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/writeback.h>
71 #include <linux/buffer_head.h> /* for invalidate_bdev() */
72 #include <linux/completion.h>
73 #include <linux/highmem.h>
74 #include <linux/gfp.h>
75 #include <linux/kthread.h>
76 #include <linux/splice.h>
78 #include <asm/uaccess.h>
80 static LIST_HEAD(loop_devices);
81 static DEFINE_MUTEX(loop_devices_mutex);
84 static int part_shift;
89 static int transfer_none(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
94 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
98 memcpy(loop_buf, raw_buf, size);
100 memcpy(raw_buf, loop_buf, size);
102 kunmap_atomic(raw_buf, KM_USER0);
103 kunmap_atomic(loop_buf, KM_USER1);
108 static int transfer_xor(struct loop_device *lo, int cmd,
109 struct page *raw_page, unsigned raw_off,
110 struct page *loop_page, unsigned loop_off,
111 int size, sector_t real_block)
113 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
114 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
115 char *in, *out, *key;
126 key = lo->lo_encrypt_key;
127 keysize = lo->lo_encrypt_key_size;
128 for (i = 0; i < size; i++)
129 *out++ = *in++ ^ key[(i & 511) % keysize];
131 kunmap_atomic(raw_buf, KM_USER0);
132 kunmap_atomic(loop_buf, KM_USER1);
137 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
139 if (unlikely(info->lo_encrypt_key_size <= 0))
144 static struct loop_func_table none_funcs = {
145 .number = LO_CRYPT_NONE,
146 .transfer = transfer_none,
149 static struct loop_func_table xor_funcs = {
150 .number = LO_CRYPT_XOR,
151 .transfer = transfer_xor,
155 /* xfer_funcs[0] is special - its release function is never called */
156 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
161 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
163 loff_t size, offset, loopsize;
165 /* Compute loopsize in bytes */
166 size = i_size_read(file->f_mapping->host);
167 offset = lo->lo_offset;
168 loopsize = size - offset;
169 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
170 loopsize = lo->lo_sizelimit;
173 * Unfortunately, if we want to do I/O on the device,
174 * the number of 512-byte sectors has to fit into a sector_t.
176 return loopsize >> 9;
180 figure_loop_size(struct loop_device *lo)
182 loff_t size = get_loop_size(lo, lo->lo_backing_file);
183 sector_t x = (sector_t)size;
185 if (unlikely((loff_t)x != size))
188 set_capacity(lo->lo_disk, x);
193 lo_do_transfer(struct loop_device *lo, int cmd,
194 struct page *rpage, unsigned roffs,
195 struct page *lpage, unsigned loffs,
196 int size, sector_t rblock)
198 if (unlikely(!lo->transfer))
201 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
205 * do_lo_send_aops - helper for writing data to a loop device
207 * This is the fast version for backing filesystems which implement the address
208 * space operations write_begin and write_end.
210 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
211 loff_t pos, struct page *unused)
213 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
214 struct address_space *mapping = file->f_mapping;
216 unsigned offset, bv_offs;
219 mutex_lock(&mapping->host->i_mutex);
220 index = pos >> PAGE_CACHE_SHIFT;
221 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
222 bv_offs = bvec->bv_offset;
226 unsigned size, copied;
231 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
232 size = PAGE_CACHE_SIZE - offset;
236 ret = pagecache_write_begin(file, mapping, pos, size, 0,
241 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
242 bvec->bv_page, bv_offs, size, IV);
244 if (unlikely(transfer_result))
247 ret = pagecache_write_end(file, mapping, pos, size, copied,
249 if (ret < 0 || ret != copied)
252 if (unlikely(transfer_result))
263 mutex_unlock(&mapping->host->i_mutex);
271 * __do_lo_send_write - helper for writing data to a loop device
273 * This helper just factors out common code between do_lo_send_direct_write()
274 * and do_lo_send_write().
276 static int __do_lo_send_write(struct file *file,
277 u8 *buf, const int len, loff_t pos)
280 mm_segment_t old_fs = get_fs();
283 bw = file->f_op->write(file, buf, len, &pos);
285 if (likely(bw == len))
287 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
288 (unsigned long long)pos, len);
295 * do_lo_send_direct_write - helper for writing data to a loop device
297 * This is the fast, non-transforming version for backing filesystems which do
298 * not implement the address space operations write_begin and write_end.
299 * It uses the write file operation which should be present on all writeable
302 static int do_lo_send_direct_write(struct loop_device *lo,
303 struct bio_vec *bvec, loff_t pos, struct page *page)
305 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
306 kmap(bvec->bv_page) + bvec->bv_offset,
308 kunmap(bvec->bv_page);
314 * do_lo_send_write - helper for writing data to a loop device
316 * This is the slow, transforming version for filesystems which do not
317 * implement the address space operations write_begin and write_end. It
318 * uses the write file operation which should be present on all writeable
321 * Using fops->write is slower than using aops->{prepare,commit}_write in the
322 * transforming case because we need to double buffer the data as we cannot do
323 * the transformations in place as we do not have direct access to the
324 * destination pages of the backing file.
326 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
327 loff_t pos, struct page *page)
329 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
330 bvec->bv_offset, bvec->bv_len, pos >> 9);
332 return __do_lo_send_write(lo->lo_backing_file,
333 page_address(page), bvec->bv_len,
335 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
336 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
342 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
344 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
346 struct bio_vec *bvec;
347 struct page *page = NULL;
350 do_lo_send = do_lo_send_aops;
351 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
352 do_lo_send = do_lo_send_direct_write;
353 if (lo->transfer != transfer_none) {
354 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
358 do_lo_send = do_lo_send_write;
361 bio_for_each_segment(bvec, bio, i) {
362 ret = do_lo_send(lo, bvec, pos, page);
374 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
379 struct lo_read_data {
380 struct loop_device *lo;
387 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
388 struct splice_desc *sd)
390 struct lo_read_data *p = sd->u.data;
391 struct loop_device *lo = p->lo;
392 struct page *page = buf->page;
396 ret = buf->ops->confirm(pipe, buf);
400 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
406 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
407 printk(KERN_ERR "loop: transfer error block %ld\n",
412 flush_dcache_page(p->page);
421 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
423 return __splice_from_pipe(pipe, sd, lo_splice_actor);
427 do_lo_receive(struct loop_device *lo,
428 struct bio_vec *bvec, int bsize, loff_t pos)
430 struct lo_read_data cookie;
431 struct splice_desc sd;
436 cookie.page = bvec->bv_page;
437 cookie.offset = bvec->bv_offset;
438 cookie.bsize = bsize;
441 sd.total_len = bvec->bv_len;
446 file = lo->lo_backing_file;
447 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
456 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
458 struct bio_vec *bvec;
461 bio_for_each_segment(bvec, bio, i) {
462 ret = do_lo_receive(lo, bvec, bsize, pos);
470 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
475 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
477 if (bio_rw(bio) == WRITE) {
478 int barrier = bio_barrier(bio);
479 struct file *file = lo->lo_backing_file;
482 if (unlikely(!file->f_op->fsync)) {
487 ret = vfs_fsync(file, file->f_path.dentry, 0);
494 ret = lo_send(lo, bio, pos);
496 if (barrier && !ret) {
497 ret = vfs_fsync(file, file->f_path.dentry, 0);
502 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
509 * Add bio to back of pending list
511 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
513 bio_list_add(&lo->lo_bio_list, bio);
517 * Grab first pending buffer
519 static struct bio *loop_get_bio(struct loop_device *lo)
521 return bio_list_pop(&lo->lo_bio_list);
524 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
526 struct loop_device *lo = q->queuedata;
527 int rw = bio_rw(old_bio);
532 BUG_ON(!lo || (rw != READ && rw != WRITE));
534 spin_lock_irq(&lo->lo_lock);
535 if (lo->lo_state != Lo_bound)
537 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
539 loop_add_bio(lo, old_bio);
540 wake_up(&lo->lo_event);
541 spin_unlock_irq(&lo->lo_lock);
545 spin_unlock_irq(&lo->lo_lock);
546 bio_io_error(old_bio);
551 * kick off io on the underlying address space
553 static void loop_unplug(struct request_queue *q)
555 struct loop_device *lo = q->queuedata;
557 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
558 blk_run_address_space(lo->lo_backing_file->f_mapping);
561 struct switch_request {
563 struct completion wait;
566 static void do_loop_switch(struct loop_device *, struct switch_request *);
568 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
570 if (unlikely(!bio->bi_bdev)) {
571 do_loop_switch(lo, bio->bi_private);
574 int ret = do_bio_filebacked(lo, bio);
580 * worker thread that handles reads/writes to file backed loop devices,
581 * to avoid blocking in our make_request_fn. it also does loop decrypting
582 * on reads for block backed loop, as that is too heavy to do from
583 * b_end_io context where irqs may be disabled.
585 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
586 * calling kthread_stop(). Therefore once kthread_should_stop() is
587 * true, make_request will not place any more requests. Therefore
588 * once kthread_should_stop() is true and lo_bio is NULL, we are
589 * done with the loop.
591 static int loop_thread(void *data)
593 struct loop_device *lo = data;
596 set_user_nice(current, -20);
598 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
600 wait_event_interruptible(lo->lo_event,
601 !bio_list_empty(&lo->lo_bio_list) ||
602 kthread_should_stop());
604 if (bio_list_empty(&lo->lo_bio_list))
606 spin_lock_irq(&lo->lo_lock);
607 bio = loop_get_bio(lo);
608 spin_unlock_irq(&lo->lo_lock);
611 loop_handle_bio(lo, bio);
618 * loop_switch performs the hard work of switching a backing store.
619 * First it needs to flush existing IO, it does this by sending a magic
620 * BIO down the pipe. The completion of this BIO does the actual switch.
622 static int loop_switch(struct loop_device *lo, struct file *file)
624 struct switch_request w;
625 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
628 init_completion(&w.wait);
630 bio->bi_private = &w;
632 loop_make_request(lo->lo_queue, bio);
633 wait_for_completion(&w.wait);
638 * Helper to flush the IOs in loop, but keeping loop thread running
640 static int loop_flush(struct loop_device *lo)
642 /* loop not yet configured, no running thread, nothing to flush */
646 return loop_switch(lo, NULL);
650 * Do the actual switch; called from the BIO completion routine
652 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
654 struct file *file = p->file;
655 struct file *old_file = lo->lo_backing_file;
656 struct address_space *mapping;
658 /* if no new file, only flush of queued bios requested */
662 mapping = file->f_mapping;
663 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
664 lo->lo_backing_file = file;
665 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
666 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
667 lo->old_gfp_mask = mapping_gfp_mask(mapping);
668 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
675 * loop_change_fd switched the backing store of a loopback device to
676 * a new file. This is useful for operating system installers to free up
677 * the original file and in High Availability environments to switch to
678 * an alternative location for the content in case of server meltdown.
679 * This can only work if the loop device is used read-only, and if the
680 * new backing store is the same size and type as the old backing store.
682 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
685 struct file *file, *old_file;
690 if (lo->lo_state != Lo_bound)
693 /* the loop device has to be read-only */
695 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
703 inode = file->f_mapping->host;
704 old_file = lo->lo_backing_file;
708 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
711 /* size of the new backing store needs to be the same */
712 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
716 error = loop_switch(lo, file);
722 ioctl_by_bdev(bdev, BLKRRPART, 0);
731 static inline int is_loop_device(struct file *file)
733 struct inode *i = file->f_mapping->host;
735 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
738 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
739 struct block_device *bdev, unsigned int arg)
741 struct file *file, *f;
743 struct address_space *mapping;
744 unsigned lo_blocksize;
749 /* This is safe, since we have a reference from open(). */
750 __module_get(THIS_MODULE);
758 if (lo->lo_state != Lo_unbound)
761 /* Avoid recursion */
763 while (is_loop_device(f)) {
764 struct loop_device *l;
766 if (f->f_mapping->host->i_bdev == bdev)
769 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
770 if (l->lo_state == Lo_unbound) {
774 f = l->lo_backing_file;
777 mapping = file->f_mapping;
778 inode = mapping->host;
780 if (!(file->f_mode & FMODE_WRITE))
781 lo_flags |= LO_FLAGS_READ_ONLY;
784 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
785 const struct address_space_operations *aops = mapping->a_ops;
787 if (aops->write_begin)
788 lo_flags |= LO_FLAGS_USE_AOPS;
789 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
790 lo_flags |= LO_FLAGS_READ_ONLY;
792 lo_blocksize = S_ISBLK(inode->i_mode) ?
793 inode->i_bdev->bd_block_size : PAGE_SIZE;
800 size = get_loop_size(lo, file);
802 if ((loff_t)(sector_t)size != size) {
807 if (!(mode & FMODE_WRITE))
808 lo_flags |= LO_FLAGS_READ_ONLY;
810 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
812 lo->lo_blocksize = lo_blocksize;
813 lo->lo_device = bdev;
814 lo->lo_flags = lo_flags;
815 lo->lo_backing_file = file;
816 lo->transfer = transfer_none;
818 lo->lo_sizelimit = 0;
819 lo->old_gfp_mask = mapping_gfp_mask(mapping);
820 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
822 bio_list_init(&lo->lo_bio_list);
825 * set queue make_request_fn, and add limits based on lower level
828 blk_queue_make_request(lo->lo_queue, loop_make_request);
829 lo->lo_queue->queuedata = lo;
830 lo->lo_queue->unplug_fn = loop_unplug;
832 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
833 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL);
835 set_capacity(lo->lo_disk, size);
836 bd_set_size(bdev, size << 9);
838 set_blocksize(bdev, lo_blocksize);
840 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
842 if (IS_ERR(lo->lo_thread)) {
843 error = PTR_ERR(lo->lo_thread);
846 lo->lo_state = Lo_bound;
847 wake_up_process(lo->lo_thread);
849 ioctl_by_bdev(bdev, BLKRRPART, 0);
853 lo->lo_thread = NULL;
854 lo->lo_device = NULL;
855 lo->lo_backing_file = NULL;
857 set_capacity(lo->lo_disk, 0);
858 invalidate_bdev(bdev);
859 bd_set_size(bdev, 0);
860 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
861 lo->lo_state = Lo_unbound;
865 /* This is safe: open() is still holding a reference. */
866 module_put(THIS_MODULE);
871 loop_release_xfer(struct loop_device *lo)
874 struct loop_func_table *xfer = lo->lo_encryption;
878 err = xfer->release(lo);
880 lo->lo_encryption = NULL;
881 module_put(xfer->owner);
887 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
888 const struct loop_info64 *i)
893 struct module *owner = xfer->owner;
895 if (!try_module_get(owner))
898 err = xfer->init(lo, i);
902 lo->lo_encryption = xfer;
907 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
909 struct file *filp = lo->lo_backing_file;
910 gfp_t gfp = lo->old_gfp_mask;
912 if (lo->lo_state != Lo_bound)
915 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
921 spin_lock_irq(&lo->lo_lock);
922 lo->lo_state = Lo_rundown;
923 spin_unlock_irq(&lo->lo_lock);
925 kthread_stop(lo->lo_thread);
927 lo->lo_queue->unplug_fn = NULL;
928 lo->lo_backing_file = NULL;
930 loop_release_xfer(lo);
933 lo->lo_device = NULL;
934 lo->lo_encryption = NULL;
936 lo->lo_sizelimit = 0;
937 lo->lo_encrypt_key_size = 0;
939 lo->lo_thread = NULL;
940 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
941 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
942 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
944 invalidate_bdev(bdev);
945 set_capacity(lo->lo_disk, 0);
947 bd_set_size(bdev, 0);
948 mapping_set_gfp_mask(filp->f_mapping, gfp);
949 lo->lo_state = Lo_unbound;
950 /* This is safe: open() is still holding a reference. */
951 module_put(THIS_MODULE);
953 ioctl_by_bdev(bdev, BLKRRPART, 0);
954 mutex_unlock(&lo->lo_ctl_mutex);
956 * Need not hold lo_ctl_mutex to fput backing file.
957 * Calling fput holding lo_ctl_mutex triggers a circular
958 * lock dependency possibility warning as fput can take
959 * bd_mutex which is usually taken before lo_ctl_mutex.
966 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
969 struct loop_func_table *xfer;
970 uid_t uid = current_uid();
972 if (lo->lo_encrypt_key_size &&
973 lo->lo_key_owner != uid &&
974 !capable(CAP_SYS_ADMIN))
976 if (lo->lo_state != Lo_bound)
978 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
981 err = loop_release_xfer(lo);
985 if (info->lo_encrypt_type) {
986 unsigned int type = info->lo_encrypt_type;
988 if (type >= MAX_LO_CRYPT)
990 xfer = xfer_funcs[type];
996 err = loop_init_xfer(lo, xfer, info);
1000 if (lo->lo_offset != info->lo_offset ||
1001 lo->lo_sizelimit != info->lo_sizelimit) {
1002 lo->lo_offset = info->lo_offset;
1003 lo->lo_sizelimit = info->lo_sizelimit;
1004 if (figure_loop_size(lo))
1008 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1009 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1010 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1011 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1015 lo->transfer = xfer->transfer;
1016 lo->ioctl = xfer->ioctl;
1018 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1019 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1020 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1022 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1023 lo->lo_init[0] = info->lo_init[0];
1024 lo->lo_init[1] = info->lo_init[1];
1025 if (info->lo_encrypt_key_size) {
1026 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1027 info->lo_encrypt_key_size);
1028 lo->lo_key_owner = uid;
1035 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1037 struct file *file = lo->lo_backing_file;
1041 if (lo->lo_state != Lo_bound)
1043 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1046 memset(info, 0, sizeof(*info));
1047 info->lo_number = lo->lo_number;
1048 info->lo_device = huge_encode_dev(stat.dev);
1049 info->lo_inode = stat.ino;
1050 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1051 info->lo_offset = lo->lo_offset;
1052 info->lo_sizelimit = lo->lo_sizelimit;
1053 info->lo_flags = lo->lo_flags;
1054 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1055 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1056 info->lo_encrypt_type =
1057 lo->lo_encryption ? lo->lo_encryption->number : 0;
1058 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1059 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1060 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1061 lo->lo_encrypt_key_size);
1067 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1069 memset(info64, 0, sizeof(*info64));
1070 info64->lo_number = info->lo_number;
1071 info64->lo_device = info->lo_device;
1072 info64->lo_inode = info->lo_inode;
1073 info64->lo_rdevice = info->lo_rdevice;
1074 info64->lo_offset = info->lo_offset;
1075 info64->lo_sizelimit = 0;
1076 info64->lo_encrypt_type = info->lo_encrypt_type;
1077 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1078 info64->lo_flags = info->lo_flags;
1079 info64->lo_init[0] = info->lo_init[0];
1080 info64->lo_init[1] = info->lo_init[1];
1081 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1082 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1084 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1085 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1089 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1091 memset(info, 0, sizeof(*info));
1092 info->lo_number = info64->lo_number;
1093 info->lo_device = info64->lo_device;
1094 info->lo_inode = info64->lo_inode;
1095 info->lo_rdevice = info64->lo_rdevice;
1096 info->lo_offset = info64->lo_offset;
1097 info->lo_encrypt_type = info64->lo_encrypt_type;
1098 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1099 info->lo_flags = info64->lo_flags;
1100 info->lo_init[0] = info64->lo_init[0];
1101 info->lo_init[1] = info64->lo_init[1];
1102 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1103 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1105 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1106 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1108 /* error in case values were truncated */
1109 if (info->lo_device != info64->lo_device ||
1110 info->lo_rdevice != info64->lo_rdevice ||
1111 info->lo_inode != info64->lo_inode ||
1112 info->lo_offset != info64->lo_offset)
1119 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1121 struct loop_info info;
1122 struct loop_info64 info64;
1124 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1126 loop_info64_from_old(&info, &info64);
1127 return loop_set_status(lo, &info64);
1131 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1133 struct loop_info64 info64;
1135 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1137 return loop_set_status(lo, &info64);
1141 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1142 struct loop_info info;
1143 struct loop_info64 info64;
1149 err = loop_get_status(lo, &info64);
1151 err = loop_info64_to_old(&info64, &info);
1152 if (!err && copy_to_user(arg, &info, sizeof(info)))
1159 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1160 struct loop_info64 info64;
1166 err = loop_get_status(lo, &info64);
1167 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1173 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1180 if (unlikely(lo->lo_state != Lo_bound))
1182 err = figure_loop_size(lo);
1185 sec = get_capacity(lo->lo_disk);
1186 /* the width of sector_t may be narrow for bit-shift */
1189 mutex_lock(&bdev->bd_mutex);
1190 bd_set_size(bdev, sz);
1191 mutex_unlock(&bdev->bd_mutex);
1197 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1198 unsigned int cmd, unsigned long arg)
1200 struct loop_device *lo = bdev->bd_disk->private_data;
1203 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1206 err = loop_set_fd(lo, mode, bdev, arg);
1208 case LOOP_CHANGE_FD:
1209 err = loop_change_fd(lo, bdev, arg);
1212 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1213 err = loop_clr_fd(lo, bdev);
1217 case LOOP_SET_STATUS:
1218 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1220 case LOOP_GET_STATUS:
1221 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1223 case LOOP_SET_STATUS64:
1224 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1226 case LOOP_GET_STATUS64:
1227 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1229 case LOOP_SET_CAPACITY:
1231 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1232 err = loop_set_capacity(lo, bdev);
1235 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1237 mutex_unlock(&lo->lo_ctl_mutex);
1243 #ifdef CONFIG_COMPAT
1244 struct compat_loop_info {
1245 compat_int_t lo_number; /* ioctl r/o */
1246 compat_dev_t lo_device; /* ioctl r/o */
1247 compat_ulong_t lo_inode; /* ioctl r/o */
1248 compat_dev_t lo_rdevice; /* ioctl r/o */
1249 compat_int_t lo_offset;
1250 compat_int_t lo_encrypt_type;
1251 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1252 compat_int_t lo_flags; /* ioctl r/o */
1253 char lo_name[LO_NAME_SIZE];
1254 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1255 compat_ulong_t lo_init[2];
1260 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1261 * - noinlined to reduce stack space usage in main part of driver
1264 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1265 struct loop_info64 *info64)
1267 struct compat_loop_info info;
1269 if (copy_from_user(&info, arg, sizeof(info)))
1272 memset(info64, 0, sizeof(*info64));
1273 info64->lo_number = info.lo_number;
1274 info64->lo_device = info.lo_device;
1275 info64->lo_inode = info.lo_inode;
1276 info64->lo_rdevice = info.lo_rdevice;
1277 info64->lo_offset = info.lo_offset;
1278 info64->lo_sizelimit = 0;
1279 info64->lo_encrypt_type = info.lo_encrypt_type;
1280 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1281 info64->lo_flags = info.lo_flags;
1282 info64->lo_init[0] = info.lo_init[0];
1283 info64->lo_init[1] = info.lo_init[1];
1284 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1285 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1287 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1288 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1293 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1294 * - noinlined to reduce stack space usage in main part of driver
1297 loop_info64_to_compat(const struct loop_info64 *info64,
1298 struct compat_loop_info __user *arg)
1300 struct compat_loop_info info;
1302 memset(&info, 0, sizeof(info));
1303 info.lo_number = info64->lo_number;
1304 info.lo_device = info64->lo_device;
1305 info.lo_inode = info64->lo_inode;
1306 info.lo_rdevice = info64->lo_rdevice;
1307 info.lo_offset = info64->lo_offset;
1308 info.lo_encrypt_type = info64->lo_encrypt_type;
1309 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1310 info.lo_flags = info64->lo_flags;
1311 info.lo_init[0] = info64->lo_init[0];
1312 info.lo_init[1] = info64->lo_init[1];
1313 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1314 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1316 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1317 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1319 /* error in case values were truncated */
1320 if (info.lo_device != info64->lo_device ||
1321 info.lo_rdevice != info64->lo_rdevice ||
1322 info.lo_inode != info64->lo_inode ||
1323 info.lo_offset != info64->lo_offset ||
1324 info.lo_init[0] != info64->lo_init[0] ||
1325 info.lo_init[1] != info64->lo_init[1])
1328 if (copy_to_user(arg, &info, sizeof(info)))
1334 loop_set_status_compat(struct loop_device *lo,
1335 const struct compat_loop_info __user *arg)
1337 struct loop_info64 info64;
1340 ret = loop_info64_from_compat(arg, &info64);
1343 return loop_set_status(lo, &info64);
1347 loop_get_status_compat(struct loop_device *lo,
1348 struct compat_loop_info __user *arg)
1350 struct loop_info64 info64;
1356 err = loop_get_status(lo, &info64);
1358 err = loop_info64_to_compat(&info64, arg);
1362 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1363 unsigned int cmd, unsigned long arg)
1365 struct loop_device *lo = bdev->bd_disk->private_data;
1369 case LOOP_SET_STATUS:
1370 mutex_lock(&lo->lo_ctl_mutex);
1371 err = loop_set_status_compat(
1372 lo, (const struct compat_loop_info __user *) arg);
1373 mutex_unlock(&lo->lo_ctl_mutex);
1375 case LOOP_GET_STATUS:
1376 mutex_lock(&lo->lo_ctl_mutex);
1377 err = loop_get_status_compat(
1378 lo, (struct compat_loop_info __user *) arg);
1379 mutex_unlock(&lo->lo_ctl_mutex);
1381 case LOOP_SET_CAPACITY:
1383 case LOOP_GET_STATUS64:
1384 case LOOP_SET_STATUS64:
1385 arg = (unsigned long) compat_ptr(arg);
1387 case LOOP_CHANGE_FD:
1388 err = lo_ioctl(bdev, mode, cmd, arg);
1398 static int lo_open(struct block_device *bdev, fmode_t mode)
1400 struct loop_device *lo = bdev->bd_disk->private_data;
1402 mutex_lock(&lo->lo_ctl_mutex);
1404 mutex_unlock(&lo->lo_ctl_mutex);
1409 static int lo_release(struct gendisk *disk, fmode_t mode)
1411 struct loop_device *lo = disk->private_data;
1414 mutex_lock(&lo->lo_ctl_mutex);
1416 if (--lo->lo_refcnt)
1419 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1421 * In autoclear mode, stop the loop thread
1422 * and remove configuration after last close.
1424 err = loop_clr_fd(lo, NULL);
1429 * Otherwise keep thread (if running) and config,
1430 * but flush possible ongoing bios in thread.
1436 mutex_unlock(&lo->lo_ctl_mutex);
1441 static struct block_device_operations lo_fops = {
1442 .owner = THIS_MODULE,
1444 .release = lo_release,
1446 #ifdef CONFIG_COMPAT
1447 .compat_ioctl = lo_compat_ioctl,
1452 * And now the modules code and kernel interface.
1454 static int max_loop;
1455 module_param(max_loop, int, 0);
1456 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1457 module_param(max_part, int, 0);
1458 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1459 MODULE_LICENSE("GPL");
1460 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1462 int loop_register_transfer(struct loop_func_table *funcs)
1464 unsigned int n = funcs->number;
1466 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1468 xfer_funcs[n] = funcs;
1472 int loop_unregister_transfer(int number)
1474 unsigned int n = number;
1475 struct loop_device *lo;
1476 struct loop_func_table *xfer;
1478 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1481 xfer_funcs[n] = NULL;
1483 list_for_each_entry(lo, &loop_devices, lo_list) {
1484 mutex_lock(&lo->lo_ctl_mutex);
1486 if (lo->lo_encryption == xfer)
1487 loop_release_xfer(lo);
1489 mutex_unlock(&lo->lo_ctl_mutex);
1495 EXPORT_SYMBOL(loop_register_transfer);
1496 EXPORT_SYMBOL(loop_unregister_transfer);
1498 static struct loop_device *loop_alloc(int i)
1500 struct loop_device *lo;
1501 struct gendisk *disk;
1503 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1507 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1511 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1513 goto out_free_queue;
1515 mutex_init(&lo->lo_ctl_mutex);
1517 lo->lo_thread = NULL;
1518 init_waitqueue_head(&lo->lo_event);
1519 spin_lock_init(&lo->lo_lock);
1520 disk->major = LOOP_MAJOR;
1521 disk->first_minor = i << part_shift;
1522 disk->fops = &lo_fops;
1523 disk->private_data = lo;
1524 disk->queue = lo->lo_queue;
1525 sprintf(disk->disk_name, "loop%d", i);
1529 blk_cleanup_queue(lo->lo_queue);
1536 static void loop_free(struct loop_device *lo)
1538 blk_cleanup_queue(lo->lo_queue);
1539 put_disk(lo->lo_disk);
1540 list_del(&lo->lo_list);
1544 static struct loop_device *loop_init_one(int i)
1546 struct loop_device *lo;
1548 list_for_each_entry(lo, &loop_devices, lo_list) {
1549 if (lo->lo_number == i)
1555 add_disk(lo->lo_disk);
1556 list_add_tail(&lo->lo_list, &loop_devices);
1561 static void loop_del_one(struct loop_device *lo)
1563 del_gendisk(lo->lo_disk);
1567 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1569 struct loop_device *lo;
1570 struct kobject *kobj;
1572 mutex_lock(&loop_devices_mutex);
1573 lo = loop_init_one(dev & MINORMASK);
1574 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1575 mutex_unlock(&loop_devices_mutex);
1581 static int __init loop_init(void)
1584 unsigned long range;
1585 struct loop_device *lo, *next;
1588 * loop module now has a feature to instantiate underlying device
1589 * structure on-demand, provided that there is an access dev node.
1590 * However, this will not work well with user space tool that doesn't
1591 * know about such "feature". In order to not break any existing
1592 * tool, we do the following:
1594 * (1) if max_loop is specified, create that many upfront, and this
1595 * also becomes a hard limit.
1596 * (2) if max_loop is not specified, create 8 loop device on module
1597 * load, user can further extend loop device by create dev node
1598 * themselves and have kernel automatically instantiate actual
1604 part_shift = fls(max_part);
1606 if (max_loop > 1UL << (MINORBITS - part_shift))
1614 range = 1UL << (MINORBITS - part_shift);
1617 if (register_blkdev(LOOP_MAJOR, "loop"))
1620 for (i = 0; i < nr; i++) {
1624 list_add_tail(&lo->lo_list, &loop_devices);
1627 /* point of no return */
1629 list_for_each_entry(lo, &loop_devices, lo_list)
1630 add_disk(lo->lo_disk);
1632 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1633 THIS_MODULE, loop_probe, NULL, NULL);
1635 printk(KERN_INFO "loop: module loaded\n");
1639 printk(KERN_INFO "loop: out of memory\n");
1641 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1644 unregister_blkdev(LOOP_MAJOR, "loop");
1648 static void __exit loop_exit(void)
1650 unsigned long range;
1651 struct loop_device *lo, *next;
1653 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1655 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1658 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1659 unregister_blkdev(LOOP_MAJOR, "loop");
1662 module_init(loop_init);
1663 module_exit(loop_exit);
1666 static int __init max_loop_setup(char *str)
1668 max_loop = simple_strtol(str, NULL, 0);
1672 __setup("max_loop=", max_loop_setup);