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);
1034 * Return maximum size of I/O possible at the supplied sector up to the current
1037 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1039 sector_t target_offset = dm_target_offset(ti, sector);
1041 return ti->len - target_offset;
1044 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1046 sector_t len = max_io_len_target_boundary(sector, ti);
1049 * Does the target need to split even further ?
1053 sector_t offset = dm_target_offset(ti, sector);
1054 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
1063 static void __map_bio(struct dm_target *ti, struct bio *clone,
1064 struct dm_target_io *tio)
1068 struct mapped_device *md;
1070 clone->bi_end_io = clone_endio;
1071 clone->bi_private = tio;
1074 * Map the clone. If r == 0 we don't need to do
1075 * anything, the target has assumed ownership of
1078 atomic_inc(&tio->io->io_count);
1079 sector = clone->bi_sector;
1080 r = ti->type->map(ti, clone, &tio->info);
1081 if (r == DM_MAPIO_REMAPPED) {
1082 /* the bio has been remapped so dispatch it */
1084 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
1085 tio->io->bio->bi_bdev->bd_dev, sector);
1087 generic_make_request(clone);
1088 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1089 /* error the io and bail out, or requeue it if needed */
1091 dec_pending(tio->io, r);
1093 * Store bio_set for cleanup.
1095 clone->bi_private = md->bs;
1099 DMWARN("unimplemented target map return value: %d", r);
1105 struct mapped_device *md;
1106 struct dm_table *map;
1110 sector_t sector_count;
1114 static void dm_bio_destructor(struct bio *bio)
1116 struct bio_set *bs = bio->bi_private;
1122 * Creates a little bio that is just does part of a bvec.
1124 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1125 unsigned short idx, unsigned int offset,
1126 unsigned int len, struct bio_set *bs)
1129 struct bio_vec *bv = bio->bi_io_vec + idx;
1131 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1132 clone->bi_destructor = dm_bio_destructor;
1133 *clone->bi_io_vec = *bv;
1135 clone->bi_sector = sector;
1136 clone->bi_bdev = bio->bi_bdev;
1137 clone->bi_rw = bio->bi_rw & ~REQ_HARDBARRIER;
1139 clone->bi_size = to_bytes(len);
1140 clone->bi_io_vec->bv_offset = offset;
1141 clone->bi_io_vec->bv_len = clone->bi_size;
1142 clone->bi_flags |= 1 << BIO_CLONED;
1144 if (bio_integrity(bio)) {
1145 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1146 bio_integrity_trim(clone,
1147 bio_sector_offset(bio, idx, offset), len);
1154 * Creates a bio that consists of range of complete bvecs.
1156 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1157 unsigned short idx, unsigned short bv_count,
1158 unsigned int len, struct bio_set *bs)
1162 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1163 __bio_clone(clone, bio);
1164 clone->bi_rw &= ~REQ_HARDBARRIER;
1165 clone->bi_destructor = dm_bio_destructor;
1166 clone->bi_sector = sector;
1167 clone->bi_idx = idx;
1168 clone->bi_vcnt = idx + bv_count;
1169 clone->bi_size = to_bytes(len);
1170 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1172 if (bio_integrity(bio)) {
1173 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1175 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1176 bio_integrity_trim(clone,
1177 bio_sector_offset(bio, idx, 0), len);
1183 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1184 struct dm_target *ti)
1186 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1190 memset(&tio->info, 0, sizeof(tio->info));
1195 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1196 unsigned request_nr, sector_t len)
1198 struct dm_target_io *tio = alloc_tio(ci, ti);
1201 tio->info.target_request_nr = request_nr;
1204 * Discard requests require the bio's inline iovecs be initialized.
1205 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1206 * and discard, so no need for concern about wasted bvec allocations.
1208 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1209 __bio_clone(clone, ci->bio);
1210 clone->bi_destructor = dm_bio_destructor;
1212 clone->bi_sector = ci->sector;
1213 clone->bi_size = to_bytes(len);
1216 __map_bio(ti, clone, tio);
1219 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1220 unsigned num_requests, sector_t len)
1222 unsigned request_nr;
1224 for (request_nr = 0; request_nr < num_requests; request_nr++)
1225 __issue_target_request(ci, ti, request_nr, len);
1228 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1230 unsigned target_nr = 0;
1231 struct dm_target *ti;
1233 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1234 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1236 ci->sector_count = 0;
1242 * Perform all io with a single clone.
1244 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1246 struct bio *clone, *bio = ci->bio;
1247 struct dm_target_io *tio;
1249 tio = alloc_tio(ci, ti);
1250 clone = clone_bio(bio, ci->sector, ci->idx,
1251 bio->bi_vcnt - ci->idx, ci->sector_count,
1253 __map_bio(ti, clone, tio);
1254 ci->sector_count = 0;
1257 static int __clone_and_map_discard(struct clone_info *ci)
1259 struct dm_target *ti;
1263 ti = dm_table_find_target(ci->map, ci->sector);
1264 if (!dm_target_is_valid(ti))
1268 * Even though the device advertised discard support,
1269 * reconfiguration might have changed that since the
1270 * check was performed.
1272 if (!ti->num_discard_requests)
1275 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1277 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1280 } while (ci->sector_count -= len);
1285 static int __clone_and_map(struct clone_info *ci)
1287 struct bio *clone, *bio = ci->bio;
1288 struct dm_target *ti;
1289 sector_t len = 0, max;
1290 struct dm_target_io *tio;
1292 if (unlikely(bio_empty_barrier(bio)))
1293 return __clone_and_map_empty_barrier(ci);
1295 if (unlikely(bio->bi_rw & REQ_DISCARD))
1296 return __clone_and_map_discard(ci);
1298 ti = dm_table_find_target(ci->map, ci->sector);
1299 if (!dm_target_is_valid(ti))
1302 max = max_io_len(ci->sector, ti);
1304 if (ci->sector_count <= max) {
1306 * Optimise for the simple case where we can do all of
1307 * the remaining io with a single clone.
1309 __clone_and_map_simple(ci, ti);
1311 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1313 * There are some bvecs that don't span targets.
1314 * Do as many of these as possible.
1317 sector_t remaining = max;
1320 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1321 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1323 if (bv_len > remaining)
1326 remaining -= bv_len;
1330 tio = alloc_tio(ci, ti);
1331 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1333 __map_bio(ti, clone, tio);
1336 ci->sector_count -= len;
1341 * Handle a bvec that must be split between two or more targets.
1343 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1344 sector_t remaining = to_sector(bv->bv_len);
1345 unsigned int offset = 0;
1349 ti = dm_table_find_target(ci->map, ci->sector);
1350 if (!dm_target_is_valid(ti))
1353 max = max_io_len(ci->sector, ti);
1356 len = min(remaining, max);
1358 tio = alloc_tio(ci, ti);
1359 clone = split_bvec(bio, ci->sector, ci->idx,
1360 bv->bv_offset + offset, len,
1363 __map_bio(ti, clone, tio);
1366 ci->sector_count -= len;
1367 offset += to_bytes(len);
1368 } while (remaining -= len);
1377 * Split the bio into several clones and submit it to targets.
1379 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1381 struct clone_info ci;
1384 ci.map = dm_get_live_table(md);
1385 if (unlikely(!ci.map)) {
1386 if (!(bio->bi_rw & REQ_HARDBARRIER))
1389 if (!md->barrier_error)
1390 md->barrier_error = -EIO;
1396 ci.io = alloc_io(md);
1398 atomic_set(&ci.io->io_count, 1);
1401 spin_lock_init(&ci.io->endio_lock);
1402 ci.sector = bio->bi_sector;
1403 ci.sector_count = bio_sectors(bio);
1404 if (unlikely(bio_empty_barrier(bio)))
1405 ci.sector_count = 1;
1406 ci.idx = bio->bi_idx;
1408 start_io_acct(ci.io);
1409 while (ci.sector_count && !error)
1410 error = __clone_and_map(&ci);
1412 /* drop the extra reference count */
1413 dec_pending(ci.io, error);
1414 dm_table_put(ci.map);
1416 /*-----------------------------------------------------------------
1418 *---------------------------------------------------------------*/
1420 static int dm_merge_bvec(struct request_queue *q,
1421 struct bvec_merge_data *bvm,
1422 struct bio_vec *biovec)
1424 struct mapped_device *md = q->queuedata;
1425 struct dm_table *map = dm_get_live_table(md);
1426 struct dm_target *ti;
1427 sector_t max_sectors;
1433 ti = dm_table_find_target(map, bvm->bi_sector);
1434 if (!dm_target_is_valid(ti))
1438 * Find maximum amount of I/O that won't need splitting
1440 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1441 (sector_t) BIO_MAX_SECTORS);
1442 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1447 * merge_bvec_fn() returns number of bytes
1448 * it can accept at this offset
1449 * max is precomputed maximal io size
1451 if (max_size && ti->type->merge)
1452 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1454 * If the target doesn't support merge method and some of the devices
1455 * provided their merge_bvec method (we know this by looking at
1456 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1457 * entries. So always set max_size to 0, and the code below allows
1460 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1469 * Always allow an entire first page
1471 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1472 max_size = biovec->bv_len;
1478 * The request function that just remaps the bio built up by
1481 static int _dm_request(struct request_queue *q, struct bio *bio)
1483 int rw = bio_data_dir(bio);
1484 struct mapped_device *md = q->queuedata;
1487 down_read(&md->io_lock);
1489 cpu = part_stat_lock();
1490 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1491 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1495 * If we're suspended or the thread is processing barriers
1496 * we have to queue this io for later.
1498 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1499 unlikely(bio->bi_rw & REQ_HARDBARRIER)) {
1500 up_read(&md->io_lock);
1502 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1503 bio_rw(bio) == READA) {
1513 __split_and_process_bio(md, bio);
1514 up_read(&md->io_lock);
1518 static int dm_make_request(struct request_queue *q, struct bio *bio)
1520 struct mapped_device *md = q->queuedata;
1522 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1525 static int dm_request_based(struct mapped_device *md)
1527 return blk_queue_stackable(md->queue);
1530 static int dm_request(struct request_queue *q, struct bio *bio)
1532 struct mapped_device *md = q->queuedata;
1534 if (dm_request_based(md))
1535 return dm_make_request(q, bio);
1537 return _dm_request(q, bio);
1540 static bool dm_rq_is_flush_request(struct request *rq)
1542 if (rq->cmd_flags & REQ_FLUSH)
1548 void dm_dispatch_request(struct request *rq)
1552 if (blk_queue_io_stat(rq->q))
1553 rq->cmd_flags |= REQ_IO_STAT;
1555 rq->start_time = jiffies;
1556 r = blk_insert_cloned_request(rq->q, rq);
1558 dm_complete_request(rq, r);
1560 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1562 static void dm_rq_bio_destructor(struct bio *bio)
1564 struct dm_rq_clone_bio_info *info = bio->bi_private;
1565 struct mapped_device *md = info->tio->md;
1567 free_bio_info(info);
1568 bio_free(bio, md->bs);
1571 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1574 struct dm_rq_target_io *tio = data;
1575 struct mapped_device *md = tio->md;
1576 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1581 info->orig = bio_orig;
1583 bio->bi_end_io = end_clone_bio;
1584 bio->bi_private = info;
1585 bio->bi_destructor = dm_rq_bio_destructor;
1590 static int setup_clone(struct request *clone, struct request *rq,
1591 struct dm_rq_target_io *tio)
1595 if (dm_rq_is_flush_request(rq)) {
1596 blk_rq_init(NULL, clone);
1597 clone->cmd_type = REQ_TYPE_FS;
1598 clone->cmd_flags |= (REQ_HARDBARRIER | WRITE);
1600 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1601 dm_rq_bio_constructor, tio);
1605 clone->cmd = rq->cmd;
1606 clone->cmd_len = rq->cmd_len;
1607 clone->sense = rq->sense;
1608 clone->buffer = rq->buffer;
1611 clone->end_io = end_clone_request;
1612 clone->end_io_data = tio;
1617 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1620 struct request *clone;
1621 struct dm_rq_target_io *tio;
1623 tio = alloc_rq_tio(md, gfp_mask);
1631 memset(&tio->info, 0, sizeof(tio->info));
1633 clone = &tio->clone;
1634 if (setup_clone(clone, rq, tio)) {
1644 * Called with the queue lock held.
1646 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1648 struct mapped_device *md = q->queuedata;
1649 struct request *clone;
1651 if (unlikely(dm_rq_is_flush_request(rq)))
1654 if (unlikely(rq->special)) {
1655 DMWARN("Already has something in rq->special.");
1656 return BLKPREP_KILL;
1659 clone = clone_rq(rq, md, GFP_ATOMIC);
1661 return BLKPREP_DEFER;
1663 rq->special = clone;
1664 rq->cmd_flags |= REQ_DONTPREP;
1671 * 0 : the request has been processed (not requeued)
1672 * !0 : the request has been requeued
1674 static int map_request(struct dm_target *ti, struct request *clone,
1675 struct mapped_device *md)
1677 int r, requeued = 0;
1678 struct dm_rq_target_io *tio = clone->end_io_data;
1681 * Hold the md reference here for the in-flight I/O.
1682 * We can't rely on the reference count by device opener,
1683 * because the device may be closed during the request completion
1684 * when all bios are completed.
1685 * See the comment in rq_completed() too.
1690 r = ti->type->map_rq(ti, clone, &tio->info);
1692 case DM_MAPIO_SUBMITTED:
1693 /* The target has taken the I/O to submit by itself later */
1695 case DM_MAPIO_REMAPPED:
1696 /* The target has remapped the I/O so dispatch it */
1697 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1698 blk_rq_pos(tio->orig));
1699 dm_dispatch_request(clone);
1701 case DM_MAPIO_REQUEUE:
1702 /* The target wants to requeue the I/O */
1703 dm_requeue_unmapped_request(clone);
1708 DMWARN("unimplemented target map return value: %d", r);
1712 /* The target wants to complete the I/O */
1713 dm_kill_unmapped_request(clone, r);
1721 * q->request_fn for request-based dm.
1722 * Called with the queue lock held.
1724 static void dm_request_fn(struct request_queue *q)
1726 struct mapped_device *md = q->queuedata;
1727 struct dm_table *map = dm_get_live_table(md);
1728 struct dm_target *ti;
1729 struct request *rq, *clone;
1732 * For suspend, check blk_queue_stopped() and increment
1733 * ->pending within a single queue_lock not to increment the
1734 * number of in-flight I/Os after the queue is stopped in
1737 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1738 rq = blk_peek_request(q);
1742 if (unlikely(dm_rq_is_flush_request(rq))) {
1743 BUG_ON(md->flush_request);
1744 md->flush_request = rq;
1745 blk_start_request(rq);
1746 queue_work(md->wq, &md->barrier_work);
1750 ti = dm_table_find_target(map, blk_rq_pos(rq));
1751 if (ti->type->busy && ti->type->busy(ti))
1754 blk_start_request(rq);
1755 clone = rq->special;
1756 atomic_inc(&md->pending[rq_data_dir(clone)]);
1758 spin_unlock(q->queue_lock);
1759 if (map_request(ti, clone, md))
1762 spin_lock_irq(q->queue_lock);
1768 spin_lock_irq(q->queue_lock);
1771 if (!elv_queue_empty(q))
1772 /* Some requests still remain, retry later */
1781 int dm_underlying_device_busy(struct request_queue *q)
1783 return blk_lld_busy(q);
1785 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1787 static int dm_lld_busy(struct request_queue *q)
1790 struct mapped_device *md = q->queuedata;
1791 struct dm_table *map = dm_get_live_table(md);
1793 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1796 r = dm_table_any_busy_target(map);
1803 static void dm_unplug_all(struct request_queue *q)
1805 struct mapped_device *md = q->queuedata;
1806 struct dm_table *map = dm_get_live_table(md);
1809 if (dm_request_based(md))
1810 generic_unplug_device(q);
1812 dm_table_unplug_all(map);
1817 static int dm_any_congested(void *congested_data, int bdi_bits)
1820 struct mapped_device *md = congested_data;
1821 struct dm_table *map;
1823 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1824 map = dm_get_live_table(md);
1827 * Request-based dm cares about only own queue for
1828 * the query about congestion status of request_queue
1830 if (dm_request_based(md))
1831 r = md->queue->backing_dev_info.state &
1834 r = dm_table_any_congested(map, bdi_bits);
1843 /*-----------------------------------------------------------------
1844 * An IDR is used to keep track of allocated minor numbers.
1845 *---------------------------------------------------------------*/
1846 static DEFINE_IDR(_minor_idr);
1848 static void free_minor(int minor)
1850 spin_lock(&_minor_lock);
1851 idr_remove(&_minor_idr, minor);
1852 spin_unlock(&_minor_lock);
1856 * See if the device with a specific minor # is free.
1858 static int specific_minor(int minor)
1862 if (minor >= (1 << MINORBITS))
1865 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1869 spin_lock(&_minor_lock);
1871 if (idr_find(&_minor_idr, minor)) {
1876 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1881 idr_remove(&_minor_idr, m);
1887 spin_unlock(&_minor_lock);
1891 static int next_free_minor(int *minor)
1895 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1899 spin_lock(&_minor_lock);
1901 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1905 if (m >= (1 << MINORBITS)) {
1906 idr_remove(&_minor_idr, m);
1914 spin_unlock(&_minor_lock);
1918 static const struct block_device_operations dm_blk_dops;
1920 static void dm_wq_work(struct work_struct *work);
1921 static void dm_rq_barrier_work(struct work_struct *work);
1923 static void dm_init_md_queue(struct mapped_device *md)
1926 * Request-based dm devices cannot be stacked on top of bio-based dm
1927 * devices. The type of this dm device has not been decided yet.
1928 * The type is decided at the first table loading time.
1929 * To prevent problematic device stacking, clear the queue flag
1930 * for request stacking support until then.
1932 * This queue is new, so no concurrency on the queue_flags.
1934 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1936 md->queue->queuedata = md;
1937 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1938 md->queue->backing_dev_info.congested_data = md;
1939 blk_queue_make_request(md->queue, dm_request);
1940 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1941 md->queue->unplug_fn = dm_unplug_all;
1942 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1946 * Allocate and initialise a blank device with a given minor.
1948 static struct mapped_device *alloc_dev(int minor)
1951 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1955 DMWARN("unable to allocate device, out of memory.");
1959 if (!try_module_get(THIS_MODULE))
1960 goto bad_module_get;
1962 /* get a minor number for the dev */
1963 if (minor == DM_ANY_MINOR)
1964 r = next_free_minor(&minor);
1966 r = specific_minor(minor);
1970 md->type = DM_TYPE_NONE;
1971 init_rwsem(&md->io_lock);
1972 mutex_init(&md->suspend_lock);
1973 mutex_init(&md->type_lock);
1974 spin_lock_init(&md->deferred_lock);
1975 spin_lock_init(&md->barrier_error_lock);
1976 rwlock_init(&md->map_lock);
1977 atomic_set(&md->holders, 1);
1978 atomic_set(&md->open_count, 0);
1979 atomic_set(&md->event_nr, 0);
1980 atomic_set(&md->uevent_seq, 0);
1981 INIT_LIST_HEAD(&md->uevent_list);
1982 spin_lock_init(&md->uevent_lock);
1984 md->queue = blk_alloc_queue(GFP_KERNEL);
1988 dm_init_md_queue(md);
1990 md->disk = alloc_disk(1);
1994 atomic_set(&md->pending[0], 0);
1995 atomic_set(&md->pending[1], 0);
1996 init_waitqueue_head(&md->wait);
1997 INIT_WORK(&md->work, dm_wq_work);
1998 INIT_WORK(&md->barrier_work, dm_rq_barrier_work);
1999 init_waitqueue_head(&md->eventq);
2001 md->disk->major = _major;
2002 md->disk->first_minor = minor;
2003 md->disk->fops = &dm_blk_dops;
2004 md->disk->queue = md->queue;
2005 md->disk->private_data = md;
2006 sprintf(md->disk->disk_name, "dm-%d", minor);
2008 format_dev_t(md->name, MKDEV(_major, minor));
2010 md->wq = create_singlethread_workqueue("kdmflush");
2014 md->bdev = bdget_disk(md->disk, 0);
2018 /* Populate the mapping, nobody knows we exist yet */
2019 spin_lock(&_minor_lock);
2020 old_md = idr_replace(&_minor_idr, md, minor);
2021 spin_unlock(&_minor_lock);
2023 BUG_ON(old_md != MINOR_ALLOCED);
2028 destroy_workqueue(md->wq);
2030 del_gendisk(md->disk);
2033 blk_cleanup_queue(md->queue);
2037 module_put(THIS_MODULE);
2043 static void unlock_fs(struct mapped_device *md);
2045 static void free_dev(struct mapped_device *md)
2047 int minor = MINOR(disk_devt(md->disk));
2051 destroy_workqueue(md->wq);
2053 mempool_destroy(md->tio_pool);
2055 mempool_destroy(md->io_pool);
2057 bioset_free(md->bs);
2058 blk_integrity_unregister(md->disk);
2059 del_gendisk(md->disk);
2062 spin_lock(&_minor_lock);
2063 md->disk->private_data = NULL;
2064 spin_unlock(&_minor_lock);
2067 blk_cleanup_queue(md->queue);
2068 module_put(THIS_MODULE);
2072 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
2074 struct dm_md_mempools *p;
2076 if (md->io_pool && md->tio_pool && md->bs)
2077 /* the md already has necessary mempools */
2080 p = dm_table_get_md_mempools(t);
2081 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
2083 md->io_pool = p->io_pool;
2085 md->tio_pool = p->tio_pool;
2091 /* mempool bind completed, now no need any mempools in the table */
2092 dm_table_free_md_mempools(t);
2096 * Bind a table to the device.
2098 static void event_callback(void *context)
2100 unsigned long flags;
2102 struct mapped_device *md = (struct mapped_device *) context;
2104 spin_lock_irqsave(&md->uevent_lock, flags);
2105 list_splice_init(&md->uevent_list, &uevents);
2106 spin_unlock_irqrestore(&md->uevent_lock, flags);
2108 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2110 atomic_inc(&md->event_nr);
2111 wake_up(&md->eventq);
2114 static void __set_size(struct mapped_device *md, sector_t size)
2116 set_capacity(md->disk, size);
2118 mutex_lock(&md->bdev->bd_inode->i_mutex);
2119 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2120 mutex_unlock(&md->bdev->bd_inode->i_mutex);
2124 * Returns old map, which caller must destroy.
2126 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2127 struct queue_limits *limits)
2129 struct dm_table *old_map;
2130 struct request_queue *q = md->queue;
2132 unsigned long flags;
2134 size = dm_table_get_size(t);
2137 * Wipe any geometry if the size of the table changed.
2139 if (size != get_capacity(md->disk))
2140 memset(&md->geometry, 0, sizeof(md->geometry));
2142 __set_size(md, size);
2144 dm_table_event_callback(t, event_callback, md);
2147 * The queue hasn't been stopped yet, if the old table type wasn't
2148 * for request-based during suspension. So stop it to prevent
2149 * I/O mapping before resume.
2150 * This must be done before setting the queue restrictions,
2151 * because request-based dm may be run just after the setting.
2153 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2156 __bind_mempools(md, t);
2158 write_lock_irqsave(&md->map_lock, flags);
2161 dm_table_set_restrictions(t, q, limits);
2162 write_unlock_irqrestore(&md->map_lock, flags);
2168 * Returns unbound table for the caller to free.
2170 static struct dm_table *__unbind(struct mapped_device *md)
2172 struct dm_table *map = md->map;
2173 unsigned long flags;
2178 dm_table_event_callback(map, NULL, NULL);
2179 write_lock_irqsave(&md->map_lock, flags);
2181 write_unlock_irqrestore(&md->map_lock, flags);
2187 * Constructor for a new device.
2189 int dm_create(int minor, struct mapped_device **result)
2191 struct mapped_device *md;
2193 md = alloc_dev(minor);
2204 * Functions to manage md->type.
2205 * All are required to hold md->type_lock.
2207 void dm_lock_md_type(struct mapped_device *md)
2209 mutex_lock(&md->type_lock);
2212 void dm_unlock_md_type(struct mapped_device *md)
2214 mutex_unlock(&md->type_lock);
2217 void dm_set_md_type(struct mapped_device *md, unsigned type)
2222 unsigned dm_get_md_type(struct mapped_device *md)
2228 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2230 static int dm_init_request_based_queue(struct mapped_device *md)
2232 struct request_queue *q = NULL;
2234 if (md->queue->elevator)
2237 /* Fully initialize the queue */
2238 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2243 md->saved_make_request_fn = md->queue->make_request_fn;
2244 dm_init_md_queue(md);
2245 blk_queue_softirq_done(md->queue, dm_softirq_done);
2246 blk_queue_prep_rq(md->queue, dm_prep_fn);
2247 blk_queue_lld_busy(md->queue, dm_lld_busy);
2248 blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN_FLUSH);
2250 elv_register_queue(md->queue);
2256 * Setup the DM device's queue based on md's type
2258 int dm_setup_md_queue(struct mapped_device *md)
2260 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2261 !dm_init_request_based_queue(md)) {
2262 DMWARN("Cannot initialize queue for request-based mapped device");
2269 static struct mapped_device *dm_find_md(dev_t dev)
2271 struct mapped_device *md;
2272 unsigned minor = MINOR(dev);
2274 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2277 spin_lock(&_minor_lock);
2279 md = idr_find(&_minor_idr, minor);
2280 if (md && (md == MINOR_ALLOCED ||
2281 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2282 dm_deleting_md(md) ||
2283 test_bit(DMF_FREEING, &md->flags))) {
2289 spin_unlock(&_minor_lock);
2294 struct mapped_device *dm_get_md(dev_t dev)
2296 struct mapped_device *md = dm_find_md(dev);
2304 void *dm_get_mdptr(struct mapped_device *md)
2306 return md->interface_ptr;
2309 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2311 md->interface_ptr = ptr;
2314 void dm_get(struct mapped_device *md)
2316 atomic_inc(&md->holders);
2317 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2320 const char *dm_device_name(struct mapped_device *md)
2324 EXPORT_SYMBOL_GPL(dm_device_name);
2326 static void __dm_destroy(struct mapped_device *md, bool wait)
2328 struct dm_table *map;
2332 spin_lock(&_minor_lock);
2333 map = dm_get_live_table(md);
2334 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2335 set_bit(DMF_FREEING, &md->flags);
2336 spin_unlock(&_minor_lock);
2338 if (!dm_suspended_md(md)) {
2339 dm_table_presuspend_targets(map);
2340 dm_table_postsuspend_targets(map);
2344 * Rare, but there may be I/O requests still going to complete,
2345 * for example. Wait for all references to disappear.
2346 * No one should increment the reference count of the mapped_device,
2347 * after the mapped_device state becomes DMF_FREEING.
2350 while (atomic_read(&md->holders))
2352 else if (atomic_read(&md->holders))
2353 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2354 dm_device_name(md), atomic_read(&md->holders));
2358 dm_table_destroy(__unbind(md));
2362 void dm_destroy(struct mapped_device *md)
2364 __dm_destroy(md, true);
2367 void dm_destroy_immediate(struct mapped_device *md)
2369 __dm_destroy(md, false);
2372 void dm_put(struct mapped_device *md)
2374 atomic_dec(&md->holders);
2376 EXPORT_SYMBOL_GPL(dm_put);
2378 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2381 DECLARE_WAITQUEUE(wait, current);
2383 dm_unplug_all(md->queue);
2385 add_wait_queue(&md->wait, &wait);
2388 set_current_state(interruptible);
2391 if (!md_in_flight(md))
2394 if (interruptible == TASK_INTERRUPTIBLE &&
2395 signal_pending(current)) {
2402 set_current_state(TASK_RUNNING);
2404 remove_wait_queue(&md->wait, &wait);
2409 static void dm_flush(struct mapped_device *md)
2411 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2413 bio_init(&md->barrier_bio);
2414 md->barrier_bio.bi_bdev = md->bdev;
2415 md->barrier_bio.bi_rw = WRITE_BARRIER;
2416 __split_and_process_bio(md, &md->barrier_bio);
2418 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2421 static void process_barrier(struct mapped_device *md, struct bio *bio)
2423 md->barrier_error = 0;
2427 if (!bio_empty_barrier(bio)) {
2428 __split_and_process_bio(md, bio);
2430 * If the request isn't supported, don't waste time with
2433 if (md->barrier_error != -EOPNOTSUPP)
2437 if (md->barrier_error != DM_ENDIO_REQUEUE)
2438 bio_endio(bio, md->barrier_error);
2440 spin_lock_irq(&md->deferred_lock);
2441 bio_list_add_head(&md->deferred, bio);
2442 spin_unlock_irq(&md->deferred_lock);
2447 * Process the deferred bios
2449 static void dm_wq_work(struct work_struct *work)
2451 struct mapped_device *md = container_of(work, struct mapped_device,
2455 down_write(&md->io_lock);
2457 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2458 spin_lock_irq(&md->deferred_lock);
2459 c = bio_list_pop(&md->deferred);
2460 spin_unlock_irq(&md->deferred_lock);
2463 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2467 up_write(&md->io_lock);
2469 if (dm_request_based(md))
2470 generic_make_request(c);
2472 if (c->bi_rw & REQ_HARDBARRIER)
2473 process_barrier(md, c);
2475 __split_and_process_bio(md, c);
2478 down_write(&md->io_lock);
2481 up_write(&md->io_lock);
2484 static void dm_queue_flush(struct mapped_device *md)
2486 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2487 smp_mb__after_clear_bit();
2488 queue_work(md->wq, &md->work);
2491 static void dm_rq_set_target_request_nr(struct request *clone, unsigned request_nr)
2493 struct dm_rq_target_io *tio = clone->end_io_data;
2495 tio->info.target_request_nr = request_nr;
2498 /* Issue barrier requests to targets and wait for their completion. */
2499 static int dm_rq_barrier(struct mapped_device *md)
2502 struct dm_table *map = dm_get_live_table(md);
2503 unsigned num_targets = dm_table_get_num_targets(map);
2504 struct dm_target *ti;
2505 struct request *clone;
2507 md->barrier_error = 0;
2509 for (i = 0; i < num_targets; i++) {
2510 ti = dm_table_get_target(map, i);
2511 for (j = 0; j < ti->num_flush_requests; j++) {
2512 clone = clone_rq(md->flush_request, md, GFP_NOIO);
2513 dm_rq_set_target_request_nr(clone, j);
2514 atomic_inc(&md->pending[rq_data_dir(clone)]);
2515 map_request(ti, clone, md);
2519 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2522 return md->barrier_error;
2525 static void dm_rq_barrier_work(struct work_struct *work)
2528 struct mapped_device *md = container_of(work, struct mapped_device,
2530 struct request_queue *q = md->queue;
2532 unsigned long flags;
2535 * Hold the md reference here and leave it at the last part so that
2536 * the md can't be deleted by device opener when the barrier request
2541 error = dm_rq_barrier(md);
2543 rq = md->flush_request;
2544 md->flush_request = NULL;
2546 if (error == DM_ENDIO_REQUEUE) {
2547 spin_lock_irqsave(q->queue_lock, flags);
2548 blk_requeue_request(q, rq);
2549 spin_unlock_irqrestore(q->queue_lock, flags);
2551 blk_end_request_all(rq, error);
2559 * Swap in a new table, returning the old one for the caller to destroy.
2561 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2563 struct dm_table *map = ERR_PTR(-EINVAL);
2564 struct queue_limits limits;
2567 mutex_lock(&md->suspend_lock);
2569 /* device must be suspended */
2570 if (!dm_suspended_md(md))
2573 r = dm_calculate_queue_limits(table, &limits);
2579 map = __bind(md, table, &limits);
2582 mutex_unlock(&md->suspend_lock);
2587 * Functions to lock and unlock any filesystem running on the
2590 static int lock_fs(struct mapped_device *md)
2594 WARN_ON(md->frozen_sb);
2596 md->frozen_sb = freeze_bdev(md->bdev);
2597 if (IS_ERR(md->frozen_sb)) {
2598 r = PTR_ERR(md->frozen_sb);
2599 md->frozen_sb = NULL;
2603 set_bit(DMF_FROZEN, &md->flags);
2608 static void unlock_fs(struct mapped_device *md)
2610 if (!test_bit(DMF_FROZEN, &md->flags))
2613 thaw_bdev(md->bdev, md->frozen_sb);
2614 md->frozen_sb = NULL;
2615 clear_bit(DMF_FROZEN, &md->flags);
2619 * We need to be able to change a mapping table under a mounted
2620 * filesystem. For example we might want to move some data in
2621 * the background. Before the table can be swapped with
2622 * dm_bind_table, dm_suspend must be called to flush any in
2623 * flight bios and ensure that any further io gets deferred.
2626 * Suspend mechanism in request-based dm.
2628 * 1. Flush all I/Os by lock_fs() if needed.
2629 * 2. Stop dispatching any I/O by stopping the request_queue.
2630 * 3. Wait for all in-flight I/Os to be completed or requeued.
2632 * To abort suspend, start the request_queue.
2634 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2636 struct dm_table *map = NULL;
2638 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2639 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2641 mutex_lock(&md->suspend_lock);
2643 if (dm_suspended_md(md)) {
2648 map = dm_get_live_table(md);
2651 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2652 * This flag is cleared before dm_suspend returns.
2655 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2657 /* This does not get reverted if there's an error later. */
2658 dm_table_presuspend_targets(map);
2661 * Flush I/O to the device.
2662 * Any I/O submitted after lock_fs() may not be flushed.
2663 * noflush takes precedence over do_lockfs.
2664 * (lock_fs() flushes I/Os and waits for them to complete.)
2666 if (!noflush && do_lockfs) {
2673 * Here we must make sure that no processes are submitting requests
2674 * to target drivers i.e. no one may be executing
2675 * __split_and_process_bio. This is called from dm_request and
2678 * To get all processes out of __split_and_process_bio in dm_request,
2679 * we take the write lock. To prevent any process from reentering
2680 * __split_and_process_bio from dm_request, we set
2681 * DMF_QUEUE_IO_TO_THREAD.
2683 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2684 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2685 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2686 * further calls to __split_and_process_bio from dm_wq_work.
2688 down_write(&md->io_lock);
2689 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2690 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2691 up_write(&md->io_lock);
2694 * Request-based dm uses md->wq for barrier (dm_rq_barrier_work) which
2695 * can be kicked until md->queue is stopped. So stop md->queue before
2698 if (dm_request_based(md))
2699 stop_queue(md->queue);
2701 flush_workqueue(md->wq);
2704 * At this point no more requests are entering target request routines.
2705 * We call dm_wait_for_completion to wait for all existing requests
2708 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2710 down_write(&md->io_lock);
2712 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2713 up_write(&md->io_lock);
2715 /* were we interrupted ? */
2719 if (dm_request_based(md))
2720 start_queue(md->queue);
2723 goto out; /* pushback list is already flushed, so skip flush */
2727 * If dm_wait_for_completion returned 0, the device is completely
2728 * quiescent now. There is no request-processing activity. All new
2729 * requests are being added to md->deferred list.
2732 set_bit(DMF_SUSPENDED, &md->flags);
2734 dm_table_postsuspend_targets(map);
2740 mutex_unlock(&md->suspend_lock);
2744 int dm_resume(struct mapped_device *md)
2747 struct dm_table *map = NULL;
2749 mutex_lock(&md->suspend_lock);
2750 if (!dm_suspended_md(md))
2753 map = dm_get_live_table(md);
2754 if (!map || !dm_table_get_size(map))
2757 r = dm_table_resume_targets(map);
2764 * Flushing deferred I/Os must be done after targets are resumed
2765 * so that mapping of targets can work correctly.
2766 * Request-based dm is queueing the deferred I/Os in its request_queue.
2768 if (dm_request_based(md))
2769 start_queue(md->queue);
2773 clear_bit(DMF_SUSPENDED, &md->flags);
2775 dm_table_unplug_all(map);
2779 mutex_unlock(&md->suspend_lock);
2784 /*-----------------------------------------------------------------
2785 * Event notification.
2786 *---------------------------------------------------------------*/
2787 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2790 char udev_cookie[DM_COOKIE_LENGTH];
2791 char *envp[] = { udev_cookie, NULL };
2794 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2796 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2797 DM_COOKIE_ENV_VAR_NAME, cookie);
2798 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2803 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2805 return atomic_add_return(1, &md->uevent_seq);
2808 uint32_t dm_get_event_nr(struct mapped_device *md)
2810 return atomic_read(&md->event_nr);
2813 int dm_wait_event(struct mapped_device *md, int event_nr)
2815 return wait_event_interruptible(md->eventq,
2816 (event_nr != atomic_read(&md->event_nr)));
2819 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2821 unsigned long flags;
2823 spin_lock_irqsave(&md->uevent_lock, flags);
2824 list_add(elist, &md->uevent_list);
2825 spin_unlock_irqrestore(&md->uevent_lock, flags);
2829 * The gendisk is only valid as long as you have a reference
2832 struct gendisk *dm_disk(struct mapped_device *md)
2837 struct kobject *dm_kobject(struct mapped_device *md)
2843 * struct mapped_device should not be exported outside of dm.c
2844 * so use this check to verify that kobj is part of md structure
2846 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2848 struct mapped_device *md;
2850 md = container_of(kobj, struct mapped_device, kobj);
2851 if (&md->kobj != kobj)
2854 if (test_bit(DMF_FREEING, &md->flags) ||
2862 int dm_suspended_md(struct mapped_device *md)
2864 return test_bit(DMF_SUSPENDED, &md->flags);
2867 int dm_suspended(struct dm_target *ti)
2869 return dm_suspended_md(dm_table_get_md(ti->table));
2871 EXPORT_SYMBOL_GPL(dm_suspended);
2873 int dm_noflush_suspending(struct dm_target *ti)
2875 return __noflush_suspending(dm_table_get_md(ti->table));
2877 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2879 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2881 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2886 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2887 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2888 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2889 if (!pools->io_pool)
2890 goto free_pools_and_out;
2892 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2893 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2894 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2895 if (!pools->tio_pool)
2896 goto free_io_pool_and_out;
2898 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2899 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2901 goto free_tio_pool_and_out;
2905 free_tio_pool_and_out:
2906 mempool_destroy(pools->tio_pool);
2908 free_io_pool_and_out:
2909 mempool_destroy(pools->io_pool);
2917 void dm_free_md_mempools(struct dm_md_mempools *pools)
2923 mempool_destroy(pools->io_pool);
2925 if (pools->tio_pool)
2926 mempool_destroy(pools->tio_pool);
2929 bioset_free(pools->bs);
2934 static const struct block_device_operations dm_blk_dops = {
2935 .open = dm_blk_open,
2936 .release = dm_blk_close,
2937 .ioctl = dm_blk_ioctl,
2938 .getgeo = dm_blk_getgeo,
2939 .owner = THIS_MODULE
2942 EXPORT_SYMBOL(dm_get_mapinfo);
2947 module_init(dm_init);
2948 module_exit(dm_exit);
2950 module_param(major, uint, 0);
2951 MODULE_PARM_DESC(major, "The major number of the device mapper");
2952 MODULE_DESCRIPTION(DM_NAME " driver");
2953 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2954 MODULE_LICENSE("GPL");