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 queue_flag_clear_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
218 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
220 queue_flag_set_unlocked(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_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
421 * We use fallocate to manipulate the space mappings used by the image
422 * a.k.a. discard/zerorange. However we do not support this if
423 * encryption is enabled, because it may give an attacker useful
426 struct file *file = lo->lo_backing_file;
429 mode |= FALLOC_FL_KEEP_SIZE;
431 if ((!file->f_op->fallocate) || lo->lo_encrypt_key_size) {
436 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
437 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
443 static int lo_req_flush(struct loop_device *lo, struct request *rq)
445 struct file *file = lo->lo_backing_file;
446 int ret = vfs_fsync(file, 0);
447 if (unlikely(ret && ret != -EINVAL))
453 static void lo_complete_rq(struct request *rq)
455 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
457 if (unlikely(req_op(cmd->rq) == REQ_OP_READ && cmd->use_aio &&
458 cmd->ret >= 0 && cmd->ret < blk_rq_bytes(cmd->rq))) {
459 struct bio *bio = cmd->rq->bio;
461 bio_advance(bio, cmd->ret);
465 blk_mq_end_request(rq, cmd->ret < 0 ? BLK_STS_IOERR : BLK_STS_OK);
468 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
470 if (!atomic_dec_and_test(&cmd->ref))
474 blk_mq_complete_request(cmd->rq);
477 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
479 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
482 lo_rw_aio_do_completion(cmd);
485 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
488 struct iov_iter iter;
489 struct bio_vec *bvec;
490 struct request *rq = cmd->rq;
491 struct bio *bio = rq->bio;
492 struct file *file = lo->lo_backing_file;
497 if (rq->bio != rq->biotail) {
498 struct req_iterator iter;
501 __rq_for_each_bio(bio, rq)
502 segments += bio_segments(bio);
503 bvec = kmalloc(sizeof(struct bio_vec) * segments, GFP_NOIO);
509 * The bios of the request may be started from the middle of
510 * the 'bvec' because of bio splitting, so we can't directly
511 * copy bio->bi_iov_vec to new bvec. The rq_for_each_segment
512 * API will take care of all details for us.
514 rq_for_each_segment(tmp, rq, iter) {
522 * Same here, this bio may be started from the middle of the
523 * 'bvec' because of bio splitting, so offset from the bvec
524 * must be passed to iov iterator
526 offset = bio->bi_iter.bi_bvec_done;
527 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
528 segments = bio_segments(bio);
530 atomic_set(&cmd->ref, 2);
532 iov_iter_bvec(&iter, ITER_BVEC | rw, bvec,
533 segments, blk_rq_bytes(rq));
534 iter.iov_offset = offset;
536 cmd->iocb.ki_pos = pos;
537 cmd->iocb.ki_filp = file;
538 cmd->iocb.ki_complete = lo_rw_aio_complete;
539 cmd->iocb.ki_flags = IOCB_DIRECT;
542 ret = call_write_iter(file, &cmd->iocb, &iter);
544 ret = call_read_iter(file, &cmd->iocb, &iter);
546 lo_rw_aio_do_completion(cmd);
548 if (ret != -EIOCBQUEUED)
549 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
553 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
555 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
556 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
559 * lo_write_simple and lo_read_simple should have been covered
560 * by io submit style function like lo_rw_aio(), one blocker
561 * is that lo_read_simple() need to call flush_dcache_page after
562 * the page is written from kernel, and it isn't easy to handle
563 * this in io submit style function which submits all segments
564 * of the req at one time. And direct read IO doesn't need to
565 * run flush_dcache_page().
567 switch (req_op(rq)) {
569 return lo_req_flush(lo, rq);
570 case REQ_OP_WRITE_ZEROES:
572 * If the caller doesn't want deallocation, call zeroout to
573 * write zeroes the range. Otherwise, punch them out.
575 return lo_fallocate(lo, rq, pos,
576 (rq->cmd_flags & REQ_NOUNMAP) ?
577 FALLOC_FL_ZERO_RANGE :
578 FALLOC_FL_PUNCH_HOLE);
580 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
583 return lo_write_transfer(lo, rq, pos);
584 else if (cmd->use_aio)
585 return lo_rw_aio(lo, cmd, pos, WRITE);
587 return lo_write_simple(lo, rq, pos);
590 return lo_read_transfer(lo, rq, pos);
591 else if (cmd->use_aio)
592 return lo_rw_aio(lo, cmd, pos, READ);
594 return lo_read_simple(lo, rq, pos);
602 static inline void loop_update_dio(struct loop_device *lo)
604 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
608 static void loop_reread_partitions(struct loop_device *lo,
609 struct block_device *bdev)
614 * bd_mutex has been held already in release path, so don't
615 * acquire it if this function is called in such case.
617 * If the reread partition isn't from release path, lo_refcnt
618 * must be at least one and it can only become zero when the
619 * current holder is released.
621 if (!atomic_read(&lo->lo_refcnt))
622 rc = __blkdev_reread_part(bdev);
624 rc = blkdev_reread_part(bdev);
626 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
627 __func__, lo->lo_number, lo->lo_file_name, rc);
630 static inline int is_loop_device(struct file *file)
632 struct inode *i = file->f_mapping->host;
634 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
637 static int loop_validate_file(struct file *file, struct block_device *bdev)
639 struct inode *inode = file->f_mapping->host;
640 struct file *f = file;
642 /* Avoid recursion */
643 while (is_loop_device(f)) {
644 struct loop_device *l;
646 if (f->f_mapping->host->i_bdev == bdev)
649 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
650 if (l->lo_state == Lo_unbound) {
653 f = l->lo_backing_file;
655 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
661 * loop_change_fd switched the backing store of a loopback device to
662 * a new file. This is useful for operating system installers to free up
663 * the original file and in High Availability environments to switch to
664 * an alternative location for the content in case of server meltdown.
665 * This can only work if the loop device is used read-only, and if the
666 * new backing store is the same size and type as the old backing store.
668 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
671 struct file *file, *old_file;
676 if (lo->lo_state != Lo_bound)
679 /* the loop device has to be read-only */
681 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
689 error = loop_validate_file(file, bdev);
693 inode = file->f_mapping->host;
694 old_file = lo->lo_backing_file;
698 /* size of the new backing store needs to be the same */
699 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
703 blk_mq_freeze_queue(lo->lo_queue);
704 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
705 lo->lo_backing_file = file;
706 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
707 mapping_set_gfp_mask(file->f_mapping,
708 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
710 blk_mq_unfreeze_queue(lo->lo_queue);
713 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
714 loop_reread_partitions(lo, bdev);
723 /* loop sysfs attributes */
725 static ssize_t loop_attr_show(struct device *dev, char *page,
726 ssize_t (*callback)(struct loop_device *, char *))
728 struct gendisk *disk = dev_to_disk(dev);
729 struct loop_device *lo = disk->private_data;
731 return callback(lo, page);
734 #define LOOP_ATTR_RO(_name) \
735 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
736 static ssize_t loop_attr_do_show_##_name(struct device *d, \
737 struct device_attribute *attr, char *b) \
739 return loop_attr_show(d, b, loop_attr_##_name##_show); \
741 static struct device_attribute loop_attr_##_name = \
742 __ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
744 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
749 spin_lock_irq(&lo->lo_lock);
750 if (lo->lo_backing_file)
751 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
752 spin_unlock_irq(&lo->lo_lock);
754 if (IS_ERR_OR_NULL(p))
758 memmove(buf, p, ret);
766 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
768 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
771 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
773 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
776 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
778 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
780 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
783 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
785 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
787 return sprintf(buf, "%s\n", partscan ? "1" : "0");
790 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
792 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
794 return sprintf(buf, "%s\n", dio ? "1" : "0");
797 LOOP_ATTR_RO(backing_file);
798 LOOP_ATTR_RO(offset);
799 LOOP_ATTR_RO(sizelimit);
800 LOOP_ATTR_RO(autoclear);
801 LOOP_ATTR_RO(partscan);
804 static struct attribute *loop_attrs[] = {
805 &loop_attr_backing_file.attr,
806 &loop_attr_offset.attr,
807 &loop_attr_sizelimit.attr,
808 &loop_attr_autoclear.attr,
809 &loop_attr_partscan.attr,
814 static struct attribute_group loop_attribute_group = {
819 static void loop_sysfs_init(struct loop_device *lo)
821 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
822 &loop_attribute_group);
825 static void loop_sysfs_exit(struct loop_device *lo)
827 if (lo->sysfs_inited)
828 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
829 &loop_attribute_group);
832 static void loop_config_discard(struct loop_device *lo)
834 struct file *file = lo->lo_backing_file;
835 struct inode *inode = file->f_mapping->host;
836 struct request_queue *q = lo->lo_queue;
839 * We use punch hole to reclaim the free space used by the
840 * image a.k.a. discard. However we do not support discard if
841 * encryption is enabled, because it may give an attacker
842 * useful information.
844 if ((!file->f_op->fallocate) ||
845 lo->lo_encrypt_key_size) {
846 q->limits.discard_granularity = 0;
847 q->limits.discard_alignment = 0;
848 blk_queue_max_discard_sectors(q, 0);
849 blk_queue_max_write_zeroes_sectors(q, 0);
850 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
854 q->limits.discard_granularity = inode->i_sb->s_blocksize;
855 q->limits.discard_alignment = 0;
857 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
858 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
859 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
862 static void loop_unprepare_queue(struct loop_device *lo)
864 kthread_flush_worker(&lo->worker);
865 kthread_stop(lo->worker_task);
868 static int loop_kthread_worker_fn(void *worker_ptr)
870 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
871 return kthread_worker_fn(worker_ptr);
874 static int loop_prepare_queue(struct loop_device *lo)
876 kthread_init_worker(&lo->worker);
877 lo->worker_task = kthread_run(loop_kthread_worker_fn,
878 &lo->worker, "loop%d", lo->lo_number);
879 if (IS_ERR(lo->worker_task))
881 set_user_nice(lo->worker_task, MIN_NICE);
885 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
886 struct block_device *bdev, unsigned int arg)
890 struct address_space *mapping;
895 /* This is safe, since we have a reference from open(). */
896 __module_get(THIS_MODULE);
904 if (lo->lo_state != Lo_unbound)
907 error = loop_validate_file(file, bdev);
911 mapping = file->f_mapping;
912 inode = mapping->host;
914 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
915 !file->f_op->write_iter)
916 lo_flags |= LO_FLAGS_READ_ONLY;
919 size = get_loop_size(lo, file);
920 if ((loff_t)(sector_t)size != size)
922 error = loop_prepare_queue(lo);
928 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
931 lo->lo_device = bdev;
932 lo->lo_flags = lo_flags;
933 lo->lo_backing_file = file;
936 lo->lo_sizelimit = 0;
937 lo->old_gfp_mask = mapping_gfp_mask(mapping);
938 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
940 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
941 blk_queue_write_cache(lo->lo_queue, true, false);
944 set_capacity(lo->lo_disk, size);
945 bd_set_size(bdev, size << 9);
947 /* let user-space know about the new size */
948 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
950 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
951 block_size(inode->i_bdev) : PAGE_SIZE);
953 lo->lo_state = Lo_bound;
955 lo->lo_flags |= LO_FLAGS_PARTSCAN;
956 if (lo->lo_flags & LO_FLAGS_PARTSCAN)
957 loop_reread_partitions(lo, bdev);
959 /* Grab the block_device to prevent its destruction after we
960 * put /dev/loopXX inode. Later in loop_clr_fd() we bdput(bdev).
968 /* This is safe: open() is still holding a reference. */
969 module_put(THIS_MODULE);
974 loop_release_xfer(struct loop_device *lo)
977 struct loop_func_table *xfer = lo->lo_encryption;
981 err = xfer->release(lo);
983 lo->lo_encryption = NULL;
984 module_put(xfer->owner);
990 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
991 const struct loop_info64 *i)
996 struct module *owner = xfer->owner;
998 if (!try_module_get(owner))
1001 err = xfer->init(lo, i);
1005 lo->lo_encryption = xfer;
1010 static int loop_clr_fd(struct loop_device *lo)
1012 struct file *filp = lo->lo_backing_file;
1013 gfp_t gfp = lo->old_gfp_mask;
1014 struct block_device *bdev = lo->lo_device;
1016 if (lo->lo_state != Lo_bound)
1020 * If we've explicitly asked to tear down the loop device,
1021 * and it has an elevated reference count, set it for auto-teardown when
1022 * the last reference goes away. This stops $!~#$@ udev from
1023 * preventing teardown because it decided that it needs to run blkid on
1024 * the loopback device whenever they appear. xfstests is notorious for
1025 * failing tests because blkid via udev races with a losetup
1026 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1027 * command to fail with EBUSY.
1029 if (atomic_read(&lo->lo_refcnt) > 1) {
1030 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1031 mutex_unlock(&lo->lo_ctl_mutex);
1038 /* freeze request queue during the transition */
1039 blk_mq_freeze_queue(lo->lo_queue);
1041 spin_lock_irq(&lo->lo_lock);
1042 lo->lo_state = Lo_rundown;
1043 lo->lo_backing_file = NULL;
1044 spin_unlock_irq(&lo->lo_lock);
1046 loop_release_xfer(lo);
1047 lo->transfer = NULL;
1049 lo->lo_device = NULL;
1050 lo->lo_encryption = NULL;
1052 lo->lo_sizelimit = 0;
1053 lo->lo_encrypt_key_size = 0;
1054 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1055 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1056 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1057 blk_queue_logical_block_size(lo->lo_queue, 512);
1058 blk_queue_physical_block_size(lo->lo_queue, 512);
1059 blk_queue_io_min(lo->lo_queue, 512);
1062 invalidate_bdev(bdev);
1064 set_capacity(lo->lo_disk, 0);
1065 loop_sysfs_exit(lo);
1067 bd_set_size(bdev, 0);
1068 /* let user-space know about this change */
1069 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1071 mapping_set_gfp_mask(filp->f_mapping, gfp);
1072 lo->lo_state = Lo_unbound;
1073 /* This is safe: open() is still holding a reference. */
1074 module_put(THIS_MODULE);
1075 blk_mq_unfreeze_queue(lo->lo_queue);
1077 if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1078 loop_reread_partitions(lo, bdev);
1081 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1082 loop_unprepare_queue(lo);
1083 mutex_unlock(&lo->lo_ctl_mutex);
1085 * Need not hold lo_ctl_mutex to fput backing file.
1086 * Calling fput holding lo_ctl_mutex triggers a circular
1087 * lock dependency possibility warning as fput can take
1088 * bd_mutex which is usually taken before lo_ctl_mutex.
1095 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1098 struct loop_func_table *xfer;
1099 kuid_t uid = current_uid();
1101 if (lo->lo_encrypt_key_size &&
1102 !uid_eq(lo->lo_key_owner, uid) &&
1103 !capable(CAP_SYS_ADMIN))
1105 if (lo->lo_state != Lo_bound)
1107 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1110 if (lo->lo_offset != info->lo_offset ||
1111 lo->lo_sizelimit != info->lo_sizelimit) {
1112 sync_blockdev(lo->lo_device);
1113 invalidate_bdev(lo->lo_device);
1116 /* I/O need to be drained during transfer transition */
1117 blk_mq_freeze_queue(lo->lo_queue);
1119 err = loop_release_xfer(lo);
1123 if (info->lo_encrypt_type) {
1124 unsigned int type = info->lo_encrypt_type;
1126 if (type >= MAX_LO_CRYPT) {
1130 xfer = xfer_funcs[type];
1138 err = loop_init_xfer(lo, xfer, info);
1142 if (lo->lo_offset != info->lo_offset ||
1143 lo->lo_sizelimit != info->lo_sizelimit) {
1144 /* kill_bdev should have truncated all the pages */
1145 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1147 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1148 __func__, lo->lo_number, lo->lo_file_name,
1149 lo->lo_device->bd_inode->i_mapping->nrpages);
1152 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1158 loop_config_discard(lo);
1160 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1161 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1162 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1163 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1167 lo->transfer = xfer->transfer;
1168 lo->ioctl = xfer->ioctl;
1170 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1171 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1172 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1174 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1175 lo->lo_init[0] = info->lo_init[0];
1176 lo->lo_init[1] = info->lo_init[1];
1177 if (info->lo_encrypt_key_size) {
1178 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1179 info->lo_encrypt_key_size);
1180 lo->lo_key_owner = uid;
1183 /* update dio if lo_offset or transfer is changed */
1184 __loop_update_dio(lo, lo->use_dio);
1187 blk_mq_unfreeze_queue(lo->lo_queue);
1189 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1190 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1191 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1192 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1193 loop_reread_partitions(lo, lo->lo_device);
1200 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1206 if (lo->lo_state != Lo_bound) {
1207 mutex_unlock(&lo->lo_ctl_mutex);
1211 memset(info, 0, sizeof(*info));
1212 info->lo_number = lo->lo_number;
1213 info->lo_offset = lo->lo_offset;
1214 info->lo_sizelimit = lo->lo_sizelimit;
1215 info->lo_flags = lo->lo_flags;
1216 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1217 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1218 info->lo_encrypt_type =
1219 lo->lo_encryption ? lo->lo_encryption->number : 0;
1220 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1221 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1222 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1223 lo->lo_encrypt_key_size);
1226 /* Drop lo_ctl_mutex while we call into the filesystem. */
1227 path = lo->lo_backing_file->f_path;
1229 mutex_unlock(&lo->lo_ctl_mutex);
1230 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1232 info->lo_device = huge_encode_dev(stat.dev);
1233 info->lo_inode = stat.ino;
1234 info->lo_rdevice = huge_encode_dev(stat.rdev);
1241 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1243 memset(info64, 0, sizeof(*info64));
1244 info64->lo_number = info->lo_number;
1245 info64->lo_device = info->lo_device;
1246 info64->lo_inode = info->lo_inode;
1247 info64->lo_rdevice = info->lo_rdevice;
1248 info64->lo_offset = info->lo_offset;
1249 info64->lo_sizelimit = 0;
1250 info64->lo_encrypt_type = info->lo_encrypt_type;
1251 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1252 info64->lo_flags = info->lo_flags;
1253 info64->lo_init[0] = info->lo_init[0];
1254 info64->lo_init[1] = info->lo_init[1];
1255 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1256 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1258 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1259 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1263 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1265 memset(info, 0, sizeof(*info));
1266 info->lo_number = info64->lo_number;
1267 info->lo_device = info64->lo_device;
1268 info->lo_inode = info64->lo_inode;
1269 info->lo_rdevice = info64->lo_rdevice;
1270 info->lo_offset = info64->lo_offset;
1271 info->lo_encrypt_type = info64->lo_encrypt_type;
1272 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1273 info->lo_flags = info64->lo_flags;
1274 info->lo_init[0] = info64->lo_init[0];
1275 info->lo_init[1] = info64->lo_init[1];
1276 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1277 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1279 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1280 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1282 /* error in case values were truncated */
1283 if (info->lo_device != info64->lo_device ||
1284 info->lo_rdevice != info64->lo_rdevice ||
1285 info->lo_inode != info64->lo_inode ||
1286 info->lo_offset != info64->lo_offset)
1293 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1295 struct loop_info info;
1296 struct loop_info64 info64;
1298 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1300 loop_info64_from_old(&info, &info64);
1301 return loop_set_status(lo, &info64);
1305 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1307 struct loop_info64 info64;
1309 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1311 return loop_set_status(lo, &info64);
1315 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1316 struct loop_info info;
1317 struct loop_info64 info64;
1321 mutex_unlock(&lo->lo_ctl_mutex);
1324 err = loop_get_status(lo, &info64);
1326 err = loop_info64_to_old(&info64, &info);
1327 if (!err && copy_to_user(arg, &info, sizeof(info)))
1334 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1335 struct loop_info64 info64;
1339 mutex_unlock(&lo->lo_ctl_mutex);
1342 err = loop_get_status(lo, &info64);
1343 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1349 static int loop_set_capacity(struct loop_device *lo)
1351 if (unlikely(lo->lo_state != Lo_bound))
1354 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1357 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1360 if (lo->lo_state != Lo_bound)
1363 __loop_update_dio(lo, !!arg);
1364 if (lo->use_dio == !!arg)
1371 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1375 if (lo->lo_state != Lo_bound)
1378 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1381 if (lo->lo_queue->limits.logical_block_size != arg) {
1382 sync_blockdev(lo->lo_device);
1383 invalidate_bdev(lo->lo_device);
1386 blk_mq_freeze_queue(lo->lo_queue);
1388 /* invalidate_bdev should have truncated all the pages */
1389 if (lo->lo_queue->limits.logical_block_size != arg &&
1390 lo->lo_device->bd_inode->i_mapping->nrpages) {
1392 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1393 __func__, lo->lo_number, lo->lo_file_name,
1394 lo->lo_device->bd_inode->i_mapping->nrpages);
1398 blk_queue_logical_block_size(lo->lo_queue, arg);
1399 blk_queue_physical_block_size(lo->lo_queue, arg);
1400 blk_queue_io_min(lo->lo_queue, arg);
1401 loop_update_dio(lo);
1403 blk_mq_unfreeze_queue(lo->lo_queue);
1408 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1409 unsigned int cmd, unsigned long arg)
1411 struct loop_device *lo = bdev->bd_disk->private_data;
1414 mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1417 err = loop_set_fd(lo, mode, bdev, arg);
1419 case LOOP_CHANGE_FD:
1420 err = loop_change_fd(lo, bdev, arg);
1423 /* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1424 err = loop_clr_fd(lo);
1428 case LOOP_SET_STATUS:
1430 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1431 err = loop_set_status_old(lo,
1432 (struct loop_info __user *)arg);
1434 case LOOP_GET_STATUS:
1435 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1436 /* loop_get_status() unlocks lo_ctl_mutex */
1438 case LOOP_SET_STATUS64:
1440 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1441 err = loop_set_status64(lo,
1442 (struct loop_info64 __user *) arg);
1444 case LOOP_GET_STATUS64:
1445 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1446 /* loop_get_status() unlocks lo_ctl_mutex */
1448 case LOOP_SET_CAPACITY:
1450 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1451 err = loop_set_capacity(lo);
1453 case LOOP_SET_DIRECT_IO:
1455 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1456 err = loop_set_dio(lo, arg);
1458 case LOOP_SET_BLOCK_SIZE:
1460 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1461 err = loop_set_block_size(lo, arg);
1464 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1466 mutex_unlock(&lo->lo_ctl_mutex);
1472 #ifdef CONFIG_COMPAT
1473 struct compat_loop_info {
1474 compat_int_t lo_number; /* ioctl r/o */
1475 compat_dev_t lo_device; /* ioctl r/o */
1476 compat_ulong_t lo_inode; /* ioctl r/o */
1477 compat_dev_t lo_rdevice; /* ioctl r/o */
1478 compat_int_t lo_offset;
1479 compat_int_t lo_encrypt_type;
1480 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1481 compat_int_t lo_flags; /* ioctl r/o */
1482 char lo_name[LO_NAME_SIZE];
1483 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1484 compat_ulong_t lo_init[2];
1489 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1490 * - noinlined to reduce stack space usage in main part of driver
1493 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1494 struct loop_info64 *info64)
1496 struct compat_loop_info info;
1498 if (copy_from_user(&info, arg, sizeof(info)))
1501 memset(info64, 0, sizeof(*info64));
1502 info64->lo_number = info.lo_number;
1503 info64->lo_device = info.lo_device;
1504 info64->lo_inode = info.lo_inode;
1505 info64->lo_rdevice = info.lo_rdevice;
1506 info64->lo_offset = info.lo_offset;
1507 info64->lo_sizelimit = 0;
1508 info64->lo_encrypt_type = info.lo_encrypt_type;
1509 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1510 info64->lo_flags = info.lo_flags;
1511 info64->lo_init[0] = info.lo_init[0];
1512 info64->lo_init[1] = info.lo_init[1];
1513 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1514 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1516 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1517 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1522 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1523 * - noinlined to reduce stack space usage in main part of driver
1526 loop_info64_to_compat(const struct loop_info64 *info64,
1527 struct compat_loop_info __user *arg)
1529 struct compat_loop_info info;
1531 memset(&info, 0, sizeof(info));
1532 info.lo_number = info64->lo_number;
1533 info.lo_device = info64->lo_device;
1534 info.lo_inode = info64->lo_inode;
1535 info.lo_rdevice = info64->lo_rdevice;
1536 info.lo_offset = info64->lo_offset;
1537 info.lo_encrypt_type = info64->lo_encrypt_type;
1538 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1539 info.lo_flags = info64->lo_flags;
1540 info.lo_init[0] = info64->lo_init[0];
1541 info.lo_init[1] = info64->lo_init[1];
1542 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1543 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1545 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1546 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1548 /* error in case values were truncated */
1549 if (info.lo_device != info64->lo_device ||
1550 info.lo_rdevice != info64->lo_rdevice ||
1551 info.lo_inode != info64->lo_inode ||
1552 info.lo_offset != info64->lo_offset ||
1553 info.lo_init[0] != info64->lo_init[0] ||
1554 info.lo_init[1] != info64->lo_init[1])
1557 if (copy_to_user(arg, &info, sizeof(info)))
1563 loop_set_status_compat(struct loop_device *lo,
1564 const struct compat_loop_info __user *arg)
1566 struct loop_info64 info64;
1569 ret = loop_info64_from_compat(arg, &info64);
1572 return loop_set_status(lo, &info64);
1576 loop_get_status_compat(struct loop_device *lo,
1577 struct compat_loop_info __user *arg)
1579 struct loop_info64 info64;
1583 mutex_unlock(&lo->lo_ctl_mutex);
1586 err = loop_get_status(lo, &info64);
1588 err = loop_info64_to_compat(&info64, arg);
1592 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1593 unsigned int cmd, unsigned long arg)
1595 struct loop_device *lo = bdev->bd_disk->private_data;
1599 case LOOP_SET_STATUS:
1600 mutex_lock(&lo->lo_ctl_mutex);
1601 err = loop_set_status_compat(
1602 lo, (const struct compat_loop_info __user *) arg);
1603 mutex_unlock(&lo->lo_ctl_mutex);
1605 case LOOP_GET_STATUS:
1606 mutex_lock(&lo->lo_ctl_mutex);
1607 err = loop_get_status_compat(
1608 lo, (struct compat_loop_info __user *) arg);
1609 /* loop_get_status() unlocks lo_ctl_mutex */
1611 case LOOP_SET_CAPACITY:
1613 case LOOP_GET_STATUS64:
1614 case LOOP_SET_STATUS64:
1615 arg = (unsigned long) compat_ptr(arg);
1617 case LOOP_CHANGE_FD:
1618 case LOOP_SET_DIRECT_IO:
1619 err = lo_ioctl(bdev, mode, cmd, arg);
1629 static int lo_open(struct block_device *bdev, fmode_t mode)
1631 struct loop_device *lo;
1634 mutex_lock(&loop_index_mutex);
1635 lo = bdev->bd_disk->private_data;
1641 atomic_inc(&lo->lo_refcnt);
1643 mutex_unlock(&loop_index_mutex);
1647 static void __lo_release(struct loop_device *lo)
1651 if (atomic_dec_return(&lo->lo_refcnt))
1654 mutex_lock(&lo->lo_ctl_mutex);
1655 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1657 * In autoclear mode, stop the loop thread
1658 * and remove configuration after last close.
1660 err = loop_clr_fd(lo);
1663 } else if (lo->lo_state == Lo_bound) {
1665 * Otherwise keep thread (if running) and config,
1666 * but flush possible ongoing bios in thread.
1668 blk_mq_freeze_queue(lo->lo_queue);
1669 blk_mq_unfreeze_queue(lo->lo_queue);
1672 mutex_unlock(&lo->lo_ctl_mutex);
1675 static void lo_release(struct gendisk *disk, fmode_t mode)
1677 mutex_lock(&loop_index_mutex);
1678 __lo_release(disk->private_data);
1679 mutex_unlock(&loop_index_mutex);
1682 static const struct block_device_operations lo_fops = {
1683 .owner = THIS_MODULE,
1685 .release = lo_release,
1687 #ifdef CONFIG_COMPAT
1688 .compat_ioctl = lo_compat_ioctl,
1693 * And now the modules code and kernel interface.
1695 static int max_loop;
1696 module_param(max_loop, int, S_IRUGO);
1697 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1698 module_param(max_part, int, S_IRUGO);
1699 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1700 MODULE_LICENSE("GPL");
1701 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1703 int loop_register_transfer(struct loop_func_table *funcs)
1705 unsigned int n = funcs->number;
1707 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1709 xfer_funcs[n] = funcs;
1713 static int unregister_transfer_cb(int id, void *ptr, void *data)
1715 struct loop_device *lo = ptr;
1716 struct loop_func_table *xfer = data;
1718 mutex_lock(&lo->lo_ctl_mutex);
1719 if (lo->lo_encryption == xfer)
1720 loop_release_xfer(lo);
1721 mutex_unlock(&lo->lo_ctl_mutex);
1725 int loop_unregister_transfer(int number)
1727 unsigned int n = number;
1728 struct loop_func_table *xfer;
1730 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1733 xfer_funcs[n] = NULL;
1734 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1738 EXPORT_SYMBOL(loop_register_transfer);
1739 EXPORT_SYMBOL(loop_unregister_transfer);
1741 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1742 const struct blk_mq_queue_data *bd)
1744 struct loop_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
1745 struct loop_device *lo = cmd->rq->q->queuedata;
1747 blk_mq_start_request(bd->rq);
1749 if (lo->lo_state != Lo_bound)
1750 return BLK_STS_IOERR;
1752 switch (req_op(cmd->rq)) {
1754 case REQ_OP_DISCARD:
1755 case REQ_OP_WRITE_ZEROES:
1756 cmd->use_aio = false;
1759 cmd->use_aio = lo->use_dio;
1763 kthread_queue_work(&lo->worker, &cmd->work);
1768 static void loop_handle_cmd(struct loop_cmd *cmd)
1770 const bool write = op_is_write(req_op(cmd->rq));
1771 struct loop_device *lo = cmd->rq->q->queuedata;
1774 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1779 ret = do_req_filebacked(lo, cmd->rq);
1781 /* complete non-aio request */
1782 if (!cmd->use_aio || ret) {
1783 cmd->ret = ret ? -EIO : 0;
1784 blk_mq_complete_request(cmd->rq);
1788 static void loop_queue_work(struct kthread_work *work)
1790 struct loop_cmd *cmd =
1791 container_of(work, struct loop_cmd, work);
1793 loop_handle_cmd(cmd);
1796 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1797 unsigned int hctx_idx, unsigned int numa_node)
1799 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1802 kthread_init_work(&cmd->work, loop_queue_work);
1807 static const struct blk_mq_ops loop_mq_ops = {
1808 .queue_rq = loop_queue_rq,
1809 .init_request = loop_init_request,
1810 .complete = lo_complete_rq,
1813 static int loop_add(struct loop_device **l, int i)
1815 struct loop_device *lo;
1816 struct gendisk *disk;
1820 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1824 lo->lo_state = Lo_unbound;
1826 /* allocate id, if @id >= 0, we're requesting that specific id */
1828 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1832 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1839 lo->tag_set.ops = &loop_mq_ops;
1840 lo->tag_set.nr_hw_queues = 1;
1841 lo->tag_set.queue_depth = 128;
1842 lo->tag_set.numa_node = NUMA_NO_NODE;
1843 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1844 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
1845 lo->tag_set.driver_data = lo;
1847 err = blk_mq_alloc_tag_set(&lo->tag_set);
1851 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
1852 if (IS_ERR_OR_NULL(lo->lo_queue)) {
1853 err = PTR_ERR(lo->lo_queue);
1854 goto out_cleanup_tags;
1856 lo->lo_queue->queuedata = lo;
1858 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1861 * By default, we do buffer IO, so it doesn't make sense to enable
1862 * merge because the I/O submitted to backing file is handled page by
1863 * page. For directio mode, merge does help to dispatch bigger request
1864 * to underlayer disk. We will enable merge once directio is enabled.
1866 queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1869 disk = lo->lo_disk = alloc_disk(1 << part_shift);
1871 goto out_free_queue;
1874 * Disable partition scanning by default. The in-kernel partition
1875 * scanning can be requested individually per-device during its
1876 * setup. Userspace can always add and remove partitions from all
1877 * devices. The needed partition minors are allocated from the
1878 * extended minor space, the main loop device numbers will continue
1879 * to match the loop minors, regardless of the number of partitions
1882 * If max_part is given, partition scanning is globally enabled for
1883 * all loop devices. The minors for the main loop devices will be
1884 * multiples of max_part.
1886 * Note: Global-for-all-devices, set-only-at-init, read-only module
1887 * parameteters like 'max_loop' and 'max_part' make things needlessly
1888 * complicated, are too static, inflexible and may surprise
1889 * userspace tools. Parameters like this in general should be avoided.
1892 disk->flags |= GENHD_FL_NO_PART_SCAN;
1893 disk->flags |= GENHD_FL_EXT_DEVT;
1894 mutex_init(&lo->lo_ctl_mutex);
1895 atomic_set(&lo->lo_refcnt, 0);
1897 spin_lock_init(&lo->lo_lock);
1898 disk->major = LOOP_MAJOR;
1899 disk->first_minor = i << part_shift;
1900 disk->fops = &lo_fops;
1901 disk->private_data = lo;
1902 disk->queue = lo->lo_queue;
1903 sprintf(disk->disk_name, "loop%d", i);
1906 return lo->lo_number;
1909 blk_cleanup_queue(lo->lo_queue);
1911 blk_mq_free_tag_set(&lo->tag_set);
1913 idr_remove(&loop_index_idr, i);
1920 static void loop_remove(struct loop_device *lo)
1922 blk_cleanup_queue(lo->lo_queue);
1923 del_gendisk(lo->lo_disk);
1924 blk_mq_free_tag_set(&lo->tag_set);
1925 put_disk(lo->lo_disk);
1929 static int find_free_cb(int id, void *ptr, void *data)
1931 struct loop_device *lo = ptr;
1932 struct loop_device **l = data;
1934 if (lo->lo_state == Lo_unbound) {
1941 static int loop_lookup(struct loop_device **l, int i)
1943 struct loop_device *lo;
1949 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1952 ret = lo->lo_number;
1957 /* lookup and return a specific i */
1958 lo = idr_find(&loop_index_idr, i);
1961 ret = lo->lo_number;
1967 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1969 struct loop_device *lo;
1970 struct kobject *kobj;
1973 mutex_lock(&loop_index_mutex);
1974 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1976 err = loop_add(&lo, MINOR(dev) >> part_shift);
1980 kobj = get_disk(lo->lo_disk);
1981 mutex_unlock(&loop_index_mutex);
1987 static long loop_control_ioctl(struct file *file, unsigned int cmd,
1990 struct loop_device *lo;
1993 mutex_lock(&loop_index_mutex);
1996 ret = loop_lookup(&lo, parm);
2001 ret = loop_add(&lo, parm);
2003 case LOOP_CTL_REMOVE:
2004 ret = loop_lookup(&lo, parm);
2007 mutex_lock(&lo->lo_ctl_mutex);
2008 if (lo->lo_state != Lo_unbound) {
2010 mutex_unlock(&lo->lo_ctl_mutex);
2013 if (atomic_read(&lo->lo_refcnt) > 0) {
2015 mutex_unlock(&lo->lo_ctl_mutex);
2018 lo->lo_disk->private_data = NULL;
2019 mutex_unlock(&lo->lo_ctl_mutex);
2020 idr_remove(&loop_index_idr, lo->lo_number);
2023 case LOOP_CTL_GET_FREE:
2024 ret = loop_lookup(&lo, -1);
2027 ret = loop_add(&lo, -1);
2029 mutex_unlock(&loop_index_mutex);
2034 static const struct file_operations loop_ctl_fops = {
2035 .open = nonseekable_open,
2036 .unlocked_ioctl = loop_control_ioctl,
2037 .compat_ioctl = loop_control_ioctl,
2038 .owner = THIS_MODULE,
2039 .llseek = noop_llseek,
2042 static struct miscdevice loop_misc = {
2043 .minor = LOOP_CTRL_MINOR,
2044 .name = "loop-control",
2045 .fops = &loop_ctl_fops,
2048 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2049 MODULE_ALIAS("devname:loop-control");
2051 static int __init loop_init(void)
2054 unsigned long range;
2055 struct loop_device *lo;
2060 part_shift = fls(max_part);
2063 * Adjust max_part according to part_shift as it is exported
2064 * to user space so that user can decide correct minor number
2065 * if [s]he want to create more devices.
2067 * Note that -1 is required because partition 0 is reserved
2068 * for the whole disk.
2070 max_part = (1UL << part_shift) - 1;
2073 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2078 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2084 * If max_loop is specified, create that many devices upfront.
2085 * This also becomes a hard limit. If max_loop is not specified,
2086 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2087 * init time. Loop devices can be requested on-demand with the
2088 * /dev/loop-control interface, or be instantiated by accessing
2089 * a 'dead' device node.
2093 range = max_loop << part_shift;
2095 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2096 range = 1UL << MINORBITS;
2099 err = misc_register(&loop_misc);
2104 if (register_blkdev(LOOP_MAJOR, "loop")) {
2109 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2110 THIS_MODULE, loop_probe, NULL, NULL);
2112 /* pre-create number of devices given by config or max_loop */
2113 mutex_lock(&loop_index_mutex);
2114 for (i = 0; i < nr; i++)
2116 mutex_unlock(&loop_index_mutex);
2118 printk(KERN_INFO "loop: module loaded\n");
2122 misc_deregister(&loop_misc);
2127 static int loop_exit_cb(int id, void *ptr, void *data)
2129 struct loop_device *lo = ptr;
2135 static void __exit loop_exit(void)
2137 unsigned long range;
2139 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2141 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2142 idr_destroy(&loop_index_idr);
2144 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2145 unregister_blkdev(LOOP_MAJOR, "loop");
2147 misc_deregister(&loop_misc);
2150 module_init(loop_init);
2151 module_exit(loop_exit);
2154 static int __init max_loop_setup(char *str)
2156 max_loop = simple_strtol(str, NULL, 0);
2160 __setup("max_loop=", max_loop_setup);