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 file_update_time(file);
243 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
244 bvec->bv_page, bv_offs, size, IV);
246 if (unlikely(transfer_result))
249 ret = pagecache_write_end(file, mapping, pos, size, copied,
251 if (ret < 0 || ret != copied)
254 if (unlikely(transfer_result))
265 mutex_unlock(&mapping->host->i_mutex);
273 * __do_lo_send_write - helper for writing data to a loop device
275 * This helper just factors out common code between do_lo_send_direct_write()
276 * and do_lo_send_write().
278 static int __do_lo_send_write(struct file *file,
279 u8 *buf, const int len, loff_t pos)
282 mm_segment_t old_fs = get_fs();
285 bw = file->f_op->write(file, buf, len, &pos);
287 if (likely(bw == len))
289 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
290 (unsigned long long)pos, len);
297 * do_lo_send_direct_write - helper for writing data to a loop device
299 * This is the fast, non-transforming version for backing filesystems which do
300 * not implement the address space operations write_begin and write_end.
301 * It uses the write file operation which should be present on all writeable
304 static int do_lo_send_direct_write(struct loop_device *lo,
305 struct bio_vec *bvec, loff_t pos, struct page *page)
307 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
308 kmap(bvec->bv_page) + bvec->bv_offset,
310 kunmap(bvec->bv_page);
316 * do_lo_send_write - helper for writing data to a loop device
318 * This is the slow, transforming version for filesystems which do not
319 * implement the address space operations write_begin and write_end. It
320 * uses the write file operation which should be present on all writeable
323 * Using fops->write is slower than using aops->{prepare,commit}_write in the
324 * transforming case because we need to double buffer the data as we cannot do
325 * the transformations in place as we do not have direct access to the
326 * destination pages of the backing file.
328 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
329 loff_t pos, struct page *page)
331 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
332 bvec->bv_offset, bvec->bv_len, pos >> 9);
334 return __do_lo_send_write(lo->lo_backing_file,
335 page_address(page), bvec->bv_len,
337 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
338 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
344 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
346 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
348 struct bio_vec *bvec;
349 struct page *page = NULL;
352 do_lo_send = do_lo_send_aops;
353 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
354 do_lo_send = do_lo_send_direct_write;
355 if (lo->transfer != transfer_none) {
356 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
360 do_lo_send = do_lo_send_write;
363 bio_for_each_segment(bvec, bio, i) {
364 ret = do_lo_send(lo, bvec, pos, page);
376 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
381 struct lo_read_data {
382 struct loop_device *lo;
389 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
390 struct splice_desc *sd)
392 struct lo_read_data *p = sd->u.data;
393 struct loop_device *lo = p->lo;
394 struct page *page = buf->page;
398 ret = buf->ops->confirm(pipe, buf);
402 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
408 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
409 printk(KERN_ERR "loop: transfer error block %ld\n",
414 flush_dcache_page(p->page);
423 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
425 return __splice_from_pipe(pipe, sd, lo_splice_actor);
429 do_lo_receive(struct loop_device *lo,
430 struct bio_vec *bvec, int bsize, loff_t pos)
432 struct lo_read_data cookie;
433 struct splice_desc sd;
438 cookie.page = bvec->bv_page;
439 cookie.offset = bvec->bv_offset;
440 cookie.bsize = bsize;
443 sd.total_len = bvec->bv_len;
448 file = lo->lo_backing_file;
449 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
458 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
460 struct bio_vec *bvec;
463 bio_for_each_segment(bvec, bio, i) {
464 ret = do_lo_receive(lo, bvec, bsize, pos);
472 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
477 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
479 if (bio_rw(bio) == WRITE) {
480 bool barrier = bio_rw_flagged(bio, BIO_RW_BARRIER);
481 struct file *file = lo->lo_backing_file;
484 if (unlikely(!file->f_op->fsync)) {
489 ret = vfs_fsync(file, file->f_path.dentry, 0);
496 ret = lo_send(lo, bio, pos);
498 if (barrier && !ret) {
499 ret = vfs_fsync(file, file->f_path.dentry, 0);
504 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
511 * Add bio to back of pending list
513 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
515 bio_list_add(&lo->lo_bio_list, bio);
519 * Grab first pending buffer
521 static struct bio *loop_get_bio(struct loop_device *lo)
523 return bio_list_pop(&lo->lo_bio_list);
526 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
528 struct loop_device *lo = q->queuedata;
529 int rw = bio_rw(old_bio);
534 BUG_ON(!lo || (rw != READ && rw != WRITE));
536 spin_lock_irq(&lo->lo_lock);
537 if (lo->lo_state != Lo_bound)
539 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
541 loop_add_bio(lo, old_bio);
542 wake_up(&lo->lo_event);
543 spin_unlock_irq(&lo->lo_lock);
547 spin_unlock_irq(&lo->lo_lock);
548 bio_io_error(old_bio);
553 * kick off io on the underlying address space
555 static void loop_unplug(struct request_queue *q)
557 struct loop_device *lo = q->queuedata;
559 queue_flag_clear_unlocked(QUEUE_FLAG_PLUGGED, q);
560 blk_run_address_space(lo->lo_backing_file->f_mapping);
563 struct switch_request {
565 struct completion wait;
568 static void do_loop_switch(struct loop_device *, struct switch_request *);
570 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
572 if (unlikely(!bio->bi_bdev)) {
573 do_loop_switch(lo, bio->bi_private);
576 int ret = do_bio_filebacked(lo, bio);
582 * worker thread that handles reads/writes to file backed loop devices,
583 * to avoid blocking in our make_request_fn. it also does loop decrypting
584 * on reads for block backed loop, as that is too heavy to do from
585 * b_end_io context where irqs may be disabled.
587 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
588 * calling kthread_stop(). Therefore once kthread_should_stop() is
589 * true, make_request will not place any more requests. Therefore
590 * once kthread_should_stop() is true and lo_bio is NULL, we are
591 * done with the loop.
593 static int loop_thread(void *data)
595 struct loop_device *lo = data;
598 set_user_nice(current, -20);
600 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
602 wait_event_interruptible(lo->lo_event,
603 !bio_list_empty(&lo->lo_bio_list) ||
604 kthread_should_stop());
606 if (bio_list_empty(&lo->lo_bio_list))
608 spin_lock_irq(&lo->lo_lock);
609 bio = loop_get_bio(lo);
610 spin_unlock_irq(&lo->lo_lock);
613 loop_handle_bio(lo, bio);
620 * loop_switch performs the hard work of switching a backing store.
621 * First it needs to flush existing IO, it does this by sending a magic
622 * BIO down the pipe. The completion of this BIO does the actual switch.
624 static int loop_switch(struct loop_device *lo, struct file *file)
626 struct switch_request w;
627 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
630 init_completion(&w.wait);
632 bio->bi_private = &w;
634 loop_make_request(lo->lo_queue, bio);
635 wait_for_completion(&w.wait);
640 * Helper to flush the IOs in loop, but keeping loop thread running
642 static int loop_flush(struct loop_device *lo)
644 /* loop not yet configured, no running thread, nothing to flush */
648 return loop_switch(lo, NULL);
652 * Do the actual switch; called from the BIO completion routine
654 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
656 struct file *file = p->file;
657 struct file *old_file = lo->lo_backing_file;
658 struct address_space *mapping;
660 /* if no new file, only flush of queued bios requested */
664 mapping = file->f_mapping;
665 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
666 lo->lo_backing_file = file;
667 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
668 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
669 lo->old_gfp_mask = mapping_gfp_mask(mapping);
670 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
677 * loop_change_fd switched the backing store of a loopback device to
678 * a new file. This is useful for operating system installers to free up
679 * the original file and in High Availability environments to switch to
680 * an alternative location for the content in case of server meltdown.
681 * This can only work if the loop device is used read-only, and if the
682 * new backing store is the same size and type as the old backing store.
684 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
687 struct file *file, *old_file;
692 if (lo->lo_state != Lo_bound)
695 /* the loop device has to be read-only */
697 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
705 inode = file->f_mapping->host;
706 old_file = lo->lo_backing_file;
710 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
713 /* size of the new backing store needs to be the same */
714 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
718 error = loop_switch(lo, file);
724 ioctl_by_bdev(bdev, BLKRRPART, 0);
733 static inline int is_loop_device(struct file *file)
735 struct inode *i = file->f_mapping->host;
737 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
740 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
741 struct block_device *bdev, unsigned int arg)
743 struct file *file, *f;
745 struct address_space *mapping;
746 unsigned lo_blocksize;
751 /* This is safe, since we have a reference from open(). */
752 __module_get(THIS_MODULE);
760 if (lo->lo_state != Lo_unbound)
763 /* Avoid recursion */
765 while (is_loop_device(f)) {
766 struct loop_device *l;
768 if (f->f_mapping->host->i_bdev == bdev)
771 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
772 if (l->lo_state == Lo_unbound) {
776 f = l->lo_backing_file;
779 mapping = file->f_mapping;
780 inode = mapping->host;
782 if (!(file->f_mode & FMODE_WRITE))
783 lo_flags |= LO_FLAGS_READ_ONLY;
786 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
787 const struct address_space_operations *aops = mapping->a_ops;
789 if (aops->write_begin)
790 lo_flags |= LO_FLAGS_USE_AOPS;
791 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
792 lo_flags |= LO_FLAGS_READ_ONLY;
794 lo_blocksize = S_ISBLK(inode->i_mode) ?
795 inode->i_bdev->bd_block_size : PAGE_SIZE;
802 size = get_loop_size(lo, file);
804 if ((loff_t)(sector_t)size != size) {
809 if (!(mode & FMODE_WRITE))
810 lo_flags |= LO_FLAGS_READ_ONLY;
812 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
814 lo->lo_blocksize = lo_blocksize;
815 lo->lo_device = bdev;
816 lo->lo_flags = lo_flags;
817 lo->lo_backing_file = file;
818 lo->transfer = transfer_none;
820 lo->lo_sizelimit = 0;
821 lo->old_gfp_mask = mapping_gfp_mask(mapping);
822 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
824 bio_list_init(&lo->lo_bio_list);
827 * set queue make_request_fn, and add limits based on lower level
830 blk_queue_make_request(lo->lo_queue, loop_make_request);
831 lo->lo_queue->queuedata = lo;
832 lo->lo_queue->unplug_fn = loop_unplug;
834 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
835 blk_queue_ordered(lo->lo_queue, QUEUE_ORDERED_DRAIN, NULL);
837 set_capacity(lo->lo_disk, size);
838 bd_set_size(bdev, size << 9);
840 set_blocksize(bdev, lo_blocksize);
842 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
844 if (IS_ERR(lo->lo_thread)) {
845 error = PTR_ERR(lo->lo_thread);
848 lo->lo_state = Lo_bound;
849 wake_up_process(lo->lo_thread);
851 ioctl_by_bdev(bdev, BLKRRPART, 0);
855 lo->lo_thread = NULL;
856 lo->lo_device = NULL;
857 lo->lo_backing_file = NULL;
859 set_capacity(lo->lo_disk, 0);
860 invalidate_bdev(bdev);
861 bd_set_size(bdev, 0);
862 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
863 lo->lo_state = Lo_unbound;
867 /* This is safe: open() is still holding a reference. */
868 module_put(THIS_MODULE);
873 loop_release_xfer(struct loop_device *lo)
876 struct loop_func_table *xfer = lo->lo_encryption;
880 err = xfer->release(lo);
882 lo->lo_encryption = NULL;
883 module_put(xfer->owner);
889 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
890 const struct loop_info64 *i)
895 struct module *owner = xfer->owner;
897 if (!try_module_get(owner))
900 err = xfer->init(lo, i);
904 lo->lo_encryption = xfer;
909 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
911 struct file *filp = lo->lo_backing_file;
912 gfp_t gfp = lo->old_gfp_mask;
914 if (lo->lo_state != Lo_bound)
917 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
923 spin_lock_irq(&lo->lo_lock);
924 lo->lo_state = Lo_rundown;
925 spin_unlock_irq(&lo->lo_lock);
927 kthread_stop(lo->lo_thread);
929 lo->lo_queue->unplug_fn = NULL;
930 lo->lo_backing_file = NULL;
932 loop_release_xfer(lo);
935 lo->lo_device = NULL;
936 lo->lo_encryption = NULL;
938 lo->lo_sizelimit = 0;
939 lo->lo_encrypt_key_size = 0;
941 lo->lo_thread = NULL;
942 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
943 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
944 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
946 invalidate_bdev(bdev);
947 set_capacity(lo->lo_disk, 0);
949 bd_set_size(bdev, 0);
950 mapping_set_gfp_mask(filp->f_mapping, gfp);
951 lo->lo_state = Lo_unbound;
952 /* This is safe: open() is still holding a reference. */
953 module_put(THIS_MODULE);
954 if (max_part > 0 && bdev)
955 ioctl_by_bdev(bdev, BLKRRPART, 0);
956 mutex_unlock(&lo->lo_ctl_mutex);
958 * Need not hold lo_ctl_mutex to fput backing file.
959 * Calling fput holding lo_ctl_mutex triggers a circular
960 * lock dependency possibility warning as fput can take
961 * bd_mutex which is usually taken before lo_ctl_mutex.
968 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
971 struct loop_func_table *xfer;
972 uid_t uid = current_uid();
974 if (lo->lo_encrypt_key_size &&
975 lo->lo_key_owner != uid &&
976 !capable(CAP_SYS_ADMIN))
978 if (lo->lo_state != Lo_bound)
980 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
983 err = loop_release_xfer(lo);
987 if (info->lo_encrypt_type) {
988 unsigned int type = info->lo_encrypt_type;
990 if (type >= MAX_LO_CRYPT)
992 xfer = xfer_funcs[type];
998 err = loop_init_xfer(lo, xfer, info);
1002 if (lo->lo_offset != info->lo_offset ||
1003 lo->lo_sizelimit != info->lo_sizelimit) {
1004 lo->lo_offset = info->lo_offset;
1005 lo->lo_sizelimit = info->lo_sizelimit;
1006 if (figure_loop_size(lo))
1010 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1011 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1012 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1013 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1017 lo->transfer = xfer->transfer;
1018 lo->ioctl = xfer->ioctl;
1020 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1021 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1022 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1024 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1025 lo->lo_init[0] = info->lo_init[0];
1026 lo->lo_init[1] = info->lo_init[1];
1027 if (info->lo_encrypt_key_size) {
1028 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1029 info->lo_encrypt_key_size);
1030 lo->lo_key_owner = uid;
1037 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1039 struct file *file = lo->lo_backing_file;
1043 if (lo->lo_state != Lo_bound)
1045 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1048 memset(info, 0, sizeof(*info));
1049 info->lo_number = lo->lo_number;
1050 info->lo_device = huge_encode_dev(stat.dev);
1051 info->lo_inode = stat.ino;
1052 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1053 info->lo_offset = lo->lo_offset;
1054 info->lo_sizelimit = lo->lo_sizelimit;
1055 info->lo_flags = lo->lo_flags;
1056 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1057 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1058 info->lo_encrypt_type =
1059 lo->lo_encryption ? lo->lo_encryption->number : 0;
1060 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1061 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1062 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1063 lo->lo_encrypt_key_size);
1069 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1071 memset(info64, 0, sizeof(*info64));
1072 info64->lo_number = info->lo_number;
1073 info64->lo_device = info->lo_device;
1074 info64->lo_inode = info->lo_inode;
1075 info64->lo_rdevice = info->lo_rdevice;
1076 info64->lo_offset = info->lo_offset;
1077 info64->lo_sizelimit = 0;
1078 info64->lo_encrypt_type = info->lo_encrypt_type;
1079 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1080 info64->lo_flags = info->lo_flags;
1081 info64->lo_init[0] = info->lo_init[0];
1082 info64->lo_init[1] = info->lo_init[1];
1083 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1084 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1086 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1087 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1091 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1093 memset(info, 0, sizeof(*info));
1094 info->lo_number = info64->lo_number;
1095 info->lo_device = info64->lo_device;
1096 info->lo_inode = info64->lo_inode;
1097 info->lo_rdevice = info64->lo_rdevice;
1098 info->lo_offset = info64->lo_offset;
1099 info->lo_encrypt_type = info64->lo_encrypt_type;
1100 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1101 info->lo_flags = info64->lo_flags;
1102 info->lo_init[0] = info64->lo_init[0];
1103 info->lo_init[1] = info64->lo_init[1];
1104 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1105 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1107 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1108 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1110 /* error in case values were truncated */
1111 if (info->lo_device != info64->lo_device ||
1112 info->lo_rdevice != info64->lo_rdevice ||
1113 info->lo_inode != info64->lo_inode ||
1114 info->lo_offset != info64->lo_offset)
1121 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1123 struct loop_info info;
1124 struct loop_info64 info64;
1126 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1128 loop_info64_from_old(&info, &info64);
1129 return loop_set_status(lo, &info64);
1133 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1135 struct loop_info64 info64;
1137 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1139 return loop_set_status(lo, &info64);
1143 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1144 struct loop_info info;
1145 struct loop_info64 info64;
1151 err = loop_get_status(lo, &info64);
1153 err = loop_info64_to_old(&info64, &info);
1154 if (!err && copy_to_user(arg, &info, sizeof(info)))
1161 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1162 struct loop_info64 info64;
1168 err = loop_get_status(lo, &info64);
1169 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1175 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1182 if (unlikely(lo->lo_state != Lo_bound))
1184 err = figure_loop_size(lo);
1187 sec = get_capacity(lo->lo_disk);
1188 /* the width of sector_t may be narrow for bit-shift */
1191 mutex_lock(&bdev->bd_mutex);
1192 bd_set_size(bdev, sz);
1193 mutex_unlock(&bdev->bd_mutex);
1199 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1200 unsigned int cmd, unsigned long arg)
1202 struct loop_device *lo = bdev->bd_disk->private_data;
1205 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1208 err = loop_set_fd(lo, mode, bdev, arg);
1210 case LOOP_CHANGE_FD:
1211 err = loop_change_fd(lo, bdev, arg);
1214 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1215 err = loop_clr_fd(lo, bdev);
1219 case LOOP_SET_STATUS:
1220 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1222 case LOOP_GET_STATUS:
1223 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1225 case LOOP_SET_STATUS64:
1226 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1228 case LOOP_GET_STATUS64:
1229 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1231 case LOOP_SET_CAPACITY:
1233 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1234 err = loop_set_capacity(lo, bdev);
1237 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1239 mutex_unlock(&lo->lo_ctl_mutex);
1245 #ifdef CONFIG_COMPAT
1246 struct compat_loop_info {
1247 compat_int_t lo_number; /* ioctl r/o */
1248 compat_dev_t lo_device; /* ioctl r/o */
1249 compat_ulong_t lo_inode; /* ioctl r/o */
1250 compat_dev_t lo_rdevice; /* ioctl r/o */
1251 compat_int_t lo_offset;
1252 compat_int_t lo_encrypt_type;
1253 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1254 compat_int_t lo_flags; /* ioctl r/o */
1255 char lo_name[LO_NAME_SIZE];
1256 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1257 compat_ulong_t lo_init[2];
1262 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1263 * - noinlined to reduce stack space usage in main part of driver
1266 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1267 struct loop_info64 *info64)
1269 struct compat_loop_info info;
1271 if (copy_from_user(&info, arg, sizeof(info)))
1274 memset(info64, 0, sizeof(*info64));
1275 info64->lo_number = info.lo_number;
1276 info64->lo_device = info.lo_device;
1277 info64->lo_inode = info.lo_inode;
1278 info64->lo_rdevice = info.lo_rdevice;
1279 info64->lo_offset = info.lo_offset;
1280 info64->lo_sizelimit = 0;
1281 info64->lo_encrypt_type = info.lo_encrypt_type;
1282 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1283 info64->lo_flags = info.lo_flags;
1284 info64->lo_init[0] = info.lo_init[0];
1285 info64->lo_init[1] = info.lo_init[1];
1286 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1287 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1289 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1290 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1295 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1296 * - noinlined to reduce stack space usage in main part of driver
1299 loop_info64_to_compat(const struct loop_info64 *info64,
1300 struct compat_loop_info __user *arg)
1302 struct compat_loop_info info;
1304 memset(&info, 0, sizeof(info));
1305 info.lo_number = info64->lo_number;
1306 info.lo_device = info64->lo_device;
1307 info.lo_inode = info64->lo_inode;
1308 info.lo_rdevice = info64->lo_rdevice;
1309 info.lo_offset = info64->lo_offset;
1310 info.lo_encrypt_type = info64->lo_encrypt_type;
1311 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1312 info.lo_flags = info64->lo_flags;
1313 info.lo_init[0] = info64->lo_init[0];
1314 info.lo_init[1] = info64->lo_init[1];
1315 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1316 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1318 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1319 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1321 /* error in case values were truncated */
1322 if (info.lo_device != info64->lo_device ||
1323 info.lo_rdevice != info64->lo_rdevice ||
1324 info.lo_inode != info64->lo_inode ||
1325 info.lo_offset != info64->lo_offset ||
1326 info.lo_init[0] != info64->lo_init[0] ||
1327 info.lo_init[1] != info64->lo_init[1])
1330 if (copy_to_user(arg, &info, sizeof(info)))
1336 loop_set_status_compat(struct loop_device *lo,
1337 const struct compat_loop_info __user *arg)
1339 struct loop_info64 info64;
1342 ret = loop_info64_from_compat(arg, &info64);
1345 return loop_set_status(lo, &info64);
1349 loop_get_status_compat(struct loop_device *lo,
1350 struct compat_loop_info __user *arg)
1352 struct loop_info64 info64;
1358 err = loop_get_status(lo, &info64);
1360 err = loop_info64_to_compat(&info64, arg);
1364 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1365 unsigned int cmd, unsigned long arg)
1367 struct loop_device *lo = bdev->bd_disk->private_data;
1371 case LOOP_SET_STATUS:
1372 mutex_lock(&lo->lo_ctl_mutex);
1373 err = loop_set_status_compat(
1374 lo, (const struct compat_loop_info __user *) arg);
1375 mutex_unlock(&lo->lo_ctl_mutex);
1377 case LOOP_GET_STATUS:
1378 mutex_lock(&lo->lo_ctl_mutex);
1379 err = loop_get_status_compat(
1380 lo, (struct compat_loop_info __user *) arg);
1381 mutex_unlock(&lo->lo_ctl_mutex);
1383 case LOOP_SET_CAPACITY:
1385 case LOOP_GET_STATUS64:
1386 case LOOP_SET_STATUS64:
1387 arg = (unsigned long) compat_ptr(arg);
1389 case LOOP_CHANGE_FD:
1390 err = lo_ioctl(bdev, mode, cmd, arg);
1400 static int lo_open(struct block_device *bdev, fmode_t mode)
1402 struct loop_device *lo = bdev->bd_disk->private_data;
1404 mutex_lock(&lo->lo_ctl_mutex);
1406 mutex_unlock(&lo->lo_ctl_mutex);
1411 static int lo_release(struct gendisk *disk, fmode_t mode)
1413 struct loop_device *lo = disk->private_data;
1416 mutex_lock(&lo->lo_ctl_mutex);
1418 if (--lo->lo_refcnt)
1421 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1423 * In autoclear mode, stop the loop thread
1424 * and remove configuration after last close.
1426 err = loop_clr_fd(lo, NULL);
1431 * Otherwise keep thread (if running) and config,
1432 * but flush possible ongoing bios in thread.
1438 mutex_unlock(&lo->lo_ctl_mutex);
1443 static const struct block_device_operations lo_fops = {
1444 .owner = THIS_MODULE,
1446 .release = lo_release,
1448 #ifdef CONFIG_COMPAT
1449 .compat_ioctl = lo_compat_ioctl,
1454 * And now the modules code and kernel interface.
1456 static int max_loop;
1457 module_param(max_loop, int, 0);
1458 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1459 module_param(max_part, int, 0);
1460 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1461 MODULE_LICENSE("GPL");
1462 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1464 int loop_register_transfer(struct loop_func_table *funcs)
1466 unsigned int n = funcs->number;
1468 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1470 xfer_funcs[n] = funcs;
1474 int loop_unregister_transfer(int number)
1476 unsigned int n = number;
1477 struct loop_device *lo;
1478 struct loop_func_table *xfer;
1480 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1483 xfer_funcs[n] = NULL;
1485 list_for_each_entry(lo, &loop_devices, lo_list) {
1486 mutex_lock(&lo->lo_ctl_mutex);
1488 if (lo->lo_encryption == xfer)
1489 loop_release_xfer(lo);
1491 mutex_unlock(&lo->lo_ctl_mutex);
1497 EXPORT_SYMBOL(loop_register_transfer);
1498 EXPORT_SYMBOL(loop_unregister_transfer);
1500 static struct loop_device *loop_alloc(int i)
1502 struct loop_device *lo;
1503 struct gendisk *disk;
1505 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1509 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1513 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1515 goto out_free_queue;
1517 mutex_init(&lo->lo_ctl_mutex);
1519 lo->lo_thread = NULL;
1520 init_waitqueue_head(&lo->lo_event);
1521 spin_lock_init(&lo->lo_lock);
1522 disk->major = LOOP_MAJOR;
1523 disk->first_minor = i << part_shift;
1524 disk->fops = &lo_fops;
1525 disk->private_data = lo;
1526 disk->queue = lo->lo_queue;
1527 sprintf(disk->disk_name, "loop%d", i);
1531 blk_cleanup_queue(lo->lo_queue);
1538 static void loop_free(struct loop_device *lo)
1540 blk_cleanup_queue(lo->lo_queue);
1541 put_disk(lo->lo_disk);
1542 list_del(&lo->lo_list);
1546 static struct loop_device *loop_init_one(int i)
1548 struct loop_device *lo;
1550 list_for_each_entry(lo, &loop_devices, lo_list) {
1551 if (lo->lo_number == i)
1557 add_disk(lo->lo_disk);
1558 list_add_tail(&lo->lo_list, &loop_devices);
1563 static void loop_del_one(struct loop_device *lo)
1565 del_gendisk(lo->lo_disk);
1569 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1571 struct loop_device *lo;
1572 struct kobject *kobj;
1574 mutex_lock(&loop_devices_mutex);
1575 lo = loop_init_one(dev & MINORMASK);
1576 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1577 mutex_unlock(&loop_devices_mutex);
1583 static int __init loop_init(void)
1586 unsigned long range;
1587 struct loop_device *lo, *next;
1590 * loop module now has a feature to instantiate underlying device
1591 * structure on-demand, provided that there is an access dev node.
1592 * However, this will not work well with user space tool that doesn't
1593 * know about such "feature". In order to not break any existing
1594 * tool, we do the following:
1596 * (1) if max_loop is specified, create that many upfront, and this
1597 * also becomes a hard limit.
1598 * (2) if max_loop is not specified, create 8 loop device on module
1599 * load, user can further extend loop device by create dev node
1600 * themselves and have kernel automatically instantiate actual
1606 part_shift = fls(max_part);
1608 if (max_loop > 1UL << (MINORBITS - part_shift))
1616 range = 1UL << (MINORBITS - part_shift);
1619 if (register_blkdev(LOOP_MAJOR, "loop"))
1622 for (i = 0; i < nr; i++) {
1626 list_add_tail(&lo->lo_list, &loop_devices);
1629 /* point of no return */
1631 list_for_each_entry(lo, &loop_devices, lo_list)
1632 add_disk(lo->lo_disk);
1634 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1635 THIS_MODULE, loop_probe, NULL, NULL);
1637 printk(KERN_INFO "loop: module loaded\n");
1641 printk(KERN_INFO "loop: out of memory\n");
1643 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1646 unregister_blkdev(LOOP_MAJOR, "loop");
1650 static void __exit loop_exit(void)
1652 unsigned long range;
1653 struct loop_device *lo, *next;
1655 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1657 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1660 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1661 unregister_blkdev(LOOP_MAJOR, "loop");
1664 module_init(loop_init);
1665 module_exit(loop_exit);
1668 static int __init max_loop_setup(char *str)
1670 max_loop = simple_strtol(str, NULL, 0);
1674 __setup("max_loop=", max_loop_setup);