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 * This is best effort. We really wouldn't know what to do with a returned
203 * error. This code is taken from the implementation of fsync.
205 static int sync_file(struct file * file, int full_sync)
207 struct address_space *mapping;
210 if (!file->f_op || !file->f_op->fsync)
213 mapping = file->f_mapping;
215 ret = filemap_fdatawrite(mapping);
219 * We need to protect against concurrent writers,
220 * which could cause livelocks in fsync_buffers_list
223 ret = file->f_op->fsync(file, file->f_dentry, 1);
225 ret2 = filemap_fdatawait(mapping);
234 * do_lo_send_aops - helper for writing data to a loop device
236 * This is the fast version for backing filesystems which implement the address
237 * space operations prepare_write and commit_write.
239 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
240 int bsize, loff_t pos, struct page *page)
242 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
243 struct address_space *mapping = file->f_mapping;
244 const struct address_space_operations *aops = mapping->a_ops;
245 struct inode *inode = file->f_dentry->d_inode;
247 unsigned offset, bv_offs;
249 unsigned long old_blocks;
251 mutex_lock(&mapping->host->i_mutex);
253 spin_lock(&inode->i_lock);
254 old_blocks = inode->i_blocks;
255 spin_unlock(&inode->i_lock);
257 index = pos >> PAGE_CACHE_SHIFT;
258 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
259 bv_offs = bvec->bv_offset;
266 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
267 size = PAGE_CACHE_SIZE - offset;
270 page = grab_cache_page(mapping, index);
273 ret = aops->prepare_write(file, page, offset,
276 if (ret == AOP_TRUNCATED_PAGE) {
277 page_cache_release(page);
282 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
283 bvec->bv_page, bv_offs, size, IV);
284 if (unlikely(transfer_result)) {
286 * The transfer failed, but we still write the data to
287 * keep prepare/commit calls balanced.
289 printk(KERN_ERR "loop: transfer error block %llu\n",
290 (unsigned long long)index);
291 zero_user_page(page, offset, size, KM_USER0);
293 flush_dcache_page(page);
294 ret = aops->commit_write(file, page, offset,
297 if (ret == AOP_TRUNCATED_PAGE) {
298 page_cache_release(page);
303 if (unlikely(transfer_result))
311 page_cache_release(page);
315 if (file->f_flags & O_SYNC) {
317 spin_lock(&inode->i_lock);
318 if (inode->i_blocks > old_blocks)
320 spin_unlock(&inode->i_lock);
321 ret = sync_file(file, full_sync);
324 mutex_unlock(&mapping->host->i_mutex);
328 page_cache_release(page);
335 * __do_lo_send_write - helper for writing data to a loop device
337 * This helper just factors out common code between do_lo_send_direct_write()
338 * and do_lo_send_write().
340 static int __do_lo_send_write(struct file *file,
341 u8 *buf, const int len, loff_t pos)
344 mm_segment_t old_fs = get_fs();
347 bw = file->f_op->write(file, buf, len, &pos);
349 if (likely(bw == len))
351 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
352 (unsigned long long)pos, len);
359 * do_lo_send_direct_write - helper for writing data to a loop device
361 * This is the fast, non-transforming version for backing filesystems which do
362 * not implement the address space operations prepare_write and commit_write.
363 * It uses the write file operation which should be present on all writeable
366 static int do_lo_send_direct_write(struct loop_device *lo,
367 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
369 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
370 kmap(bvec->bv_page) + bvec->bv_offset,
372 kunmap(bvec->bv_page);
378 * do_lo_send_write - helper for writing data to a loop device
380 * This is the slow, transforming version for filesystems which do not
381 * implement the address space operations prepare_write and commit_write. It
382 * uses the write file operation which should be present on all writeable
385 * Using fops->write is slower than using aops->{prepare,commit}_write in the
386 * transforming case because we need to double buffer the data as we cannot do
387 * the transformations in place as we do not have direct access to the
388 * destination pages of the backing file.
390 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
391 int bsize, loff_t pos, struct page *page)
393 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
394 bvec->bv_offset, bvec->bv_len, pos >> 9);
396 return __do_lo_send_write(lo->lo_backing_file,
397 page_address(page), bvec->bv_len,
399 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
400 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
406 static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
409 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
411 struct bio_vec *bvec;
412 struct page *page = NULL;
415 do_lo_send = do_lo_send_aops;
416 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
417 do_lo_send = do_lo_send_direct_write;
418 if (lo->transfer != transfer_none) {
419 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
423 do_lo_send = do_lo_send_write;
426 bio_for_each_segment(bvec, bio, i) {
427 ret = do_lo_send(lo, bvec, bsize, pos, page);
439 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
444 struct lo_read_data {
445 struct loop_device *lo;
452 lo_read_actor(read_descriptor_t *desc, struct page *page,
453 unsigned long offset, unsigned long size)
455 unsigned long count = desc->count;
456 struct lo_read_data *p = desc->arg.data;
457 struct loop_device *lo = p->lo;
460 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
465 if (lo_do_transfer(lo, READ, page, offset, p->page, p->offset, size, IV)) {
467 printk(KERN_ERR "loop: transfer error block %ld\n",
469 desc->error = -EINVAL;
472 flush_dcache_page(p->page);
474 desc->count = count - size;
475 desc->written += size;
481 do_lo_receive(struct loop_device *lo,
482 struct bio_vec *bvec, int bsize, loff_t pos)
484 struct lo_read_data cookie;
489 cookie.page = bvec->bv_page;
490 cookie.offset = bvec->bv_offset;
491 cookie.bsize = bsize;
492 file = lo->lo_backing_file;
493 retval = file->f_op->sendfile(file, &pos, bvec->bv_len,
494 lo_read_actor, &cookie);
495 return (retval < 0)? retval: 0;
499 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
501 struct bio_vec *bvec;
504 bio_for_each_segment(bvec, bio, i) {
505 ret = do_lo_receive(lo, bvec, bsize, pos);
513 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
517 int sync = bio_sync(bio);
518 int barrier = bio_barrier(bio);
521 ret = sync_file(lo->lo_backing_file, 1);
526 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
527 if (bio_rw(bio) == WRITE)
528 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
530 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
532 if ((barrier || sync) && !ret)
533 ret = sync_file(lo->lo_backing_file, 1);
539 * Add bio to back of pending list
541 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
543 if (lo->lo_biotail) {
544 lo->lo_biotail->bi_next = bio;
545 lo->lo_biotail = bio;
547 lo->lo_bio = lo->lo_biotail = bio;
551 * Grab first pending buffer
553 static struct bio *loop_get_bio(struct loop_device *lo)
557 if ((bio = lo->lo_bio)) {
558 if (bio == lo->lo_biotail)
559 lo->lo_biotail = NULL;
560 lo->lo_bio = bio->bi_next;
567 static int loop_make_request(request_queue_t *q, struct bio *old_bio)
569 struct loop_device *lo = q->queuedata;
570 int rw = bio_rw(old_bio);
575 BUG_ON(!lo || (rw != READ && rw != WRITE));
577 spin_lock_irq(&lo->lo_lock);
578 if (lo->lo_state != Lo_bound)
580 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
582 loop_add_bio(lo, old_bio);
583 wake_up(&lo->lo_event);
584 spin_unlock_irq(&lo->lo_lock);
588 spin_unlock_irq(&lo->lo_lock);
589 bio_io_error(old_bio, old_bio->bi_size);
594 * kick off io on the underlying address space
596 static void loop_unplug(request_queue_t *q)
598 struct loop_device *lo = q->queuedata;
600 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
601 blk_run_address_space(lo->lo_backing_file->f_mapping);
604 struct switch_request {
606 struct completion wait;
609 static void do_loop_switch(struct loop_device *, struct switch_request *);
611 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
613 if (unlikely(!bio->bi_bdev)) {
614 do_loop_switch(lo, bio->bi_private);
617 int ret = do_bio_filebacked(lo, bio);
618 bio_endio(bio, bio->bi_size, ret);
623 * worker thread that handles reads/writes to file backed loop devices,
624 * to avoid blocking in our make_request_fn. it also does loop decrypting
625 * on reads for block backed loop, as that is too heavy to do from
626 * b_end_io context where irqs may be disabled.
628 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
629 * calling kthread_stop(). Therefore once kthread_should_stop() is
630 * true, make_request will not place any more requests. Therefore
631 * once kthread_should_stop() is true and lo_bio is NULL, we are
632 * done with the loop.
634 static int loop_thread(void *data)
636 struct loop_device *lo = data;
640 * loop can be used in an encrypted device,
641 * hence, it mustn't be stopped at all
642 * because it could be indirectly used during suspension
644 current->flags |= PF_NOFREEZE;
646 set_user_nice(current, -20);
648 while (!kthread_should_stop() || lo->lo_bio) {
650 wait_event_interruptible(lo->lo_event,
651 lo->lo_bio || kthread_should_stop());
655 spin_lock_irq(&lo->lo_lock);
656 bio = loop_get_bio(lo);
657 spin_unlock_irq(&lo->lo_lock);
660 loop_handle_bio(lo, bio);
667 * loop_switch performs the hard work of switching a backing store.
668 * First it needs to flush existing IO, it does this by sending a magic
669 * BIO down the pipe. The completion of this BIO does the actual switch.
671 static int loop_switch(struct loop_device *lo, struct file *file)
673 struct switch_request w;
674 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
677 init_completion(&w.wait);
679 bio->bi_private = &w;
681 loop_make_request(lo->lo_queue, bio);
682 wait_for_completion(&w.wait);
687 * Do the actual switch; called from the BIO completion routine
689 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
691 struct file *file = p->file;
692 struct file *old_file = lo->lo_backing_file;
693 struct address_space *mapping = file->f_mapping;
695 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
696 lo->lo_backing_file = file;
697 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
698 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
699 lo->old_gfp_mask = mapping_gfp_mask(mapping);
700 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
706 * loop_change_fd switched the backing store of a loopback device to
707 * a new file. This is useful for operating system installers to free up
708 * the original file and in High Availability environments to switch to
709 * an alternative location for the content in case of server meltdown.
710 * This can only work if the loop device is used read-only, and if the
711 * new backing store is the same size and type as the old backing store.
713 static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
714 struct block_device *bdev, unsigned int arg)
716 struct file *file, *old_file;
721 if (lo->lo_state != Lo_bound)
724 /* the loop device has to be read-only */
726 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
734 inode = file->f_mapping->host;
735 old_file = lo->lo_backing_file;
739 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
742 /* new backing store needs to support loop (eg sendfile) */
743 if (!inode->i_fop->sendfile)
746 /* size of the new backing store needs to be the same */
747 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
751 error = loop_switch(lo, file);
764 static inline int is_loop_device(struct file *file)
766 struct inode *i = file->f_mapping->host;
768 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
771 static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
772 struct block_device *bdev, unsigned int arg)
774 struct file *file, *f;
776 struct address_space *mapping;
777 unsigned lo_blocksize;
782 /* This is safe, since we have a reference from open(). */
783 __module_get(THIS_MODULE);
791 if (lo->lo_state != Lo_unbound)
794 /* Avoid recursion */
796 while (is_loop_device(f)) {
797 struct loop_device *l;
799 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
802 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
803 if (l->lo_state == Lo_unbound) {
807 f = l->lo_backing_file;
810 mapping = file->f_mapping;
811 inode = mapping->host;
813 if (!(file->f_mode & FMODE_WRITE))
814 lo_flags |= LO_FLAGS_READ_ONLY;
816 if ((file->f_flags & O_SYNC) && (!file->f_op || !file->f_op->fsync))
820 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
821 const struct address_space_operations *aops = mapping->a_ops;
823 * If we can't read - sorry. If we only can't write - well,
824 * it's going to be read-only.
826 if (!file->f_op->sendfile)
828 if (aops->prepare_write && aops->commit_write)
829 lo_flags |= LO_FLAGS_USE_AOPS;
830 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
831 lo_flags |= LO_FLAGS_READ_ONLY;
833 lo_blocksize = S_ISBLK(inode->i_mode) ?
834 inode->i_bdev->bd_block_size : PAGE_SIZE;
841 size = get_loop_size(lo, file);
843 if ((loff_t)(sector_t)size != size) {
848 if (!(lo_file->f_mode & FMODE_WRITE))
849 lo_flags |= LO_FLAGS_READ_ONLY;
851 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
853 lo->lo_blocksize = lo_blocksize;
854 lo->lo_device = bdev;
855 lo->lo_flags = lo_flags;
856 lo->lo_backing_file = file;
857 lo->transfer = transfer_none;
859 lo->lo_sizelimit = 0;
860 lo->old_gfp_mask = mapping_gfp_mask(mapping);
861 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
863 lo->lo_bio = lo->lo_biotail = NULL;
866 * set queue make_request_fn, and add limits based on lower level
869 blk_queue_make_request(lo->lo_queue, loop_make_request);
870 lo->lo_queue->queuedata = lo;
871 lo->lo_queue->unplug_fn = loop_unplug;
873 set_capacity(lo->lo_disk, size);
874 bd_set_size(bdev, size << 9);
876 set_blocksize(bdev, lo_blocksize);
878 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
880 if (IS_ERR(lo->lo_thread)) {
881 error = PTR_ERR(lo->lo_thread);
884 lo->lo_state = Lo_bound;
885 wake_up_process(lo->lo_thread);
889 lo->lo_thread = NULL;
890 lo->lo_device = NULL;
891 lo->lo_backing_file = NULL;
893 set_capacity(lo->lo_disk, 0);
894 invalidate_bdev(bdev);
895 bd_set_size(bdev, 0);
896 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
897 lo->lo_state = Lo_unbound;
901 /* This is safe: open() is still holding a reference. */
902 module_put(THIS_MODULE);
907 loop_release_xfer(struct loop_device *lo)
910 struct loop_func_table *xfer = lo->lo_encryption;
914 err = xfer->release(lo);
916 lo->lo_encryption = NULL;
917 module_put(xfer->owner);
923 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
924 const struct loop_info64 *i)
929 struct module *owner = xfer->owner;
931 if (!try_module_get(owner))
934 err = xfer->init(lo, i);
938 lo->lo_encryption = xfer;
943 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
945 struct file *filp = lo->lo_backing_file;
946 gfp_t gfp = lo->old_gfp_mask;
948 if (lo->lo_state != Lo_bound)
951 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
957 spin_lock_irq(&lo->lo_lock);
958 lo->lo_state = Lo_rundown;
959 spin_unlock_irq(&lo->lo_lock);
961 kthread_stop(lo->lo_thread);
963 lo->lo_backing_file = NULL;
965 loop_release_xfer(lo);
968 lo->lo_device = NULL;
969 lo->lo_encryption = NULL;
971 lo->lo_sizelimit = 0;
972 lo->lo_encrypt_key_size = 0;
974 lo->lo_thread = NULL;
975 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
976 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
977 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
978 invalidate_bdev(bdev);
979 set_capacity(lo->lo_disk, 0);
980 bd_set_size(bdev, 0);
981 mapping_set_gfp_mask(filp->f_mapping, gfp);
982 lo->lo_state = Lo_unbound;
984 /* This is safe: open() is still holding a reference. */
985 module_put(THIS_MODULE);
990 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
993 struct loop_func_table *xfer;
995 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
996 !capable(CAP_SYS_ADMIN))
998 if (lo->lo_state != Lo_bound)
1000 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1003 err = loop_release_xfer(lo);
1007 if (info->lo_encrypt_type) {
1008 unsigned int type = info->lo_encrypt_type;
1010 if (type >= MAX_LO_CRYPT)
1012 xfer = xfer_funcs[type];
1018 err = loop_init_xfer(lo, xfer, info);
1022 if (lo->lo_offset != info->lo_offset ||
1023 lo->lo_sizelimit != info->lo_sizelimit) {
1024 lo->lo_offset = info->lo_offset;
1025 lo->lo_sizelimit = info->lo_sizelimit;
1026 if (figure_loop_size(lo))
1030 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1031 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1032 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1033 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1037 lo->transfer = xfer->transfer;
1038 lo->ioctl = xfer->ioctl;
1040 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1041 lo->lo_init[0] = info->lo_init[0];
1042 lo->lo_init[1] = info->lo_init[1];
1043 if (info->lo_encrypt_key_size) {
1044 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1045 info->lo_encrypt_key_size);
1046 lo->lo_key_owner = current->uid;
1053 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1055 struct file *file = lo->lo_backing_file;
1059 if (lo->lo_state != Lo_bound)
1061 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1064 memset(info, 0, sizeof(*info));
1065 info->lo_number = lo->lo_number;
1066 info->lo_device = huge_encode_dev(stat.dev);
1067 info->lo_inode = stat.ino;
1068 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1069 info->lo_offset = lo->lo_offset;
1070 info->lo_sizelimit = lo->lo_sizelimit;
1071 info->lo_flags = lo->lo_flags;
1072 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1073 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1074 info->lo_encrypt_type =
1075 lo->lo_encryption ? lo->lo_encryption->number : 0;
1076 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1077 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1078 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1079 lo->lo_encrypt_key_size);
1085 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1087 memset(info64, 0, sizeof(*info64));
1088 info64->lo_number = info->lo_number;
1089 info64->lo_device = info->lo_device;
1090 info64->lo_inode = info->lo_inode;
1091 info64->lo_rdevice = info->lo_rdevice;
1092 info64->lo_offset = info->lo_offset;
1093 info64->lo_sizelimit = 0;
1094 info64->lo_encrypt_type = info->lo_encrypt_type;
1095 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1096 info64->lo_flags = info->lo_flags;
1097 info64->lo_init[0] = info->lo_init[0];
1098 info64->lo_init[1] = info->lo_init[1];
1099 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1100 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1102 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1103 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1107 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1109 memset(info, 0, sizeof(*info));
1110 info->lo_number = info64->lo_number;
1111 info->lo_device = info64->lo_device;
1112 info->lo_inode = info64->lo_inode;
1113 info->lo_rdevice = info64->lo_rdevice;
1114 info->lo_offset = info64->lo_offset;
1115 info->lo_encrypt_type = info64->lo_encrypt_type;
1116 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1117 info->lo_flags = info64->lo_flags;
1118 info->lo_init[0] = info64->lo_init[0];
1119 info->lo_init[1] = info64->lo_init[1];
1120 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1121 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1123 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1124 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1126 /* error in case values were truncated */
1127 if (info->lo_device != info64->lo_device ||
1128 info->lo_rdevice != info64->lo_rdevice ||
1129 info->lo_inode != info64->lo_inode ||
1130 info->lo_offset != info64->lo_offset)
1137 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1139 struct loop_info info;
1140 struct loop_info64 info64;
1142 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1144 loop_info64_from_old(&info, &info64);
1145 return loop_set_status(lo, &info64);
1149 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1151 struct loop_info64 info64;
1153 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1155 return loop_set_status(lo, &info64);
1159 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1160 struct loop_info info;
1161 struct loop_info64 info64;
1167 err = loop_get_status(lo, &info64);
1169 err = loop_info64_to_old(&info64, &info);
1170 if (!err && copy_to_user(arg, &info, sizeof(info)))
1177 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1178 struct loop_info64 info64;
1184 err = loop_get_status(lo, &info64);
1185 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1191 static int lo_ioctl(struct inode * inode, struct file * file,
1192 unsigned int cmd, unsigned long arg)
1194 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1197 mutex_lock(&lo->lo_ctl_mutex);
1200 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1202 case LOOP_CHANGE_FD:
1203 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1206 err = loop_clr_fd(lo, inode->i_bdev);
1208 case LOOP_SET_STATUS:
1209 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1211 case LOOP_GET_STATUS:
1212 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1214 case LOOP_SET_STATUS64:
1215 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1217 case LOOP_GET_STATUS64:
1218 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1221 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1223 mutex_unlock(&lo->lo_ctl_mutex);
1227 #ifdef CONFIG_COMPAT
1228 struct compat_loop_info {
1229 compat_int_t lo_number; /* ioctl r/o */
1230 compat_dev_t lo_device; /* ioctl r/o */
1231 compat_ulong_t lo_inode; /* ioctl r/o */
1232 compat_dev_t lo_rdevice; /* ioctl r/o */
1233 compat_int_t lo_offset;
1234 compat_int_t lo_encrypt_type;
1235 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1236 compat_int_t lo_flags; /* ioctl r/o */
1237 char lo_name[LO_NAME_SIZE];
1238 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1239 compat_ulong_t lo_init[2];
1244 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1245 * - noinlined to reduce stack space usage in main part of driver
1248 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1249 struct loop_info64 *info64)
1251 struct compat_loop_info info;
1253 if (copy_from_user(&info, arg, sizeof(info)))
1256 memset(info64, 0, sizeof(*info64));
1257 info64->lo_number = info.lo_number;
1258 info64->lo_device = info.lo_device;
1259 info64->lo_inode = info.lo_inode;
1260 info64->lo_rdevice = info.lo_rdevice;
1261 info64->lo_offset = info.lo_offset;
1262 info64->lo_sizelimit = 0;
1263 info64->lo_encrypt_type = info.lo_encrypt_type;
1264 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1265 info64->lo_flags = info.lo_flags;
1266 info64->lo_init[0] = info.lo_init[0];
1267 info64->lo_init[1] = info.lo_init[1];
1268 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1269 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1271 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1272 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1277 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1278 * - noinlined to reduce stack space usage in main part of driver
1281 loop_info64_to_compat(const struct loop_info64 *info64,
1282 struct compat_loop_info __user *arg)
1284 struct compat_loop_info info;
1286 memset(&info, 0, sizeof(info));
1287 info.lo_number = info64->lo_number;
1288 info.lo_device = info64->lo_device;
1289 info.lo_inode = info64->lo_inode;
1290 info.lo_rdevice = info64->lo_rdevice;
1291 info.lo_offset = info64->lo_offset;
1292 info.lo_encrypt_type = info64->lo_encrypt_type;
1293 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1294 info.lo_flags = info64->lo_flags;
1295 info.lo_init[0] = info64->lo_init[0];
1296 info.lo_init[1] = info64->lo_init[1];
1297 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1298 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1300 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1301 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1303 /* error in case values were truncated */
1304 if (info.lo_device != info64->lo_device ||
1305 info.lo_rdevice != info64->lo_rdevice ||
1306 info.lo_inode != info64->lo_inode ||
1307 info.lo_offset != info64->lo_offset ||
1308 info.lo_init[0] != info64->lo_init[0] ||
1309 info.lo_init[1] != info64->lo_init[1])
1312 if (copy_to_user(arg, &info, sizeof(info)))
1318 loop_set_status_compat(struct loop_device *lo,
1319 const struct compat_loop_info __user *arg)
1321 struct loop_info64 info64;
1324 ret = loop_info64_from_compat(arg, &info64);
1327 return loop_set_status(lo, &info64);
1331 loop_get_status_compat(struct loop_device *lo,
1332 struct compat_loop_info __user *arg)
1334 struct loop_info64 info64;
1340 err = loop_get_status(lo, &info64);
1342 err = loop_info64_to_compat(&info64, arg);
1346 static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1348 struct inode *inode = file->f_path.dentry->d_inode;
1349 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1354 case LOOP_SET_STATUS:
1355 mutex_lock(&lo->lo_ctl_mutex);
1356 err = loop_set_status_compat(
1357 lo, (const struct compat_loop_info __user *) arg);
1358 mutex_unlock(&lo->lo_ctl_mutex);
1360 case LOOP_GET_STATUS:
1361 mutex_lock(&lo->lo_ctl_mutex);
1362 err = loop_get_status_compat(
1363 lo, (struct compat_loop_info __user *) arg);
1364 mutex_unlock(&lo->lo_ctl_mutex);
1367 case LOOP_GET_STATUS64:
1368 case LOOP_SET_STATUS64:
1369 arg = (unsigned long) compat_ptr(arg);
1371 case LOOP_CHANGE_FD:
1372 err = lo_ioctl(inode, file, cmd, arg);
1383 static int lo_open(struct inode *inode, struct file *file)
1385 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1387 mutex_lock(&lo->lo_ctl_mutex);
1389 mutex_unlock(&lo->lo_ctl_mutex);
1394 static int lo_release(struct inode *inode, struct file *file)
1396 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1398 mutex_lock(&lo->lo_ctl_mutex);
1400 mutex_unlock(&lo->lo_ctl_mutex);
1405 static struct block_device_operations lo_fops = {
1406 .owner = THIS_MODULE,
1408 .release = lo_release,
1410 #ifdef CONFIG_COMPAT
1411 .compat_ioctl = lo_compat_ioctl,
1416 * And now the modules code and kernel interface.
1418 static int max_loop;
1419 module_param(max_loop, int, 0);
1420 MODULE_PARM_DESC(max_loop, "obsolete, loop device is created on-demand");
1421 MODULE_LICENSE("GPL");
1422 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1424 int loop_register_transfer(struct loop_func_table *funcs)
1426 unsigned int n = funcs->number;
1428 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1430 xfer_funcs[n] = funcs;
1434 int loop_unregister_transfer(int number)
1436 unsigned int n = number;
1437 struct loop_device *lo;
1438 struct loop_func_table *xfer;
1440 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1443 xfer_funcs[n] = NULL;
1445 list_for_each_entry(lo, &loop_devices, lo_list) {
1446 mutex_lock(&lo->lo_ctl_mutex);
1448 if (lo->lo_encryption == xfer)
1449 loop_release_xfer(lo);
1451 mutex_unlock(&lo->lo_ctl_mutex);
1457 EXPORT_SYMBOL(loop_register_transfer);
1458 EXPORT_SYMBOL(loop_unregister_transfer);
1460 static struct loop_device *loop_init_one(int i)
1462 struct loop_device *lo;
1463 struct gendisk *disk;
1465 list_for_each_entry(lo, &loop_devices, lo_list) {
1466 if (lo->lo_number == i)
1470 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1474 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1478 disk = lo->lo_disk = alloc_disk(1);
1480 goto out_free_queue;
1482 mutex_init(&lo->lo_ctl_mutex);
1484 lo->lo_thread = NULL;
1485 init_waitqueue_head(&lo->lo_event);
1486 spin_lock_init(&lo->lo_lock);
1487 disk->major = LOOP_MAJOR;
1488 disk->first_minor = i;
1489 disk->fops = &lo_fops;
1490 disk->private_data = lo;
1491 disk->queue = lo->lo_queue;
1492 sprintf(disk->disk_name, "loop%d", i);
1494 list_add_tail(&lo->lo_list, &loop_devices);
1498 blk_cleanup_queue(lo->lo_queue);
1505 static void loop_del_one(struct loop_device *lo)
1507 del_gendisk(lo->lo_disk);
1508 blk_cleanup_queue(lo->lo_queue);
1509 put_disk(lo->lo_disk);
1510 list_del(&lo->lo_list);
1514 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1516 struct loop_device *lo;
1517 struct kobject *kobj;
1519 mutex_lock(&loop_devices_mutex);
1520 lo = loop_init_one(dev & MINORMASK);
1521 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1522 mutex_unlock(&loop_devices_mutex);
1528 static int __init loop_init(void)
1530 if (register_blkdev(LOOP_MAJOR, "loop"))
1532 blk_register_region(MKDEV(LOOP_MAJOR, 0), 1UL << MINORBITS,
1533 THIS_MODULE, loop_probe, NULL, NULL);
1536 printk(KERN_INFO "loop: the max_loop option is obsolete "
1537 "and will be removed in March 2008\n");
1540 printk(KERN_INFO "loop: module loaded\n");
1544 static void __exit loop_exit(void)
1546 struct loop_device *lo, *next;
1548 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1551 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), 1UL << MINORBITS);
1552 if (unregister_blkdev(LOOP_MAJOR, "loop"))
1553 printk(KERN_WARNING "loop: cannot unregister blkdev\n");
1556 module_init(loop_init);
1557 module_exit(loop_exit);
1560 static int __init max_loop_setup(char *str)
1562 max_loop = simple_strtol(str, NULL, 0);
1566 __setup("max_loop=", max_loop_setup);