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/smp_lock.h>
19 #include <linux/mempool.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/hdreg.h>
23 #include <linux/delay.h>
25 #include <trace/events/block.h>
27 #define DM_MSG_PREFIX "core"
30 * Cookies are numeric values sent with CHANGE and REMOVE
31 * uevents while resuming, removing or renaming the device.
33 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
34 #define DM_COOKIE_LENGTH 24
36 static const char *_name = DM_NAME;
38 static unsigned int major = 0;
39 static unsigned int _major = 0;
41 static DEFINE_SPINLOCK(_minor_lock);
44 * One of these is allocated per bio.
47 struct mapped_device *md;
51 unsigned long start_time;
52 spinlock_t endio_lock;
57 * One of these is allocated per target within a bio. Hopefully
58 * this will be simplified out one day.
67 * For request-based dm.
68 * One of these is allocated per request.
70 struct dm_rq_target_io {
71 struct mapped_device *md;
73 struct request *orig, clone;
79 * For request-based dm.
80 * One of these is allocated per bio.
82 struct dm_rq_clone_bio_info {
84 struct dm_rq_target_io *tio;
87 union map_info *dm_get_mapinfo(struct bio *bio)
89 if (bio && bio->bi_private)
90 return &((struct dm_target_io *)bio->bi_private)->info;
94 union map_info *dm_get_rq_mapinfo(struct request *rq)
96 if (rq && rq->end_io_data)
97 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
100 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
102 #define MINOR_ALLOCED ((void *)-1)
105 * Bits for the md->flags field.
107 #define DMF_BLOCK_IO_FOR_SUSPEND 0
108 #define DMF_SUSPENDED 1
110 #define DMF_FREEING 3
111 #define DMF_DELETING 4
112 #define DMF_NOFLUSH_SUSPENDING 5
113 #define DMF_QUEUE_IO_TO_THREAD 6
116 * Work processed by per-device workqueue.
118 struct mapped_device {
119 struct rw_semaphore io_lock;
120 struct mutex suspend_lock;
127 struct request_queue *queue;
129 /* Protect queue and type against concurrent access. */
130 struct mutex type_lock;
132 struct gendisk *disk;
138 * A list of ios that arrived while we were suspended.
141 wait_queue_head_t wait;
142 struct work_struct work;
143 struct bio_list deferred;
144 spinlock_t deferred_lock;
147 * An error from the barrier request currently being processed.
152 * Protect barrier_error from concurrent endio processing
153 * in request-based dm.
155 spinlock_t barrier_error_lock;
158 * Processing queue (flush/barriers)
160 struct workqueue_struct *wq;
161 struct work_struct barrier_work;
163 /* A pointer to the currently processing pre/post flush request */
164 struct request *flush_request;
167 * The current mapping.
169 struct dm_table *map;
172 * io objects are allocated from here.
183 wait_queue_head_t eventq;
185 struct list_head uevent_list;
186 spinlock_t uevent_lock; /* Protect access to uevent_list */
189 * freeze/thaw support require holding onto a super block
191 struct super_block *frozen_sb;
192 struct block_device *bdev;
194 /* forced geometry settings */
195 struct hd_geometry geometry;
197 /* For saving the address of __make_request for request based dm */
198 make_request_fn *saved_make_request_fn;
203 /* zero-length barrier that will be cloned and submitted to targets */
204 struct bio barrier_bio;
208 * For mempools pre-allocation at the table loading time.
210 struct dm_md_mempools {
217 static struct kmem_cache *_io_cache;
218 static struct kmem_cache *_tio_cache;
219 static struct kmem_cache *_rq_tio_cache;
220 static struct kmem_cache *_rq_bio_info_cache;
222 static int __init local_init(void)
226 /* allocate a slab for the dm_ios */
227 _io_cache = KMEM_CACHE(dm_io, 0);
231 /* allocate a slab for the target ios */
232 _tio_cache = KMEM_CACHE(dm_target_io, 0);
234 goto out_free_io_cache;
236 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
238 goto out_free_tio_cache;
240 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
241 if (!_rq_bio_info_cache)
242 goto out_free_rq_tio_cache;
244 r = dm_uevent_init();
246 goto out_free_rq_bio_info_cache;
249 r = register_blkdev(_major, _name);
251 goto out_uevent_exit;
260 out_free_rq_bio_info_cache:
261 kmem_cache_destroy(_rq_bio_info_cache);
262 out_free_rq_tio_cache:
263 kmem_cache_destroy(_rq_tio_cache);
265 kmem_cache_destroy(_tio_cache);
267 kmem_cache_destroy(_io_cache);
272 static void local_exit(void)
274 kmem_cache_destroy(_rq_bio_info_cache);
275 kmem_cache_destroy(_rq_tio_cache);
276 kmem_cache_destroy(_tio_cache);
277 kmem_cache_destroy(_io_cache);
278 unregister_blkdev(_major, _name);
283 DMINFO("cleaned up");
286 static int (*_inits[])(void) __initdata = {
296 static void (*_exits[])(void) = {
306 static int __init dm_init(void)
308 const int count = ARRAY_SIZE(_inits);
312 for (i = 0; i < count; i++) {
327 static void __exit dm_exit(void)
329 int i = ARRAY_SIZE(_exits);
336 * Block device functions
338 int dm_deleting_md(struct mapped_device *md)
340 return test_bit(DMF_DELETING, &md->flags);
343 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
345 struct mapped_device *md;
348 spin_lock(&_minor_lock);
350 md = bdev->bd_disk->private_data;
354 if (test_bit(DMF_FREEING, &md->flags) ||
355 dm_deleting_md(md)) {
361 atomic_inc(&md->open_count);
364 spin_unlock(&_minor_lock);
367 return md ? 0 : -ENXIO;
370 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
372 struct mapped_device *md = disk->private_data;
375 atomic_dec(&md->open_count);
382 int dm_open_count(struct mapped_device *md)
384 return atomic_read(&md->open_count);
388 * Guarantees nothing is using the device before it's deleted.
390 int dm_lock_for_deletion(struct mapped_device *md)
394 spin_lock(&_minor_lock);
396 if (dm_open_count(md))
399 set_bit(DMF_DELETING, &md->flags);
401 spin_unlock(&_minor_lock);
406 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
408 struct mapped_device *md = bdev->bd_disk->private_data;
410 return dm_get_geometry(md, geo);
413 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
414 unsigned int cmd, unsigned long arg)
416 struct mapped_device *md = bdev->bd_disk->private_data;
417 struct dm_table *map = dm_get_live_table(md);
418 struct dm_target *tgt;
421 if (!map || !dm_table_get_size(map))
424 /* We only support devices that have a single target */
425 if (dm_table_get_num_targets(map) != 1)
428 tgt = dm_table_get_target(map, 0);
430 if (dm_suspended_md(md)) {
435 if (tgt->type->ioctl)
436 r = tgt->type->ioctl(tgt, cmd, arg);
444 static struct dm_io *alloc_io(struct mapped_device *md)
446 return mempool_alloc(md->io_pool, GFP_NOIO);
449 static void free_io(struct mapped_device *md, struct dm_io *io)
451 mempool_free(io, md->io_pool);
454 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
456 mempool_free(tio, md->tio_pool);
459 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
462 return mempool_alloc(md->tio_pool, gfp_mask);
465 static void free_rq_tio(struct dm_rq_target_io *tio)
467 mempool_free(tio, tio->md->tio_pool);
470 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
472 return mempool_alloc(md->io_pool, GFP_ATOMIC);
475 static void free_bio_info(struct dm_rq_clone_bio_info *info)
477 mempool_free(info, info->tio->md->io_pool);
480 static int md_in_flight(struct mapped_device *md)
482 return atomic_read(&md->pending[READ]) +
483 atomic_read(&md->pending[WRITE]);
486 static void start_io_acct(struct dm_io *io)
488 struct mapped_device *md = io->md;
490 int rw = bio_data_dir(io->bio);
492 io->start_time = jiffies;
494 cpu = part_stat_lock();
495 part_round_stats(cpu, &dm_disk(md)->part0);
497 dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]);
500 static void end_io_acct(struct dm_io *io)
502 struct mapped_device *md = io->md;
503 struct bio *bio = io->bio;
504 unsigned long duration = jiffies - io->start_time;
506 int rw = bio_data_dir(bio);
508 cpu = part_stat_lock();
509 part_round_stats(cpu, &dm_disk(md)->part0);
510 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
514 * After this is decremented the bio must not be touched if it is
517 dm_disk(md)->part0.in_flight[rw] = pending =
518 atomic_dec_return(&md->pending[rw]);
519 pending += atomic_read(&md->pending[rw^0x1]);
521 /* nudge anyone waiting on suspend queue */
527 * Add the bio to the list of deferred io.
529 static void queue_io(struct mapped_device *md, struct bio *bio)
531 down_write(&md->io_lock);
533 spin_lock_irq(&md->deferred_lock);
534 bio_list_add(&md->deferred, bio);
535 spin_unlock_irq(&md->deferred_lock);
537 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
538 queue_work(md->wq, &md->work);
540 up_write(&md->io_lock);
544 * Everyone (including functions in this file), should use this
545 * function to access the md->map field, and make sure they call
546 * dm_table_put() when finished.
548 struct dm_table *dm_get_live_table(struct mapped_device *md)
553 read_lock_irqsave(&md->map_lock, flags);
557 read_unlock_irqrestore(&md->map_lock, flags);
563 * Get the geometry associated with a dm device
565 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
573 * Set the geometry of a device.
575 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
577 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
579 if (geo->start > sz) {
580 DMWARN("Start sector is beyond the geometry limits.");
589 /*-----------------------------------------------------------------
591 * A more elegant soln is in the works that uses the queue
592 * merge fn, unfortunately there are a couple of changes to
593 * the block layer that I want to make for this. So in the
594 * interests of getting something for people to use I give
595 * you this clearly demarcated crap.
596 *---------------------------------------------------------------*/
598 static int __noflush_suspending(struct mapped_device *md)
600 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
604 * Decrements the number of outstanding ios that a bio has been
605 * cloned into, completing the original io if necc.
607 static void dec_pending(struct dm_io *io, int error)
612 struct mapped_device *md = io->md;
614 /* Push-back supersedes any I/O errors */
615 if (unlikely(error)) {
616 spin_lock_irqsave(&io->endio_lock, flags);
617 if (!(io->error > 0 && __noflush_suspending(md)))
619 spin_unlock_irqrestore(&io->endio_lock, flags);
622 if (atomic_dec_and_test(&io->io_count)) {
623 if (io->error == DM_ENDIO_REQUEUE) {
625 * Target requested pushing back the I/O.
627 spin_lock_irqsave(&md->deferred_lock, flags);
628 if (__noflush_suspending(md)) {
629 if (!(io->bio->bi_rw & REQ_HARDBARRIER))
630 bio_list_add_head(&md->deferred,
633 /* noflush suspend was interrupted. */
635 spin_unlock_irqrestore(&md->deferred_lock, flags);
638 io_error = io->error;
641 if (bio->bi_rw & REQ_HARDBARRIER) {
643 * There can be just one barrier request so we use
644 * a per-device variable for error reporting.
645 * Note that you can't touch the bio after end_io_acct
647 * We ignore -EOPNOTSUPP for empty flush reported by
648 * underlying devices. We assume that if the device
649 * doesn't support empty barriers, it doesn't need
650 * cache flushing commands.
652 if (!md->barrier_error &&
653 !(bio_empty_barrier(bio) && io_error == -EOPNOTSUPP))
654 md->barrier_error = io_error;
661 if (io_error != DM_ENDIO_REQUEUE) {
662 trace_block_bio_complete(md->queue, bio);
664 bio_endio(bio, io_error);
670 static void clone_endio(struct bio *bio, int error)
673 struct dm_target_io *tio = bio->bi_private;
674 struct dm_io *io = tio->io;
675 struct mapped_device *md = tio->io->md;
676 dm_endio_fn endio = tio->ti->type->end_io;
678 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
682 r = endio(tio->ti, bio, error, &tio->info);
683 if (r < 0 || r == DM_ENDIO_REQUEUE)
685 * error and requeue request are handled
689 else if (r == DM_ENDIO_INCOMPLETE)
690 /* The target will handle the io */
693 DMWARN("unimplemented target endio return value: %d", r);
699 * Store md for cleanup instead of tio which is about to get freed.
701 bio->bi_private = md->bs;
705 dec_pending(io, error);
709 * Partial completion handling for request-based dm
711 static void end_clone_bio(struct bio *clone, int error)
713 struct dm_rq_clone_bio_info *info = clone->bi_private;
714 struct dm_rq_target_io *tio = info->tio;
715 struct bio *bio = info->orig;
716 unsigned int nr_bytes = info->orig->bi_size;
722 * An error has already been detected on the request.
723 * Once error occurred, just let clone->end_io() handle
729 * Don't notice the error to the upper layer yet.
730 * The error handling decision is made by the target driver,
731 * when the request is completed.
738 * I/O for the bio successfully completed.
739 * Notice the data completion to the upper layer.
743 * bios are processed from the head of the list.
744 * So the completing bio should always be rq->bio.
745 * If it's not, something wrong is happening.
747 if (tio->orig->bio != bio)
748 DMERR("bio completion is going in the middle of the request");
751 * Update the original request.
752 * Do not use blk_end_request() here, because it may complete
753 * the original request before the clone, and break the ordering.
755 blk_update_request(tio->orig, 0, nr_bytes);
758 static void store_barrier_error(struct mapped_device *md, int error)
762 spin_lock_irqsave(&md->barrier_error_lock, flags);
764 * Basically, the first error is taken, but:
765 * -EOPNOTSUPP supersedes any I/O error.
766 * Requeue request supersedes any I/O error but -EOPNOTSUPP.
768 if (!md->barrier_error || error == -EOPNOTSUPP ||
769 (md->barrier_error != -EOPNOTSUPP &&
770 error == DM_ENDIO_REQUEUE))
771 md->barrier_error = error;
772 spin_unlock_irqrestore(&md->barrier_error_lock, flags);
776 * Don't touch any member of the md after calling this function because
777 * the md may be freed in dm_put() at the end of this function.
778 * Or do dm_get() before calling this function and dm_put() later.
780 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
782 atomic_dec(&md->pending[rw]);
784 /* nudge anyone waiting on suspend queue */
785 if (!md_in_flight(md))
789 blk_run_queue(md->queue);
792 * dm_put() must be at the end of this function. See the comment above
797 static void free_rq_clone(struct request *clone)
799 struct dm_rq_target_io *tio = clone->end_io_data;
801 blk_rq_unprep_clone(clone);
806 * Complete the clone and the original request.
807 * Must be called without queue lock.
809 static void dm_end_request(struct request *clone, int error)
811 int rw = rq_data_dir(clone);
813 bool is_barrier = clone->cmd_flags & REQ_HARDBARRIER;
814 struct dm_rq_target_io *tio = clone->end_io_data;
815 struct mapped_device *md = tio->md;
816 struct request *rq = tio->orig;
818 if (rq->cmd_type == REQ_TYPE_BLOCK_PC && !is_barrier) {
819 rq->errors = clone->errors;
820 rq->resid_len = clone->resid_len;
824 * We are using the sense buffer of the original
826 * So setting the length of the sense data is enough.
828 rq->sense_len = clone->sense_len;
831 free_rq_clone(clone);
833 if (unlikely(is_barrier)) {
835 store_barrier_error(md, error);
838 blk_end_request_all(rq, error);
840 rq_completed(md, rw, run_queue);
843 static void dm_unprep_request(struct request *rq)
845 struct request *clone = rq->special;
848 rq->cmd_flags &= ~REQ_DONTPREP;
850 free_rq_clone(clone);
854 * Requeue the original request of a clone.
856 void dm_requeue_unmapped_request(struct request *clone)
858 int rw = rq_data_dir(clone);
859 struct dm_rq_target_io *tio = clone->end_io_data;
860 struct mapped_device *md = tio->md;
861 struct request *rq = tio->orig;
862 struct request_queue *q = rq->q;
865 if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) {
867 * Barrier clones share an original request.
868 * Leave it to dm_end_request(), which handles this special
871 dm_end_request(clone, DM_ENDIO_REQUEUE);
875 dm_unprep_request(rq);
877 spin_lock_irqsave(q->queue_lock, flags);
878 if (elv_queue_empty(q))
880 blk_requeue_request(q, rq);
881 spin_unlock_irqrestore(q->queue_lock, flags);
883 rq_completed(md, rw, 0);
885 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
887 static void __stop_queue(struct request_queue *q)
892 static void stop_queue(struct request_queue *q)
896 spin_lock_irqsave(q->queue_lock, flags);
898 spin_unlock_irqrestore(q->queue_lock, flags);
901 static void __start_queue(struct request_queue *q)
903 if (blk_queue_stopped(q))
907 static void start_queue(struct request_queue *q)
911 spin_lock_irqsave(q->queue_lock, flags);
913 spin_unlock_irqrestore(q->queue_lock, flags);
916 static void dm_done(struct request *clone, int error, bool mapped)
919 struct dm_rq_target_io *tio = clone->end_io_data;
920 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
922 if (mapped && rq_end_io)
923 r = rq_end_io(tio->ti, clone, error, &tio->info);
926 /* The target wants to complete the I/O */
927 dm_end_request(clone, r);
928 else if (r == DM_ENDIO_INCOMPLETE)
929 /* The target will handle the I/O */
931 else if (r == DM_ENDIO_REQUEUE)
932 /* The target wants to requeue the I/O */
933 dm_requeue_unmapped_request(clone);
935 DMWARN("unimplemented target endio return value: %d", r);
941 * Request completion handler for request-based dm
943 static void dm_softirq_done(struct request *rq)
946 struct request *clone = rq->completion_data;
947 struct dm_rq_target_io *tio = clone->end_io_data;
949 if (rq->cmd_flags & REQ_FAILED)
952 dm_done(clone, tio->error, mapped);
956 * Complete the clone and the original request with the error status
957 * through softirq context.
959 static void dm_complete_request(struct request *clone, int error)
961 struct dm_rq_target_io *tio = clone->end_io_data;
962 struct request *rq = tio->orig;
964 if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) {
966 * Barrier clones share an original request. So can't use
967 * softirq_done with the original.
968 * Pass the clone to dm_done() directly in this special case.
969 * It is safe (even if clone->q->queue_lock is held here)
970 * because there is no I/O dispatching during the completion
973 dm_done(clone, error, true);
978 rq->completion_data = clone;
979 blk_complete_request(rq);
983 * Complete the not-mapped clone and the original request with the error status
984 * through softirq context.
985 * Target's rq_end_io() function isn't called.
986 * This may be used when the target's map_rq() function fails.
988 void dm_kill_unmapped_request(struct request *clone, int error)
990 struct dm_rq_target_io *tio = clone->end_io_data;
991 struct request *rq = tio->orig;
993 if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) {
995 * Barrier clones share an original request.
996 * Leave it to dm_end_request(), which handles this special
1000 dm_end_request(clone, error);
1004 rq->cmd_flags |= REQ_FAILED;
1005 dm_complete_request(clone, error);
1007 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1010 * Called with the queue lock held
1012 static void end_clone_request(struct request *clone, int error)
1015 * For just cleaning up the information of the queue in which
1016 * the clone was dispatched.
1017 * The clone is *NOT* freed actually here because it is alloced from
1018 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1020 __blk_put_request(clone->q, clone);
1023 * Actual request completion is done in a softirq context which doesn't
1024 * hold the queue lock. Otherwise, deadlock could occur because:
1025 * - another request may be submitted by the upper level driver
1026 * of the stacking during the completion
1027 * - the submission which requires queue lock may be done
1028 * against this queue
1030 dm_complete_request(clone, error);
1033 static sector_t max_io_len(struct mapped_device *md,
1034 sector_t sector, struct dm_target *ti)
1036 sector_t offset = sector - ti->begin;
1037 sector_t len = ti->len - offset;
1040 * Does the target need to split even further ?
1044 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
1053 static void __map_bio(struct dm_target *ti, struct bio *clone,
1054 struct dm_target_io *tio)
1058 struct mapped_device *md;
1060 clone->bi_end_io = clone_endio;
1061 clone->bi_private = tio;
1064 * Map the clone. If r == 0 we don't need to do
1065 * anything, the target has assumed ownership of
1068 atomic_inc(&tio->io->io_count);
1069 sector = clone->bi_sector;
1070 r = ti->type->map(ti, clone, &tio->info);
1071 if (r == DM_MAPIO_REMAPPED) {
1072 /* the bio has been remapped so dispatch it */
1074 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
1075 tio->io->bio->bi_bdev->bd_dev, sector);
1077 generic_make_request(clone);
1078 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1079 /* error the io and bail out, or requeue it if needed */
1081 dec_pending(tio->io, r);
1083 * Store bio_set for cleanup.
1085 clone->bi_private = md->bs;
1089 DMWARN("unimplemented target map return value: %d", r);
1095 struct mapped_device *md;
1096 struct dm_table *map;
1100 sector_t sector_count;
1104 static void dm_bio_destructor(struct bio *bio)
1106 struct bio_set *bs = bio->bi_private;
1112 * Creates a little bio that is just does part of a bvec.
1114 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1115 unsigned short idx, unsigned int offset,
1116 unsigned int len, struct bio_set *bs)
1119 struct bio_vec *bv = bio->bi_io_vec + idx;
1121 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1122 clone->bi_destructor = dm_bio_destructor;
1123 *clone->bi_io_vec = *bv;
1125 clone->bi_sector = sector;
1126 clone->bi_bdev = bio->bi_bdev;
1127 clone->bi_rw = bio->bi_rw & ~REQ_HARDBARRIER;
1129 clone->bi_size = to_bytes(len);
1130 clone->bi_io_vec->bv_offset = offset;
1131 clone->bi_io_vec->bv_len = clone->bi_size;
1132 clone->bi_flags |= 1 << BIO_CLONED;
1134 if (bio_integrity(bio)) {
1135 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1136 bio_integrity_trim(clone,
1137 bio_sector_offset(bio, idx, offset), len);
1144 * Creates a bio that consists of range of complete bvecs.
1146 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1147 unsigned short idx, unsigned short bv_count,
1148 unsigned int len, struct bio_set *bs)
1152 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1153 __bio_clone(clone, bio);
1154 clone->bi_rw &= ~REQ_HARDBARRIER;
1155 clone->bi_destructor = dm_bio_destructor;
1156 clone->bi_sector = sector;
1157 clone->bi_idx = idx;
1158 clone->bi_vcnt = idx + bv_count;
1159 clone->bi_size = to_bytes(len);
1160 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1162 if (bio_integrity(bio)) {
1163 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1165 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1166 bio_integrity_trim(clone,
1167 bio_sector_offset(bio, idx, 0), len);
1173 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1174 struct dm_target *ti)
1176 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1180 memset(&tio->info, 0, sizeof(tio->info));
1185 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1186 unsigned request_nr)
1188 struct dm_target_io *tio = alloc_tio(ci, ti);
1191 tio->info.target_request_nr = request_nr;
1194 * Discard requests require the bio's inline iovecs be initialized.
1195 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1196 * and discard, so no need for concern about wasted bvec allocations.
1198 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1199 __bio_clone(clone, ci->bio);
1200 clone->bi_destructor = dm_bio_destructor;
1202 __map_bio(ti, clone, tio);
1205 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1206 unsigned num_requests)
1208 unsigned request_nr;
1210 for (request_nr = 0; request_nr < num_requests; request_nr++)
1211 __issue_target_request(ci, ti, request_nr);
1214 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1216 unsigned target_nr = 0;
1217 struct dm_target *ti;
1219 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1220 __issue_target_requests(ci, ti, ti->num_flush_requests);
1222 ci->sector_count = 0;
1228 * Perform all io with a single clone.
1230 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1232 struct bio *clone, *bio = ci->bio;
1233 struct dm_target_io *tio;
1235 tio = alloc_tio(ci, ti);
1236 clone = clone_bio(bio, ci->sector, ci->idx,
1237 bio->bi_vcnt - ci->idx, ci->sector_count,
1239 __map_bio(ti, clone, tio);
1240 ci->sector_count = 0;
1243 static int __clone_and_map_discard(struct clone_info *ci)
1245 struct dm_target *ti;
1248 ti = dm_table_find_target(ci->map, ci->sector);
1249 if (!dm_target_is_valid(ti))
1253 * Even though the device advertised discard support,
1254 * reconfiguration might have changed that since the
1255 * check was performed.
1258 if (!ti->num_discard_requests)
1261 max = max_io_len(ci->md, ci->sector, ti);
1263 if (ci->sector_count > max)
1265 * FIXME: Handle a discard that spans two or more targets.
1269 __issue_target_requests(ci, ti, ti->num_discard_requests);
1271 ci->sector_count = 0;
1276 static int __clone_and_map(struct clone_info *ci)
1278 struct bio *clone, *bio = ci->bio;
1279 struct dm_target *ti;
1280 sector_t len = 0, max;
1281 struct dm_target_io *tio;
1283 if (unlikely(bio_empty_barrier(bio)))
1284 return __clone_and_map_empty_barrier(ci);
1286 if (unlikely(bio->bi_rw & REQ_DISCARD))
1287 return __clone_and_map_discard(ci);
1289 ti = dm_table_find_target(ci->map, ci->sector);
1290 if (!dm_target_is_valid(ti))
1293 max = max_io_len(ci->md, ci->sector, ti);
1295 if (ci->sector_count <= max) {
1297 * Optimise for the simple case where we can do all of
1298 * the remaining io with a single clone.
1300 __clone_and_map_simple(ci, ti);
1302 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1304 * There are some bvecs that don't span targets.
1305 * Do as many of these as possible.
1308 sector_t remaining = max;
1311 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1312 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1314 if (bv_len > remaining)
1317 remaining -= bv_len;
1321 tio = alloc_tio(ci, ti);
1322 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1324 __map_bio(ti, clone, tio);
1327 ci->sector_count -= len;
1332 * Handle a bvec that must be split between two or more targets.
1334 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1335 sector_t remaining = to_sector(bv->bv_len);
1336 unsigned int offset = 0;
1340 ti = dm_table_find_target(ci->map, ci->sector);
1341 if (!dm_target_is_valid(ti))
1344 max = max_io_len(ci->md, ci->sector, ti);
1347 len = min(remaining, max);
1349 tio = alloc_tio(ci, ti);
1350 clone = split_bvec(bio, ci->sector, ci->idx,
1351 bv->bv_offset + offset, len,
1354 __map_bio(ti, clone, tio);
1357 ci->sector_count -= len;
1358 offset += to_bytes(len);
1359 } while (remaining -= len);
1368 * Split the bio into several clones and submit it to targets.
1370 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1372 struct clone_info ci;
1375 ci.map = dm_get_live_table(md);
1376 if (unlikely(!ci.map)) {
1377 if (!(bio->bi_rw & REQ_HARDBARRIER))
1380 if (!md->barrier_error)
1381 md->barrier_error = -EIO;
1387 ci.io = alloc_io(md);
1389 atomic_set(&ci.io->io_count, 1);
1392 spin_lock_init(&ci.io->endio_lock);
1393 ci.sector = bio->bi_sector;
1394 ci.sector_count = bio_sectors(bio);
1395 if (unlikely(bio_empty_barrier(bio)))
1396 ci.sector_count = 1;
1397 ci.idx = bio->bi_idx;
1399 start_io_acct(ci.io);
1400 while (ci.sector_count && !error)
1401 error = __clone_and_map(&ci);
1403 /* drop the extra reference count */
1404 dec_pending(ci.io, error);
1405 dm_table_put(ci.map);
1407 /*-----------------------------------------------------------------
1409 *---------------------------------------------------------------*/
1411 static int dm_merge_bvec(struct request_queue *q,
1412 struct bvec_merge_data *bvm,
1413 struct bio_vec *biovec)
1415 struct mapped_device *md = q->queuedata;
1416 struct dm_table *map = dm_get_live_table(md);
1417 struct dm_target *ti;
1418 sector_t max_sectors;
1424 ti = dm_table_find_target(map, bvm->bi_sector);
1425 if (!dm_target_is_valid(ti))
1429 * Find maximum amount of I/O that won't need splitting
1431 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
1432 (sector_t) BIO_MAX_SECTORS);
1433 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1438 * merge_bvec_fn() returns number of bytes
1439 * it can accept at this offset
1440 * max is precomputed maximal io size
1442 if (max_size && ti->type->merge)
1443 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1445 * If the target doesn't support merge method and some of the devices
1446 * provided their merge_bvec method (we know this by looking at
1447 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1448 * entries. So always set max_size to 0, and the code below allows
1451 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1460 * Always allow an entire first page
1462 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1463 max_size = biovec->bv_len;
1469 * The request function that just remaps the bio built up by
1472 static int _dm_request(struct request_queue *q, struct bio *bio)
1474 int rw = bio_data_dir(bio);
1475 struct mapped_device *md = q->queuedata;
1478 down_read(&md->io_lock);
1480 cpu = part_stat_lock();
1481 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1482 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1486 * If we're suspended or the thread is processing barriers
1487 * we have to queue this io for later.
1489 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1490 unlikely(bio->bi_rw & REQ_HARDBARRIER)) {
1491 up_read(&md->io_lock);
1493 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1494 bio_rw(bio) == READA) {
1504 __split_and_process_bio(md, bio);
1505 up_read(&md->io_lock);
1509 static int dm_make_request(struct request_queue *q, struct bio *bio)
1511 struct mapped_device *md = q->queuedata;
1513 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1516 static int dm_request_based(struct mapped_device *md)
1518 return blk_queue_stackable(md->queue);
1521 static int dm_request(struct request_queue *q, struct bio *bio)
1523 struct mapped_device *md = q->queuedata;
1525 if (dm_request_based(md))
1526 return dm_make_request(q, bio);
1528 return _dm_request(q, bio);
1531 static bool dm_rq_is_flush_request(struct request *rq)
1533 if (rq->cmd_flags & REQ_FLUSH)
1539 void dm_dispatch_request(struct request *rq)
1543 if (blk_queue_io_stat(rq->q))
1544 rq->cmd_flags |= REQ_IO_STAT;
1546 rq->start_time = jiffies;
1547 r = blk_insert_cloned_request(rq->q, rq);
1549 dm_complete_request(rq, r);
1551 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1553 static void dm_rq_bio_destructor(struct bio *bio)
1555 struct dm_rq_clone_bio_info *info = bio->bi_private;
1556 struct mapped_device *md = info->tio->md;
1558 free_bio_info(info);
1559 bio_free(bio, md->bs);
1562 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1565 struct dm_rq_target_io *tio = data;
1566 struct mapped_device *md = tio->md;
1567 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1572 info->orig = bio_orig;
1574 bio->bi_end_io = end_clone_bio;
1575 bio->bi_private = info;
1576 bio->bi_destructor = dm_rq_bio_destructor;
1581 static int setup_clone(struct request *clone, struct request *rq,
1582 struct dm_rq_target_io *tio)
1586 if (dm_rq_is_flush_request(rq)) {
1587 blk_rq_init(NULL, clone);
1588 clone->cmd_type = REQ_TYPE_FS;
1589 clone->cmd_flags |= (REQ_HARDBARRIER | WRITE);
1591 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1592 dm_rq_bio_constructor, tio);
1596 clone->cmd = rq->cmd;
1597 clone->cmd_len = rq->cmd_len;
1598 clone->sense = rq->sense;
1599 clone->buffer = rq->buffer;
1602 clone->end_io = end_clone_request;
1603 clone->end_io_data = tio;
1608 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1611 struct request *clone;
1612 struct dm_rq_target_io *tio;
1614 tio = alloc_rq_tio(md, gfp_mask);
1622 memset(&tio->info, 0, sizeof(tio->info));
1624 clone = &tio->clone;
1625 if (setup_clone(clone, rq, tio)) {
1635 * Called with the queue lock held.
1637 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1639 struct mapped_device *md = q->queuedata;
1640 struct request *clone;
1642 if (unlikely(dm_rq_is_flush_request(rq)))
1645 if (unlikely(rq->special)) {
1646 DMWARN("Already has something in rq->special.");
1647 return BLKPREP_KILL;
1650 clone = clone_rq(rq, md, GFP_ATOMIC);
1652 return BLKPREP_DEFER;
1654 rq->special = clone;
1655 rq->cmd_flags |= REQ_DONTPREP;
1662 * 0 : the request has been processed (not requeued)
1663 * !0 : the request has been requeued
1665 static int map_request(struct dm_target *ti, struct request *clone,
1666 struct mapped_device *md)
1668 int r, requeued = 0;
1669 struct dm_rq_target_io *tio = clone->end_io_data;
1672 * Hold the md reference here for the in-flight I/O.
1673 * We can't rely on the reference count by device opener,
1674 * because the device may be closed during the request completion
1675 * when all bios are completed.
1676 * See the comment in rq_completed() too.
1681 r = ti->type->map_rq(ti, clone, &tio->info);
1683 case DM_MAPIO_SUBMITTED:
1684 /* The target has taken the I/O to submit by itself later */
1686 case DM_MAPIO_REMAPPED:
1687 /* The target has remapped the I/O so dispatch it */
1688 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1689 blk_rq_pos(tio->orig));
1690 dm_dispatch_request(clone);
1692 case DM_MAPIO_REQUEUE:
1693 /* The target wants to requeue the I/O */
1694 dm_requeue_unmapped_request(clone);
1699 DMWARN("unimplemented target map return value: %d", r);
1703 /* The target wants to complete the I/O */
1704 dm_kill_unmapped_request(clone, r);
1712 * q->request_fn for request-based dm.
1713 * Called with the queue lock held.
1715 static void dm_request_fn(struct request_queue *q)
1717 struct mapped_device *md = q->queuedata;
1718 struct dm_table *map = dm_get_live_table(md);
1719 struct dm_target *ti;
1720 struct request *rq, *clone;
1723 * For suspend, check blk_queue_stopped() and increment
1724 * ->pending within a single queue_lock not to increment the
1725 * number of in-flight I/Os after the queue is stopped in
1728 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1729 rq = blk_peek_request(q);
1733 if (unlikely(dm_rq_is_flush_request(rq))) {
1734 BUG_ON(md->flush_request);
1735 md->flush_request = rq;
1736 blk_start_request(rq);
1737 queue_work(md->wq, &md->barrier_work);
1741 ti = dm_table_find_target(map, blk_rq_pos(rq));
1742 if (ti->type->busy && ti->type->busy(ti))
1745 blk_start_request(rq);
1746 clone = rq->special;
1747 atomic_inc(&md->pending[rq_data_dir(clone)]);
1749 spin_unlock(q->queue_lock);
1750 if (map_request(ti, clone, md))
1753 spin_lock_irq(q->queue_lock);
1759 spin_lock_irq(q->queue_lock);
1762 if (!elv_queue_empty(q))
1763 /* Some requests still remain, retry later */
1772 int dm_underlying_device_busy(struct request_queue *q)
1774 return blk_lld_busy(q);
1776 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1778 static int dm_lld_busy(struct request_queue *q)
1781 struct mapped_device *md = q->queuedata;
1782 struct dm_table *map = dm_get_live_table(md);
1784 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1787 r = dm_table_any_busy_target(map);
1794 static void dm_unplug_all(struct request_queue *q)
1796 struct mapped_device *md = q->queuedata;
1797 struct dm_table *map = dm_get_live_table(md);
1800 if (dm_request_based(md))
1801 generic_unplug_device(q);
1803 dm_table_unplug_all(map);
1808 static int dm_any_congested(void *congested_data, int bdi_bits)
1811 struct mapped_device *md = congested_data;
1812 struct dm_table *map;
1814 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1815 map = dm_get_live_table(md);
1818 * Request-based dm cares about only own queue for
1819 * the query about congestion status of request_queue
1821 if (dm_request_based(md))
1822 r = md->queue->backing_dev_info.state &
1825 r = dm_table_any_congested(map, bdi_bits);
1834 /*-----------------------------------------------------------------
1835 * An IDR is used to keep track of allocated minor numbers.
1836 *---------------------------------------------------------------*/
1837 static DEFINE_IDR(_minor_idr);
1839 static void free_minor(int minor)
1841 spin_lock(&_minor_lock);
1842 idr_remove(&_minor_idr, minor);
1843 spin_unlock(&_minor_lock);
1847 * See if the device with a specific minor # is free.
1849 static int specific_minor(int minor)
1853 if (minor >= (1 << MINORBITS))
1856 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1860 spin_lock(&_minor_lock);
1862 if (idr_find(&_minor_idr, minor)) {
1867 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1872 idr_remove(&_minor_idr, m);
1878 spin_unlock(&_minor_lock);
1882 static int next_free_minor(int *minor)
1886 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1890 spin_lock(&_minor_lock);
1892 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1896 if (m >= (1 << MINORBITS)) {
1897 idr_remove(&_minor_idr, m);
1905 spin_unlock(&_minor_lock);
1909 static const struct block_device_operations dm_blk_dops;
1911 static void dm_wq_work(struct work_struct *work);
1912 static void dm_rq_barrier_work(struct work_struct *work);
1914 static void dm_init_md_queue(struct mapped_device *md)
1917 * Request-based dm devices cannot be stacked on top of bio-based dm
1918 * devices. The type of this dm device has not been decided yet.
1919 * The type is decided at the first table loading time.
1920 * To prevent problematic device stacking, clear the queue flag
1921 * for request stacking support until then.
1923 * This queue is new, so no concurrency on the queue_flags.
1925 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1927 md->queue->queuedata = md;
1928 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1929 md->queue->backing_dev_info.congested_data = md;
1930 blk_queue_make_request(md->queue, dm_request);
1931 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1932 md->queue->unplug_fn = dm_unplug_all;
1933 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1937 * Allocate and initialise a blank device with a given minor.
1939 static struct mapped_device *alloc_dev(int minor)
1942 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1946 DMWARN("unable to allocate device, out of memory.");
1950 if (!try_module_get(THIS_MODULE))
1951 goto bad_module_get;
1953 /* get a minor number for the dev */
1954 if (minor == DM_ANY_MINOR)
1955 r = next_free_minor(&minor);
1957 r = specific_minor(minor);
1961 md->type = DM_TYPE_NONE;
1962 init_rwsem(&md->io_lock);
1963 mutex_init(&md->suspend_lock);
1964 mutex_init(&md->type_lock);
1965 spin_lock_init(&md->deferred_lock);
1966 spin_lock_init(&md->barrier_error_lock);
1967 rwlock_init(&md->map_lock);
1968 atomic_set(&md->holders, 1);
1969 atomic_set(&md->open_count, 0);
1970 atomic_set(&md->event_nr, 0);
1971 atomic_set(&md->uevent_seq, 0);
1972 INIT_LIST_HEAD(&md->uevent_list);
1973 spin_lock_init(&md->uevent_lock);
1975 md->queue = blk_alloc_queue(GFP_KERNEL);
1979 dm_init_md_queue(md);
1981 md->disk = alloc_disk(1);
1985 atomic_set(&md->pending[0], 0);
1986 atomic_set(&md->pending[1], 0);
1987 init_waitqueue_head(&md->wait);
1988 INIT_WORK(&md->work, dm_wq_work);
1989 INIT_WORK(&md->barrier_work, dm_rq_barrier_work);
1990 init_waitqueue_head(&md->eventq);
1992 md->disk->major = _major;
1993 md->disk->first_minor = minor;
1994 md->disk->fops = &dm_blk_dops;
1995 md->disk->queue = md->queue;
1996 md->disk->private_data = md;
1997 sprintf(md->disk->disk_name, "dm-%d", minor);
1999 format_dev_t(md->name, MKDEV(_major, minor));
2001 md->wq = create_singlethread_workqueue("kdmflush");
2005 md->bdev = bdget_disk(md->disk, 0);
2009 /* Populate the mapping, nobody knows we exist yet */
2010 spin_lock(&_minor_lock);
2011 old_md = idr_replace(&_minor_idr, md, minor);
2012 spin_unlock(&_minor_lock);
2014 BUG_ON(old_md != MINOR_ALLOCED);
2019 destroy_workqueue(md->wq);
2021 del_gendisk(md->disk);
2024 blk_cleanup_queue(md->queue);
2028 module_put(THIS_MODULE);
2034 static void unlock_fs(struct mapped_device *md);
2036 static void free_dev(struct mapped_device *md)
2038 int minor = MINOR(disk_devt(md->disk));
2042 destroy_workqueue(md->wq);
2044 mempool_destroy(md->tio_pool);
2046 mempool_destroy(md->io_pool);
2048 bioset_free(md->bs);
2049 blk_integrity_unregister(md->disk);
2050 del_gendisk(md->disk);
2053 spin_lock(&_minor_lock);
2054 md->disk->private_data = NULL;
2055 spin_unlock(&_minor_lock);
2058 blk_cleanup_queue(md->queue);
2059 module_put(THIS_MODULE);
2063 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2065 struct dm_md_mempools *p;
2067 if (md->io_pool && md->tio_pool && md->bs)
2068 /* the md already has necessary mempools */
2071 p = dm_table_get_md_mempools(t);
2072 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
2074 md->io_pool = p->io_pool;
2076 md->tio_pool = p->tio_pool;
2082 /* mempool bind completed, now no need any mempools in the table */
2083 dm_table_free_md_mempools(t);
2087 * Bind a table to the device.
2089 static void event_callback(void *context)
2091 unsigned long flags;
2093 struct mapped_device *md = (struct mapped_device *) context;
2095 spin_lock_irqsave(&md->uevent_lock, flags);
2096 list_splice_init(&md->uevent_list, &uevents);
2097 spin_unlock_irqrestore(&md->uevent_lock, flags);
2099 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2101 atomic_inc(&md->event_nr);
2102 wake_up(&md->eventq);
2105 static void __set_size(struct mapped_device *md, sector_t size)
2107 set_capacity(md->disk, size);
2109 mutex_lock(&md->bdev->bd_inode->i_mutex);
2110 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2111 mutex_unlock(&md->bdev->bd_inode->i_mutex);
2115 * Returns old map, which caller must destroy.
2117 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2118 struct queue_limits *limits)
2120 struct dm_table *old_map;
2121 struct request_queue *q = md->queue;
2123 unsigned long flags;
2125 size = dm_table_get_size(t);
2128 * Wipe any geometry if the size of the table changed.
2130 if (size != get_capacity(md->disk))
2131 memset(&md->geometry, 0, sizeof(md->geometry));
2133 __set_size(md, size);
2135 dm_table_event_callback(t, event_callback, md);
2138 * The queue hasn't been stopped yet, if the old table type wasn't
2139 * for request-based during suspension. So stop it to prevent
2140 * I/O mapping before resume.
2141 * This must be done before setting the queue restrictions,
2142 * because request-based dm may be run just after the setting.
2144 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2147 __bind_mempools(md, t);
2149 write_lock_irqsave(&md->map_lock, flags);
2152 dm_table_set_restrictions(t, q, limits);
2153 write_unlock_irqrestore(&md->map_lock, flags);
2159 * Returns unbound table for the caller to free.
2161 static struct dm_table *__unbind(struct mapped_device *md)
2163 struct dm_table *map = md->map;
2164 unsigned long flags;
2169 dm_table_event_callback(map, NULL, NULL);
2170 write_lock_irqsave(&md->map_lock, flags);
2172 write_unlock_irqrestore(&md->map_lock, flags);
2178 * Constructor for a new device.
2180 int dm_create(int minor, struct mapped_device **result)
2182 struct mapped_device *md;
2184 md = alloc_dev(minor);
2195 * Functions to manage md->type.
2196 * All are required to hold md->type_lock.
2198 void dm_lock_md_type(struct mapped_device *md)
2200 mutex_lock(&md->type_lock);
2203 void dm_unlock_md_type(struct mapped_device *md)
2205 mutex_unlock(&md->type_lock);
2208 void dm_set_md_type(struct mapped_device *md, unsigned type)
2213 unsigned dm_get_md_type(struct mapped_device *md)
2219 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2221 static int dm_init_request_based_queue(struct mapped_device *md)
2223 struct request_queue *q = NULL;
2225 if (md->queue->elevator)
2228 /* Fully initialize the queue */
2229 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2234 md->saved_make_request_fn = md->queue->make_request_fn;
2235 dm_init_md_queue(md);
2236 blk_queue_softirq_done(md->queue, dm_softirq_done);
2237 blk_queue_prep_rq(md->queue, dm_prep_fn);
2238 blk_queue_lld_busy(md->queue, dm_lld_busy);
2239 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN_FLUSH);
2241 elv_register_queue(md->queue);
2247 * Setup the DM device's queue based on md's type
2249 int dm_setup_md_queue(struct mapped_device *md)
2251 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2252 !dm_init_request_based_queue(md)) {
2253 DMWARN("Cannot initialize queue for request-based mapped device");
2260 static struct mapped_device *dm_find_md(dev_t dev)
2262 struct mapped_device *md;
2263 unsigned minor = MINOR(dev);
2265 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2268 spin_lock(&_minor_lock);
2270 md = idr_find(&_minor_idr, minor);
2271 if (md && (md == MINOR_ALLOCED ||
2272 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2273 dm_deleting_md(md) ||
2274 test_bit(DMF_FREEING, &md->flags))) {
2280 spin_unlock(&_minor_lock);
2285 struct mapped_device *dm_get_md(dev_t dev)
2287 struct mapped_device *md = dm_find_md(dev);
2295 void *dm_get_mdptr(struct mapped_device *md)
2297 return md->interface_ptr;
2300 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2302 md->interface_ptr = ptr;
2305 void dm_get(struct mapped_device *md)
2307 atomic_inc(&md->holders);
2308 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2311 const char *dm_device_name(struct mapped_device *md)
2315 EXPORT_SYMBOL_GPL(dm_device_name);
2317 static void __dm_destroy(struct mapped_device *md, bool wait)
2319 struct dm_table *map;
2323 spin_lock(&_minor_lock);
2324 map = dm_get_live_table(md);
2325 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2326 set_bit(DMF_FREEING, &md->flags);
2327 spin_unlock(&_minor_lock);
2329 if (!dm_suspended_md(md)) {
2330 dm_table_presuspend_targets(map);
2331 dm_table_postsuspend_targets(map);
2335 * Rare, but there may be I/O requests still going to complete,
2336 * for example. Wait for all references to disappear.
2337 * No one should increment the reference count of the mapped_device,
2338 * after the mapped_device state becomes DMF_FREEING.
2341 while (atomic_read(&md->holders))
2343 else if (atomic_read(&md->holders))
2344 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2345 dm_device_name(md), atomic_read(&md->holders));
2349 dm_table_destroy(__unbind(md));
2353 void dm_destroy(struct mapped_device *md)
2355 __dm_destroy(md, true);
2358 void dm_destroy_immediate(struct mapped_device *md)
2360 __dm_destroy(md, false);
2363 void dm_put(struct mapped_device *md)
2365 atomic_dec(&md->holders);
2367 EXPORT_SYMBOL_GPL(dm_put);
2369 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2372 DECLARE_WAITQUEUE(wait, current);
2374 dm_unplug_all(md->queue);
2376 add_wait_queue(&md->wait, &wait);
2379 set_current_state(interruptible);
2382 if (!md_in_flight(md))
2385 if (interruptible == TASK_INTERRUPTIBLE &&
2386 signal_pending(current)) {
2393 set_current_state(TASK_RUNNING);
2395 remove_wait_queue(&md->wait, &wait);
2400 static void dm_flush(struct mapped_device *md)
2402 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2404 bio_init(&md->barrier_bio);
2405 md->barrier_bio.bi_bdev = md->bdev;
2406 md->barrier_bio.bi_rw = WRITE_BARRIER;
2407 __split_and_process_bio(md, &md->barrier_bio);
2409 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2412 static void process_barrier(struct mapped_device *md, struct bio *bio)
2414 md->barrier_error = 0;
2418 if (!bio_empty_barrier(bio)) {
2419 __split_and_process_bio(md, bio);
2421 * If the request isn't supported, don't waste time with
2424 if (md->barrier_error != -EOPNOTSUPP)
2428 if (md->barrier_error != DM_ENDIO_REQUEUE)
2429 bio_endio(bio, md->barrier_error);
2431 spin_lock_irq(&md->deferred_lock);
2432 bio_list_add_head(&md->deferred, bio);
2433 spin_unlock_irq(&md->deferred_lock);
2438 * Process the deferred bios
2440 static void dm_wq_work(struct work_struct *work)
2442 struct mapped_device *md = container_of(work, struct mapped_device,
2446 down_write(&md->io_lock);
2448 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2449 spin_lock_irq(&md->deferred_lock);
2450 c = bio_list_pop(&md->deferred);
2451 spin_unlock_irq(&md->deferred_lock);
2454 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2458 up_write(&md->io_lock);
2460 if (dm_request_based(md))
2461 generic_make_request(c);
2463 if (c->bi_rw & REQ_HARDBARRIER)
2464 process_barrier(md, c);
2466 __split_and_process_bio(md, c);
2469 down_write(&md->io_lock);
2472 up_write(&md->io_lock);
2475 static void dm_queue_flush(struct mapped_device *md)
2477 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2478 smp_mb__after_clear_bit();
2479 queue_work(md->wq, &md->work);
2482 static void dm_rq_set_target_request_nr(struct request *clone, unsigned request_nr)
2484 struct dm_rq_target_io *tio = clone->end_io_data;
2486 tio->info.target_request_nr = request_nr;
2489 /* Issue barrier requests to targets and wait for their completion. */
2490 static int dm_rq_barrier(struct mapped_device *md)
2493 struct dm_table *map = dm_get_live_table(md);
2494 unsigned num_targets = dm_table_get_num_targets(map);
2495 struct dm_target *ti;
2496 struct request *clone;
2498 md->barrier_error = 0;
2500 for (i = 0; i < num_targets; i++) {
2501 ti = dm_table_get_target(map, i);
2502 for (j = 0; j < ti->num_flush_requests; j++) {
2503 clone = clone_rq(md->flush_request, md, GFP_NOIO);
2504 dm_rq_set_target_request_nr(clone, j);
2505 atomic_inc(&md->pending[rq_data_dir(clone)]);
2506 map_request(ti, clone, md);
2510 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2513 return md->barrier_error;
2516 static void dm_rq_barrier_work(struct work_struct *work)
2519 struct mapped_device *md = container_of(work, struct mapped_device,
2521 struct request_queue *q = md->queue;
2523 unsigned long flags;
2526 * Hold the md reference here and leave it at the last part so that
2527 * the md can't be deleted by device opener when the barrier request
2532 error = dm_rq_barrier(md);
2534 rq = md->flush_request;
2535 md->flush_request = NULL;
2537 if (error == DM_ENDIO_REQUEUE) {
2538 spin_lock_irqsave(q->queue_lock, flags);
2539 blk_requeue_request(q, rq);
2540 spin_unlock_irqrestore(q->queue_lock, flags);
2542 blk_end_request_all(rq, error);
2550 * Swap in a new table, returning the old one for the caller to destroy.
2552 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2554 struct dm_table *map = ERR_PTR(-EINVAL);
2555 struct queue_limits limits;
2558 mutex_lock(&md->suspend_lock);
2560 /* device must be suspended */
2561 if (!dm_suspended_md(md))
2564 r = dm_calculate_queue_limits(table, &limits);
2570 map = __bind(md, table, &limits);
2573 mutex_unlock(&md->suspend_lock);
2578 * Functions to lock and unlock any filesystem running on the
2581 static int lock_fs(struct mapped_device *md)
2585 WARN_ON(md->frozen_sb);
2587 md->frozen_sb = freeze_bdev(md->bdev);
2588 if (IS_ERR(md->frozen_sb)) {
2589 r = PTR_ERR(md->frozen_sb);
2590 md->frozen_sb = NULL;
2594 set_bit(DMF_FROZEN, &md->flags);
2599 static void unlock_fs(struct mapped_device *md)
2601 if (!test_bit(DMF_FROZEN, &md->flags))
2604 thaw_bdev(md->bdev, md->frozen_sb);
2605 md->frozen_sb = NULL;
2606 clear_bit(DMF_FROZEN, &md->flags);
2610 * We need to be able to change a mapping table under a mounted
2611 * filesystem. For example we might want to move some data in
2612 * the background. Before the table can be swapped with
2613 * dm_bind_table, dm_suspend must be called to flush any in
2614 * flight bios and ensure that any further io gets deferred.
2617 * Suspend mechanism in request-based dm.
2619 * 1. Flush all I/Os by lock_fs() if needed.
2620 * 2. Stop dispatching any I/O by stopping the request_queue.
2621 * 3. Wait for all in-flight I/Os to be completed or requeued.
2623 * To abort suspend, start the request_queue.
2625 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2627 struct dm_table *map = NULL;
2629 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2630 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2632 mutex_lock(&md->suspend_lock);
2634 if (dm_suspended_md(md)) {
2639 map = dm_get_live_table(md);
2642 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2643 * This flag is cleared before dm_suspend returns.
2646 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2648 /* This does not get reverted if there's an error later. */
2649 dm_table_presuspend_targets(map);
2652 * Flush I/O to the device.
2653 * Any I/O submitted after lock_fs() may not be flushed.
2654 * noflush takes precedence over do_lockfs.
2655 * (lock_fs() flushes I/Os and waits for them to complete.)
2657 if (!noflush && do_lockfs) {
2664 * Here we must make sure that no processes are submitting requests
2665 * to target drivers i.e. no one may be executing
2666 * __split_and_process_bio. This is called from dm_request and
2669 * To get all processes out of __split_and_process_bio in dm_request,
2670 * we take the write lock. To prevent any process from reentering
2671 * __split_and_process_bio from dm_request, we set
2672 * DMF_QUEUE_IO_TO_THREAD.
2674 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2675 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2676 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2677 * further calls to __split_and_process_bio from dm_wq_work.
2679 down_write(&md->io_lock);
2680 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2681 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2682 up_write(&md->io_lock);
2685 * Request-based dm uses md->wq for barrier (dm_rq_barrier_work) which
2686 * can be kicked until md->queue is stopped. So stop md->queue before
2689 if (dm_request_based(md))
2690 stop_queue(md->queue);
2692 flush_workqueue(md->wq);
2695 * At this point no more requests are entering target request routines.
2696 * We call dm_wait_for_completion to wait for all existing requests
2699 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2701 down_write(&md->io_lock);
2703 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2704 up_write(&md->io_lock);
2706 /* were we interrupted ? */
2710 if (dm_request_based(md))
2711 start_queue(md->queue);
2714 goto out; /* pushback list is already flushed, so skip flush */
2718 * If dm_wait_for_completion returned 0, the device is completely
2719 * quiescent now. There is no request-processing activity. All new
2720 * requests are being added to md->deferred list.
2723 set_bit(DMF_SUSPENDED, &md->flags);
2725 dm_table_postsuspend_targets(map);
2731 mutex_unlock(&md->suspend_lock);
2735 int dm_resume(struct mapped_device *md)
2738 struct dm_table *map = NULL;
2740 mutex_lock(&md->suspend_lock);
2741 if (!dm_suspended_md(md))
2744 map = dm_get_live_table(md);
2745 if (!map || !dm_table_get_size(map))
2748 r = dm_table_resume_targets(map);
2755 * Flushing deferred I/Os must be done after targets are resumed
2756 * so that mapping of targets can work correctly.
2757 * Request-based dm is queueing the deferred I/Os in its request_queue.
2759 if (dm_request_based(md))
2760 start_queue(md->queue);
2764 clear_bit(DMF_SUSPENDED, &md->flags);
2766 dm_table_unplug_all(map);
2770 mutex_unlock(&md->suspend_lock);
2775 /*-----------------------------------------------------------------
2776 * Event notification.
2777 *---------------------------------------------------------------*/
2778 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2781 char udev_cookie[DM_COOKIE_LENGTH];
2782 char *envp[] = { udev_cookie, NULL };
2785 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2787 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2788 DM_COOKIE_ENV_VAR_NAME, cookie);
2789 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2794 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2796 return atomic_add_return(1, &md->uevent_seq);
2799 uint32_t dm_get_event_nr(struct mapped_device *md)
2801 return atomic_read(&md->event_nr);
2804 int dm_wait_event(struct mapped_device *md, int event_nr)
2806 return wait_event_interruptible(md->eventq,
2807 (event_nr != atomic_read(&md->event_nr)));
2810 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2812 unsigned long flags;
2814 spin_lock_irqsave(&md->uevent_lock, flags);
2815 list_add(elist, &md->uevent_list);
2816 spin_unlock_irqrestore(&md->uevent_lock, flags);
2820 * The gendisk is only valid as long as you have a reference
2823 struct gendisk *dm_disk(struct mapped_device *md)
2828 struct kobject *dm_kobject(struct mapped_device *md)
2834 * struct mapped_device should not be exported outside of dm.c
2835 * so use this check to verify that kobj is part of md structure
2837 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2839 struct mapped_device *md;
2841 md = container_of(kobj, struct mapped_device, kobj);
2842 if (&md->kobj != kobj)
2845 if (test_bit(DMF_FREEING, &md->flags) ||
2853 int dm_suspended_md(struct mapped_device *md)
2855 return test_bit(DMF_SUSPENDED, &md->flags);
2858 int dm_suspended(struct dm_target *ti)
2860 return dm_suspended_md(dm_table_get_md(ti->table));
2862 EXPORT_SYMBOL_GPL(dm_suspended);
2864 int dm_noflush_suspending(struct dm_target *ti)
2866 return __noflush_suspending(dm_table_get_md(ti->table));
2868 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2870 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2872 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2877 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2878 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2879 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2880 if (!pools->io_pool)
2881 goto free_pools_and_out;
2883 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2884 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2885 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2886 if (!pools->tio_pool)
2887 goto free_io_pool_and_out;
2889 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2890 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2892 goto free_tio_pool_and_out;
2896 free_tio_pool_and_out:
2897 mempool_destroy(pools->tio_pool);
2899 free_io_pool_and_out:
2900 mempool_destroy(pools->io_pool);
2908 void dm_free_md_mempools(struct dm_md_mempools *pools)
2914 mempool_destroy(pools->io_pool);
2916 if (pools->tio_pool)
2917 mempool_destroy(pools->tio_pool);
2920 bioset_free(pools->bs);
2925 static const struct block_device_operations dm_blk_dops = {
2926 .open = dm_blk_open,
2927 .release = dm_blk_close,
2928 .ioctl = dm_blk_ioctl,
2929 .getgeo = dm_blk_getgeo,
2930 .owner = THIS_MODULE
2933 EXPORT_SYMBOL(dm_get_mapinfo);
2938 module_init(dm_init);
2939 module_exit(dm_exit);
2941 module_param(major, uint, 0);
2942 MODULE_PARM_DESC(major, "The major number of the device mapper");
2943 MODULE_DESCRIPTION(DM_NAME " driver");
2944 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2945 MODULE_LICENSE("GPL");