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/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
28 * Cookies are numeric values sent with CHANGE and REMOVE
29 * uevents while resuming, removing or renaming the device.
31 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
32 #define DM_COOKIE_LENGTH 24
34 static const char *_name = DM_NAME;
36 static unsigned int major = 0;
37 static unsigned int _major = 0;
39 static DEFINE_SPINLOCK(_minor_lock);
42 * One of these is allocated per bio.
45 struct mapped_device *md;
49 unsigned long start_time;
50 spinlock_t endio_lock;
55 * One of these is allocated per target within a bio. Hopefully
56 * this will be simplified out one day.
65 * For request-based dm.
66 * One of these is allocated per request.
68 struct dm_rq_target_io {
69 struct mapped_device *md;
71 struct request *orig, clone;
77 * For request-based dm.
78 * One of these is allocated per bio.
80 struct dm_rq_clone_bio_info {
82 struct dm_rq_target_io *tio;
85 union map_info *dm_get_mapinfo(struct bio *bio)
87 if (bio && bio->bi_private)
88 return &((struct dm_target_io *)bio->bi_private)->info;
92 union map_info *dm_get_rq_mapinfo(struct request *rq)
94 if (rq && rq->end_io_data)
95 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
98 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
100 #define MINOR_ALLOCED ((void *)-1)
103 * Bits for the md->flags field.
105 #define DMF_BLOCK_IO_FOR_SUSPEND 0
106 #define DMF_SUSPENDED 1
108 #define DMF_FREEING 3
109 #define DMF_DELETING 4
110 #define DMF_NOFLUSH_SUSPENDING 5
111 #define DMF_QUEUE_IO_TO_THREAD 6
114 * Work processed by per-device workqueue.
116 struct mapped_device {
117 struct rw_semaphore io_lock;
118 struct mutex suspend_lock;
125 struct request_queue *queue;
126 struct gendisk *disk;
132 * A list of ios that arrived while we were suspended.
135 wait_queue_head_t wait;
136 struct work_struct work;
137 struct bio_list deferred;
138 spinlock_t deferred_lock;
141 * An error from the barrier request currently being processed.
146 * Processing queue (flush/barriers)
148 struct workqueue_struct *wq;
151 * The current mapping.
153 struct dm_table *map;
156 * io objects are allocated from here.
167 wait_queue_head_t eventq;
169 struct list_head uevent_list;
170 spinlock_t uevent_lock; /* Protect access to uevent_list */
173 * freeze/thaw support require holding onto a super block
175 struct super_block *frozen_sb;
176 struct block_device *bdev;
178 /* forced geometry settings */
179 struct hd_geometry geometry;
181 /* marker of flush suspend for request-based dm */
182 struct request suspend_rq;
184 /* For saving the address of __make_request for request based dm */
185 make_request_fn *saved_make_request_fn;
190 /* zero-length barrier that will be cloned and submitted to targets */
191 struct bio barrier_bio;
195 * For mempools pre-allocation at the table loading time.
197 struct dm_md_mempools {
204 static struct kmem_cache *_io_cache;
205 static struct kmem_cache *_tio_cache;
206 static struct kmem_cache *_rq_tio_cache;
207 static struct kmem_cache *_rq_bio_info_cache;
209 static int __init local_init(void)
213 /* allocate a slab for the dm_ios */
214 _io_cache = KMEM_CACHE(dm_io, 0);
218 /* allocate a slab for the target ios */
219 _tio_cache = KMEM_CACHE(dm_target_io, 0);
221 goto out_free_io_cache;
223 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
225 goto out_free_tio_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(_tio_cache);
254 kmem_cache_destroy(_io_cache);
259 static void local_exit(void)
261 kmem_cache_destroy(_rq_bio_info_cache);
262 kmem_cache_destroy(_rq_tio_cache);
263 kmem_cache_destroy(_tio_cache);
264 kmem_cache_destroy(_io_cache);
265 unregister_blkdev(_major, _name);
270 DMINFO("cleaned up");
273 static int (*_inits[])(void) __initdata = {
283 static void (*_exits[])(void) = {
293 static int __init dm_init(void)
295 const int count = ARRAY_SIZE(_inits);
299 for (i = 0; i < count; i++) {
314 static void __exit dm_exit(void)
316 int i = ARRAY_SIZE(_exits);
323 * Block device functions
325 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
327 struct mapped_device *md;
329 spin_lock(&_minor_lock);
331 md = bdev->bd_disk->private_data;
335 if (test_bit(DMF_FREEING, &md->flags) ||
336 test_bit(DMF_DELETING, &md->flags)) {
342 atomic_inc(&md->open_count);
345 spin_unlock(&_minor_lock);
347 return md ? 0 : -ENXIO;
350 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
352 struct mapped_device *md = disk->private_data;
353 atomic_dec(&md->open_count);
358 int dm_open_count(struct mapped_device *md)
360 return atomic_read(&md->open_count);
364 * Guarantees nothing is using the device before it's deleted.
366 int dm_lock_for_deletion(struct mapped_device *md)
370 spin_lock(&_minor_lock);
372 if (dm_open_count(md))
375 set_bit(DMF_DELETING, &md->flags);
377 spin_unlock(&_minor_lock);
382 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
384 struct mapped_device *md = bdev->bd_disk->private_data;
386 return dm_get_geometry(md, geo);
389 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
390 unsigned int cmd, unsigned long arg)
392 struct mapped_device *md = bdev->bd_disk->private_data;
393 struct dm_table *map = dm_get_table(md);
394 struct dm_target *tgt;
397 if (!map || !dm_table_get_size(map))
400 /* We only support devices that have a single target */
401 if (dm_table_get_num_targets(map) != 1)
404 tgt = dm_table_get_target(map, 0);
406 if (dm_suspended(md)) {
411 if (tgt->type->ioctl)
412 r = tgt->type->ioctl(tgt, cmd, arg);
420 static struct dm_io *alloc_io(struct mapped_device *md)
422 return mempool_alloc(md->io_pool, GFP_NOIO);
425 static void free_io(struct mapped_device *md, struct dm_io *io)
427 mempool_free(io, md->io_pool);
430 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
432 mempool_free(tio, md->tio_pool);
435 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md)
437 return mempool_alloc(md->tio_pool, GFP_ATOMIC);
440 static void free_rq_tio(struct dm_rq_target_io *tio)
442 mempool_free(tio, tio->md->tio_pool);
445 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
447 return mempool_alloc(md->io_pool, GFP_ATOMIC);
450 static void free_bio_info(struct dm_rq_clone_bio_info *info)
452 mempool_free(info, info->tio->md->io_pool);
455 static void start_io_acct(struct dm_io *io)
457 struct mapped_device *md = io->md;
459 int rw = bio_data_dir(io->bio);
461 io->start_time = jiffies;
463 cpu = part_stat_lock();
464 part_round_stats(cpu, &dm_disk(md)->part0);
466 dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]);
469 static void end_io_acct(struct dm_io *io)
471 struct mapped_device *md = io->md;
472 struct bio *bio = io->bio;
473 unsigned long duration = jiffies - io->start_time;
475 int rw = bio_data_dir(bio);
477 cpu = part_stat_lock();
478 part_round_stats(cpu, &dm_disk(md)->part0);
479 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
483 * After this is decremented the bio must not be touched if it is
486 dm_disk(md)->part0.in_flight[rw] = pending =
487 atomic_dec_return(&md->pending[rw]);
488 pending += atomic_read(&md->pending[rw^0x1]);
490 /* nudge anyone waiting on suspend queue */
496 * Add the bio to the list of deferred io.
498 static void queue_io(struct mapped_device *md, struct bio *bio)
500 down_write(&md->io_lock);
502 spin_lock_irq(&md->deferred_lock);
503 bio_list_add(&md->deferred, bio);
504 spin_unlock_irq(&md->deferred_lock);
506 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
507 queue_work(md->wq, &md->work);
509 up_write(&md->io_lock);
513 * Everyone (including functions in this file), should use this
514 * function to access the md->map field, and make sure they call
515 * dm_table_put() when finished.
517 struct dm_table *dm_get_table(struct mapped_device *md)
522 read_lock_irqsave(&md->map_lock, flags);
526 read_unlock_irqrestore(&md->map_lock, flags);
532 * Get the geometry associated with a dm device
534 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
542 * Set the geometry of a device.
544 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
546 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
548 if (geo->start > sz) {
549 DMWARN("Start sector is beyond the geometry limits.");
558 /*-----------------------------------------------------------------
560 * A more elegant soln is in the works that uses the queue
561 * merge fn, unfortunately there are a couple of changes to
562 * the block layer that I want to make for this. So in the
563 * interests of getting something for people to use I give
564 * you this clearly demarcated crap.
565 *---------------------------------------------------------------*/
567 static int __noflush_suspending(struct mapped_device *md)
569 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
573 * Decrements the number of outstanding ios that a bio has been
574 * cloned into, completing the original io if necc.
576 static void dec_pending(struct dm_io *io, int error)
581 struct mapped_device *md = io->md;
583 /* Push-back supersedes any I/O errors */
584 if (unlikely(error)) {
585 spin_lock_irqsave(&io->endio_lock, flags);
586 if (!(io->error > 0 && __noflush_suspending(md)))
588 spin_unlock_irqrestore(&io->endio_lock, flags);
591 if (atomic_dec_and_test(&io->io_count)) {
592 if (io->error == DM_ENDIO_REQUEUE) {
594 * Target requested pushing back the I/O.
596 spin_lock_irqsave(&md->deferred_lock, flags);
597 if (__noflush_suspending(md)) {
598 if (!bio_rw_flagged(io->bio, BIO_RW_BARRIER))
599 bio_list_add_head(&md->deferred,
602 /* noflush suspend was interrupted. */
604 spin_unlock_irqrestore(&md->deferred_lock, flags);
607 io_error = io->error;
610 if (bio_rw_flagged(bio, BIO_RW_BARRIER)) {
612 * There can be just one barrier request so we use
613 * a per-device variable for error reporting.
614 * Note that you can't touch the bio after end_io_acct
616 if (!md->barrier_error && io_error != -EOPNOTSUPP)
617 md->barrier_error = io_error;
622 if (io_error != DM_ENDIO_REQUEUE) {
623 trace_block_bio_complete(md->queue, bio);
625 bio_endio(bio, io_error);
633 static void clone_endio(struct bio *bio, int error)
636 struct dm_target_io *tio = bio->bi_private;
637 struct dm_io *io = tio->io;
638 struct mapped_device *md = tio->io->md;
639 dm_endio_fn endio = tio->ti->type->end_io;
641 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
645 r = endio(tio->ti, bio, error, &tio->info);
646 if (r < 0 || r == DM_ENDIO_REQUEUE)
648 * error and requeue request are handled
652 else if (r == DM_ENDIO_INCOMPLETE)
653 /* The target will handle the io */
656 DMWARN("unimplemented target endio return value: %d", r);
662 * Store md for cleanup instead of tio which is about to get freed.
664 bio->bi_private = md->bs;
668 dec_pending(io, error);
672 * Partial completion handling for request-based dm
674 static void end_clone_bio(struct bio *clone, int error)
676 struct dm_rq_clone_bio_info *info = clone->bi_private;
677 struct dm_rq_target_io *tio = info->tio;
678 struct bio *bio = info->orig;
679 unsigned int nr_bytes = info->orig->bi_size;
685 * An error has already been detected on the request.
686 * Once error occurred, just let clone->end_io() handle
692 * Don't notice the error to the upper layer yet.
693 * The error handling decision is made by the target driver,
694 * when the request is completed.
701 * I/O for the bio successfully completed.
702 * Notice the data completion to the upper layer.
706 * bios are processed from the head of the list.
707 * So the completing bio should always be rq->bio.
708 * If it's not, something wrong is happening.
710 if (tio->orig->bio != bio)
711 DMERR("bio completion is going in the middle of the request");
714 * Update the original request.
715 * Do not use blk_end_request() here, because it may complete
716 * the original request before the clone, and break the ordering.
718 blk_update_request(tio->orig, 0, nr_bytes);
722 * Don't touch any member of the md after calling this function because
723 * the md may be freed in dm_put() at the end of this function.
724 * Or do dm_get() before calling this function and dm_put() later.
726 static void rq_completed(struct mapped_device *md, int run_queue)
728 int wakeup_waiters = 0;
729 struct request_queue *q = md->queue;
732 spin_lock_irqsave(q->queue_lock, flags);
733 if (!queue_in_flight(q))
735 spin_unlock_irqrestore(q->queue_lock, flags);
737 /* nudge anyone waiting on suspend queue */
745 * dm_put() must be at the end of this function. See the comment above
750 static void free_rq_clone(struct request *clone)
752 struct dm_rq_target_io *tio = clone->end_io_data;
754 blk_rq_unprep_clone(clone);
758 static void dm_unprep_request(struct request *rq)
760 struct request *clone = rq->special;
763 rq->cmd_flags &= ~REQ_DONTPREP;
765 free_rq_clone(clone);
769 * Requeue the original request of a clone.
771 void dm_requeue_unmapped_request(struct request *clone)
773 struct dm_rq_target_io *tio = clone->end_io_data;
774 struct mapped_device *md = tio->md;
775 struct request *rq = tio->orig;
776 struct request_queue *q = rq->q;
779 dm_unprep_request(rq);
781 spin_lock_irqsave(q->queue_lock, flags);
782 if (elv_queue_empty(q))
784 blk_requeue_request(q, rq);
785 spin_unlock_irqrestore(q->queue_lock, flags);
789 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
791 static void __stop_queue(struct request_queue *q)
796 static void stop_queue(struct request_queue *q)
800 spin_lock_irqsave(q->queue_lock, flags);
802 spin_unlock_irqrestore(q->queue_lock, flags);
805 static void __start_queue(struct request_queue *q)
807 if (blk_queue_stopped(q))
811 static void start_queue(struct request_queue *q)
815 spin_lock_irqsave(q->queue_lock, flags);
817 spin_unlock_irqrestore(q->queue_lock, flags);
821 * Complete the clone and the original request.
822 * Must be called without queue lock.
824 static void dm_end_request(struct request *clone, int error)
826 struct dm_rq_target_io *tio = clone->end_io_data;
827 struct mapped_device *md = tio->md;
828 struct request *rq = tio->orig;
830 if (blk_pc_request(rq)) {
831 rq->errors = clone->errors;
832 rq->resid_len = clone->resid_len;
836 * We are using the sense buffer of the original
838 * So setting the length of the sense data is enough.
840 rq->sense_len = clone->sense_len;
843 free_rq_clone(clone);
845 blk_end_request_all(rq, error);
851 * Request completion handler for request-based dm
853 static void dm_softirq_done(struct request *rq)
855 struct request *clone = rq->completion_data;
856 struct dm_rq_target_io *tio = clone->end_io_data;
857 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
858 int error = tio->error;
860 if (!(rq->cmd_flags & REQ_FAILED) && rq_end_io)
861 error = rq_end_io(tio->ti, clone, error, &tio->info);
864 /* The target wants to complete the I/O */
865 dm_end_request(clone, error);
866 else if (error == DM_ENDIO_INCOMPLETE)
867 /* The target will handle the I/O */
869 else if (error == DM_ENDIO_REQUEUE)
870 /* The target wants to requeue the I/O */
871 dm_requeue_unmapped_request(clone);
873 DMWARN("unimplemented target endio return value: %d", error);
879 * Complete the clone and the original request with the error status
880 * through softirq context.
882 static void dm_complete_request(struct request *clone, int error)
884 struct dm_rq_target_io *tio = clone->end_io_data;
885 struct request *rq = tio->orig;
888 rq->completion_data = clone;
889 blk_complete_request(rq);
893 * Complete the not-mapped clone and the original request with the error status
894 * through softirq context.
895 * Target's rq_end_io() function isn't called.
896 * This may be used when the target's map_rq() function fails.
898 void dm_kill_unmapped_request(struct request *clone, int error)
900 struct dm_rq_target_io *tio = clone->end_io_data;
901 struct request *rq = tio->orig;
903 rq->cmd_flags |= REQ_FAILED;
904 dm_complete_request(clone, error);
906 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
909 * Called with the queue lock held
911 static void end_clone_request(struct request *clone, int error)
914 * For just cleaning up the information of the queue in which
915 * the clone was dispatched.
916 * The clone is *NOT* freed actually here because it is alloced from
917 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
919 __blk_put_request(clone->q, clone);
922 * Actual request completion is done in a softirq context which doesn't
923 * hold the queue lock. Otherwise, deadlock could occur because:
924 * - another request may be submitted by the upper level driver
925 * of the stacking during the completion
926 * - the submission which requires queue lock may be done
929 dm_complete_request(clone, error);
932 static sector_t max_io_len(struct mapped_device *md,
933 sector_t sector, struct dm_target *ti)
935 sector_t offset = sector - ti->begin;
936 sector_t len = ti->len - offset;
939 * Does the target need to split even further ?
943 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
952 static void __map_bio(struct dm_target *ti, struct bio *clone,
953 struct dm_target_io *tio)
957 struct mapped_device *md;
959 clone->bi_end_io = clone_endio;
960 clone->bi_private = tio;
963 * Map the clone. If r == 0 we don't need to do
964 * anything, the target has assumed ownership of
967 atomic_inc(&tio->io->io_count);
968 sector = clone->bi_sector;
969 r = ti->type->map(ti, clone, &tio->info);
970 if (r == DM_MAPIO_REMAPPED) {
971 /* the bio has been remapped so dispatch it */
973 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
974 tio->io->bio->bi_bdev->bd_dev, sector);
976 generic_make_request(clone);
977 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
978 /* error the io and bail out, or requeue it if needed */
980 dec_pending(tio->io, r);
982 * Store bio_set for cleanup.
984 clone->bi_private = md->bs;
988 DMWARN("unimplemented target map return value: %d", r);
994 struct mapped_device *md;
995 struct dm_table *map;
999 sector_t sector_count;
1003 static void dm_bio_destructor(struct bio *bio)
1005 struct bio_set *bs = bio->bi_private;
1011 * Creates a little bio that is just does part of a bvec.
1013 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1014 unsigned short idx, unsigned int offset,
1015 unsigned int len, struct bio_set *bs)
1018 struct bio_vec *bv = bio->bi_io_vec + idx;
1020 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1021 clone->bi_destructor = dm_bio_destructor;
1022 *clone->bi_io_vec = *bv;
1024 clone->bi_sector = sector;
1025 clone->bi_bdev = bio->bi_bdev;
1026 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
1028 clone->bi_size = to_bytes(len);
1029 clone->bi_io_vec->bv_offset = offset;
1030 clone->bi_io_vec->bv_len = clone->bi_size;
1031 clone->bi_flags |= 1 << BIO_CLONED;
1033 if (bio_integrity(bio)) {
1034 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1035 bio_integrity_trim(clone,
1036 bio_sector_offset(bio, idx, offset), len);
1043 * Creates a bio that consists of range of complete bvecs.
1045 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1046 unsigned short idx, unsigned short bv_count,
1047 unsigned int len, struct bio_set *bs)
1051 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1052 __bio_clone(clone, bio);
1053 clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
1054 clone->bi_destructor = dm_bio_destructor;
1055 clone->bi_sector = sector;
1056 clone->bi_idx = idx;
1057 clone->bi_vcnt = idx + bv_count;
1058 clone->bi_size = to_bytes(len);
1059 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1061 if (bio_integrity(bio)) {
1062 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1064 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1065 bio_integrity_trim(clone,
1066 bio_sector_offset(bio, idx, 0), len);
1072 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1073 struct dm_target *ti)
1075 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1079 memset(&tio->info, 0, sizeof(tio->info));
1084 static void __flush_target(struct clone_info *ci, struct dm_target *ti,
1087 struct dm_target_io *tio = alloc_tio(ci, ti);
1090 tio->info.flush_request = flush_nr;
1092 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1093 __bio_clone(clone, ci->bio);
1094 clone->bi_destructor = dm_bio_destructor;
1096 __map_bio(ti, clone, tio);
1099 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1101 unsigned target_nr = 0, flush_nr;
1102 struct dm_target *ti;
1104 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1105 for (flush_nr = 0; flush_nr < ti->num_flush_requests;
1107 __flush_target(ci, ti, flush_nr);
1109 ci->sector_count = 0;
1114 static int __clone_and_map(struct clone_info *ci)
1116 struct bio *clone, *bio = ci->bio;
1117 struct dm_target *ti;
1118 sector_t len = 0, max;
1119 struct dm_target_io *tio;
1121 if (unlikely(bio_empty_barrier(bio)))
1122 return __clone_and_map_empty_barrier(ci);
1124 ti = dm_table_find_target(ci->map, ci->sector);
1125 if (!dm_target_is_valid(ti))
1128 max = max_io_len(ci->md, ci->sector, ti);
1131 * Allocate a target io object.
1133 tio = alloc_tio(ci, ti);
1135 if (ci->sector_count <= max) {
1137 * Optimise for the simple case where we can do all of
1138 * the remaining io with a single clone.
1140 clone = clone_bio(bio, ci->sector, ci->idx,
1141 bio->bi_vcnt - ci->idx, ci->sector_count,
1143 __map_bio(ti, clone, tio);
1144 ci->sector_count = 0;
1146 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1148 * There are some bvecs that don't span targets.
1149 * Do as many of these as possible.
1152 sector_t remaining = max;
1155 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1156 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1158 if (bv_len > remaining)
1161 remaining -= bv_len;
1165 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1167 __map_bio(ti, clone, tio);
1170 ci->sector_count -= len;
1175 * Handle a bvec that must be split between two or more targets.
1177 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1178 sector_t remaining = to_sector(bv->bv_len);
1179 unsigned int offset = 0;
1183 ti = dm_table_find_target(ci->map, ci->sector);
1184 if (!dm_target_is_valid(ti))
1187 max = max_io_len(ci->md, ci->sector, ti);
1189 tio = alloc_tio(ci, ti);
1192 len = min(remaining, max);
1194 clone = split_bvec(bio, ci->sector, ci->idx,
1195 bv->bv_offset + offset, len,
1198 __map_bio(ti, clone, tio);
1201 ci->sector_count -= len;
1202 offset += to_bytes(len);
1203 } while (remaining -= len);
1212 * Split the bio into several clones and submit it to targets.
1214 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1216 struct clone_info ci;
1219 ci.map = dm_get_table(md);
1220 if (unlikely(!ci.map)) {
1221 if (!bio_rw_flagged(bio, BIO_RW_BARRIER))
1224 if (!md->barrier_error)
1225 md->barrier_error = -EIO;
1231 ci.io = alloc_io(md);
1233 atomic_set(&ci.io->io_count, 1);
1236 spin_lock_init(&ci.io->endio_lock);
1237 ci.sector = bio->bi_sector;
1238 ci.sector_count = bio_sectors(bio);
1239 if (unlikely(bio_empty_barrier(bio)))
1240 ci.sector_count = 1;
1241 ci.idx = bio->bi_idx;
1243 start_io_acct(ci.io);
1244 while (ci.sector_count && !error)
1245 error = __clone_and_map(&ci);
1247 /* drop the extra reference count */
1248 dec_pending(ci.io, error);
1249 dm_table_put(ci.map);
1251 /*-----------------------------------------------------------------
1253 *---------------------------------------------------------------*/
1255 static int dm_merge_bvec(struct request_queue *q,
1256 struct bvec_merge_data *bvm,
1257 struct bio_vec *biovec)
1259 struct mapped_device *md = q->queuedata;
1260 struct dm_table *map = dm_get_table(md);
1261 struct dm_target *ti;
1262 sector_t max_sectors;
1268 ti = dm_table_find_target(map, bvm->bi_sector);
1269 if (!dm_target_is_valid(ti))
1273 * Find maximum amount of I/O that won't need splitting
1275 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
1276 (sector_t) BIO_MAX_SECTORS);
1277 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1282 * merge_bvec_fn() returns number of bytes
1283 * it can accept at this offset
1284 * max is precomputed maximal io size
1286 if (max_size && ti->type->merge)
1287 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1289 * If the target doesn't support merge method and some of the devices
1290 * provided their merge_bvec method (we know this by looking at
1291 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1292 * entries. So always set max_size to 0, and the code below allows
1295 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1304 * Always allow an entire first page
1306 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1307 max_size = biovec->bv_len;
1313 * The request function that just remaps the bio built up by
1316 static int _dm_request(struct request_queue *q, struct bio *bio)
1318 int rw = bio_data_dir(bio);
1319 struct mapped_device *md = q->queuedata;
1322 down_read(&md->io_lock);
1324 cpu = part_stat_lock();
1325 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1326 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1330 * If we're suspended or the thread is processing barriers
1331 * we have to queue this io for later.
1333 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1334 unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
1335 up_read(&md->io_lock);
1337 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1338 bio_rw(bio) == READA) {
1348 __split_and_process_bio(md, bio);
1349 up_read(&md->io_lock);
1353 static int dm_make_request(struct request_queue *q, struct bio *bio)
1355 struct mapped_device *md = q->queuedata;
1357 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
1358 bio_endio(bio, -EOPNOTSUPP);
1362 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1365 static int dm_request_based(struct mapped_device *md)
1367 return blk_queue_stackable(md->queue);
1370 static int dm_request(struct request_queue *q, struct bio *bio)
1372 struct mapped_device *md = q->queuedata;
1374 if (dm_request_based(md))
1375 return dm_make_request(q, bio);
1377 return _dm_request(q, bio);
1380 void dm_dispatch_request(struct request *rq)
1384 if (blk_queue_io_stat(rq->q))
1385 rq->cmd_flags |= REQ_IO_STAT;
1387 rq->start_time = jiffies;
1388 r = blk_insert_cloned_request(rq->q, rq);
1390 dm_complete_request(rq, r);
1392 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1394 static void dm_rq_bio_destructor(struct bio *bio)
1396 struct dm_rq_clone_bio_info *info = bio->bi_private;
1397 struct mapped_device *md = info->tio->md;
1399 free_bio_info(info);
1400 bio_free(bio, md->bs);
1403 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1406 struct dm_rq_target_io *tio = data;
1407 struct mapped_device *md = tio->md;
1408 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1413 info->orig = bio_orig;
1415 bio->bi_end_io = end_clone_bio;
1416 bio->bi_private = info;
1417 bio->bi_destructor = dm_rq_bio_destructor;
1422 static int setup_clone(struct request *clone, struct request *rq,
1423 struct dm_rq_target_io *tio)
1425 int r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1426 dm_rq_bio_constructor, tio);
1431 clone->cmd = rq->cmd;
1432 clone->cmd_len = rq->cmd_len;
1433 clone->sense = rq->sense;
1434 clone->buffer = rq->buffer;
1435 clone->end_io = end_clone_request;
1436 clone->end_io_data = tio;
1441 static int dm_rq_flush_suspending(struct mapped_device *md)
1443 return !md->suspend_rq.special;
1447 * Called with the queue lock held.
1449 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1451 struct mapped_device *md = q->queuedata;
1452 struct dm_rq_target_io *tio;
1453 struct request *clone;
1455 if (unlikely(rq == &md->suspend_rq)) {
1456 if (dm_rq_flush_suspending(md))
1459 /* The flush suspend was interrupted */
1460 return BLKPREP_KILL;
1463 if (unlikely(rq->special)) {
1464 DMWARN("Already has something in rq->special.");
1465 return BLKPREP_KILL;
1468 tio = alloc_rq_tio(md); /* Only one for each original request */
1471 return BLKPREP_DEFER;
1477 memset(&tio->info, 0, sizeof(tio->info));
1479 clone = &tio->clone;
1480 if (setup_clone(clone, rq, tio)) {
1483 return BLKPREP_DEFER;
1486 rq->special = clone;
1487 rq->cmd_flags |= REQ_DONTPREP;
1492 static void map_request(struct dm_target *ti, struct request *rq,
1493 struct mapped_device *md)
1496 struct request *clone = rq->special;
1497 struct dm_rq_target_io *tio = clone->end_io_data;
1500 * Hold the md reference here for the in-flight I/O.
1501 * We can't rely on the reference count by device opener,
1502 * because the device may be closed during the request completion
1503 * when all bios are completed.
1504 * See the comment in rq_completed() too.
1509 r = ti->type->map_rq(ti, clone, &tio->info);
1511 case DM_MAPIO_SUBMITTED:
1512 /* The target has taken the I/O to submit by itself later */
1514 case DM_MAPIO_REMAPPED:
1515 /* The target has remapped the I/O so dispatch it */
1516 dm_dispatch_request(clone);
1518 case DM_MAPIO_REQUEUE:
1519 /* The target wants to requeue the I/O */
1520 dm_requeue_unmapped_request(clone);
1524 DMWARN("unimplemented target map return value: %d", r);
1528 /* The target wants to complete the I/O */
1529 dm_kill_unmapped_request(clone, r);
1535 * q->request_fn for request-based dm.
1536 * Called with the queue lock held.
1538 static void dm_request_fn(struct request_queue *q)
1540 struct mapped_device *md = q->queuedata;
1541 struct dm_table *map = dm_get_table(md);
1542 struct dm_target *ti;
1546 * For noflush suspend, check blk_queue_stopped() to immediately
1547 * quit I/O dispatching.
1549 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1550 rq = blk_peek_request(q);
1554 if (unlikely(rq == &md->suspend_rq)) { /* Flush suspend maker */
1555 if (queue_in_flight(q))
1556 /* Not quiet yet. Wait more */
1559 /* This device should be quiet now */
1561 blk_start_request(rq);
1562 __blk_end_request_all(rq, 0);
1567 ti = dm_table_find_target(map, blk_rq_pos(rq));
1568 if (ti->type->busy && ti->type->busy(ti))
1571 blk_start_request(rq);
1572 spin_unlock(q->queue_lock);
1573 map_request(ti, rq, md);
1574 spin_lock_irq(q->queue_lock);
1580 if (!elv_queue_empty(q))
1581 /* Some requests still remain, retry later */
1590 int dm_underlying_device_busy(struct request_queue *q)
1592 return blk_lld_busy(q);
1594 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1596 static int dm_lld_busy(struct request_queue *q)
1599 struct mapped_device *md = q->queuedata;
1600 struct dm_table *map = dm_get_table(md);
1602 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1605 r = dm_table_any_busy_target(map);
1612 static void dm_unplug_all(struct request_queue *q)
1614 struct mapped_device *md = q->queuedata;
1615 struct dm_table *map = dm_get_table(md);
1618 if (dm_request_based(md))
1619 generic_unplug_device(q);
1621 dm_table_unplug_all(map);
1626 static int dm_any_congested(void *congested_data, int bdi_bits)
1629 struct mapped_device *md = congested_data;
1630 struct dm_table *map;
1632 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1633 map = dm_get_table(md);
1636 * Request-based dm cares about only own queue for
1637 * the query about congestion status of request_queue
1639 if (dm_request_based(md))
1640 r = md->queue->backing_dev_info.state &
1643 r = dm_table_any_congested(map, bdi_bits);
1652 /*-----------------------------------------------------------------
1653 * An IDR is used to keep track of allocated minor numbers.
1654 *---------------------------------------------------------------*/
1655 static DEFINE_IDR(_minor_idr);
1657 static void free_minor(int minor)
1659 spin_lock(&_minor_lock);
1660 idr_remove(&_minor_idr, minor);
1661 spin_unlock(&_minor_lock);
1665 * See if the device with a specific minor # is free.
1667 static int specific_minor(int minor)
1671 if (minor >= (1 << MINORBITS))
1674 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1678 spin_lock(&_minor_lock);
1680 if (idr_find(&_minor_idr, minor)) {
1685 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1690 idr_remove(&_minor_idr, m);
1696 spin_unlock(&_minor_lock);
1700 static int next_free_minor(int *minor)
1704 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1708 spin_lock(&_minor_lock);
1710 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1714 if (m >= (1 << MINORBITS)) {
1715 idr_remove(&_minor_idr, m);
1723 spin_unlock(&_minor_lock);
1727 static const struct block_device_operations dm_blk_dops;
1729 static void dm_wq_work(struct work_struct *work);
1732 * Allocate and initialise a blank device with a given minor.
1734 static struct mapped_device *alloc_dev(int minor)
1737 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1741 DMWARN("unable to allocate device, out of memory.");
1745 if (!try_module_get(THIS_MODULE))
1746 goto bad_module_get;
1748 /* get a minor number for the dev */
1749 if (minor == DM_ANY_MINOR)
1750 r = next_free_minor(&minor);
1752 r = specific_minor(minor);
1756 init_rwsem(&md->io_lock);
1757 mutex_init(&md->suspend_lock);
1758 spin_lock_init(&md->deferred_lock);
1759 rwlock_init(&md->map_lock);
1760 atomic_set(&md->holders, 1);
1761 atomic_set(&md->open_count, 0);
1762 atomic_set(&md->event_nr, 0);
1763 atomic_set(&md->uevent_seq, 0);
1764 INIT_LIST_HEAD(&md->uevent_list);
1765 spin_lock_init(&md->uevent_lock);
1767 md->queue = blk_init_queue(dm_request_fn, NULL);
1772 * Request-based dm devices cannot be stacked on top of bio-based dm
1773 * devices. The type of this dm device has not been decided yet,
1774 * although we initialized the queue using blk_init_queue().
1775 * The type is decided at the first table loading time.
1776 * To prevent problematic device stacking, clear the queue flag
1777 * for request stacking support until then.
1779 * This queue is new, so no concurrency on the queue_flags.
1781 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1782 md->saved_make_request_fn = md->queue->make_request_fn;
1783 md->queue->queuedata = md;
1784 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1785 md->queue->backing_dev_info.congested_data = md;
1786 blk_queue_make_request(md->queue, dm_request);
1787 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1788 md->queue->unplug_fn = dm_unplug_all;
1789 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1790 blk_queue_softirq_done(md->queue, dm_softirq_done);
1791 blk_queue_prep_rq(md->queue, dm_prep_fn);
1792 blk_queue_lld_busy(md->queue, dm_lld_busy);
1794 md->disk = alloc_disk(1);
1798 atomic_set(&md->pending[0], 0);
1799 atomic_set(&md->pending[1], 0);
1800 init_waitqueue_head(&md->wait);
1801 INIT_WORK(&md->work, dm_wq_work);
1802 init_waitqueue_head(&md->eventq);
1804 md->disk->major = _major;
1805 md->disk->first_minor = minor;
1806 md->disk->fops = &dm_blk_dops;
1807 md->disk->queue = md->queue;
1808 md->disk->private_data = md;
1809 sprintf(md->disk->disk_name, "dm-%d", minor);
1811 format_dev_t(md->name, MKDEV(_major, minor));
1813 md->wq = create_singlethread_workqueue("kdmflush");
1817 md->bdev = bdget_disk(md->disk, 0);
1821 /* Populate the mapping, nobody knows we exist yet */
1822 spin_lock(&_minor_lock);
1823 old_md = idr_replace(&_minor_idr, md, minor);
1824 spin_unlock(&_minor_lock);
1826 BUG_ON(old_md != MINOR_ALLOCED);
1831 destroy_workqueue(md->wq);
1833 del_gendisk(md->disk);
1836 blk_cleanup_queue(md->queue);
1840 module_put(THIS_MODULE);
1846 static void unlock_fs(struct mapped_device *md);
1848 static void free_dev(struct mapped_device *md)
1850 int minor = MINOR(disk_devt(md->disk));
1854 destroy_workqueue(md->wq);
1856 mempool_destroy(md->tio_pool);
1858 mempool_destroy(md->io_pool);
1860 bioset_free(md->bs);
1861 blk_integrity_unregister(md->disk);
1862 del_gendisk(md->disk);
1865 spin_lock(&_minor_lock);
1866 md->disk->private_data = NULL;
1867 spin_unlock(&_minor_lock);
1870 blk_cleanup_queue(md->queue);
1871 module_put(THIS_MODULE);
1875 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1877 struct dm_md_mempools *p;
1879 if (md->io_pool && md->tio_pool && md->bs)
1880 /* the md already has necessary mempools */
1883 p = dm_table_get_md_mempools(t);
1884 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1886 md->io_pool = p->io_pool;
1888 md->tio_pool = p->tio_pool;
1894 /* mempool bind completed, now no need any mempools in the table */
1895 dm_table_free_md_mempools(t);
1899 * Bind a table to the device.
1901 static void event_callback(void *context)
1903 unsigned long flags;
1905 struct mapped_device *md = (struct mapped_device *) context;
1907 spin_lock_irqsave(&md->uevent_lock, flags);
1908 list_splice_init(&md->uevent_list, &uevents);
1909 spin_unlock_irqrestore(&md->uevent_lock, flags);
1911 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1913 atomic_inc(&md->event_nr);
1914 wake_up(&md->eventq);
1917 static void __set_size(struct mapped_device *md, sector_t size)
1919 set_capacity(md->disk, size);
1921 mutex_lock(&md->bdev->bd_inode->i_mutex);
1922 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1923 mutex_unlock(&md->bdev->bd_inode->i_mutex);
1926 static int __bind(struct mapped_device *md, struct dm_table *t,
1927 struct queue_limits *limits)
1929 struct request_queue *q = md->queue;
1931 unsigned long flags;
1933 size = dm_table_get_size(t);
1936 * Wipe any geometry if the size of the table changed.
1938 if (size != get_capacity(md->disk))
1939 memset(&md->geometry, 0, sizeof(md->geometry));
1941 __set_size(md, size);
1944 dm_table_destroy(t);
1948 dm_table_event_callback(t, event_callback, md);
1951 * The queue hasn't been stopped yet, if the old table type wasn't
1952 * for request-based during suspension. So stop it to prevent
1953 * I/O mapping before resume.
1954 * This must be done before setting the queue restrictions,
1955 * because request-based dm may be run just after the setting.
1957 if (dm_table_request_based(t) && !blk_queue_stopped(q))
1960 __bind_mempools(md, t);
1962 write_lock_irqsave(&md->map_lock, flags);
1964 dm_table_set_restrictions(t, q, limits);
1965 write_unlock_irqrestore(&md->map_lock, flags);
1970 static void __unbind(struct mapped_device *md)
1972 struct dm_table *map = md->map;
1973 unsigned long flags;
1978 dm_table_event_callback(map, NULL, NULL);
1979 write_lock_irqsave(&md->map_lock, flags);
1981 write_unlock_irqrestore(&md->map_lock, flags);
1982 dm_table_destroy(map);
1986 * Constructor for a new device.
1988 int dm_create(int minor, struct mapped_device **result)
1990 struct mapped_device *md;
1992 md = alloc_dev(minor);
2002 static struct mapped_device *dm_find_md(dev_t dev)
2004 struct mapped_device *md;
2005 unsigned minor = MINOR(dev);
2007 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2010 spin_lock(&_minor_lock);
2012 md = idr_find(&_minor_idr, minor);
2013 if (md && (md == MINOR_ALLOCED ||
2014 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2015 test_bit(DMF_FREEING, &md->flags))) {
2021 spin_unlock(&_minor_lock);
2026 struct mapped_device *dm_get_md(dev_t dev)
2028 struct mapped_device *md = dm_find_md(dev);
2036 void *dm_get_mdptr(struct mapped_device *md)
2038 return md->interface_ptr;
2041 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2043 md->interface_ptr = ptr;
2046 void dm_get(struct mapped_device *md)
2048 atomic_inc(&md->holders);
2051 const char *dm_device_name(struct mapped_device *md)
2055 EXPORT_SYMBOL_GPL(dm_device_name);
2057 void dm_put(struct mapped_device *md)
2059 struct dm_table *map;
2061 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2063 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
2064 map = dm_get_table(md);
2065 idr_replace(&_minor_idr, MINOR_ALLOCED,
2066 MINOR(disk_devt(dm_disk(md))));
2067 set_bit(DMF_FREEING, &md->flags);
2068 spin_unlock(&_minor_lock);
2069 if (!dm_suspended(md)) {
2070 dm_table_presuspend_targets(map);
2071 dm_table_postsuspend_targets(map);
2079 EXPORT_SYMBOL_GPL(dm_put);
2081 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2084 DECLARE_WAITQUEUE(wait, current);
2085 struct request_queue *q = md->queue;
2086 unsigned long flags;
2088 dm_unplug_all(md->queue);
2090 add_wait_queue(&md->wait, &wait);
2093 set_current_state(interruptible);
2096 if (dm_request_based(md)) {
2097 spin_lock_irqsave(q->queue_lock, flags);
2098 if (!queue_in_flight(q) && blk_queue_stopped(q)) {
2099 spin_unlock_irqrestore(q->queue_lock, flags);
2102 spin_unlock_irqrestore(q->queue_lock, flags);
2103 } else if (!atomic_read(&md->pending[0]) &&
2104 !atomic_read(&md->pending[1]))
2107 if (interruptible == TASK_INTERRUPTIBLE &&
2108 signal_pending(current)) {
2115 set_current_state(TASK_RUNNING);
2117 remove_wait_queue(&md->wait, &wait);
2122 static void dm_flush(struct mapped_device *md)
2124 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2126 bio_init(&md->barrier_bio);
2127 md->barrier_bio.bi_bdev = md->bdev;
2128 md->barrier_bio.bi_rw = WRITE_BARRIER;
2129 __split_and_process_bio(md, &md->barrier_bio);
2131 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2134 static void process_barrier(struct mapped_device *md, struct bio *bio)
2136 md->barrier_error = 0;
2140 if (!bio_empty_barrier(bio)) {
2141 __split_and_process_bio(md, bio);
2145 if (md->barrier_error != DM_ENDIO_REQUEUE)
2146 bio_endio(bio, md->barrier_error);
2148 spin_lock_irq(&md->deferred_lock);
2149 bio_list_add_head(&md->deferred, bio);
2150 spin_unlock_irq(&md->deferred_lock);
2155 * Process the deferred bios
2157 static void dm_wq_work(struct work_struct *work)
2159 struct mapped_device *md = container_of(work, struct mapped_device,
2163 down_write(&md->io_lock);
2165 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2166 spin_lock_irq(&md->deferred_lock);
2167 c = bio_list_pop(&md->deferred);
2168 spin_unlock_irq(&md->deferred_lock);
2171 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2175 up_write(&md->io_lock);
2177 if (dm_request_based(md))
2178 generic_make_request(c);
2180 if (bio_rw_flagged(c, BIO_RW_BARRIER))
2181 process_barrier(md, c);
2183 __split_and_process_bio(md, c);
2186 down_write(&md->io_lock);
2189 up_write(&md->io_lock);
2192 static void dm_queue_flush(struct mapped_device *md)
2194 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2195 smp_mb__after_clear_bit();
2196 queue_work(md->wq, &md->work);
2200 * Swap in a new table (destroying old one).
2202 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
2204 struct queue_limits limits;
2207 mutex_lock(&md->suspend_lock);
2209 /* device must be suspended */
2210 if (!dm_suspended(md))
2213 r = dm_calculate_queue_limits(table, &limits);
2217 /* cannot change the device type, once a table is bound */
2219 (dm_table_get_type(md->map) != dm_table_get_type(table))) {
2220 DMWARN("can't change the device type after a table is bound");
2225 r = __bind(md, table, &limits);
2228 mutex_unlock(&md->suspend_lock);
2232 static void dm_rq_invalidate_suspend_marker(struct mapped_device *md)
2234 md->suspend_rq.special = (void *)0x1;
2237 static void dm_rq_abort_suspend(struct mapped_device *md, int noflush)
2239 struct request_queue *q = md->queue;
2240 unsigned long flags;
2242 spin_lock_irqsave(q->queue_lock, flags);
2244 dm_rq_invalidate_suspend_marker(md);
2246 spin_unlock_irqrestore(q->queue_lock, flags);
2249 static void dm_rq_start_suspend(struct mapped_device *md, int noflush)
2251 struct request *rq = &md->suspend_rq;
2252 struct request_queue *q = md->queue;
2258 blk_insert_request(q, rq, 0, NULL);
2262 static int dm_rq_suspend_available(struct mapped_device *md, int noflush)
2265 struct request *rq = &md->suspend_rq;
2266 struct request_queue *q = md->queue;
2267 unsigned long flags;
2272 /* The marker must be protected by queue lock if it is in use */
2273 spin_lock_irqsave(q->queue_lock, flags);
2274 if (unlikely(rq->ref_count)) {
2276 * This can happen, when the previous flush suspend was
2277 * interrupted, the marker is still in the queue and
2278 * this flush suspend has been invoked, because we don't
2279 * remove the marker at the time of suspend interruption.
2280 * We have only one marker per mapped_device, so we can't
2281 * start another flush suspend while it is in use.
2283 BUG_ON(!rq->special); /* The marker should be invalidated */
2284 DMWARN("Invalidating the previous flush suspend is still in"
2285 " progress. Please retry later.");
2288 spin_unlock_irqrestore(q->queue_lock, flags);
2294 * Functions to lock and unlock any filesystem running on the
2297 static int lock_fs(struct mapped_device *md)
2301 WARN_ON(md->frozen_sb);
2303 md->frozen_sb = freeze_bdev(md->bdev);
2304 if (IS_ERR(md->frozen_sb)) {
2305 r = PTR_ERR(md->frozen_sb);
2306 md->frozen_sb = NULL;
2310 set_bit(DMF_FROZEN, &md->flags);
2315 static void unlock_fs(struct mapped_device *md)
2317 if (!test_bit(DMF_FROZEN, &md->flags))
2320 thaw_bdev(md->bdev, md->frozen_sb);
2321 md->frozen_sb = NULL;
2322 clear_bit(DMF_FROZEN, &md->flags);
2326 * We need to be able to change a mapping table under a mounted
2327 * filesystem. For example we might want to move some data in
2328 * the background. Before the table can be swapped with
2329 * dm_bind_table, dm_suspend must be called to flush any in
2330 * flight bios and ensure that any further io gets deferred.
2333 * Suspend mechanism in request-based dm.
2335 * After the suspend starts, further incoming requests are kept in
2336 * the request_queue and deferred.
2337 * Remaining requests in the request_queue at the start of suspend are flushed
2338 * if it is flush suspend.
2339 * The suspend completes when the following conditions have been satisfied,
2341 * 1. q->in_flight is 0 (which means no in_flight request)
2342 * 2. queue has been stopped (which means no request dispatching)
2347 * Noflush suspend doesn't need to dispatch remaining requests.
2348 * So stop the queue immediately. Then, wait for all in_flight requests
2349 * to be completed or requeued.
2351 * To abort noflush suspend, start the queue.
2356 * Flush suspend needs to dispatch remaining requests. So stop the queue
2357 * after the remaining requests are completed. (Requeued request must be also
2358 * re-dispatched and completed. Until then, we can't stop the queue.)
2360 * During flushing the remaining requests, further incoming requests are also
2361 * inserted to the same queue. To distinguish which requests are to be
2362 * flushed, we insert a marker request to the queue at the time of starting
2363 * flush suspend, like a barrier.
2364 * The dispatching is blocked when the marker is found on the top of the queue.
2365 * And the queue is stopped when all in_flight requests are completed, since
2366 * that means the remaining requests are completely flushed.
2367 * Then, the marker is removed from the queue.
2369 * To abort flush suspend, we also need to take care of the marker, not only
2370 * starting the queue.
2371 * We don't remove the marker forcibly from the queue since it's against
2372 * the block-layer manner. Instead, we put a invalidated mark on the marker.
2373 * When the invalidated marker is found on the top of the queue, it is
2374 * immediately removed from the queue, so it doesn't block dispatching.
2375 * Because we have only one marker per mapped_device, we can't start another
2376 * flush suspend until the invalidated marker is removed from the queue.
2377 * So fail and return with -EBUSY in such a case.
2379 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2381 struct dm_table *map = NULL;
2383 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2384 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2386 mutex_lock(&md->suspend_lock);
2388 if (dm_suspended(md)) {
2393 if (dm_request_based(md) && !dm_rq_suspend_available(md, noflush)) {
2398 map = dm_get_table(md);
2401 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2402 * This flag is cleared before dm_suspend returns.
2405 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2407 /* This does not get reverted if there's an error later. */
2408 dm_table_presuspend_targets(map);
2411 * Flush I/O to the device. noflush supersedes do_lockfs,
2412 * because lock_fs() needs to flush I/Os.
2414 if (!noflush && do_lockfs) {
2421 * Here we must make sure that no processes are submitting requests
2422 * to target drivers i.e. no one may be executing
2423 * __split_and_process_bio. This is called from dm_request and
2426 * To get all processes out of __split_and_process_bio in dm_request,
2427 * we take the write lock. To prevent any process from reentering
2428 * __split_and_process_bio from dm_request, we set
2429 * DMF_QUEUE_IO_TO_THREAD.
2431 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2432 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2433 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2434 * further calls to __split_and_process_bio from dm_wq_work.
2436 down_write(&md->io_lock);
2437 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2438 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2439 up_write(&md->io_lock);
2441 flush_workqueue(md->wq);
2443 if (dm_request_based(md))
2444 dm_rq_start_suspend(md, noflush);
2447 * At this point no more requests are entering target request routines.
2448 * We call dm_wait_for_completion to wait for all existing requests
2451 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2453 down_write(&md->io_lock);
2455 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2456 up_write(&md->io_lock);
2458 /* were we interrupted ? */
2462 if (dm_request_based(md))
2463 dm_rq_abort_suspend(md, noflush);
2466 goto out; /* pushback list is already flushed, so skip flush */
2470 * If dm_wait_for_completion returned 0, the device is completely
2471 * quiescent now. There is no request-processing activity. All new
2472 * requests are being added to md->deferred list.
2475 dm_table_postsuspend_targets(map);
2477 set_bit(DMF_SUSPENDED, &md->flags);
2483 mutex_unlock(&md->suspend_lock);
2487 int dm_resume(struct mapped_device *md)
2490 struct dm_table *map = NULL;
2492 mutex_lock(&md->suspend_lock);
2493 if (!dm_suspended(md))
2496 map = dm_get_table(md);
2497 if (!map || !dm_table_get_size(map))
2500 r = dm_table_resume_targets(map);
2507 * Flushing deferred I/Os must be done after targets are resumed
2508 * so that mapping of targets can work correctly.
2509 * Request-based dm is queueing the deferred I/Os in its request_queue.
2511 if (dm_request_based(md))
2512 start_queue(md->queue);
2516 clear_bit(DMF_SUSPENDED, &md->flags);
2518 dm_table_unplug_all(map);
2522 mutex_unlock(&md->suspend_lock);
2527 /*-----------------------------------------------------------------
2528 * Event notification.
2529 *---------------------------------------------------------------*/
2530 void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2533 char udev_cookie[DM_COOKIE_LENGTH];
2534 char *envp[] = { udev_cookie, NULL };
2537 kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2539 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2540 DM_COOKIE_ENV_VAR_NAME, cookie);
2541 kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
2545 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2547 return atomic_add_return(1, &md->uevent_seq);
2550 uint32_t dm_get_event_nr(struct mapped_device *md)
2552 return atomic_read(&md->event_nr);
2555 int dm_wait_event(struct mapped_device *md, int event_nr)
2557 return wait_event_interruptible(md->eventq,
2558 (event_nr != atomic_read(&md->event_nr)));
2561 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2563 unsigned long flags;
2565 spin_lock_irqsave(&md->uevent_lock, flags);
2566 list_add(elist, &md->uevent_list);
2567 spin_unlock_irqrestore(&md->uevent_lock, flags);
2571 * The gendisk is only valid as long as you have a reference
2574 struct gendisk *dm_disk(struct mapped_device *md)
2579 struct kobject *dm_kobject(struct mapped_device *md)
2585 * struct mapped_device should not be exported outside of dm.c
2586 * so use this check to verify that kobj is part of md structure
2588 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2590 struct mapped_device *md;
2592 md = container_of(kobj, struct mapped_device, kobj);
2593 if (&md->kobj != kobj)
2596 if (test_bit(DMF_FREEING, &md->flags) ||
2597 test_bit(DMF_DELETING, &md->flags))
2604 int dm_suspended(struct mapped_device *md)
2606 return test_bit(DMF_SUSPENDED, &md->flags);
2609 int dm_noflush_suspending(struct dm_target *ti)
2611 struct mapped_device *md = dm_table_get_md(ti->table);
2612 int r = __noflush_suspending(md);
2618 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2620 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2622 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2627 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2628 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2629 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2630 if (!pools->io_pool)
2631 goto free_pools_and_out;
2633 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2634 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2635 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2636 if (!pools->tio_pool)
2637 goto free_io_pool_and_out;
2639 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2640 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2642 goto free_tio_pool_and_out;
2646 free_tio_pool_and_out:
2647 mempool_destroy(pools->tio_pool);
2649 free_io_pool_and_out:
2650 mempool_destroy(pools->io_pool);
2658 void dm_free_md_mempools(struct dm_md_mempools *pools)
2664 mempool_destroy(pools->io_pool);
2666 if (pools->tio_pool)
2667 mempool_destroy(pools->tio_pool);
2670 bioset_free(pools->bs);
2675 static const struct block_device_operations dm_blk_dops = {
2676 .open = dm_blk_open,
2677 .release = dm_blk_close,
2678 .ioctl = dm_blk_ioctl,
2679 .getgeo = dm_blk_getgeo,
2680 .owner = THIS_MODULE
2683 EXPORT_SYMBOL(dm_get_mapinfo);
2688 module_init(dm_init);
2689 module_exit(dm_exit);
2691 module_param(major, uint, 0);
2692 MODULE_PARM_DESC(major, "The major number of the device mapper");
2693 MODULE_DESCRIPTION(DM_NAME " driver");
2694 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2695 MODULE_LICENSE("GPL");