2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations write_begin is not available on the backing filesystem.
44 * Anton Altaparmakov, 16 Feb 2005
47 * - Advisory locking is ignored here.
48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
52 #include <linux/module.h>
53 #include <linux/moduleparam.h>
54 #include <linux/sched.h>
56 #include <linux/file.h>
57 #include <linux/stat.h>
58 #include <linux/errno.h>
59 #include <linux/major.h>
60 #include <linux/wait.h>
61 #include <linux/blkdev.h>
62 #include <linux/blkpg.h>
63 #include <linux/init.h>
64 #include <linux/swap.h>
65 #include <linux/slab.h>
66 #include <linux/loop.h>
67 #include <linux/compat.h>
68 #include <linux/suspend.h>
69 #include <linux/freezer.h>
70 #include <linux/mutex.h>
71 #include <linux/writeback.h>
72 #include <linux/buffer_head.h> /* for invalidate_bdev() */
73 #include <linux/completion.h>
74 #include <linux/highmem.h>
75 #include <linux/kthread.h>
76 #include <linux/splice.h>
77 #include <linux/sysfs.h>
79 #include <asm/uaccess.h>
81 static LIST_HEAD(loop_devices);
82 static DEFINE_MUTEX(loop_devices_mutex);
85 static int part_shift;
90 static int transfer_none(struct loop_device *lo, int cmd,
91 struct page *raw_page, unsigned raw_off,
92 struct page *loop_page, unsigned loop_off,
93 int size, sector_t real_block)
95 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
96 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
99 memcpy(loop_buf, raw_buf, size);
101 memcpy(raw_buf, loop_buf, size);
103 kunmap_atomic(loop_buf, KM_USER1);
104 kunmap_atomic(raw_buf, KM_USER0);
109 static int transfer_xor(struct loop_device *lo, int cmd,
110 struct page *raw_page, unsigned raw_off,
111 struct page *loop_page, unsigned loop_off,
112 int size, sector_t real_block)
114 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
115 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
116 char *in, *out, *key;
127 key = lo->lo_encrypt_key;
128 keysize = lo->lo_encrypt_key_size;
129 for (i = 0; i < size; i++)
130 *out++ = *in++ ^ key[(i & 511) % keysize];
132 kunmap_atomic(loop_buf, KM_USER1);
133 kunmap_atomic(raw_buf, KM_USER0);
138 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
140 if (unlikely(info->lo_encrypt_key_size <= 0))
145 static struct loop_func_table none_funcs = {
146 .number = LO_CRYPT_NONE,
147 .transfer = transfer_none,
150 static struct loop_func_table xor_funcs = {
151 .number = LO_CRYPT_XOR,
152 .transfer = transfer_xor,
156 /* xfer_funcs[0] is special - its release function is never called */
157 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164 loff_t size, offset, loopsize;
166 /* Compute loopsize in bytes */
167 size = i_size_read(file->f_mapping->host);
168 offset = lo->lo_offset;
169 loopsize = size - offset;
170 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
171 loopsize = lo->lo_sizelimit;
174 * Unfortunately, if we want to do I/O on the device,
175 * the number of 512-byte sectors has to fit into a sector_t.
177 return loopsize >> 9;
181 figure_loop_size(struct loop_device *lo)
183 loff_t size = get_loop_size(lo, lo->lo_backing_file);
184 sector_t x = (sector_t)size;
186 if (unlikely((loff_t)x != size))
189 set_capacity(lo->lo_disk, x);
194 lo_do_transfer(struct loop_device *lo, int cmd,
195 struct page *rpage, unsigned roffs,
196 struct page *lpage, unsigned loffs,
197 int size, sector_t rblock)
199 if (unlikely(!lo->transfer))
202 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
206 * do_lo_send_aops - helper for writing data to a loop device
208 * This is the fast version for backing filesystems which implement the address
209 * space operations write_begin and write_end.
211 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
212 loff_t pos, struct page *unused)
214 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
215 struct address_space *mapping = file->f_mapping;
217 unsigned offset, bv_offs;
220 mutex_lock(&mapping->host->i_mutex);
221 index = pos >> PAGE_CACHE_SHIFT;
222 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
223 bv_offs = bvec->bv_offset;
227 unsigned size, copied;
232 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
233 size = PAGE_CACHE_SIZE - offset;
237 ret = pagecache_write_begin(file, mapping, pos, size, 0,
242 file_update_time(file);
244 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
245 bvec->bv_page, bv_offs, size, IV);
247 if (unlikely(transfer_result))
250 ret = pagecache_write_end(file, mapping, pos, size, copied,
252 if (ret < 0 || ret != copied)
255 if (unlikely(transfer_result))
266 mutex_unlock(&mapping->host->i_mutex);
274 * __do_lo_send_write - helper for writing data to a loop device
276 * This helper just factors out common code between do_lo_send_direct_write()
277 * and do_lo_send_write().
279 static int __do_lo_send_write(struct file *file,
280 u8 *buf, const int len, loff_t pos)
283 mm_segment_t old_fs = get_fs();
286 bw = file->f_op->write(file, buf, len, &pos);
288 if (likely(bw == len))
290 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
291 (unsigned long long)pos, len);
298 * do_lo_send_direct_write - helper for writing data to a loop device
300 * This is the fast, non-transforming version for backing filesystems which do
301 * not implement the address space operations write_begin and write_end.
302 * It uses the write file operation which should be present on all writeable
305 static int do_lo_send_direct_write(struct loop_device *lo,
306 struct bio_vec *bvec, loff_t pos, struct page *page)
308 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
309 kmap(bvec->bv_page) + bvec->bv_offset,
311 kunmap(bvec->bv_page);
317 * do_lo_send_write - helper for writing data to a loop device
319 * This is the slow, transforming version for filesystems which do not
320 * implement the address space operations write_begin and write_end. It
321 * uses the write file operation which should be present on all writeable
324 * Using fops->write is slower than using aops->{prepare,commit}_write in the
325 * transforming case because we need to double buffer the data as we cannot do
326 * the transformations in place as we do not have direct access to the
327 * destination pages of the backing file.
329 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
330 loff_t pos, struct page *page)
332 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
333 bvec->bv_offset, bvec->bv_len, pos >> 9);
335 return __do_lo_send_write(lo->lo_backing_file,
336 page_address(page), bvec->bv_len,
338 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
339 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
345 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
347 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
349 struct bio_vec *bvec;
350 struct page *page = NULL;
353 do_lo_send = do_lo_send_aops;
354 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
355 do_lo_send = do_lo_send_direct_write;
356 if (lo->transfer != transfer_none) {
357 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
361 do_lo_send = do_lo_send_write;
364 bio_for_each_segment(bvec, bio, i) {
365 ret = do_lo_send(lo, bvec, pos, page);
377 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
382 struct lo_read_data {
383 struct loop_device *lo;
390 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
391 struct splice_desc *sd)
393 struct lo_read_data *p = sd->u.data;
394 struct loop_device *lo = p->lo;
395 struct page *page = buf->page;
399 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
405 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
406 printk(KERN_ERR "loop: transfer error block %ld\n",
411 flush_dcache_page(p->page);
420 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
422 return __splice_from_pipe(pipe, sd, lo_splice_actor);
426 do_lo_receive(struct loop_device *lo,
427 struct bio_vec *bvec, int bsize, loff_t pos)
429 struct lo_read_data cookie;
430 struct splice_desc sd;
435 cookie.page = bvec->bv_page;
436 cookie.offset = bvec->bv_offset;
437 cookie.bsize = bsize;
440 sd.total_len = bvec->bv_len;
445 file = lo->lo_backing_file;
446 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
455 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
457 struct bio_vec *bvec;
460 bio_for_each_segment(bvec, bio, i) {
461 ret = do_lo_receive(lo, bvec, bsize, pos);
469 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
474 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
476 if (bio_rw(bio) == WRITE) {
477 struct file *file = lo->lo_backing_file;
479 if (bio->bi_rw & REQ_FLUSH) {
480 ret = vfs_fsync(file, 0);
481 if (unlikely(ret && ret != -EINVAL)) {
487 ret = lo_send(lo, bio, pos);
489 if ((bio->bi_rw & REQ_FUA) && !ret) {
490 ret = vfs_fsync(file, 0);
491 if (unlikely(ret && ret != -EINVAL))
495 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
502 * Add bio to back of pending list
504 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
506 bio_list_add(&lo->lo_bio_list, bio);
510 * Grab first pending buffer
512 static struct bio *loop_get_bio(struct loop_device *lo)
514 return bio_list_pop(&lo->lo_bio_list);
517 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
519 struct loop_device *lo = q->queuedata;
520 int rw = bio_rw(old_bio);
525 BUG_ON(!lo || (rw != READ && rw != WRITE));
527 spin_lock_irq(&lo->lo_lock);
528 if (lo->lo_state != Lo_bound)
530 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
532 loop_add_bio(lo, old_bio);
533 wake_up(&lo->lo_event);
534 spin_unlock_irq(&lo->lo_lock);
538 spin_unlock_irq(&lo->lo_lock);
539 bio_io_error(old_bio);
543 struct switch_request {
545 struct completion wait;
548 static void do_loop_switch(struct loop_device *, struct switch_request *);
550 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
552 if (unlikely(!bio->bi_bdev)) {
553 do_loop_switch(lo, bio->bi_private);
556 int ret = do_bio_filebacked(lo, bio);
562 * worker thread that handles reads/writes to file backed loop devices,
563 * to avoid blocking in our make_request_fn. it also does loop decrypting
564 * on reads for block backed loop, as that is too heavy to do from
565 * b_end_io context where irqs may be disabled.
567 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
568 * calling kthread_stop(). Therefore once kthread_should_stop() is
569 * true, make_request will not place any more requests. Therefore
570 * once kthread_should_stop() is true and lo_bio is NULL, we are
571 * done with the loop.
573 static int loop_thread(void *data)
575 struct loop_device *lo = data;
578 set_user_nice(current, -20);
580 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
582 wait_event_interruptible(lo->lo_event,
583 !bio_list_empty(&lo->lo_bio_list) ||
584 kthread_should_stop());
586 if (bio_list_empty(&lo->lo_bio_list))
588 spin_lock_irq(&lo->lo_lock);
589 bio = loop_get_bio(lo);
590 spin_unlock_irq(&lo->lo_lock);
593 loop_handle_bio(lo, bio);
600 * loop_switch performs the hard work of switching a backing store.
601 * First it needs to flush existing IO, it does this by sending a magic
602 * BIO down the pipe. The completion of this BIO does the actual switch.
604 static int loop_switch(struct loop_device *lo, struct file *file)
606 struct switch_request w;
607 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
610 init_completion(&w.wait);
612 bio->bi_private = &w;
614 loop_make_request(lo->lo_queue, bio);
615 wait_for_completion(&w.wait);
620 * Helper to flush the IOs in loop, but keeping loop thread running
622 static int loop_flush(struct loop_device *lo)
624 /* loop not yet configured, no running thread, nothing to flush */
628 return loop_switch(lo, NULL);
632 * Do the actual switch; called from the BIO completion routine
634 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
636 struct file *file = p->file;
637 struct file *old_file = lo->lo_backing_file;
638 struct address_space *mapping;
640 /* if no new file, only flush of queued bios requested */
644 mapping = file->f_mapping;
645 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
646 lo->lo_backing_file = file;
647 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
648 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
649 lo->old_gfp_mask = mapping_gfp_mask(mapping);
650 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
657 * loop_change_fd switched the backing store of a loopback device to
658 * a new file. This is useful for operating system installers to free up
659 * the original file and in High Availability environments to switch to
660 * an alternative location for the content in case of server meltdown.
661 * This can only work if the loop device is used read-only, and if the
662 * new backing store is the same size and type as the old backing store.
664 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
667 struct file *file, *old_file;
672 if (lo->lo_state != Lo_bound)
675 /* the loop device has to be read-only */
677 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
685 inode = file->f_mapping->host;
686 old_file = lo->lo_backing_file;
690 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
693 /* size of the new backing store needs to be the same */
694 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
698 error = loop_switch(lo, file);
704 ioctl_by_bdev(bdev, BLKRRPART, 0);
713 static inline int is_loop_device(struct file *file)
715 struct inode *i = file->f_mapping->host;
717 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
720 /* loop sysfs attributes */
722 static ssize_t loop_attr_show(struct device *dev, char *page,
723 ssize_t (*callback)(struct loop_device *, char *))
725 struct loop_device *l, *lo = NULL;
727 mutex_lock(&loop_devices_mutex);
728 list_for_each_entry(l, &loop_devices, lo_list)
729 if (disk_to_dev(l->lo_disk) == dev) {
733 mutex_unlock(&loop_devices_mutex);
735 return lo ? callback(lo, page) : -EIO;
738 #define LOOP_ATTR_RO(_name) \
739 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
740 static ssize_t loop_attr_do_show_##_name(struct device *d, \
741 struct device_attribute *attr, char *b) \
743 return loop_attr_show(d, b, loop_attr_##_name##_show); \
745 static struct device_attribute loop_attr_##_name = \
746 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
748 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
753 spin_lock_irq(&lo->lo_lock);
754 if (lo->lo_backing_file)
755 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
756 spin_unlock_irq(&lo->lo_lock);
758 if (IS_ERR_OR_NULL(p))
762 memmove(buf, p, ret);
770 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
772 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
775 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
777 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
780 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
782 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
784 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
787 LOOP_ATTR_RO(backing_file);
788 LOOP_ATTR_RO(offset);
789 LOOP_ATTR_RO(sizelimit);
790 LOOP_ATTR_RO(autoclear);
792 static struct attribute *loop_attrs[] = {
793 &loop_attr_backing_file.attr,
794 &loop_attr_offset.attr,
795 &loop_attr_sizelimit.attr,
796 &loop_attr_autoclear.attr,
800 static struct attribute_group loop_attribute_group = {
805 static int loop_sysfs_init(struct loop_device *lo)
807 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
808 &loop_attribute_group);
811 static void loop_sysfs_exit(struct loop_device *lo)
813 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
814 &loop_attribute_group);
817 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
818 struct block_device *bdev, unsigned int arg)
820 struct file *file, *f;
822 struct address_space *mapping;
823 unsigned lo_blocksize;
828 /* This is safe, since we have a reference from open(). */
829 __module_get(THIS_MODULE);
837 if (lo->lo_state != Lo_unbound)
840 /* Avoid recursion */
842 while (is_loop_device(f)) {
843 struct loop_device *l;
845 if (f->f_mapping->host->i_bdev == bdev)
848 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
849 if (l->lo_state == Lo_unbound) {
853 f = l->lo_backing_file;
856 mapping = file->f_mapping;
857 inode = mapping->host;
859 if (!(file->f_mode & FMODE_WRITE))
860 lo_flags |= LO_FLAGS_READ_ONLY;
863 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
864 const struct address_space_operations *aops = mapping->a_ops;
866 if (aops->write_begin)
867 lo_flags |= LO_FLAGS_USE_AOPS;
868 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
869 lo_flags |= LO_FLAGS_READ_ONLY;
871 lo_blocksize = S_ISBLK(inode->i_mode) ?
872 inode->i_bdev->bd_block_size : PAGE_SIZE;
879 size = get_loop_size(lo, file);
881 if ((loff_t)(sector_t)size != size) {
886 if (!(mode & FMODE_WRITE))
887 lo_flags |= LO_FLAGS_READ_ONLY;
889 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
891 lo->lo_blocksize = lo_blocksize;
892 lo->lo_device = bdev;
893 lo->lo_flags = lo_flags;
894 lo->lo_backing_file = file;
895 lo->transfer = transfer_none;
897 lo->lo_sizelimit = 0;
898 lo->old_gfp_mask = mapping_gfp_mask(mapping);
899 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
901 bio_list_init(&lo->lo_bio_list);
904 * set queue make_request_fn, and add limits based on lower level
907 blk_queue_make_request(lo->lo_queue, loop_make_request);
908 lo->lo_queue->queuedata = lo;
910 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
911 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
913 set_capacity(lo->lo_disk, size);
914 bd_set_size(bdev, size << 9);
916 /* let user-space know about the new size */
917 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
919 set_blocksize(bdev, lo_blocksize);
921 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
923 if (IS_ERR(lo->lo_thread)) {
924 error = PTR_ERR(lo->lo_thread);
927 lo->lo_state = Lo_bound;
928 wake_up_process(lo->lo_thread);
930 ioctl_by_bdev(bdev, BLKRRPART, 0);
935 lo->lo_thread = NULL;
936 lo->lo_device = NULL;
937 lo->lo_backing_file = NULL;
939 set_capacity(lo->lo_disk, 0);
940 invalidate_bdev(bdev);
941 bd_set_size(bdev, 0);
942 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
943 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
944 lo->lo_state = Lo_unbound;
948 /* This is safe: open() is still holding a reference. */
949 module_put(THIS_MODULE);
954 loop_release_xfer(struct loop_device *lo)
957 struct loop_func_table *xfer = lo->lo_encryption;
961 err = xfer->release(lo);
963 lo->lo_encryption = NULL;
964 module_put(xfer->owner);
970 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
971 const struct loop_info64 *i)
976 struct module *owner = xfer->owner;
978 if (!try_module_get(owner))
981 err = xfer->init(lo, i);
985 lo->lo_encryption = xfer;
990 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
992 struct file *filp = lo->lo_backing_file;
993 gfp_t gfp = lo->old_gfp_mask;
995 if (lo->lo_state != Lo_bound)
998 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
1004 spin_lock_irq(&lo->lo_lock);
1005 lo->lo_state = Lo_rundown;
1006 spin_unlock_irq(&lo->lo_lock);
1008 kthread_stop(lo->lo_thread);
1010 spin_lock_irq(&lo->lo_lock);
1011 lo->lo_backing_file = NULL;
1012 spin_unlock_irq(&lo->lo_lock);
1014 loop_release_xfer(lo);
1015 lo->transfer = NULL;
1017 lo->lo_device = NULL;
1018 lo->lo_encryption = NULL;
1020 lo->lo_sizelimit = 0;
1021 lo->lo_encrypt_key_size = 0;
1023 lo->lo_thread = NULL;
1024 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1025 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1026 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1028 invalidate_bdev(bdev);
1029 set_capacity(lo->lo_disk, 0);
1030 loop_sysfs_exit(lo);
1032 bd_set_size(bdev, 0);
1033 /* let user-space know about this change */
1034 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1036 mapping_set_gfp_mask(filp->f_mapping, gfp);
1037 lo->lo_state = Lo_unbound;
1038 /* This is safe: open() is still holding a reference. */
1039 module_put(THIS_MODULE);
1040 if (max_part > 0 && bdev)
1041 ioctl_by_bdev(bdev, BLKRRPART, 0);
1042 mutex_unlock(&lo->lo_ctl_mutex);
1044 * Need not hold lo_ctl_mutex to fput backing file.
1045 * Calling fput holding lo_ctl_mutex triggers a circular
1046 * lock dependency possibility warning as fput can take
1047 * bd_mutex which is usually taken before lo_ctl_mutex.
1054 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1057 struct loop_func_table *xfer;
1058 uid_t uid = current_uid();
1060 if (lo->lo_encrypt_key_size &&
1061 lo->lo_key_owner != uid &&
1062 !capable(CAP_SYS_ADMIN))
1064 if (lo->lo_state != Lo_bound)
1066 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1069 err = loop_release_xfer(lo);
1073 if (info->lo_encrypt_type) {
1074 unsigned int type = info->lo_encrypt_type;
1076 if (type >= MAX_LO_CRYPT)
1078 xfer = xfer_funcs[type];
1084 err = loop_init_xfer(lo, xfer, info);
1088 if (lo->lo_offset != info->lo_offset ||
1089 lo->lo_sizelimit != info->lo_sizelimit) {
1090 lo->lo_offset = info->lo_offset;
1091 lo->lo_sizelimit = info->lo_sizelimit;
1092 if (figure_loop_size(lo))
1096 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1097 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1098 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1099 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1103 lo->transfer = xfer->transfer;
1104 lo->ioctl = xfer->ioctl;
1106 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1107 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1108 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1110 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1111 lo->lo_init[0] = info->lo_init[0];
1112 lo->lo_init[1] = info->lo_init[1];
1113 if (info->lo_encrypt_key_size) {
1114 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1115 info->lo_encrypt_key_size);
1116 lo->lo_key_owner = uid;
1123 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1125 struct file *file = lo->lo_backing_file;
1129 if (lo->lo_state != Lo_bound)
1131 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1134 memset(info, 0, sizeof(*info));
1135 info->lo_number = lo->lo_number;
1136 info->lo_device = huge_encode_dev(stat.dev);
1137 info->lo_inode = stat.ino;
1138 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1139 info->lo_offset = lo->lo_offset;
1140 info->lo_sizelimit = lo->lo_sizelimit;
1141 info->lo_flags = lo->lo_flags;
1142 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1143 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1144 info->lo_encrypt_type =
1145 lo->lo_encryption ? lo->lo_encryption->number : 0;
1146 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1147 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1148 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1149 lo->lo_encrypt_key_size);
1155 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1157 memset(info64, 0, sizeof(*info64));
1158 info64->lo_number = info->lo_number;
1159 info64->lo_device = info->lo_device;
1160 info64->lo_inode = info->lo_inode;
1161 info64->lo_rdevice = info->lo_rdevice;
1162 info64->lo_offset = info->lo_offset;
1163 info64->lo_sizelimit = 0;
1164 info64->lo_encrypt_type = info->lo_encrypt_type;
1165 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1166 info64->lo_flags = info->lo_flags;
1167 info64->lo_init[0] = info->lo_init[0];
1168 info64->lo_init[1] = info->lo_init[1];
1169 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1170 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1172 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1173 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1177 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1179 memset(info, 0, sizeof(*info));
1180 info->lo_number = info64->lo_number;
1181 info->lo_device = info64->lo_device;
1182 info->lo_inode = info64->lo_inode;
1183 info->lo_rdevice = info64->lo_rdevice;
1184 info->lo_offset = info64->lo_offset;
1185 info->lo_encrypt_type = info64->lo_encrypt_type;
1186 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1187 info->lo_flags = info64->lo_flags;
1188 info->lo_init[0] = info64->lo_init[0];
1189 info->lo_init[1] = info64->lo_init[1];
1190 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1191 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1193 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1194 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1196 /* error in case values were truncated */
1197 if (info->lo_device != info64->lo_device ||
1198 info->lo_rdevice != info64->lo_rdevice ||
1199 info->lo_inode != info64->lo_inode ||
1200 info->lo_offset != info64->lo_offset)
1207 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1209 struct loop_info info;
1210 struct loop_info64 info64;
1212 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1214 loop_info64_from_old(&info, &info64);
1215 return loop_set_status(lo, &info64);
1219 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1221 struct loop_info64 info64;
1223 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1225 return loop_set_status(lo, &info64);
1229 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1230 struct loop_info info;
1231 struct loop_info64 info64;
1237 err = loop_get_status(lo, &info64);
1239 err = loop_info64_to_old(&info64, &info);
1240 if (!err && copy_to_user(arg, &info, sizeof(info)))
1247 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1248 struct loop_info64 info64;
1254 err = loop_get_status(lo, &info64);
1255 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1261 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1268 if (unlikely(lo->lo_state != Lo_bound))
1270 err = figure_loop_size(lo);
1273 sec = get_capacity(lo->lo_disk);
1274 /* the width of sector_t may be narrow for bit-shift */
1277 mutex_lock(&bdev->bd_mutex);
1278 bd_set_size(bdev, sz);
1279 /* let user-space know about the new size */
1280 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1281 mutex_unlock(&bdev->bd_mutex);
1287 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1288 unsigned int cmd, unsigned long arg)
1290 struct loop_device *lo = bdev->bd_disk->private_data;
1293 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1296 err = loop_set_fd(lo, mode, bdev, arg);
1298 case LOOP_CHANGE_FD:
1299 err = loop_change_fd(lo, bdev, arg);
1302 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1303 err = loop_clr_fd(lo, bdev);
1307 case LOOP_SET_STATUS:
1308 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1310 case LOOP_GET_STATUS:
1311 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1313 case LOOP_SET_STATUS64:
1314 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1316 case LOOP_GET_STATUS64:
1317 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1319 case LOOP_SET_CAPACITY:
1321 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1322 err = loop_set_capacity(lo, bdev);
1325 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1327 mutex_unlock(&lo->lo_ctl_mutex);
1333 #ifdef CONFIG_COMPAT
1334 struct compat_loop_info {
1335 compat_int_t lo_number; /* ioctl r/o */
1336 compat_dev_t lo_device; /* ioctl r/o */
1337 compat_ulong_t lo_inode; /* ioctl r/o */
1338 compat_dev_t lo_rdevice; /* ioctl r/o */
1339 compat_int_t lo_offset;
1340 compat_int_t lo_encrypt_type;
1341 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1342 compat_int_t lo_flags; /* ioctl r/o */
1343 char lo_name[LO_NAME_SIZE];
1344 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1345 compat_ulong_t lo_init[2];
1350 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1351 * - noinlined to reduce stack space usage in main part of driver
1354 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1355 struct loop_info64 *info64)
1357 struct compat_loop_info info;
1359 if (copy_from_user(&info, arg, sizeof(info)))
1362 memset(info64, 0, sizeof(*info64));
1363 info64->lo_number = info.lo_number;
1364 info64->lo_device = info.lo_device;
1365 info64->lo_inode = info.lo_inode;
1366 info64->lo_rdevice = info.lo_rdevice;
1367 info64->lo_offset = info.lo_offset;
1368 info64->lo_sizelimit = 0;
1369 info64->lo_encrypt_type = info.lo_encrypt_type;
1370 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1371 info64->lo_flags = info.lo_flags;
1372 info64->lo_init[0] = info.lo_init[0];
1373 info64->lo_init[1] = info.lo_init[1];
1374 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1375 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1377 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1378 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1383 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1384 * - noinlined to reduce stack space usage in main part of driver
1387 loop_info64_to_compat(const struct loop_info64 *info64,
1388 struct compat_loop_info __user *arg)
1390 struct compat_loop_info info;
1392 memset(&info, 0, sizeof(info));
1393 info.lo_number = info64->lo_number;
1394 info.lo_device = info64->lo_device;
1395 info.lo_inode = info64->lo_inode;
1396 info.lo_rdevice = info64->lo_rdevice;
1397 info.lo_offset = info64->lo_offset;
1398 info.lo_encrypt_type = info64->lo_encrypt_type;
1399 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1400 info.lo_flags = info64->lo_flags;
1401 info.lo_init[0] = info64->lo_init[0];
1402 info.lo_init[1] = info64->lo_init[1];
1403 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1404 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1406 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1407 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1409 /* error in case values were truncated */
1410 if (info.lo_device != info64->lo_device ||
1411 info.lo_rdevice != info64->lo_rdevice ||
1412 info.lo_inode != info64->lo_inode ||
1413 info.lo_offset != info64->lo_offset ||
1414 info.lo_init[0] != info64->lo_init[0] ||
1415 info.lo_init[1] != info64->lo_init[1])
1418 if (copy_to_user(arg, &info, sizeof(info)))
1424 loop_set_status_compat(struct loop_device *lo,
1425 const struct compat_loop_info __user *arg)
1427 struct loop_info64 info64;
1430 ret = loop_info64_from_compat(arg, &info64);
1433 return loop_set_status(lo, &info64);
1437 loop_get_status_compat(struct loop_device *lo,
1438 struct compat_loop_info __user *arg)
1440 struct loop_info64 info64;
1446 err = loop_get_status(lo, &info64);
1448 err = loop_info64_to_compat(&info64, arg);
1452 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1453 unsigned int cmd, unsigned long arg)
1455 struct loop_device *lo = bdev->bd_disk->private_data;
1459 case LOOP_SET_STATUS:
1460 mutex_lock(&lo->lo_ctl_mutex);
1461 err = loop_set_status_compat(
1462 lo, (const struct compat_loop_info __user *) arg);
1463 mutex_unlock(&lo->lo_ctl_mutex);
1465 case LOOP_GET_STATUS:
1466 mutex_lock(&lo->lo_ctl_mutex);
1467 err = loop_get_status_compat(
1468 lo, (struct compat_loop_info __user *) arg);
1469 mutex_unlock(&lo->lo_ctl_mutex);
1471 case LOOP_SET_CAPACITY:
1473 case LOOP_GET_STATUS64:
1474 case LOOP_SET_STATUS64:
1475 arg = (unsigned long) compat_ptr(arg);
1477 case LOOP_CHANGE_FD:
1478 err = lo_ioctl(bdev, mode, cmd, arg);
1488 static int lo_open(struct block_device *bdev, fmode_t mode)
1490 struct loop_device *lo = bdev->bd_disk->private_data;
1492 mutex_lock(&lo->lo_ctl_mutex);
1494 mutex_unlock(&lo->lo_ctl_mutex);
1499 static int lo_release(struct gendisk *disk, fmode_t mode)
1501 struct loop_device *lo = disk->private_data;
1504 mutex_lock(&lo->lo_ctl_mutex);
1506 if (--lo->lo_refcnt)
1509 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1511 * In autoclear mode, stop the loop thread
1512 * and remove configuration after last close.
1514 err = loop_clr_fd(lo, NULL);
1519 * Otherwise keep thread (if running) and config,
1520 * but flush possible ongoing bios in thread.
1526 mutex_unlock(&lo->lo_ctl_mutex);
1531 static const struct block_device_operations lo_fops = {
1532 .owner = THIS_MODULE,
1534 .release = lo_release,
1536 #ifdef CONFIG_COMPAT
1537 .compat_ioctl = lo_compat_ioctl,
1542 * And now the modules code and kernel interface.
1544 static int max_loop;
1545 module_param(max_loop, int, S_IRUGO);
1546 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1547 module_param(max_part, int, S_IRUGO);
1548 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1549 MODULE_LICENSE("GPL");
1550 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1552 int loop_register_transfer(struct loop_func_table *funcs)
1554 unsigned int n = funcs->number;
1556 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1558 xfer_funcs[n] = funcs;
1562 int loop_unregister_transfer(int number)
1564 unsigned int n = number;
1565 struct loop_device *lo;
1566 struct loop_func_table *xfer;
1568 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1571 xfer_funcs[n] = NULL;
1573 list_for_each_entry(lo, &loop_devices, lo_list) {
1574 mutex_lock(&lo->lo_ctl_mutex);
1576 if (lo->lo_encryption == xfer)
1577 loop_release_xfer(lo);
1579 mutex_unlock(&lo->lo_ctl_mutex);
1585 EXPORT_SYMBOL(loop_register_transfer);
1586 EXPORT_SYMBOL(loop_unregister_transfer);
1588 static struct loop_device *loop_alloc(int i)
1590 struct loop_device *lo;
1591 struct gendisk *disk;
1593 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1597 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1601 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1603 goto out_free_queue;
1605 mutex_init(&lo->lo_ctl_mutex);
1607 lo->lo_thread = NULL;
1608 init_waitqueue_head(&lo->lo_event);
1609 spin_lock_init(&lo->lo_lock);
1610 disk->major = LOOP_MAJOR;
1611 disk->first_minor = i << part_shift;
1612 disk->fops = &lo_fops;
1613 disk->private_data = lo;
1614 disk->queue = lo->lo_queue;
1615 sprintf(disk->disk_name, "loop%d", i);
1619 blk_cleanup_queue(lo->lo_queue);
1626 static void loop_free(struct loop_device *lo)
1628 blk_cleanup_queue(lo->lo_queue);
1629 put_disk(lo->lo_disk);
1630 list_del(&lo->lo_list);
1634 static struct loop_device *loop_init_one(int i)
1636 struct loop_device *lo;
1638 list_for_each_entry(lo, &loop_devices, lo_list) {
1639 if (lo->lo_number == i)
1645 add_disk(lo->lo_disk);
1646 list_add_tail(&lo->lo_list, &loop_devices);
1651 static void loop_del_one(struct loop_device *lo)
1653 del_gendisk(lo->lo_disk);
1657 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1659 struct loop_device *lo;
1660 struct kobject *kobj;
1662 mutex_lock(&loop_devices_mutex);
1663 lo = loop_init_one(MINOR(dev) >> part_shift);
1664 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1665 mutex_unlock(&loop_devices_mutex);
1671 static int __init loop_init(void)
1674 unsigned long range;
1675 struct loop_device *lo, *next;
1678 * loop module now has a feature to instantiate underlying device
1679 * structure on-demand, provided that there is an access dev node.
1680 * However, this will not work well with user space tool that doesn't
1681 * know about such "feature". In order to not break any existing
1682 * tool, we do the following:
1684 * (1) if max_loop is specified, create that many upfront, and this
1685 * also becomes a hard limit.
1686 * (2) if max_loop is not specified, create 8 loop device on module
1687 * load, user can further extend loop device by create dev node
1688 * themselves and have kernel automatically instantiate actual
1694 part_shift = fls(max_part);
1697 * Adjust max_part according to part_shift as it is exported
1698 * to user space so that user can decide correct minor number
1699 * if [s]he want to create more devices.
1701 * Note that -1 is required because partition 0 is reserved
1702 * for the whole disk.
1704 max_part = (1UL << part_shift) - 1;
1707 if ((1UL << part_shift) > DISK_MAX_PARTS)
1710 if (max_loop > 1UL << (MINORBITS - part_shift))
1715 range = max_loop << part_shift;
1718 range = 1UL << MINORBITS;
1721 if (register_blkdev(LOOP_MAJOR, "loop"))
1724 for (i = 0; i < nr; i++) {
1728 list_add_tail(&lo->lo_list, &loop_devices);
1731 /* point of no return */
1733 list_for_each_entry(lo, &loop_devices, lo_list)
1734 add_disk(lo->lo_disk);
1736 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1737 THIS_MODULE, loop_probe, NULL, NULL);
1739 printk(KERN_INFO "loop: module loaded\n");
1743 printk(KERN_INFO "loop: out of memory\n");
1745 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1748 unregister_blkdev(LOOP_MAJOR, "loop");
1752 static void __exit loop_exit(void)
1754 unsigned long range;
1755 struct loop_device *lo, *next;
1757 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1759 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1762 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1763 unregister_blkdev(LOOP_MAJOR, "loop");
1766 module_init(loop_init);
1767 module_exit(loop_exit);
1770 static int __init max_loop_setup(char *str)
1772 max_loop = simple_strtol(str, NULL, 0);
1776 __setup("max_loop=", max_loop_setup);