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>
78 #include <linux/miscdevice.h>
79 #include <asm/uaccess.h>
81 static DEFINE_IDR(loop_index_idr);
82 static DEFINE_MUTEX(loop_index_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_write - helper for writing data to a loop device
208 * This helper just factors out common code between do_lo_send_direct_write()
209 * and do_lo_send_write().
211 static int __do_lo_send_write(struct file *file,
212 u8 *buf, const int len, loff_t pos)
215 mm_segment_t old_fs = get_fs();
218 bw = file->f_op->write(file, buf, len, &pos);
220 if (likely(bw == len))
222 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
223 (unsigned long long)pos, len);
230 * do_lo_send_direct_write - helper for writing data to a loop device
232 * This is the fast, non-transforming version that does not need double
235 static int do_lo_send_direct_write(struct loop_device *lo,
236 struct bio_vec *bvec, loff_t pos, struct page *page)
238 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
239 kmap(bvec->bv_page) + bvec->bv_offset,
241 kunmap(bvec->bv_page);
247 * do_lo_send_write - helper for writing data to a loop device
249 * This is the slow, transforming version that needs to double buffer the
250 * data as it cannot do the transformations in place without having direct
251 * access to the destination pages of the backing file.
253 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
254 loff_t pos, struct page *page)
256 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
257 bvec->bv_offset, bvec->bv_len, pos >> 9);
259 return __do_lo_send_write(lo->lo_backing_file,
260 page_address(page), bvec->bv_len,
262 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
263 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
269 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
271 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
273 struct bio_vec *bvec;
274 struct page *page = NULL;
277 if (lo->transfer != transfer_none) {
278 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
282 do_lo_send = do_lo_send_write;
284 do_lo_send = do_lo_send_direct_write;
287 bio_for_each_segment(bvec, bio, i) {
288 ret = do_lo_send(lo, bvec, pos, page);
300 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
305 struct lo_read_data {
306 struct loop_device *lo;
313 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
314 struct splice_desc *sd)
316 struct lo_read_data *p = sd->u.data;
317 struct loop_device *lo = p->lo;
318 struct page *page = buf->page;
322 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
328 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
329 printk(KERN_ERR "loop: transfer error block %ld\n",
334 flush_dcache_page(p->page);
343 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
345 return __splice_from_pipe(pipe, sd, lo_splice_actor);
349 do_lo_receive(struct loop_device *lo,
350 struct bio_vec *bvec, int bsize, loff_t pos)
352 struct lo_read_data cookie;
353 struct splice_desc sd;
358 cookie.page = bvec->bv_page;
359 cookie.offset = bvec->bv_offset;
360 cookie.bsize = bsize;
363 sd.total_len = bvec->bv_len;
368 file = lo->lo_backing_file;
369 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
378 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
380 struct bio_vec *bvec;
383 bio_for_each_segment(bvec, bio, i) {
384 ret = do_lo_receive(lo, bvec, bsize, pos);
392 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
397 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
399 if (bio_rw(bio) == WRITE) {
400 struct file *file = lo->lo_backing_file;
402 if (bio->bi_rw & REQ_FLUSH) {
403 ret = vfs_fsync(file, 0);
404 if (unlikely(ret && ret != -EINVAL)) {
410 ret = lo_send(lo, bio, pos);
412 if ((bio->bi_rw & REQ_FUA) && !ret) {
413 ret = vfs_fsync(file, 0);
414 if (unlikely(ret && ret != -EINVAL))
418 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
425 * Add bio to back of pending list
427 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
429 bio_list_add(&lo->lo_bio_list, bio);
433 * Grab first pending buffer
435 static struct bio *loop_get_bio(struct loop_device *lo)
437 return bio_list_pop(&lo->lo_bio_list);
440 static void loop_make_request(struct request_queue *q, struct bio *old_bio)
442 struct loop_device *lo = q->queuedata;
443 int rw = bio_rw(old_bio);
448 BUG_ON(!lo || (rw != READ && rw != WRITE));
450 spin_lock_irq(&lo->lo_lock);
451 if (lo->lo_state != Lo_bound)
453 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
455 loop_add_bio(lo, old_bio);
456 wake_up(&lo->lo_event);
457 spin_unlock_irq(&lo->lo_lock);
461 spin_unlock_irq(&lo->lo_lock);
462 bio_io_error(old_bio);
465 struct switch_request {
467 struct completion wait;
470 static void do_loop_switch(struct loop_device *, struct switch_request *);
472 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
474 if (unlikely(!bio->bi_bdev)) {
475 do_loop_switch(lo, bio->bi_private);
478 int ret = do_bio_filebacked(lo, bio);
484 * worker thread that handles reads/writes to file backed loop devices,
485 * to avoid blocking in our make_request_fn. it also does loop decrypting
486 * on reads for block backed loop, as that is too heavy to do from
487 * b_end_io context where irqs may be disabled.
489 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
490 * calling kthread_stop(). Therefore once kthread_should_stop() is
491 * true, make_request will not place any more requests. Therefore
492 * once kthread_should_stop() is true and lo_bio is NULL, we are
493 * done with the loop.
495 static int loop_thread(void *data)
497 struct loop_device *lo = data;
500 set_user_nice(current, -20);
502 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
504 wait_event_interruptible(lo->lo_event,
505 !bio_list_empty(&lo->lo_bio_list) ||
506 kthread_should_stop());
508 if (bio_list_empty(&lo->lo_bio_list))
510 spin_lock_irq(&lo->lo_lock);
511 bio = loop_get_bio(lo);
512 spin_unlock_irq(&lo->lo_lock);
515 loop_handle_bio(lo, bio);
522 * loop_switch performs the hard work of switching a backing store.
523 * First it needs to flush existing IO, it does this by sending a magic
524 * BIO down the pipe. The completion of this BIO does the actual switch.
526 static int loop_switch(struct loop_device *lo, struct file *file)
528 struct switch_request w;
529 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
532 init_completion(&w.wait);
534 bio->bi_private = &w;
536 loop_make_request(lo->lo_queue, bio);
537 wait_for_completion(&w.wait);
542 * Helper to flush the IOs in loop, but keeping loop thread running
544 static int loop_flush(struct loop_device *lo)
546 /* loop not yet configured, no running thread, nothing to flush */
550 return loop_switch(lo, NULL);
554 * Do the actual switch; called from the BIO completion routine
556 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
558 struct file *file = p->file;
559 struct file *old_file = lo->lo_backing_file;
560 struct address_space *mapping;
562 /* if no new file, only flush of queued bios requested */
566 mapping = file->f_mapping;
567 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
568 lo->lo_backing_file = file;
569 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
570 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
571 lo->old_gfp_mask = mapping_gfp_mask(mapping);
572 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
579 * loop_change_fd switched the backing store of a loopback device to
580 * a new file. This is useful for operating system installers to free up
581 * the original file and in High Availability environments to switch to
582 * an alternative location for the content in case of server meltdown.
583 * This can only work if the loop device is used read-only, and if the
584 * new backing store is the same size and type as the old backing store.
586 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
589 struct file *file, *old_file;
594 if (lo->lo_state != Lo_bound)
597 /* the loop device has to be read-only */
599 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
607 inode = file->f_mapping->host;
608 old_file = lo->lo_backing_file;
612 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
615 /* size of the new backing store needs to be the same */
616 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
620 error = loop_switch(lo, file);
626 ioctl_by_bdev(bdev, BLKRRPART, 0);
635 static inline int is_loop_device(struct file *file)
637 struct inode *i = file->f_mapping->host;
639 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
642 /* loop sysfs attributes */
644 static ssize_t loop_attr_show(struct device *dev, char *page,
645 ssize_t (*callback)(struct loop_device *, char *))
647 struct gendisk *disk = dev_to_disk(dev);
648 struct loop_device *lo = disk->private_data;
650 return callback(lo, page);
653 #define LOOP_ATTR_RO(_name) \
654 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
655 static ssize_t loop_attr_do_show_##_name(struct device *d, \
656 struct device_attribute *attr, char *b) \
658 return loop_attr_show(d, b, loop_attr_##_name##_show); \
660 static struct device_attribute loop_attr_##_name = \
661 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
663 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
668 spin_lock_irq(&lo->lo_lock);
669 if (lo->lo_backing_file)
670 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
671 spin_unlock_irq(&lo->lo_lock);
673 if (IS_ERR_OR_NULL(p))
677 memmove(buf, p, ret);
685 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
687 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
690 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
692 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
695 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
697 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
699 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
702 LOOP_ATTR_RO(backing_file);
703 LOOP_ATTR_RO(offset);
704 LOOP_ATTR_RO(sizelimit);
705 LOOP_ATTR_RO(autoclear);
707 static struct attribute *loop_attrs[] = {
708 &loop_attr_backing_file.attr,
709 &loop_attr_offset.attr,
710 &loop_attr_sizelimit.attr,
711 &loop_attr_autoclear.attr,
715 static struct attribute_group loop_attribute_group = {
720 static int loop_sysfs_init(struct loop_device *lo)
722 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
723 &loop_attribute_group);
726 static void loop_sysfs_exit(struct loop_device *lo)
728 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
729 &loop_attribute_group);
732 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
733 struct block_device *bdev, unsigned int arg)
735 struct file *file, *f;
737 struct address_space *mapping;
738 unsigned lo_blocksize;
743 /* This is safe, since we have a reference from open(). */
744 __module_get(THIS_MODULE);
752 if (lo->lo_state != Lo_unbound)
755 /* Avoid recursion */
757 while (is_loop_device(f)) {
758 struct loop_device *l;
760 if (f->f_mapping->host->i_bdev == bdev)
763 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
764 if (l->lo_state == Lo_unbound) {
768 f = l->lo_backing_file;
771 mapping = file->f_mapping;
772 inode = mapping->host;
775 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
778 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
780 lo_flags |= LO_FLAGS_READ_ONLY;
782 lo_blocksize = S_ISBLK(inode->i_mode) ?
783 inode->i_bdev->bd_block_size : PAGE_SIZE;
786 size = get_loop_size(lo, file);
787 if ((loff_t)(sector_t)size != size)
792 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
794 lo->lo_blocksize = lo_blocksize;
795 lo->lo_device = bdev;
796 lo->lo_flags = lo_flags;
797 lo->lo_backing_file = file;
798 lo->transfer = transfer_none;
800 lo->lo_sizelimit = 0;
801 lo->old_gfp_mask = mapping_gfp_mask(mapping);
802 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
804 bio_list_init(&lo->lo_bio_list);
807 * set queue make_request_fn, and add limits based on lower level
810 blk_queue_make_request(lo->lo_queue, loop_make_request);
811 lo->lo_queue->queuedata = lo;
813 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
814 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
816 set_capacity(lo->lo_disk, size);
817 bd_set_size(bdev, size << 9);
819 /* let user-space know about the new size */
820 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
822 set_blocksize(bdev, lo_blocksize);
824 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
826 if (IS_ERR(lo->lo_thread)) {
827 error = PTR_ERR(lo->lo_thread);
830 lo->lo_state = Lo_bound;
831 wake_up_process(lo->lo_thread);
833 ioctl_by_bdev(bdev, BLKRRPART, 0);
838 lo->lo_thread = NULL;
839 lo->lo_device = NULL;
840 lo->lo_backing_file = NULL;
842 set_capacity(lo->lo_disk, 0);
843 invalidate_bdev(bdev);
844 bd_set_size(bdev, 0);
845 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
846 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
847 lo->lo_state = Lo_unbound;
851 /* This is safe: open() is still holding a reference. */
852 module_put(THIS_MODULE);
857 loop_release_xfer(struct loop_device *lo)
860 struct loop_func_table *xfer = lo->lo_encryption;
864 err = xfer->release(lo);
866 lo->lo_encryption = NULL;
867 module_put(xfer->owner);
873 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
874 const struct loop_info64 *i)
879 struct module *owner = xfer->owner;
881 if (!try_module_get(owner))
884 err = xfer->init(lo, i);
888 lo->lo_encryption = xfer;
893 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
895 struct file *filp = lo->lo_backing_file;
896 gfp_t gfp = lo->old_gfp_mask;
898 if (lo->lo_state != Lo_bound)
901 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
907 spin_lock_irq(&lo->lo_lock);
908 lo->lo_state = Lo_rundown;
909 spin_unlock_irq(&lo->lo_lock);
911 kthread_stop(lo->lo_thread);
913 spin_lock_irq(&lo->lo_lock);
914 lo->lo_backing_file = NULL;
915 spin_unlock_irq(&lo->lo_lock);
917 loop_release_xfer(lo);
920 lo->lo_device = NULL;
921 lo->lo_encryption = NULL;
923 lo->lo_sizelimit = 0;
924 lo->lo_encrypt_key_size = 0;
926 lo->lo_thread = NULL;
927 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
928 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
929 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
931 invalidate_bdev(bdev);
932 set_capacity(lo->lo_disk, 0);
935 bd_set_size(bdev, 0);
936 /* let user-space know about this change */
937 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
939 mapping_set_gfp_mask(filp->f_mapping, gfp);
940 lo->lo_state = Lo_unbound;
941 /* This is safe: open() is still holding a reference. */
942 module_put(THIS_MODULE);
943 if (max_part > 0 && bdev)
944 ioctl_by_bdev(bdev, BLKRRPART, 0);
945 mutex_unlock(&lo->lo_ctl_mutex);
947 * Need not hold lo_ctl_mutex to fput backing file.
948 * Calling fput holding lo_ctl_mutex triggers a circular
949 * lock dependency possibility warning as fput can take
950 * bd_mutex which is usually taken before lo_ctl_mutex.
957 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
960 struct loop_func_table *xfer;
961 uid_t uid = current_uid();
963 if (lo->lo_encrypt_key_size &&
964 lo->lo_key_owner != uid &&
965 !capable(CAP_SYS_ADMIN))
967 if (lo->lo_state != Lo_bound)
969 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
972 err = loop_release_xfer(lo);
976 if (info->lo_encrypt_type) {
977 unsigned int type = info->lo_encrypt_type;
979 if (type >= MAX_LO_CRYPT)
981 xfer = xfer_funcs[type];
987 err = loop_init_xfer(lo, xfer, info);
991 if (lo->lo_offset != info->lo_offset ||
992 lo->lo_sizelimit != info->lo_sizelimit) {
993 lo->lo_offset = info->lo_offset;
994 lo->lo_sizelimit = info->lo_sizelimit;
995 if (figure_loop_size(lo))
999 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1000 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1001 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1002 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1006 lo->transfer = xfer->transfer;
1007 lo->ioctl = xfer->ioctl;
1009 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1010 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1011 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1013 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1014 lo->lo_init[0] = info->lo_init[0];
1015 lo->lo_init[1] = info->lo_init[1];
1016 if (info->lo_encrypt_key_size) {
1017 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1018 info->lo_encrypt_key_size);
1019 lo->lo_key_owner = uid;
1026 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1028 struct file *file = lo->lo_backing_file;
1032 if (lo->lo_state != Lo_bound)
1034 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1037 memset(info, 0, sizeof(*info));
1038 info->lo_number = lo->lo_number;
1039 info->lo_device = huge_encode_dev(stat.dev);
1040 info->lo_inode = stat.ino;
1041 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1042 info->lo_offset = lo->lo_offset;
1043 info->lo_sizelimit = lo->lo_sizelimit;
1044 info->lo_flags = lo->lo_flags;
1045 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1046 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1047 info->lo_encrypt_type =
1048 lo->lo_encryption ? lo->lo_encryption->number : 0;
1049 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1050 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1051 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1052 lo->lo_encrypt_key_size);
1058 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1060 memset(info64, 0, sizeof(*info64));
1061 info64->lo_number = info->lo_number;
1062 info64->lo_device = info->lo_device;
1063 info64->lo_inode = info->lo_inode;
1064 info64->lo_rdevice = info->lo_rdevice;
1065 info64->lo_offset = info->lo_offset;
1066 info64->lo_sizelimit = 0;
1067 info64->lo_encrypt_type = info->lo_encrypt_type;
1068 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1069 info64->lo_flags = info->lo_flags;
1070 info64->lo_init[0] = info->lo_init[0];
1071 info64->lo_init[1] = info->lo_init[1];
1072 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1073 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1075 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1076 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1080 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1082 memset(info, 0, sizeof(*info));
1083 info->lo_number = info64->lo_number;
1084 info->lo_device = info64->lo_device;
1085 info->lo_inode = info64->lo_inode;
1086 info->lo_rdevice = info64->lo_rdevice;
1087 info->lo_offset = info64->lo_offset;
1088 info->lo_encrypt_type = info64->lo_encrypt_type;
1089 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1090 info->lo_flags = info64->lo_flags;
1091 info->lo_init[0] = info64->lo_init[0];
1092 info->lo_init[1] = info64->lo_init[1];
1093 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1094 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1096 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1097 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1099 /* error in case values were truncated */
1100 if (info->lo_device != info64->lo_device ||
1101 info->lo_rdevice != info64->lo_rdevice ||
1102 info->lo_inode != info64->lo_inode ||
1103 info->lo_offset != info64->lo_offset)
1110 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1112 struct loop_info info;
1113 struct loop_info64 info64;
1115 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1117 loop_info64_from_old(&info, &info64);
1118 return loop_set_status(lo, &info64);
1122 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1124 struct loop_info64 info64;
1126 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1128 return loop_set_status(lo, &info64);
1132 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1133 struct loop_info info;
1134 struct loop_info64 info64;
1140 err = loop_get_status(lo, &info64);
1142 err = loop_info64_to_old(&info64, &info);
1143 if (!err && copy_to_user(arg, &info, sizeof(info)))
1150 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1151 struct loop_info64 info64;
1157 err = loop_get_status(lo, &info64);
1158 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1164 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1171 if (unlikely(lo->lo_state != Lo_bound))
1173 err = figure_loop_size(lo);
1176 sec = get_capacity(lo->lo_disk);
1177 /* the width of sector_t may be narrow for bit-shift */
1180 mutex_lock(&bdev->bd_mutex);
1181 bd_set_size(bdev, sz);
1182 /* let user-space know about the new size */
1183 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1184 mutex_unlock(&bdev->bd_mutex);
1190 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1191 unsigned int cmd, unsigned long arg)
1193 struct loop_device *lo = bdev->bd_disk->private_data;
1196 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1199 err = loop_set_fd(lo, mode, bdev, arg);
1201 case LOOP_CHANGE_FD:
1202 err = loop_change_fd(lo, bdev, arg);
1205 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1206 err = loop_clr_fd(lo, bdev);
1210 case LOOP_SET_STATUS:
1211 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1213 case LOOP_GET_STATUS:
1214 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1216 case LOOP_SET_STATUS64:
1217 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1219 case LOOP_GET_STATUS64:
1220 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1222 case LOOP_SET_CAPACITY:
1224 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1225 err = loop_set_capacity(lo, bdev);
1228 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1230 mutex_unlock(&lo->lo_ctl_mutex);
1236 #ifdef CONFIG_COMPAT
1237 struct compat_loop_info {
1238 compat_int_t lo_number; /* ioctl r/o */
1239 compat_dev_t lo_device; /* ioctl r/o */
1240 compat_ulong_t lo_inode; /* ioctl r/o */
1241 compat_dev_t lo_rdevice; /* ioctl r/o */
1242 compat_int_t lo_offset;
1243 compat_int_t lo_encrypt_type;
1244 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1245 compat_int_t lo_flags; /* ioctl r/o */
1246 char lo_name[LO_NAME_SIZE];
1247 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1248 compat_ulong_t lo_init[2];
1253 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1254 * - noinlined to reduce stack space usage in main part of driver
1257 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1258 struct loop_info64 *info64)
1260 struct compat_loop_info info;
1262 if (copy_from_user(&info, arg, sizeof(info)))
1265 memset(info64, 0, sizeof(*info64));
1266 info64->lo_number = info.lo_number;
1267 info64->lo_device = info.lo_device;
1268 info64->lo_inode = info.lo_inode;
1269 info64->lo_rdevice = info.lo_rdevice;
1270 info64->lo_offset = info.lo_offset;
1271 info64->lo_sizelimit = 0;
1272 info64->lo_encrypt_type = info.lo_encrypt_type;
1273 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1274 info64->lo_flags = info.lo_flags;
1275 info64->lo_init[0] = info.lo_init[0];
1276 info64->lo_init[1] = info.lo_init[1];
1277 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1278 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1280 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1281 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1286 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1287 * - noinlined to reduce stack space usage in main part of driver
1290 loop_info64_to_compat(const struct loop_info64 *info64,
1291 struct compat_loop_info __user *arg)
1293 struct compat_loop_info info;
1295 memset(&info, 0, sizeof(info));
1296 info.lo_number = info64->lo_number;
1297 info.lo_device = info64->lo_device;
1298 info.lo_inode = info64->lo_inode;
1299 info.lo_rdevice = info64->lo_rdevice;
1300 info.lo_offset = info64->lo_offset;
1301 info.lo_encrypt_type = info64->lo_encrypt_type;
1302 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1303 info.lo_flags = info64->lo_flags;
1304 info.lo_init[0] = info64->lo_init[0];
1305 info.lo_init[1] = info64->lo_init[1];
1306 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1307 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1309 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1310 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1312 /* error in case values were truncated */
1313 if (info.lo_device != info64->lo_device ||
1314 info.lo_rdevice != info64->lo_rdevice ||
1315 info.lo_inode != info64->lo_inode ||
1316 info.lo_offset != info64->lo_offset ||
1317 info.lo_init[0] != info64->lo_init[0] ||
1318 info.lo_init[1] != info64->lo_init[1])
1321 if (copy_to_user(arg, &info, sizeof(info)))
1327 loop_set_status_compat(struct loop_device *lo,
1328 const struct compat_loop_info __user *arg)
1330 struct loop_info64 info64;
1333 ret = loop_info64_from_compat(arg, &info64);
1336 return loop_set_status(lo, &info64);
1340 loop_get_status_compat(struct loop_device *lo,
1341 struct compat_loop_info __user *arg)
1343 struct loop_info64 info64;
1349 err = loop_get_status(lo, &info64);
1351 err = loop_info64_to_compat(&info64, arg);
1355 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1356 unsigned int cmd, unsigned long arg)
1358 struct loop_device *lo = bdev->bd_disk->private_data;
1362 case LOOP_SET_STATUS:
1363 mutex_lock(&lo->lo_ctl_mutex);
1364 err = loop_set_status_compat(
1365 lo, (const struct compat_loop_info __user *) arg);
1366 mutex_unlock(&lo->lo_ctl_mutex);
1368 case LOOP_GET_STATUS:
1369 mutex_lock(&lo->lo_ctl_mutex);
1370 err = loop_get_status_compat(
1371 lo, (struct compat_loop_info __user *) arg);
1372 mutex_unlock(&lo->lo_ctl_mutex);
1374 case LOOP_SET_CAPACITY:
1376 case LOOP_GET_STATUS64:
1377 case LOOP_SET_STATUS64:
1378 arg = (unsigned long) compat_ptr(arg);
1380 case LOOP_CHANGE_FD:
1381 err = lo_ioctl(bdev, mode, cmd, arg);
1391 static int lo_open(struct block_device *bdev, fmode_t mode)
1393 struct loop_device *lo;
1396 mutex_lock(&loop_index_mutex);
1397 lo = bdev->bd_disk->private_data;
1403 mutex_lock(&lo->lo_ctl_mutex);
1405 mutex_unlock(&lo->lo_ctl_mutex);
1407 mutex_unlock(&loop_index_mutex);
1411 static int lo_release(struct gendisk *disk, fmode_t mode)
1413 struct loop_device *lo = disk->private_data;
1416 mutex_lock(&lo->lo_ctl_mutex);
1418 if (--lo->lo_refcnt)
1421 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1423 * In autoclear mode, stop the loop thread
1424 * and remove configuration after last close.
1426 err = loop_clr_fd(lo, NULL);
1431 * Otherwise keep thread (if running) and config,
1432 * but flush possible ongoing bios in thread.
1438 mutex_unlock(&lo->lo_ctl_mutex);
1443 static const struct block_device_operations lo_fops = {
1444 .owner = THIS_MODULE,
1446 .release = lo_release,
1448 #ifdef CONFIG_COMPAT
1449 .compat_ioctl = lo_compat_ioctl,
1454 * And now the modules code and kernel interface.
1456 static int max_loop;
1457 module_param(max_loop, int, S_IRUGO);
1458 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1459 module_param(max_part, int, S_IRUGO);
1460 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1461 MODULE_LICENSE("GPL");
1462 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1464 int loop_register_transfer(struct loop_func_table *funcs)
1466 unsigned int n = funcs->number;
1468 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1470 xfer_funcs[n] = funcs;
1474 static int unregister_transfer_cb(int id, void *ptr, void *data)
1476 struct loop_device *lo = ptr;
1477 struct loop_func_table *xfer = data;
1479 mutex_lock(&lo->lo_ctl_mutex);
1480 if (lo->lo_encryption == xfer)
1481 loop_release_xfer(lo);
1482 mutex_unlock(&lo->lo_ctl_mutex);
1486 int loop_unregister_transfer(int number)
1488 unsigned int n = number;
1489 struct loop_func_table *xfer;
1491 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1494 xfer_funcs[n] = NULL;
1495 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1499 EXPORT_SYMBOL(loop_register_transfer);
1500 EXPORT_SYMBOL(loop_unregister_transfer);
1502 static int loop_add(struct loop_device **l, int i)
1504 struct loop_device *lo;
1505 struct gendisk *disk;
1508 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1514 err = idr_pre_get(&loop_index_idr, GFP_KERNEL);
1521 /* create specific i in the index */
1522 err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1523 if (err >= 0 && i != m) {
1524 idr_remove(&loop_index_idr, m);
1527 } else if (i == -1) {
1530 /* get next free nr */
1531 err = idr_get_new(&loop_index_idr, lo, &m);
1540 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1544 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1546 goto out_free_queue;
1548 mutex_init(&lo->lo_ctl_mutex);
1550 lo->lo_thread = NULL;
1551 init_waitqueue_head(&lo->lo_event);
1552 spin_lock_init(&lo->lo_lock);
1553 disk->major = LOOP_MAJOR;
1554 disk->first_minor = i << part_shift;
1555 disk->fops = &lo_fops;
1556 disk->private_data = lo;
1557 disk->queue = lo->lo_queue;
1558 sprintf(disk->disk_name, "loop%d", i);
1561 return lo->lo_number;
1564 blk_cleanup_queue(lo->lo_queue);
1571 static void loop_remove(struct loop_device *lo)
1573 del_gendisk(lo->lo_disk);
1574 blk_cleanup_queue(lo->lo_queue);
1575 put_disk(lo->lo_disk);
1579 static int find_free_cb(int id, void *ptr, void *data)
1581 struct loop_device *lo = ptr;
1582 struct loop_device **l = data;
1584 if (lo->lo_state == Lo_unbound) {
1591 static int loop_lookup(struct loop_device **l, int i)
1593 struct loop_device *lo;
1599 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1602 ret = lo->lo_number;
1607 /* lookup and return a specific i */
1608 lo = idr_find(&loop_index_idr, i);
1611 ret = lo->lo_number;
1617 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1619 struct loop_device *lo;
1620 struct kobject *kobj;
1623 mutex_lock(&loop_index_mutex);
1624 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1626 err = loop_add(&lo, MINOR(dev) >> part_shift);
1628 kobj = ERR_PTR(err);
1630 kobj = get_disk(lo->lo_disk);
1631 mutex_unlock(&loop_index_mutex);
1637 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1640 struct loop_device *lo;
1643 mutex_lock(&loop_index_mutex);
1646 ret = loop_lookup(&lo, parm);
1651 ret = loop_add(&lo, parm);
1653 case LOOP_CTL_REMOVE:
1654 ret = loop_lookup(&lo, parm);
1657 mutex_lock(&lo->lo_ctl_mutex);
1658 if (lo->lo_state != Lo_unbound) {
1660 mutex_unlock(&lo->lo_ctl_mutex);
1663 if (lo->lo_refcnt > 0) {
1665 mutex_unlock(&lo->lo_ctl_mutex);
1668 lo->lo_disk->private_data = NULL;
1669 mutex_unlock(&lo->lo_ctl_mutex);
1670 idr_remove(&loop_index_idr, lo->lo_number);
1673 case LOOP_CTL_GET_FREE:
1674 ret = loop_lookup(&lo, -1);
1677 ret = loop_add(&lo, -1);
1679 mutex_unlock(&loop_index_mutex);
1684 static const struct file_operations loop_ctl_fops = {
1685 .open = nonseekable_open,
1686 .unlocked_ioctl = loop_control_ioctl,
1687 .compat_ioctl = loop_control_ioctl,
1688 .owner = THIS_MODULE,
1689 .llseek = noop_llseek,
1692 static struct miscdevice loop_misc = {
1693 .minor = LOOP_CTRL_MINOR,
1694 .name = "loop-control",
1695 .fops = &loop_ctl_fops,
1698 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1699 MODULE_ALIAS("devname:loop-control");
1701 static int __init loop_init(void)
1704 unsigned long range;
1705 struct loop_device *lo;
1708 err = misc_register(&loop_misc);
1714 part_shift = fls(max_part);
1717 * Adjust max_part according to part_shift as it is exported
1718 * to user space so that user can decide correct minor number
1719 * if [s]he want to create more devices.
1721 * Note that -1 is required because partition 0 is reserved
1722 * for the whole disk.
1724 max_part = (1UL << part_shift) - 1;
1727 if ((1UL << part_shift) > DISK_MAX_PARTS)
1730 if (max_loop > 1UL << (MINORBITS - part_shift))
1734 * If max_loop is specified, create that many devices upfront.
1735 * This also becomes a hard limit. If max_loop is not specified,
1736 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1737 * init time. Loop devices can be requested on-demand with the
1738 * /dev/loop-control interface, or be instantiated by accessing
1739 * a 'dead' device node.
1743 range = max_loop << part_shift;
1745 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1746 range = 1UL << MINORBITS;
1749 if (register_blkdev(LOOP_MAJOR, "loop"))
1752 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1753 THIS_MODULE, loop_probe, NULL, NULL);
1755 /* pre-create number of devices given by config or max_loop */
1756 mutex_lock(&loop_index_mutex);
1757 for (i = 0; i < nr; i++)
1759 mutex_unlock(&loop_index_mutex);
1761 printk(KERN_INFO "loop: module loaded\n");
1765 static int loop_exit_cb(int id, void *ptr, void *data)
1767 struct loop_device *lo = ptr;
1773 static void __exit loop_exit(void)
1775 unsigned long range;
1777 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1779 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1780 idr_remove_all(&loop_index_idr);
1781 idr_destroy(&loop_index_idr);
1783 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1784 unregister_blkdev(LOOP_MAJOR, "loop");
1786 misc_deregister(&loop_misc);
1789 module_init(loop_init);
1790 module_exit(loop_exit);
1793 static int __init max_loop_setup(char *str)
1795 max_loop = simple_strtol(str, NULL, 0);
1799 __setup("max_loop=", max_loop_setup);