 lo->lo_key_owner = uid;
1036 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1038 struct file *file = lo->lo_backing_file;
1042 if (lo->lo_state != Lo_bound)
1044 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1047 memset(info, 0, sizeof(*info));
1048 info->lo_number = lo->lo_number;
1049 info->lo_device = huge_encode_dev(stat.dev);
1050 info->lo_inode = stat.ino;
1051 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1052 info->lo_offset = lo->lo_offset;
1053 info->lo_sizelimit = lo->lo_sizelimit;
1054 info->lo_flags = lo->lo_flags;
1055 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1056 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1057 info->lo_encrypt_type =
1058 lo->lo_encryption ? lo->lo_encryption->number : 0;
1059 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1060 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1061 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1062 lo->lo_encrypt_key_size);
1068 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1070 memset(info64, 0, sizeof(*info64));
1071 info64->lo_number = info->lo_number;
1072 info64->lo_device = info->lo_device;
1073 info64->lo_inode = info->lo_inode;
1074 info64->lo_rdevice = info->lo_rdevice;
1075 info64->lo_offset = info->lo_offset;
1076 info64->lo_sizelimit = 0;
1077 info64->lo_encrypt_type = info->lo_encrypt_type;
1078 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1079 info64->lo_flags = info->lo_flags;
1080 info64->lo_init[0] = info->lo_init[0];
1081 info64->lo_init[1] = info->lo_init[1];
1082 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1083 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1085 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1086 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1090 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1092 memset(info, 0, sizeof(*info));
1093 info->lo_number = info64->lo_number;
1094 info->lo_device = info64->lo_device;
1095 info->lo_inode = info64->lo_inode;
1096 info->lo_rdevice = info64->lo_rdevice;
1097 info->lo_offset = info64->lo_offset;
1098 info->lo_encrypt_type = info64->lo_encrypt_type;
1099 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1100 info->lo_flags = info64->lo_flags;
1101 info->lo_init[0] = info64->lo_init[0];
1102 info->lo_init[1] = info64->lo_init[1];
1103 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1104 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1106 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1107 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1109 /* error in case values were truncated */
1110 if (info->lo_device != info64->lo_device ||
1111 info->lo_rdevice != info64->lo_rdevice ||
1112 info->lo_inode != info64->lo_inode ||
1113 info->lo_offset != info64->lo_offset)
1120 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1122 struct loop_info info;
1123 struct loop_info64 info64;
1125 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1127 loop_info64_from_old(&info, &info64);
1128 return loop_set_status(lo, &info64);
1132 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1134 struct loop_info64 info64;
1136 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1138 return loop_set_status(lo, &info64);
1142 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1143 struct loop_info info;
1144 struct loop_info64 info64;
1150 err = loop_get_status(lo, &info64);
1152 err = loop_info64_to_old(&info64, &info);
1153 if (!err && copy_to_user(arg, &info, sizeof(info)))
1160 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1161 struct loop_info64 info64;
1167 err = loop_get_status(lo, &info64);
1168 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1174 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1181 if (unlikely(lo->lo_state != Lo_bound))
1183 err = figure_loop_size(lo);
1186 sec = get_capacity(lo->lo_disk);
1187 /* the width of sector_t may be narrow for bit-shift */
1190 mutex_lock(&bdev->bd_mutex);
1191 bd_set_size(bdev, sz);
1192 mutex_unlock(&bdev->bd_mutex);
1198 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1199 unsigned int cmd, unsigned long arg)
1201 struct loop_device *lo = bdev->bd_disk->private_data;
1204 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1207 err = loop_set_fd(lo, mode, bdev, arg);
1209 case LOOP_CHANGE_FD:
1210 err = loop_change_fd(lo, bdev, arg);
1213 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1214 err = loop_clr_fd(lo, bdev);
1218 case LOOP_SET_STATUS:
1219 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1221 case LOOP_GET_STATUS:
1222 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1224 case LOOP_SET_STATUS64:
1225 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1227 case LOOP_GET_STATUS64:
1228 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1230 case LOOP_SET_CAPACITY:
1232 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1233 err = loop_set_capacity(lo, bdev);
1236 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1238 mutex_unlock(&lo->lo_ctl_mutex);
1244 #ifdef CONFIG_COMPAT
1245 struct compat_loop_info {
1246 compat_int_t lo_number; /* ioctl r/o */
1247 compat_dev_t lo_device; /* ioctl r/o */
1248 compat_ulong_t lo_inode; /* ioctl r/o */
1249 compat_dev_t lo_rdevice; /* ioctl r/o */
1250 compat_int_t lo_offset;
1251 compat_int_t lo_encrypt_type;
1252 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1253 compat_int_t lo_flags; /* ioctl r/o */
1254 char lo_name[LO_NAME_SIZE];
1255 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1256 compat_ulong_t lo_init[2];
1261 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1262 * - noinlined to reduce stack space usage in main part of driver
1265 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1266 struct loop_info64 *info64)
1268 struct compat_loop_info info;
1270 if (copy_from_user(&info, arg, sizeof(info)))
1273 memset(info64, 0, sizeof(*info64));
1274 info64->lo_number = info.lo_number;
1275 info64->lo_device = info.lo_device;
1276 info64->lo_inode = info.lo_inode;
1277 info64->lo_rdevice = info.lo_rdevice;
1278 info64->lo_offset = info.lo_offset;
1279 info64->lo_sizelimit = 0;
1280 info64->lo_encrypt_type = info.lo_encrypt_type;
1281 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1282 info64->lo_flags = info.lo_flags;
1283 info64->lo_init[0] = info.lo_init[0];
1284 info64->lo_init[1] = info.lo_init[1];
1285 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1286 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1288 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1289 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1294 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1295 * - noinlined to reduce stack space usage in main part of driver
1298 loop_info64_to_compat(const struct loop_info64 *info64,
1299 struct compat_loop_info __user *arg)
1301 struct compat_loop_info info;
1303 memset(&info, 0, sizeof(info));
1304 info.lo_number = info64->lo_number;
1305 info.lo_device = info64->lo_device;
1306 info.lo_inode = info64->lo_inode;
1307 info.lo_rdevice = info64->lo_rdevice;
1308 info.lo_offset = info64->lo_offset;
1309 info.lo_encrypt_type = info64->lo_encrypt_type;
1310 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1311 info.lo_flags = info64->lo_flags;
1312 info.lo_init[0] = info64->lo_init[0];
1313 info.lo_init[1] = info64->lo_init[1];
1314 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1315 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1317 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1318 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1320 /* error in case values were truncated */
1321 if (info.lo_device != info64->lo_device ||
1322 info.lo_rdevice != info64->lo_rdevice ||
1323 info.lo_inode != info64->lo_inode ||
1324 info.lo_offset != info64->lo_offset ||
1325 info.lo_init[0] != info64->lo_init[0] ||
1326 info.lo_init[1] != info64->lo_init[1])
1329 if (copy_to_user(arg, &info, sizeof(info)))
1335 loop_set_status_compat(struct loop_device *lo,
1336 const struct compat_loop_info __user *arg)
1338 struct loop_info64 info64;
1341 ret = loop_info64_from_compat(arg, &info64);
1344 return loop_set_status(lo, &info64);
1348 loop_get_status_compat(struct loop_device *lo,
1349 struct compat_loop_info __user *arg)
1351 struct loop_info64 info64;
1357 err = loop_get_status(lo, &info64);
1359 err = loop_info64_to_compat(&info64, arg);
1363 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1364 unsigned int cmd, unsigned long arg)
1366 struct loop_device *lo = bdev->bd_disk->private_data;
1370 case LOOP_SET_STATUS:
1371 mutex_lock(&lo->lo_ctl_mutex);
1372 err = loop_set_status_compat(
1373 lo, (const struct compat_loop_info __user *) arg);
1374 mutex_unlock(&lo->lo_ctl_mutex);
1376 case LOOP_GET_STATUS:
1377 mutex_lock(&lo->lo_ctl_mutex);
1378 err = loop_get_status_compat(
1379 lo, (struct compat_loop_info __user *) arg);
1380 mutex_unlock(&lo->lo_ctl_mutex);
1382 case LOOP_SET_CAPACITY:
1384 case LOOP_GET_STATUS64:
1385 case LOOP_SET_STATUS64:
1386 arg = (unsigned long) compat_ptr(arg);
1388 case LOOP_CHANGE_FD:
1389 err = lo_ioctl(bdev, mode, cmd, arg);
1399 static int lo_open(struct block_device *bdev, fmode_t mode)
1401 struct loop_device *lo = bdev->bd_disk->private_data;
1403 mutex_lock(&lo->lo_ctl_mutex);
1405 mutex_unlock(&lo->lo_ctl_mutex);
1410 static int lo_release(struct gendisk *disk, fmode_t mode)
1412 struct loop_device *lo = disk->private_data;
1415 mutex_lock(&lo->lo_ctl_mutex);
1417 if (--lo->lo_refcnt)
1420 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1422 * In autoclear mode, stop the loop thread
1423 * and remove configuration after last close.
1425 err = loop_clr_fd(lo, NULL);
1430 * Otherwise keep thread (if running) and config,
1431 * but flush possible ongoing bios in thread.
1437 mutex_unlock(&lo->lo_ctl_mutex);
1442 static struct block_device_operations lo_fops = {
1443 .owner = THIS_MODULE,
1445 .release = lo_release,
1447 #ifdef CONFIG_COMPAT
1448 .compat_ioctl = lo_compat_ioctl,
1453 * And now the modules code and kernel interface.
1455 static int max_loop;
1456 module_param(max_loop, int, 0);
1457 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1458 module_param(max_part, int, 0);
1459 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1460 MODULE_LICENSE("GPL");
1461 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1463 int loop_register_transfer(struct loop_func_table *funcs)
1465 unsigned int n = funcs->number;
1467 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1469 xfer_funcs[n] = funcs;
1473 int loop_unregister_transfer(int number)
1475 unsigned int n = number;
1476 struct loop_device *lo;
1477 struct loop_func_table *xfer;
1479 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1482 xfer_funcs[n] = NULL;
1484 list_for_each_entry(lo, &loop_devices, lo_list) {
1485 mutex_lock(&lo->lo_ctl_mutex);
1487 if (lo->lo_encryption == xfer)
1488 loop_release_xfer(lo);
1490 mutex_unlock(&lo->lo_ctl_mutex);
1496 EXPORT_SYMBOL(loop_register_transfer);
1497 EXPORT_SYMBOL(loop_unregister_transfer);
1499 static struct loop_device *loop_alloc(int i)
1501 struct loop_device *lo;
1502 struct gendisk *disk;
1504 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1508 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1512 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1514 goto out_free_queue;
1516 mutex_init(&lo->lo_ctl_mutex);
1518 lo->lo_thread = NULL;
1519 init_waitqueue_head(&lo->lo_event);
1520 spin_lock_init(&lo->lo_lock);
1521 disk->major = LOOP_MAJOR;
1522 disk->first_minor = i << part_shift;
1523 disk->fops = &lo_fops;
1524 disk->private_data = lo;
1525 disk->queue = lo->lo_queue;
1526 sprintf(disk->disk_name, "loop%d", i);
1530 blk_cleanup_queue(lo->lo_queue);
1537 static void loop_free(struct loop_device *lo)
1539 blk_cleanup_queue(lo->lo_queue);
1540 put_disk(lo->lo_disk);
1541 list_del(&lo->lo_list);
1545 static struct loop_device *loop_init_one(int i)
1547 struct loop_device *lo;
1549 list_for_each_entry(lo, &loop_devices, lo_list) {
1550 if (lo->lo_number == i)
1556 add_disk(lo->lo_disk);
1557 list_add_tail(&lo->lo_list, &loop_devices);
1562 static void loop_del_one(struct loop_device *lo)
1564 del_gendisk(lo->lo_disk);
1568 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1570 struct loop_device *lo;
1571 struct kobject *kobj;
1573 mutex_lock(&loop_devices_mutex);
1574 lo = loop_init_one(dev & MINORMASK);
1575 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1576 mutex_unlock(&loop_devices_mutex);
1582 static int __init loop_init(void)
1585 unsigned long range;
1586 struct loop_device *lo, *next;
1589 * loop module now has a feature to instantiate underlying device
1590 * structure on-demand, provided that there is an access dev node.
1591 * However, this will not work well with user space tool that doesn't
1592 * know about such "feature". In order to not break any existing
1593 * tool, we do the following:
1595 * (1) if max_loop is specified, create that many upfront, and this
1596 * also becomes a hard limit.
1597 * (2) if max_loop is not specified, create 8 loop device on module
1598 * load, user can further extend loop device by create dev node
1599 * themselves and have kernel automatically instantiate actual
1605 part_shift = fls(max_part);
1607 if (max_loop > 1UL << (MINORBITS - part_shift))
1615 range = 1UL << (MINORBITS - part_shift);
1618 if (register_blkdev(LOOP_MAJOR, "loop"))
1621 for (i = 0; i < nr; i++) {
1625 list_add_tail(&lo->lo_list, &loop_devices);
1628 /* point of no return */
1630 list_for_each_entry(lo, &loop_devices, lo_list)
1631 add_disk(lo->lo_disk);
1633 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1634 THIS_MODULE, loop_probe, NULL, NULL);
1636 printk(KERN_INFO "loop: module loaded\n");
1640 printk(KERN_INFO "loop: out of memory\n");
1642 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1645 unregister_blkdev(LOOP_MAJOR, "loop");
1649 static void __exit loop_exit(void)
1651 unsigned long range;
1652 struct loop_device *lo, *next;
1654 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1656 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1659 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1660 unregister_blkdev(LOOP_MAJOR, "loop");
1663 module_init(loop_init);
1664 module_exit(loop_exit);
1667 static int __init max_loop_setup(char *str)
1669 max_loop = simple_strtol(str, NULL, 0);
1673 __setup("max_loop=", max_loop_setup);