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 prepare_write and/or commit_write are not available on the
45 * Anton Altaparmakov, 16 Feb 2005
48 * - Advisory locking is ignored here.
49 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
53 #include <linux/module.h>
54 #include <linux/moduleparam.h>
55 #include <linux/sched.h>
57 #include <linux/file.h>
58 #include <linux/stat.h>
59 #include <linux/errno.h>
60 #include <linux/major.h>
61 #include <linux/wait.h>
62 #include <linux/blkdev.h>
63 #include <linux/blkpg.h>
64 #include <linux/init.h>
65 #include <linux/smp_lock.h>
66 #include <linux/swap.h>
67 #include <linux/slab.h>
68 #include <linux/loop.h>
69 #include <linux/compat.h>
70 #include <linux/suspend.h>
71 #include <linux/writeback.h>
72 #include <linux/buffer_head.h> /* for invalidate_bdev() */
73 #include <linux/completion.h>
74 #include <linux/highmem.h>
75 #include <linux/gfp.h>
76 #include <linux/kthread.h>
78 #include <asm/uaccess.h>
80 static LIST_HEAD(loop_devices);
81 static DEFINE_MUTEX(loop_devices_mutex);
86 static int transfer_none(struct loop_device *lo, int cmd,
87 struct page *raw_page, unsigned raw_off,
88 struct page *loop_page, unsigned loop_off,
89 int size, sector_t real_block)
91 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
92 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
95 memcpy(loop_buf, raw_buf, size);
97 memcpy(raw_buf, loop_buf, size);
99 kunmap_atomic(raw_buf, KM_USER0);
100 kunmap_atomic(loop_buf, KM_USER1);
105 static int transfer_xor(struct loop_device *lo, int cmd,
106 struct page *raw_page, unsigned raw_off,
107 struct page *loop_page, unsigned loop_off,
108 int size, sector_t real_block)
110 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
111 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
112 char *in, *out, *key;
123 key = lo->lo_encrypt_key;
124 keysize = lo->lo_encrypt_key_size;
125 for (i = 0; i < size; i++)
126 *out++ = *in++ ^ key[(i & 511) % keysize];
128 kunmap_atomic(raw_buf, KM_USER0);
129 kunmap_atomic(loop_buf, KM_USER1);
134 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
136 if (unlikely(info->lo_encrypt_key_size <= 0))
141 static struct loop_func_table none_funcs = {
142 .number = LO_CRYPT_NONE,
143 .transfer = transfer_none,
146 static struct loop_func_table xor_funcs = {
147 .number = LO_CRYPT_XOR,
148 .transfer = transfer_xor,
152 /* xfer_funcs[0] is special - its release function is never called */
153 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
158 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
160 loff_t size, offset, loopsize;
162 /* Compute loopsize in bytes */
163 size = i_size_read(file->f_mapping->host);
164 offset = lo->lo_offset;
165 loopsize = size - offset;
166 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
167 loopsize = lo->lo_sizelimit;
170 * Unfortunately, if we want to do I/O on the device,
171 * the number of 512-byte sectors has to fit into a sector_t.
173 return loopsize >> 9;
177 figure_loop_size(struct loop_device *lo)
179 loff_t size = get_loop_size(lo, lo->lo_backing_file);
180 sector_t x = (sector_t)size;
182 if (unlikely((loff_t)x != size))
185 set_capacity(lo->lo_disk, x);
190 lo_do_transfer(struct loop_device *lo, int cmd,
191 struct page *rpage, unsigned roffs,
192 struct page *lpage, unsigned loffs,
193 int size, sector_t rblock)
195 if (unlikely(!lo->transfer))
198 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
202 * do_lo_send_aops - helper for writing data to a loop device
204 * This is the fast version for backing filesystems which implement the address
205 * space operations prepare_write and commit_write.
207 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
208 int bsize, loff_t pos, struct page *page)
210 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
211 struct address_space *mapping = file->f_mapping;
212 const struct address_space_operations *aops = mapping->a_ops;
214 unsigned offset, bv_offs;
217 mutex_lock(&mapping->host->i_mutex);
218 index = pos >> PAGE_CACHE_SHIFT;
219 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
220 bv_offs = bvec->bv_offset;
227 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
228 size = PAGE_CACHE_SIZE - offset;
231 page = grab_cache_page(mapping, index);
234 ret = aops->prepare_write(file, page, offset,
237 if (ret == AOP_TRUNCATED_PAGE) {
238 page_cache_release(page);
243 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
244 bvec->bv_page, bv_offs, size, IV);
245 if (unlikely(transfer_result)) {
247 * The transfer failed, but we still write the data to
248 * keep prepare/commit calls balanced.
250 printk(KERN_ERR "loop: transfer error block %llu\n",
251 (unsigned long long)index);
252 zero_user_page(page, offset, size, KM_USER0);
254 flush_dcache_page(page);
255 ret = aops->commit_write(file, page, offset,
258 if (ret == AOP_TRUNCATED_PAGE) {
259 page_cache_release(page);
264 if (unlikely(transfer_result))
272 page_cache_release(page);
276 mutex_unlock(&mapping->host->i_mutex);
280 page_cache_release(page);
287 * __do_lo_send_write - helper for writing data to a loop device
289 * This helper just factors out common code between do_lo_send_direct_write()
290 * and do_lo_send_write().
292 static int __do_lo_send_write(struct file *file,
293 u8 *buf, const int len, loff_t pos)
296 mm_segment_t old_fs = get_fs();
299 bw = file->f_op->write(file, buf, len, &pos);
301 if (likely(bw == len))
303 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
304 (unsigned long long)pos, len);
311 * do_lo_send_direct_write - helper for writing data to a loop device
313 * This is the fast, non-transforming version for backing filesystems which do
314 * not implement the address space operations prepare_write and commit_write.
315 * It uses the write file operation which should be present on all writeable
318 static int do_lo_send_direct_write(struct loop_device *lo,
319 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
321 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
322 kmap(bvec->bv_page) + bvec->bv_offset,
324 kunmap(bvec->bv_page);
330 * do_lo_send_write - helper for writing data to a loop device
332 * This is the slow, transforming version for filesystems which do not
333 * implement the address space operations prepare_write and commit_write. It
334 * uses the write file operation which should be present on all writeable
337 * Using fops->write is slower than using aops->{prepare,commit}_write in the
338 * transforming case because we need to double buffer the data as we cannot do
339 * the transformations in place as we do not have direct access to the
340 * destination pages of the backing file.
342 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
343 int bsize, loff_t pos, struct page *page)
345 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
346 bvec->bv_offset, bvec->bv_len, pos >> 9);
348 return __do_lo_send_write(lo->lo_backing_file,
349 page_address(page), bvec->bv_len,
351 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
352 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
358 static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
361 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
363 struct bio_vec *bvec;
364 struct page *page = NULL;
367 do_lo_send = do_lo_send_aops;
368 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
369 do_lo_send = do_lo_send_direct_write;
370 if (lo->transfer != transfer_none) {
371 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
375 do_lo_send = do_lo_send_write;
378 bio_for_each_segment(bvec, bio, i) {
379 ret = do_lo_send(lo, bvec, bsize, pos, page);
391 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
396 struct lo_read_data {
397 struct loop_device *lo;
404 lo_read_actor(read_descriptor_t *desc, struct page *page,
405 unsigned long offset, unsigned long size)
407 unsigned long count = desc->count;
408 struct lo_read_data *p = desc->arg.data;
409 struct loop_device *lo = p->lo;
412 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
417 if (lo_do_transfer(lo, READ, page, offset, p->page, p->offset, size, IV)) {
419 printk(KERN_ERR "loop: transfer error block %ld\n",
421 desc->error = -EINVAL;
424 flush_dcache_page(p->page);
426 desc->count = count - size;
427 desc->written += size;
433 do_lo_receive(struct loop_device *lo,
434 struct bio_vec *bvec, int bsize, loff_t pos)
436 struct lo_read_data cookie;
441 cookie.page = bvec->bv_page;
442 cookie.offset = bvec->bv_offset;
443 cookie.bsize = bsize;
444 file = lo->lo_backing_file;
445 retval = file->f_op->sendfile(file, &pos, bvec->bv_len,
446 lo_read_actor, &cookie);
447 return (retval < 0)? retval: 0;
451 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
453 struct bio_vec *bvec;
456 bio_for_each_segment(bvec, bio, i) {
457 ret = do_lo_receive(lo, bvec, bsize, pos);
465 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
470 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
471 if (bio_rw(bio) == WRITE)
472 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
474 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
479 * Add bio to back of pending list
481 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
483 if (lo->lo_biotail) {
484 lo->lo_biotail->bi_next = bio;
485 lo->lo_biotail = bio;
487 lo->lo_bio = lo->lo_biotail = bio;
491 * Grab first pending buffer
493 static struct bio *loop_get_bio(struct loop_device *lo)
497 if ((bio = lo->lo_bio)) {
498 if (bio == lo->lo_biotail)
499 lo->lo_biotail = NULL;
500 lo->lo_bio = bio->bi_next;
507 static int loop_make_request(request_queue_t *q, struct bio *old_bio)
509 struct loop_device *lo = q->queuedata;
510 int rw = bio_rw(old_bio);
515 BUG_ON(!lo || (rw != READ && rw != WRITE));
517 spin_lock_irq(&lo->lo_lock);
518 if (lo->lo_state != Lo_bound)
520 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
522 loop_add_bio(lo, old_bio);
523 wake_up(&lo->lo_event);
524 spin_unlock_irq(&lo->lo_lock);
528 spin_unlock_irq(&lo->lo_lock);
529 bio_io_error(old_bio, old_bio->bi_size);
534 * kick off io on the underlying address space
536 static void loop_unplug(request_queue_t *q)
538 struct loop_device *lo = q->queuedata;
540 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
541 blk_run_address_space(lo->lo_backing_file->f_mapping);
544 struct switch_request {
546 struct completion wait;
549 static void do_loop_switch(struct loop_device *, struct switch_request *);
551 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
553 if (unlikely(!bio->bi_bdev)) {
554 do_loop_switch(lo, bio->bi_private);
557 int ret = do_bio_filebacked(lo, bio);
558 bio_endio(bio, bio->bi_size, ret);
563 * worker thread that handles reads/writes to file backed loop devices,
564 * to avoid blocking in our make_request_fn. it also does loop decrypting
565 * on reads for block backed loop, as that is too heavy to do from
566 * b_end_io context where irqs may be disabled.
568 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
569 * calling kthread_stop(). Therefore once kthread_should_stop() is
570 * true, make_request will not place any more requests. Therefore
571 * once kthread_should_stop() is true and lo_bio is NULL, we are
572 * done with the loop.
574 static int loop_thread(void *data)
576 struct loop_device *lo = data;
580 * loop can be used in an encrypted device,
581 * hence, it mustn't be stopped at all
582 * because it could be indirectly used during suspension
584 current->flags |= PF_NOFREEZE;
586 set_user_nice(current, -20);
588 while (!kthread_should_stop() || lo->lo_bio) {
590 wait_event_interruptible(lo->lo_event,
591 lo->lo_bio || kthread_should_stop());
595 spin_lock_irq(&lo->lo_lock);
596 bio = loop_get_bio(lo);
597 spin_unlock_irq(&lo->lo_lock);
600 loop_handle_bio(lo, bio);
607 * loop_switch performs the hard work of switching a backing store.
608 * First it needs to flush existing IO, it does this by sending a magic
609 * BIO down the pipe. The completion of this BIO does the actual switch.
611 static int loop_switch(struct loop_device *lo, struct file *file)
613 struct switch_request w;
614 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
617 init_completion(&w.wait);
619 bio->bi_private = &w;
621 loop_make_request(lo->lo_queue, bio);
622 wait_for_completion(&w.wait);
627 * Do the actual switch; called from the BIO completion routine
629 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
631 struct file *file = p->file;
632 struct file *old_file = lo->lo_backing_file;
633 struct address_space *mapping = file->f_mapping;
635 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
636 lo->lo_backing_file = file;
637 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
638 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
639 lo->old_gfp_mask = mapping_gfp_mask(mapping);
640 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
646 * loop_change_fd switched the backing store of a loopback device to
647 * a new file. This is useful for operating system installers to free up
648 * the original file and in High Availability environments to switch to
649 * an alternative location for the content in case of server meltdown.
650 * This can only work if the loop device is used read-only, and if the
651 * new backing store is the same size and type as the old backing store.
653 static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
654 struct block_device *bdev, unsigned int arg)
656 struct file *file, *old_file;
661 if (lo->lo_state != Lo_bound)
664 /* the loop device has to be read-only */
666 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
674 inode = file->f_mapping->host;
675 old_file = lo->lo_backing_file;
679 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
682 /* new backing store needs to support loop (eg sendfile) */
683 if (!inode->i_fop->sendfile)
686 /* size of the new backing store needs to be the same */
687 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
691 error = loop_switch(lo, file);
704 static inline int is_loop_device(struct file *file)
706 struct inode *i = file->f_mapping->host;
708 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
711 static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
712 struct block_device *bdev, unsigned int arg)
714 struct file *file, *f;
716 struct address_space *mapping;
717 unsigned lo_blocksize;
722 /* This is safe, since we have a reference from open(). */
723 __module_get(THIS_MODULE);
731 if (lo->lo_state != Lo_unbound)
734 /* Avoid recursion */
736 while (is_loop_device(f)) {
737 struct loop_device *l;
739 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
742 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
743 if (l->lo_state == Lo_unbound) {
747 f = l->lo_backing_file;
750 mapping = file->f_mapping;
751 inode = mapping->host;
753 if (!(file->f_mode & FMODE_WRITE))
754 lo_flags |= LO_FLAGS_READ_ONLY;
757 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
758 const struct address_space_operations *aops = mapping->a_ops;
760 * If we can't read - sorry. If we only can't write - well,
761 * it's going to be read-only.
763 if (!file->f_op->sendfile)
765 if (aops->prepare_write && aops->commit_write)
766 lo_flags |= LO_FLAGS_USE_AOPS;
767 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
768 lo_flags |= LO_FLAGS_READ_ONLY;
770 lo_blocksize = S_ISBLK(inode->i_mode) ?
771 inode->i_bdev->bd_block_size : PAGE_SIZE;
778 size = get_loop_size(lo, file);
780 if ((loff_t)(sector_t)size != size) {
785 if (!(lo_file->f_mode & FMODE_WRITE))
786 lo_flags |= LO_FLAGS_READ_ONLY;
788 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
790 lo->lo_blocksize = lo_blocksize;
791 lo->lo_device = bdev;
792 lo->lo_flags = lo_flags;
793 lo->lo_backing_file = file;
794 lo->transfer = transfer_none;
796 lo->lo_sizelimit = 0;
797 lo->old_gfp_mask = mapping_gfp_mask(mapping);
798 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
800 lo->lo_bio = lo->lo_biotail = NULL;
803 * set queue make_request_fn, and add limits based on lower level
806 blk_queue_make_request(lo->lo_queue, loop_make_request);
807 lo->lo_queue->queuedata = lo;
808 lo->lo_queue->unplug_fn = loop_unplug;
810 set_capacity(lo->lo_disk, size);
811 bd_set_size(bdev, size << 9);
813 set_blocksize(bdev, lo_blocksize);
815 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
817 if (IS_ERR(lo->lo_thread)) {
818 error = PTR_ERR(lo->lo_thread);
821 lo->lo_state = Lo_bound;
822 wake_up_process(lo->lo_thread);
826 lo->lo_thread = NULL;
827 lo->lo_device = NULL;
828 lo->lo_backing_file = NULL;
830 set_capacity(lo->lo_disk, 0);
831 invalidate_bdev(bdev);
832 bd_set_size(bdev, 0);
833 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
834 lo->lo_state = Lo_unbound;
838 /* This is safe: open() is still holding a reference. */
839 module_put(THIS_MODULE);
844 loop_release_xfer(struct loop_device *lo)
847 struct loop_func_table *xfer = lo->lo_encryption;
851 err = xfer->release(lo);
853 lo->lo_encryption = NULL;
854 module_put(xfer->owner);
860 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
861 const struct loop_info64 *i)
866 struct module *owner = xfer->owner;
868 if (!try_module_get(owner))
871 err = xfer->init(lo, i);
875 lo->lo_encryption = xfer;
880 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
882 struct file *filp = lo->lo_backing_file;
883 gfp_t gfp = lo->old_gfp_mask;
885 if (lo->lo_state != Lo_bound)
888 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
894 spin_lock_irq(&lo->lo_lock);
895 lo->lo_state = Lo_rundown;
896 spin_unlock_irq(&lo->lo_lock);
898 kthread_stop(lo->lo_thread);
900 lo->lo_backing_file = NULL;
902 loop_release_xfer(lo);
905 lo->lo_device = NULL;
906 lo->lo_encryption = NULL;
908 lo->lo_sizelimit = 0;
909 lo->lo_encrypt_key_size = 0;
911 lo->lo_thread = NULL;
912 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
913 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
914 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
915 invalidate_bdev(bdev);
916 set_capacity(lo->lo_disk, 0);
917 bd_set_size(bdev, 0);
918 mapping_set_gfp_mask(filp->f_mapping, gfp);
919 lo->lo_state = Lo_unbound;
921 /* This is safe: open() is still holding a reference. */
922 module_put(THIS_MODULE);
927 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
930 struct loop_func_table *xfer;
932 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
933 !capable(CAP_SYS_ADMIN))
935 if (lo->lo_state != Lo_bound)
937 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
940 err = loop_release_xfer(lo);
944 if (info->lo_encrypt_type) {
945 unsigned int type = info->lo_encrypt_type;
947 if (type >= MAX_LO_CRYPT)
949 xfer = xfer_funcs[type];
955 err = loop_init_xfer(lo, xfer, info);
959 if (lo->lo_offset != info->lo_offset ||
960 lo->lo_sizelimit != info->lo_sizelimit) {
961 lo->lo_offset = info->lo_offset;
962 lo->lo_sizelimit = info->lo_sizelimit;
963 if (figure_loop_size(lo))
967 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
968 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
969 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
970 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
974 lo->transfer = xfer->transfer;
975 lo->ioctl = xfer->ioctl;
977 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
978 lo->lo_init[0] = info->lo_init[0];
979 lo->lo_init[1] = info->lo_init[1];
980 if (info->lo_encrypt_key_size) {
981 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
982 info->lo_encrypt_key_size);
983 lo->lo_key_owner = current->uid;
990 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
992 struct file *file = lo->lo_backing_file;
996 if (lo->lo_state != Lo_bound)
998 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1001 memset(info, 0, sizeof(*info));
1002 info->lo_number = lo->lo_number;
1003 info->lo_device = huge_encode_dev(stat.dev);
1004 info->lo_inode = stat.ino;
1005 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1006 info->lo_offset = lo->lo_offset;
1007 info->lo_sizelimit = lo->lo_sizelimit;
1008 info->lo_flags = lo->lo_flags;
1009 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1010 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1011 info->lo_encrypt_type =
1012 lo->lo_encryption ? lo->lo_encryption->number : 0;
1013 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1014 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1015 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1016 lo->lo_encrypt_key_size);
1022 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1024 memset(info64, 0, sizeof(*info64));
1025 info64->lo_number = info->lo_number;
1026 info64->lo_device = info->lo_device;
1027 info64->lo_inode = info->lo_inode;
1028 info64->lo_rdevice = info->lo_rdevice;
1029 info64->lo_offset = info->lo_offset;
1030 info64->lo_sizelimit = 0;
1031 info64->lo_encrypt_type = info->lo_encrypt_type;
1032 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1033 info64->lo_flags = info->lo_flags;
1034 info64->lo_init[0] = info->lo_init[0];
1035 info64->lo_init[1] = info->lo_init[1];
1036 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1037 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1039 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1040 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1044 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1046 memset(info, 0, sizeof(*info));
1047 info->lo_number = info64->lo_number;
1048 info->lo_device = info64->lo_device;
1049 info->lo_inode = info64->lo_inode;
1050 info->lo_rdevice = info64->lo_rdevice;
1051 info->lo_offset = info64->lo_offset;
1052 info->lo_encrypt_type = info64->lo_encrypt_type;
1053 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1054 info->lo_flags = info64->lo_flags;
1055 info->lo_init[0] = info64->lo_init[0];
1056 info->lo_init[1] = info64->lo_init[1];
1057 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1058 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1060 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1061 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1063 /* error in case values were truncated */
1064 if (info->lo_device != info64->lo_device ||
1065 info->lo_rdevice != info64->lo_rdevice ||
1066 info->lo_inode != info64->lo_inode ||
1067 info->lo_offset != info64->lo_offset)
1074 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1076 struct loop_info info;
1077 struct loop_info64 info64;
1079 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1081 loop_info64_from_old(&info, &info64);
1082 return loop_set_status(lo, &info64);
1086 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1088 struct loop_info64 info64;
1090 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1092 return loop_set_status(lo, &info64);
1096 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1097 struct loop_info info;
1098 struct loop_info64 info64;
1104 err = loop_get_status(lo, &info64);
1106 err = loop_info64_to_old(&info64, &info);
1107 if (!err && copy_to_user(arg, &info, sizeof(info)))
1114 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1115 struct loop_info64 info64;
1121 err = loop_get_status(lo, &info64);
1122 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1128 static int lo_ioctl(struct inode * inode, struct file * file,
1129 unsigned int cmd, unsigned long arg)
1131 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1134 mutex_lock(&lo->lo_ctl_mutex);
1137 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1139 case LOOP_CHANGE_FD:
1140 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1143 err = loop_clr_fd(lo, inode->i_bdev);
1145 case LOOP_SET_STATUS:
1146 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1148 case LOOP_GET_STATUS:
1149 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1151 case LOOP_SET_STATUS64:
1152 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1154 case LOOP_GET_STATUS64:
1155 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1158 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1160 mutex_unlock(&lo->lo_ctl_mutex);
1164 #ifdef CONFIG_COMPAT
1165 struct compat_loop_info {
1166 compat_int_t lo_number; /* ioctl r/o */
1167 compat_dev_t lo_device; /* ioctl r/o */
1168 compat_ulong_t lo_inode; /* ioctl r/o */
1169 compat_dev_t lo_rdevice; /* ioctl r/o */
1170 compat_int_t lo_offset;
1171 compat_int_t lo_encrypt_type;
1172 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1173 compat_int_t lo_flags; /* ioctl r/o */
1174 char lo_name[LO_NAME_SIZE];
1175 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1176 compat_ulong_t lo_init[2];
1181 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1182 * - noinlined to reduce stack space usage in main part of driver
1185 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1186 struct loop_info64 *info64)
1188 struct compat_loop_info info;
1190 if (copy_from_user(&info, arg, sizeof(info)))
1193 memset(info64, 0, sizeof(*info64));
1194 info64->lo_number = info.lo_number;
1195 info64->lo_device = info.lo_device;
1196 info64->lo_inode = info.lo_inode;
1197 info64->lo_rdevice = info.lo_rdevice;
1198 info64->lo_offset = info.lo_offset;
1199 info64->lo_sizelimit = 0;
1200 info64->lo_encrypt_type = info.lo_encrypt_type;
1201 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1202 info64->lo_flags = info.lo_flags;
1203 info64->lo_init[0] = info.lo_init[0];
1204 info64->lo_init[1] = info.lo_init[1];
1205 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1206 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1208 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1209 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1214 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1215 * - noinlined to reduce stack space usage in main part of driver
1218 loop_info64_to_compat(const struct loop_info64 *info64,
1219 struct compat_loop_info __user *arg)
1221 struct compat_loop_info info;
1223 memset(&info, 0, sizeof(info));
1224 info.lo_number = info64->lo_number;
1225 info.lo_device = info64->lo_device;
1226 info.lo_inode = info64->lo_inode;
1227 info.lo_rdevice = info64->lo_rdevice;
1228 info.lo_offset = info64->lo_offset;
1229 info.lo_encrypt_type = info64->lo_encrypt_type;
1230 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1231 info.lo_flags = info64->lo_flags;
1232 info.lo_init[0] = info64->lo_init[0];
1233 info.lo_init[1] = info64->lo_init[1];
1234 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1235 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1237 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1238 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1240 /* error in case values were truncated */
1241 if (info.lo_device != info64->lo_device ||
1242 info.lo_rdevice != info64->lo_rdevice ||
1243 info.lo_inode != info64->lo_inode ||
1244 info.lo_offset != info64->lo_offset ||
1245 info.lo_init[0] != info64->lo_init[0] ||
1246 info.lo_init[1] != info64->lo_init[1])
1249 if (copy_to_user(arg, &info, sizeof(info)))
1255 loop_set_status_compat(struct loop_device *lo,
1256 const struct compat_loop_info __user *arg)
1258 struct loop_info64 info64;
1261 ret = loop_info64_from_compat(arg, &info64);
1264 return loop_set_status(lo, &info64);
1268 loop_get_status_compat(struct loop_device *lo,
1269 struct compat_loop_info __user *arg)
1271 struct loop_info64 info64;
1277 err = loop_get_status(lo, &info64);
1279 err = loop_info64_to_compat(&info64, arg);
1283 static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1285 struct inode *inode = file->f_path.dentry->d_inode;
1286 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1291 case LOOP_SET_STATUS:
1292 mutex_lock(&lo->lo_ctl_mutex);
1293 err = loop_set_status_compat(
1294 lo, (const struct compat_loop_info __user *) arg);
1295 mutex_unlock(&lo->lo_ctl_mutex);
1297 case LOOP_GET_STATUS:
1298 mutex_lock(&lo->lo_ctl_mutex);
1299 err = loop_get_status_compat(
1300 lo, (struct compat_loop_info __user *) arg);
1301 mutex_unlock(&lo->lo_ctl_mutex);
1304 case LOOP_GET_STATUS64:
1305 case LOOP_SET_STATUS64:
1306 arg = (unsigned long) compat_ptr(arg);
1308 case LOOP_CHANGE_FD:
1309 err = lo_ioctl(inode, file, cmd, arg);
1320 static int lo_open(struct inode *inode, struct file *file)
1322 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1324 mutex_lock(&lo->lo_ctl_mutex);
1326 mutex_unlock(&lo->lo_ctl_mutex);
1331 static int lo_release(struct inode *inode, struct file *file)
1333 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1335 mutex_lock(&lo->lo_ctl_mutex);
1337 mutex_unlock(&lo->lo_ctl_mutex);
1342 static struct block_device_operations lo_fops = {
1343 .owner = THIS_MODULE,
1345 .release = lo_release,
1347 #ifdef CONFIG_COMPAT
1348 .compat_ioctl = lo_compat_ioctl,
1353 * And now the modules code and kernel interface.
1355 static int max_loop;
1356 module_param(max_loop, int, 0);
1357 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1358 MODULE_LICENSE("GPL");
1359 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1361 int loop_register_transfer(struct loop_func_table *funcs)
1363 unsigned int n = funcs->number;
1365 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1367 xfer_funcs[n] = funcs;
1371 int loop_unregister_transfer(int number)
1373 unsigned int n = number;
1374 struct loop_device *lo;
1375 struct loop_func_table *xfer;
1377 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1380 xfer_funcs[n] = NULL;
1382 list_for_each_entry(lo, &loop_devices, lo_list) {
1383 mutex_lock(&lo->lo_ctl_mutex);
1385 if (lo->lo_encryption == xfer)
1386 loop_release_xfer(lo);
1388 mutex_unlock(&lo->lo_ctl_mutex);
1394 EXPORT_SYMBOL(loop_register_transfer);
1395 EXPORT_SYMBOL(loop_unregister_transfer);
1397 static struct loop_device *loop_alloc(int i)
1399 struct loop_device *lo;
1400 struct gendisk *disk;
1402 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1406 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1410 disk = lo->lo_disk = alloc_disk(1);
1412 goto out_free_queue;
1414 mutex_init(&lo->lo_ctl_mutex);
1416 lo->lo_thread = NULL;
1417 init_waitqueue_head(&lo->lo_event);
1418 spin_lock_init(&lo->lo_lock);
1419 disk->major = LOOP_MAJOR;
1420 disk->first_minor = i;
1421 disk->fops = &lo_fops;
1422 disk->private_data = lo;
1423 disk->queue = lo->lo_queue;
1424 sprintf(disk->disk_name, "loop%d", i);
1428 blk_cleanup_queue(lo->lo_queue);
1435 static void loop_free(struct loop_device *lo)
1437 blk_cleanup_queue(lo->lo_queue);
1438 put_disk(lo->lo_disk);
1439 list_del(&lo->lo_list);
1443 static struct loop_device *loop_init_one(int i)
1445 struct loop_device *lo;
1447 list_for_each_entry(lo, &loop_devices, lo_list) {
1448 if (lo->lo_number == i)
1454 add_disk(lo->lo_disk);
1455 list_add_tail(&lo->lo_list, &loop_devices);
1460 static void loop_del_one(struct loop_device *lo)
1462 del_gendisk(lo->lo_disk);
1466 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1468 struct loop_device *lo;
1469 struct kobject *kobj;
1471 mutex_lock(&loop_devices_mutex);
1472 lo = loop_init_one(dev & MINORMASK);
1473 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1474 mutex_unlock(&loop_devices_mutex);
1480 static int __init loop_init(void)
1483 unsigned long range;
1484 struct loop_device *lo, *next;
1487 * loop module now has a feature to instantiate underlying device
1488 * structure on-demand, provided that there is an access dev node.
1489 * However, this will not work well with user space tool that doesn't
1490 * know about such "feature". In order to not break any existing
1491 * tool, we do the following:
1493 * (1) if max_loop is specified, create that many upfront, and this
1494 * also becomes a hard limit.
1495 * (2) if max_loop is not specified, create 8 loop device on module
1496 * load, user can further extend loop device by create dev node
1497 * themselves and have kernel automatically instantiate actual
1500 if (max_loop > 1UL << MINORBITS)
1508 range = 1UL << MINORBITS;
1511 if (register_blkdev(LOOP_MAJOR, "loop"))
1514 for (i = 0; i < nr; i++) {
1518 list_add_tail(&lo->lo_list, &loop_devices);
1521 /* point of no return */
1523 list_for_each_entry(lo, &loop_devices, lo_list)
1524 add_disk(lo->lo_disk);
1526 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1527 THIS_MODULE, loop_probe, NULL, NULL);
1529 printk(KERN_INFO "loop: module loaded\n");
1533 printk(KERN_INFO "loop: out of memory\n");
1535 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1538 unregister_blkdev(LOOP_MAJOR, "loop");
1542 static void __exit loop_exit(void)
1544 unsigned long range;
1545 struct loop_device *lo, *next;
1547 range = max_loop ? max_loop : 1UL << MINORBITS;
1549 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1552 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1553 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1554 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
1557 module_init(loop_init);
1558 module_exit(loop_exit);
1561 static int __init max_loop_setup(char *str)
1563 max_loop = simple_strtol(str, NULL, 0);
1567 __setup("max_loop=", max_loop_setup);