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 mutex_lock(&lo->lo_ctl_mutex);
754 if (lo->lo_backing_file)
755 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
756 mutex_unlock(&lo->lo_ctl_mutex);
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 lo->lo_backing_file = NULL;
1012 loop_release_xfer(lo);
1013 lo->transfer = NULL;
1015 lo->lo_device = NULL;
1016 lo->lo_encryption = NULL;
1018 lo->lo_sizelimit = 0;
1019 lo->lo_encrypt_key_size = 0;
1021 lo->lo_thread = NULL;
1022 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1023 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1024 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1026 invalidate_bdev(bdev);
1027 set_capacity(lo->lo_disk, 0);
1028 loop_sysfs_exit(lo);
1030 bd_set_size(bdev, 0);
1031 /* let user-space know about this change */
1032 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1034 mapping_set_gfp_mask(filp->f_mapping, gfp);
1035 lo->lo_state = Lo_unbound;
1036 /* This is safe: open() is still holding a reference. */
1037 module_put(THIS_MODULE);
1038 if (max_part > 0 && bdev)
1039 ioctl_by_bdev(bdev, BLKRRPART, 0);
1040 mutex_unlock(&lo->lo_ctl_mutex);
1042 * Need not hold lo_ctl_mutex to fput backing file.
1043 * Calling fput holding lo_ctl_mutex triggers a circular
1044 * lock dependency possibility warning as fput can take
1045 * bd_mutex which is usually taken before lo_ctl_mutex.
1052 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1055 struct loop_func_table *xfer;
1056 uid_t uid = current_uid();
1058 if (lo->lo_encrypt_key_size &&
1059 lo->lo_key_owner != uid &&
1060 !capable(CAP_SYS_ADMIN))
1062 if (lo->lo_state != Lo_bound)
1064 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1067 err = loop_release_xfer(lo);
1071 if (info->lo_encrypt_type) {
1072 unsigned int type = info->lo_encrypt_type;
1074 if (type >= MAX_LO_CRYPT)
1076 xfer = xfer_funcs[type];
1082 err = loop_init_xfer(lo, xfer, info);
1086 if (lo->lo_offset != info->lo_offset ||
1087 lo->lo_sizelimit != info->lo_sizelimit) {
1088 lo->lo_offset = info->lo_offset;
1089 lo->lo_sizelimit = info->lo_sizelimit;
1090 if (figure_loop_size(lo))
1094 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1095 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1096 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1097 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1101 lo->transfer = xfer->transfer;
1102 lo->ioctl = xfer->ioctl;
1104 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1105 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1106 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1108 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1109 lo->lo_init[0] = info->lo_init[0];
1110 lo->lo_init[1] = info->lo_init[1];
1111 if (info->lo_encrypt_key_size) {
1112 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1113 info->lo_encrypt_key_size);
1114 lo->lo_key_owner = uid;
1121 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1123 struct file *file = lo->lo_backing_file;
1127 if (lo->lo_state != Lo_bound)
1129 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1132 memset(info, 0, sizeof(*info));
1133 info->lo_number = lo->lo_number;
1134 info->lo_device = huge_encode_dev(stat.dev);
1135 info->lo_inode = stat.ino;
1136 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1137 info->lo_offset = lo->lo_offset;
1138 info->lo_sizelimit = lo->lo_sizelimit;
1139 info->lo_flags = lo->lo_flags;
1140 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1141 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1142 info->lo_encrypt_type =
1143 lo->lo_encryption ? lo->lo_encryption->number : 0;
1144 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1145 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1146 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1147 lo->lo_encrypt_key_size);
1153 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1155 memset(info64, 0, sizeof(*info64));
1156 info64->lo_number = info->lo_number;
1157 info64->lo_device = info->lo_device;
1158 info64->lo_inode = info->lo_inode;
1159 info64->lo_rdevice = info->lo_rdevice;
1160 info64->lo_offset = info->lo_offset;
1161 info64->lo_sizelimit = 0;
1162 info64->lo_encrypt_type = info->lo_encrypt_type;
1163 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1164 info64->lo_flags = info->lo_flags;
1165 info64->lo_init[0] = info->lo_init[0];
1166 info64->lo_init[1] = info->lo_init[1];
1167 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1168 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1170 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1171 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1175 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1177 memset(info, 0, sizeof(*info));
1178 info->lo_number = info64->lo_number;
1179 info->lo_device = info64->lo_device;
1180 info->lo_inode = info64->lo_inode;
1181 info->lo_rdevice = info64->lo_rdevice;
1182 info->lo_offset = info64->lo_offset;
1183 info->lo_encrypt_type = info64->lo_encrypt_type;
1184 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1185 info->lo_flags = info64->lo_flags;
1186 info->lo_init[0] = info64->lo_init[0];
1187 info->lo_init[1] = info64->lo_init[1];
1188 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1189 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1191 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1192 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1194 /* error in case values were truncated */
1195 if (info->lo_device != info64->lo_device ||
1196 info->lo_rdevice != info64->lo_rdevice ||
1197 info->lo_inode != info64->lo_inode ||
1198 info->lo_offset != info64->lo_offset)
1205 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1207 struct loop_info info;
1208 struct loop_info64 info64;
1210 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1212 loop_info64_from_old(&info, &info64);
1213 return loop_set_status(lo, &info64);
1217 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1219 struct loop_info64 info64;
1221 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1223 return loop_set_status(lo, &info64);
1227 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1228 struct loop_info info;
1229 struct loop_info64 info64;
1235 err = loop_get_status(lo, &info64);
1237 err = loop_info64_to_old(&info64, &info);
1238 if (!err && copy_to_user(arg, &info, sizeof(info)))
1245 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1246 struct loop_info64 info64;
1252 err = loop_get_status(lo, &info64);
1253 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1259 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1266 if (unlikely(lo->lo_state != Lo_bound))
1268 err = figure_loop_size(lo);
1271 sec = get_capacity(lo->lo_disk);
1272 /* the width of sector_t may be narrow for bit-shift */
1275 mutex_lock(&bdev->bd_mutex);
1276 bd_set_size(bdev, sz);
1277 /* let user-space know about the new size */
1278 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1279 mutex_unlock(&bdev->bd_mutex);
1285 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1286 unsigned int cmd, unsigned long arg)
1288 struct loop_device *lo = bdev->bd_disk->private_data;
1291 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1294 err = loop_set_fd(lo, mode, bdev, arg);
1296 case LOOP_CHANGE_FD:
1297 err = loop_change_fd(lo, bdev, arg);
1300 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1301 err = loop_clr_fd(lo, bdev);
1305 case LOOP_SET_STATUS:
1306 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1308 case LOOP_GET_STATUS:
1309 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1311 case LOOP_SET_STATUS64:
1312 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1314 case LOOP_GET_STATUS64:
1315 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1317 case LOOP_SET_CAPACITY:
1319 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1320 err = loop_set_capacity(lo, bdev);
1323 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1325 mutex_unlock(&lo->lo_ctl_mutex);
1331 #ifdef CONFIG_COMPAT
1332 struct compat_loop_info {
1333 compat_int_t lo_number; /* ioctl r/o */
1334 compat_dev_t lo_device; /* ioctl r/o */
1335 compat_ulong_t lo_inode; /* ioctl r/o */
1336 compat_dev_t lo_rdevice; /* ioctl r/o */
1337 compat_int_t lo_offset;
1338 compat_int_t lo_encrypt_type;
1339 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1340 compat_int_t lo_flags; /* ioctl r/o */
1341 char lo_name[LO_NAME_SIZE];
1342 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1343 compat_ulong_t lo_init[2];
1348 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1349 * - noinlined to reduce stack space usage in main part of driver
1352 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1353 struct loop_info64 *info64)
1355 struct compat_loop_info info;
1357 if (copy_from_user(&info, arg, sizeof(info)))
1360 memset(info64, 0, sizeof(*info64));
1361 info64->lo_number = info.lo_number;
1362 info64->lo_device = info.lo_device;
1363 info64->lo_inode = info.lo_inode;
1364 info64->lo_rdevice = info.lo_rdevice;
1365 info64->lo_offset = info.lo_offset;
1366 info64->lo_sizelimit = 0;
1367 info64->lo_encrypt_type = info.lo_encrypt_type;
1368 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1369 info64->lo_flags = info.lo_flags;
1370 info64->lo_init[0] = info.lo_init[0];
1371 info64->lo_init[1] = info.lo_init[1];
1372 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1373 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1375 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1376 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1381 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1382 * - noinlined to reduce stack space usage in main part of driver
1385 loop_info64_to_compat(const struct loop_info64 *info64,
1386 struct compat_loop_info __user *arg)
1388 struct compat_loop_info info;
1390 memset(&info, 0, sizeof(info));
1391 info.lo_number = info64->lo_number;
1392 info.lo_device = info64->lo_device;
1393 info.lo_inode = info64->lo_inode;
1394 info.lo_rdevice = info64->lo_rdevice;
1395 info.lo_offset = info64->lo_offset;
1396 info.lo_encrypt_type = info64->lo_encrypt_type;
1397 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1398 info.lo_flags = info64->lo_flags;
1399 info.lo_init[0] = info64->lo_init[0];
1400 info.lo_init[1] = info64->lo_init[1];
1401 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1402 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1404 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1405 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1407 /* error in case values were truncated */
1408 if (info.lo_device != info64->lo_device ||
1409 info.lo_rdevice != info64->lo_rdevice ||
1410 info.lo_inode != info64->lo_inode ||
1411 info.lo_offset != info64->lo_offset ||
1412 info.lo_init[0] != info64->lo_init[0] ||
1413 info.lo_init[1] != info64->lo_init[1])
1416 if (copy_to_user(arg, &info, sizeof(info)))
1422 loop_set_status_compat(struct loop_device *lo,
1423 const struct compat_loop_info __user *arg)
1425 struct loop_info64 info64;
1428 ret = loop_info64_from_compat(arg, &info64);
1431 return loop_set_status(lo, &info64);
1435 loop_get_status_compat(struct loop_device *lo,
1436 struct compat_loop_info __user *arg)
1438 struct loop_info64 info64;
1444 err = loop_get_status(lo, &info64);
1446 err = loop_info64_to_compat(&info64, arg);
1450 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1451 unsigned int cmd, unsigned long arg)
1453 struct loop_device *lo = bdev->bd_disk->private_data;
1457 case LOOP_SET_STATUS:
1458 mutex_lock(&lo->lo_ctl_mutex);
1459 err = loop_set_status_compat(
1460 lo, (const struct compat_loop_info __user *) arg);
1461 mutex_unlock(&lo->lo_ctl_mutex);
1463 case LOOP_GET_STATUS:
1464 mutex_lock(&lo->lo_ctl_mutex);
1465 err = loop_get_status_compat(
1466 lo, (struct compat_loop_info __user *) arg);
1467 mutex_unlock(&lo->lo_ctl_mutex);
1469 case LOOP_SET_CAPACITY:
1471 case LOOP_GET_STATUS64:
1472 case LOOP_SET_STATUS64:
1473 arg = (unsigned long) compat_ptr(arg);
1475 case LOOP_CHANGE_FD:
1476 err = lo_ioctl(bdev, mode, cmd, arg);
1486 static int lo_open(struct block_device *bdev, fmode_t mode)
1488 struct loop_device *lo = bdev->bd_disk->private_data;
1490 mutex_lock(&lo->lo_ctl_mutex);
1492 mutex_unlock(&lo->lo_ctl_mutex);
1497 static int lo_release(struct gendisk *disk, fmode_t mode)
1499 struct loop_device *lo = disk->private_data;
1502 mutex_lock(&lo->lo_ctl_mutex);
1504 if (--lo->lo_refcnt)
1507 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1509 * In autoclear mode, stop the loop thread
1510 * and remove configuration after last close.
1512 err = loop_clr_fd(lo, NULL);
1517 * Otherwise keep thread (if running) and config,
1518 * but flush possible ongoing bios in thread.
1524 mutex_unlock(&lo->lo_ctl_mutex);
1529 static const struct block_device_operations lo_fops = {
1530 .owner = THIS_MODULE,
1532 .release = lo_release,
1534 #ifdef CONFIG_COMPAT
1535 .compat_ioctl = lo_compat_ioctl,
1540 * And now the modules code and kernel interface.
1542 static int max_loop;
1543 module_param(max_loop, int, S_IRUGO);
1544 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1545 module_param(max_part, int, S_IRUGO);
1546 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1547 MODULE_LICENSE("GPL");
1548 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1550 int loop_register_transfer(struct loop_func_table *funcs)
1552 unsigned int n = funcs->number;
1554 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1556 xfer_funcs[n] = funcs;
1560 int loop_unregister_transfer(int number)
1562 unsigned int n = number;
1563 struct loop_device *lo;
1564 struct loop_func_table *xfer;
1566 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1569 xfer_funcs[n] = NULL;
1571 list_for_each_entry(lo, &loop_devices, lo_list) {
1572 mutex_lock(&lo->lo_ctl_mutex);
1574 if (lo->lo_encryption == xfer)
1575 loop_release_xfer(lo);
1577 mutex_unlock(&lo->lo_ctl_mutex);
1583 EXPORT_SYMBOL(loop_register_transfer);
1584 EXPORT_SYMBOL(loop_unregister_transfer);
1586 static struct loop_device *loop_alloc(int i)
1588 struct loop_device *lo;
1589 struct gendisk *disk;
1591 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1595 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1599 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1601 goto out_free_queue;
1603 mutex_init(&lo->lo_ctl_mutex);
1605 lo->lo_thread = NULL;
1606 init_waitqueue_head(&lo->lo_event);
1607 spin_lock_init(&lo->lo_lock);
1608 disk->major = LOOP_MAJOR;
1609 disk->first_minor = i << part_shift;
1610 disk->fops = &lo_fops;
1611 disk->private_data = lo;
1612 disk->queue = lo->lo_queue;
1613 sprintf(disk->disk_name, "loop%d", i);
1617 blk_cleanup_queue(lo->lo_queue);
1624 static void loop_free(struct loop_device *lo)
1626 blk_cleanup_queue(lo->lo_queue);
1627 put_disk(lo->lo_disk);
1628 list_del(&lo->lo_list);
1632 static struct loop_device *loop_init_one(int i)
1634 struct loop_device *lo;
1636 list_for_each_entry(lo, &loop_devices, lo_list) {
1637 if (lo->lo_number == i)
1643 add_disk(lo->lo_disk);
1644 list_add_tail(&lo->lo_list, &loop_devices);
1649 static void loop_del_one(struct loop_device *lo)
1651 del_gendisk(lo->lo_disk);
1655 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1657 struct loop_device *lo;
1658 struct kobject *kobj;
1660 mutex_lock(&loop_devices_mutex);
1661 lo = loop_init_one(MINOR(dev) >> part_shift);
1662 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1663 mutex_unlock(&loop_devices_mutex);
1669 static int __init loop_init(void)
1672 unsigned long range;
1673 struct loop_device *lo, *next;
1676 * loop module now has a feature to instantiate underlying device
1677 * structure on-demand, provided that there is an access dev node.
1678 * However, this will not work well with user space tool that doesn't
1679 * know about such "feature". In order to not break any existing
1680 * tool, we do the following:
1682 * (1) if max_loop is specified, create that many upfront, and this
1683 * also becomes a hard limit.
1684 * (2) if max_loop is not specified, create 8 loop device on module
1685 * load, user can further extend loop device by create dev node
1686 * themselves and have kernel automatically instantiate actual
1692 part_shift = fls(max_part);
1695 * Adjust max_part according to part_shift as it is exported
1696 * to user space so that user can decide correct minor number
1697 * if [s]he want to create more devices.
1699 * Note that -1 is required because partition 0 is reserved
1700 * for the whole disk.
1702 max_part = (1UL << part_shift) - 1;
1705 if ((1UL << part_shift) > DISK_MAX_PARTS)
1708 if (max_loop > 1UL << (MINORBITS - part_shift))
1713 range = max_loop << part_shift;
1716 range = 1UL << MINORBITS;
1719 if (register_blkdev(LOOP_MAJOR, "loop"))
1722 for (i = 0; i < nr; i++) {
1726 list_add_tail(&lo->lo_list, &loop_devices);
1729 /* point of no return */
1731 list_for_each_entry(lo, &loop_devices, lo_list)
1732 add_disk(lo->lo_disk);
1734 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1735 THIS_MODULE, loop_probe, NULL, NULL);
1737 printk(KERN_INFO "loop: module loaded\n");
1741 printk(KERN_INFO "loop: out of memory\n");
1743 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1746 unregister_blkdev(LOOP_MAJOR, "loop");
1750 static void __exit loop_exit(void)
1752 unsigned long range;
1753 struct loop_device *lo, *next;
1755 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1757 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1760 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1761 unregister_blkdev(LOOP_MAJOR, "loop");
1764 module_init(loop_init);
1765 module_exit(loop_exit);
1768 static int __init max_loop_setup(char *str)
1770 max_loop = simple_strtol(str, NULL, 0);
1774 __setup("max_loop=", max_loop_setup);