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>
79 #include <linux/ioprio.h>
80 #include <linux/blk-cgroup.h>
84 #include <linux/uaccess.h>
86 static DEFINE_IDR(loop_index_idr);
87 static DEFINE_MUTEX(loop_ctl_mutex);
90 static int part_shift;
92 static int transfer_xor(struct loop_device *lo, int cmd,
93 struct page *raw_page, unsigned raw_off,
94 struct page *loop_page, unsigned loop_off,
95 int size, sector_t real_block)
97 char *raw_buf = kmap_atomic(raw_page) + raw_off;
98 char *loop_buf = kmap_atomic(loop_page) + loop_off;
110 key = lo->lo_encrypt_key;
111 keysize = lo->lo_encrypt_key_size;
112 for (i = 0; i < size; i++)
113 *out++ = *in++ ^ key[(i & 511) % keysize];
115 kunmap_atomic(loop_buf);
116 kunmap_atomic(raw_buf);
121 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
123 if (unlikely(info->lo_encrypt_key_size <= 0))
128 static struct loop_func_table none_funcs = {
129 .number = LO_CRYPT_NONE,
132 static struct loop_func_table xor_funcs = {
133 .number = LO_CRYPT_XOR,
134 .transfer = transfer_xor,
138 /* xfer_funcs[0] is special - its release function is never called */
139 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
144 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
148 /* Compute loopsize in bytes */
149 loopsize = i_size_read(file->f_mapping->host);
152 /* offset is beyond i_size, weird but possible */
156 if (sizelimit > 0 && sizelimit < loopsize)
157 loopsize = sizelimit;
159 * Unfortunately, if we want to do I/O on the device,
160 * the number of 512-byte sectors has to fit into a sector_t.
162 return loopsize >> 9;
165 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
167 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
170 static void __loop_update_dio(struct loop_device *lo, bool dio)
172 struct file *file = lo->lo_backing_file;
173 struct address_space *mapping = file->f_mapping;
174 struct inode *inode = mapping->host;
175 unsigned short sb_bsize = 0;
176 unsigned dio_align = 0;
179 if (inode->i_sb->s_bdev) {
180 sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
181 dio_align = sb_bsize - 1;
185 * We support direct I/O only if lo_offset is aligned with the
186 * logical I/O size of backing device, and the logical block
187 * size of loop is bigger than the backing device's and the loop
188 * needn't transform transfer.
190 * TODO: the above condition may be loosed in the future, and
191 * direct I/O may be switched runtime at that time because most
192 * of requests in sane applications should be PAGE_SIZE aligned
195 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
196 !(lo->lo_offset & dio_align) &&
197 mapping->a_ops->direct_IO &&
206 if (lo->use_dio == use_dio)
209 /* flush dirty pages before changing direct IO */
213 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
214 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
215 * will get updated by ioctl(LOOP_GET_STATUS)
217 if (lo->lo_state == Lo_bound)
218 blk_mq_freeze_queue(lo->lo_queue);
219 lo->use_dio = use_dio;
221 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
222 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
224 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
225 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
227 if (lo->lo_state == Lo_bound)
228 blk_mq_unfreeze_queue(lo->lo_queue);
232 figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
234 loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
235 sector_t x = (sector_t)size;
236 struct block_device *bdev = lo->lo_device;
238 if (unlikely((loff_t)x != size))
240 if (lo->lo_offset != offset)
241 lo->lo_offset = offset;
242 if (lo->lo_sizelimit != sizelimit)
243 lo->lo_sizelimit = sizelimit;
244 set_capacity(lo->lo_disk, x);
245 bd_set_size(bdev, (loff_t)get_capacity(bdev->bd_disk) << 9);
246 /* let user-space know about the new size */
247 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
252 lo_do_transfer(struct loop_device *lo, int cmd,
253 struct page *rpage, unsigned roffs,
254 struct page *lpage, unsigned loffs,
255 int size, sector_t rblock)
259 ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
263 printk_ratelimited(KERN_ERR
264 "loop: Transfer error at byte offset %llu, length %i.\n",
265 (unsigned long long)rblock << 9, size);
269 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
274 iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
276 file_start_write(file);
277 bw = vfs_iter_write(file, &i, ppos, 0);
278 file_end_write(file);
280 if (likely(bw == bvec->bv_len))
283 printk_ratelimited(KERN_ERR
284 "loop: Write error at byte offset %llu, length %i.\n",
285 (unsigned long long)*ppos, bvec->bv_len);
291 static int lo_write_simple(struct loop_device *lo, struct request *rq,
295 struct req_iterator iter;
298 rq_for_each_segment(bvec, rq, iter) {
299 ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
309 * This is the slow, transforming version that needs to double buffer the
310 * data as it cannot do the transformations in place without having direct
311 * access to the destination pages of the backing file.
313 static int lo_write_transfer(struct loop_device *lo, struct request *rq,
316 struct bio_vec bvec, b;
317 struct req_iterator iter;
321 page = alloc_page(GFP_NOIO);
325 rq_for_each_segment(bvec, rq, iter) {
326 ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page,
327 bvec.bv_offset, bvec.bv_len, pos >> 9);
333 b.bv_len = bvec.bv_len;
334 ret = lo_write_bvec(lo->lo_backing_file, &b, &pos);
343 static int lo_read_simple(struct loop_device *lo, struct request *rq,
347 struct req_iterator iter;
351 rq_for_each_segment(bvec, rq, iter) {
352 iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
353 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
357 flush_dcache_page(bvec.bv_page);
359 if (len != bvec.bv_len) {
362 __rq_for_each_bio(bio, rq)
372 static int lo_read_transfer(struct loop_device *lo, struct request *rq,
375 struct bio_vec bvec, b;
376 struct req_iterator iter;
382 page = alloc_page(GFP_NOIO);
386 rq_for_each_segment(bvec, rq, iter) {
391 b.bv_len = bvec.bv_len;
393 iov_iter_bvec(&i, READ, &b, 1, b.bv_len);
394 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
400 ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page,
401 bvec.bv_offset, len, offset >> 9);
405 flush_dcache_page(bvec.bv_page);
407 if (len != bvec.bv_len) {
410 __rq_for_each_bio(bio, rq)
422 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
426 * We use fallocate to manipulate the space mappings used by the image
427 * a.k.a. discard/zerorange. However we do not support this if
428 * encryption is enabled, because it may give an attacker useful
431 struct file *file = lo->lo_backing_file;
432 struct request_queue *q = lo->lo_queue;
435 mode |= FALLOC_FL_KEEP_SIZE;
437 if (!blk_queue_discard(q)) {
442 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
443 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
449 static int lo_req_flush(struct loop_device *lo, struct request *rq)
451 struct file *file = lo->lo_backing_file;
452 int ret = vfs_fsync(file, 0);
453 if (unlikely(ret && ret != -EINVAL))
459 static void lo_complete_rq(struct request *rq)
461 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
462 blk_status_t ret = BLK_STS_OK;
464 if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
465 req_op(rq) != REQ_OP_READ) {
467 ret = errno_to_blk_status(cmd->ret);
472 * Short READ - if we got some data, advance our request and
473 * retry it. If we got no data, end the rest with EIO.
476 blk_update_request(rq, BLK_STS_OK, cmd->ret);
478 blk_mq_requeue_request(rq, true);
481 struct bio *bio = rq->bio;
490 blk_mq_end_request(rq, ret);
494 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
496 struct request *rq = blk_mq_rq_from_pdu(cmd);
498 if (!atomic_dec_and_test(&cmd->ref))
502 blk_mq_complete_request(rq);
505 static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2)
507 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
512 lo_rw_aio_do_completion(cmd);
515 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
518 struct iov_iter iter;
519 struct req_iterator rq_iter;
520 struct bio_vec *bvec;
521 struct request *rq = blk_mq_rq_from_pdu(cmd);
522 struct bio *bio = rq->bio;
523 struct file *file = lo->lo_backing_file;
529 rq_for_each_bvec(tmp, rq, rq_iter)
532 if (rq->bio != rq->biotail) {
534 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
541 * The bios of the request may be started from the middle of
542 * the 'bvec' because of bio splitting, so we can't directly
543 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
544 * API will take care of all details for us.
546 rq_for_each_bvec(tmp, rq, rq_iter) {
554 * Same here, this bio may be started from the middle of the
555 * 'bvec' because of bio splitting, so offset from the bvec
556 * must be passed to iov iterator
558 offset = bio->bi_iter.bi_bvec_done;
559 bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
561 atomic_set(&cmd->ref, 2);
563 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
564 iter.iov_offset = offset;
566 cmd->iocb.ki_pos = pos;
567 cmd->iocb.ki_filp = file;
568 cmd->iocb.ki_complete = lo_rw_aio_complete;
569 cmd->iocb.ki_flags = IOCB_DIRECT;
570 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
572 kthread_associate_blkcg(cmd->css);
575 ret = call_write_iter(file, &cmd->iocb, &iter);
577 ret = call_read_iter(file, &cmd->iocb, &iter);
579 lo_rw_aio_do_completion(cmd);
580 kthread_associate_blkcg(NULL);
582 if (ret != -EIOCBQUEUED)
583 cmd->iocb.ki_complete(&cmd->iocb, ret, 0);
587 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
589 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
590 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
593 * lo_write_simple and lo_read_simple should have been covered
594 * by io submit style function like lo_rw_aio(), one blocker
595 * is that lo_read_simple() need to call flush_dcache_page after
596 * the page is written from kernel, and it isn't easy to handle
597 * this in io submit style function which submits all segments
598 * of the req at one time. And direct read IO doesn't need to
599 * run flush_dcache_page().
601 switch (req_op(rq)) {
603 return lo_req_flush(lo, rq);
604 case REQ_OP_WRITE_ZEROES:
606 * If the caller doesn't want deallocation, call zeroout to
607 * write zeroes the range. Otherwise, punch them out.
609 return lo_fallocate(lo, rq, pos,
610 (rq->cmd_flags & REQ_NOUNMAP) ?
611 FALLOC_FL_ZERO_RANGE :
612 FALLOC_FL_PUNCH_HOLE);
614 return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
617 return lo_write_transfer(lo, rq, pos);
618 else if (cmd->use_aio)
619 return lo_rw_aio(lo, cmd, pos, WRITE);
621 return lo_write_simple(lo, rq, pos);
624 return lo_read_transfer(lo, rq, pos);
625 else if (cmd->use_aio)
626 return lo_rw_aio(lo, cmd, pos, READ);
628 return lo_read_simple(lo, rq, pos);
635 static inline void loop_update_dio(struct loop_device *lo)
637 __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) |
641 static void loop_reread_partitions(struct loop_device *lo,
642 struct block_device *bdev)
646 mutex_lock(&bdev->bd_mutex);
647 rc = bdev_disk_changed(bdev, false);
648 mutex_unlock(&bdev->bd_mutex);
650 pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
651 __func__, lo->lo_number, lo->lo_file_name, rc);
654 static inline int is_loop_device(struct file *file)
656 struct inode *i = file->f_mapping->host;
658 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
661 static int loop_validate_file(struct file *file, struct block_device *bdev)
663 struct inode *inode = file->f_mapping->host;
664 struct file *f = file;
666 /* Avoid recursion */
667 while (is_loop_device(f)) {
668 struct loop_device *l;
670 if (f->f_mapping->host->i_bdev == bdev)
673 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
674 if (l->lo_state != Lo_bound) {
677 f = l->lo_backing_file;
679 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
685 * loop_change_fd switched the backing store of a loopback device to
686 * a new file. This is useful for operating system installers to free up
687 * the original file and in High Availability environments to switch to
688 * an alternative location for the content in case of server meltdown.
689 * This can only work if the loop device is used read-only, and if the
690 * new backing store is the same size and type as the old backing store.
692 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
695 struct file *file = NULL, *old_file;
699 error = mutex_lock_killable(&loop_ctl_mutex);
703 if (lo->lo_state != Lo_bound)
706 /* the loop device has to be read-only */
708 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
716 error = loop_validate_file(file, bdev);
720 old_file = lo->lo_backing_file;
724 /* size of the new backing store needs to be the same */
725 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
729 blk_mq_freeze_queue(lo->lo_queue);
730 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
731 lo->lo_backing_file = file;
732 lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
733 mapping_set_gfp_mask(file->f_mapping,
734 lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
736 blk_mq_unfreeze_queue(lo->lo_queue);
737 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
738 mutex_unlock(&loop_ctl_mutex);
740 * We must drop file reference outside of loop_ctl_mutex as dropping
741 * the file ref can take bd_mutex which creates circular locking
746 loop_reread_partitions(lo, bdev);
750 mutex_unlock(&loop_ctl_mutex);
756 /* loop sysfs attributes */
758 static ssize_t loop_attr_show(struct device *dev, char *page,
759 ssize_t (*callback)(struct loop_device *, char *))
761 struct gendisk *disk = dev_to_disk(dev);
762 struct loop_device *lo = disk->private_data;
764 return callback(lo, page);
767 #define LOOP_ATTR_RO(_name) \
768 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
769 static ssize_t loop_attr_do_show_##_name(struct device *d, \
770 struct device_attribute *attr, char *b) \
772 return loop_attr_show(d, b, loop_attr_##_name##_show); \
774 static struct device_attribute loop_attr_##_name = \
775 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
777 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
782 spin_lock_irq(&lo->lo_lock);
783 if (lo->lo_backing_file)
784 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
785 spin_unlock_irq(&lo->lo_lock);
787 if (IS_ERR_OR_NULL(p))
791 memmove(buf, p, ret);
799 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
801 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
804 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
806 return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
809 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
811 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
813 return sprintf(buf, "%s\n", autoclear ? "1" : "0");
816 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
818 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
820 return sprintf(buf, "%s\n", partscan ? "1" : "0");
823 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
825 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
827 return sprintf(buf, "%s\n", dio ? "1" : "0");
830 LOOP_ATTR_RO(backing_file);
831 LOOP_ATTR_RO(offset);
832 LOOP_ATTR_RO(sizelimit);
833 LOOP_ATTR_RO(autoclear);
834 LOOP_ATTR_RO(partscan);
837 static struct attribute *loop_attrs[] = {
838 &loop_attr_backing_file.attr,
839 &loop_attr_offset.attr,
840 &loop_attr_sizelimit.attr,
841 &loop_attr_autoclear.attr,
842 &loop_attr_partscan.attr,
847 static struct attribute_group loop_attribute_group = {
852 static void loop_sysfs_init(struct loop_device *lo)
854 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
855 &loop_attribute_group);
858 static void loop_sysfs_exit(struct loop_device *lo)
860 if (lo->sysfs_inited)
861 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
862 &loop_attribute_group);
865 static void loop_config_discard(struct loop_device *lo)
867 struct file *file = lo->lo_backing_file;
868 struct inode *inode = file->f_mapping->host;
869 struct request_queue *q = lo->lo_queue;
872 * If the backing device is a block device, mirror its zeroing
873 * capability. Set the discard sectors to the block device's zeroing
874 * capabilities because loop discards result in blkdev_issue_zeroout(),
875 * not blkdev_issue_discard(). This maintains consistent behavior with
876 * file-backed loop devices: discarded regions read back as zero.
878 if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
879 struct request_queue *backingq;
881 backingq = bdev_get_queue(inode->i_bdev);
882 blk_queue_max_discard_sectors(q,
883 backingq->limits.max_write_zeroes_sectors);
885 blk_queue_max_write_zeroes_sectors(q,
886 backingq->limits.max_write_zeroes_sectors);
889 * We use punch hole to reclaim the free space used by the
890 * image a.k.a. discard. However we do not support discard if
891 * encryption is enabled, because it may give an attacker
892 * useful information.
894 } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) {
895 q->limits.discard_granularity = 0;
896 q->limits.discard_alignment = 0;
897 blk_queue_max_discard_sectors(q, 0);
898 blk_queue_max_write_zeroes_sectors(q, 0);
901 q->limits.discard_granularity = inode->i_sb->s_blocksize;
902 q->limits.discard_alignment = 0;
904 blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
905 blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
908 if (q->limits.max_write_zeroes_sectors)
909 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
911 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
914 static void loop_unprepare_queue(struct loop_device *lo)
916 kthread_flush_worker(&lo->worker);
917 kthread_stop(lo->worker_task);
920 static int loop_kthread_worker_fn(void *worker_ptr)
922 current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO;
923 return kthread_worker_fn(worker_ptr);
926 static int loop_prepare_queue(struct loop_device *lo)
928 kthread_init_worker(&lo->worker);
929 lo->worker_task = kthread_run(loop_kthread_worker_fn,
930 &lo->worker, "loop%d", lo->lo_number);
931 if (IS_ERR(lo->worker_task))
933 set_user_nice(lo->worker_task, MIN_NICE);
937 static void loop_update_rotational(struct loop_device *lo)
939 struct file *file = lo->lo_backing_file;
940 struct inode *file_inode = file->f_mapping->host;
941 struct block_device *file_bdev = file_inode->i_sb->s_bdev;
942 struct request_queue *q = lo->lo_queue;
945 /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
947 nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev));
950 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
952 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
955 static int loop_set_fd(struct loop_device *lo, fmode_t mode,
956 struct block_device *bdev, unsigned int arg)
960 struct address_space *mapping;
961 struct block_device *claimed_bdev = NULL;
967 /* This is safe, since we have a reference from open(). */
968 __module_get(THIS_MODULE);
976 * If we don't hold exclusive handle for the device, upgrade to it
977 * here to avoid changing device under exclusive owner.
979 if (!(mode & FMODE_EXCL)) {
980 claimed_bdev = bd_start_claiming(bdev, loop_set_fd);
981 if (IS_ERR(claimed_bdev)) {
982 error = PTR_ERR(claimed_bdev);
987 error = mutex_lock_killable(&loop_ctl_mutex);
992 if (lo->lo_state != Lo_unbound)
995 error = loop_validate_file(file, bdev);
999 mapping = file->f_mapping;
1000 inode = mapping->host;
1002 if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1003 !file->f_op->write_iter)
1004 lo_flags |= LO_FLAGS_READ_ONLY;
1007 size = get_loop_size(lo, file);
1008 if ((loff_t)(sector_t)size != size)
1010 error = loop_prepare_queue(lo);
1016 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
1018 lo->use_dio = false;
1019 lo->lo_device = bdev;
1020 lo->lo_flags = lo_flags;
1021 lo->lo_backing_file = file;
1022 lo->transfer = NULL;
1024 lo->lo_sizelimit = 0;
1025 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1026 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1028 if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1029 blk_queue_write_cache(lo->lo_queue, true, false);
1031 if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) {
1032 /* In case of direct I/O, match underlying block size */
1033 unsigned short bsize = bdev_logical_block_size(
1034 inode->i_sb->s_bdev);
1036 blk_queue_logical_block_size(lo->lo_queue, bsize);
1037 blk_queue_physical_block_size(lo->lo_queue, bsize);
1038 blk_queue_io_min(lo->lo_queue, bsize);
1041 loop_update_rotational(lo);
1042 loop_update_dio(lo);
1043 set_capacity(lo->lo_disk, size);
1044 bd_set_size(bdev, size << 9);
1045 loop_sysfs_init(lo);
1046 /* let user-space know about the new size */
1047 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1049 set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
1050 block_size(inode->i_bdev) : PAGE_SIZE);
1052 lo->lo_state = Lo_bound;
1054 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1055 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1057 /* Grab the block_device to prevent its destruction after we
1058 * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev).
1061 mutex_unlock(&loop_ctl_mutex);
1063 loop_reread_partitions(lo, bdev);
1065 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1069 mutex_unlock(&loop_ctl_mutex);
1072 bd_abort_claiming(bdev, claimed_bdev, loop_set_fd);
1076 /* This is safe: open() is still holding a reference. */
1077 module_put(THIS_MODULE);
1082 loop_release_xfer(struct loop_device *lo)
1085 struct loop_func_table *xfer = lo->lo_encryption;
1089 err = xfer->release(lo);
1090 lo->transfer = NULL;
1091 lo->lo_encryption = NULL;
1092 module_put(xfer->owner);
1098 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
1099 const struct loop_info64 *i)
1104 struct module *owner = xfer->owner;
1106 if (!try_module_get(owner))
1109 err = xfer->init(lo, i);
1113 lo->lo_encryption = xfer;
1118 static int __loop_clr_fd(struct loop_device *lo, bool release)
1120 struct file *filp = NULL;
1121 gfp_t gfp = lo->old_gfp_mask;
1122 struct block_device *bdev = lo->lo_device;
1124 bool partscan = false;
1127 mutex_lock(&loop_ctl_mutex);
1128 if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) {
1133 filp = lo->lo_backing_file;
1139 /* freeze request queue during the transition */
1140 blk_mq_freeze_queue(lo->lo_queue);
1142 spin_lock_irq(&lo->lo_lock);
1143 lo->lo_backing_file = NULL;
1144 spin_unlock_irq(&lo->lo_lock);
1146 loop_release_xfer(lo);
1147 lo->transfer = NULL;
1149 lo->lo_device = NULL;
1150 lo->lo_encryption = NULL;
1152 lo->lo_sizelimit = 0;
1153 lo->lo_encrypt_key_size = 0;
1154 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1155 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1156 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1157 blk_queue_logical_block_size(lo->lo_queue, 512);
1158 blk_queue_physical_block_size(lo->lo_queue, 512);
1159 blk_queue_io_min(lo->lo_queue, 512);
1162 invalidate_bdev(bdev);
1163 bdev->bd_inode->i_mapping->wb_err = 0;
1165 set_capacity(lo->lo_disk, 0);
1166 loop_sysfs_exit(lo);
1168 bd_set_size(bdev, 0);
1169 /* let user-space know about this change */
1170 kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1172 mapping_set_gfp_mask(filp->f_mapping, gfp);
1173 /* This is safe: open() is still holding a reference. */
1174 module_put(THIS_MODULE);
1175 blk_mq_unfreeze_queue(lo->lo_queue);
1177 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev;
1178 lo_number = lo->lo_number;
1179 loop_unprepare_queue(lo);
1181 mutex_unlock(&loop_ctl_mutex);
1184 * bd_mutex has been held already in release path, so don't
1185 * acquire it if this function is called in such case.
1187 * If the reread partition isn't from release path, lo_refcnt
1188 * must be at least one and it can only become zero when the
1189 * current holder is released.
1192 mutex_lock(&bdev->bd_mutex);
1193 err = bdev_disk_changed(bdev, false);
1195 mutex_unlock(&bdev->bd_mutex);
1197 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1198 __func__, lo_number, err);
1199 /* Device is gone, no point in returning error */
1204 * lo->lo_state is set to Lo_unbound here after above partscan has
1207 * There cannot be anybody else entering __loop_clr_fd() as
1208 * lo->lo_backing_file is already cleared and Lo_rundown state
1209 * protects us from all the other places trying to change the 'lo'
1212 mutex_lock(&loop_ctl_mutex);
1215 lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1216 lo->lo_state = Lo_unbound;
1217 mutex_unlock(&loop_ctl_mutex);
1220 * Need not hold loop_ctl_mutex to fput backing file.
1221 * Calling fput holding loop_ctl_mutex triggers a circular
1222 * lock dependency possibility warning as fput can take
1223 * bd_mutex which is usually taken before loop_ctl_mutex.
1230 static int loop_clr_fd(struct loop_device *lo)
1234 err = mutex_lock_killable(&loop_ctl_mutex);
1237 if (lo->lo_state != Lo_bound) {
1238 mutex_unlock(&loop_ctl_mutex);
1242 * If we've explicitly asked to tear down the loop device,
1243 * and it has an elevated reference count, set it for auto-teardown when
1244 * the last reference goes away. This stops $!~#$@ udev from
1245 * preventing teardown because it decided that it needs to run blkid on
1246 * the loopback device whenever they appear. xfstests is notorious for
1247 * failing tests because blkid via udev races with a losetup
1248 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1249 * command to fail with EBUSY.
1251 if (atomic_read(&lo->lo_refcnt) > 1) {
1252 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1253 mutex_unlock(&loop_ctl_mutex);
1256 lo->lo_state = Lo_rundown;
1257 mutex_unlock(&loop_ctl_mutex);
1259 return __loop_clr_fd(lo, false);
1263 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1266 struct loop_func_table *xfer;
1267 kuid_t uid = current_uid();
1268 struct block_device *bdev;
1269 bool partscan = false;
1271 err = mutex_lock_killable(&loop_ctl_mutex);
1274 if (lo->lo_encrypt_key_size &&
1275 !uid_eq(lo->lo_key_owner, uid) &&
1276 !capable(CAP_SYS_ADMIN)) {
1280 if (lo->lo_state != Lo_bound) {
1284 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) {
1289 if (lo->lo_offset != info->lo_offset ||
1290 lo->lo_sizelimit != info->lo_sizelimit) {
1291 sync_blockdev(lo->lo_device);
1292 kill_bdev(lo->lo_device);
1295 /* I/O need to be drained during transfer transition */
1296 blk_mq_freeze_queue(lo->lo_queue);
1298 err = loop_release_xfer(lo);
1302 if (info->lo_encrypt_type) {
1303 unsigned int type = info->lo_encrypt_type;
1305 if (type >= MAX_LO_CRYPT) {
1309 xfer = xfer_funcs[type];
1317 err = loop_init_xfer(lo, xfer, info);
1321 if (lo->lo_offset != info->lo_offset ||
1322 lo->lo_sizelimit != info->lo_sizelimit) {
1323 /* kill_bdev should have truncated all the pages */
1324 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1326 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1327 __func__, lo->lo_number, lo->lo_file_name,
1328 lo->lo_device->bd_inode->i_mapping->nrpages);
1331 if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
1337 loop_config_discard(lo);
1339 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1340 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1341 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1342 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1346 lo->transfer = xfer->transfer;
1347 lo->ioctl = xfer->ioctl;
1349 if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1350 (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1351 lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1353 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1354 lo->lo_init[0] = info->lo_init[0];
1355 lo->lo_init[1] = info->lo_init[1];
1356 if (info->lo_encrypt_key_size) {
1357 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1358 info->lo_encrypt_key_size);
1359 lo->lo_key_owner = uid;
1362 /* update dio if lo_offset or transfer is changed */
1363 __loop_update_dio(lo, lo->use_dio);
1366 blk_mq_unfreeze_queue(lo->lo_queue);
1368 if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
1369 !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1370 lo->lo_flags |= LO_FLAGS_PARTSCAN;
1371 lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1372 bdev = lo->lo_device;
1376 mutex_unlock(&loop_ctl_mutex);
1378 loop_reread_partitions(lo, bdev);
1384 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1390 ret = mutex_lock_killable(&loop_ctl_mutex);
1393 if (lo->lo_state != Lo_bound) {
1394 mutex_unlock(&loop_ctl_mutex);
1398 memset(info, 0, sizeof(*info));
1399 info->lo_number = lo->lo_number;
1400 info->lo_offset = lo->lo_offset;
1401 info->lo_sizelimit = lo->lo_sizelimit;
1402 info->lo_flags = lo->lo_flags;
1403 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1404 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1405 info->lo_encrypt_type =
1406 lo->lo_encryption ? lo->lo_encryption->number : 0;
1407 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1408 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1409 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1410 lo->lo_encrypt_key_size);
1413 /* Drop loop_ctl_mutex while we call into the filesystem. */
1414 path = lo->lo_backing_file->f_path;
1416 mutex_unlock(&loop_ctl_mutex);
1417 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1419 info->lo_device = huge_encode_dev(stat.dev);
1420 info->lo_inode = stat.ino;
1421 info->lo_rdevice = huge_encode_dev(stat.rdev);
1428 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1430 memset(info64, 0, sizeof(*info64));
1431 info64->lo_number = info->lo_number;
1432 info64->lo_device = info->lo_device;
1433 info64->lo_inode = info->lo_inode;
1434 info64->lo_rdevice = info->lo_rdevice;
1435 info64->lo_offset = info->lo_offset;
1436 info64->lo_sizelimit = 0;
1437 info64->lo_encrypt_type = info->lo_encrypt_type;
1438 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1439 info64->lo_flags = info->lo_flags;
1440 info64->lo_init[0] = info->lo_init[0];
1441 info64->lo_init[1] = info->lo_init[1];
1442 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1443 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1445 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1446 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1450 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1452 memset(info, 0, sizeof(*info));
1453 info->lo_number = info64->lo_number;
1454 info->lo_device = info64->lo_device;
1455 info->lo_inode = info64->lo_inode;
1456 info->lo_rdevice = info64->lo_rdevice;
1457 info->lo_offset = info64->lo_offset;
1458 info->lo_encrypt_type = info64->lo_encrypt_type;
1459 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1460 info->lo_flags = info64->lo_flags;
1461 info->lo_init[0] = info64->lo_init[0];
1462 info->lo_init[1] = info64->lo_init[1];
1463 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1464 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1466 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1467 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1469 /* error in case values were truncated */
1470 if (info->lo_device != info64->lo_device ||
1471 info->lo_rdevice != info64->lo_rdevice ||
1472 info->lo_inode != info64->lo_inode ||
1473 info->lo_offset != info64->lo_offset)
1480 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1482 struct loop_info info;
1483 struct loop_info64 info64;
1485 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1487 loop_info64_from_old(&info, &info64);
1488 return loop_set_status(lo, &info64);
1492 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1494 struct loop_info64 info64;
1496 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1498 return loop_set_status(lo, &info64);
1502 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1503 struct loop_info info;
1504 struct loop_info64 info64;
1509 err = loop_get_status(lo, &info64);
1511 err = loop_info64_to_old(&info64, &info);
1512 if (!err && copy_to_user(arg, &info, sizeof(info)))
1519 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1520 struct loop_info64 info64;
1525 err = loop_get_status(lo, &info64);
1526 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1532 static int loop_set_capacity(struct loop_device *lo)
1534 if (unlikely(lo->lo_state != Lo_bound))
1537 return figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1540 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1543 if (lo->lo_state != Lo_bound)
1546 __loop_update_dio(lo, !!arg);
1547 if (lo->use_dio == !!arg)
1554 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1558 if (lo->lo_state != Lo_bound)
1561 if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
1564 if (lo->lo_queue->limits.logical_block_size == arg)
1567 sync_blockdev(lo->lo_device);
1568 kill_bdev(lo->lo_device);
1570 blk_mq_freeze_queue(lo->lo_queue);
1572 /* kill_bdev should have truncated all the pages */
1573 if (lo->lo_device->bd_inode->i_mapping->nrpages) {
1575 pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
1576 __func__, lo->lo_number, lo->lo_file_name,
1577 lo->lo_device->bd_inode->i_mapping->nrpages);
1581 blk_queue_logical_block_size(lo->lo_queue, arg);
1582 blk_queue_physical_block_size(lo->lo_queue, arg);
1583 blk_queue_io_min(lo->lo_queue, arg);
1584 loop_update_dio(lo);
1586 blk_mq_unfreeze_queue(lo->lo_queue);
1591 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1596 err = mutex_lock_killable(&loop_ctl_mutex);
1600 case LOOP_SET_CAPACITY:
1601 err = loop_set_capacity(lo);
1603 case LOOP_SET_DIRECT_IO:
1604 err = loop_set_dio(lo, arg);
1606 case LOOP_SET_BLOCK_SIZE:
1607 err = loop_set_block_size(lo, arg);
1610 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1612 mutex_unlock(&loop_ctl_mutex);
1616 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1617 unsigned int cmd, unsigned long arg)
1619 struct loop_device *lo = bdev->bd_disk->private_data;
1624 return loop_set_fd(lo, mode, bdev, arg);
1625 case LOOP_CHANGE_FD:
1626 return loop_change_fd(lo, bdev, arg);
1628 return loop_clr_fd(lo);
1629 case LOOP_SET_STATUS:
1631 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1632 err = loop_set_status_old(lo,
1633 (struct loop_info __user *)arg);
1636 case LOOP_GET_STATUS:
1637 return loop_get_status_old(lo, (struct loop_info __user *) arg);
1638 case LOOP_SET_STATUS64:
1640 if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1641 err = loop_set_status64(lo,
1642 (struct loop_info64 __user *) arg);
1645 case LOOP_GET_STATUS64:
1646 return loop_get_status64(lo, (struct loop_info64 __user *) arg);
1647 case LOOP_SET_CAPACITY:
1648 case LOOP_SET_DIRECT_IO:
1649 case LOOP_SET_BLOCK_SIZE:
1650 if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1654 err = lo_simple_ioctl(lo, cmd, arg);
1661 #ifdef CONFIG_COMPAT
1662 struct compat_loop_info {
1663 compat_int_t lo_number; /* ioctl r/o */
1664 compat_dev_t lo_device; /* ioctl r/o */
1665 compat_ulong_t lo_inode; /* ioctl r/o */
1666 compat_dev_t lo_rdevice; /* ioctl r/o */
1667 compat_int_t lo_offset;
1668 compat_int_t lo_encrypt_type;
1669 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1670 compat_int_t lo_flags; /* ioctl r/o */
1671 char lo_name[LO_NAME_SIZE];
1672 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1673 compat_ulong_t lo_init[2];
1678 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1679 * - noinlined to reduce stack space usage in main part of driver
1682 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1683 struct loop_info64 *info64)
1685 struct compat_loop_info info;
1687 if (copy_from_user(&info, arg, sizeof(info)))
1690 memset(info64, 0, sizeof(*info64));
1691 info64->lo_number = info.lo_number;
1692 info64->lo_device = info.lo_device;
1693 info64->lo_inode = info.lo_inode;
1694 info64->lo_rdevice = info.lo_rdevice;
1695 info64->lo_offset = info.lo_offset;
1696 info64->lo_sizelimit = 0;
1697 info64->lo_encrypt_type = info.lo_encrypt_type;
1698 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1699 info64->lo_flags = info.lo_flags;
1700 info64->lo_init[0] = info.lo_init[0];
1701 info64->lo_init[1] = info.lo_init[1];
1702 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1703 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1705 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1706 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1711 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1712 * - noinlined to reduce stack space usage in main part of driver
1715 loop_info64_to_compat(const struct loop_info64 *info64,
1716 struct compat_loop_info __user *arg)
1718 struct compat_loop_info info;
1720 memset(&info, 0, sizeof(info));
1721 info.lo_number = info64->lo_number;
1722 info.lo_device = info64->lo_device;
1723 info.lo_inode = info64->lo_inode;
1724 info.lo_rdevice = info64->lo_rdevice;
1725 info.lo_offset = info64->lo_offset;
1726 info.lo_encrypt_type = info64->lo_encrypt_type;
1727 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1728 info.lo_flags = info64->lo_flags;
1729 info.lo_init[0] = info64->lo_init[0];
1730 info.lo_init[1] = info64->lo_init[1];
1731 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1732 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1734 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1735 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1737 /* error in case values were truncated */
1738 if (info.lo_device != info64->lo_device ||
1739 info.lo_rdevice != info64->lo_rdevice ||
1740 info.lo_inode != info64->lo_inode ||
1741 info.lo_offset != info64->lo_offset ||
1742 info.lo_init[0] != info64->lo_init[0] ||
1743 info.lo_init[1] != info64->lo_init[1])
1746 if (copy_to_user(arg, &info, sizeof(info)))
1752 loop_set_status_compat(struct loop_device *lo,
1753 const struct compat_loop_info __user *arg)
1755 struct loop_info64 info64;
1758 ret = loop_info64_from_compat(arg, &info64);
1761 return loop_set_status(lo, &info64);
1765 loop_get_status_compat(struct loop_device *lo,
1766 struct compat_loop_info __user *arg)
1768 struct loop_info64 info64;
1773 err = loop_get_status(lo, &info64);
1775 err = loop_info64_to_compat(&info64, arg);
1779 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1780 unsigned int cmd, unsigned long arg)
1782 struct loop_device *lo = bdev->bd_disk->private_data;
1786 case LOOP_SET_STATUS:
1787 err = loop_set_status_compat(lo,
1788 (const struct compat_loop_info __user *)arg);
1790 case LOOP_GET_STATUS:
1791 err = loop_get_status_compat(lo,
1792 (struct compat_loop_info __user *)arg);
1794 case LOOP_SET_CAPACITY:
1796 case LOOP_GET_STATUS64:
1797 case LOOP_SET_STATUS64:
1798 arg = (unsigned long) compat_ptr(arg);
1801 case LOOP_CHANGE_FD:
1802 case LOOP_SET_BLOCK_SIZE:
1803 case LOOP_SET_DIRECT_IO:
1804 err = lo_ioctl(bdev, mode, cmd, arg);
1814 static int lo_open(struct block_device *bdev, fmode_t mode)
1816 struct loop_device *lo;
1819 err = mutex_lock_killable(&loop_ctl_mutex);
1822 lo = bdev->bd_disk->private_data;
1828 atomic_inc(&lo->lo_refcnt);
1830 mutex_unlock(&loop_ctl_mutex);
1834 static void lo_release(struct gendisk *disk, fmode_t mode)
1836 struct loop_device *lo;
1838 mutex_lock(&loop_ctl_mutex);
1839 lo = disk->private_data;
1840 if (atomic_dec_return(&lo->lo_refcnt))
1843 if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
1844 if (lo->lo_state != Lo_bound)
1846 lo->lo_state = Lo_rundown;
1847 mutex_unlock(&loop_ctl_mutex);
1849 * In autoclear mode, stop the loop thread
1850 * and remove configuration after last close.
1852 __loop_clr_fd(lo, true);
1854 } else if (lo->lo_state == Lo_bound) {
1856 * Otherwise keep thread (if running) and config,
1857 * but flush possible ongoing bios in thread.
1859 blk_mq_freeze_queue(lo->lo_queue);
1860 blk_mq_unfreeze_queue(lo->lo_queue);
1864 mutex_unlock(&loop_ctl_mutex);
1867 static const struct block_device_operations lo_fops = {
1868 .owner = THIS_MODULE,
1870 .release = lo_release,
1872 #ifdef CONFIG_COMPAT
1873 .compat_ioctl = lo_compat_ioctl,
1878 * And now the modules code and kernel interface.
1880 static int max_loop;
1881 module_param(max_loop, int, 0444);
1882 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1883 module_param(max_part, int, 0444);
1884 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1885 MODULE_LICENSE("GPL");
1886 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1888 int loop_register_transfer(struct loop_func_table *funcs)
1890 unsigned int n = funcs->number;
1892 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1894 xfer_funcs[n] = funcs;
1898 static int unregister_transfer_cb(int id, void *ptr, void *data)
1900 struct loop_device *lo = ptr;
1901 struct loop_func_table *xfer = data;
1903 mutex_lock(&loop_ctl_mutex);
1904 if (lo->lo_encryption == xfer)
1905 loop_release_xfer(lo);
1906 mutex_unlock(&loop_ctl_mutex);
1910 int loop_unregister_transfer(int number)
1912 unsigned int n = number;
1913 struct loop_func_table *xfer;
1915 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1918 xfer_funcs[n] = NULL;
1919 idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1923 EXPORT_SYMBOL(loop_register_transfer);
1924 EXPORT_SYMBOL(loop_unregister_transfer);
1926 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1927 const struct blk_mq_queue_data *bd)
1929 struct request *rq = bd->rq;
1930 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1931 struct loop_device *lo = rq->q->queuedata;
1933 blk_mq_start_request(rq);
1935 if (lo->lo_state != Lo_bound)
1936 return BLK_STS_IOERR;
1938 switch (req_op(rq)) {
1940 case REQ_OP_DISCARD:
1941 case REQ_OP_WRITE_ZEROES:
1942 cmd->use_aio = false;
1945 cmd->use_aio = lo->use_dio;
1949 /* always use the first bio's css */
1950 #ifdef CONFIG_BLK_CGROUP
1951 if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) {
1952 cmd->css = &bio_blkcg(rq->bio)->css;
1957 kthread_queue_work(&lo->worker, &cmd->work);
1962 static void loop_handle_cmd(struct loop_cmd *cmd)
1964 struct request *rq = blk_mq_rq_from_pdu(cmd);
1965 const bool write = op_is_write(req_op(rq));
1966 struct loop_device *lo = rq->q->queuedata;
1969 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1974 ret = do_req_filebacked(lo, rq);
1976 /* complete non-aio request */
1977 if (!cmd->use_aio || ret) {
1978 if (ret == -EOPNOTSUPP)
1981 cmd->ret = ret ? -EIO : 0;
1982 blk_mq_complete_request(rq);
1986 static void loop_queue_work(struct kthread_work *work)
1988 struct loop_cmd *cmd =
1989 container_of(work, struct loop_cmd, work);
1991 loop_handle_cmd(cmd);
1994 static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq,
1995 unsigned int hctx_idx, unsigned int numa_node)
1997 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1999 kthread_init_work(&cmd->work, loop_queue_work);
2003 static const struct blk_mq_ops loop_mq_ops = {
2004 .queue_rq = loop_queue_rq,
2005 .init_request = loop_init_request,
2006 .complete = lo_complete_rq,
2009 static int loop_add(struct loop_device **l, int i)
2011 struct loop_device *lo;
2012 struct gendisk *disk;
2016 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2020 lo->lo_state = Lo_unbound;
2022 /* allocate id, if @id >= 0, we're requesting that specific id */
2024 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2028 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2035 lo->tag_set.ops = &loop_mq_ops;
2036 lo->tag_set.nr_hw_queues = 1;
2037 lo->tag_set.queue_depth = 128;
2038 lo->tag_set.numa_node = NUMA_NO_NODE;
2039 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2040 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
2041 lo->tag_set.driver_data = lo;
2043 err = blk_mq_alloc_tag_set(&lo->tag_set);
2047 lo->lo_queue = blk_mq_init_queue(&lo->tag_set);
2048 if (IS_ERR(lo->lo_queue)) {
2049 err = PTR_ERR(lo->lo_queue);
2050 goto out_cleanup_tags;
2052 lo->lo_queue->queuedata = lo;
2054 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2057 * By default, we do buffer IO, so it doesn't make sense to enable
2058 * merge because the I/O submitted to backing file is handled page by
2059 * page. For directio mode, merge does help to dispatch bigger request
2060 * to underlayer disk. We will enable merge once directio is enabled.
2062 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2065 disk = lo->lo_disk = alloc_disk(1 << part_shift);
2067 goto out_free_queue;
2070 * Disable partition scanning by default. The in-kernel partition
2071 * scanning can be requested individually per-device during its
2072 * setup. Userspace can always add and remove partitions from all
2073 * devices. The needed partition minors are allocated from the
2074 * extended minor space, the main loop device numbers will continue
2075 * to match the loop minors, regardless of the number of partitions
2078 * If max_part is given, partition scanning is globally enabled for
2079 * all loop devices. The minors for the main loop devices will be
2080 * multiples of max_part.
2082 * Note: Global-for-all-devices, set-only-at-init, read-only module
2083 * parameteters like 'max_loop' and 'max_part' make things needlessly
2084 * complicated, are too static, inflexible and may surprise
2085 * userspace tools. Parameters like this in general should be avoided.
2088 disk->flags |= GENHD_FL_NO_PART_SCAN;
2089 disk->flags |= GENHD_FL_EXT_DEVT;
2090 atomic_set(&lo->lo_refcnt, 0);
2092 spin_lock_init(&lo->lo_lock);
2093 disk->major = LOOP_MAJOR;
2094 disk->first_minor = i << part_shift;
2095 disk->fops = &lo_fops;
2096 disk->private_data = lo;
2097 disk->queue = lo->lo_queue;
2098 sprintf(disk->disk_name, "loop%d", i);
2101 return lo->lo_number;
2104 blk_cleanup_queue(lo->lo_queue);
2106 blk_mq_free_tag_set(&lo->tag_set);
2108 idr_remove(&loop_index_idr, i);
2115 static void loop_remove(struct loop_device *lo)
2117 del_gendisk(lo->lo_disk);
2118 blk_cleanup_queue(lo->lo_queue);
2119 blk_mq_free_tag_set(&lo->tag_set);
2120 put_disk(lo->lo_disk);
2124 static int find_free_cb(int id, void *ptr, void *data)
2126 struct loop_device *lo = ptr;
2127 struct loop_device **l = data;
2129 if (lo->lo_state == Lo_unbound) {
2136 static int loop_lookup(struct loop_device **l, int i)
2138 struct loop_device *lo;
2144 err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
2147 ret = lo->lo_number;
2152 /* lookup and return a specific i */
2153 lo = idr_find(&loop_index_idr, i);
2156 ret = lo->lo_number;
2162 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
2164 struct loop_device *lo;
2165 struct kobject *kobj;
2168 mutex_lock(&loop_ctl_mutex);
2169 err = loop_lookup(&lo, MINOR(dev) >> part_shift);
2171 err = loop_add(&lo, MINOR(dev) >> part_shift);
2175 kobj = get_disk_and_module(lo->lo_disk);
2176 mutex_unlock(&loop_ctl_mutex);
2182 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2185 struct loop_device *lo;
2188 ret = mutex_lock_killable(&loop_ctl_mutex);
2195 ret = loop_lookup(&lo, parm);
2200 ret = loop_add(&lo, parm);
2202 case LOOP_CTL_REMOVE:
2203 ret = loop_lookup(&lo, parm);
2206 if (lo->lo_state != Lo_unbound) {
2210 if (atomic_read(&lo->lo_refcnt) > 0) {
2214 lo->lo_disk->private_data = NULL;
2215 idr_remove(&loop_index_idr, lo->lo_number);
2218 case LOOP_CTL_GET_FREE:
2219 ret = loop_lookup(&lo, -1);
2222 ret = loop_add(&lo, -1);
2224 mutex_unlock(&loop_ctl_mutex);
2229 static const struct file_operations loop_ctl_fops = {
2230 .open = nonseekable_open,
2231 .unlocked_ioctl = loop_control_ioctl,
2232 .compat_ioctl = loop_control_ioctl,
2233 .owner = THIS_MODULE,
2234 .llseek = noop_llseek,
2237 static struct miscdevice loop_misc = {
2238 .minor = LOOP_CTRL_MINOR,
2239 .name = "loop-control",
2240 .fops = &loop_ctl_fops,
2243 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2244 MODULE_ALIAS("devname:loop-control");
2246 static int __init loop_init(void)
2249 unsigned long range;
2250 struct loop_device *lo;
2255 part_shift = fls(max_part);
2258 * Adjust max_part according to part_shift as it is exported
2259 * to user space so that user can decide correct minor number
2260 * if [s]he want to create more devices.
2262 * Note that -1 is required because partition 0 is reserved
2263 * for the whole disk.
2265 max_part = (1UL << part_shift) - 1;
2268 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2273 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2279 * If max_loop is specified, create that many devices upfront.
2280 * This also becomes a hard limit. If max_loop is not specified,
2281 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2282 * init time. Loop devices can be requested on-demand with the
2283 * /dev/loop-control interface, or be instantiated by accessing
2284 * a 'dead' device node.
2288 range = max_loop << part_shift;
2290 nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2291 range = 1UL << MINORBITS;
2294 err = misc_register(&loop_misc);
2299 if (register_blkdev(LOOP_MAJOR, "loop")) {
2304 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
2305 THIS_MODULE, loop_probe, NULL, NULL);
2307 /* pre-create number of devices given by config or max_loop */
2308 mutex_lock(&loop_ctl_mutex);
2309 for (i = 0; i < nr; i++)
2311 mutex_unlock(&loop_ctl_mutex);
2313 printk(KERN_INFO "loop: module loaded\n");
2317 misc_deregister(&loop_misc);
2322 static int loop_exit_cb(int id, void *ptr, void *data)
2324 struct loop_device *lo = ptr;
2330 static void __exit loop_exit(void)
2332 unsigned long range;
2334 range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
2336 idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
2337 idr_destroy(&loop_index_idr);
2339 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
2340 unregister_blkdev(LOOP_MAJOR, "loop");
2342 misc_deregister(&loop_misc);
2345 module_init(loop_init);
2346 module_exit(loop_exit);
2349 static int __init max_loop_setup(char *str)
2351 max_loop = simple_strtol(str, NULL, 0);
2355 __setup("max_loop=", max_loop_setup);