2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/moduleparam.h>
15 #include <linux/blkpg.h>
16 #include <linux/bio.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mempool.h>
19 #include <linux/slab.h>
20 #include <linux/idr.h>
21 #include <linux/hdreg.h>
23 #include <trace/events/block.h>
25 #define DM_MSG_PREFIX "core"
28 * Cookies are numeric values sent with CHANGE and REMOVE
29 * uevents while resuming, removing or renaming the device.
31 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
32 #define DM_COOKIE_LENGTH 24
34 static const char *_name = DM_NAME;
36 static unsigned int major = 0;
37 static unsigned int _major = 0;
39 static DEFINE_SPINLOCK(_minor_lock);
42 * One of these is allocated per bio.
45 struct mapped_device *md;
49 unsigned long start_time;
54 * One of these is allocated per target within a bio. Hopefully
55 * this will be simplified out one day.
64 * For request-based dm.
65 * One of these is allocated per request.
67 struct dm_rq_target_io {
68 struct mapped_device *md;
70 struct request *orig, clone;
76 * For request-based dm.
77 * One of these is allocated per bio.
79 struct dm_rq_clone_bio_info {
81 struct dm_rq_target_io *tio;
84 union map_info *dm_get_mapinfo(struct bio *bio)
86 if (bio && bio->bi_private)
87 return &((struct dm_target_io *)bio->bi_private)->info;
91 union map_info *dm_get_rq_mapinfo(struct request *rq)
93 if (rq && rq->end_io_data)
94 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
97 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
99 #define MINOR_ALLOCED ((void *)-1)
102 * Bits for the md->flags field.
104 #define DMF_BLOCK_IO_FOR_SUSPEND 0
105 #define DMF_SUSPENDED 1
107 #define DMF_FREEING 3
108 #define DMF_DELETING 4
109 #define DMF_NOFLUSH_SUSPENDING 5
110 #define DMF_QUEUE_IO_TO_THREAD 6
113 * Work processed by per-device workqueue.
115 struct mapped_device {
116 struct rw_semaphore io_lock;
117 struct mutex suspend_lock;
124 struct request_queue *queue;
125 struct gendisk *disk;
131 * A list of ios that arrived while we were suspended.
134 wait_queue_head_t wait;
135 struct work_struct work;
136 struct bio_list deferred;
137 spinlock_t deferred_lock;
140 * An error from the barrier request currently being processed.
145 * Processing queue (flush/barriers)
147 struct workqueue_struct *wq;
150 * The current mapping.
152 struct dm_table *map;
155 * io objects are allocated from here.
166 wait_queue_head_t eventq;
168 struct list_head uevent_list;
169 spinlock_t uevent_lock; /* Protect access to uevent_list */
172 * freeze/thaw support require holding onto a super block
174 struct super_block *frozen_sb;
175 struct block_device *bdev;
177 /* forced geometry settings */
178 struct hd_geometry geometry;
180 /* marker of flush suspend for request-based dm */
181 struct request suspend_rq;
183 /* For saving the address of __make_request for request based dm */
184 make_request_fn *saved_make_request_fn;
189 /* zero-length barrier that will be cloned and submitted to targets */
190 struct bio barrier_bio;
194 * For mempools pre-allocation at the table loading time.
196 struct dm_md_mempools {
203 static struct kmem_cache *_io_cache;
204 static struct kmem_cache *_tio_cache;
205 static struct kmem_cache *_rq_tio_cache;
206 static struct kmem_cache *_rq_bio_info_cache;
208 static int __init local_init(void)
212 /* allocate a slab for the dm_ios */
213 _io_cache = KMEM_CACHE(dm_io, 0);
217 /* allocate a slab for the target ios */
218 _tio_cache = KMEM_CACHE(dm_target_io, 0);
220 goto out_free_io_cache;
222 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
224 goto out_free_tio_cache;
226 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
227 if (!_rq_bio_info_cache)
228 goto out_free_rq_tio_cache;
230 r = dm_uevent_init();
232 goto out_free_rq_bio_info_cache;
235 r = register_blkdev(_major, _name);
237 goto out_uevent_exit;
246 out_free_rq_bio_info_cache:
247 kmem_cache_destroy(_rq_bio_info_cache);
248 out_free_rq_tio_cache:
249 kmem_cache_destroy(_rq_tio_cache);
251 kmem_cache_destroy(_tio_cache);
253 kmem_cache_destroy(_io_cache);
258 static void local_exit(void)
260 kmem_cache_destroy(_rq_bio_info_cache);
261 kmem_cache_destroy(_rq_tio_cache);
262 kmem_cache_destroy(_tio_cache);
263 kmem_cache_destroy(_io_cache);
264 unregister_blkdev(_major, _name);
269 DMINFO("cleaned up");
272 static int (*_inits[])(void) __initdata = {
281 static void (*_exits[])(void) = {
290 static int __init dm_init(void)
292 const int count = ARRAY_SIZE(_inits);
296 for (i = 0; i < count; i++) {
311 static void __exit dm_exit(void)
313 int i = ARRAY_SIZE(_exits);
320 * Block device functions
322 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
324 struct mapped_device *md;
326 spin_lock(&_minor_lock);
328 md = bdev->bd_disk->private_data;
332 if (test_bit(DMF_FREEING, &md->flags) ||
333 test_bit(DMF_DELETING, &md->flags)) {
339 atomic_inc(&md->open_count);
342 spin_unlock(&_minor_lock);
344 return md ? 0 : -ENXIO;
347 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
349 struct mapped_device *md = disk->private_data;
350 atomic_dec(&md->open_count);
355 int dm_open_count(struct mapped_device *md)
357 return atomic_read(&md->open_count);
361 * Guarantees nothing is using the device before it's deleted.
363 int dm_lock_for_deletion(struct mapped_device *md)
367 spin_lock(&_minor_lock);
369 if (dm_open_count(md))
372 set_bit(DMF_DELETING, &md->flags);
374 spin_unlock(&_minor_lock);
379 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
381 struct mapped_device *md = bdev->bd_disk->private_data;
383 return dm_get_geometry(md, geo);
386 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
387 unsigned int cmd, unsigned long arg)
389 struct mapped_device *md = bdev->bd_disk->private_data;
390 struct dm_table *map = dm_get_table(md);
391 struct dm_target *tgt;
394 if (!map || !dm_table_get_size(map))
397 /* We only support devices that have a single target */
398 if (dm_table_get_num_targets(map) != 1)
401 tgt = dm_table_get_target(map, 0);
403 if (dm_suspended(md)) {
408 if (tgt->type->ioctl)
409 r = tgt->type->ioctl(tgt, cmd, arg);
417 static struct dm_io *alloc_io(struct mapped_device *md)
419 return mempool_alloc(md->io_pool, GFP_NOIO);
422 static void free_io(struct mapped_device *md, struct dm_io *io)
424 mempool_free(io, md->io_pool);
427 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
429 mempool_free(tio, md->tio_pool);
432 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md)
434 return mempool_alloc(md->tio_pool, GFP_ATOMIC);
437 static void free_rq_tio(struct dm_rq_target_io *tio)
439 mempool_free(tio, tio->md->tio_pool);
442 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
444 return mempool_alloc(md->io_pool, GFP_ATOMIC);
447 static void free_bio_info(struct dm_rq_clone_bio_info *info)
449 mempool_free(info, info->tio->md->io_pool);
452 static void start_io_acct(struct dm_io *io)
454 struct mapped_device *md = io->md;
457 io->start_time = jiffies;
459 cpu = part_stat_lock();
460 part_round_stats(cpu, &dm_disk(md)->part0);
462 dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending);
465 static void end_io_acct(struct dm_io *io)
467 struct mapped_device *md = io->md;
468 struct bio *bio = io->bio;
469 unsigned long duration = jiffies - io->start_time;
471 int rw = bio_data_dir(bio);
473 cpu = part_stat_lock();
474 part_round_stats(cpu, &dm_disk(md)->part0);
475 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
479 * After this is decremented the bio must not be touched if it is
482 dm_disk(md)->part0.in_flight = pending =
483 atomic_dec_return(&md->pending);
485 /* nudge anyone waiting on suspend queue */
491 * Add the bio to the list of deferred io.
493 static void queue_io(struct mapped_device *md, struct bio *bio)
495 down_write(&md->io_lock);
497 spin_lock_irq(&md->deferred_lock);
498 bio_list_add(&md->deferred, bio);
499 spin_unlock_irq(&md->deferred_lock);
501 if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags))
502 queue_work(md->wq, &md->work);
504 up_write(&md->io_lock);
508 * Everyone (including functions in this file), should use this
509 * function to access the md->map field, and make sure they call
510 * dm_table_put() when finished.
512 struct dm_table *dm_get_table(struct mapped_device *md)
517 read_lock_irqsave(&md->map_lock, flags);
521 read_unlock_irqrestore(&md->map_lock, flags);
527 * Get the geometry associated with a dm device
529 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
537 * Set the geometry of a device.
539 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
541 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
543 if (geo->start > sz) {
544 DMWARN("Start sector is beyond the geometry limits.");
553 /*-----------------------------------------------------------------
555 * A more elegant soln is in the works that uses the queue
556 * merge fn, unfortunately there are a couple of changes to
557 * the block layer that I want to make for this. So in the
558 * interests of getting something for people to use I give
559 * you this clearly demarcated crap.
560 *---------------------------------------------------------------*/
562 static int __noflush_suspending(struct mapped_device *md)
564 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
568 * Decrements the number of outstanding ios that a bio has been
569 * cloned into, completing the original io if necc.
571 static void dec_pending(struct dm_io *io, int error)
576 struct mapped_device *md = io->md;
578 /* Push-back supersedes any I/O errors */
579 if (error && !(io->error > 0 && __noflush_suspending(md)))
582 if (atomic_dec_and_test(&io->io_count)) {
583 if (io->error == DM_ENDIO_REQUEUE) {
585 * Target requested pushing back the I/O.
587 spin_lock_irqsave(&md->deferred_lock, flags);
588 if (__noflush_suspending(md)) {
589 if (!bio_barrier(io->bio))
590 bio_list_add_head(&md->deferred,
593 /* noflush suspend was interrupted. */
595 spin_unlock_irqrestore(&md->deferred_lock, flags);
598 io_error = io->error;
601 if (bio_barrier(bio)) {
603 * There can be just one barrier request so we use
604 * a per-device variable for error reporting.
605 * Note that you can't touch the bio after end_io_acct
607 if (!md->barrier_error && io_error != -EOPNOTSUPP)
608 md->barrier_error = io_error;
613 if (io_error != DM_ENDIO_REQUEUE) {
614 trace_block_bio_complete(md->queue, bio);
616 bio_endio(bio, io_error);
624 static void clone_endio(struct bio *bio, int error)
627 struct dm_target_io *tio = bio->bi_private;
628 struct dm_io *io = tio->io;
629 struct mapped_device *md = tio->io->md;
630 dm_endio_fn endio = tio->ti->type->end_io;
632 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
636 r = endio(tio->ti, bio, error, &tio->info);
637 if (r < 0 || r == DM_ENDIO_REQUEUE)
639 * error and requeue request are handled
643 else if (r == DM_ENDIO_INCOMPLETE)
644 /* The target will handle the io */
647 DMWARN("unimplemented target endio return value: %d", r);
653 * Store md for cleanup instead of tio which is about to get freed.
655 bio->bi_private = md->bs;
659 dec_pending(io, error);
663 * Partial completion handling for request-based dm
665 static void end_clone_bio(struct bio *clone, int error)
667 struct dm_rq_clone_bio_info *info = clone->bi_private;
668 struct dm_rq_target_io *tio = info->tio;
669 struct bio *bio = info->orig;
670 unsigned int nr_bytes = info->orig->bi_size;
676 * An error has already been detected on the request.
677 * Once error occurred, just let clone->end_io() handle
683 * Don't notice the error to the upper layer yet.
684 * The error handling decision is made by the target driver,
685 * when the request is completed.
692 * I/O for the bio successfully completed.
693 * Notice the data completion to the upper layer.
697 * bios are processed from the head of the list.
698 * So the completing bio should always be rq->bio.
699 * If it's not, something wrong is happening.
701 if (tio->orig->bio != bio)
702 DMERR("bio completion is going in the middle of the request");
705 * Update the original request.
706 * Do not use blk_end_request() here, because it may complete
707 * the original request before the clone, and break the ordering.
709 blk_update_request(tio->orig, 0, nr_bytes);
713 * Don't touch any member of the md after calling this function because
714 * the md may be freed in dm_put() at the end of this function.
715 * Or do dm_get() before calling this function and dm_put() later.
717 static void rq_completed(struct mapped_device *md, int run_queue)
719 int wakeup_waiters = 0;
720 struct request_queue *q = md->queue;
723 spin_lock_irqsave(q->queue_lock, flags);
724 if (!queue_in_flight(q))
726 spin_unlock_irqrestore(q->queue_lock, flags);
728 /* nudge anyone waiting on suspend queue */
736 * dm_put() must be at the end of this function. See the comment above
741 static void free_rq_clone(struct request *clone)
743 struct dm_rq_target_io *tio = clone->end_io_data;
745 blk_rq_unprep_clone(clone);
749 static void dm_unprep_request(struct request *rq)
751 struct request *clone = rq->special;
754 rq->cmd_flags &= ~REQ_DONTPREP;
756 free_rq_clone(clone);
760 * Requeue the original request of a clone.
762 void dm_requeue_unmapped_request(struct request *clone)
764 struct dm_rq_target_io *tio = clone->end_io_data;
765 struct mapped_device *md = tio->md;
766 struct request *rq = tio->orig;
767 struct request_queue *q = rq->q;
770 dm_unprep_request(rq);
772 spin_lock_irqsave(q->queue_lock, flags);
773 if (elv_queue_empty(q))
775 blk_requeue_request(q, rq);
776 spin_unlock_irqrestore(q->queue_lock, flags);
780 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
782 static void __stop_queue(struct request_queue *q)
787 static void stop_queue(struct request_queue *q)
791 spin_lock_irqsave(q->queue_lock, flags);
793 spin_unlock_irqrestore(q->queue_lock, flags);
796 static void __start_queue(struct request_queue *q)
798 if (blk_queue_stopped(q))
802 static void start_queue(struct request_queue *q)
806 spin_lock_irqsave(q->queue_lock, flags);
808 spin_unlock_irqrestore(q->queue_lock, flags);
812 * Complete the clone and the original request.
813 * Must be called without queue lock.
815 static void dm_end_request(struct request *clone, int error)
817 struct dm_rq_target_io *tio = clone->end_io_data;
818 struct mapped_device *md = tio->md;
819 struct request *rq = tio->orig;
821 if (blk_pc_request(rq)) {
822 rq->errors = clone->errors;
823 rq->resid_len = clone->resid_len;
827 * We are using the sense buffer of the original
829 * So setting the length of the sense data is enough.
831 rq->sense_len = clone->sense_len;
834 free_rq_clone(clone);
836 blk_end_request_all(rq, error);
842 * Request completion handler for request-based dm
844 static void dm_softirq_done(struct request *rq)
846 struct request *clone = rq->completion_data;
847 struct dm_rq_target_io *tio = clone->end_io_data;
848 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
849 int error = tio->error;
851 if (!(rq->cmd_flags & REQ_FAILED) && rq_end_io)
852 error = rq_end_io(tio->ti, clone, error, &tio->info);
855 /* The target wants to complete the I/O */
856 dm_end_request(clone, error);
857 else if (error == DM_ENDIO_INCOMPLETE)
858 /* The target will handle the I/O */
860 else if (error == DM_ENDIO_REQUEUE)
861 /* The target wants to requeue the I/O */
862 dm_requeue_unmapped_request(clone);
864 DMWARN("unimplemented target endio return value: %d", error);
870 * Complete the clone and the original request with the error status
871 * through softirq context.
873 static void dm_complete_request(struct request *clone, int error)
875 struct dm_rq_target_io *tio = clone->end_io_data;
876 struct request *rq = tio->orig;
879 rq->completion_data = clone;
880 blk_complete_request(rq);
884 * Complete the not-mapped clone and the original request with the error status
885 * through softirq context.
886 * Target's rq_end_io() function isn't called.
887 * This may be used when the target's map_rq() function fails.
889 void dm_kill_unmapped_request(struct request *clone, int error)
891 struct dm_rq_target_io *tio = clone->end_io_data;
892 struct request *rq = tio->orig;
894 rq->cmd_flags |= REQ_FAILED;
895 dm_complete_request(clone, error);
897 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
900 * Called with the queue lock held
902 static void end_clone_request(struct request *clone, int error)
905 * For just cleaning up the information of the queue in which
906 * the clone was dispatched.
907 * The clone is *NOT* freed actually here because it is alloced from
908 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
910 __blk_put_request(clone->q, clone);
913 * Actual request completion is done in a softirq context which doesn't
914 * hold the queue lock. Otherwise, deadlock could occur because:
915 * - another request may be submitted by the upper level driver
916 * of the stacking during the completion
917 * - the submission which requires queue lock may be done
920 dm_complete_request(clone, error);
923 static sector_t max_io_len(struct mapped_device *md,
924 sector_t sector, struct dm_target *ti)
926 sector_t offset = sector - ti->begin;
927 sector_t len = ti->len - offset;
930 * Does the target need to split even further ?
934 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
943 static void __map_bio(struct dm_target *ti, struct bio *clone,
944 struct dm_target_io *tio)
948 struct mapped_device *md;
950 clone->bi_end_io = clone_endio;
951 clone->bi_private = tio;
954 * Map the clone. If r == 0 we don't need to do
955 * anything, the target has assumed ownership of
958 atomic_inc(&tio->io->io_count);
959 sector = clone->bi_sector;
960 r = ti->type->map(ti, clone, &tio->info);
961 if (r == DM_MAPIO_REMAPPED) {
962 /* the bio has been remapped so dispatch it */
964 trace_block_remap(bdev_get_queue(clone->bi_bdev), clone,
965 tio->io->bio->bi_bdev->bd_dev, sector);
967 generic_make_request(clone);
968 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
969 /* error the io and bail out, or requeue it if needed */
971 dec_pending(tio->io, r);
973 * Store bio_set for cleanup.
975 clone->bi_private = md->bs;
979 DMWARN("unimplemented target map return value: %d", r);
985 struct mapped_device *md;
986 struct dm_table *map;
990 sector_t sector_count;
994 static void dm_bio_destructor(struct bio *bio)
996 struct bio_set *bs = bio->bi_private;
1002 * Creates a little bio that is just does part of a bvec.
1004 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1005 unsigned short idx, unsigned int offset,
1006 unsigned int len, struct bio_set *bs)
1009 struct bio_vec *bv = bio->bi_io_vec + idx;
1011 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1012 clone->bi_destructor = dm_bio_destructor;
1013 *clone->bi_io_vec = *bv;
1015 clone->bi_sector = sector;
1016 clone->bi_bdev = bio->bi_bdev;
1017 clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER);
1019 clone->bi_size = to_bytes(len);
1020 clone->bi_io_vec->bv_offset = offset;
1021 clone->bi_io_vec->bv_len = clone->bi_size;
1022 clone->bi_flags |= 1 << BIO_CLONED;
1024 if (bio_integrity(bio)) {
1025 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1026 bio_integrity_trim(clone,
1027 bio_sector_offset(bio, idx, offset), len);
1034 * Creates a bio that consists of range of complete bvecs.
1036 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1037 unsigned short idx, unsigned short bv_count,
1038 unsigned int len, struct bio_set *bs)
1042 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1043 __bio_clone(clone, bio);
1044 clone->bi_rw &= ~(1 << BIO_RW_BARRIER);
1045 clone->bi_destructor = dm_bio_destructor;
1046 clone->bi_sector = sector;
1047 clone->bi_idx = idx;
1048 clone->bi_vcnt = idx + bv_count;
1049 clone->bi_size = to_bytes(len);
1050 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1052 if (bio_integrity(bio)) {
1053 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1055 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1056 bio_integrity_trim(clone,
1057 bio_sector_offset(bio, idx, 0), len);
1063 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1064 struct dm_target *ti)
1066 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1070 memset(&tio->info, 0, sizeof(tio->info));
1075 static void __flush_target(struct clone_info *ci, struct dm_target *ti,
1078 struct dm_target_io *tio = alloc_tio(ci, ti);
1081 tio->info.flush_request = flush_nr;
1083 clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs);
1084 __bio_clone(clone, ci->bio);
1085 clone->bi_destructor = dm_bio_destructor;
1087 __map_bio(ti, clone, tio);
1090 static int __clone_and_map_empty_barrier(struct clone_info *ci)
1092 unsigned target_nr = 0, flush_nr;
1093 struct dm_target *ti;
1095 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1096 for (flush_nr = 0; flush_nr < ti->num_flush_requests;
1098 __flush_target(ci, ti, flush_nr);
1100 ci->sector_count = 0;
1105 static int __clone_and_map(struct clone_info *ci)
1107 struct bio *clone, *bio = ci->bio;
1108 struct dm_target *ti;
1109 sector_t len = 0, max;
1110 struct dm_target_io *tio;
1112 if (unlikely(bio_empty_barrier(bio)))
1113 return __clone_and_map_empty_barrier(ci);
1115 ti = dm_table_find_target(ci->map, ci->sector);
1116 if (!dm_target_is_valid(ti))
1119 max = max_io_len(ci->md, ci->sector, ti);
1122 * Allocate a target io object.
1124 tio = alloc_tio(ci, ti);
1126 if (ci->sector_count <= max) {
1128 * Optimise for the simple case where we can do all of
1129 * the remaining io with a single clone.
1131 clone = clone_bio(bio, ci->sector, ci->idx,
1132 bio->bi_vcnt - ci->idx, ci->sector_count,
1134 __map_bio(ti, clone, tio);
1135 ci->sector_count = 0;
1137 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1139 * There are some bvecs that don't span targets.
1140 * Do as many of these as possible.
1143 sector_t remaining = max;
1146 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1147 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1149 if (bv_len > remaining)
1152 remaining -= bv_len;
1156 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1158 __map_bio(ti, clone, tio);
1161 ci->sector_count -= len;
1166 * Handle a bvec that must be split between two or more targets.
1168 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1169 sector_t remaining = to_sector(bv->bv_len);
1170 unsigned int offset = 0;
1174 ti = dm_table_find_target(ci->map, ci->sector);
1175 if (!dm_target_is_valid(ti))
1178 max = max_io_len(ci->md, ci->sector, ti);
1180 tio = alloc_tio(ci, ti);
1183 len = min(remaining, max);
1185 clone = split_bvec(bio, ci->sector, ci->idx,
1186 bv->bv_offset + offset, len,
1189 __map_bio(ti, clone, tio);
1192 ci->sector_count -= len;
1193 offset += to_bytes(len);
1194 } while (remaining -= len);
1203 * Split the bio into several clones and submit it to targets.
1205 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1207 struct clone_info ci;
1210 ci.map = dm_get_table(md);
1211 if (unlikely(!ci.map)) {
1212 if (!bio_barrier(bio))
1215 if (!md->barrier_error)
1216 md->barrier_error = -EIO;
1222 ci.io = alloc_io(md);
1224 atomic_set(&ci.io->io_count, 1);
1227 ci.sector = bio->bi_sector;
1228 ci.sector_count = bio_sectors(bio);
1229 if (unlikely(bio_empty_barrier(bio)))
1230 ci.sector_count = 1;
1231 ci.idx = bio->bi_idx;
1233 start_io_acct(ci.io);
1234 while (ci.sector_count && !error)
1235 error = __clone_and_map(&ci);
1237 /* drop the extra reference count */
1238 dec_pending(ci.io, error);
1239 dm_table_put(ci.map);
1241 /*-----------------------------------------------------------------
1243 *---------------------------------------------------------------*/
1245 static int dm_merge_bvec(struct request_queue *q,
1246 struct bvec_merge_data *bvm,
1247 struct bio_vec *biovec)
1249 struct mapped_device *md = q->queuedata;
1250 struct dm_table *map = dm_get_table(md);
1251 struct dm_target *ti;
1252 sector_t max_sectors;
1258 ti = dm_table_find_target(map, bvm->bi_sector);
1259 if (!dm_target_is_valid(ti))
1263 * Find maximum amount of I/O that won't need splitting
1265 max_sectors = min(max_io_len(md, bvm->bi_sector, ti),
1266 (sector_t) BIO_MAX_SECTORS);
1267 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1272 * merge_bvec_fn() returns number of bytes
1273 * it can accept at this offset
1274 * max is precomputed maximal io size
1276 if (max_size && ti->type->merge)
1277 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1279 * If the target doesn't support merge method and some of the devices
1280 * provided their merge_bvec method (we know this by looking at
1281 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1282 * entries. So always set max_size to 0, and the code below allows
1285 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1294 * Always allow an entire first page
1296 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1297 max_size = biovec->bv_len;
1303 * The request function that just remaps the bio built up by
1306 static int _dm_request(struct request_queue *q, struct bio *bio)
1308 int rw = bio_data_dir(bio);
1309 struct mapped_device *md = q->queuedata;
1312 down_read(&md->io_lock);
1314 cpu = part_stat_lock();
1315 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1316 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1320 * If we're suspended or the thread is processing barriers
1321 * we have to queue this io for later.
1323 if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) ||
1324 unlikely(bio_barrier(bio))) {
1325 up_read(&md->io_lock);
1327 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) &&
1328 bio_rw(bio) == READA) {
1338 __split_and_process_bio(md, bio);
1339 up_read(&md->io_lock);
1343 static int dm_make_request(struct request_queue *q, struct bio *bio)
1345 struct mapped_device *md = q->queuedata;
1347 if (unlikely(bio_barrier(bio))) {
1348 bio_endio(bio, -EOPNOTSUPP);
1352 return md->saved_make_request_fn(q, bio); /* call __make_request() */
1355 static int dm_request_based(struct mapped_device *md)
1357 return blk_queue_stackable(md->queue);
1360 static int dm_request(struct request_queue *q, struct bio *bio)
1362 struct mapped_device *md = q->queuedata;
1364 if (dm_request_based(md))
1365 return dm_make_request(q, bio);
1367 return _dm_request(q, bio);
1370 void dm_dispatch_request(struct request *rq)
1374 if (blk_queue_io_stat(rq->q))
1375 rq->cmd_flags |= REQ_IO_STAT;
1377 rq->start_time = jiffies;
1378 r = blk_insert_cloned_request(rq->q, rq);
1380 dm_complete_request(rq, r);
1382 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1384 static void dm_rq_bio_destructor(struct bio *bio)
1386 struct dm_rq_clone_bio_info *info = bio->bi_private;
1387 struct mapped_device *md = info->tio->md;
1389 free_bio_info(info);
1390 bio_free(bio, md->bs);
1393 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1396 struct dm_rq_target_io *tio = data;
1397 struct mapped_device *md = tio->md;
1398 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1403 info->orig = bio_orig;
1405 bio->bi_end_io = end_clone_bio;
1406 bio->bi_private = info;
1407 bio->bi_destructor = dm_rq_bio_destructor;
1412 static int setup_clone(struct request *clone, struct request *rq,
1413 struct dm_rq_target_io *tio)
1415 int r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1416 dm_rq_bio_constructor, tio);
1421 clone->cmd = rq->cmd;
1422 clone->cmd_len = rq->cmd_len;
1423 clone->sense = rq->sense;
1424 clone->buffer = rq->buffer;
1425 clone->end_io = end_clone_request;
1426 clone->end_io_data = tio;
1431 static int dm_rq_flush_suspending(struct mapped_device *md)
1433 return !md->suspend_rq.special;
1437 * Called with the queue lock held.
1439 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1441 struct mapped_device *md = q->queuedata;
1442 struct dm_rq_target_io *tio;
1443 struct request *clone;
1445 if (unlikely(rq == &md->suspend_rq)) {
1446 if (dm_rq_flush_suspending(md))
1449 /* The flush suspend was interrupted */
1450 return BLKPREP_KILL;
1453 if (unlikely(rq->special)) {
1454 DMWARN("Already has something in rq->special.");
1455 return BLKPREP_KILL;
1458 tio = alloc_rq_tio(md); /* Only one for each original request */
1461 return BLKPREP_DEFER;
1467 memset(&tio->info, 0, sizeof(tio->info));
1469 clone = &tio->clone;
1470 if (setup_clone(clone, rq, tio)) {
1473 return BLKPREP_DEFER;
1476 rq->special = clone;
1477 rq->cmd_flags |= REQ_DONTPREP;
1482 static void map_request(struct dm_target *ti, struct request *rq,
1483 struct mapped_device *md)
1486 struct request *clone = rq->special;
1487 struct dm_rq_target_io *tio = clone->end_io_data;
1490 * Hold the md reference here for the in-flight I/O.
1491 * We can't rely on the reference count by device opener,
1492 * because the device may be closed during the request completion
1493 * when all bios are completed.
1494 * See the comment in rq_completed() too.
1499 r = ti->type->map_rq(ti, clone, &tio->info);
1501 case DM_MAPIO_SUBMITTED:
1502 /* The target has taken the I/O to submit by itself later */
1504 case DM_MAPIO_REMAPPED:
1505 /* The target has remapped the I/O so dispatch it */
1506 dm_dispatch_request(clone);
1508 case DM_MAPIO_REQUEUE:
1509 /* The target wants to requeue the I/O */
1510 dm_requeue_unmapped_request(clone);
1514 DMWARN("unimplemented target map return value: %d", r);
1518 /* The target wants to complete the I/O */
1519 dm_kill_unmapped_request(clone, r);
1525 * q->request_fn for request-based dm.
1526 * Called with the queue lock held.
1528 static void dm_request_fn(struct request_queue *q)
1530 struct mapped_device *md = q->queuedata;
1531 struct dm_table *map = dm_get_table(md);
1532 struct dm_target *ti;
1536 * For noflush suspend, check blk_queue_stopped() to immediately
1537 * quit I/O dispatching.
1539 while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) {
1540 rq = blk_peek_request(q);
1544 if (unlikely(rq == &md->suspend_rq)) { /* Flush suspend maker */
1545 if (queue_in_flight(q))
1546 /* Not quiet yet. Wait more */
1549 /* This device should be quiet now */
1551 blk_start_request(rq);
1552 __blk_end_request_all(rq, 0);
1557 ti = dm_table_find_target(map, blk_rq_pos(rq));
1558 if (ti->type->busy && ti->type->busy(ti))
1561 blk_start_request(rq);
1562 spin_unlock(q->queue_lock);
1563 map_request(ti, rq, md);
1564 spin_lock_irq(q->queue_lock);
1570 if (!elv_queue_empty(q))
1571 /* Some requests still remain, retry later */
1580 int dm_underlying_device_busy(struct request_queue *q)
1582 return blk_lld_busy(q);
1584 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1586 static int dm_lld_busy(struct request_queue *q)
1589 struct mapped_device *md = q->queuedata;
1590 struct dm_table *map = dm_get_table(md);
1592 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1595 r = dm_table_any_busy_target(map);
1602 static void dm_unplug_all(struct request_queue *q)
1604 struct mapped_device *md = q->queuedata;
1605 struct dm_table *map = dm_get_table(md);
1608 if (dm_request_based(md))
1609 generic_unplug_device(q);
1611 dm_table_unplug_all(map);
1616 static int dm_any_congested(void *congested_data, int bdi_bits)
1619 struct mapped_device *md = congested_data;
1620 struct dm_table *map;
1622 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1623 map = dm_get_table(md);
1626 * Request-based dm cares about only own queue for
1627 * the query about congestion status of request_queue
1629 if (dm_request_based(md))
1630 r = md->queue->backing_dev_info.state &
1633 r = dm_table_any_congested(map, bdi_bits);
1642 /*-----------------------------------------------------------------
1643 * An IDR is used to keep track of allocated minor numbers.
1644 *---------------------------------------------------------------*/
1645 static DEFINE_IDR(_minor_idr);
1647 static void free_minor(int minor)
1649 spin_lock(&_minor_lock);
1650 idr_remove(&_minor_idr, minor);
1651 spin_unlock(&_minor_lock);
1655 * See if the device with a specific minor # is free.
1657 static int specific_minor(int minor)
1661 if (minor >= (1 << MINORBITS))
1664 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1668 spin_lock(&_minor_lock);
1670 if (idr_find(&_minor_idr, minor)) {
1675 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1680 idr_remove(&_minor_idr, m);
1686 spin_unlock(&_minor_lock);
1690 static int next_free_minor(int *minor)
1694 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1698 spin_lock(&_minor_lock);
1700 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1704 if (m >= (1 << MINORBITS)) {
1705 idr_remove(&_minor_idr, m);
1713 spin_unlock(&_minor_lock);
1717 static struct block_device_operations dm_blk_dops;
1719 static void dm_wq_work(struct work_struct *work);
1722 * Allocate and initialise a blank device with a given minor.
1724 static struct mapped_device *alloc_dev(int minor)
1727 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1731 DMWARN("unable to allocate device, out of memory.");
1735 if (!try_module_get(THIS_MODULE))
1736 goto bad_module_get;
1738 /* get a minor number for the dev */
1739 if (minor == DM_ANY_MINOR)
1740 r = next_free_minor(&minor);
1742 r = specific_minor(minor);
1746 init_rwsem(&md->io_lock);
1747 mutex_init(&md->suspend_lock);
1748 spin_lock_init(&md->deferred_lock);
1749 rwlock_init(&md->map_lock);
1750 atomic_set(&md->holders, 1);
1751 atomic_set(&md->open_count, 0);
1752 atomic_set(&md->event_nr, 0);
1753 atomic_set(&md->uevent_seq, 0);
1754 INIT_LIST_HEAD(&md->uevent_list);
1755 spin_lock_init(&md->uevent_lock);
1757 md->queue = blk_init_queue(dm_request_fn, NULL);
1762 * Request-based dm devices cannot be stacked on top of bio-based dm
1763 * devices. The type of this dm device has not been decided yet,
1764 * although we initialized the queue using blk_init_queue().
1765 * The type is decided at the first table loading time.
1766 * To prevent problematic device stacking, clear the queue flag
1767 * for request stacking support until then.
1769 * This queue is new, so no concurrency on the queue_flags.
1771 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1772 md->saved_make_request_fn = md->queue->make_request_fn;
1773 md->queue->queuedata = md;
1774 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1775 md->queue->backing_dev_info.congested_data = md;
1776 blk_queue_make_request(md->queue, dm_request);
1777 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1778 md->queue->unplug_fn = dm_unplug_all;
1779 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1780 blk_queue_softirq_done(md->queue, dm_softirq_done);
1781 blk_queue_prep_rq(md->queue, dm_prep_fn);
1782 blk_queue_lld_busy(md->queue, dm_lld_busy);
1784 md->disk = alloc_disk(1);
1788 atomic_set(&md->pending, 0);
1789 init_waitqueue_head(&md->wait);
1790 INIT_WORK(&md->work, dm_wq_work);
1791 init_waitqueue_head(&md->eventq);
1793 md->disk->major = _major;
1794 md->disk->first_minor = minor;
1795 md->disk->fops = &dm_blk_dops;
1796 md->disk->queue = md->queue;
1797 md->disk->private_data = md;
1798 sprintf(md->disk->disk_name, "dm-%d", minor);
1800 format_dev_t(md->name, MKDEV(_major, minor));
1802 md->wq = create_singlethread_workqueue("kdmflush");
1806 md->bdev = bdget_disk(md->disk, 0);
1810 /* Populate the mapping, nobody knows we exist yet */
1811 spin_lock(&_minor_lock);
1812 old_md = idr_replace(&_minor_idr, md, minor);
1813 spin_unlock(&_minor_lock);
1815 BUG_ON(old_md != MINOR_ALLOCED);
1820 destroy_workqueue(md->wq);
1824 blk_cleanup_queue(md->queue);
1828 module_put(THIS_MODULE);
1834 static void unlock_fs(struct mapped_device *md);
1836 static void free_dev(struct mapped_device *md)
1838 int minor = MINOR(disk_devt(md->disk));
1842 destroy_workqueue(md->wq);
1844 mempool_destroy(md->tio_pool);
1846 mempool_destroy(md->io_pool);
1848 bioset_free(md->bs);
1849 blk_integrity_unregister(md->disk);
1850 del_gendisk(md->disk);
1853 spin_lock(&_minor_lock);
1854 md->disk->private_data = NULL;
1855 spin_unlock(&_minor_lock);
1858 blk_cleanup_queue(md->queue);
1859 module_put(THIS_MODULE);
1863 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1865 struct dm_md_mempools *p;
1867 if (md->io_pool && md->tio_pool && md->bs)
1868 /* the md already has necessary mempools */
1871 p = dm_table_get_md_mempools(t);
1872 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1874 md->io_pool = p->io_pool;
1876 md->tio_pool = p->tio_pool;
1882 /* mempool bind completed, now no need any mempools in the table */
1883 dm_table_free_md_mempools(t);
1887 * Bind a table to the device.
1889 static void event_callback(void *context)
1891 unsigned long flags;
1893 struct mapped_device *md = (struct mapped_device *) context;
1895 spin_lock_irqsave(&md->uevent_lock, flags);
1896 list_splice_init(&md->uevent_list, &uevents);
1897 spin_unlock_irqrestore(&md->uevent_lock, flags);
1899 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1901 atomic_inc(&md->event_nr);
1902 wake_up(&md->eventq);
1905 static void __set_size(struct mapped_device *md, sector_t size)
1907 set_capacity(md->disk, size);
1909 mutex_lock(&md->bdev->bd_inode->i_mutex);
1910 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1911 mutex_unlock(&md->bdev->bd_inode->i_mutex);
1914 static int __bind(struct mapped_device *md, struct dm_table *t,
1915 struct queue_limits *limits)
1917 struct request_queue *q = md->queue;
1919 unsigned long flags;
1921 size = dm_table_get_size(t);
1924 * Wipe any geometry if the size of the table changed.
1926 if (size != get_capacity(md->disk))
1927 memset(&md->geometry, 0, sizeof(md->geometry));
1929 __set_size(md, size);
1932 dm_table_destroy(t);
1936 dm_table_event_callback(t, event_callback, md);
1939 * The queue hasn't been stopped yet, if the old table type wasn't
1940 * for request-based during suspension. So stop it to prevent
1941 * I/O mapping before resume.
1942 * This must be done before setting the queue restrictions,
1943 * because request-based dm may be run just after the setting.
1945 if (dm_table_request_based(t) && !blk_queue_stopped(q))
1948 __bind_mempools(md, t);
1950 write_lock_irqsave(&md->map_lock, flags);
1952 dm_table_set_restrictions(t, q, limits);
1953 write_unlock_irqrestore(&md->map_lock, flags);
1958 static void __unbind(struct mapped_device *md)
1960 struct dm_table *map = md->map;
1961 unsigned long flags;
1966 dm_table_event_callback(map, NULL, NULL);
1967 write_lock_irqsave(&md->map_lock, flags);
1969 write_unlock_irqrestore(&md->map_lock, flags);
1970 dm_table_destroy(map);
1974 * Constructor for a new device.
1976 int dm_create(int minor, struct mapped_device **result)
1978 struct mapped_device *md;
1980 md = alloc_dev(minor);
1990 static struct mapped_device *dm_find_md(dev_t dev)
1992 struct mapped_device *md;
1993 unsigned minor = MINOR(dev);
1995 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
1998 spin_lock(&_minor_lock);
2000 md = idr_find(&_minor_idr, minor);
2001 if (md && (md == MINOR_ALLOCED ||
2002 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2003 test_bit(DMF_FREEING, &md->flags))) {
2009 spin_unlock(&_minor_lock);
2014 struct mapped_device *dm_get_md(dev_t dev)
2016 struct mapped_device *md = dm_find_md(dev);
2024 void *dm_get_mdptr(struct mapped_device *md)
2026 return md->interface_ptr;
2029 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2031 md->interface_ptr = ptr;
2034 void dm_get(struct mapped_device *md)
2036 atomic_inc(&md->holders);
2039 const char *dm_device_name(struct mapped_device *md)
2043 EXPORT_SYMBOL_GPL(dm_device_name);
2045 void dm_put(struct mapped_device *md)
2047 struct dm_table *map;
2049 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2051 if (atomic_dec_and_lock(&md->holders, &_minor_lock)) {
2052 map = dm_get_table(md);
2053 idr_replace(&_minor_idr, MINOR_ALLOCED,
2054 MINOR(disk_devt(dm_disk(md))));
2055 set_bit(DMF_FREEING, &md->flags);
2056 spin_unlock(&_minor_lock);
2057 if (!dm_suspended(md)) {
2058 dm_table_presuspend_targets(map);
2059 dm_table_postsuspend_targets(map);
2067 EXPORT_SYMBOL_GPL(dm_put);
2069 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2072 DECLARE_WAITQUEUE(wait, current);
2073 struct request_queue *q = md->queue;
2074 unsigned long flags;
2076 dm_unplug_all(md->queue);
2078 add_wait_queue(&md->wait, &wait);
2081 set_current_state(interruptible);
2084 if (dm_request_based(md)) {
2085 spin_lock_irqsave(q->queue_lock, flags);
2086 if (!queue_in_flight(q) && blk_queue_stopped(q)) {
2087 spin_unlock_irqrestore(q->queue_lock, flags);
2090 spin_unlock_irqrestore(q->queue_lock, flags);
2091 } else if (!atomic_read(&md->pending))
2094 if (interruptible == TASK_INTERRUPTIBLE &&
2095 signal_pending(current)) {
2102 set_current_state(TASK_RUNNING);
2104 remove_wait_queue(&md->wait, &wait);
2109 static void dm_flush(struct mapped_device *md)
2111 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2113 bio_init(&md->barrier_bio);
2114 md->barrier_bio.bi_bdev = md->bdev;
2115 md->barrier_bio.bi_rw = WRITE_BARRIER;
2116 __split_and_process_bio(md, &md->barrier_bio);
2118 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2121 static void process_barrier(struct mapped_device *md, struct bio *bio)
2123 md->barrier_error = 0;
2127 if (!bio_empty_barrier(bio)) {
2128 __split_and_process_bio(md, bio);
2132 if (md->barrier_error != DM_ENDIO_REQUEUE)
2133 bio_endio(bio, md->barrier_error);
2135 spin_lock_irq(&md->deferred_lock);
2136 bio_list_add_head(&md->deferred, bio);
2137 spin_unlock_irq(&md->deferred_lock);
2142 * Process the deferred bios
2144 static void dm_wq_work(struct work_struct *work)
2146 struct mapped_device *md = container_of(work, struct mapped_device,
2150 down_write(&md->io_lock);
2152 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2153 spin_lock_irq(&md->deferred_lock);
2154 c = bio_list_pop(&md->deferred);
2155 spin_unlock_irq(&md->deferred_lock);
2158 clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2162 up_write(&md->io_lock);
2164 if (dm_request_based(md))
2165 generic_make_request(c);
2168 process_barrier(md, c);
2170 __split_and_process_bio(md, c);
2173 down_write(&md->io_lock);
2176 up_write(&md->io_lock);
2179 static void dm_queue_flush(struct mapped_device *md)
2181 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2182 smp_mb__after_clear_bit();
2183 queue_work(md->wq, &md->work);
2187 * Swap in a new table (destroying old one).
2189 int dm_swap_table(struct mapped_device *md, struct dm_table *table)
2191 struct queue_limits limits;
2194 mutex_lock(&md->suspend_lock);
2196 /* device must be suspended */
2197 if (!dm_suspended(md))
2200 r = dm_calculate_queue_limits(table, &limits);
2204 /* cannot change the device type, once a table is bound */
2206 (dm_table_get_type(md->map) != dm_table_get_type(table))) {
2207 DMWARN("can't change the device type after a table is bound");
2212 r = __bind(md, table, &limits);
2215 mutex_unlock(&md->suspend_lock);
2219 static void dm_rq_invalidate_suspend_marker(struct mapped_device *md)
2221 md->suspend_rq.special = (void *)0x1;
2224 static void dm_rq_abort_suspend(struct mapped_device *md, int noflush)
2226 struct request_queue *q = md->queue;
2227 unsigned long flags;
2229 spin_lock_irqsave(q->queue_lock, flags);
2231 dm_rq_invalidate_suspend_marker(md);
2233 spin_unlock_irqrestore(q->queue_lock, flags);
2236 static void dm_rq_start_suspend(struct mapped_device *md, int noflush)
2238 struct request *rq = &md->suspend_rq;
2239 struct request_queue *q = md->queue;
2245 blk_insert_request(q, rq, 0, NULL);
2249 static int dm_rq_suspend_available(struct mapped_device *md, int noflush)
2252 struct request *rq = &md->suspend_rq;
2253 struct request_queue *q = md->queue;
2254 unsigned long flags;
2259 /* The marker must be protected by queue lock if it is in use */
2260 spin_lock_irqsave(q->queue_lock, flags);
2261 if (unlikely(rq->ref_count)) {
2263 * This can happen, when the previous flush suspend was
2264 * interrupted, the marker is still in the queue and
2265 * this flush suspend has been invoked, because we don't
2266 * remove the marker at the time of suspend interruption.
2267 * We have only one marker per mapped_device, so we can't
2268 * start another flush suspend while it is in use.
2270 BUG_ON(!rq->special); /* The marker should be invalidated */
2271 DMWARN("Invalidating the previous flush suspend is still in"
2272 " progress. Please retry later.");
2275 spin_unlock_irqrestore(q->queue_lock, flags);
2281 * Functions to lock and unlock any filesystem running on the
2284 static int lock_fs(struct mapped_device *md)
2288 WARN_ON(md->frozen_sb);
2290 md->frozen_sb = freeze_bdev(md->bdev);
2291 if (IS_ERR(md->frozen_sb)) {
2292 r = PTR_ERR(md->frozen_sb);
2293 md->frozen_sb = NULL;
2297 set_bit(DMF_FROZEN, &md->flags);
2302 static void unlock_fs(struct mapped_device *md)
2304 if (!test_bit(DMF_FROZEN, &md->flags))
2307 thaw_bdev(md->bdev, md->frozen_sb);
2308 md->frozen_sb = NULL;
2309 clear_bit(DMF_FROZEN, &md->flags);
2313 * We need to be able to change a mapping table under a mounted
2314 * filesystem. For example we might want to move some data in
2315 * the background. Before the table can be swapped with
2316 * dm_bind_table, dm_suspend must be called to flush any in
2317 * flight bios and ensure that any further io gets deferred.
2320 * Suspend mechanism in request-based dm.
2322 * After the suspend starts, further incoming requests are kept in
2323 * the request_queue and deferred.
2324 * Remaining requests in the request_queue at the start of suspend are flushed
2325 * if it is flush suspend.
2326 * The suspend completes when the following conditions have been satisfied,
2328 * 1. q->in_flight is 0 (which means no in_flight request)
2329 * 2. queue has been stopped (which means no request dispatching)
2334 * Noflush suspend doesn't need to dispatch remaining requests.
2335 * So stop the queue immediately. Then, wait for all in_flight requests
2336 * to be completed or requeued.
2338 * To abort noflush suspend, start the queue.
2343 * Flush suspend needs to dispatch remaining requests. So stop the queue
2344 * after the remaining requests are completed. (Requeued request must be also
2345 * re-dispatched and completed. Until then, we can't stop the queue.)
2347 * During flushing the remaining requests, further incoming requests are also
2348 * inserted to the same queue. To distinguish which requests are to be
2349 * flushed, we insert a marker request to the queue at the time of starting
2350 * flush suspend, like a barrier.
2351 * The dispatching is blocked when the marker is found on the top of the queue.
2352 * And the queue is stopped when all in_flight requests are completed, since
2353 * that means the remaining requests are completely flushed.
2354 * Then, the marker is removed from the queue.
2356 * To abort flush suspend, we also need to take care of the marker, not only
2357 * starting the queue.
2358 * We don't remove the marker forcibly from the queue since it's against
2359 * the block-layer manner. Instead, we put a invalidated mark on the marker.
2360 * When the invalidated marker is found on the top of the queue, it is
2361 * immediately removed from the queue, so it doesn't block dispatching.
2362 * Because we have only one marker per mapped_device, we can't start another
2363 * flush suspend until the invalidated marker is removed from the queue.
2364 * So fail and return with -EBUSY in such a case.
2366 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2368 struct dm_table *map = NULL;
2370 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2371 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2373 mutex_lock(&md->suspend_lock);
2375 if (dm_suspended(md)) {
2380 if (dm_request_based(md) && !dm_rq_suspend_available(md, noflush)) {
2385 map = dm_get_table(md);
2388 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2389 * This flag is cleared before dm_suspend returns.
2392 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2394 /* This does not get reverted if there's an error later. */
2395 dm_table_presuspend_targets(map);
2398 * Flush I/O to the device. noflush supersedes do_lockfs,
2399 * because lock_fs() needs to flush I/Os.
2401 if (!noflush && do_lockfs) {
2408 * Here we must make sure that no processes are submitting requests
2409 * to target drivers i.e. no one may be executing
2410 * __split_and_process_bio. This is called from dm_request and
2413 * To get all processes out of __split_and_process_bio in dm_request,
2414 * we take the write lock. To prevent any process from reentering
2415 * __split_and_process_bio from dm_request, we set
2416 * DMF_QUEUE_IO_TO_THREAD.
2418 * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND
2419 * and call flush_workqueue(md->wq). flush_workqueue will wait until
2420 * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any
2421 * further calls to __split_and_process_bio from dm_wq_work.
2423 down_write(&md->io_lock);
2424 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2425 set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags);
2426 up_write(&md->io_lock);
2428 flush_workqueue(md->wq);
2430 if (dm_request_based(md))
2431 dm_rq_start_suspend(md, noflush);
2434 * At this point no more requests are entering target request routines.
2435 * We call dm_wait_for_completion to wait for all existing requests
2438 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2440 down_write(&md->io_lock);
2442 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2443 up_write(&md->io_lock);
2445 /* were we interrupted ? */
2449 if (dm_request_based(md))
2450 dm_rq_abort_suspend(md, noflush);
2453 goto out; /* pushback list is already flushed, so skip flush */
2457 * If dm_wait_for_completion returned 0, the device is completely
2458 * quiescent now. There is no request-processing activity. All new
2459 * requests are being added to md->deferred list.
2462 dm_table_postsuspend_targets(map);
2464 set_bit(DMF_SUSPENDED, &md->flags);
2470 mutex_unlock(&md->suspend_lock);
2474 int dm_resume(struct mapped_device *md)
2477 struct dm_table *map = NULL;
2479 mutex_lock(&md->suspend_lock);
2480 if (!dm_suspended(md))
2483 map = dm_get_table(md);
2484 if (!map || !dm_table_get_size(map))
2487 r = dm_table_resume_targets(map);
2494 * Flushing deferred I/Os must be done after targets are resumed
2495 * so that mapping of targets can work correctly.
2496 * Request-based dm is queueing the deferred I/Os in its request_queue.
2498 if (dm_request_based(md))
2499 start_queue(md->queue);
2503 clear_bit(DMF_SUSPENDED, &md->flags);
2505 dm_table_unplug_all(map);
2509 mutex_unlock(&md->suspend_lock);
2514 /*-----------------------------------------------------------------
2515 * Event notification.
2516 *---------------------------------------------------------------*/
2517 void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2520 char udev_cookie[DM_COOKIE_LENGTH];
2521 char *envp[] = { udev_cookie, NULL };
2524 kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2526 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2527 DM_COOKIE_ENV_VAR_NAME, cookie);
2528 kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp);
2532 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2534 return atomic_add_return(1, &md->uevent_seq);
2537 uint32_t dm_get_event_nr(struct mapped_device *md)
2539 return atomic_read(&md->event_nr);
2542 int dm_wait_event(struct mapped_device *md, int event_nr)
2544 return wait_event_interruptible(md->eventq,
2545 (event_nr != atomic_read(&md->event_nr)));
2548 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2550 unsigned long flags;
2552 spin_lock_irqsave(&md->uevent_lock, flags);
2553 list_add(elist, &md->uevent_list);
2554 spin_unlock_irqrestore(&md->uevent_lock, flags);
2558 * The gendisk is only valid as long as you have a reference
2561 struct gendisk *dm_disk(struct mapped_device *md)
2566 struct kobject *dm_kobject(struct mapped_device *md)
2572 * struct mapped_device should not be exported outside of dm.c
2573 * so use this check to verify that kobj is part of md structure
2575 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2577 struct mapped_device *md;
2579 md = container_of(kobj, struct mapped_device, kobj);
2580 if (&md->kobj != kobj)
2583 if (test_bit(DMF_FREEING, &md->flags) ||
2584 test_bit(DMF_DELETING, &md->flags))
2591 int dm_suspended(struct mapped_device *md)
2593 return test_bit(DMF_SUSPENDED, &md->flags);
2596 int dm_noflush_suspending(struct dm_target *ti)
2598 struct mapped_device *md = dm_table_get_md(ti->table);
2599 int r = __noflush_suspending(md);
2605 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2607 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type)
2609 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2614 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2615 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2616 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2617 if (!pools->io_pool)
2618 goto free_pools_and_out;
2620 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2621 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2622 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2623 if (!pools->tio_pool)
2624 goto free_io_pool_and_out;
2626 pools->bs = (type == DM_TYPE_BIO_BASED) ?
2627 bioset_create(16, 0) : bioset_create(MIN_IOS, 0);
2629 goto free_tio_pool_and_out;
2633 free_tio_pool_and_out:
2634 mempool_destroy(pools->tio_pool);
2636 free_io_pool_and_out:
2637 mempool_destroy(pools->io_pool);
2645 void dm_free_md_mempools(struct dm_md_mempools *pools)
2651 mempool_destroy(pools->io_pool);
2653 if (pools->tio_pool)
2654 mempool_destroy(pools->tio_pool);
2657 bioset_free(pools->bs);
2662 static struct block_device_operations dm_blk_dops = {
2663 .open = dm_blk_open,
2664 .release = dm_blk_close,
2665 .ioctl = dm_blk_ioctl,
2666 .getgeo = dm_blk_getgeo,
2667 .owner = THIS_MODULE
2670 EXPORT_SYMBOL(dm_get_mapinfo);
2675 module_init(dm_init);
2676 module_exit(dm_exit);
2678 module_param(major, uint, 0);
2679 MODULE_PARM_DESC(major, "The major number of the device mapper");
2680 MODULE_DESCRIPTION(DM_NAME " driver");
2681 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2682 MODULE_LICENSE("GPL");