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/compat.h>
67 #include <linux/suspend.h>
68 #include <linux/freezer.h>
69 #include <linux/mutex.h>
70 #include <linux/writeback.h>
71 #include <linux/completion.h>
72 #include <linux/highmem.h>
73 #include <linux/kthread.h>
74 #include <linux/splice.h>
75 #include <linux/sysfs.h>
76 #include <linux/miscdevice.h>
77 #include <linux/falloc.h>
78 #include <linux/uio.h>
81 #include <linux/uaccess.h>
83 static DEFINE_IDR(loop_index_idr);
84 static DEFINE_MUTEX(loop_index_mutex);
87 static int part_shift;
89 static int transfer_xor(struct loop_device *lo, int cmd,
90 struct page *raw_page, unsigned raw_off,
91 struct page *loop_page, unsigned loop_off,
92 int size, sector_t real_block)
94 char *raw_buf = kmap_atomic(raw_page) + raw_off;
95 char *loop_buf = kmap_atomic(loop_page) + loop_off;
107 key = lo->lo_encrypt_key;
108 keysize = lo->lo_encrypt_key_size;
109 for (i = 0; i < size; i++)
110 *out++ = *in++ ^ key[(i & 511) % keysize];
112 kunmap_atomic(loop_buf);
113 kunmap_atomic(raw_buf);
118 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
120 if (unlikely(info->lo_encrypt_key_size <= 0))
125 static struct loop_func_table none_funcs = {
126 .number = LO_CRYPT_NONE,
129 static struct loop_func_table xor_funcs = {
130 .number = LO_CRYPT_XOR,
131 .transfer = transfer_xor,
135 /* xfer_funcs[0] is special - its release function is never called */
136 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
141 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
145 /* Compute loopsize in bytes */
146 loopsize = i_size_read(file->f_mapping->host);
149 /* offset is beyond i_size, weird but possible */
153 if (sizelimit > 0 && sizelimit < loopsize)
154 loopsize = sizelimit;
156 * Unfortunately, if we want to do I/O on the device,
157 * the number of 512-byte sectors has to fit into a sector_t.
159 return loopsize >> 9;
162 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
164 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
167 static void __loop_update_dio(struct loop_device *lo, bool dio)
169 struct file *file = lo->lo_backing_file;
170 struct address_space *mapping = file->f_mapping;
171 struct inode *inode = mapping->host;
172 unsigned short sb_bsize = 0;
173 unsigned dio_align = 0;
176 if (inode->i_sb->s_bdev) {
177 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
178 dio_align = sb_bsize - 1;
182 * We support direct I/O only if lo_offset is aligned with the
183 * logical I/O size of backing device, and the logical block
184 * size of loop is bigger than the backing device's and the loop
185 * needn't transform transfer.
187 * TODO: the above condition may be loosed in the future, and
188 * direct I/O may be switched runtime at that time because most
189 * of requests in sane applications should be PAGE_SIZE aligned
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
193 !(lo->lo_offset & dio_align) &&
194 mapping->a_ops->direct_IO &&
203 if (lo->use_dio == use_dio)
206 /* flush dirty pages before changing direct IO */
210 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
211 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
212 * will get updated by ioctl(LOOP_GET_STATUS)
214 blk_mq_freeze_queue(lo->lo_queue);
215 lo->use_dio = use_dio;
217 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
218 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
220 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
221 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
223 blk_mq_unfreeze_queue(lo->lo_queue);
227 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
229 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
230 sector_t x = (sector_t)size;
231 struct block_device *bdev = lo->lo_device;
233 if (unlikely((loff_t)x != size))
235 if (lo->lo_offset != offset)
236 lo->lo_offset = offset;
237 if (lo->lo_sizelimit != sizelimit)
238 lo->lo_sizelimit = sizelimit;
239 set_capacity(lo->lo_disk, x);
240 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
241 /* let user-space know about the new size */
242 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
247 lo_do_transfer(struct loop_device *lo, int cmd,
248 struct page *rpage, unsigned roffs,
249 struct page *lpage, unsigned loffs,
250 int size, sector_t rblock)
254 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
258 printk_ratelimited(KERN_ERR
259 "loop: Transfer error at byte offset %llu, length %i.\n",
260 (unsigned long long)rblock << 9, size);
264 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
269 iov_iter_bvec(&i, ITER_BVEC | WRITE, bvec, 1, bvec->bv_len);
271 file_start_write(file);
272 bw = vfs_iter_write(file, &i, ppos, 0);
273 file_end_write(file);
275 if (likely(bw == bvec->bv_len))
278 printk_ratelimited(KERN_ERR
279 "loop: Write error at byte offset %llu, length %i.\n",
280 (unsigned long long)*ppos, bvec->bv_len);
286 static int lo_write_simple(struct loop_device *lo, struct request *rq,
290 struct req_iterator iter;
293 rq_for_each_segment(bvec, rq, iter) {
294 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
304 * This is the slow, transforming version that needs to double buffer the
305 * data as it cannot do the transformations in place without having direct
306 * access to the destination pages of the backing file.
308 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
311 struct bio_vec bvec, b;
312 struct req_iterator iter;
316 page = alloc_page(GFP_NOIO);
320 rq_for_each_segment(bvec, rq, iter) {
321 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
322 bvec.bv_offset, bvec.bv_len, pos >> 9);
328 b.bv_len = bvec.bv_len;
329 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
338 static int lo_read_simple(struct loop_device *lo, struct request *rq,
342 struct req_iterator iter;
346 rq_for_each_segment(bvec, rq, iter) {
347 iov_iter_bvec(&i, ITER_BVEC, &bvec, 1, bvec.bv_len);
348 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
352 flush_dcache_page(bvec.bv_page);
354 if (len != bvec.bv_len) {
357 __rq_for_each_bio(bio, rq)
367 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
370 struct bio_vec bvec, b;
371 struct req_iterator iter;
377 page = alloc_page(GFP_NOIO);
381 rq_for_each_segment(bvec, rq, iter) {
386 b.bv_len = bvec.bv_len;
388 iov_iter_bvec(&i, ITER_BVEC, &b, 1, b.bv_len);
389 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
395 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
396 bvec.bv_offset, len, offset >> 9);
400 flush_dcache_page(bvec.bv_page);
402 if (len != bvec.bv_len) {
405 __rq_for_each_bio(bio, rq)
417 static int lo_discard(struct loop_device *lo, struct request *rq, loff_t pos)
420 * We use punch hole to reclaim the free space used by the
421 * image a.k.a. discard. However we do not support discard if
422 * encryption is enabled, because it may give an attacker
423 * useful information.
425 struct file *file = lo->lo_backing_file;
426 int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
429 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
434 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
435 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
441 static int lo_req_flush(struct loop_device *lo, struct request *rq)
443 struct file *file = lo->lo_backing_file;
444 int ret = vfs_fsync(file, 0);
445 if (unlikely(ret && ret != -EINVAL))
451 static void lo_complete_rq(struct request *rq)
453 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
454 blk_status_t ret = BLK_STS_OK;
456 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
457 req_op(rq) != REQ_OP_READ) {
464 * Short READ - if we got some data, advance our request and
465 * retry it. If we got no data, end the rest with EIO.
468 blk_update_request(rq, BLK_STS_OK, cmd->ret);
470 blk_mq_requeue_request(rq, true);
473 struct bio *bio = rq->bio;
482 blk_mq_end_request(rq, ret);
486 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
488 struct request *rq = blk_mq_rq_from_pdu(cmd);
490 if (!atomic_dec_and_test(&cmd->ref))
494 blk_mq_complete_request(rq);
497 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
499 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
504 lo_rw_aio_do_completion(cmd);
507 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
510 struct iov_iter iter;
511 struct bio_vec *bvec;
512 struct request *rq = blk_mq_rq_from_pdu(cmd);
513 struct bio *bio = rq->bio;
514 struct file *file = lo->lo_backing_file;
519 if (rq->bio != rq->biotail) {
520 struct req_iterator iter;
523 __rq_for_each_bio(bio, rq)
524 segments += bio_segments(bio);
525 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO);
531 * The bios of the request may be started from the middle of
532 * the 'bvec' because of bio splitting, so we can't directly
533 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
534 * API will take care of all details for us.
536 rq_for_each_segment(tmp, rq, iter) {
544 * Same here, this bio may be started from the middle of the
545 * 'bvec' because of bio splitting, so offset from the bvec
546 * must be passed to iov iterator
548 offset = bio->bi_iter.bi_bvec_done;
549 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
550 segments = bio_segments(bio);
552 atomic_set(&cmd->ref, 2);
554 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
555 segments, blk_rq_bytes(rq));
556 iter.iov_offset = offset;
558 cmd->iocb.ki_pos = pos;
559 cmd->iocb.ki_filp = file;
560 cmd->iocb.ki_complete = lo_rw_aio_complete;
561 cmd->iocb.ki_flags = IOCB_DIRECT;
563 kthread_associate_blkcg(cmd->css);
566 ret = call_write_iter(file, &cmd->iocb, &iter);
568 ret = call_read_iter(file, &cmd->iocb, &iter);
570 lo_rw_aio_do_completion(cmd);
571 kthread_associate_blkcg(NULL);
573 if (ret != -EIOCBQUEUED)
574 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
578 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
580 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
581 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
584 * lo_write_simple and lo_read_simple should have been covered
585 * by io submit style function like lo_rw_aio(), one blocker
586 * is that lo_read_simple() need to call flush_dcache_page after
587 * the page is written from kernel, and it isn't easy to handle
588 * this in io submit style function which submits all segments
589 * of the req at one time. And direct read IO doesn't need to
590 * run flush_dcache_page().
592 switch (req_op(rq)) {
594 return lo_req_flush(lo, rq);
596 case REQ_OP_WRITE_ZEROES:
597 return lo_discard(lo, rq, pos);
600 return lo_write_transfer(lo, rq, pos);
601 else if (cmd->use_aio)
602 return lo_rw_aio(lo, cmd, pos, WRITE);
604 return lo_write_simple(lo, rq, pos);
607 return lo_read_transfer(lo, rq, pos);
608 else if (cmd->use_aio)
609 return lo_rw_aio(lo, cmd, pos, READ);
611 return lo_read_simple(lo, rq, pos);
619 static inline void loop_update_dio(struct loop_device *lo)
621 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
625 static void loop_reread_partitions(struct loop_device *lo,
626 struct block_device *bdev)
631 * bd_mutex has been held already in release path, so don't
632 * acquire it if this function is called in such case.
634 * If the reread partition isn't from release path, lo_refcnt
635 * must be at least one and it can only become zero when the
636 * current holder is released.
638 if (!atomic_read(&lo->lo_refcnt))
639 rc = __blkdev_reread_part(bdev);
641 rc = blkdev_reread_part(bdev);
643 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
644 __func__, lo->lo_number, lo->lo_file_name, rc);
648 * loop_change_fd switched the backing store of a loopback device to
649 * a new file. This is useful for operating system installers to free up
650 * the original file and in High Availability environments to switch to
651 * an alternative location for the content in case of server meltdown.
652 * This can only work if the loop device is used read-only, and if the
653 * new backing store is the same size and type as the old backing store.
655 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
658 struct file *file, *old_file;
663 if (lo->lo_state != Lo_bound)
666 /* the loop device has to be read-only */
668 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
676 inode = file->f_mapping->host;
677 old_file = lo->lo_backing_file;
681 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
684 /* size of the new backing store needs to be the same */
685 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
689 blk_mq_freeze_queue(lo->lo_queue);
690 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
691 lo->lo_backing_file = file;
692 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
693 mapping_set_gfp_mask(file->f_mapping,
694 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
696 blk_mq_unfreeze_queue(lo->lo_queue);
699 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
700 loop_reread_partitions(lo, bdev);
709 static inline int is_loop_device(struct file *file)
711 struct inode *i = file->f_mapping->host;
713 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
716 /* loop sysfs attributes */
718 static ssize_t loop_attr_show(struct device *dev, char *page,
719 ssize_t (*callback)(struct loop_device *, char *))
721 struct gendisk *disk = dev_to_disk(dev);
722 struct loop_device *lo = disk->private_data;
724 return callback(lo, page);
727 #define LOOP_ATTR_RO(_name) \
728 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
729 static ssize_t loop_attr_do_show_##_name(struct device *d, \
730 struct device_attribute *attr, char *b) \
732 return loop_attr_show(d, b, loop_attr_##_name##_show); \
734 static struct device_attribute loop_attr_##_name = \
735 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
737 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
742 spin_lock_irq(&lo->lo_lock);
743 if (lo->lo_backing_file)
744 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
745 spin_unlock_irq(&lo->lo_lock);
747 if (IS_ERR_OR_NULL(p))
751 memmove(buf, p, ret);
759 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
761 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
764 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
766 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
769 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
771 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
773 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
776 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
778 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
780 return sprintf(buf, "%s\n", partscan ? "1" : "0");
783 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
785 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
787 return sprintf(buf, "%s\n", dio ? "1" : "0");
790 LOOP_ATTR_RO(backing_file);
791 LOOP_ATTR_RO(offset);
792 LOOP_ATTR_RO(sizelimit);
793 LOOP_ATTR_RO(autoclear);
794 LOOP_ATTR_RO(partscan);
797 static struct attribute *loop_attrs[] = {
798 &loop_attr_backing_file.attr,
799 &loop_attr_offset.attr,
800 &loop_attr_sizelimit.attr,
801 &loop_attr_autoclear.attr,
802 &loop_attr_partscan.attr,
807 static struct attribute_group loop_attribute_group = {
812 static void loop_sysfs_init(struct loop_device *lo)
814 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
815 &loop_attribute_group);
818 static void loop_sysfs_exit(struct loop_device *lo)
820 if (lo->sysfs_inited)
821 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
822 &loop_attribute_group);
825 static void loop_config_discard(struct loop_device *lo)
827 struct file *file = lo->lo_backing_file;
828 struct inode *inode = file->f_mapping->host;
829 struct request_queue *q = lo->lo_queue;
832 * We use punch hole to reclaim the free space used by the
833 * image a.k.a. discard. However we do not support discard if
834 * encryption is enabled, because it may give an attacker
835 * useful information.
837 if ((!file->f_op->fallocate) ||
838 lo->lo_encrypt_key_size) {
839 q->limits.discard_granularity = 0;
840 q->limits.discard_alignment = 0;
841 blk_queue_max_discard_sectors(q, 0);
842 blk_queue_max_write_zeroes_sectors(q, 0);
843 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
847 q->limits.discard_granularity = inode->i_sb->s_blocksize;
848 q->limits.discard_alignment = 0;
850 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
851 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
852 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
855 static void loop_unprepare_queue(struct loop_device *lo)
857 kthread_flush_worker(&lo->worker);
858 kthread_stop(lo->worker_task);
861 static int loop_kthread_worker_fn(void *worker_ptr)
863 current->flags |= PF_LESS_THROTTLE;
864 return kthread_worker_fn(worker_ptr);
867 static int loop_prepare_queue(struct loop_device *lo)
869 kthread_init_worker(&lo->worker);
870 lo->worker_task = kthread_run(loop_kthread_worker_fn,
871 &lo->worker, "loop%d", lo->lo_number);
872 if (IS_ERR(lo->worker_task))
874 set_user_nice(lo->worker_task, MIN_NICE);
878 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
879 struct block_device *bdev, unsigned int arg)
881 struct file *file, *f;
883 struct address_space *mapping;
888 /* This is safe, since we have a reference from open(). */
889 __module_get(THIS_MODULE);
897 if (lo->lo_state != Lo_unbound)
900 /* Avoid recursion */
902 while (is_loop_device(f)) {
903 struct loop_device *l;
905 if (f->f_mapping->host->i_bdev == bdev)
908 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
909 if (l->lo_state == Lo_unbound) {
913 f = l->lo_backing_file;
916 mapping = file->f_mapping;
917 inode = mapping->host;
920 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
923 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
924 !file->f_op->write_iter)
925 lo_flags |= LO_FLAGS_READ_ONLY;
928 size = get_loop_size(lo, file);
929 if ((loff_t)(sector_t)size != size)
931 error = loop_prepare_queue(lo);
937 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
940 lo->lo_device = bdev;
941 lo->lo_flags = lo_flags;
942 lo->lo_backing_file = file;
945 lo->lo_sizelimit = 0;
946 lo->old_gfp_mask = mapping_gfp_mask(mapping);
947 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
949 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
950 blk_queue_write_cache(lo->lo_queue, true, false);
953 set_capacity(lo->lo_disk, size);
954 bd_set_size(bdev, size << 9);
956 /* let user-space know about the new size */
957 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
959 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
960 block_size(inode->i_bdev) : PAGE_SIZE);
962 lo->lo_state = Lo_bound;
964 lo->lo_flags |= LO_FLAGS_PARTSCAN;
965 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
966 loop_reread_partitions(lo, bdev);
968 /* Grab the block_device to prevent its destruction after we
969 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
977 /* This is safe: open() is still holding a reference. */
978 module_put(THIS_MODULE);
983 loop_release_xfer(struct loop_device *lo)
986 struct loop_func_table *xfer = lo->lo_encryption;
990 err = xfer->release(lo);
992 lo->lo_encryption = NULL;
993 module_put(xfer->owner);
999 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1000 const struct loop_info64 *i)
1005 struct module *owner = xfer->owner;
1007 if (!try_module_get(owner))
1010 err = xfer->init(lo, i);
1014 lo->lo_encryption = xfer;
1019 static int loop_clr_fd(struct loop_device *lo)
1021 struct file *filp = lo->lo_backing_file;
1022 gfp_t gfp = lo->old_gfp_mask;
1023 struct block_device *bdev = lo->lo_device;
1025 if (lo->lo_state != Lo_bound)
1029 * If we've explicitly asked to tear down the loop device,
1030 * and it has an elevated reference count, set it for auto-teardown when
1031 * the last reference goes away. This stops $!~#$@ udev from
1032 * preventing teardown because it decided that it needs to run blkid on
1033 * the loopback device whenever they appear. xfstests is notorious for
1034 * failing tests because blkid via udev races with a losetup
1035 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1036 * command to fail with EBUSY.
1038 if (atomic_read(&lo->lo_refcnt) > 1) {
1039 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1040 mutex_unlock(&lo->lo_ctl_mutex);
1047 /* freeze request queue during the transition */
1048 blk_mq_freeze_queue(lo->lo_queue);
1050 spin_lock_irq(&lo->lo_lock);
1051 lo->lo_state = Lo_rundown;
1052 lo->lo_backing_file = NULL;
1053 spin_unlock_irq(&lo->lo_lock);
1055 loop_release_xfer(lo);
1056 lo->transfer = NULL;
1058 lo->lo_device = NULL;
1059 lo->lo_encryption = NULL;
1061 lo->lo_sizelimit = 0;
1062 lo->lo_encrypt_key_size = 0;
1063 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1064 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1065 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1066 blk_queue_logical_block_size(lo->lo_queue, 512);
1067 blk_queue_physical_block_size(lo->lo_queue, 512);
1068 blk_queue_io_min(lo->lo_queue, 512);
1071 invalidate_bdev(bdev);
1073 set_capacity(lo->lo_disk, 0);
1074 loop_sysfs_exit(lo);
1076 bd_set_size(bdev, 0);
1077 /* let user-space know about this change */
1078 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1080 mapping_set_gfp_mask(filp->f_mapping, gfp);
1081 lo->lo_state = Lo_unbound;
1082 /* This is safe: open() is still holding a reference. */
1083 module_put(THIS_MODULE);
1084 blk_mq_unfreeze_queue(lo->lo_queue);
1086 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1087 loop_reread_partitions(lo, bdev);
1090 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1091 loop_unprepare_queue(lo);
1092 mutex_unlock(&lo->lo_ctl_mutex);
1094 * Need not hold lo_ctl_mutex to fput backing file.
1095 * Calling fput holding lo_ctl_mutex triggers a circular
1096 * lock dependency possibility warning as fput can take
1097 * bd_mutex which is usually taken before lo_ctl_mutex.
1104 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1107 struct loop_func_table *xfer;
1108 kuid_t uid = current_uid();
1110 if (lo->lo_encrypt_key_size &&
1111 !uid_eq(lo->lo_key_owner, uid) &&
1112 !capable(CAP_SYS_ADMIN))
1114 if (lo->lo_state != Lo_bound)
1116 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1119 /* I/O need to be drained during transfer transition */
1120 blk_mq_freeze_queue(lo->lo_queue);
1122 err = loop_release_xfer(lo);
1126 if (info->lo_encrypt_type) {
1127 unsigned int type = info->lo_encrypt_type;
1129 if (type >= MAX_LO_CRYPT) {
1133 xfer = xfer_funcs[type];
1141 err = loop_init_xfer(lo, xfer, info);
1145 if (lo->lo_offset != info->lo_offset ||
1146 lo->lo_sizelimit != info->lo_sizelimit) {
1147 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1153 loop_config_discard(lo);
1155 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1156 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1157 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1158 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1162 lo->transfer = xfer->transfer;
1163 lo->ioctl = xfer->ioctl;
1165 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1166 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1167 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1169 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1170 lo->lo_init[0] = info->lo_init[0];
1171 lo->lo_init[1] = info->lo_init[1];
1172 if (info->lo_encrypt_key_size) {
1173 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1174 info->lo_encrypt_key_size);
1175 lo->lo_key_owner = uid;
1178 /* update dio if lo_offset or transfer is changed */
1179 __loop_update_dio(lo, lo->use_dio);
1182 blk_mq_unfreeze_queue(lo->lo_queue);
1184 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1185 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1186 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1187 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1188 loop_reread_partitions(lo, lo->lo_device);
1195 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1201 if (lo->lo_state != Lo_bound) {
1202 mutex_unlock(&lo->lo_ctl_mutex);
1206 memset(info, 0, sizeof(*info));
1207 info->lo_number = lo->lo_number;
1208 info->lo_offset = lo->lo_offset;
1209 info->lo_sizelimit = lo->lo_sizelimit;
1210 info->lo_flags = lo->lo_flags;
1211 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1212 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1213 info->lo_encrypt_type =
1214 lo->lo_encryption ? lo->lo_encryption->number : 0;
1215 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1216 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1217 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1218 lo->lo_encrypt_key_size);
1221 /* Drop lo_ctl_mutex while we call into the filesystem. */
1222 file = get_file(lo->lo_backing_file);
1223 mutex_unlock(&lo->lo_ctl_mutex);
1224 ret = vfs_getattr(&file->f_path, &stat, STATX_INO,
1225 AT_STATX_SYNC_AS_STAT);
1227 info->lo_device = huge_encode_dev(stat.dev);
1228 info->lo_inode = stat.ino;
1229 info->lo_rdevice = huge_encode_dev(stat.rdev);
1236 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1238 memset(info64, 0, sizeof(*info64));
1239 info64->lo_number = info->lo_number;
1240 info64->lo_device = info->lo_device;
1241 info64->lo_inode = info->lo_inode;
1242 info64->lo_rdevice = info->lo_rdevice;
1243 info64->lo_offset = info->lo_offset;
1244 info64->lo_sizelimit = 0;
1245 info64->lo_encrypt_type = info->lo_encrypt_type;
1246 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1247 info64->lo_flags = info->lo_flags;
1248 info64->lo_init[0] = info->lo_init[0];
1249 info64->lo_init[1] = info->lo_init[1];
1250 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1251 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1253 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1254 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1258 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1260 memset(info, 0, sizeof(*info));
1261 info->lo_number = info64->lo_number;
1262 info->lo_device = info64->lo_device;
1263 info->lo_inode = info64->lo_inode;
1264 info->lo_rdevice = info64->lo_rdevice;
1265 info->lo_offset = info64->lo_offset;
1266 info->lo_encrypt_type = info64->lo_encrypt_type;
1267 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1268 info->lo_flags = info64->lo_flags;
1269 info->lo_init[0] = info64->lo_init[0];
1270 info->lo_init[1] = info64->lo_init[1];
1271 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1272 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1274 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1275 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1277 /* error in case values were truncated */
1278 if (info->lo_device != info64->lo_device ||
1279 info->lo_rdevice != info64->lo_rdevice ||
1280 info->lo_inode != info64->lo_inode ||
1281 info->lo_offset != info64->lo_offset)
1288 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1290 struct loop_info info;
1291 struct loop_info64 info64;
1293 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1295 loop_info64_from_old(&info, &info64);
1296 return loop_set_status(lo, &info64);
1300 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1302 struct loop_info64 info64;
1304 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1306 return loop_set_status(lo, &info64);
1310 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1311 struct loop_info info;
1312 struct loop_info64 info64;
1316 mutex_unlock(&lo->lo_ctl_mutex);
1319 err = loop_get_status(lo, &info64);
1321 err = loop_info64_to_old(&info64, &info);
1322 if (!err && copy_to_user(arg, &info, sizeof(info)))
1329 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1330 struct loop_info64 info64;
1334 mutex_unlock(&lo->lo_ctl_mutex);
1337 err = loop_get_status(lo, &info64);
1338 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1344 static int loop_set_capacity(struct loop_device *lo)
1346 if (unlikely(lo->lo_state != Lo_bound))
1349 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1352 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1355 if (lo->lo_state != Lo_bound)
1358 __loop_update_dio(lo, !!arg);
1359 if (lo->use_dio == !!arg)
1366 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1368 if (lo->lo_state != Lo_bound)
1371 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1374 blk_mq_freeze_queue(lo->lo_queue);
1376 blk_queue_logical_block_size(lo->lo_queue, arg);
1377 blk_queue_physical_block_size(lo->lo_queue, arg);
1378 blk_queue_io_min(lo->lo_queue, arg);
1379 loop_update_dio(lo);
1381 blk_mq_unfreeze_queue(lo->lo_queue);
1386 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1387 unsigned int cmd, unsigned long arg)
1389 struct loop_device *lo = bdev->bd_disk->private_data;
1392 err = mutex_lock_killable_nested(&lo->lo_ctl_mutex, 1);
1398 err = loop_set_fd(lo, mode, bdev, arg);
1400 case LOOP_CHANGE_FD:
1401 err = loop_change_fd(lo, bdev, arg);
1404 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1405 err = loop_clr_fd(lo);
1409 case LOOP_SET_STATUS:
1411 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1412 err = loop_set_status_old(lo,
1413 (struct loop_info __user *)arg);
1415 case LOOP_GET_STATUS:
1416 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1417 /* loop_get_status() unlocks lo_ctl_mutex */
1419 case LOOP_SET_STATUS64:
1421 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1422 err = loop_set_status64(lo,
1423 (struct loop_info64 __user *) arg);
1425 case LOOP_GET_STATUS64:
1426 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1427 /* loop_get_status() unlocks lo_ctl_mutex */
1429 case LOOP_SET_CAPACITY:
1431 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1432 err = loop_set_capacity(lo);
1434 case LOOP_SET_DIRECT_IO:
1436 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1437 err = loop_set_dio(lo, arg);
1439 case LOOP_SET_BLOCK_SIZE:
1441 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1442 err = loop_set_block_size(lo, arg);
1445 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1447 mutex_unlock(&lo->lo_ctl_mutex);
1453 #ifdef CONFIG_COMPAT
1454 struct compat_loop_info {
1455 compat_int_t lo_number; /* ioctl r/o */
1456 compat_dev_t lo_device; /* ioctl r/o */
1457 compat_ulong_t lo_inode; /* ioctl r/o */
1458 compat_dev_t lo_rdevice; /* ioctl r/o */
1459 compat_int_t lo_offset;
1460 compat_int_t lo_encrypt_type;
1461 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1462 compat_int_t lo_flags; /* ioctl r/o */
1463 char lo_name[LO_NAME_SIZE];
1464 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1465 compat_ulong_t lo_init[2];
1470 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1471 * - noinlined to reduce stack space usage in main part of driver
1474 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1475 struct loop_info64 *info64)
1477 struct compat_loop_info info;
1479 if (copy_from_user(&info, arg, sizeof(info)))
1482 memset(info64, 0, sizeof(*info64));
1483 info64->lo_number = info.lo_number;
1484 info64->lo_device = info.lo_device;
1485 info64->lo_inode = info.lo_inode;
1486 info64->lo_rdevice = info.lo_rdevice;
1487 info64->lo_offset = info.lo_offset;
1488 info64->lo_sizelimit = 0;
1489 info64->lo_encrypt_type = info.lo_encrypt_type;
1490 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1491 info64->lo_flags = info.lo_flags;
1492 info64->lo_init[0] = info.lo_init[0];
1493 info64->lo_init[1] = info.lo_init[1];
1494 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1495 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1497 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1498 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1503 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1504 * - noinlined to reduce stack space usage in main part of driver
1507 loop_info64_to_compat(const struct loop_info64 *info64,
1508 struct compat_loop_info __user *arg)
1510 struct compat_loop_info info;
1512 memset(&info, 0, sizeof(info));
1513 info.lo_number = info64->lo_number;
1514 info.lo_device = info64->lo_device;
1515 info.lo_inode = info64->lo_inode;
1516 info.lo_rdevice = info64->lo_rdevice;
1517 info.lo_offset = info64->lo_offset;
1518 info.lo_encrypt_type = info64->lo_encrypt_type;
1519 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1520 info.lo_flags = info64->lo_flags;
1521 info.lo_init[0] = info64->lo_init[0];
1522 info.lo_init[1] = info64->lo_init[1];
1523 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1524 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1526 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1527 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1529 /* error in case values were truncated */
1530 if (info.lo_device != info64->lo_device ||
1531 info.lo_rdevice != info64->lo_rdevice ||
1532 info.lo_inode != info64->lo_inode ||
1533 info.lo_offset != info64->lo_offset ||
1534 info.lo_init[0] != info64->lo_init[0] ||
1535 info.lo_init[1] != info64->lo_init[1])
1538 if (copy_to_user(arg, &info, sizeof(info)))
1544 loop_set_status_compat(struct loop_device *lo,
1545 const struct compat_loop_info __user *arg)
1547 struct loop_info64 info64;
1550 ret = loop_info64_from_compat(arg, &info64);
1553 return loop_set_status(lo, &info64);
1557 loop_get_status_compat(struct loop_device *lo,
1558 struct compat_loop_info __user *arg)
1560 struct loop_info64 info64;
1564 mutex_unlock(&lo->lo_ctl_mutex);
1567 err = loop_get_status(lo, &info64);
1569 err = loop_info64_to_compat(&info64, arg);
1573 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1574 unsigned int cmd, unsigned long arg)
1576 struct loop_device *lo = bdev->bd_disk->private_data;
1580 case LOOP_SET_STATUS:
1581 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1583 err = loop_set_status_compat(lo,
1584 (const struct compat_loop_info __user *)arg);
1585 mutex_unlock(&lo->lo_ctl_mutex);
1588 case LOOP_GET_STATUS:
1589 err = mutex_lock_killable(&lo->lo_ctl_mutex);
1591 err = loop_get_status_compat(lo,
1592 (struct compat_loop_info __user *)arg);
1593 /* loop_get_status() unlocks lo_ctl_mutex */
1596 case LOOP_SET_CAPACITY:
1598 case LOOP_GET_STATUS64:
1599 case LOOP_SET_STATUS64:
1600 arg = (unsigned long) compat_ptr(arg);
1602 case LOOP_CHANGE_FD:
1603 err = lo_ioctl(bdev, mode, cmd, arg);
1613 static int lo_open(struct block_device *bdev, fmode_t mode)
1615 struct loop_device *lo;
1618 mutex_lock(&loop_index_mutex);
1619 lo = bdev->bd_disk->private_data;
1625 atomic_inc(&lo->lo_refcnt);
1627 mutex_unlock(&loop_index_mutex);
1631 static void __lo_release(struct loop_device *lo)
1635 if (atomic_dec_return(&lo->lo_refcnt))
1638 mutex_lock(&lo->lo_ctl_mutex);
1639 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1641 * In autoclear mode, stop the loop thread
1642 * and remove configuration after last close.
1644 err = loop_clr_fd(lo);
1647 } else if (lo->lo_state == Lo_bound) {
1649 * Otherwise keep thread (if running) and config,
1650 * but flush possible ongoing bios in thread.
1652 blk_mq_freeze_queue(lo->lo_queue);
1653 blk_mq_unfreeze_queue(lo->lo_queue);
1656 mutex_unlock(&lo->lo_ctl_mutex);
1659 static void lo_release(struct gendisk *disk, fmode_t mode)
1661 mutex_lock(&loop_index_mutex);
1662 __lo_release(disk->private_data);
1663 mutex_unlock(&loop_index_mutex);
1666 static const struct block_device_operations lo_fops = {
1667 .owner = THIS_MODULE,
1669 .release = lo_release,
1671 #ifdef CONFIG_COMPAT
1672 .compat_ioctl = lo_compat_ioctl,
1677 * And now the modules code and kernel interface.
1679 static int max_loop;
1680 module_param(max_loop, int, S_IRUGO);
1681 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1682 module_param(max_part, int, S_IRUGO);
1683 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1684 MODULE_LICENSE("GPL");
1685 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1687 int loop_register_transfer(struct loop_func_table *funcs)
1689 unsigned int n = funcs->number;
1691 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1693 xfer_funcs[n] = funcs;
1697 static int unregister_transfer_cb(int id, void *ptr, void *data)
1699 struct loop_device *lo = ptr;
1700 struct loop_func_table *xfer = data;
1702 mutex_lock(&lo->lo_ctl_mutex);
1703 if (lo->lo_encryption == xfer)
1704 loop_release_xfer(lo);
1705 mutex_unlock(&lo->lo_ctl_mutex);
1709 int loop_unregister_transfer(int number)
1711 unsigned int n = number;
1712 struct loop_func_table *xfer;
1714 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1717 xfer_funcs[n] = NULL;
1718 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1722 EXPORT_SYMBOL(loop_register_transfer);
1723 EXPORT_SYMBOL(loop_unregister_transfer);
1725 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1726 const struct blk_mq_queue_data *bd)
1728 struct request *rq = bd->rq;
1729 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1730 struct loop_device *lo = rq->q->queuedata;
1732 blk_mq_start_request(rq);
1734 if (lo->lo_state != Lo_bound)
1735 return BLK_STS_IOERR;
1737 switch (req_op(rq)) {
1739 case REQ_OP_DISCARD:
1740 case REQ_OP_WRITE_ZEROES:
1741 cmd->use_aio = false;
1744 cmd->use_aio = lo->use_dio;
1748 /* always use the first bio's css */
1749 #ifdef CONFIG_BLK_CGROUP
1750 if (cmd->use_aio && rq->bio && rq->bio->bi_css) {
1751 cmd->css = rq->bio->bi_css;
1756 kthread_queue_work(&lo->worker, &cmd->work);
1761 static void loop_handle_cmd(struct loop_cmd *cmd)
1763 struct request *rq = blk_mq_rq_from_pdu(cmd);
1764 const bool write = op_is_write(req_op(rq));
1765 struct loop_device *lo = rq->q->queuedata;
1768 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1773 ret = do_req_filebacked(lo, rq);
1775 /* complete non-aio request */
1776 if (!cmd->use_aio || ret) {
1777 cmd->ret = ret ? -EIO : 0;
1778 blk_mq_complete_request(rq);
1782 static void loop_queue_work(struct kthread_work *work)
1784 struct loop_cmd *cmd =
1785 container_of(work, struct loop_cmd, work);
1787 loop_handle_cmd(cmd);
1790 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1791 unsigned int hctx_idx, unsigned int numa_node)
1793 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1795 kthread_init_work(&cmd->work, loop_queue_work);
1799 static const struct blk_mq_ops loop_mq_ops = {
1800 .queue_rq = loop_queue_rq,
1801 .init_request = loop_init_request,
1802 .complete = lo_complete_rq,
1805 static int loop_add(struct loop_device **l, int i)
1807 struct loop_device *lo;
1808 struct gendisk *disk;
1812 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1816 lo->lo_state = Lo_unbound;
1818 /* allocate id, if @id >= 0, we're requesting that specific id */
1820 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1824 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1831 lo->tag_set.ops = &loop_mq_ops;
1832 lo->tag_set.nr_hw_queues = 1;
1833 lo->tag_set.queue_depth = 128;
1834 lo->tag_set.numa_node = NUMA_NO_NODE;
1835 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1836 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1837 lo->tag_set.driver_data = lo;
1839 err = blk_mq_alloc_tag_set(&lo->tag_set);
1843 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1844 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1845 err = PTR_ERR(lo->lo_queue);
1846 goto out_cleanup_tags;
1848 lo->lo_queue->queuedata = lo;
1850 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1853 * By default, we do buffer IO, so it doesn't make sense to enable
1854 * merge because the I/O submitted to backing file is handled page by
1855 * page. For directio mode, merge does help to dispatch bigger request
1856 * to underlayer disk. We will enable merge once directio is enabled.
1858 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1861 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1863 goto out_free_queue;
1866 * Disable partition scanning by default. The in-kernel partition
1867 * scanning can be requested individually per-device during its
1868 * setup. Userspace can always add and remove partitions from all
1869 * devices. The needed partition minors are allocated from the
1870 * extended minor space, the main loop device numbers will continue
1871 * to match the loop minors, regardless of the number of partitions
1874 * If max_part is given, partition scanning is globally enabled for
1875 * all loop devices. The minors for the main loop devices will be
1876 * multiples of max_part.
1878 * Note: Global-for-all-devices, set-only-at-init, read-only module
1879 * parameteters like 'max_loop' and 'max_part' make things needlessly
1880 * complicated, are too static, inflexible and may surprise
1881 * userspace tools. Parameters like this in general should be avoided.
1884 disk->flags |= GENHD_FL_NO_PART_SCAN;
1885 disk->flags |= GENHD_FL_EXT_DEVT;
1886 mutex_init(&lo->lo_ctl_mutex);
1887 atomic_set(&lo->lo_refcnt, 0);
1889 spin_lock_init(&lo->lo_lock);
1890 disk->major = LOOP_MAJOR;
1891 disk->first_minor = i << part_shift;
1892 disk->fops = &lo_fops;
1893 disk->private_data = lo;
1894 disk->queue = lo->lo_queue;
1895 sprintf(disk->disk_name, "loop%d", i);
1898 return lo->lo_number;
1901 blk_cleanup_queue(lo->lo_queue);
1903 blk_mq_free_tag_set(&lo->tag_set);
1905 idr_remove(&loop_index_idr, i);
1912 static void loop_remove(struct loop_device *lo)
1914 del_gendisk(lo->lo_disk);
1915 blk_cleanup_queue(lo->lo_queue);
1916 blk_mq_free_tag_set(&lo->tag_set);
1917 put_disk(lo->lo_disk);
1921 static int find_free_cb(int id, void *ptr, void *data)
1923 struct loop_device *lo = ptr;
1924 struct loop_device **l = data;
1926 if (lo->lo_state == Lo_unbound) {
1933 static int loop_lookup(struct loop_device **l, int i)
1935 struct loop_device *lo;
1941 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1944 ret = lo->lo_number;
1949 /* lookup and return a specific i */
1950 lo = idr_find(&loop_index_idr, i);
1953 ret = lo->lo_number;
1959 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1961 struct loop_device *lo;
1962 struct kobject *kobj;
1965 mutex_lock(&loop_index_mutex);
1966 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1968 err = loop_add(&lo, MINOR(dev) >> part_shift);
1972 kobj = get_disk_and_module(lo->lo_disk);
1973 mutex_unlock(&loop_index_mutex);
1979 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1982 struct loop_device *lo;
1985 mutex_lock(&loop_index_mutex);
1988 ret = loop_lookup(&lo, parm);
1993 ret = loop_add(&lo, parm);
1995 case LOOP_CTL_REMOVE:
1996 ret = loop_lookup(&lo, parm);
1999 ret = mutex_lock_killable(&lo->lo_ctl_mutex);
2002 if (lo->lo_state != Lo_unbound) {
2004 mutex_unlock(&lo->lo_ctl_mutex);
2007 if (atomic_read(&lo->lo_refcnt) > 0) {
2009 mutex_unlock(&lo->lo_ctl_mutex);
2012 lo->lo_disk->private_data = NULL;
2013 mutex_unlock(&lo->lo_ctl_mutex);
2014 idr_remove(&loop_index_idr, lo->lo_number);
2017 case LOOP_CTL_GET_FREE:
2018 ret = loop_lookup(&lo, -1);
2021 ret = loop_add(&lo, -1);
2023 mutex_unlock(&loop_index_mutex);
2028 static const struct file_operations loop_ctl_fops = {
2029 .open = nonseekable_open,
2030 .unlocked_ioctl = loop_control_ioctl,
2031 .compat_ioctl = loop_control_ioctl,
2032 .owner = THIS_MODULE,
2033 .llseek = noop_llseek,
2036 static struct miscdevice loop_misc = {
2037 .minor = LOOP_CTRL_MINOR,
2038 .name = "loop-control",
2039 .fops = &loop_ctl_fops,
2042 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2043 MODULE_ALIAS("devname:loop-control");
2045 static int __init loop_init(void)
2048 unsigned long range;
2049 struct loop_device *lo;
2054 part_shift = fls(max_part);
2057 * Adjust max_part according to part_shift as it is exported
2058 * to user space so that user can decide correct minor number
2059 * if [s]he want to create more devices.
2061 * Note that -1 is required because partition 0 is reserved
2062 * for the whole disk.
2064 max_part = (1UL << part_shift) - 1;
2067 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2072 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2078 * If max_loop is specified, create that many devices upfront.
2079 * This also becomes a hard limit. If max_loop is not specified,
2080 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2081 * init time. Loop devices can be requested on-demand with the
2082 * /dev/loop-control interface, or be instantiated by accessing
2083 * a 'dead' device node.
2087 range = max_loop << part_shift;
2089 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2090 range = 1UL << MINORBITS;
2093 err = misc_register(&loop_misc);
2098 if (register_blkdev(LOOP_MAJOR, "loop")) {
2103 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2104 THIS_MODULE, loop_probe, NULL, NULL);
2106 /* pre-create number of devices given by config or max_loop */
2107 mutex_lock(&loop_index_mutex);
2108 for (i = 0; i < nr; i++)
2110 mutex_unlock(&loop_index_mutex);
2112 printk(KERN_INFO "loop: module loaded\n");
2116 misc_deregister(&loop_misc);
2121 static int loop_exit_cb(int id, void *ptr, void *data)
2123 struct loop_device *lo = ptr;
2129 static void __exit loop_exit(void)
2131 unsigned long range;
2133 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2135 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2136 idr_destroy(&loop_index_idr);
2138 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2139 unregister_blkdev(LOOP_MAJOR, "loop");
2141 misc_deregister(&loop_misc);
2144 module_init(loop_init);
2145 module_exit(loop_exit);
2148 static int __init max_loop_setup(char *str)
2150 max_loop = simple_strtol(str, NULL, 0);
2154 __setup("max_loop=", max_loop_setup);