2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
29 * ratelimit state to be used in DMXXX_LIMIT().
31 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
32 DEFAULT_RATELIMIT_INTERVAL,
33 DEFAULT_RATELIMIT_BURST);
34 EXPORT_SYMBOL(dm_ratelimit_state);
38 * Cookies are numeric values sent with CHANGE and REMOVE
39 * uevents while resuming, removing or renaming the device.
41 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
42 #define DM_COOKIE_LENGTH 24
44 static const char *_name = DM_NAME;
46 static unsigned int major = 0;
47 static unsigned int _major = 0;
49 static DEFINE_IDR(_minor_idr);
51 static DEFINE_SPINLOCK(_minor_lock);
54 * One of these is allocated per bio.
57 struct mapped_device *md;
61 unsigned long start_time;
62 spinlock_t endio_lock;
66 * For request-based dm.
67 * One of these is allocated per request.
69 struct dm_rq_target_io {
70 struct mapped_device *md;
72 struct request *orig, clone;
78 * For request-based dm - the bio clones we allocate are embedded in these
81 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
82 * the bioset is created - this means the bio has to come at the end of the
85 struct dm_rq_clone_bio_info {
87 struct dm_rq_target_io *tio;
91 union map_info *dm_get_mapinfo(struct bio *bio)
93 if (bio && bio->bi_private)
94 return &((struct dm_target_io *)bio->bi_private)->info;
98 union map_info *dm_get_rq_mapinfo(struct request *rq)
100 if (rq && rq->end_io_data)
101 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
104 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
106 #define MINOR_ALLOCED ((void *)-1)
109 * Bits for the md->flags field.
111 #define DMF_BLOCK_IO_FOR_SUSPEND 0
112 #define DMF_SUSPENDED 1
114 #define DMF_FREEING 3
115 #define DMF_DELETING 4
116 #define DMF_NOFLUSH_SUSPENDING 5
117 #define DMF_MERGE_IS_OPTIONAL 6
120 * Work processed by per-device workqueue.
122 struct mapped_device {
123 struct rw_semaphore io_lock;
124 struct mutex suspend_lock;
131 struct request_queue *queue;
133 /* Protect queue and type against concurrent access. */
134 struct mutex type_lock;
136 struct target_type *immutable_target_type;
138 struct gendisk *disk;
144 * A list of ios that arrived while we were suspended.
147 wait_queue_head_t wait;
148 struct work_struct work;
149 struct bio_list deferred;
150 spinlock_t deferred_lock;
153 * Processing queue (flush)
155 struct workqueue_struct *wq;
158 * The current mapping.
160 struct dm_table *map;
163 * io objects are allocated from here.
174 wait_queue_head_t eventq;
176 struct list_head uevent_list;
177 spinlock_t uevent_lock; /* Protect access to uevent_list */
180 * freeze/thaw support require holding onto a super block
182 struct super_block *frozen_sb;
183 struct block_device *bdev;
185 /* forced geometry settings */
186 struct hd_geometry geometry;
191 /* zero-length flush that will be cloned and submitted to targets */
192 struct bio flush_bio;
196 * For mempools pre-allocation at the table loading time.
198 struct dm_md_mempools {
205 static struct kmem_cache *_io_cache;
206 static struct kmem_cache *_rq_tio_cache;
209 * Unused now, and needs to be deleted. But since io_pool is overloaded and it's
210 * still used for _io_cache, I'm leaving this for a later cleanup
212 static struct kmem_cache *_rq_bio_info_cache;
214 static int __init local_init(void)
218 /* allocate a slab for the dm_ios */
219 _io_cache = KMEM_CACHE(dm_io, 0);
223 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
225 goto out_free_io_cache;
227 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
228 if (!_rq_bio_info_cache)
229 goto out_free_rq_tio_cache;
231 r = dm_uevent_init();
233 goto out_free_rq_bio_info_cache;
236 r = register_blkdev(_major, _name);
238 goto out_uevent_exit;
247 out_free_rq_bio_info_cache:
248 kmem_cache_destroy(_rq_bio_info_cache);
249 out_free_rq_tio_cache:
250 kmem_cache_destroy(_rq_tio_cache);
252 kmem_cache_destroy(_io_cache);
257 static void local_exit(void)
259 kmem_cache_destroy(_rq_bio_info_cache);
260 kmem_cache_destroy(_rq_tio_cache);
261 kmem_cache_destroy(_io_cache);
262 unregister_blkdev(_major, _name);
267 DMINFO("cleaned up");
270 static int (*_inits[])(void) __initdata = {
280 static void (*_exits[])(void) = {
290 static int __init dm_init(void)
292 const int count = ARRAY_SIZE(_inits);
296 for (i = 0; i < count; i++) {
311 static void __exit dm_exit(void)
313 int i = ARRAY_SIZE(_exits);
319 * Should be empty by this point.
321 idr_destroy(&_minor_idr);
325 * Block device functions
327 int dm_deleting_md(struct mapped_device *md)
329 return test_bit(DMF_DELETING, &md->flags);
332 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
334 struct mapped_device *md;
336 spin_lock(&_minor_lock);
338 md = bdev->bd_disk->private_data;
342 if (test_bit(DMF_FREEING, &md->flags) ||
343 dm_deleting_md(md)) {
349 atomic_inc(&md->open_count);
352 spin_unlock(&_minor_lock);
354 return md ? 0 : -ENXIO;
357 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
359 struct mapped_device *md = disk->private_data;
361 spin_lock(&_minor_lock);
363 atomic_dec(&md->open_count);
366 spin_unlock(&_minor_lock);
371 int dm_open_count(struct mapped_device *md)
373 return atomic_read(&md->open_count);
377 * Guarantees nothing is using the device before it's deleted.
379 int dm_lock_for_deletion(struct mapped_device *md)
383 spin_lock(&_minor_lock);
385 if (dm_open_count(md))
388 set_bit(DMF_DELETING, &md->flags);
390 spin_unlock(&_minor_lock);
395 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
397 struct mapped_device *md = bdev->bd_disk->private_data;
399 return dm_get_geometry(md, geo);
402 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
403 unsigned int cmd, unsigned long arg)
405 struct mapped_device *md = bdev->bd_disk->private_data;
406 struct dm_table *map = dm_get_live_table(md);
407 struct dm_target *tgt;
410 if (!map || !dm_table_get_size(map))
413 /* We only support devices that have a single target */
414 if (dm_table_get_num_targets(map) != 1)
417 tgt = dm_table_get_target(map, 0);
419 if (dm_suspended_md(md)) {
424 if (tgt->type->ioctl)
425 r = tgt->type->ioctl(tgt, cmd, arg);
433 static struct dm_io *alloc_io(struct mapped_device *md)
435 return mempool_alloc(md->io_pool, GFP_NOIO);
438 static void free_io(struct mapped_device *md, struct dm_io *io)
440 mempool_free(io, md->io_pool);
443 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
445 bio_put(&tio->clone);
448 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
451 return mempool_alloc(md->tio_pool, gfp_mask);
454 static void free_rq_tio(struct dm_rq_target_io *tio)
456 mempool_free(tio, tio->md->tio_pool);
459 static int md_in_flight(struct mapped_device *md)
461 return atomic_read(&md->pending[READ]) +
462 atomic_read(&md->pending[WRITE]);
465 static void start_io_acct(struct dm_io *io)
467 struct mapped_device *md = io->md;
469 int rw = bio_data_dir(io->bio);
471 io->start_time = jiffies;
473 cpu = part_stat_lock();
474 part_round_stats(cpu, &dm_disk(md)->part0);
476 atomic_set(&dm_disk(md)->part0.in_flight[rw],
477 atomic_inc_return(&md->pending[rw]));
480 static void end_io_acct(struct dm_io *io)
482 struct mapped_device *md = io->md;
483 struct bio *bio = io->bio;
484 unsigned long duration = jiffies - io->start_time;
486 int rw = bio_data_dir(bio);
488 cpu = part_stat_lock();
489 part_round_stats(cpu, &dm_disk(md)->part0);
490 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
494 * After this is decremented the bio must not be touched if it is
497 pending = atomic_dec_return(&md->pending[rw]);
498 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
499 pending += atomic_read(&md->pending[rw^0x1]);
501 /* nudge anyone waiting on suspend queue */
507 * Add the bio to the list of deferred io.
509 static void queue_io(struct mapped_device *md, struct bio *bio)
513 spin_lock_irqsave(&md->deferred_lock, flags);
514 bio_list_add(&md->deferred, bio);
515 spin_unlock_irqrestore(&md->deferred_lock, flags);
516 queue_work(md->wq, &md->work);
520 * Everyone (including functions in this file), should use this
521 * function to access the md->map field, and make sure they call
522 * dm_table_put() when finished.
524 struct dm_table *dm_get_live_table(struct mapped_device *md)
529 read_lock_irqsave(&md->map_lock, flags);
533 read_unlock_irqrestore(&md->map_lock, flags);
539 * Get the geometry associated with a dm device
541 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
549 * Set the geometry of a device.
551 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
553 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
555 if (geo->start > sz) {
556 DMWARN("Start sector is beyond the geometry limits.");
565 /*-----------------------------------------------------------------
567 * A more elegant soln is in the works that uses the queue
568 * merge fn, unfortunately there are a couple of changes to
569 * the block layer that I want to make for this. So in the
570 * interests of getting something for people to use I give
571 * you this clearly demarcated crap.
572 *---------------------------------------------------------------*/
574 static int __noflush_suspending(struct mapped_device *md)
576 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
580 * Decrements the number of outstanding ios that a bio has been
581 * cloned into, completing the original io if necc.
583 static void dec_pending(struct dm_io *io, int error)
588 struct mapped_device *md = io->md;
590 /* Push-back supersedes any I/O errors */
591 if (unlikely(error)) {
592 spin_lock_irqsave(&io->endio_lock, flags);
593 if (!(io->error > 0 && __noflush_suspending(md)))
595 spin_unlock_irqrestore(&io->endio_lock, flags);
598 if (atomic_dec_and_test(&io->io_count)) {
599 if (io->error == DM_ENDIO_REQUEUE) {
601 * Target requested pushing back the I/O.
603 spin_lock_irqsave(&md->deferred_lock, flags);
604 if (__noflush_suspending(md))
605 bio_list_add_head(&md->deferred, io->bio);
607 /* noflush suspend was interrupted. */
609 spin_unlock_irqrestore(&md->deferred_lock, flags);
612 io_error = io->error;
617 if (io_error == DM_ENDIO_REQUEUE)
620 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
622 * Preflush done for flush with data, reissue
625 bio->bi_rw &= ~REQ_FLUSH;
628 /* done with normal IO or empty flush */
629 trace_block_bio_complete(md->queue, bio, io_error);
630 bio_endio(bio, io_error);
635 static void clone_endio(struct bio *bio, int error)
638 struct dm_target_io *tio = bio->bi_private;
639 struct dm_io *io = tio->io;
640 struct mapped_device *md = tio->io->md;
641 dm_endio_fn endio = tio->ti->type->end_io;
643 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
647 r = endio(tio->ti, bio, error);
648 if (r < 0 || r == DM_ENDIO_REQUEUE)
650 * error and requeue request are handled
654 else if (r == DM_ENDIO_INCOMPLETE)
655 /* The target will handle the io */
658 DMWARN("unimplemented target endio return value: %d", r);
664 dec_pending(io, error);
668 * Partial completion handling for request-based dm
670 static void end_clone_bio(struct bio *clone, int error)
672 struct dm_rq_clone_bio_info *info = clone->bi_private;
673 struct dm_rq_target_io *tio = info->tio;
674 struct bio *bio = info->orig;
675 unsigned int nr_bytes = info->orig->bi_size;
681 * An error has already been detected on the request.
682 * Once error occurred, just let clone->end_io() handle
688 * Don't notice the error to the upper layer yet.
689 * The error handling decision is made by the target driver,
690 * when the request is completed.
697 * I/O for the bio successfully completed.
698 * Notice the data completion to the upper layer.
702 * bios are processed from the head of the list.
703 * So the completing bio should always be rq->bio.
704 * If it's not, something wrong is happening.
706 if (tio->orig->bio != bio)
707 DMERR("bio completion is going in the middle of the request");
710 * Update the original request.
711 * Do not use blk_end_request() here, because it may complete
712 * the original request before the clone, and break the ordering.
714 blk_update_request(tio->orig, 0, nr_bytes);
718 * Don't touch any member of the md after calling this function because
719 * the md may be freed in dm_put() at the end of this function.
720 * Or do dm_get() before calling this function and dm_put() later.
722 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
724 atomic_dec(&md->pending[rw]);
726 /* nudge anyone waiting on suspend queue */
727 if (!md_in_flight(md))
731 * Run this off this callpath, as drivers could invoke end_io while
732 * inside their request_fn (and holding the queue lock). Calling
733 * back into ->request_fn() could deadlock attempting to grab the
737 blk_run_queue_async(md->queue);
740 * dm_put() must be at the end of this function. See the comment above
745 static void free_rq_clone(struct request *clone)
747 struct dm_rq_target_io *tio = clone->end_io_data;
749 blk_rq_unprep_clone(clone);
754 * Complete the clone and the original request.
755 * Must be called without queue lock.
757 static void dm_end_request(struct request *clone, int error)
759 int rw = rq_data_dir(clone);
760 struct dm_rq_target_io *tio = clone->end_io_data;
761 struct mapped_device *md = tio->md;
762 struct request *rq = tio->orig;
764 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
765 rq->errors = clone->errors;
766 rq->resid_len = clone->resid_len;
770 * We are using the sense buffer of the original
772 * So setting the length of the sense data is enough.
774 rq->sense_len = clone->sense_len;
777 free_rq_clone(clone);
778 blk_end_request_all(rq, error);
779 rq_completed(md, rw, true);
782 static void dm_unprep_request(struct request *rq)
784 struct request *clone = rq->special;
787 rq->cmd_flags &= ~REQ_DONTPREP;
789 free_rq_clone(clone);
793 * Requeue the original request of a clone.
795 void dm_requeue_unmapped_request(struct request *clone)
797 int rw = rq_data_dir(clone);
798 struct dm_rq_target_io *tio = clone->end_io_data;
799 struct mapped_device *md = tio->md;
800 struct request *rq = tio->orig;
801 struct request_queue *q = rq->q;
804 dm_unprep_request(rq);
806 spin_lock_irqsave(q->queue_lock, flags);
807 blk_requeue_request(q, rq);
808 spin_unlock_irqrestore(q->queue_lock, flags);
810 rq_completed(md, rw, 0);
812 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
814 static void __stop_queue(struct request_queue *q)
819 static void stop_queue(struct request_queue *q)
823 spin_lock_irqsave(q->queue_lock, flags);
825 spin_unlock_irqrestore(q->queue_lock, flags);
828 static void __start_queue(struct request_queue *q)
830 if (blk_queue_stopped(q))
834 static void start_queue(struct request_queue *q)
838 spin_lock_irqsave(q->queue_lock, flags);
840 spin_unlock_irqrestore(q->queue_lock, flags);
843 static void dm_done(struct request *clone, int error, bool mapped)
846 struct dm_rq_target_io *tio = clone->end_io_data;
847 dm_request_endio_fn rq_end_io = NULL;
850 rq_end_io = tio->ti->type->rq_end_io;
852 if (mapped && rq_end_io)
853 r = rq_end_io(tio->ti, clone, error, &tio->info);
857 /* The target wants to complete the I/O */
858 dm_end_request(clone, r);
859 else if (r == DM_ENDIO_INCOMPLETE)
860 /* The target will handle the I/O */
862 else if (r == DM_ENDIO_REQUEUE)
863 /* The target wants to requeue the I/O */
864 dm_requeue_unmapped_request(clone);
866 DMWARN("unimplemented target endio return value: %d", r);
872 * Request completion handler for request-based dm
874 static void dm_softirq_done(struct request *rq)
877 struct request *clone = rq->completion_data;
878 struct dm_rq_target_io *tio = clone->end_io_data;
880 if (rq->cmd_flags & REQ_FAILED)
883 dm_done(clone, tio->error, mapped);
887 * Complete the clone and the original request with the error status
888 * through softirq context.
890 static void dm_complete_request(struct request *clone, int error)
892 struct dm_rq_target_io *tio = clone->end_io_data;
893 struct request *rq = tio->orig;
896 rq->completion_data = clone;
897 blk_complete_request(rq);
901 * Complete the not-mapped clone and the original request with the error status
902 * through softirq context.
903 * Target's rq_end_io() function isn't called.
904 * This may be used when the target's map_rq() function fails.
906 void dm_kill_unmapped_request(struct request *clone, int error)
908 struct dm_rq_target_io *tio = clone->end_io_data;
909 struct request *rq = tio->orig;
911 rq->cmd_flags |= REQ_FAILED;
912 dm_complete_request(clone, error);
914 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
917 * Called with the queue lock held
919 static void end_clone_request(struct request *clone, int error)
922 * For just cleaning up the information of the queue in which
923 * the clone was dispatched.
924 * The clone is *NOT* freed actually here because it is alloced from
925 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
927 __blk_put_request(clone->q, clone);
930 * Actual request completion is done in a softirq context which doesn't
931 * hold the queue lock. Otherwise, deadlock could occur because:
932 * - another request may be submitted by the upper level driver
933 * of the stacking during the completion
934 * - the submission which requires queue lock may be done
937 dm_complete_request(clone, error);
941 * Return maximum size of I/O possible at the supplied sector up to the current
944 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
946 sector_t target_offset = dm_target_offset(ti, sector);
948 return ti->len - target_offset;
951 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
953 sector_t len = max_io_len_target_boundary(sector, ti);
954 sector_t offset, max_len;
957 * Does the target need to split even further?
959 if (ti->max_io_len) {
960 offset = dm_target_offset(ti, sector);
961 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
962 max_len = sector_div(offset, ti->max_io_len);
964 max_len = offset & (ti->max_io_len - 1);
965 max_len = ti->max_io_len - max_len;
974 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
976 if (len > UINT_MAX) {
977 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
978 (unsigned long long)len, UINT_MAX);
979 ti->error = "Maximum size of target IO is too large";
983 ti->max_io_len = (uint32_t) len;
987 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
989 static void __map_bio(struct dm_target *ti, struct dm_target_io *tio)
993 struct mapped_device *md;
994 struct bio *clone = &tio->clone;
996 clone->bi_end_io = clone_endio;
997 clone->bi_private = tio;
1000 * Map the clone. If r == 0 we don't need to do
1001 * anything, the target has assumed ownership of
1004 atomic_inc(&tio->io->io_count);
1005 sector = clone->bi_sector;
1006 r = ti->type->map(ti, clone);
1007 if (r == DM_MAPIO_REMAPPED) {
1008 /* the bio has been remapped so dispatch it */
1010 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1011 tio->io->bio->bi_bdev->bd_dev, sector);
1013 generic_make_request(clone);
1014 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1015 /* error the io and bail out, or requeue it if needed */
1017 dec_pending(tio->io, r);
1020 DMWARN("unimplemented target map return value: %d", r);
1026 struct mapped_device *md;
1027 struct dm_table *map;
1031 sector_t sector_count;
1036 * Creates a little bio that just does part of a bvec.
1038 static void split_bvec(struct dm_target_io *tio, struct bio *bio,
1039 sector_t sector, unsigned short idx, unsigned int offset,
1040 unsigned int len, struct bio_set *bs)
1042 struct bio *clone = &tio->clone;
1043 struct bio_vec *bv = bio->bi_io_vec + idx;
1045 *clone->bi_io_vec = *bv;
1047 clone->bi_sector = sector;
1048 clone->bi_bdev = bio->bi_bdev;
1049 clone->bi_rw = bio->bi_rw;
1051 clone->bi_size = to_bytes(len);
1052 clone->bi_io_vec->bv_offset = offset;
1053 clone->bi_io_vec->bv_len = clone->bi_size;
1054 clone->bi_flags |= 1 << BIO_CLONED;
1056 if (bio_integrity(bio)) {
1057 bio_integrity_clone(clone, bio, GFP_NOIO);
1058 bio_integrity_trim(clone,
1059 bio_sector_offset(bio, idx, offset), len);
1064 * Creates a bio that consists of range of complete bvecs.
1066 static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1067 sector_t sector, unsigned short idx,
1068 unsigned short bv_count, unsigned int len,
1071 struct bio *clone = &tio->clone;
1073 __bio_clone(clone, bio);
1074 clone->bi_sector = sector;
1075 clone->bi_idx = idx;
1076 clone->bi_vcnt = idx + bv_count;
1077 clone->bi_size = to_bytes(len);
1078 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1080 if (bio_integrity(bio)) {
1081 bio_integrity_clone(clone, bio, GFP_NOIO);
1083 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1084 bio_integrity_trim(clone,
1085 bio_sector_offset(bio, idx, 0), len);
1089 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1090 struct dm_target *ti, int nr_iovecs)
1092 struct dm_target_io *tio;
1095 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1096 tio = container_of(clone, struct dm_target_io, clone);
1100 memset(&tio->info, 0, sizeof(tio->info));
1101 tio->target_request_nr = 0;
1106 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1107 unsigned request_nr, sector_t len)
1109 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs);
1110 struct bio *clone = &tio->clone;
1112 tio->target_request_nr = request_nr;
1115 * Discard requests require the bio's inline iovecs be initialized.
1116 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1117 * and discard, so no need for concern about wasted bvec allocations.
1120 __bio_clone(clone, ci->bio);
1122 clone->bi_sector = ci->sector;
1123 clone->bi_size = to_bytes(len);
1129 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1130 unsigned num_requests, sector_t len)
1132 unsigned request_nr;
1134 for (request_nr = 0; request_nr < num_requests; request_nr++)
1135 __issue_target_request(ci, ti, request_nr, len);
1138 static int __clone_and_map_empty_flush(struct clone_info *ci)
1140 unsigned target_nr = 0;
1141 struct dm_target *ti;
1143 BUG_ON(bio_has_data(ci->bio));
1144 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1145 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1151 * Perform all io with a single clone.
1153 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1155 struct bio *bio = ci->bio;
1156 struct dm_target_io *tio;
1158 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1159 clone_bio(tio, bio, ci->sector, ci->idx, bio->bi_vcnt - ci->idx,
1160 ci->sector_count, ci->md->bs);
1162 ci->sector_count = 0;
1165 typedef unsigned (*get_num_requests_fn)(struct dm_target *ti);
1167 static unsigned get_num_discard_requests(struct dm_target *ti)
1169 return ti->num_discard_requests;
1172 static unsigned get_num_write_same_requests(struct dm_target *ti)
1174 return ti->num_write_same_requests;
1177 typedef bool (*is_split_required_fn)(struct dm_target *ti);
1179 static bool is_split_required_for_discard(struct dm_target *ti)
1181 return ti->split_discard_requests;
1184 static int __clone_and_map_changing_extent_only(struct clone_info *ci,
1185 get_num_requests_fn get_num_requests,
1186 is_split_required_fn is_split_required)
1188 struct dm_target *ti;
1190 unsigned num_requests;
1193 ti = dm_table_find_target(ci->map, ci->sector);
1194 if (!dm_target_is_valid(ti))
1198 * Even though the device advertised support for this type of
1199 * request, that does not mean every target supports it, and
1200 * reconfiguration might also have changed that since the
1201 * check was performed.
1203 num_requests = get_num_requests ? get_num_requests(ti) : 0;
1207 if (is_split_required && !is_split_required(ti))
1208 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1210 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1212 __issue_target_requests(ci, ti, num_requests, len);
1215 } while (ci->sector_count -= len);
1220 static int __clone_and_map_discard(struct clone_info *ci)
1222 return __clone_and_map_changing_extent_only(ci, get_num_discard_requests,
1223 is_split_required_for_discard);
1226 static int __clone_and_map_write_same(struct clone_info *ci)
1228 return __clone_and_map_changing_extent_only(ci, get_num_write_same_requests, NULL);
1231 static int __clone_and_map(struct clone_info *ci)
1233 struct bio *bio = ci->bio;
1234 struct dm_target *ti;
1235 sector_t len = 0, max;
1236 struct dm_target_io *tio;
1238 if (unlikely(bio->bi_rw & REQ_DISCARD))
1239 return __clone_and_map_discard(ci);
1240 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1241 return __clone_and_map_write_same(ci);
1243 ti = dm_table_find_target(ci->map, ci->sector);
1244 if (!dm_target_is_valid(ti))
1247 max = max_io_len(ci->sector, ti);
1249 if (ci->sector_count <= max) {
1251 * Optimise for the simple case where we can do all of
1252 * the remaining io with a single clone.
1254 __clone_and_map_simple(ci, ti);
1256 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1258 * There are some bvecs that don't span targets.
1259 * Do as many of these as possible.
1262 sector_t remaining = max;
1265 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1266 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1268 if (bv_len > remaining)
1271 remaining -= bv_len;
1275 tio = alloc_tio(ci, ti, bio->bi_max_vecs);
1276 clone_bio(tio, bio, ci->sector, ci->idx, i - ci->idx, len,
1281 ci->sector_count -= len;
1286 * Handle a bvec that must be split between two or more targets.
1288 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1289 sector_t remaining = to_sector(bv->bv_len);
1290 unsigned int offset = 0;
1294 ti = dm_table_find_target(ci->map, ci->sector);
1295 if (!dm_target_is_valid(ti))
1298 max = max_io_len(ci->sector, ti);
1301 len = min(remaining, max);
1303 tio = alloc_tio(ci, ti, 1);
1304 split_bvec(tio, bio, ci->sector, ci->idx,
1305 bv->bv_offset + offset, len, ci->md->bs);
1310 ci->sector_count -= len;
1311 offset += to_bytes(len);
1312 } while (remaining -= len);
1321 * Split the bio into several clones and submit it to targets.
1323 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1325 struct clone_info ci;
1328 ci.map = dm_get_live_table(md);
1329 if (unlikely(!ci.map)) {
1335 ci.io = alloc_io(md);
1337 atomic_set(&ci.io->io_count, 1);
1340 spin_lock_init(&ci.io->endio_lock);
1341 ci.sector = bio->bi_sector;
1342 ci.idx = bio->bi_idx;
1344 start_io_acct(ci.io);
1345 if (bio->bi_rw & REQ_FLUSH) {
1346 ci.bio = &ci.md->flush_bio;
1347 ci.sector_count = 0;
1348 error = __clone_and_map_empty_flush(&ci);
1349 /* dec_pending submits any data associated with flush */
1352 ci.sector_count = bio_sectors(bio);
1353 while (ci.sector_count && !error)
1354 error = __clone_and_map(&ci);
1357 /* drop the extra reference count */
1358 dec_pending(ci.io, error);
1359 dm_table_put(ci.map);
1361 /*-----------------------------------------------------------------
1363 *---------------------------------------------------------------*/
1365 static int dm_merge_bvec(struct request_queue *q,
1366 struct bvec_merge_data *bvm,
1367 struct bio_vec *biovec)
1369 struct mapped_device *md = q->queuedata;
1370 struct dm_table *map = dm_get_live_table(md);
1371 struct dm_target *ti;
1372 sector_t max_sectors;
1378 ti = dm_table_find_target(map, bvm->bi_sector);
1379 if (!dm_target_is_valid(ti))
1383 * Find maximum amount of I/O that won't need splitting
1385 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1386 (sector_t) BIO_MAX_SECTORS);
1387 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1392 * merge_bvec_fn() returns number of bytes
1393 * it can accept at this offset
1394 * max is precomputed maximal io size
1396 if (max_size && ti->type->merge)
1397 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1399 * If the target doesn't support merge method and some of the devices
1400 * provided their merge_bvec method (we know this by looking at
1401 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1402 * entries. So always set max_size to 0, and the code below allows
1405 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1414 * Always allow an entire first page
1416 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1417 max_size = biovec->bv_len;
1423 * The request function that just remaps the bio built up by
1426 static void _dm_request(struct request_queue *q, struct bio *bio)
1428 int rw = bio_data_dir(bio);
1429 struct mapped_device *md = q->queuedata;
1432 down_read(&md->io_lock);
1434 cpu = part_stat_lock();
1435 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1436 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1439 /* if we're suspended, we have to queue this io for later */
1440 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1441 up_read(&md->io_lock);
1443 if (bio_rw(bio) != READA)
1450 __split_and_process_bio(md, bio);
1451 up_read(&md->io_lock);
1455 static int dm_request_based(struct mapped_device *md)
1457 return blk_queue_stackable(md->queue);
1460 static void dm_request(struct request_queue *q, struct bio *bio)
1462 struct mapped_device *md = q->queuedata;
1464 if (dm_request_based(md))
1465 blk_queue_bio(q, bio);
1467 _dm_request(q, bio);
1470 void dm_dispatch_request(struct request *rq)
1474 if (blk_queue_io_stat(rq->q))
1475 rq->cmd_flags |= REQ_IO_STAT;
1477 rq->start_time = jiffies;
1478 r = blk_insert_cloned_request(rq->q, rq);
1480 dm_complete_request(rq, r);
1482 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1484 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1487 struct dm_rq_target_io *tio = data;
1488 struct dm_rq_clone_bio_info *info =
1489 container_of(bio, struct dm_rq_clone_bio_info, clone);
1491 info->orig = bio_orig;
1493 bio->bi_end_io = end_clone_bio;
1494 bio->bi_private = info;
1499 static int setup_clone(struct request *clone, struct request *rq,
1500 struct dm_rq_target_io *tio)
1504 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1505 dm_rq_bio_constructor, tio);
1509 clone->cmd = rq->cmd;
1510 clone->cmd_len = rq->cmd_len;
1511 clone->sense = rq->sense;
1512 clone->buffer = rq->buffer;
1513 clone->end_io = end_clone_request;
1514 clone->end_io_data = tio;
1519 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1522 struct request *clone;
1523 struct dm_rq_target_io *tio;
1525 tio = alloc_rq_tio(md, gfp_mask);
1533 memset(&tio->info, 0, sizeof(tio->info));
1535 clone = &tio->clone;
1536 if (setup_clone(clone, rq, tio)) {
1546 * Called with the queue lock held.
1548 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1550 struct mapped_device *md = q->queuedata;
1551 struct request *clone;
1553 if (unlikely(rq->special)) {
1554 DMWARN("Already has something in rq->special.");
1555 return BLKPREP_KILL;
1558 clone = clone_rq(rq, md, GFP_ATOMIC);
1560 return BLKPREP_DEFER;
1562 rq->special = clone;
1563 rq->cmd_flags |= REQ_DONTPREP;
1570 * 0 : the request has been processed (not requeued)
1571 * !0 : the request has been requeued
1573 static int map_request(struct dm_target *ti, struct request *clone,
1574 struct mapped_device *md)
1576 int r, requeued = 0;
1577 struct dm_rq_target_io *tio = clone->end_io_data;
1580 r = ti->type->map_rq(ti, clone, &tio->info);
1582 case DM_MAPIO_SUBMITTED:
1583 /* The target has taken the I/O to submit by itself later */
1585 case DM_MAPIO_REMAPPED:
1586 /* The target has remapped the I/O so dispatch it */
1587 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1588 blk_rq_pos(tio->orig));
1589 dm_dispatch_request(clone);
1591 case DM_MAPIO_REQUEUE:
1592 /* The target wants to requeue the I/O */
1593 dm_requeue_unmapped_request(clone);
1598 DMWARN("unimplemented target map return value: %d", r);
1602 /* The target wants to complete the I/O */
1603 dm_kill_unmapped_request(clone, r);
1610 static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1612 struct request *clone;
1614 blk_start_request(orig);
1615 clone = orig->special;
1616 atomic_inc(&md->pending[rq_data_dir(clone)]);
1619 * Hold the md reference here for the in-flight I/O.
1620 * We can't rely on the reference count by device opener,
1621 * because the device may be closed during the request completion
1622 * when all bios are completed.
1623 * See the comment in rq_completed() too.
1631 * q->request_fn for request-based dm.
1632 * Called with the queue lock held.
1634 static void dm_request_fn(struct request_queue *q)
1636 struct mapped_device *md = q->queuedata;
1637 struct dm_table *map = dm_get_live_table(md);
1638 struct dm_target *ti;
1639 struct request *rq, *clone;
1643 * For suspend, check blk_queue_stopped() and increment
1644 * ->pending within a single queue_lock not to increment the
1645 * number of in-flight I/Os after the queue is stopped in
1648 while (!blk_queue_stopped(q)) {
1649 rq = blk_peek_request(q);
1653 /* always use block 0 to find the target for flushes for now */
1655 if (!(rq->cmd_flags & REQ_FLUSH))
1656 pos = blk_rq_pos(rq);
1658 ti = dm_table_find_target(map, pos);
1659 if (!dm_target_is_valid(ti)) {
1661 * Must perform setup, that dm_done() requires,
1662 * before calling dm_kill_unmapped_request
1664 DMERR_LIMIT("request attempted access beyond the end of device");
1665 clone = dm_start_request(md, rq);
1666 dm_kill_unmapped_request(clone, -EIO);
1670 if (ti->type->busy && ti->type->busy(ti))
1673 clone = dm_start_request(md, rq);
1675 spin_unlock(q->queue_lock);
1676 if (map_request(ti, clone, md))
1679 BUG_ON(!irqs_disabled());
1680 spin_lock(q->queue_lock);
1686 BUG_ON(!irqs_disabled());
1687 spin_lock(q->queue_lock);
1690 blk_delay_queue(q, HZ / 10);
1695 int dm_underlying_device_busy(struct request_queue *q)
1697 return blk_lld_busy(q);
1699 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1701 static int dm_lld_busy(struct request_queue *q)
1704 struct mapped_device *md = q->queuedata;
1705 struct dm_table *map = dm_get_live_table(md);
1707 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1710 r = dm_table_any_busy_target(map);
1717 static int dm_any_congested(void *congested_data, int bdi_bits)
1720 struct mapped_device *md = congested_data;
1721 struct dm_table *map;
1723 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1724 map = dm_get_live_table(md);
1727 * Request-based dm cares about only own queue for
1728 * the query about congestion status of request_queue
1730 if (dm_request_based(md))
1731 r = md->queue->backing_dev_info.state &
1734 r = dm_table_any_congested(map, bdi_bits);
1743 /*-----------------------------------------------------------------
1744 * An IDR is used to keep track of allocated minor numbers.
1745 *---------------------------------------------------------------*/
1746 static void free_minor(int minor)
1748 spin_lock(&_minor_lock);
1749 idr_remove(&_minor_idr, minor);
1750 spin_unlock(&_minor_lock);
1754 * See if the device with a specific minor # is free.
1756 static int specific_minor(int minor)
1760 if (minor >= (1 << MINORBITS))
1763 idr_preload(GFP_KERNEL);
1764 spin_lock(&_minor_lock);
1766 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1768 spin_unlock(&_minor_lock);
1771 return r == -ENOSPC ? -EBUSY : r;
1775 static int next_free_minor(int *minor)
1779 idr_preload(GFP_KERNEL);
1780 spin_lock(&_minor_lock);
1782 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1784 spin_unlock(&_minor_lock);
1792 static const struct block_device_operations dm_blk_dops;
1794 static void dm_wq_work(struct work_struct *work);
1796 static void dm_init_md_queue(struct mapped_device *md)
1799 * Request-based dm devices cannot be stacked on top of bio-based dm
1800 * devices. The type of this dm device has not been decided yet.
1801 * The type is decided at the first table loading time.
1802 * To prevent problematic device stacking, clear the queue flag
1803 * for request stacking support until then.
1805 * This queue is new, so no concurrency on the queue_flags.
1807 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1809 md->queue->queuedata = md;
1810 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1811 md->queue->backing_dev_info.congested_data = md;
1812 blk_queue_make_request(md->queue, dm_request);
1813 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1814 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1818 * Allocate and initialise a blank device with a given minor.
1820 static struct mapped_device *alloc_dev(int minor)
1823 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1827 DMWARN("unable to allocate device, out of memory.");
1831 if (!try_module_get(THIS_MODULE))
1832 goto bad_module_get;
1834 /* get a minor number for the dev */
1835 if (minor == DM_ANY_MINOR)
1836 r = next_free_minor(&minor);
1838 r = specific_minor(minor);
1842 md->type = DM_TYPE_NONE;
1843 init_rwsem(&md->io_lock);
1844 mutex_init(&md->suspend_lock);
1845 mutex_init(&md->type_lock);
1846 spin_lock_init(&md->deferred_lock);
1847 rwlock_init(&md->map_lock);
1848 atomic_set(&md->holders, 1);
1849 atomic_set(&md->open_count, 0);
1850 atomic_set(&md->event_nr, 0);
1851 atomic_set(&md->uevent_seq, 0);
1852 INIT_LIST_HEAD(&md->uevent_list);
1853 spin_lock_init(&md->uevent_lock);
1855 md->queue = blk_alloc_queue(GFP_KERNEL);
1859 dm_init_md_queue(md);
1861 md->disk = alloc_disk(1);
1865 atomic_set(&md->pending[0], 0);
1866 atomic_set(&md->pending[1], 0);
1867 init_waitqueue_head(&md->wait);
1868 INIT_WORK(&md->work, dm_wq_work);
1869 init_waitqueue_head(&md->eventq);
1871 md->disk->major = _major;
1872 md->disk->first_minor = minor;
1873 md->disk->fops = &dm_blk_dops;
1874 md->disk->queue = md->queue;
1875 md->disk->private_data = md;
1876 sprintf(md->disk->disk_name, "dm-%d", minor);
1878 format_dev_t(md->name, MKDEV(_major, minor));
1880 md->wq = alloc_workqueue("kdmflush",
1881 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1885 md->bdev = bdget_disk(md->disk, 0);
1889 bio_init(&md->flush_bio);
1890 md->flush_bio.bi_bdev = md->bdev;
1891 md->flush_bio.bi_rw = WRITE_FLUSH;
1893 /* Populate the mapping, nobody knows we exist yet */
1894 spin_lock(&_minor_lock);
1895 old_md = idr_replace(&_minor_idr, md, minor);
1896 spin_unlock(&_minor_lock);
1898 BUG_ON(old_md != MINOR_ALLOCED);
1903 destroy_workqueue(md->wq);
1905 del_gendisk(md->disk);
1908 blk_cleanup_queue(md->queue);
1912 module_put(THIS_MODULE);
1918 static void unlock_fs(struct mapped_device *md);
1920 static void free_dev(struct mapped_device *md)
1922 int minor = MINOR(disk_devt(md->disk));
1926 destroy_workqueue(md->wq);
1928 mempool_destroy(md->tio_pool);
1930 mempool_destroy(md->io_pool);
1932 bioset_free(md->bs);
1933 blk_integrity_unregister(md->disk);
1934 del_gendisk(md->disk);
1937 spin_lock(&_minor_lock);
1938 md->disk->private_data = NULL;
1939 spin_unlock(&_minor_lock);
1942 blk_cleanup_queue(md->queue);
1943 module_put(THIS_MODULE);
1947 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1949 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1951 if (md->io_pool && (md->tio_pool || dm_table_get_type(t) == DM_TYPE_BIO_BASED) && md->bs) {
1953 * The md already has necessary mempools. Reload just the
1954 * bioset because front_pad may have changed because
1955 * a different table was loaded.
1957 bioset_free(md->bs);
1963 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1965 md->io_pool = p->io_pool;
1967 md->tio_pool = p->tio_pool;
1973 /* mempool bind completed, now no need any mempools in the table */
1974 dm_table_free_md_mempools(t);
1978 * Bind a table to the device.
1980 static void event_callback(void *context)
1982 unsigned long flags;
1984 struct mapped_device *md = (struct mapped_device *) context;
1986 spin_lock_irqsave(&md->uevent_lock, flags);
1987 list_splice_init(&md->uevent_list, &uevents);
1988 spin_unlock_irqrestore(&md->uevent_lock, flags);
1990 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1992 atomic_inc(&md->event_nr);
1993 wake_up(&md->eventq);
1997 * Protected by md->suspend_lock obtained by dm_swap_table().
1999 static void __set_size(struct mapped_device *md, sector_t size)
2001 set_capacity(md->disk, size);
2003 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2007 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2009 * If this function returns 0, then the device is either a non-dm
2010 * device without a merge_bvec_fn, or it is a dm device that is
2011 * able to split any bios it receives that are too big.
2013 int dm_queue_merge_is_compulsory(struct request_queue *q)
2015 struct mapped_device *dev_md;
2017 if (!q->merge_bvec_fn)
2020 if (q->make_request_fn == dm_request) {
2021 dev_md = q->queuedata;
2022 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2029 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2030 struct dm_dev *dev, sector_t start,
2031 sector_t len, void *data)
2033 struct block_device *bdev = dev->bdev;
2034 struct request_queue *q = bdev_get_queue(bdev);
2036 return dm_queue_merge_is_compulsory(q);
2040 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2041 * on the properties of the underlying devices.
2043 static int dm_table_merge_is_optional(struct dm_table *table)
2046 struct dm_target *ti;
2048 while (i < dm_table_get_num_targets(table)) {
2049 ti = dm_table_get_target(table, i++);
2051 if (ti->type->iterate_devices &&
2052 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2060 * Returns old map, which caller must destroy.
2062 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2063 struct queue_limits *limits)
2065 struct dm_table *old_map;
2066 struct request_queue *q = md->queue;
2068 unsigned long flags;
2069 int merge_is_optional;
2071 size = dm_table_get_size(t);
2074 * Wipe any geometry if the size of the table changed.
2076 if (size != get_capacity(md->disk))
2077 memset(&md->geometry, 0, sizeof(md->geometry));
2079 __set_size(md, size);
2081 dm_table_event_callback(t, event_callback, md);
2084 * The queue hasn't been stopped yet, if the old table type wasn't
2085 * for request-based during suspension. So stop it to prevent
2086 * I/O mapping before resume.
2087 * This must be done before setting the queue restrictions,
2088 * because request-based dm may be run just after the setting.
2090 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2093 __bind_mempools(md, t);
2095 merge_is_optional = dm_table_merge_is_optional(t);
2097 write_lock_irqsave(&md->map_lock, flags);
2100 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2102 dm_table_set_restrictions(t, q, limits);
2103 if (merge_is_optional)
2104 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2106 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2107 write_unlock_irqrestore(&md->map_lock, flags);
2113 * Returns unbound table for the caller to free.
2115 static struct dm_table *__unbind(struct mapped_device *md)
2117 struct dm_table *map = md->map;
2118 unsigned long flags;
2123 dm_table_event_callback(map, NULL, NULL);
2124 write_lock_irqsave(&md->map_lock, flags);
2126 write_unlock_irqrestore(&md->map_lock, flags);
2132 * Constructor for a new device.
2134 int dm_create(int minor, struct mapped_device **result)
2136 struct mapped_device *md;
2138 md = alloc_dev(minor);
2149 * Functions to manage md->type.
2150 * All are required to hold md->type_lock.
2152 void dm_lock_md_type(struct mapped_device *md)
2154 mutex_lock(&md->type_lock);
2157 void dm_unlock_md_type(struct mapped_device *md)
2159 mutex_unlock(&md->type_lock);
2162 void dm_set_md_type(struct mapped_device *md, unsigned type)
2167 unsigned dm_get_md_type(struct mapped_device *md)
2172 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2174 return md->immutable_target_type;
2178 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2180 static int dm_init_request_based_queue(struct mapped_device *md)
2182 struct request_queue *q = NULL;
2184 if (md->queue->elevator)
2187 /* Fully initialize the queue */
2188 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2193 dm_init_md_queue(md);
2194 blk_queue_softirq_done(md->queue, dm_softirq_done);
2195 blk_queue_prep_rq(md->queue, dm_prep_fn);
2196 blk_queue_lld_busy(md->queue, dm_lld_busy);
2198 elv_register_queue(md->queue);
2204 * Setup the DM device's queue based on md's type
2206 int dm_setup_md_queue(struct mapped_device *md)
2208 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2209 !dm_init_request_based_queue(md)) {
2210 DMWARN("Cannot initialize queue for request-based mapped device");
2217 static struct mapped_device *dm_find_md(dev_t dev)
2219 struct mapped_device *md;
2220 unsigned minor = MINOR(dev);
2222 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2225 spin_lock(&_minor_lock);
2227 md = idr_find(&_minor_idr, minor);
2228 if (md && (md == MINOR_ALLOCED ||
2229 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2230 dm_deleting_md(md) ||
2231 test_bit(DMF_FREEING, &md->flags))) {
2237 spin_unlock(&_minor_lock);
2242 struct mapped_device *dm_get_md(dev_t dev)
2244 struct mapped_device *md = dm_find_md(dev);
2251 EXPORT_SYMBOL_GPL(dm_get_md);
2253 void *dm_get_mdptr(struct mapped_device *md)
2255 return md->interface_ptr;
2258 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2260 md->interface_ptr = ptr;
2263 void dm_get(struct mapped_device *md)
2265 atomic_inc(&md->holders);
2266 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2269 const char *dm_device_name(struct mapped_device *md)
2273 EXPORT_SYMBOL_GPL(dm_device_name);
2275 static void __dm_destroy(struct mapped_device *md, bool wait)
2277 struct dm_table *map;
2281 spin_lock(&_minor_lock);
2282 map = dm_get_live_table(md);
2283 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2284 set_bit(DMF_FREEING, &md->flags);
2285 spin_unlock(&_minor_lock);
2287 if (!dm_suspended_md(md)) {
2288 dm_table_presuspend_targets(map);
2289 dm_table_postsuspend_targets(map);
2293 * Rare, but there may be I/O requests still going to complete,
2294 * for example. Wait for all references to disappear.
2295 * No one should increment the reference count of the mapped_device,
2296 * after the mapped_device state becomes DMF_FREEING.
2299 while (atomic_read(&md->holders))
2301 else if (atomic_read(&md->holders))
2302 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2303 dm_device_name(md), atomic_read(&md->holders));
2307 dm_table_destroy(__unbind(md));
2311 void dm_destroy(struct mapped_device *md)
2313 __dm_destroy(md, true);
2316 void dm_destroy_immediate(struct mapped_device *md)
2318 __dm_destroy(md, false);
2321 void dm_put(struct mapped_device *md)
2323 atomic_dec(&md->holders);
2325 EXPORT_SYMBOL_GPL(dm_put);
2327 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2330 DECLARE_WAITQUEUE(wait, current);
2332 add_wait_queue(&md->wait, &wait);
2335 set_current_state(interruptible);
2337 if (!md_in_flight(md))
2340 if (interruptible == TASK_INTERRUPTIBLE &&
2341 signal_pending(current)) {
2348 set_current_state(TASK_RUNNING);
2350 remove_wait_queue(&md->wait, &wait);
2356 * Process the deferred bios
2358 static void dm_wq_work(struct work_struct *work)
2360 struct mapped_device *md = container_of(work, struct mapped_device,
2364 down_read(&md->io_lock);
2366 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2367 spin_lock_irq(&md->deferred_lock);
2368 c = bio_list_pop(&md->deferred);
2369 spin_unlock_irq(&md->deferred_lock);
2374 up_read(&md->io_lock);
2376 if (dm_request_based(md))
2377 generic_make_request(c);
2379 __split_and_process_bio(md, c);
2381 down_read(&md->io_lock);
2384 up_read(&md->io_lock);
2387 static void dm_queue_flush(struct mapped_device *md)
2389 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2390 smp_mb__after_clear_bit();
2391 queue_work(md->wq, &md->work);
2395 * Swap in a new table, returning the old one for the caller to destroy.
2397 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2399 struct dm_table *live_map, *map = ERR_PTR(-EINVAL);
2400 struct queue_limits limits;
2403 mutex_lock(&md->suspend_lock);
2405 /* device must be suspended */
2406 if (!dm_suspended_md(md))
2410 * If the new table has no data devices, retain the existing limits.
2411 * This helps multipath with queue_if_no_path if all paths disappear,
2412 * then new I/O is queued based on these limits, and then some paths
2415 if (dm_table_has_no_data_devices(table)) {
2416 live_map = dm_get_live_table(md);
2418 limits = md->queue->limits;
2419 dm_table_put(live_map);
2422 r = dm_calculate_queue_limits(table, &limits);
2428 map = __bind(md, table, &limits);
2431 mutex_unlock(&md->suspend_lock);
2436 * Functions to lock and unlock any filesystem running on the
2439 static int lock_fs(struct mapped_device *md)
2443 WARN_ON(md->frozen_sb);
2445 md->frozen_sb = freeze_bdev(md->bdev);
2446 if (IS_ERR(md->frozen_sb)) {
2447 r = PTR_ERR(md->frozen_sb);
2448 md->frozen_sb = NULL;
2452 set_bit(DMF_FROZEN, &md->flags);
2457 static void unlock_fs(struct mapped_device *md)
2459 if (!test_bit(DMF_FROZEN, &md->flags))
2462 thaw_bdev(md->bdev, md->frozen_sb);
2463 md->frozen_sb = NULL;
2464 clear_bit(DMF_FROZEN, &md->flags);
2468 * We need to be able to change a mapping table under a mounted
2469 * filesystem. For example we might want to move some data in
2470 * the background. Before the table can be swapped with
2471 * dm_bind_table, dm_suspend must be called to flush any in
2472 * flight bios and ensure that any further io gets deferred.
2475 * Suspend mechanism in request-based dm.
2477 * 1. Flush all I/Os by lock_fs() if needed.
2478 * 2. Stop dispatching any I/O by stopping the request_queue.
2479 * 3. Wait for all in-flight I/Os to be completed or requeued.
2481 * To abort suspend, start the request_queue.
2483 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2485 struct dm_table *map = NULL;
2487 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2488 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2490 mutex_lock(&md->suspend_lock);
2492 if (dm_suspended_md(md)) {
2497 map = dm_get_live_table(md);
2500 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2501 * This flag is cleared before dm_suspend returns.
2504 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2506 /* This does not get reverted if there's an error later. */
2507 dm_table_presuspend_targets(map);
2510 * Flush I/O to the device.
2511 * Any I/O submitted after lock_fs() may not be flushed.
2512 * noflush takes precedence over do_lockfs.
2513 * (lock_fs() flushes I/Os and waits for them to complete.)
2515 if (!noflush && do_lockfs) {
2522 * Here we must make sure that no processes are submitting requests
2523 * to target drivers i.e. no one may be executing
2524 * __split_and_process_bio. This is called from dm_request and
2527 * To get all processes out of __split_and_process_bio in dm_request,
2528 * we take the write lock. To prevent any process from reentering
2529 * __split_and_process_bio from dm_request and quiesce the thread
2530 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2531 * flush_workqueue(md->wq).
2533 down_write(&md->io_lock);
2534 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2535 up_write(&md->io_lock);
2538 * Stop md->queue before flushing md->wq in case request-based
2539 * dm defers requests to md->wq from md->queue.
2541 if (dm_request_based(md))
2542 stop_queue(md->queue);
2544 flush_workqueue(md->wq);
2547 * At this point no more requests are entering target request routines.
2548 * We call dm_wait_for_completion to wait for all existing requests
2551 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2553 down_write(&md->io_lock);
2555 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2556 up_write(&md->io_lock);
2558 /* were we interrupted ? */
2562 if (dm_request_based(md))
2563 start_queue(md->queue);
2566 goto out; /* pushback list is already flushed, so skip flush */
2570 * If dm_wait_for_completion returned 0, the device is completely
2571 * quiescent now. There is no request-processing activity. All new
2572 * requests are being added to md->deferred list.
2575 set_bit(DMF_SUSPENDED, &md->flags);
2577 dm_table_postsuspend_targets(map);
2583 mutex_unlock(&md->suspend_lock);
2587 int dm_resume(struct mapped_device *md)
2590 struct dm_table *map = NULL;
2592 mutex_lock(&md->suspend_lock);
2593 if (!dm_suspended_md(md))
2596 map = dm_get_live_table(md);
2597 if (!map || !dm_table_get_size(map))
2600 r = dm_table_resume_targets(map);
2607 * Flushing deferred I/Os must be done after targets are resumed
2608 * so that mapping of targets can work correctly.
2609 * Request-based dm is queueing the deferred I/Os in its request_queue.
2611 if (dm_request_based(md))
2612 start_queue(md->queue);
2616 clear_bit(DMF_SUSPENDED, &md->flags);
2621 mutex_unlock(&md->suspend_lock);
2626 /*-----------------------------------------------------------------
2627 * Event notification.
2628 *---------------------------------------------------------------*/
2629 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2632 char udev_cookie[DM_COOKIE_LENGTH];
2633 char *envp[] = { udev_cookie, NULL };
2636 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2638 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2639 DM_COOKIE_ENV_VAR_NAME, cookie);
2640 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2645 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2647 return atomic_add_return(1, &md->uevent_seq);
2650 uint32_t dm_get_event_nr(struct mapped_device *md)
2652 return atomic_read(&md->event_nr);
2655 int dm_wait_event(struct mapped_device *md, int event_nr)
2657 return wait_event_interruptible(md->eventq,
2658 (event_nr != atomic_read(&md->event_nr)));
2661 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2663 unsigned long flags;
2665 spin_lock_irqsave(&md->uevent_lock, flags);
2666 list_add(elist, &md->uevent_list);
2667 spin_unlock_irqrestore(&md->uevent_lock, flags);
2671 * The gendisk is only valid as long as you have a reference
2674 struct gendisk *dm_disk(struct mapped_device *md)
2679 struct kobject *dm_kobject(struct mapped_device *md)
2685 * struct mapped_device should not be exported outside of dm.c
2686 * so use this check to verify that kobj is part of md structure
2688 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2690 struct mapped_device *md;
2692 md = container_of(kobj, struct mapped_device, kobj);
2693 if (&md->kobj != kobj)
2696 if (test_bit(DMF_FREEING, &md->flags) ||
2704 int dm_suspended_md(struct mapped_device *md)
2706 return test_bit(DMF_SUSPENDED, &md->flags);
2709 int dm_suspended(struct dm_target *ti)
2711 return dm_suspended_md(dm_table_get_md(ti->table));
2713 EXPORT_SYMBOL_GPL(dm_suspended);
2715 int dm_noflush_suspending(struct dm_target *ti)
2717 return __noflush_suspending(dm_table_get_md(ti->table));
2719 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2721 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2723 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2724 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2729 per_bio_data_size = roundup(per_bio_data_size, __alignof__(struct dm_target_io));
2731 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2732 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2733 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2734 if (!pools->io_pool)
2735 goto free_pools_and_out;
2737 pools->tio_pool = NULL;
2738 if (type == DM_TYPE_REQUEST_BASED) {
2739 pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2740 if (!pools->tio_pool)
2741 goto free_io_pool_and_out;
2744 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2745 bioset_create(pool_size,
2746 per_bio_data_size + offsetof(struct dm_target_io, clone)) :
2747 bioset_create(pool_size,
2748 offsetof(struct dm_rq_clone_bio_info, clone));
2750 goto free_tio_pool_and_out;
2752 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2753 goto free_bioset_and_out;
2757 free_bioset_and_out:
2758 bioset_free(pools->bs);
2760 free_tio_pool_and_out:
2761 if (pools->tio_pool)
2762 mempool_destroy(pools->tio_pool);
2764 free_io_pool_and_out:
2765 mempool_destroy(pools->io_pool);
2773 void dm_free_md_mempools(struct dm_md_mempools *pools)
2779 mempool_destroy(pools->io_pool);
2781 if (pools->tio_pool)
2782 mempool_destroy(pools->tio_pool);
2785 bioset_free(pools->bs);
2790 static const struct block_device_operations dm_blk_dops = {
2791 .open = dm_blk_open,
2792 .release = dm_blk_close,
2793 .ioctl = dm_blk_ioctl,
2794 .getgeo = dm_blk_getgeo,
2795 .owner = THIS_MODULE
2798 EXPORT_SYMBOL(dm_get_mapinfo);
2803 module_init(dm_init);
2804 module_exit(dm_exit);
2806 module_param(major, uint, 0);
2807 MODULE_PARM_DESC(major, "The major number of the device mapper");
2808 MODULE_DESCRIPTION(DM_NAME " driver");
2809 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2810 MODULE_LICENSE("GPL");