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 DEFINE_MUTEX(loop_mutex);
82 static LIST_HEAD(loop_devices);
83 static DEFINE_MUTEX(loop_devices_mutex);
86 static int part_shift;
91 static int transfer_none(struct loop_device *lo, int cmd,
92 struct page *raw_page, unsigned raw_off,
93 struct page *loop_page, unsigned loop_off,
94 int size, sector_t real_block)
96 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
97 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
100 memcpy(loop_buf, raw_buf, size);
102 memcpy(raw_buf, loop_buf, size);
104 kunmap_atomic(loop_buf, KM_USER1);
105 kunmap_atomic(raw_buf, KM_USER0);
110 static int transfer_xor(struct loop_device *lo, int cmd,
111 struct page *raw_page, unsigned raw_off,
112 struct page *loop_page, unsigned loop_off,
113 int size, sector_t real_block)
115 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
116 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
117 char *in, *out, *key;
128 key = lo->lo_encrypt_key;
129 keysize = lo->lo_encrypt_key_size;
130 for (i = 0; i < size; i++)
131 *out++ = *in++ ^ key[(i & 511) % keysize];
133 kunmap_atomic(loop_buf, KM_USER1);
134 kunmap_atomic(raw_buf, KM_USER0);
139 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
141 if (unlikely(info->lo_encrypt_key_size <= 0))
146 static struct loop_func_table none_funcs = {
147 .number = LO_CRYPT_NONE,
148 .transfer = transfer_none,
151 static struct loop_func_table xor_funcs = {
152 .number = LO_CRYPT_XOR,
153 .transfer = transfer_xor,
157 /* xfer_funcs[0] is special - its release function is never called */
158 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
163 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
165 loff_t size, offset, loopsize;
167 /* Compute loopsize in bytes */
168 size = i_size_read(file->f_mapping->host);
169 offset = lo->lo_offset;
170 loopsize = size - offset;
171 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
172 loopsize = lo->lo_sizelimit;
175 * Unfortunately, if we want to do I/O on the device,
176 * the number of 512-byte sectors has to fit into a sector_t.
178 return loopsize >> 9;
182 figure_loop_size(struct loop_device *lo)
184 loff_t size = get_loop_size(lo, lo->lo_backing_file);
185 sector_t x = (sector_t)size;
187 if (unlikely((loff_t)x != size))
190 set_capacity(lo->lo_disk, x);
195 lo_do_transfer(struct loop_device *lo, int cmd,
196 struct page *rpage, unsigned roffs,
197 struct page *lpage, unsigned loffs,
198 int size, sector_t rblock)
200 if (unlikely(!lo->transfer))
203 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
207 * do_lo_send_aops - helper for writing data to a loop device
209 * This is the fast version for backing filesystems which implement the address
210 * space operations write_begin and write_end.
212 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
213 loff_t pos, struct page *unused)
215 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
216 struct address_space *mapping = file->f_mapping;
218 unsigned offset, bv_offs;
221 mutex_lock(&mapping->host->i_mutex);
222 index = pos >> PAGE_CACHE_SHIFT;
223 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
224 bv_offs = bvec->bv_offset;
228 unsigned size, copied;
233 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
234 size = PAGE_CACHE_SIZE - offset;
238 ret = pagecache_write_begin(file, mapping, pos, size, 0,
243 file_update_time(file);
245 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
246 bvec->bv_page, bv_offs, size, IV);
248 if (unlikely(transfer_result))
251 ret = pagecache_write_end(file, mapping, pos, size, copied,
253 if (ret < 0 || ret != copied)
256 if (unlikely(transfer_result))
267 mutex_unlock(&mapping->host->i_mutex);
275 * __do_lo_send_write - helper for writing data to a loop device
277 * This helper just factors out common code between do_lo_send_direct_write()
278 * and do_lo_send_write().
280 static int __do_lo_send_write(struct file *file,
281 u8 *buf, const int len, loff_t pos)
284 mm_segment_t old_fs = get_fs();
287 bw = file->f_op->write(file, buf, len, &pos);
289 if (likely(bw == len))
291 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
292 (unsigned long long)pos, len);
299 * do_lo_send_direct_write - helper for writing data to a loop device
301 * This is the fast, non-transforming version for backing filesystems which do
302 * not implement the address space operations write_begin and write_end.
303 * It uses the write file operation which should be present on all writeable
306 static int do_lo_send_direct_write(struct loop_device *lo,
307 struct bio_vec *bvec, loff_t pos, struct page *page)
309 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
310 kmap(bvec->bv_page) + bvec->bv_offset,
312 kunmap(bvec->bv_page);
318 * do_lo_send_write - helper for writing data to a loop device
320 * This is the slow, transforming version for filesystems which do not
321 * implement the address space operations write_begin and write_end. It
322 * uses the write file operation which should be present on all writeable
325 * Using fops->write is slower than using aops->{prepare,commit}_write in the
326 * transforming case because we need to double buffer the data as we cannot do
327 * the transformations in place as we do not have direct access to the
328 * destination pages of the backing file.
330 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
331 loff_t pos, struct page *page)
333 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
334 bvec->bv_offset, bvec->bv_len, pos >> 9);
336 return __do_lo_send_write(lo->lo_backing_file,
337 page_address(page), bvec->bv_len,
339 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
340 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
346 static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
348 int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
350 struct bio_vec *bvec;
351 struct page *page = NULL;
354 do_lo_send = do_lo_send_aops;
355 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
356 do_lo_send = do_lo_send_direct_write;
357 if (lo->transfer != transfer_none) {
358 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
362 do_lo_send = do_lo_send_write;
365 bio_for_each_segment(bvec, bio, i) {
366 ret = do_lo_send(lo, bvec, pos, page);
378 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
383 struct lo_read_data {
384 struct loop_device *lo;
391 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
392 struct splice_desc *sd)
394 struct lo_read_data *p = sd->u.data;
395 struct loop_device *lo = p->lo;
396 struct page *page = buf->page;
400 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
406 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
407 printk(KERN_ERR "loop: transfer error block %ld\n",
412 flush_dcache_page(p->page);
421 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
423 return __splice_from_pipe(pipe, sd, lo_splice_actor);
427 do_lo_receive(struct loop_device *lo,
428 struct bio_vec *bvec, int bsize, loff_t pos)
430 struct lo_read_data cookie;
431 struct splice_desc sd;
436 cookie.page = bvec->bv_page;
437 cookie.offset = bvec->bv_offset;
438 cookie.bsize = bsize;
441 sd.total_len = bvec->bv_len;
446 file = lo->lo_backing_file;
447 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
456 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
458 struct bio_vec *bvec;
461 bio_for_each_segment(bvec, bio, i) {
462 ret = do_lo_receive(lo, bvec, bsize, pos);
470 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
475 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
477 if (bio_rw(bio) == WRITE) {
478 struct file *file = lo->lo_backing_file;
480 if (bio->bi_rw & REQ_FLUSH) {
481 ret = vfs_fsync(file, 0);
482 if (unlikely(ret && ret != -EINVAL)) {
488 ret = lo_send(lo, bio, pos);
490 if ((bio->bi_rw & REQ_FUA) && !ret) {
491 ret = vfs_fsync(file, 0);
492 if (unlikely(ret && ret != -EINVAL))
496 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
503 * Add bio to back of pending list
505 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
507 bio_list_add(&lo->lo_bio_list, bio);
511 * Grab first pending buffer
513 static struct bio *loop_get_bio(struct loop_device *lo)
515 return bio_list_pop(&lo->lo_bio_list);
518 static int loop_make_request(struct request_queue *q, struct bio *old_bio)
520 struct loop_device *lo = q->queuedata;
521 int rw = bio_rw(old_bio);
526 BUG_ON(!lo || (rw != READ && rw != WRITE));
528 spin_lock_irq(&lo->lo_lock);
529 if (lo->lo_state != Lo_bound)
531 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
533 loop_add_bio(lo, old_bio);
534 wake_up(&lo->lo_event);
535 spin_unlock_irq(&lo->lo_lock);
539 spin_unlock_irq(&lo->lo_lock);
540 bio_io_error(old_bio);
544 struct switch_request {
546 struct completion wait;
549 static void do_loop_switch(struct loop_device *, struct switch_request *);
551 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
553 if (unlikely(!bio->bi_bdev)) {
554 do_loop_switch(lo, bio->bi_private);
557 int ret = do_bio_filebacked(lo, bio);
563 * worker thread that handles reads/writes to file backed loop devices,
564 * to avoid blocking in our make_request_fn. it also does loop decrypting
565 * on reads for block backed loop, as that is too heavy to do from
566 * b_end_io context where irqs may be disabled.
568 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
569 * calling kthread_stop(). Therefore once kthread_should_stop() is
570 * true, make_request will not place any more requests. Therefore
571 * once kthread_should_stop() is true and lo_bio is NULL, we are
572 * done with the loop.
574 static int loop_thread(void *data)
576 struct loop_device *lo = data;
579 set_user_nice(current, -20);
581 while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
583 wait_event_interruptible(lo->lo_event,
584 !bio_list_empty(&lo->lo_bio_list) ||
585 kthread_should_stop());
587 if (bio_list_empty(&lo->lo_bio_list))
589 spin_lock_irq(&lo->lo_lock);
590 bio = loop_get_bio(lo);
591 spin_unlock_irq(&lo->lo_lock);
594 loop_handle_bio(lo, bio);
601 * loop_switch performs the hard work of switching a backing store.
602 * First it needs to flush existing IO, it does this by sending a magic
603 * BIO down the pipe. The completion of this BIO does the actual switch.
605 static int loop_switch(struct loop_device *lo, struct file *file)
607 struct switch_request w;
608 struct bio *bio = bio_alloc(GFP_KERNEL, 0);
611 init_completion(&w.wait);
613 bio->bi_private = &w;
615 loop_make_request(lo->lo_queue, bio);
616 wait_for_completion(&w.wait);
621 * Helper to flush the IOs in loop, but keeping loop thread running
623 static int loop_flush(struct loop_device *lo)
625 /* loop not yet configured, no running thread, nothing to flush */
629 return loop_switch(lo, NULL);
633 * Do the actual switch; called from the BIO completion routine
635 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
637 struct file *file = p->file;
638 struct file *old_file = lo->lo_backing_file;
639 struct address_space *mapping;
641 /* if no new file, only flush of queued bios requested */
645 mapping = file->f_mapping;
646 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
647 lo->lo_backing_file = file;
648 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
649 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
650 lo->old_gfp_mask = mapping_gfp_mask(mapping);
651 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
658 * loop_change_fd switched the backing store of a loopback device to
659 * a new file. This is useful for operating system installers to free up
660 * the original file and in High Availability environments to switch to
661 * an alternative location for the content in case of server meltdown.
662 * This can only work if the loop device is used read-only, and if the
663 * new backing store is the same size and type as the old backing store.
665 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
668 struct file *file, *old_file;
673 if (lo->lo_state != Lo_bound)
676 /* the loop device has to be read-only */
678 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
686 inode = file->f_mapping->host;
687 old_file = lo->lo_backing_file;
691 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
694 /* size of the new backing store needs to be the same */
695 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
699 error = loop_switch(lo, file);
705 ioctl_by_bdev(bdev, BLKRRPART, 0);
714 static inline int is_loop_device(struct file *file)
716 struct inode *i = file->f_mapping->host;
718 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
721 /* loop sysfs attributes */
723 static ssize_t loop_attr_show(struct device *dev, char *page,
724 ssize_t (*callback)(struct loop_device *, char *))
726 struct loop_device *l, *lo = NULL;
728 mutex_lock(&loop_devices_mutex);
729 list_for_each_entry(l, &loop_devices, lo_list)
730 if (disk_to_dev(l->lo_disk) == dev) {
734 mutex_unlock(&loop_devices_mutex);
736 return lo ? callback(lo, page) : -EIO;
739 #define LOOP_ATTR_RO(_name) \
740 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
741 static ssize_t loop_attr_do_show_##_name(struct device *d, \
742 struct device_attribute *attr, char *b) \
744 return loop_attr_show(d, b, loop_attr_##_name##_show); \
746 static struct device_attribute loop_attr_##_name = \
747 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
749 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
754 mutex_lock(&lo->lo_ctl_mutex);
755 if (lo->lo_backing_file)
756 p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
757 mutex_unlock(&lo->lo_ctl_mutex);
759 if (IS_ERR_OR_NULL(p))
763 memmove(buf, p, ret);
771 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
773 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
776 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
778 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
781 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
783 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
785 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
788 LOOP_ATTR_RO(backing_file);
789 LOOP_ATTR_RO(offset);
790 LOOP_ATTR_RO(sizelimit);
791 LOOP_ATTR_RO(autoclear);
793 static struct attribute *loop_attrs[] = {
794 &loop_attr_backing_file.attr,
795 &loop_attr_offset.attr,
796 &loop_attr_sizelimit.attr,
797 &loop_attr_autoclear.attr,
801 static struct attribute_group loop_attribute_group = {
806 static int loop_sysfs_init(struct loop_device *lo)
808 return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
809 &loop_attribute_group);
812 static void loop_sysfs_exit(struct loop_device *lo)
814 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
815 &loop_attribute_group);
818 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
819 struct block_device *bdev, unsigned int arg)
821 struct file *file, *f;
823 struct address_space *mapping;
824 unsigned lo_blocksize;
829 /* This is safe, since we have a reference from open(). */
830 __module_get(THIS_MODULE);
838 if (lo->lo_state != Lo_unbound)
841 /* Avoid recursion */
843 while (is_loop_device(f)) {
844 struct loop_device *l;
846 if (f->f_mapping->host->i_bdev == bdev)
849 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
850 if (l->lo_state == Lo_unbound) {
854 f = l->lo_backing_file;
857 mapping = file->f_mapping;
858 inode = mapping->host;
860 if (!(file->f_mode & FMODE_WRITE))
861 lo_flags |= LO_FLAGS_READ_ONLY;
864 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
865 const struct address_space_operations *aops = mapping->a_ops;
867 if (aops->write_begin)
868 lo_flags |= LO_FLAGS_USE_AOPS;
869 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
870 lo_flags |= LO_FLAGS_READ_ONLY;
872 lo_blocksize = S_ISBLK(inode->i_mode) ?
873 inode->i_bdev->bd_block_size : PAGE_SIZE;
880 size = get_loop_size(lo, file);
882 if ((loff_t)(sector_t)size != size) {
887 if (!(mode & FMODE_WRITE))
888 lo_flags |= LO_FLAGS_READ_ONLY;
890 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
892 lo->lo_blocksize = lo_blocksize;
893 lo->lo_device = bdev;
894 lo->lo_flags = lo_flags;
895 lo->lo_backing_file = file;
896 lo->transfer = transfer_none;
898 lo->lo_sizelimit = 0;
899 lo->old_gfp_mask = mapping_gfp_mask(mapping);
900 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
902 bio_list_init(&lo->lo_bio_list);
905 * set queue make_request_fn, and add limits based on lower level
908 blk_queue_make_request(lo->lo_queue, loop_make_request);
909 lo->lo_queue->queuedata = lo;
911 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
912 blk_queue_flush(lo->lo_queue, REQ_FLUSH);
914 set_capacity(lo->lo_disk, size);
915 bd_set_size(bdev, size << 9);
917 /* let user-space know about the new size */
918 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
920 set_blocksize(bdev, lo_blocksize);
922 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
924 if (IS_ERR(lo->lo_thread)) {
925 error = PTR_ERR(lo->lo_thread);
928 lo->lo_state = Lo_bound;
929 wake_up_process(lo->lo_thread);
931 ioctl_by_bdev(bdev, BLKRRPART, 0);
936 lo->lo_thread = NULL;
937 lo->lo_device = NULL;
938 lo->lo_backing_file = NULL;
940 set_capacity(lo->lo_disk, 0);
941 invalidate_bdev(bdev);
942 bd_set_size(bdev, 0);
943 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
944 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
945 lo->lo_state = Lo_unbound;
949 /* This is safe: open() is still holding a reference. */
950 module_put(THIS_MODULE);
955 loop_release_xfer(struct loop_device *lo)
958 struct loop_func_table *xfer = lo->lo_encryption;
962 err = xfer->release(lo);
964 lo->lo_encryption = NULL;
965 module_put(xfer->owner);
971 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
972 const struct loop_info64 *i)
977 struct module *owner = xfer->owner;
979 if (!try_module_get(owner))
982 err = xfer->init(lo, i);
986 lo->lo_encryption = xfer;
991 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
993 struct file *filp = lo->lo_backing_file;
994 gfp_t gfp = lo->old_gfp_mask;
996 if (lo->lo_state != Lo_bound)
999 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
1005 spin_lock_irq(&lo->lo_lock);
1006 lo->lo_state = Lo_rundown;
1007 spin_unlock_irq(&lo->lo_lock);
1009 kthread_stop(lo->lo_thread);
1011 lo->lo_backing_file = NULL;
1013 loop_release_xfer(lo);
1014 lo->transfer = NULL;
1016 lo->lo_device = NULL;
1017 lo->lo_encryption = NULL;
1019 lo->lo_sizelimit = 0;
1020 lo->lo_encrypt_key_size = 0;
1022 lo->lo_thread = NULL;
1023 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1024 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1025 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1027 invalidate_bdev(bdev);
1028 set_capacity(lo->lo_disk, 0);
1029 loop_sysfs_exit(lo);
1031 bd_set_size(bdev, 0);
1032 /* let user-space know about this change */
1033 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1035 mapping_set_gfp_mask(filp->f_mapping, gfp);
1036 lo->lo_state = Lo_unbound;
1037 /* This is safe: open() is still holding a reference. */
1038 module_put(THIS_MODULE);
1039 if (max_part > 0 && bdev)
1040 ioctl_by_bdev(bdev, BLKRRPART, 0);
1041 mutex_unlock(&lo->lo_ctl_mutex);
1043 * Need not hold lo_ctl_mutex to fput backing file.
1044 * Calling fput holding lo_ctl_mutex triggers a circular
1045 * lock dependency possibility warning as fput can take
1046 * bd_mutex which is usually taken before lo_ctl_mutex.
1053 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1056 struct loop_func_table *xfer;
1057 uid_t uid = current_uid();
1059 if (lo->lo_encrypt_key_size &&
1060 lo->lo_key_owner != uid &&
1061 !capable(CAP_SYS_ADMIN))
1063 if (lo->lo_state != Lo_bound)
1065 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1068 err = loop_release_xfer(lo);
1072 if (info->lo_encrypt_type) {
1073 unsigned int type = info->lo_encrypt_type;
1075 if (type >= MAX_LO_CRYPT)
1077 xfer = xfer_funcs[type];
1083 err = loop_init_xfer(lo, xfer, info);
1087 if (lo->lo_offset != info->lo_offset ||
1088 lo->lo_sizelimit != info->lo_sizelimit) {
1089 lo->lo_offset = info->lo_offset;
1090 lo->lo_sizelimit = info->lo_sizelimit;
1091 if (figure_loop_size(lo))
1095 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1096 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1097 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1098 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1102 lo->transfer = xfer->transfer;
1103 lo->ioctl = xfer->ioctl;
1105 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1106 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1107 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1109 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1110 lo->lo_init[0] = info->lo_init[0];
1111 lo->lo_init[1] = info->lo_init[1];
1112 if (info->lo_encrypt_key_size) {
1113 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1114 info->lo_encrypt_key_size);
1115 lo->lo_key_owner = uid;
1122 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1124 struct file *file = lo->lo_backing_file;
1128 if (lo->lo_state != Lo_bound)
1130 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1133 memset(info, 0, sizeof(*info));
1134 info->lo_number = lo->lo_number;
1135 info->lo_device = huge_encode_dev(stat.dev);
1136 info->lo_inode = stat.ino;
1137 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1138 info->lo_offset = lo->lo_offset;
1139 info->lo_sizelimit = lo->lo_sizelimit;
1140 info->lo_flags = lo->lo_flags;
1141 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1142 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1143 info->lo_encrypt_type =
1144 lo->lo_encryption ? lo->lo_encryption->number : 0;
1145 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1146 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1147 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1148 lo->lo_encrypt_key_size);
1154 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1156 memset(info64, 0, sizeof(*info64));
1157 info64->lo_number = info->lo_number;
1158 info64->lo_device = info->lo_device;
1159 info64->lo_inode = info->lo_inode;
1160 info64->lo_rdevice = info->lo_rdevice;
1161 info64->lo_offset = info->lo_offset;
1162 info64->lo_sizelimit = 0;
1163 info64->lo_encrypt_type = info->lo_encrypt_type;
1164 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1165 info64->lo_flags = info->lo_flags;
1166 info64->lo_init[0] = info->lo_init[0];
1167 info64->lo_init[1] = info->lo_init[1];
1168 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1169 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1171 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1172 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1176 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1178 memset(info, 0, sizeof(*info));
1179 info->lo_number = info64->lo_number;
1180 info->lo_device = info64->lo_device;
1181 info->lo_inode = info64->lo_inode;
1182 info->lo_rdevice = info64->lo_rdevice;
1183 info->lo_offset = info64->lo_offset;
1184 info->lo_encrypt_type = info64->lo_encrypt_type;
1185 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1186 info->lo_flags = info64->lo_flags;
1187 info->lo_init[0] = info64->lo_init[0];
1188 info->lo_init[1] = info64->lo_init[1];
1189 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1190 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1192 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1193 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1195 /* error in case values were truncated */
1196 if (info->lo_device != info64->lo_device ||
1197 info->lo_rdevice != info64->lo_rdevice ||
1198 info->lo_inode != info64->lo_inode ||
1199 info->lo_offset != info64->lo_offset)
1206 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1208 struct loop_info info;
1209 struct loop_info64 info64;
1211 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1213 loop_info64_from_old(&info, &info64);
1214 return loop_set_status(lo, &info64);
1218 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1220 struct loop_info64 info64;
1222 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1224 return loop_set_status(lo, &info64);
1228 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1229 struct loop_info info;
1230 struct loop_info64 info64;
1236 err = loop_get_status(lo, &info64);
1238 err = loop_info64_to_old(&info64, &info);
1239 if (!err && copy_to_user(arg, &info, sizeof(info)))
1246 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1247 struct loop_info64 info64;
1253 err = loop_get_status(lo, &info64);
1254 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1260 static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1267 if (unlikely(lo->lo_state != Lo_bound))
1269 err = figure_loop_size(lo);
1272 sec = get_capacity(lo->lo_disk);
1273 /* the width of sector_t may be narrow for bit-shift */
1276 mutex_lock(&bdev->bd_mutex);
1277 bd_set_size(bdev, sz);
1278 /* let user-space know about the new size */
1279 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1280 mutex_unlock(&bdev->bd_mutex);
1286 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1287 unsigned int cmd, unsigned long arg)
1289 struct loop_device *lo = bdev->bd_disk->private_data;
1292 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1295 err = loop_set_fd(lo, mode, bdev, arg);
1297 case LOOP_CHANGE_FD:
1298 err = loop_change_fd(lo, bdev, arg);
1301 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1302 err = loop_clr_fd(lo, bdev);
1306 case LOOP_SET_STATUS:
1307 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1309 case LOOP_GET_STATUS:
1310 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1312 case LOOP_SET_STATUS64:
1313 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1315 case LOOP_GET_STATUS64:
1316 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1318 case LOOP_SET_CAPACITY:
1320 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1321 err = loop_set_capacity(lo, bdev);
1324 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1326 mutex_unlock(&lo->lo_ctl_mutex);
1332 #ifdef CONFIG_COMPAT
1333 struct compat_loop_info {
1334 compat_int_t lo_number; /* ioctl r/o */
1335 compat_dev_t lo_device; /* ioctl r/o */
1336 compat_ulong_t lo_inode; /* ioctl r/o */
1337 compat_dev_t lo_rdevice; /* ioctl r/o */
1338 compat_int_t lo_offset;
1339 compat_int_t lo_encrypt_type;
1340 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1341 compat_int_t lo_flags; /* ioctl r/o */
1342 char lo_name[LO_NAME_SIZE];
1343 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1344 compat_ulong_t lo_init[2];
1349 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1350 * - noinlined to reduce stack space usage in main part of driver
1353 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1354 struct loop_info64 *info64)
1356 struct compat_loop_info info;
1358 if (copy_from_user(&info, arg, sizeof(info)))
1361 memset(info64, 0, sizeof(*info64));
1362 info64->lo_number = info.lo_number;
1363 info64->lo_device = info.lo_device;
1364 info64->lo_inode = info.lo_inode;
1365 info64->lo_rdevice = info.lo_rdevice;
1366 info64->lo_offset = info.lo_offset;
1367 info64->lo_sizelimit = 0;
1368 info64->lo_encrypt_type = info.lo_encrypt_type;
1369 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1370 info64->lo_flags = info.lo_flags;
1371 info64->lo_init[0] = info.lo_init[0];
1372 info64->lo_init[1] = info.lo_init[1];
1373 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1374 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1376 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1377 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1382 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1383 * - noinlined to reduce stack space usage in main part of driver
1386 loop_info64_to_compat(const struct loop_info64 *info64,
1387 struct compat_loop_info __user *arg)
1389 struct compat_loop_info info;
1391 memset(&info, 0, sizeof(info));
1392 info.lo_number = info64->lo_number;
1393 info.lo_device = info64->lo_device;
1394 info.lo_inode = info64->lo_inode;
1395 info.lo_rdevice = info64->lo_rdevice;
1396 info.lo_offset = info64->lo_offset;
1397 info.lo_encrypt_type = info64->lo_encrypt_type;
1398 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1399 info.lo_flags = info64->lo_flags;
1400 info.lo_init[0] = info64->lo_init[0];
1401 info.lo_init[1] = info64->lo_init[1];
1402 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1403 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1405 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1406 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1408 /* error in case values were truncated */
1409 if (info.lo_device != info64->lo_device ||
1410 info.lo_rdevice != info64->lo_rdevice ||
1411 info.lo_inode != info64->lo_inode ||
1412 info.lo_offset != info64->lo_offset ||
1413 info.lo_init[0] != info64->lo_init[0] ||
1414 info.lo_init[1] != info64->lo_init[1])
1417 if (copy_to_user(arg, &info, sizeof(info)))
1423 loop_set_status_compat(struct loop_device *lo,
1424 const struct compat_loop_info __user *arg)
1426 struct loop_info64 info64;
1429 ret = loop_info64_from_compat(arg, &info64);
1432 return loop_set_status(lo, &info64);
1436 loop_get_status_compat(struct loop_device *lo,
1437 struct compat_loop_info __user *arg)
1439 struct loop_info64 info64;
1445 err = loop_get_status(lo, &info64);
1447 err = loop_info64_to_compat(&info64, arg);
1451 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1452 unsigned int cmd, unsigned long arg)
1454 struct loop_device *lo = bdev->bd_disk->private_data;
1458 case LOOP_SET_STATUS:
1459 mutex_lock(&lo->lo_ctl_mutex);
1460 err = loop_set_status_compat(
1461 lo, (const struct compat_loop_info __user *) arg);
1462 mutex_unlock(&lo->lo_ctl_mutex);
1464 case LOOP_GET_STATUS:
1465 mutex_lock(&lo->lo_ctl_mutex);
1466 err = loop_get_status_compat(
1467 lo, (struct compat_loop_info __user *) arg);
1468 mutex_unlock(&lo->lo_ctl_mutex);
1470 case LOOP_SET_CAPACITY:
1472 case LOOP_GET_STATUS64:
1473 case LOOP_SET_STATUS64:
1474 arg = (unsigned long) compat_ptr(arg);
1476 case LOOP_CHANGE_FD:
1477 err = lo_ioctl(bdev, mode, cmd, arg);
1487 static int lo_open(struct block_device *bdev, fmode_t mode)
1489 struct loop_device *lo = bdev->bd_disk->private_data;
1491 mutex_lock(&loop_mutex);
1492 mutex_lock(&lo->lo_ctl_mutex);
1494 mutex_unlock(&lo->lo_ctl_mutex);
1495 mutex_unlock(&loop_mutex);
1500 static int lo_release(struct gendisk *disk, fmode_t mode)
1502 struct loop_device *lo = disk->private_data;
1505 mutex_lock(&loop_mutex);
1506 mutex_lock(&lo->lo_ctl_mutex);
1508 if (--lo->lo_refcnt)
1511 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1513 * In autoclear mode, stop the loop thread
1514 * and remove configuration after last close.
1516 err = loop_clr_fd(lo, NULL);
1521 * Otherwise keep thread (if running) and config,
1522 * but flush possible ongoing bios in thread.
1528 mutex_unlock(&lo->lo_ctl_mutex);
1530 mutex_unlock(&loop_mutex);
1534 static const struct block_device_operations lo_fops = {
1535 .owner = THIS_MODULE,
1537 .release = lo_release,
1539 #ifdef CONFIG_COMPAT
1540 .compat_ioctl = lo_compat_ioctl,
1545 * And now the modules code and kernel interface.
1547 static int max_loop;
1548 module_param(max_loop, int, 0);
1549 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1550 module_param(max_part, int, 0);
1551 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1552 MODULE_LICENSE("GPL");
1553 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1555 int loop_register_transfer(struct loop_func_table *funcs)
1557 unsigned int n = funcs->number;
1559 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1561 xfer_funcs[n] = funcs;
1565 int loop_unregister_transfer(int number)
1567 unsigned int n = number;
1568 struct loop_device *lo;
1569 struct loop_func_table *xfer;
1571 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1574 xfer_funcs[n] = NULL;
1576 list_for_each_entry(lo, &loop_devices, lo_list) {
1577 mutex_lock(&lo->lo_ctl_mutex);
1579 if (lo->lo_encryption == xfer)
1580 loop_release_xfer(lo);
1582 mutex_unlock(&lo->lo_ctl_mutex);
1588 EXPORT_SYMBOL(loop_register_transfer);
1589 EXPORT_SYMBOL(loop_unregister_transfer);
1591 static struct loop_device *loop_alloc(int i)
1593 struct loop_device *lo;
1594 struct gendisk *disk;
1596 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1600 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1604 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1606 goto out_free_queue;
1608 mutex_init(&lo->lo_ctl_mutex);
1610 lo->lo_thread = NULL;
1611 init_waitqueue_head(&lo->lo_event);
1612 spin_lock_init(&lo->lo_lock);
1613 disk->major = LOOP_MAJOR;
1614 disk->first_minor = i << part_shift;
1615 disk->fops = &lo_fops;
1616 disk->private_data = lo;
1617 disk->queue = lo->lo_queue;
1618 sprintf(disk->disk_name, "loop%d", i);
1622 blk_cleanup_queue(lo->lo_queue);
1629 static void loop_free(struct loop_device *lo)
1631 if (!lo->lo_queue->queue_lock)
1632 lo->lo_queue->queue_lock = &lo->lo_queue->__queue_lock;
1634 blk_cleanup_queue(lo->lo_queue);
1635 put_disk(lo->lo_disk);
1636 list_del(&lo->lo_list);
1640 static struct loop_device *loop_init_one(int i)
1642 struct loop_device *lo;
1644 list_for_each_entry(lo, &loop_devices, lo_list) {
1645 if (lo->lo_number == i)
1651 add_disk(lo->lo_disk);
1652 list_add_tail(&lo->lo_list, &loop_devices);
1657 static void loop_del_one(struct loop_device *lo)
1659 del_gendisk(lo->lo_disk);
1663 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1665 struct loop_device *lo;
1666 struct kobject *kobj;
1668 mutex_lock(&loop_devices_mutex);
1669 lo = loop_init_one(dev & MINORMASK);
1670 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1671 mutex_unlock(&loop_devices_mutex);
1677 static int __init loop_init(void)
1680 unsigned long range;
1681 struct loop_device *lo, *next;
1684 * loop module now has a feature to instantiate underlying device
1685 * structure on-demand, provided that there is an access dev node.
1686 * However, this will not work well with user space tool that doesn't
1687 * know about such "feature". In order to not break any existing
1688 * tool, we do the following:
1690 * (1) if max_loop is specified, create that many upfront, and this
1691 * also becomes a hard limit.
1692 * (2) if max_loop is not specified, create 8 loop device on module
1693 * load, user can further extend loop device by create dev node
1694 * themselves and have kernel automatically instantiate actual
1700 part_shift = fls(max_part);
1702 if (max_loop > 1UL << (MINORBITS - part_shift))
1710 range = 1UL << (MINORBITS - part_shift);
1713 if (register_blkdev(LOOP_MAJOR, "loop"))
1716 for (i = 0; i < nr; i++) {
1720 list_add_tail(&lo->lo_list, &loop_devices);
1723 /* point of no return */
1725 list_for_each_entry(lo, &loop_devices, lo_list)
1726 add_disk(lo->lo_disk);
1728 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1729 THIS_MODULE, loop_probe, NULL, NULL);
1731 printk(KERN_INFO "loop: module loaded\n");
1735 printk(KERN_INFO "loop: out of memory\n");
1737 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1740 unregister_blkdev(LOOP_MAJOR, "loop");
1744 static void __exit loop_exit(void)
1746 unsigned long range;
1747 struct loop_device *lo, *next;
1749 range = max_loop ? max_loop : 1UL << (MINORBITS - part_shift);
1751 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1754 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1755 unregister_blkdev(LOOP_MAJOR, "loop");
1758 module_init(loop_init);
1759 module_exit(loop_exit);
1762 static int __init max_loop_setup(char *str)
1764 max_loop = simple_strtol(str, NULL, 0);
1768 __setup("max_loop=", max_loop_setup);