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 int md_in_flight(struct mapped_device *md)
457 return atomic_read(&md->pending[READ]) +
458 atomic_read(&md->pending[WRITE]);
461 static void start_io_acct(struct dm_io *io)
463 struct mapped_device *md = io->md;
465 int rw = bio_data_dir(io->bio);
467 io->start_time = jiffies;
469 cpu = part_stat_lock();
470 part_round_stats(cpu, &dm_disk(md)->part0);
472 dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]);
475 static void end_io_acct(struct dm_io *io)
477 struct mapped_device *md = io->md;
478 struct bio *bio = io->bio;
479 unsigned long duration = jiffies - io->start_time;
481 int rw = bio_data_dir(bio);
483 cpu = part_stat_lock();
484 part_round_stats(cpu, &dm_disk(md)->part0);
485 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
489 * After this is decremented the bio must not be touched if it is
492 dm_disk(md)->part0.in_flight[rw] = pending =
493 atomic_dec_return(&md->pending[rw]);
494 pending += atomic_read(&md->pending[rw^0x1]);
496 /* nudge anyone waiting on suspend queue */
502 * Add the bio to the list of deferred io.
504 static void queue_io(struct mapped_device *md, struct bio *bio)
506 down_write(&md->io_lock);
508 spin_lock_irq(&md->deferred_lock);
509 bio_list_add(&md->deferred, bio);
510 spin_unlock_irq(&md->deferred_lock);
512 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
513 queue_work(md->wq, &md->work);
515 up_write(&md->io_lock);
519 * Everyone (including functions in this file), should use this
520 * function to access the md->map field, and make sure they call
521 * dm_table_put() when finished.
523 struct dm_table *dm_get_table(struct mapped_device *md)
528 read_lock_irqsave(&md->map_lock, flags);
532 read_unlock_irqrestore(&md->map_lock, flags);
538 * Get the geometry associated with a dm device
540 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
548 * Set the geometry of a device.
550 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
552 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
554 if (geo->start > sz) {
555 DMWARN("Start sector is beyond the geometry limits.");
564 /*-----------------------------------------------------------------
566 * A more elegant soln is in the works that uses the queue
567 * merge fn, unfortunately there are a couple of changes to
568 * the block layer that I want to make for this. So in the
569 * interests of getting something for people to use I give
570 * you this clearly demarcated crap.
571 *---------------------------------------------------------------*/
573 static int __noflush_suspending(struct mapped_device *md)
575 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
579 * Decrements the number of outstanding ios that a bio has been
580 * cloned into, completing the original io if necc.
582 static void dec_pending(struct dm_io *io, int error)
587 struct mapped_device *md = io->md;
589 /* Push-back supersedes any I/O errors */
590 if (unlikely(error)) {
591 spin_lock_irqsave(&io->endio_lock, flags);
592 if (!(io->error > 0 && __noflush_suspending(md)))
594 spin_unlock_irqrestore(&io->endio_lock, flags);
597 if (atomic_dec_and_test(&io->io_count)) {
598 if (io->error == DM_ENDIO_REQUEUE) {
600 * Target requested pushing back the I/O.
602 spin_lock_irqsave(&md->deferred_lock, flags);
603 if (__noflush_suspending(md)) {
604 if (!bio_rw_flagged(io->bio, BIO_RW_BARRIER))
605 bio_list_add_head(&md->deferred,
608 /* noflush suspend was interrupted. */
610 spin_unlock_irqrestore(&md->deferred_lock, flags);
613 io_error = io->error;
616 if (bio_rw_flagged(bio, BIO_RW_BARRIER)) {
618 * There can be just one barrier request so we use
619 * a per-device variable for error reporting.
620 * Note that you can't touch the bio after end_io_acct
622 if (!md->barrier_error && io_error != -EOPNOTSUPP)
623 md->barrier_error = io_error;
628 if (io_error != DM_ENDIO_REQUEUE) {
629 trace_block_bio_complete(md->queue, bio);
631 bio_endio(bio, io_error);
639 static void clone_endio(struct bio *bio, int error)
642 struct dm_target_io *tio = bio->bi_private;
643 struct dm_io *io = tio->io;
644 struct mapped_device *md = tio->io->md;
645 dm_endio_fn endio = tio->ti->type->end_io;
647 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
651 r = endio(tio->ti, bio, error, &tio->info);
652 if (r < 0 || r == DM_ENDIO_REQUEUE)
654 * error and requeue request are handled
658 else if (r == DM_ENDIO_INCOMPLETE)
659 /* The target will handle the io */
662 DMWARN("unimplemented target endio return value: %d", r);
668 * Store md for cleanup instead of tio which is about to get freed.
670 bio->bi_private = md->bs;
674 dec_pending(io, error);
678 * Partial completion handling for request-based dm
680 static void end_clone_bio(struct bio *clone, int error)
682 struct dm_rq_clone_bio_info *info = clone->bi_private;
683 struct dm_rq_target_io *tio = info->tio;
684 struct bio *bio = info->orig;
685 unsigned int nr_bytes = info->orig->bi_size;
691 * An error has already been detected on the request.
692 * Once error occurred, just let clone->end_io() handle
698 * Don't notice the error to the upper layer yet.
699 * The error handling decision is made by the target driver,
700 * when the request is completed.
707 * I/O for the bio successfully completed.
708 * Notice the data completion to the upper layer.
712 * bios are processed from the head of the list.
713 * So the completing bio should always be rq->bio.
714 * If it's not, something wrong is happening.
716 if (tio->orig->bio != bio)
717 DMERR("bio completion is going in the middle of the request");
720 * Update the original request.
721 * Do not use blk_end_request() here, because it may complete
722 * the original request before the clone, and break the ordering.
724 blk_update_request(tio->orig, 0, nr_bytes);
728 * Don't touch any member of the md after calling this function because
729 * the md may be freed in dm_put() at the end of this function.
730 * Or do dm_get() before calling this function and dm_put() later.
732 static void rq_completed(struct mapped_device *md, int run_queue)
734 int wakeup_waiters = 0;
735 struct request_queue *q = md->queue;
738 spin_lock_irqsave(q->queue_lock, flags);
739 if (!queue_in_flight(q))
741 spin_unlock_irqrestore(q->queue_lock, flags);
743 /* nudge anyone waiting on suspend queue */
751 * dm_put() must be at the end of this function. See the comment above
756 static void free_rq_clone(struct request *clone)
758 struct dm_rq_target_io *tio = clone->end_io_data;
760 blk_rq_unprep_clone(clone);
764 static void dm_unprep_request(struct request *rq)
766 struct request *clone = rq->special;
769 rq->cmd_flags &= ~REQ_DONTPREP;
771 free_rq_clone(clone);
775 * Requeue the original request of a clone.
777 void dm_requeue_unmapped_request(struct request *clone)
779 struct dm_rq_target_io *tio = clone->end_io_data;
780 struct mapped_device *md = tio->md;
781 struct request *rq = tio->orig;
782 struct request_queue *q = rq->q;
785 dm_unprep_request(rq);
787 spin_lock_irqsave(q->queue_lock, flags);
788 if (elv_queue_empty(q))
790 blk_requeue_request(q, rq);
791 spin_unlock_irqrestore(q->queue_lock, flags);
795 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
797 static void __stop_queue(struct request_queue *q)
802 static void stop_queue(struct request_queue *q)
806 spin_lock_irqsave(q->queue_lock, flags);
808 spin_unlock_irqrestore(q->queue_lock, flags);
811 static void __start_queue(struct request_queue *q)
813 if (blk_queue_stopped(q))
817 static void start_queue(struct request_queue *q)
821 spin_lock_irqsave(q->queue_lock, flags);
823 spin_unlock_irqrestore(q->queue_lock, flags);
827 * Complete the clone and the original request.
828 * Must be called without queue lock.
830 static void dm_end_request(struct request *clone, int error)
832 struct dm_rq_target_io *tio = clone->end_io_data;
833 struct mapped_device *md = tio->md;
834 struct request *rq = tio->orig;
836 if (blk_pc_request(rq)) {
837 rq->errors = clone->errors;
838 rq->resid_len = clone->resid_len;
842 * We are using the sense buffer of the original
844 * So setting the length of the sense data is enough.
846 rq->sense_len = clone->sense_len;
849 free_rq_clone(clone);
851 blk_end_request_all(rq, error);
857 * Request completion handler for request-based dm
859 static void dm_softirq_done(struct request *rq)
861 struct request *clone = rq->completion_data;
862 struct dm_rq_target_io *tio = clone->end_io_data;
863 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
864 int error = tio->error;
866 if (!(rq->cmd_flags & REQ_FAILED) && rq_end_io)
867 error = rq_end_io(tio->ti, clone, error, &tio->info);
870 /* The target wants to complete the I/O */
871 dm_end_request(clone, error);
872 else if (error == DM_ENDIO_INCOMPLETE)
873 /* The target will handle the I/O */
875 else if (error == DM_ENDIO_REQUEUE)
876 /* The target wants to requeue the I/O */
877 dm_requeue_unmapped_request(clone);
879 DMWARN("unimplemented target endio return value: %d", error);
885 * Complete the clone and the original request with the error status
886 * through softirq context.
888 static void dm_complete_request(struct request *clone, int error)
890 struct dm_rq_target_io *tio = clone->end_io_data;
891 struct request *rq = tio->orig;
894 rq->completion_data = clone;
895 blk_complete_request(rq);
899 * Complete the not-mapped clone and the original request with the error status
900 * through softirq context.
901 * Target's rq_end_io() function isn't called.
902 * This may be used when the target's map_rq() function fails.
904 void dm_kill_unmapped_request(struct request *clone, int error)
906 struct dm_rq_target_io *tio = clone->end_io_data;
907 struct request *rq = tio->orig;
909 rq->cmd_flags |= REQ_FAILED;
910 dm_complete_request(clone, error);
912 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
915 * Called with the queue lock held
917 static void end_clone_request(struct request *clone, int error)
920 * For just cleaning up the information of the queue in which
921 * the clone was dispatched.
922 * The clone is *NOT* freed actually here because it is alloced from
923 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
925 __blk_put_request(clone->q, clone);
928 * Actual request completion is done in a softirq context which doesn't
929 * hold the queue lock. Otherwise, deadlock could occur because:
930 * - another request may be submitted by the upper level driver
931 * of the stacking during the completion
932 * - the submission which requires queue lock may be done
935 dm_complete_request(clone, error);
938 static sector_t max_io_len(struct mapped_device *md,
939 sector_t sector, struct dm_target *ti)
941 sector_t offset = sector - ti->begin;
942 sector_t len = ti->len - offset;
945 * Does the target need to split even further ?
949 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
958 static void __map_bio(struct dm_target *ti, struct bio *clone,
959 struct dm_target_io *tio)
963 struct mapped_device *md;
965 clone->bi_end_io = clone_endio;
966 clone->bi_private = tio;
969 * Map the clone. If r == 0 we don't need to do
970 * anything, the target has assumed ownership of
973 atomic_inc(&tio->io->io_count);
974 sector = clone->bi_sector;
975 r = ti->type->map(ti, clone, &tio->info);
976 if (r == DM_MAPIO_REMAPPED) {
977 /* the bio has been remapped so dispatch it */
979 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
980 tio->io->bio->bi_bdev->bd_dev, sector);
982 generic_make_request(clone);
983 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
984 /* error the io and bail out, or requeue it if needed */
986 dec_pending(tio->io, r);
988 * Store bio_set for cleanup.
990 clone->bi_private = md->bs;
994 DMWARN("unimplemented target map return value: %d", r);
1000 struct mapped_device *md;
1001 struct dm_table *map;
1005 sector_t sector_count;
1009 static void dm_bio_destructor(struct bio *bio)
1011 struct bio_set *bs = bio->bi_private;
1017 * Creates a little bio that is just does part of a bvec.
1019 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1020 unsigned short idx, unsigned int offset,
1021 unsigned int len, struct bio_set *bs)
1024 struct bio_vec *bv = bio->bi_io_vec + idx;
1026 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1027 clone->bi_destructor = dm_bio_destructor;
1028 *clone->bi_io_vec = *bv;
1030 clone->bi_sector = sector;
1031 clone->bi_bdev = bio->bi_bdev;
1032 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
1034 clone->bi_size = to_bytes(len);
1035 clone->bi_io_vec->bv_offset = offset;
1036 clone->bi_io_vec->bv_len = clone->bi_size;
1037 clone->bi_flags |= 1 << BIO_CLONED;
1039 if (bio_integrity(bio)) {
1040 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1041 bio_integrity_trim(clone,
1042 bio_sector_offset(bio, idx, offset), len);
1049 * Creates a bio that consists of range of complete bvecs.
1051 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1052 unsigned short idx, unsigned short bv_count,
1053 unsigned int len, struct bio_set *bs)
1057 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1058 __bio_clone(clone, bio);
1059 clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
1060 clone->bi_destructor = dm_bio_destructor;
1061 clone->bi_sector = sector;
1062 clone->bi_idx = idx;
1063 clone->bi_vcnt = idx + bv_count;
1064 clone->bi_size = to_bytes(len);
1065 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1067 if (bio_integrity(bio)) {
1068 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1070 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1071 bio_integrity_trim(clone,
1072 bio_sector_offset(bio, idx, 0), len);
1078 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1079 struct dm_target *ti)
1081 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1085 memset(&tio->info, 0, sizeof(tio->info));
1090 static void __flush_target(struct clone_info *ci, struct dm_target *ti,
1093 struct dm_target_io *tio = alloc_tio(ci, ti);
1096 tio->info.flush_request = flush_nr;
1098 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1099 __bio_clone(clone, ci->bio);
1100 clone->bi_destructor = dm_bio_destructor;
1102 __map_bio(ti, clone, tio);
1105 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1107 unsigned target_nr = 0, flush_nr;
1108 struct dm_target *ti;
1110 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1111 for (flush_nr = 0; flush_nr < ti->num_flush_requests;
1113 __flush_target(ci, ti, flush_nr);
1115 ci->sector_count = 0;
1120 static int __clone_and_map(struct clone_info *ci)
1122 struct bio *clone, *bio = ci->bio;
1123 struct dm_target *ti;
1124 sector_t len = 0, max;
1125 struct dm_target_io *tio;
1127 if (unlikely(bio_empty_barrier(bio)))
1128 return __clone_and_map_empty_barrier(ci);
1130 ti = dm_table_find_target(ci->map, ci->sector);
1131 if (!dm_target_is_valid(ti))
1134 max = max_io_len(ci->md, ci->sector, ti);
1137 * Allocate a target io object.
1139 tio = alloc_tio(ci, ti);
1141 if (ci->sector_count <= max) {
1143 * Optimise for the simple case where we can do all of
1144 * the remaining io with a single clone.
1146 clone = clone_bio(bio, ci->sector, ci->idx,
1147 bio->bi_vcnt - ci->idx, ci->sector_count,
1149 __map_bio(ti, clone, tio);
1150 ci->sector_count = 0;
1152 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1154 * There are some bvecs that don't span targets.
1155 * Do as many of these as possible.
1158 sector_t remaining = max;
1161 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1162 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1164 if (bv_len > remaining)
1167 remaining -= bv_len;
1171 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1173 __map_bio(ti, clone, tio);
1176 ci->sector_count -= len;
1181 * Handle a bvec that must be split between two or more targets.
1183 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1184 sector_t remaining = to_sector(bv->bv_len);
1185 unsigned int offset = 0;
1189 ti = dm_table_find_target(ci->map, ci->sector);
1190 if (!dm_target_is_valid(ti))
1193 max = max_io_len(ci->md, ci->sector, ti);
1195 tio = alloc_tio(ci, ti);
1198 len = min(remaining, max);
1200 clone = split_bvec(bio, ci->sector, ci->idx,
1201 bv->bv_offset + offset, len,
1204 __map_bio(ti, clone, tio);
1207 ci->sector_count -= len;
1208 offset += to_bytes(len);
1209 } while (remaining -= len);
1218 * Split the bio into several clones and submit it to targets.
1220 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1222 struct clone_info ci;
1225 ci.map = dm_get_table(md);
1226 if (unlikely(!ci.map)) {
1227 if (!bio_rw_flagged(bio, BIO_RW_BARRIER))
1230 if (!md->barrier_error)
1231 md->barrier_error = -EIO;
1237 ci.io = alloc_io(md);
1239 atomic_set(&ci.io->io_count, 1);
1242 spin_lock_init(&ci.io->endio_lock);
1243 ci.sector = bio->bi_sector;
1244 ci.sector_count = bio_sectors(bio);
1245 if (unlikely(bio_empty_barrier(bio)))
1246 ci.sector_count = 1;
1247 ci.idx = bio->bi_idx;
1249 start_io_acct(ci.io);
1250 while (ci.sector_count && !error)
1251 error = __clone_and_map(&ci);
1253 /* drop the extra reference count */
1254 dec_pending(ci.io, error);
1255 dm_table_put(ci.map);
1257 /*-----------------------------------------------------------------
1259 *---------------------------------------------------------------*/
1261 static int dm_merge_bvec(struct request_queue *q,
1262 struct bvec_merge_data *bvm,
1263 struct bio_vec *biovec)
1265 struct mapped_device *md = q->queuedata;
1266 struct dm_table *map = dm_get_table(md);
1267 struct dm_target *ti;
1268 sector_t max_sectors;
1274 ti = dm_table_find_target(map, bvm->bi_sector);
1275 if (!dm_target_is_valid(ti))
1279 * Find maximum amount of I/O that won't need splitting
1281 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
1282 (sector_t) BIO_MAX_SECTORS);
1283 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1288 * merge_bvec_fn() returns number of bytes
1289 * it can accept at this offset
1290 * max is precomputed maximal io size
1292 if (max_size && ti->type->merge)
1293 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1295 * If the target doesn't support merge method and some of the devices
1296 * provided their merge_bvec method (we know this by looking at
1297 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1298 * entries. So always set max_size to 0, and the code below allows
1301 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1310 * Always allow an entire first page
1312 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1313 max_size = biovec->bv_len;
1319 * The request function that just remaps the bio built up by
1322 static int _dm_request(struct request_queue *q, struct bio *bio)
1324 int rw = bio_data_dir(bio);
1325 struct mapped_device *md = q->queuedata;
1328 down_read(&md->io_lock);
1330 cpu = part_stat_lock();
1331 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1332 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1336 * If we're suspended or the thread is processing barriers
1337 * we have to queue this io for later.
1339 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1340 unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
1341 up_read(&md->io_lock);
1343 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1344 bio_rw(bio) == READA) {
1354 __split_and_process_bio(md, bio);
1355 up_read(&md->io_lock);
1359 static int dm_make_request(struct request_queue *q, struct bio *bio)
1361 struct mapped_device *md = q->queuedata;
1363 if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
1364 bio_endio(bio, -EOPNOTSUPP);
1368 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1371 static int dm_request_based(struct mapped_device *md)
1373 return blk_queue_stackable(md->queue);
1376 static int dm_request(struct request_queue *q, struct bio *bio)
1378 struct mapped_device *md = q->queuedata;
1380 if (dm_request_based(md))
1381 return dm_make_request(q, bio);
1383 return _dm_request(q, bio);
1386 void dm_dispatch_request(struct request *rq)
1390 if (blk_queue_io_stat(rq->q))
1391 rq->cmd_flags |= REQ_IO_STAT;
1393 rq->start_time = jiffies;
1394 r = blk_insert_cloned_request(rq->q, rq);
1396 dm_complete_request(rq, r);
1398 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1400 static void dm_rq_bio_destructor(struct bio *bio)
1402 struct dm_rq_clone_bio_info *info = bio->bi_private;
1403 struct mapped_device *md = info->tio->md;
1405 free_bio_info(info);
1406 bio_free(bio, md->bs);
1409 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1412 struct dm_rq_target_io *tio = data;
1413 struct mapped_device *md = tio->md;
1414 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1419 info->orig = bio_orig;
1421 bio->bi_end_io = end_clone_bio;
1422 bio->bi_private = info;
1423 bio->bi_destructor = dm_rq_bio_destructor;
1428 static int setup_clone(struct request *clone, struct request *rq,
1429 struct dm_rq_target_io *tio)
1431 int r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1432 dm_rq_bio_constructor, tio);
1437 clone->cmd = rq->cmd;
1438 clone->cmd_len = rq->cmd_len;
1439 clone->sense = rq->sense;
1440 clone->buffer = rq->buffer;
1441 clone->end_io = end_clone_request;
1442 clone->end_io_data = tio;
1447 static int dm_rq_flush_suspending(struct mapped_device *md)
1449 return !md->suspend_rq.special;
1453 * Called with the queue lock held.
1455 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1457 struct mapped_device *md = q->queuedata;
1458 struct dm_rq_target_io *tio;
1459 struct request *clone;
1461 if (unlikely(rq == &md->suspend_rq)) {
1462 if (dm_rq_flush_suspending(md))
1465 /* The flush suspend was interrupted */
1466 return BLKPREP_KILL;
1469 if (unlikely(rq->special)) {
1470 DMWARN("Already has something in rq->special.");
1471 return BLKPREP_KILL;
1474 tio = alloc_rq_tio(md); /* Only one for each original request */
1477 return BLKPREP_DEFER;
1483 memset(&tio->info, 0, sizeof(tio->info));
1485 clone = &tio->clone;
1486 if (setup_clone(clone, rq, tio)) {
1489 return BLKPREP_DEFER;
1492 rq->special = clone;
1493 rq->cmd_flags |= REQ_DONTPREP;
1498 static void map_request(struct dm_target *ti, struct request *clone,
1499 struct mapped_device *md)
1502 struct dm_rq_target_io *tio = clone->end_io_data;
1505 * Hold the md reference here for the in-flight I/O.
1506 * We can't rely on the reference count by device opener,
1507 * because the device may be closed during the request completion
1508 * when all bios are completed.
1509 * See the comment in rq_completed() too.
1514 r = ti->type->map_rq(ti, clone, &tio->info);
1516 case DM_MAPIO_SUBMITTED:
1517 /* The target has taken the I/O to submit by itself later */
1519 case DM_MAPIO_REMAPPED:
1520 /* The target has remapped the I/O so dispatch it */
1521 dm_dispatch_request(clone);
1523 case DM_MAPIO_REQUEUE:
1524 /* The target wants to requeue the I/O */
1525 dm_requeue_unmapped_request(clone);
1529 DMWARN("unimplemented target map return value: %d", r);
1533 /* The target wants to complete the I/O */
1534 dm_kill_unmapped_request(clone, r);
1540 * q->request_fn for request-based dm.
1541 * Called with the queue lock held.
1543 static void dm_request_fn(struct request_queue *q)
1545 struct mapped_device *md = q->queuedata;
1546 struct dm_table *map = dm_get_table(md);
1547 struct dm_target *ti;
1551 * For noflush suspend, check blk_queue_stopped() to immediately
1552 * quit I/O dispatching.
1554 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1555 rq = blk_peek_request(q);
1559 if (unlikely(rq == &md->suspend_rq)) { /* Flush suspend maker */
1560 if (queue_in_flight(q))
1561 /* Not quiet yet. Wait more */
1564 /* This device should be quiet now */
1566 blk_start_request(rq);
1567 __blk_end_request_all(rq, 0);
1572 ti = dm_table_find_target(map, blk_rq_pos(rq));
1573 if (ti->type->busy && ti->type->busy(ti))
1576 blk_start_request(rq);
1577 spin_unlock(q->queue_lock);
1578 map_request(ti, rq->special, md);
1579 spin_lock_irq(q->queue_lock);
1585 if (!elv_queue_empty(q))
1586 /* Some requests still remain, retry later */
1595 int dm_underlying_device_busy(struct request_queue *q)
1597 return blk_lld_busy(q);
1599 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1601 static int dm_lld_busy(struct request_queue *q)
1604 struct mapped_device *md = q->queuedata;
1605 struct dm_table *map = dm_get_table(md);
1607 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1610 r = dm_table_any_busy_target(map);
1617 static void dm_unplug_all(struct request_queue *q)
1619 struct mapped_device *md = q->queuedata;
1620 struct dm_table *map = dm_get_table(md);
1623 if (dm_request_based(md))
1624 generic_unplug_device(q);
1626 dm_table_unplug_all(map);
1631 static int dm_any_congested(void *congested_data, int bdi_bits)
1634 struct mapped_device *md = congested_data;
1635 struct dm_table *map;
1637 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1638 map = dm_get_table(md);
1641 * Request-based dm cares about only own queue for
1642 * the query about congestion status of request_queue
1644 if (dm_request_based(md))
1645 r = md->queue->backing_dev_info.state &
1648 r = dm_table_any_congested(map, bdi_bits);
1657 /*-----------------------------------------------------------------
1658 * An IDR is used to keep track of allocated minor numbers.
1659 *---------------------------------------------------------------*/
1660 static DEFINE_IDR(_minor_idr);
1662 static void free_minor(int minor)
1664 spin_lock(&_minor_lock);
1665 idr_remove(&_minor_idr, minor);
1666 spin_unlock(&_minor_lock);
1670 * See if the device with a specific minor # is free.
1672 static int specific_minor(int minor)
1676 if (minor >= (1 << MINORBITS))
1679 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1683 spin_lock(&_minor_lock);
1685 if (idr_find(&_minor_idr, minor)) {
1690 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1695 idr_remove(&_minor_idr, m);
1701 spin_unlock(&_minor_lock);
1705 static int next_free_minor(int *minor)
1709 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1713 spin_lock(&_minor_lock);
1715 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1719 if (m >= (1 << MINORBITS)) {
1720 idr_remove(&_minor_idr, m);
1728 spin_unlock(&_minor_lock);
1732 static const struct block_device_operations dm_blk_dops;
1734 static void dm_wq_work(struct work_struct *work);
1737 * Allocate and initialise a blank device with a given minor.
1739 static struct mapped_device *alloc_dev(int minor)
1742 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1746 DMWARN("unable to allocate device, out of memory.");
1750 if (!try_module_get(THIS_MODULE))
1751 goto bad_module_get;
1753 /* get a minor number for the dev */
1754 if (minor == DM_ANY_MINOR)
1755 r = next_free_minor(&minor);
1757 r = specific_minor(minor);
1761 init_rwsem(&md->io_lock);
1762 mutex_init(&md->suspend_lock);
1763 spin_lock_init(&md->deferred_lock);
1764 rwlock_init(&md->map_lock);
1765 atomic_set(&md->holders, 1);
1766 atomic_set(&md->open_count, 0);
1767 atomic_set(&md->event_nr, 0);
1768 atomic_set(&md->uevent_seq, 0);
1769 INIT_LIST_HEAD(&md->uevent_list);
1770 spin_lock_init(&md->uevent_lock);
1772 md->queue = blk_init_queue(dm_request_fn, NULL);
1777 * Request-based dm devices cannot be stacked on top of bio-based dm
1778 * devices. The type of this dm device has not been decided yet,
1779 * although we initialized the queue using blk_init_queue().
1780 * The type is decided at the first table loading time.
1781 * To prevent problematic device stacking, clear the queue flag
1782 * for request stacking support until then.
1784 * This queue is new, so no concurrency on the queue_flags.
1786 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1787 md->saved_make_request_fn = md->queue->make_request_fn;
1788 md->queue->queuedata = md;
1789 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1790 md->queue->backing_dev_info.congested_data = md;
1791 blk_queue_make_request(md->queue, dm_request);
1792 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1793 md->queue->unplug_fn = dm_unplug_all;
1794 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1795 blk_queue_softirq_done(md->queue, dm_softirq_done);
1796 blk_queue_prep_rq(md->queue, dm_prep_fn);
1797 blk_queue_lld_busy(md->queue, dm_lld_busy);
1799 md->disk = alloc_disk(1);
1803 atomic_set(&md->pending[0], 0);
1804 atomic_set(&md->pending[1], 0);
1805 init_waitqueue_head(&md->wait);
1806 INIT_WORK(&md->work, dm_wq_work);
1807 init_waitqueue_head(&md->eventq);
1809 md->disk->major = _major;
1810 md->disk->first_minor = minor;
1811 md->disk->fops = &dm_blk_dops;
1812 md->disk->queue = md->queue;
1813 md->disk->private_data = md;
1814 sprintf(md->disk->disk_name, "dm-%d", minor);
1816 format_dev_t(md->name, MKDEV(_major, minor));
1818 md->wq = create_singlethread_workqueue("kdmflush");
1822 md->bdev = bdget_disk(md->disk, 0);
1826 /* Populate the mapping, nobody knows we exist yet */
1827 spin_lock(&_minor_lock);
1828 old_md = idr_replace(&_minor_idr, md, minor);
1829 spin_unlock(&_minor_lock);
1831 BUG_ON(old_md != MINOR_ALLOCED);
1836 destroy_workqueue(md->wq);
1838 del_gendisk(md->disk);
1841 blk_cleanup_queue(md->queue);
1845 module_put(THIS_MODULE);
1851 static void unlock_fs(struct mapped_device *md);
1853 static void free_dev(struct mapped_device *md)
1855 int minor = MINOR(disk_devt(md->disk));
1859 destroy_workqueue(md->wq);
1861 mempool_destroy(md->tio_pool);
1863 mempool_destroy(md->io_pool);
1865 bioset_free(md->bs);
1866 blk_integrity_unregister(md->disk);
1867 del_gendisk(md->disk);
1870 spin_lock(&_minor_lock);
1871 md->disk->private_data = NULL;
1872 spin_unlock(&_minor_lock);
1875 blk_cleanup_queue(md->queue);
1876 module_put(THIS_MODULE);
1880 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1882 struct dm_md_mempools *p;
1884 if (md->io_pool && md->tio_pool && md->bs)
1885 /* the md already has necessary mempools */
1888 p = dm_table_get_md_mempools(t);
1889 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1891 md->io_pool = p->io_pool;
1893 md->tio_pool = p->tio_pool;
1899 /* mempool bind completed, now no need any mempools in the table */
1900 dm_table_free_md_mempools(t);
1904 * Bind a table to the device.
1906 static void event_callback(void *context)
1908 unsigned long flags;
1910 struct mapped_device *md = (struct mapped_device *) context;
1912 spin_lock_irqsave(&md->uevent_lock, flags);
1913 list_splice_init(&md->uevent_list, &uevents);
1914 spin_unlock_irqrestore(&md->uevent_lock, flags);
1916 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1918 atomic_inc(&md->event_nr);
1919 wake_up(&md->eventq);
1922 static void __set_size(struct mapped_device *md, sector_t size)
1924 set_capacity(md->disk, size);
1926 mutex_lock(&md->bdev->bd_inode->i_mutex);
1927 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1928 mutex_unlock(&md->bdev->bd_inode->i_mutex);
1931 static int __bind(struct mapped_device *md, struct dm_table *t,
1932 struct queue_limits *limits)
1934 struct request_queue *q = md->queue;
1936 unsigned long flags;
1938 size = dm_table_get_size(t);
1941 * Wipe any geometry if the size of the table changed.
1943 if (size != get_capacity(md->disk))
1944 memset(&md->geometry, 0, sizeof(md->geometry));
1946 __set_size(md, size);
1949 dm_table_destroy(t);
1953 dm_table_event_callback(t, event_callback, md);
1956 * The queue hasn't been stopped yet, if the old table type wasn't
1957 * for request-based during suspension. So stop it to prevent
1958 * I/O mapping before resume.
1959 * This must be done before setting the queue restrictions,
1960 * because request-based dm may be run just after the setting.
1962 if (dm_table_request_based(t) && !blk_queue_stopped(q))
1965 __bind_mempools(md, t);
1967 write_lock_irqsave(&md->map_lock, flags);
1969 dm_table_set_restrictions(t, q, limits);
1970 write_unlock_irqrestore(&md->map_lock, flags);
1975 static void __unbind(struct mapped_device *md)
1977 struct dm_table *map = md->map;
1978 unsigned long flags;
1983 dm_table_event_callback(map, NULL, NULL);
1984 write_lock_irqsave(&md->map_lock, flags);
1986 write_unlock_irqrestore(&md->map_lock, flags);
1987 dm_table_destroy(map);
1991 * Constructor for a new device.
1993 int dm_create(int minor, struct mapped_device **result)
1995 struct mapped_device *md;
1997 md = alloc_dev(minor);
2007 static struct mapped_device *dm_find_md(dev_t dev)
2009 struct mapped_device *md;
2010 unsigned minor = MINOR(dev);
2012 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2015 spin_lock(&_minor_lock);
2017 md = idr_find(&_minor_idr, minor);
2018 if (md && (md == MINOR_ALLOCED ||
2019 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2020 test_bit(DMF_FREEING, &md->flags))) {
2026 spin_unlock(&_minor_lock);
2031 struct mapped_device *dm_get_md(dev_t dev)
2033 struct mapped_device *md = dm_find_md(dev);
2041 void *dm_get_mdptr(struct mapped_device *md)
2043 return md->interface_ptr;
2046 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2048 md->interface_ptr = ptr;
2051 void dm_get(struct mapped_device *md)
2053 atomic_inc(&md->holders);
2056 const char *dm_device_name(struct mapped_device *md)
2060 EXPORT_SYMBOL_GPL(dm_device_name);
2062 void dm_put(struct mapped_device *md)
2064 struct dm_table *map;
2066 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2068 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
2069 map = dm_get_table(md);
2070 idr_replace(&_minor_idr, MINOR_ALLOCED,
2071 MINOR(disk_devt(dm_disk(md))));
2072 set_bit(DMF_FREEING, &md->flags);
2073 spin_unlock(&_minor_lock);
2074 if (!dm_suspended(md)) {
2075 dm_table_presuspend_targets(map);
2076 dm_table_postsuspend_targets(map);
2084 EXPORT_SYMBOL_GPL(dm_put);
2086 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2089 DECLARE_WAITQUEUE(wait, current);
2090 struct request_queue *q = md->queue;
2091 unsigned long flags;
2093 dm_unplug_all(md->queue);
2095 add_wait_queue(&md->wait, &wait);
2098 set_current_state(interruptible);
2101 if (dm_request_based(md)) {
2102 spin_lock_irqsave(q->queue_lock, flags);
2103 if (!queue_in_flight(q) && blk_queue_stopped(q)) {
2104 spin_unlock_irqrestore(q->queue_lock, flags);
2107 spin_unlock_irqrestore(q->queue_lock, flags);
2108 } else if (!md_in_flight(md))
2111 if (interruptible == TASK_INTERRUPTIBLE &&
2112 signal_pending(current)) {
2119 set_current_state(TASK_RUNNING);
2121 remove_wait_queue(&md->wait, &wait);
2126 static void dm_flush(struct mapped_device *md)
2128 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2130 bio_init(&md->barrier_bio);
2131 md->barrier_bio.bi_bdev = md->bdev;
2132 md->barrier_bio.bi_rw = WRITE_BARRIER;
2133 __split_and_process_bio(md, &md->barrier_bio);
2135 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2138 static void process_barrier(struct mapped_device *md, struct bio *bio)
2140 md->barrier_error = 0;
2144 if (!bio_empty_barrier(bio)) {
2145 __split_and_process_bio(md, bio);
2149 if (md->barrier_error != DM_ENDIO_REQUEUE)
2150 bio_endio(bio, md->barrier_error);
2152 spin_lock_irq(&md->deferred_lock);
2153 bio_list_add_head(&md->deferred, bio);
2154 spin_unlock_irq(&md->deferred_lock);
2159 * Process the deferred bios
2161 static void dm_wq_work(struct work_struct *work)
2163 struct mapped_device *md = container_of(work, struct mapped_device,
2167 down_write(&md->io_lock);
2169 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2170 spin_lock_irq(&md->deferred_lock);
2171 c = bio_list_pop(&md->deferred);
2172 spin_unlock_irq(&md->deferred_lock);
2175 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2179 up_write(&md->io_lock);
2181 if (dm_request_based(md))
2182 generic_make_request(c);
2184 if (bio_rw_flagged(c, BIO_RW_BARRIER))
2185 process_barrier(md, c);
2187 __split_and_process_bio(md, c);
2190 down_write(&md->io_lock);
2193 up_write(&md->io_lock);
2196 static void dm_queue_flush(struct mapped_device *md)
2198 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2199 smp_mb__after_clear_bit();
2200 queue_work(md->wq, &md->work);
2204 * Swap in a new table (destroying old one).
2206 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
2208 struct queue_limits limits;
2211 mutex_lock(&md->suspend_lock);
2213 /* device must be suspended */
2214 if (!dm_suspended(md))
2217 r = dm_calculate_queue_limits(table, &limits);
2221 /* cannot change the device type, once a table is bound */
2223 (dm_table_get_type(md->map) != dm_table_get_type(table))) {
2224 DMWARN("can't change the device type after a table is bound");
2229 r = __bind(md, table, &limits);
2232 mutex_unlock(&md->suspend_lock);
2236 static void dm_rq_invalidate_suspend_marker(struct mapped_device *md)
2238 md->suspend_rq.special = (void *)0x1;
2241 static void dm_rq_abort_suspend(struct mapped_device *md, int noflush)
2243 struct request_queue *q = md->queue;
2244 unsigned long flags;
2246 spin_lock_irqsave(q->queue_lock, flags);
2248 dm_rq_invalidate_suspend_marker(md);
2250 spin_unlock_irqrestore(q->queue_lock, flags);
2253 static void dm_rq_start_suspend(struct mapped_device *md, int noflush)
2255 struct request *rq = &md->suspend_rq;
2256 struct request_queue *q = md->queue;
2262 blk_insert_request(q, rq, 0, NULL);
2266 static int dm_rq_suspend_available(struct mapped_device *md, int noflush)
2269 struct request *rq = &md->suspend_rq;
2270 struct request_queue *q = md->queue;
2271 unsigned long flags;
2276 /* The marker must be protected by queue lock if it is in use */
2277 spin_lock_irqsave(q->queue_lock, flags);
2278 if (unlikely(rq->ref_count)) {
2280 * This can happen, when the previous flush suspend was
2281 * interrupted, the marker is still in the queue and
2282 * this flush suspend has been invoked, because we don't
2283 * remove the marker at the time of suspend interruption.
2284 * We have only one marker per mapped_device, so we can't
2285 * start another flush suspend while it is in use.
2287 BUG_ON(!rq->special); /* The marker should be invalidated */
2288 DMWARN("Invalidating the previous flush suspend is still in"
2289 " progress. Please retry later.");
2292 spin_unlock_irqrestore(q->queue_lock, flags);
2298 * Functions to lock and unlock any filesystem running on the
2301 static int lock_fs(struct mapped_device *md)
2305 WARN_ON(md->frozen_sb);
2307 md->frozen_sb = freeze_bdev(md->bdev);
2308 if (IS_ERR(md->frozen_sb)) {
2309 r = PTR_ERR(md->frozen_sb);
2310 md->frozen_sb = NULL;
2314 set_bit(DMF_FROZEN, &md->flags);
2319 static void unlock_fs(struct mapped_device *md)
2321 if (!test_bit(DMF_FROZEN, &md->flags))
2324 thaw_bdev(md->bdev, md->frozen_sb);
2325 md->frozen_sb = NULL;
2326 clear_bit(DMF_FROZEN, &md->flags);
2330 * We need to be able to change a mapping table under a mounted
2331 * filesystem. For example we might want to move some data in
2332 * the background. Before the table can be swapped with
2333 * dm_bind_table, dm_suspend must be called to flush any in
2334 * flight bios and ensure that any further io gets deferred.
2337 * Suspend mechanism in request-based dm.
2339 * After the suspend starts, further incoming requests are kept in
2340 * the request_queue and deferred.
2341 * Remaining requests in the request_queue at the start of suspend are flushed
2342 * if it is flush suspend.
2343 * The suspend completes when the following conditions have been satisfied,
2345 * 1. q->in_flight is 0 (which means no in_flight request)
2346 * 2. queue has been stopped (which means no request dispatching)
2351 * Noflush suspend doesn't need to dispatch remaining requests.
2352 * So stop the queue immediately. Then, wait for all in_flight requests
2353 * to be completed or requeued.
2355 * To abort noflush suspend, start the queue.
2360 * Flush suspend needs to dispatch remaining requests. So stop the queue
2361 * after the remaining requests are completed. (Requeued request must be also
2362 * re-dispatched and completed. Until then, we can't stop the queue.)
2364 * During flushing the remaining requests, further incoming requests are also
2365 * inserted to the same queue. To distinguish which requests are to be
2366 * flushed, we insert a marker request to the queue at the time of starting
2367 * flush suspend, like a barrier.
2368 * The dispatching is blocked when the marker is found on the top of the queue.
2369 * And the queue is stopped when all in_flight requests are completed, since
2370 * that means the remaining requests are completely flushed.
2371 * Then, the marker is removed from the queue.
2373 * To abort flush suspend, we also need to take care of the marker, not only
2374 * starting the queue.
2375 * We don't remove the marker forcibly from the queue since it's against
2376 * the block-layer manner. Instead, we put a invalidated mark on the marker.
2377 * When the invalidated marker is found on the top of the queue, it is
2378 * immediately removed from the queue, so it doesn't block dispatching.
2379 * Because we have only one marker per mapped_device, we can't start another
2380 * flush suspend until the invalidated marker is removed from the queue.
2381 * So fail and return with -EBUSY in such a case.
2383 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2385 struct dm_table *map = NULL;
2387 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2388 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2390 mutex_lock(&md->suspend_lock);
2392 if (dm_suspended(md)) {
2397 if (dm_request_based(md) && !dm_rq_suspend_available(md, noflush)) {
2402 map = dm_get_table(md);
2405 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2406 * This flag is cleared before dm_suspend returns.
2409 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2411 /* This does not get reverted if there's an error later. */
2412 dm_table_presuspend_targets(map);
2415 * Flush I/O to the device. noflush supersedes do_lockfs,
2416 * because lock_fs() needs to flush I/Os.
2418 if (!noflush && do_lockfs) {
2425 * Here we must make sure that no processes are submitting requests
2426 * to target drivers i.e. no one may be executing
2427 * __split_and_process_bio. This is called from dm_request and
2430 * To get all processes out of __split_and_process_bio in dm_request,
2431 * we take the write lock. To prevent any process from reentering
2432 * __split_and_process_bio from dm_request, we set
2433 * DMF_QUEUE_IO_TO_THREAD.
2435 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2436 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2437 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2438 * further calls to __split_and_process_bio from dm_wq_work.
2440 down_write(&md->io_lock);
2441 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2442 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2443 up_write(&md->io_lock);
2445 flush_workqueue(md->wq);
2447 if (dm_request_based(md))
2448 dm_rq_start_suspend(md, noflush);
2451 * At this point no more requests are entering target request routines.
2452 * We call dm_wait_for_completion to wait for all existing requests
2455 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2457 down_write(&md->io_lock);
2459 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2460 up_write(&md->io_lock);
2462 /* were we interrupted ? */
2466 if (dm_request_based(md))
2467 dm_rq_abort_suspend(md, noflush);
2470 goto out; /* pushback list is already flushed, so skip flush */
2474 * If dm_wait_for_completion returned 0, the device is completely
2475 * quiescent now. There is no request-processing activity. All new
2476 * requests are being added to md->deferred list.
2479 dm_table_postsuspend_targets(map);
2481 set_bit(DMF_SUSPENDED, &md->flags);
2487 mutex_unlock(&md->suspend_lock);
2491 int dm_resume(struct mapped_device *md)
2494 struct dm_table *map = NULL;
2496 mutex_lock(&md->suspend_lock);
2497 if (!dm_suspended(md))
2500 map = dm_get_table(md);
2501 if (!map || !dm_table_get_size(map))
2504 r = dm_table_resume_targets(map);
2511 * Flushing deferred I/Os must be done after targets are resumed
2512 * so that mapping of targets can work correctly.
2513 * Request-based dm is queueing the deferred I/Os in its request_queue.
2515 if (dm_request_based(md))
2516 start_queue(md->queue);
2520 clear_bit(DMF_SUSPENDED, &md->flags);
2522 dm_table_unplug_all(map);
2526 mutex_unlock(&md->suspend_lock);
2531 /*-----------------------------------------------------------------
2532 * Event notification.
2533 *---------------------------------------------------------------*/
2534 void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2537 char udev_cookie[DM_COOKIE_LENGTH];
2538 char *envp[] = { udev_cookie, NULL };
2541 kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2543 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2544 DM_COOKIE_ENV_VAR_NAME, cookie);
2545 kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
2549 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2551 return atomic_add_return(1, &md->uevent_seq);
2554 uint32_t dm_get_event_nr(struct mapped_device *md)
2556 return atomic_read(&md->event_nr);
2559 int dm_wait_event(struct mapped_device *md, int event_nr)
2561 return wait_event_interruptible(md->eventq,
2562 (event_nr != atomic_read(&md->event_nr)));
2565 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2567 unsigned long flags;
2569 spin_lock_irqsave(&md->uevent_lock, flags);
2570 list_add(elist, &md->uevent_list);
2571 spin_unlock_irqrestore(&md->uevent_lock, flags);
2575 * The gendisk is only valid as long as you have a reference
2578 struct gendisk *dm_disk(struct mapped_device *md)
2583 struct kobject *dm_kobject(struct mapped_device *md)
2589 * struct mapped_device should not be exported outside of dm.c
2590 * so use this check to verify that kobj is part of md structure
2592 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2594 struct mapped_device *md;
2596 md = container_of(kobj, struct mapped_device, kobj);
2597 if (&md->kobj != kobj)
2600 if (test_bit(DMF_FREEING, &md->flags) ||
2601 test_bit(DMF_DELETING, &md->flags))
2608 int dm_suspended(struct mapped_device *md)
2610 return test_bit(DMF_SUSPENDED, &md->flags);
2613 int dm_noflush_suspending(struct dm_target *ti)
2615 struct mapped_device *md = dm_table_get_md(ti->table);
2616 int r = __noflush_suspending(md);
2622 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2624 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2626 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2631 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2632 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2633 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2634 if (!pools->io_pool)
2635 goto free_pools_and_out;
2637 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2638 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2639 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2640 if (!pools->tio_pool)
2641 goto free_io_pool_and_out;
2643 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2644 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2646 goto free_tio_pool_and_out;
2650 free_tio_pool_and_out:
2651 mempool_destroy(pools->tio_pool);
2653 free_io_pool_and_out:
2654 mempool_destroy(pools->io_pool);
2662 void dm_free_md_mempools(struct dm_md_mempools *pools)
2668 mempool_destroy(pools->io_pool);
2670 if (pools->tio_pool)
2671 mempool_destroy(pools->tio_pool);
2674 bioset_free(pools->bs);
2679 static const struct block_device_operations dm_blk_dops = {
2680 .open = dm_blk_open,
2681 .release = dm_blk_close,
2682 .ioctl = dm_blk_ioctl,
2683 .getgeo = dm_blk_getgeo,
2684 .owner = THIS_MODULE
2687 EXPORT_SYMBOL(dm_get_mapinfo);
2692 module_init(dm_init);
2693 module_exit(dm_exit);
2695 module_param(major, uint, 0);
2696 MODULE_PARM_DESC(major, "The major number of the device mapper");
2697 MODULE_DESCRIPTION(DM_NAME " driver");
2698 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2699 MODULE_LICENSE("GPL");