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/swap.h>
65 #include <linux/slab.h>
66 #include <linux/loop.h>
67 #include <linux/compat.h>
68 #include <linux/suspend.h>
69 #include <linux/freezer.h>
70 #include <linux/mutex.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/kthread.h>
76 #include <linux/splice.h>
78 #include <asm/uaccess.h>
80 static DEFINE_MUTEX(loop_mutex);
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 file_update_time(file);
244 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
245 bvec->bv_page, bv_offs, size, IV);
247 if (unlikely(transfer_result))
250 ret = pagecache_write_end(file, mapping, pos, size, copied,
252 if (ret < 0 || ret != copied)
255 if (unlikely(transfer_result))
266 mutex_unlock(&mapping->host->i_mutex);
274 * __do_lo_send_write - helper for writing data to a loop device
276 * This helper just factors out common code between do_lo_send_direct_write()
277 * and do_lo_send_write().
279 static int __do_lo_send_write(struct file *file,
280 u8 *buf, const int len, loff_t pos)
283 mm_segment_t old_fs = get_fs();
286 bw = file->f_op->write(file, buf, len, &pos);
288 if (likely(bw == len))
290 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
291 (unsigned long long)pos, len);
298 * do_lo_send_direct_write - helper for writing data to a loop device
300 * This is the fast, non-transforming version for backing filesystems which do
301 * not implement the address space operations write_begin and write_end.
302 * It uses the write file operation which should be present on all writeable
305 static int do_lo_send_direct_write(struct loop_device *lo,
306 struct bio_vec *bvec, loff_t pos, struct page *page)
308 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
309 kmap(bvec->bv_page) + bvec->bv_offset,
311 kunmap(bvec->bv_page);
317 * do_lo_send_write - helper for writing data to a loop device
319 * This is the slow, transforming version for filesystems which do not
320 * implement the address space operations write_begin and write_end. It
321 * uses the write file operation which should be present on all writeable
324 * Using fops->write is slower than using aops->{prepare,commit}_write in the
325 * transforming case because we need to double buffer the data as we cannot do
326 * the transformations in place as we do not have direct access to the
327 * destination pages of the backing file.
329 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
330 loff_t pos, struct page *page)
332 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
333 bvec->bv_offset, bvec->bv_len, pos >> 9);
335 return __do_lo_send_write(lo->lo_backing_file,
336 page_address(page), bvec->bv_len,
338 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
339 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
345 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
347 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
349 struct bio_vec *bvec;
350 struct page *page = NULL;
353 do_lo_send = do_lo_send_aops;
354 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
355 do_lo_send = do_lo_send_direct_write;
356 if (lo->transfer != transfer_none) {
357 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
361 do_lo_send = do_lo_send_write;
364 bio_for_each_segment(bvec, bio, i) {
365 ret = do_lo_send(lo, bvec, pos, page);
377 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
382 struct lo_read_data {
383 struct loop_device *lo;
390 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
391 struct splice_desc *sd)
393 struct lo_read_data *p = sd->u.data;
394 struct loop_device *lo = p->lo;
395 struct page *page = buf->page;
399 ret = buf->ops->confirm(pipe, buf);
403 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
409 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
410 printk(KERN_ERR "loop: transfer error block %ld\n",
415 flush_dcache_page(p->page);
424 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
426 return __splice_from_pipe(pipe, sd, lo_splice_actor);
430 do_lo_receive(struct loop_device *lo,
431 struct bio_vec *bvec, int bsize, loff_t pos)
433 struct lo_read_data cookie;
434 struct splice_desc sd;
439 cookie.page = bvec->bv_page;
440 cookie.offset = bvec->bv_offset;
441 cookie.bsize = bsize;
444 sd.total_len = bvec->bv_len;
449 file = lo->lo_backing_file;
450 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
459 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
461 struct bio_vec *bvec;
464 bio_for_each_segment(bvec, bio, i) {
465 ret = do_lo_receive(lo, bvec, bsize, pos);
473 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
478 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
480 if (bio_rw(bio) == WRITE) {
481 bool barrier = !!(bio->bi_rw & REQ_HARDBARRIER);
482 struct file *file = lo->lo_backing_file;
485 if (unlikely(!file->f_op->fsync)) {
490 ret = vfs_fsync(file, 0);
497 ret = lo_send(lo, bio, pos);
499 if (barrier && !ret) {
500 ret = vfs_fsync(file, 0);
505 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
512 * Add bio to back of pending list
514 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
516 bio_list_add(&lo->lo_bio_list, bio);
520 * Grab first pending buffer
522 static struct bio *loop_get_bio(struct loop_device *lo)
524 return bio_list_pop(&lo->lo_bio_list);
527 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
529 struct loop_device *lo = q->queuedata;
530 int rw = bio_rw(old_bio);
535 BUG_ON(!lo || (rw != READ && rw != WRITE));
537 spin_lock_irq(&lo->lo_lock);
538 if (lo->lo_state != Lo_bound)
540 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
542 loop_add_bio(lo, old_bio);
543 wake_up(&lo->lo_event);
544 spin_unlock_irq(&lo->lo_lock);
548 spin_unlock_irq(&lo->lo_lock);
549 bio_io_error(old_bio);
554 * kick off io on the underlying address space
556 static void loop_unplug(struct request_queue *q)
558 struct loop_device *lo = q->queuedata;
560 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
561 blk_run_address_space(lo->lo_backing_file->f_mapping);
564 struct switch_request {
566 struct completion wait;
569 static void do_loop_switch(struct loop_device *, struct switch_request *);
571 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
573 if (unlikely(!bio->bi_bdev)) {
574 do_loop_switch(lo, bio->bi_private);
577 int ret = do_bio_filebacked(lo, bio);
583 * worker thread that handles reads/writes to file backed loop devices,
584 * to avoid blocking in our make_request_fn. it also does loop decrypting
585 * on reads for block backed loop, as that is too heavy to do from
586 * b_end_io context where irqs may be disabled.
588 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
589 * calling kthread_stop(). Therefore once kthread_should_stop() is
590 * true, make_request will not place any more requests. Therefore
591 * once kthread_should_stop() is true and lo_bio is NULL, we are
592 * done with the loop.
594 static int loop_thread(void *data)
596 struct loop_device *lo = data;
599 set_user_nice(current, -20);
601 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
603 wait_event_interruptible(lo->lo_event,
604 !bio_list_empty(&lo->lo_bio_list) ||
605 kthread_should_stop());
607 if (bio_list_empty(&lo->lo_bio_list))
609 spin_lock_irq(&lo->lo_lock);
610 bio = loop_get_bio(lo);
611 spin_unlock_irq(&lo->lo_lock);
614 loop_handle_bio(lo, bio);
621 * loop_switch performs the hard work of switching a backing store.
622 * First it needs to flush existing IO, it does this by sending a magic
623 * BIO down the pipe. The completion of this BIO does the actual switch.
625 static int loop_switch(struct loop_device *lo, struct file *file)
627 struct switch_request w;
628 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
631 init_completion(&w.wait);
633 bio->bi_private = &w;
635 loop_make_request(lo->lo_queue, bio);
636 wait_for_completion(&w.wait);
641 * Helper to flush the IOs in loop, but keeping loop thread running
643 static int loop_flush(struct loop_device *lo)
645 /* loop not yet configured, no running thread, nothing to flush */
649 return loop_switch(lo, NULL);
653 * Do the actual switch; called from the BIO completion routine
655 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
657 struct file *file = p->file;
658 struct file *old_file = lo->lo_backing_file;
659 struct address_space *mapping;
661 /* if no new file, only flush of queued bios requested */
665 mapping = file->f_mapping;
666 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
667 lo->lo_backing_file = file;
668 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
669 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
670 lo->old_gfp_mask = mapping_gfp_mask(mapping);
671 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
678 * loop_change_fd switched the backing store of a loopback device to
679 * a new file. This is useful for operating system installers to free up
680 * the original file and in High Availability environments to switch to
681 * an alternative location for the content in case of server meltdown.
682 * This can only work if the loop device is used read-only, and if the
683 * new backing store is the same size and type as the old backing store.
685 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
688 struct file *file, *old_file;
693 if (lo->lo_state != Lo_bound)
696 /* the loop device has to be read-only */
698 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
706 inode = file->f_mapping->host;
707 old_file = lo->lo_backing_file;
711 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
714 /* size of the new backing store needs to be the same */
715 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
719 error = loop_switch(lo, file);
725 ioctl_by_bdev(bdev, BLKRRPART, 0);
734 static inline int is_loop_device(struct file *file)
736 struct inode *i = file->f_mapping->host;
738 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
741 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
742 struct block_device *bdev, unsigned int arg)
744 struct file *file, *f;
746 struct address_space *mapping;
747 unsigned lo_blocksize;
752 /* This is safe, since we have a reference from open(). */
753 __module_get(THIS_MODULE);
761 if (lo->lo_state != Lo_unbound)
764 /* Avoid recursion */
766 while (is_loop_device(f)) {
767 struct loop_device *l;
769 if (f->f_mapping->host->i_bdev == bdev)
772 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
773 if (l->lo_state == Lo_unbound) {
777 f = l->lo_backing_file;
780 mapping = file->f_mapping;
781 inode = mapping->host;
783 if (!(file->f_mode & FMODE_WRITE))
784 lo_flags |= LO_FLAGS_READ_ONLY;
787 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
788 const struct address_space_operations *aops = mapping->a_ops;
790 if (aops->write_begin)
791 lo_flags |= LO_FLAGS_USE_AOPS;
792 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
793 lo_flags |= LO_FLAGS_READ_ONLY;
795 lo_blocksize = S_ISBLK(inode->i_mode) ?
796 inode->i_bdev->bd_block_size : PAGE_SIZE;
803 size = get_loop_size(lo, file);
805 if ((loff_t)(sector_t)size != size) {
810 if (!(mode & FMODE_WRITE))
811 lo_flags |= LO_FLAGS_READ_ONLY;
813 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
815 lo->lo_blocksize = lo_blocksize;
816 lo->lo_device = bdev;
817 lo->lo_flags = lo_flags;
818 lo->lo_backing_file = file;
819 lo->transfer = transfer_none;
821 lo->lo_sizelimit = 0;
822 lo->old_gfp_mask = mapping_gfp_mask(mapping);
823 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
825 bio_list_init(&lo->lo_bio_list);
828 * set queue make_request_fn, and add limits based on lower level
831 blk_queue_make_request(lo->lo_queue, loop_make_request);
832 lo->lo_queue->queuedata = lo;
833 lo->lo_queue->unplug_fn = loop_unplug;
835 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
836 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN);
838 set_capacity(lo->lo_disk, size);
839 bd_set_size(bdev, size << 9);
840 /* let user-space know about the new size */
841 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
843 set_blocksize(bdev, lo_blocksize);
845 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
847 if (IS_ERR(lo->lo_thread)) {
848 error = PTR_ERR(lo->lo_thread);
851 lo->lo_state = Lo_bound;
852 wake_up_process(lo->lo_thread);
854 ioctl_by_bdev(bdev, BLKRRPART, 0);
858 lo->lo_thread = NULL;
859 lo->lo_device = NULL;
860 lo->lo_backing_file = NULL;
862 set_capacity(lo->lo_disk, 0);
863 invalidate_bdev(bdev);
864 bd_set_size(bdev, 0);
865 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
866 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
867 lo->lo_state = Lo_unbound;
871 /* This is safe: open() is still holding a reference. */
872 module_put(THIS_MODULE);
877 loop_release_xfer(struct loop_device *lo)
880 struct loop_func_table *xfer = lo->lo_encryption;
884 err = xfer->release(lo);
886 lo->lo_encryption = NULL;
887 module_put(xfer->owner);
893 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
894 const struct loop_info64 *i)
899 struct module *owner = xfer->owner;
901 if (!try_module_get(owner))
904 err = xfer->init(lo, i);
908 lo->lo_encryption = xfer;
913 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
915 struct file *filp = lo->lo_backing_file;
916 gfp_t gfp = lo->old_gfp_mask;
918 if (lo->lo_state != Lo_bound)
921 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
927 spin_lock_irq(&lo->lo_lock);
928 lo->lo_state = Lo_rundown;
929 spin_unlock_irq(&lo->lo_lock);
931 kthread_stop(lo->lo_thread);
933 lo->lo_queue->unplug_fn = NULL;
934 lo->lo_backing_file = NULL;
936 loop_release_xfer(lo);
939 lo->lo_device = NULL;
940 lo->lo_encryption = NULL;
942 lo->lo_sizelimit = 0;
943 lo->lo_encrypt_key_size = 0;
945 lo->lo_thread = NULL;
946 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
947 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
948 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
950 invalidate_bdev(bdev);
951 set_capacity(lo->lo_disk, 0);
953 bd_set_size(bdev, 0);
954 /* let user-space know about this change */
955 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
957 mapping_set_gfp_mask(filp->f_mapping, gfp);
958 lo->lo_state = Lo_unbound;
959 /* This is safe: open() is still holding a reference. */
960 module_put(THIS_MODULE);
961 if (max_part > 0 && bdev)
962 ioctl_by_bdev(bdev, BLKRRPART, 0);
963 mutex_unlock(&lo->lo_ctl_mutex);
965 * Need not hold lo_ctl_mutex to fput backing file.
966 * Calling fput holding lo_ctl_mutex triggers a circular
967 * lock dependency possibility warning as fput can take
968 * bd_mutex which is usually taken before lo_ctl_mutex.
975 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
978 struct loop_func_table *xfer;
979 uid_t uid = current_uid();
981 if (lo->lo_encrypt_key_size &&
982 lo->lo_key_owner != uid &&
983 !capable(CAP_SYS_ADMIN))
985 if (lo->lo_state != Lo_bound)
987 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
990 err = loop_release_xfer(lo);
994 if (info->lo_encrypt_type) {
995 unsigned int type = info->lo_encrypt_type;
997 if (type >= MAX_LO_CRYPT)
999 xfer = xfer_funcs[type];
1005 err = loop_init_xfer(lo, xfer, info);
1009 if (lo->lo_offset != info->lo_offset ||
1010 lo->lo_sizelimit != info->lo_sizelimit) {
1011 lo->lo_offset = info->lo_offset;
1012 lo->lo_sizelimit = info->lo_sizelimit;
1013 if (figure_loop_size(lo))
1017 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1018 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1019 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1020 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1024 lo->transfer = xfer->transfer;
1025 lo->ioctl = xfer->ioctl;
1027 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1028 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1029 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1031 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1032 lo->lo_init[0] = info->lo_init[0];
1033 lo->lo_init[1] = info->lo_init[1];
1034 if (info->lo_encrypt_key_size) {
1035 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1036 info->lo_encrypt_key_size);
1037 lo->lo_key_owner = uid;
1044 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1046 struct file *file = lo->lo_backing_file;
1050 if (lo->lo_state != Lo_bound)
1052 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1055 memset(info, 0, sizeof(*info));
1056 info->lo_number = lo->lo_number;
1057 info->lo_device = huge_encode_dev(stat.dev);
1058 info->lo_inode = stat.ino;
1059 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1060 info->lo_offset = lo->lo_offset;
1061 info->lo_sizelimit = lo->lo_sizelimit;
1062 info->lo_flags = lo->lo_flags;
1063 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1064 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1065 info->lo_encrypt_type =
1066 lo->lo_encryption ? lo->lo_encryption->number : 0;
1067 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1068 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1069 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1070 lo->lo_encrypt_key_size);
1076 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1078 memset(info64, 0, sizeof(*info64));
1079 info64->lo_number = info->lo_number;
1080 info64->lo_device = info->lo_device;
1081 info64->lo_inode = info->lo_inode;
1082 info64->lo_rdevice = info->lo_rdevice;
1083 info64->lo_offset = info->lo_offset;
1084 info64->lo_sizelimit = 0;
1085 info64->lo_encrypt_type = info->lo_encrypt_type;
1086 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1087 info64->lo_flags = info->lo_flags;
1088 info64->lo_init[0] = info->lo_init[0];
1089 info64->lo_init[1] = info->lo_init[1];
1090 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1091 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1093 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1094 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1098 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1100 memset(info, 0, sizeof(*info));
1101 info->lo_number = info64->lo_number;
1102 info->lo_device = info64->lo_device;
1103 info->lo_inode = info64->lo_inode;
1104 info->lo_rdevice = info64->lo_rdevice;
1105 info->lo_offset = info64->lo_offset;
1106 info->lo_encrypt_type = info64->lo_encrypt_type;
1107 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1108 info->lo_flags = info64->lo_flags;
1109 info->lo_init[0] = info64->lo_init[0];
1110 info->lo_init[1] = info64->lo_init[1];
1111 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1112 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1114 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1115 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1117 /* error in case values were truncated */
1118 if (info->lo_device != info64->lo_device ||
1119 info->lo_rdevice != info64->lo_rdevice ||
1120 info->lo_inode != info64->lo_inode ||
1121 info->lo_offset != info64->lo_offset)
1128 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1130 struct loop_info info;
1131 struct loop_info64 info64;
1133 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1135 loop_info64_from_old(&info, &info64);
1136 return loop_set_status(lo, &info64);
1140 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1142 struct loop_info64 info64;
1144 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1146 return loop_set_status(lo, &info64);
1150 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1151 struct loop_info info;
1152 struct loop_info64 info64;
1158 err = loop_get_status(lo, &info64);
1160 err = loop_info64_to_old(&info64, &info);
1161 if (!err && copy_to_user(arg, &info, sizeof(info)))
1168 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1169 struct loop_info64 info64;
1175 err = loop_get_status(lo, &info64);
1176 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1182 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1189 if (unlikely(lo->lo_state != Lo_bound))
1191 err = figure_loop_size(lo);
1194 sec = get_capacity(lo->lo_disk);
1195 /* the width of sector_t may be narrow for bit-shift */
1198 mutex_lock(&bdev->bd_mutex);
1199 bd_set_size(bdev, sz);
1200 /* let user-space know about the new size */
1201 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1202 mutex_unlock(&bdev->bd_mutex);
1208 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1209 unsigned int cmd, unsigned long arg)
1211 struct loop_device *lo = bdev->bd_disk->private_data;
1214 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1217 err = loop_set_fd(lo, mode, bdev, arg);
1219 case LOOP_CHANGE_FD:
1220 err = loop_change_fd(lo, bdev, arg);
1223 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1224 err = loop_clr_fd(lo, bdev);
1228 case LOOP_SET_STATUS:
1229 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1231 case LOOP_GET_STATUS:
1232 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1234 case LOOP_SET_STATUS64:
1235 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1237 case LOOP_GET_STATUS64:
1238 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1240 case LOOP_SET_CAPACITY:
1242 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1243 err = loop_set_capacity(lo, bdev);
1246 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1248 mutex_unlock(&lo->lo_ctl_mutex);
1254 #ifdef CONFIG_COMPAT
1255 struct compat_loop_info {
1256 compat_int_t lo_number; /* ioctl r/o */
1257 compat_dev_t lo_device; /* ioctl r/o */
1258 compat_ulong_t lo_inode; /* ioctl r/o */
1259 compat_dev_t lo_rdevice; /* ioctl r/o */
1260 compat_int_t lo_offset;
1261 compat_int_t lo_encrypt_type;
1262 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1263 compat_int_t lo_flags; /* ioctl r/o */
1264 char lo_name[LO_NAME_SIZE];
1265 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1266 compat_ulong_t lo_init[2];
1271 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1272 * - noinlined to reduce stack space usage in main part of driver
1275 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1276 struct loop_info64 *info64)
1278 struct compat_loop_info info;
1280 if (copy_from_user(&info, arg, sizeof(info)))
1283 memset(info64, 0, sizeof(*info64));
1284 info64->lo_number = info.lo_number;
1285 info64->lo_device = info.lo_device;
1286 info64->lo_inode = info.lo_inode;
1287 info64->lo_rdevice = info.lo_rdevice;
1288 info64->lo_offset = info.lo_offset;
1289 info64->lo_sizelimit = 0;
1290 info64->lo_encrypt_type = info.lo_encrypt_type;
1291 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1292 info64->lo_flags = info.lo_flags;
1293 info64->lo_init[0] = info.lo_init[0];
1294 info64->lo_init[1] = info.lo_init[1];
1295 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1296 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1298 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1299 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1304 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1305 * - noinlined to reduce stack space usage in main part of driver
1308 loop_info64_to_compat(const struct loop_info64 *info64,
1309 struct compat_loop_info __user *arg)
1311 struct compat_loop_info info;
1313 memset(&info, 0, sizeof(info));
1314 info.lo_number = info64->lo_number;
1315 info.lo_device = info64->lo_device;
1316 info.lo_inode = info64->lo_inode;
1317 info.lo_rdevice = info64->lo_rdevice;
1318 info.lo_offset = info64->lo_offset;
1319 info.lo_encrypt_type = info64->lo_encrypt_type;
1320 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1321 info.lo_flags = info64->lo_flags;
1322 info.lo_init[0] = info64->lo_init[0];
1323 info.lo_init[1] = info64->lo_init[1];
1324 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1325 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1327 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1328 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1330 /* error in case values were truncated */
1331 if (info.lo_device != info64->lo_device ||
1332 info.lo_rdevice != info64->lo_rdevice ||
1333 info.lo_inode != info64->lo_inode ||
1334 info.lo_offset != info64->lo_offset ||
1335 info.lo_init[0] != info64->lo_init[0] ||
1336 info.lo_init[1] != info64->lo_init[1])
1339 if (copy_to_user(arg, &info, sizeof(info)))
1345 loop_set_status_compat(struct loop_device *lo,
1346 const struct compat_loop_info __user *arg)
1348 struct loop_info64 info64;
1351 ret = loop_info64_from_compat(arg, &info64);
1354 return loop_set_status(lo, &info64);
1358 loop_get_status_compat(struct loop_device *lo,
1359 struct compat_loop_info __user *arg)
1361 struct loop_info64 info64;
1367 err = loop_get_status(lo, &info64);
1369 err = loop_info64_to_compat(&info64, arg);
1373 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1374 unsigned int cmd, unsigned long arg)
1376 struct loop_device *lo = bdev->bd_disk->private_data;
1380 case LOOP_SET_STATUS:
1381 mutex_lock(&lo->lo_ctl_mutex);
1382 err = loop_set_status_compat(
1383 lo, (const struct compat_loop_info __user *) arg);
1384 mutex_unlock(&lo->lo_ctl_mutex);
1386 case LOOP_GET_STATUS:
1387 mutex_lock(&lo->lo_ctl_mutex);
1388 err = loop_get_status_compat(
1389 lo, (struct compat_loop_info __user *) arg);
1390 mutex_unlock(&lo->lo_ctl_mutex);
1392 case LOOP_SET_CAPACITY:
1394 case LOOP_GET_STATUS64:
1395 case LOOP_SET_STATUS64:
1396 arg = (unsigned long) compat_ptr(arg);
1398 case LOOP_CHANGE_FD:
1399 err = lo_ioctl(bdev, mode, cmd, arg);
1409 static int lo_open(struct block_device *bdev, fmode_t mode)
1411 struct loop_device *lo = bdev->bd_disk->private_data;
1413 mutex_lock(&loop_mutex);
1414 mutex_lock(&lo->lo_ctl_mutex);
1416 mutex_unlock(&lo->lo_ctl_mutex);
1417 mutex_unlock(&loop_mutex);
1422 static int lo_release(struct gendisk *disk, fmode_t mode)
1424 struct loop_device *lo = disk->private_data;
1427 mutex_lock(&loop_mutex);
1428 mutex_lock(&lo->lo_ctl_mutex);
1430 if (--lo->lo_refcnt)
1433 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1435 * In autoclear mode, stop the loop thread
1436 * and remove configuration after last close.
1438 err = loop_clr_fd(lo, NULL);
1443 * Otherwise keep thread (if running) and config,
1444 * but flush possible ongoing bios in thread.
1450 mutex_unlock(&lo->lo_ctl_mutex);
1452 mutex_unlock(&loop_mutex);
1456 static const struct block_device_operations lo_fops = {
1457 .owner = THIS_MODULE,
1459 .release = lo_release,
1461 #ifdef CONFIG_COMPAT
1462 .compat_ioctl = lo_compat_ioctl,
1467 * And now the modules code and kernel interface.
1469 static int max_loop;
1470 module_param(max_loop, int, 0);
1471 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1472 module_param(max_part, int, 0);
1473 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1474 MODULE_LICENSE("GPL");
1475 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1477 int loop_register_transfer(struct loop_func_table *funcs)
1479 unsigned int n = funcs->number;
1481 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1483 xfer_funcs[n] = funcs;
1487 int loop_unregister_transfer(int number)
1489 unsigned int n = number;
1490 struct loop_device *lo;
1491 struct loop_func_table *xfer;
1493 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1496 xfer_funcs[n] = NULL;
1498 list_for_each_entry(lo, &loop_devices, lo_list) {
1499 mutex_lock(&lo->lo_ctl_mutex);
1501 if (lo->lo_encryption == xfer)
1502 loop_release_xfer(lo);
1504 mutex_unlock(&lo->lo_ctl_mutex);
1510 EXPORT_SYMBOL(loop_register_transfer);
1511 EXPORT_SYMBOL(loop_unregister_transfer);
1513 static struct loop_device *loop_alloc(int i)
1515 struct loop_device *lo;
1516 struct gendisk *disk;
1518 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1522 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1526 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1528 goto out_free_queue;
1530 mutex_init(&lo->lo_ctl_mutex);
1532 lo->lo_thread = NULL;
1533 init_waitqueue_head(&lo->lo_event);
1534 spin_lock_init(&lo->lo_lock);
1535 disk->major = LOOP_MAJOR;
1536 disk->first_minor = i << part_shift;
1537 disk->fops = &lo_fops;
1538 disk->private_data = lo;
1539 disk->queue = lo->lo_queue;
1540 sprintf(disk->disk_name, "loop%d", i);
1544 blk_cleanup_queue(lo->lo_queue);
1551 static void loop_free(struct loop_device *lo)
1553 blk_cleanup_queue(lo->lo_queue);
1554 put_disk(lo->lo_disk);
1555 list_del(&lo->lo_list);
1559 static struct loop_device *loop_init_one(int i)
1561 struct loop_device *lo;
1563 list_for_each_entry(lo, &loop_devices, lo_list) {
1564 if (lo->lo_number == i)
1570 add_disk(lo->lo_disk);
1571 list_add_tail(&lo->lo_list, &loop_devices);
1576 static void loop_del_one(struct loop_device *lo)
1578 del_gendisk(lo->lo_disk);
1582 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1584 struct loop_device *lo;
1585 struct kobject *kobj;
1587 mutex_lock(&loop_devices_mutex);
1588 lo = loop_init_one(dev & MINORMASK);
1589 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1590 mutex_unlock(&loop_devices_mutex);
1596 static int __init loop_init(void)
1599 unsigned long range;
1600 struct loop_device *lo, *next;
1603 * loop module now has a feature to instantiate underlying device
1604 * structure on-demand, provided that there is an access dev node.
1605 * However, this will not work well with user space tool that doesn't
1606 * know about such "feature". In order to not break any existing
1607 * tool, we do the following:
1609 * (1) if max_loop is specified, create that many upfront, and this
1610 * also becomes a hard limit.
1611 * (2) if max_loop is not specified, create 8 loop device on module
1612 * load, user can further extend loop device by create dev node
1613 * themselves and have kernel automatically instantiate actual
1619 part_shift = fls(max_part);
1621 if (max_loop > 1UL << (MINORBITS - part_shift))
1629 range = 1UL << (MINORBITS - part_shift);
1632 if (register_blkdev(LOOP_MAJOR, "loop"))
1635 for (i = 0; i < nr; i++) {
1639 list_add_tail(&lo->lo_list, &loop_devices);
1642 /* point of no return */
1644 list_for_each_entry(lo, &loop_devices, lo_list)
1645 add_disk(lo->lo_disk);
1647 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1648 THIS_MODULE, loop_probe, NULL, NULL);
1650 printk(KERN_INFO "loop: module loaded\n");
1654 printk(KERN_INFO "loop: out of memory\n");
1656 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1659 unregister_blkdev(LOOP_MAJOR, "loop");
1663 static void __exit loop_exit(void)
1665 unsigned long range;
1666 struct loop_device *lo, *next;
1668 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1670 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1673 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1674 unregister_blkdev(LOOP_MAJOR, "loop");
1677 module_init(loop_init);
1678 module_exit(loop_exit);
1681 static int __init max_loop_setup(char *str)
1683 max_loop = simple_strtol(str, NULL, 0);
1687 __setup("max_loop=", max_loop_setup);