nilfs2: unbreak compat ioctl
[zen-stable.git] / drivers / md / dm.c
blob4720f68f817e66eb1e08da1535512b9cd6455fff
1 /*
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.
6 */
8 #include "dm.h"
9 #include "dm-uevent.h"
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>
22 #include <linux/delay.h>
24 #include <trace/events/block.h>
26 #define DM_MSG_PREFIX "core"
28 #ifdef CONFIG_PRINTK
30 * ratelimit state to be used in DMXXX_LIMIT().
32 DEFINE_RATELIMIT_STATE(dm_ratelimit_state,
33 DEFAULT_RATELIMIT_INTERVAL,
34 DEFAULT_RATELIMIT_BURST);
35 EXPORT_SYMBOL(dm_ratelimit_state);
36 #endif
39 * Cookies are numeric values sent with CHANGE and REMOVE
40 * uevents while resuming, removing or renaming the device.
42 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
43 #define DM_COOKIE_LENGTH 24
45 static const char *_name = DM_NAME;
47 static unsigned int major = 0;
48 static unsigned int _major = 0;
50 static DEFINE_IDR(_minor_idr);
52 static DEFINE_SPINLOCK(_minor_lock);
54 * For bio-based dm.
55 * One of these is allocated per bio.
57 struct dm_io {
58 struct mapped_device *md;
59 int error;
60 atomic_t io_count;
61 struct bio *bio;
62 unsigned long start_time;
63 spinlock_t endio_lock;
67 * For bio-based dm.
68 * One of these is allocated per target within a bio. Hopefully
69 * this will be simplified out one day.
71 struct dm_target_io {
72 struct dm_io *io;
73 struct dm_target *ti;
74 union map_info info;
78 * For request-based dm.
79 * One of these is allocated per request.
81 struct dm_rq_target_io {
82 struct mapped_device *md;
83 struct dm_target *ti;
84 struct request *orig, clone;
85 int error;
86 union map_info info;
90 * For request-based dm.
91 * One of these is allocated per bio.
93 struct dm_rq_clone_bio_info {
94 struct bio *orig;
95 struct dm_rq_target_io *tio;
98 union map_info *dm_get_mapinfo(struct bio *bio)
100 if (bio && bio->bi_private)
101 return &((struct dm_target_io *)bio->bi_private)->info;
102 return NULL;
105 union map_info *dm_get_rq_mapinfo(struct request *rq)
107 if (rq && rq->end_io_data)
108 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
109 return NULL;
111 EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
113 #define MINOR_ALLOCED ((void *)-1)
116 * Bits for the md->flags field.
118 #define DMF_BLOCK_IO_FOR_SUSPEND 0
119 #define DMF_SUSPENDED 1
120 #define DMF_FROZEN 2
121 #define DMF_FREEING 3
122 #define DMF_DELETING 4
123 #define DMF_NOFLUSH_SUSPENDING 5
124 #define DMF_MERGE_IS_OPTIONAL 6
127 * Work processed by per-device workqueue.
129 struct mapped_device {
130 struct rw_semaphore io_lock;
131 struct mutex suspend_lock;
132 rwlock_t map_lock;
133 atomic_t holders;
134 atomic_t open_count;
136 unsigned long flags;
138 struct request_queue *queue;
139 unsigned type;
140 /* Protect queue and type against concurrent access. */
141 struct mutex type_lock;
143 struct target_type *immutable_target_type;
145 struct gendisk *disk;
146 char name[16];
148 void *interface_ptr;
151 * A list of ios that arrived while we were suspended.
153 atomic_t pending[2];
154 wait_queue_head_t wait;
155 struct work_struct work;
156 struct bio_list deferred;
157 spinlock_t deferred_lock;
160 * Processing queue (flush)
162 struct workqueue_struct *wq;
165 * The current mapping.
167 struct dm_table *map;
170 * io objects are allocated from here.
172 mempool_t *io_pool;
173 mempool_t *tio_pool;
175 struct bio_set *bs;
178 * Event handling.
180 atomic_t event_nr;
181 wait_queue_head_t eventq;
182 atomic_t uevent_seq;
183 struct list_head uevent_list;
184 spinlock_t uevent_lock; /* Protect access to uevent_list */
187 * freeze/thaw support require holding onto a super block
189 struct super_block *frozen_sb;
190 struct block_device *bdev;
192 /* forced geometry settings */
193 struct hd_geometry geometry;
195 /* sysfs handle */
196 struct kobject kobj;
198 /* zero-length flush that will be cloned and submitted to targets */
199 struct bio flush_bio;
203 * For mempools pre-allocation at the table loading time.
205 struct dm_md_mempools {
206 mempool_t *io_pool;
207 mempool_t *tio_pool;
208 struct bio_set *bs;
211 #define MIN_IOS 256
212 static struct kmem_cache *_io_cache;
213 static struct kmem_cache *_tio_cache;
214 static struct kmem_cache *_rq_tio_cache;
215 static struct kmem_cache *_rq_bio_info_cache;
217 static int __init local_init(void)
219 int r = -ENOMEM;
221 /* allocate a slab for the dm_ios */
222 _io_cache = KMEM_CACHE(dm_io, 0);
223 if (!_io_cache)
224 return r;
226 /* allocate a slab for the target ios */
227 _tio_cache = KMEM_CACHE(dm_target_io, 0);
228 if (!_tio_cache)
229 goto out_free_io_cache;
231 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
232 if (!_rq_tio_cache)
233 goto out_free_tio_cache;
235 _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0);
236 if (!_rq_bio_info_cache)
237 goto out_free_rq_tio_cache;
239 r = dm_uevent_init();
240 if (r)
241 goto out_free_rq_bio_info_cache;
243 _major = major;
244 r = register_blkdev(_major, _name);
245 if (r < 0)
246 goto out_uevent_exit;
248 if (!_major)
249 _major = r;
251 return 0;
253 out_uevent_exit:
254 dm_uevent_exit();
255 out_free_rq_bio_info_cache:
256 kmem_cache_destroy(_rq_bio_info_cache);
257 out_free_rq_tio_cache:
258 kmem_cache_destroy(_rq_tio_cache);
259 out_free_tio_cache:
260 kmem_cache_destroy(_tio_cache);
261 out_free_io_cache:
262 kmem_cache_destroy(_io_cache);
264 return r;
267 static void local_exit(void)
269 kmem_cache_destroy(_rq_bio_info_cache);
270 kmem_cache_destroy(_rq_tio_cache);
271 kmem_cache_destroy(_tio_cache);
272 kmem_cache_destroy(_io_cache);
273 unregister_blkdev(_major, _name);
274 dm_uevent_exit();
276 _major = 0;
278 DMINFO("cleaned up");
281 static int (*_inits[])(void) __initdata = {
282 local_init,
283 dm_target_init,
284 dm_linear_init,
285 dm_stripe_init,
286 dm_io_init,
287 dm_kcopyd_init,
288 dm_interface_init,
291 static void (*_exits[])(void) = {
292 local_exit,
293 dm_target_exit,
294 dm_linear_exit,
295 dm_stripe_exit,
296 dm_io_exit,
297 dm_kcopyd_exit,
298 dm_interface_exit,
301 static int __init dm_init(void)
303 const int count = ARRAY_SIZE(_inits);
305 int r, i;
307 for (i = 0; i < count; i++) {
308 r = _inits[i]();
309 if (r)
310 goto bad;
313 return 0;
315 bad:
316 while (i--)
317 _exits[i]();
319 return r;
322 static void __exit dm_exit(void)
324 int i = ARRAY_SIZE(_exits);
326 while (i--)
327 _exits[i]();
330 * Should be empty by this point.
332 idr_remove_all(&_minor_idr);
333 idr_destroy(&_minor_idr);
337 * Block device functions
339 int dm_deleting_md(struct mapped_device *md)
341 return test_bit(DMF_DELETING, &md->flags);
344 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
346 struct mapped_device *md;
348 spin_lock(&_minor_lock);
350 md = bdev->bd_disk->private_data;
351 if (!md)
352 goto out;
354 if (test_bit(DMF_FREEING, &md->flags) ||
355 dm_deleting_md(md)) {
356 md = NULL;
357 goto out;
360 dm_get(md);
361 atomic_inc(&md->open_count);
363 out:
364 spin_unlock(&_minor_lock);
366 return md ? 0 : -ENXIO;
369 static int dm_blk_close(struct gendisk *disk, fmode_t mode)
371 struct mapped_device *md = disk->private_data;
373 spin_lock(&_minor_lock);
375 atomic_dec(&md->open_count);
376 dm_put(md);
378 spin_unlock(&_minor_lock);
380 return 0;
383 int dm_open_count(struct mapped_device *md)
385 return atomic_read(&md->open_count);
389 * Guarantees nothing is using the device before it's deleted.
391 int dm_lock_for_deletion(struct mapped_device *md)
393 int r = 0;
395 spin_lock(&_minor_lock);
397 if (dm_open_count(md))
398 r = -EBUSY;
399 else
400 set_bit(DMF_DELETING, &md->flags);
402 spin_unlock(&_minor_lock);
404 return r;
407 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
409 struct mapped_device *md = bdev->bd_disk->private_data;
411 return dm_get_geometry(md, geo);
414 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
415 unsigned int cmd, unsigned long arg)
417 struct mapped_device *md = bdev->bd_disk->private_data;
418 struct dm_table *map = dm_get_live_table(md);
419 struct dm_target *tgt;
420 int r = -ENOTTY;
422 if (!map || !dm_table_get_size(map))
423 goto out;
425 /* We only support devices that have a single target */
426 if (dm_table_get_num_targets(map) != 1)
427 goto out;
429 tgt = dm_table_get_target(map, 0);
431 if (dm_suspended_md(md)) {
432 r = -EAGAIN;
433 goto out;
436 if (tgt->type->ioctl)
437 r = tgt->type->ioctl(tgt, cmd, arg);
439 out:
440 dm_table_put(map);
442 return r;
445 static struct dm_io *alloc_io(struct mapped_device *md)
447 return mempool_alloc(md->io_pool, GFP_NOIO);
450 static void free_io(struct mapped_device *md, struct dm_io *io)
452 mempool_free(io, md->io_pool);
455 static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
457 mempool_free(tio, md->tio_pool);
460 static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
461 gfp_t gfp_mask)
463 return mempool_alloc(md->tio_pool, gfp_mask);
466 static void free_rq_tio(struct dm_rq_target_io *tio)
468 mempool_free(tio, tio->md->tio_pool);
471 static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md)
473 return mempool_alloc(md->io_pool, GFP_ATOMIC);
476 static void free_bio_info(struct dm_rq_clone_bio_info *info)
478 mempool_free(info, info->tio->md->io_pool);
481 static int md_in_flight(struct mapped_device *md)
483 return atomic_read(&md->pending[READ]) +
484 atomic_read(&md->pending[WRITE]);
487 static void start_io_acct(struct dm_io *io)
489 struct mapped_device *md = io->md;
490 int cpu;
491 int rw = bio_data_dir(io->bio);
493 io->start_time = jiffies;
495 cpu = part_stat_lock();
496 part_round_stats(cpu, &dm_disk(md)->part0);
497 part_stat_unlock();
498 atomic_set(&dm_disk(md)->part0.in_flight[rw],
499 atomic_inc_return(&md->pending[rw]));
502 static void end_io_acct(struct dm_io *io)
504 struct mapped_device *md = io->md;
505 struct bio *bio = io->bio;
506 unsigned long duration = jiffies - io->start_time;
507 int pending, cpu;
508 int rw = bio_data_dir(bio);
510 cpu = part_stat_lock();
511 part_round_stats(cpu, &dm_disk(md)->part0);
512 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
513 part_stat_unlock();
516 * After this is decremented the bio must not be touched if it is
517 * a flush.
519 pending = atomic_dec_return(&md->pending[rw]);
520 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
521 pending += atomic_read(&md->pending[rw^0x1]);
523 /* nudge anyone waiting on suspend queue */
524 if (!pending)
525 wake_up(&md->wait);
529 * Add the bio to the list of deferred io.
531 static void queue_io(struct mapped_device *md, struct bio *bio)
533 unsigned long flags;
535 spin_lock_irqsave(&md->deferred_lock, flags);
536 bio_list_add(&md->deferred, bio);
537 spin_unlock_irqrestore(&md->deferred_lock, flags);
538 queue_work(md->wq, &md->work);
542 * Everyone (including functions in this file), should use this
543 * function to access the md->map field, and make sure they call
544 * dm_table_put() when finished.
546 struct dm_table *dm_get_live_table(struct mapped_device *md)
548 struct dm_table *t;
549 unsigned long flags;
551 read_lock_irqsave(&md->map_lock, flags);
552 t = md->map;
553 if (t)
554 dm_table_get(t);
555 read_unlock_irqrestore(&md->map_lock, flags);
557 return t;
561 * Get the geometry associated with a dm device
563 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
565 *geo = md->geometry;
567 return 0;
571 * Set the geometry of a device.
573 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
575 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
577 if (geo->start > sz) {
578 DMWARN("Start sector is beyond the geometry limits.");
579 return -EINVAL;
582 md->geometry = *geo;
584 return 0;
587 /*-----------------------------------------------------------------
588 * CRUD START:
589 * A more elegant soln is in the works that uses the queue
590 * merge fn, unfortunately there are a couple of changes to
591 * the block layer that I want to make for this. So in the
592 * interests of getting something for people to use I give
593 * you this clearly demarcated crap.
594 *---------------------------------------------------------------*/
596 static int __noflush_suspending(struct mapped_device *md)
598 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
602 * Decrements the number of outstanding ios that a bio has been
603 * cloned into, completing the original io if necc.
605 static void dec_pending(struct dm_io *io, int error)
607 unsigned long flags;
608 int io_error;
609 struct bio *bio;
610 struct mapped_device *md = io->md;
612 /* Push-back supersedes any I/O errors */
613 if (unlikely(error)) {
614 spin_lock_irqsave(&io->endio_lock, flags);
615 if (!(io->error > 0 && __noflush_suspending(md)))
616 io->error = error;
617 spin_unlock_irqrestore(&io->endio_lock, flags);
620 if (atomic_dec_and_test(&io->io_count)) {
621 if (io->error == DM_ENDIO_REQUEUE) {
623 * Target requested pushing back the I/O.
625 spin_lock_irqsave(&md->deferred_lock, flags);
626 if (__noflush_suspending(md))
627 bio_list_add_head(&md->deferred, io->bio);
628 else
629 /* noflush suspend was interrupted. */
630 io->error = -EIO;
631 spin_unlock_irqrestore(&md->deferred_lock, flags);
634 io_error = io->error;
635 bio = io->bio;
636 end_io_acct(io);
637 free_io(md, io);
639 if (io_error == DM_ENDIO_REQUEUE)
640 return;
642 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) {
644 * Preflush done for flush with data, reissue
645 * without REQ_FLUSH.
647 bio->bi_rw &= ~REQ_FLUSH;
648 queue_io(md, bio);
649 } else {
650 /* done with normal IO or empty flush */
651 trace_block_bio_complete(md->queue, bio, io_error);
652 bio_endio(bio, io_error);
657 static void clone_endio(struct bio *bio, int error)
659 int r = 0;
660 struct dm_target_io *tio = bio->bi_private;
661 struct dm_io *io = tio->io;
662 struct mapped_device *md = tio->io->md;
663 dm_endio_fn endio = tio->ti->type->end_io;
665 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
666 error = -EIO;
668 if (endio) {
669 r = endio(tio->ti, bio, error, &tio->info);
670 if (r < 0 || r == DM_ENDIO_REQUEUE)
672 * error and requeue request are handled
673 * in dec_pending().
675 error = r;
676 else if (r == DM_ENDIO_INCOMPLETE)
677 /* The target will handle the io */
678 return;
679 else if (r) {
680 DMWARN("unimplemented target endio return value: %d", r);
681 BUG();
686 * Store md for cleanup instead of tio which is about to get freed.
688 bio->bi_private = md->bs;
690 free_tio(md, tio);
691 bio_put(bio);
692 dec_pending(io, error);
696 * Partial completion handling for request-based dm
698 static void end_clone_bio(struct bio *clone, int error)
700 struct dm_rq_clone_bio_info *info = clone->bi_private;
701 struct dm_rq_target_io *tio = info->tio;
702 struct bio *bio = info->orig;
703 unsigned int nr_bytes = info->orig->bi_size;
705 bio_put(clone);
707 if (tio->error)
709 * An error has already been detected on the request.
710 * Once error occurred, just let clone->end_io() handle
711 * the remainder.
713 return;
714 else if (error) {
716 * Don't notice the error to the upper layer yet.
717 * The error handling decision is made by the target driver,
718 * when the request is completed.
720 tio->error = error;
721 return;
725 * I/O for the bio successfully completed.
726 * Notice the data completion to the upper layer.
730 * bios are processed from the head of the list.
731 * So the completing bio should always be rq->bio.
732 * If it's not, something wrong is happening.
734 if (tio->orig->bio != bio)
735 DMERR("bio completion is going in the middle of the request");
738 * Update the original request.
739 * Do not use blk_end_request() here, because it may complete
740 * the original request before the clone, and break the ordering.
742 blk_update_request(tio->orig, 0, nr_bytes);
746 * Don't touch any member of the md after calling this function because
747 * the md may be freed in dm_put() at the end of this function.
748 * Or do dm_get() before calling this function and dm_put() later.
750 static void rq_completed(struct mapped_device *md, int rw, int run_queue)
752 atomic_dec(&md->pending[rw]);
754 /* nudge anyone waiting on suspend queue */
755 if (!md_in_flight(md))
756 wake_up(&md->wait);
758 if (run_queue)
759 blk_run_queue(md->queue);
762 * dm_put() must be at the end of this function. See the comment above
764 dm_put(md);
767 static void free_rq_clone(struct request *clone)
769 struct dm_rq_target_io *tio = clone->end_io_data;
771 blk_rq_unprep_clone(clone);
772 free_rq_tio(tio);
776 * Complete the clone and the original request.
777 * Must be called without queue lock.
779 static void dm_end_request(struct request *clone, int error)
781 int rw = rq_data_dir(clone);
782 struct dm_rq_target_io *tio = clone->end_io_data;
783 struct mapped_device *md = tio->md;
784 struct request *rq = tio->orig;
786 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
787 rq->errors = clone->errors;
788 rq->resid_len = clone->resid_len;
790 if (rq->sense)
792 * We are using the sense buffer of the original
793 * request.
794 * So setting the length of the sense data is enough.
796 rq->sense_len = clone->sense_len;
799 free_rq_clone(clone);
800 blk_end_request_all(rq, error);
801 rq_completed(md, rw, true);
804 static void dm_unprep_request(struct request *rq)
806 struct request *clone = rq->special;
808 rq->special = NULL;
809 rq->cmd_flags &= ~REQ_DONTPREP;
811 free_rq_clone(clone);
815 * Requeue the original request of a clone.
817 void dm_requeue_unmapped_request(struct request *clone)
819 int rw = rq_data_dir(clone);
820 struct dm_rq_target_io *tio = clone->end_io_data;
821 struct mapped_device *md = tio->md;
822 struct request *rq = tio->orig;
823 struct request_queue *q = rq->q;
824 unsigned long flags;
826 dm_unprep_request(rq);
828 spin_lock_irqsave(q->queue_lock, flags);
829 blk_requeue_request(q, rq);
830 spin_unlock_irqrestore(q->queue_lock, flags);
832 rq_completed(md, rw, 0);
834 EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
836 static void __stop_queue(struct request_queue *q)
838 blk_stop_queue(q);
841 static void stop_queue(struct request_queue *q)
843 unsigned long flags;
845 spin_lock_irqsave(q->queue_lock, flags);
846 __stop_queue(q);
847 spin_unlock_irqrestore(q->queue_lock, flags);
850 static void __start_queue(struct request_queue *q)
852 if (blk_queue_stopped(q))
853 blk_start_queue(q);
856 static void start_queue(struct request_queue *q)
858 unsigned long flags;
860 spin_lock_irqsave(q->queue_lock, flags);
861 __start_queue(q);
862 spin_unlock_irqrestore(q->queue_lock, flags);
865 static void dm_done(struct request *clone, int error, bool mapped)
867 int r = error;
868 struct dm_rq_target_io *tio = clone->end_io_data;
869 dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io;
871 if (mapped && rq_end_io)
872 r = rq_end_io(tio->ti, clone, error, &tio->info);
874 if (r <= 0)
875 /* The target wants to complete the I/O */
876 dm_end_request(clone, r);
877 else if (r == DM_ENDIO_INCOMPLETE)
878 /* The target will handle the I/O */
879 return;
880 else if (r == DM_ENDIO_REQUEUE)
881 /* The target wants to requeue the I/O */
882 dm_requeue_unmapped_request(clone);
883 else {
884 DMWARN("unimplemented target endio return value: %d", r);
885 BUG();
890 * Request completion handler for request-based dm
892 static void dm_softirq_done(struct request *rq)
894 bool mapped = true;
895 struct request *clone = rq->completion_data;
896 struct dm_rq_target_io *tio = clone->end_io_data;
898 if (rq->cmd_flags & REQ_FAILED)
899 mapped = false;
901 dm_done(clone, tio->error, mapped);
905 * Complete the clone and the original request with the error status
906 * through softirq context.
908 static void dm_complete_request(struct request *clone, int error)
910 struct dm_rq_target_io *tio = clone->end_io_data;
911 struct request *rq = tio->orig;
913 tio->error = error;
914 rq->completion_data = clone;
915 blk_complete_request(rq);
919 * Complete the not-mapped clone and the original request with the error status
920 * through softirq context.
921 * Target's rq_end_io() function isn't called.
922 * This may be used when the target's map_rq() function fails.
924 void dm_kill_unmapped_request(struct request *clone, int error)
926 struct dm_rq_target_io *tio = clone->end_io_data;
927 struct request *rq = tio->orig;
929 rq->cmd_flags |= REQ_FAILED;
930 dm_complete_request(clone, error);
932 EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
935 * Called with the queue lock held
937 static void end_clone_request(struct request *clone, int error)
940 * For just cleaning up the information of the queue in which
941 * the clone was dispatched.
942 * The clone is *NOT* freed actually here because it is alloced from
943 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
945 __blk_put_request(clone->q, clone);
948 * Actual request completion is done in a softirq context which doesn't
949 * hold the queue lock. Otherwise, deadlock could occur because:
950 * - another request may be submitted by the upper level driver
951 * of the stacking during the completion
952 * - the submission which requires queue lock may be done
953 * against this queue
955 dm_complete_request(clone, error);
959 * Return maximum size of I/O possible at the supplied sector up to the current
960 * target boundary.
962 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
964 sector_t target_offset = dm_target_offset(ti, sector);
966 return ti->len - target_offset;
969 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
971 sector_t len = max_io_len_target_boundary(sector, ti);
974 * Does the target need to split even further ?
976 if (ti->split_io) {
977 sector_t boundary;
978 sector_t offset = dm_target_offset(ti, sector);
979 boundary = ((offset + ti->split_io) & ~(ti->split_io - 1))
980 - offset;
981 if (len > boundary)
982 len = boundary;
985 return len;
988 static void __map_bio(struct dm_target *ti, struct bio *clone,
989 struct dm_target_io *tio)
991 int r;
992 sector_t sector;
993 struct mapped_device *md;
995 clone->bi_end_io = clone_endio;
996 clone->bi_private = tio;
999 * Map the clone. If r == 0 we don't need to do
1000 * anything, the target has assumed ownership of
1001 * this io.
1003 atomic_inc(&tio->io->io_count);
1004 sector = clone->bi_sector;
1005 r = ti->type->map(ti, clone, &tio->info);
1006 if (r == DM_MAPIO_REMAPPED) {
1007 /* the bio has been remapped so dispatch it */
1009 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1010 tio->io->bio->bi_bdev->bd_dev, sector);
1012 generic_make_request(clone);
1013 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1014 /* error the io and bail out, or requeue it if needed */
1015 md = tio->io->md;
1016 dec_pending(tio->io, r);
1018 * Store bio_set for cleanup.
1020 clone->bi_private = md->bs;
1021 bio_put(clone);
1022 free_tio(md, tio);
1023 } else if (r) {
1024 DMWARN("unimplemented target map return value: %d", r);
1025 BUG();
1029 struct clone_info {
1030 struct mapped_device *md;
1031 struct dm_table *map;
1032 struct bio *bio;
1033 struct dm_io *io;
1034 sector_t sector;
1035 sector_t sector_count;
1036 unsigned short idx;
1039 static void dm_bio_destructor(struct bio *bio)
1041 struct bio_set *bs = bio->bi_private;
1043 bio_free(bio, bs);
1047 * Creates a little bio that just does part of a bvec.
1049 static struct bio *split_bvec(struct bio *bio, sector_t sector,
1050 unsigned short idx, unsigned int offset,
1051 unsigned int len, struct bio_set *bs)
1053 struct bio *clone;
1054 struct bio_vec *bv = bio->bi_io_vec + idx;
1056 clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
1057 clone->bi_destructor = dm_bio_destructor;
1058 *clone->bi_io_vec = *bv;
1060 clone->bi_sector = sector;
1061 clone->bi_bdev = bio->bi_bdev;
1062 clone->bi_rw = bio->bi_rw;
1063 clone->bi_vcnt = 1;
1064 clone->bi_size = to_bytes(len);
1065 clone->bi_io_vec->bv_offset = offset;
1066 clone->bi_io_vec->bv_len = clone->bi_size;
1067 clone->bi_flags |= 1 << BIO_CLONED;
1069 if (bio_integrity(bio)) {
1070 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1071 bio_integrity_trim(clone,
1072 bio_sector_offset(bio, idx, offset), len);
1075 return clone;
1079 * Creates a bio that consists of range of complete bvecs.
1081 static struct bio *clone_bio(struct bio *bio, sector_t sector,
1082 unsigned short idx, unsigned short bv_count,
1083 unsigned int len, struct bio_set *bs)
1085 struct bio *clone;
1087 clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
1088 __bio_clone(clone, bio);
1089 clone->bi_destructor = dm_bio_destructor;
1090 clone->bi_sector = sector;
1091 clone->bi_idx = idx;
1092 clone->bi_vcnt = idx + bv_count;
1093 clone->bi_size = to_bytes(len);
1094 clone->bi_flags &= ~(1 << BIO_SEG_VALID);
1096 if (bio_integrity(bio)) {
1097 bio_integrity_clone(clone, bio, GFP_NOIO, bs);
1099 if (idx != bio->bi_idx || clone->bi_size < bio->bi_size)
1100 bio_integrity_trim(clone,
1101 bio_sector_offset(bio, idx, 0), len);
1104 return clone;
1107 static struct dm_target_io *alloc_tio(struct clone_info *ci,
1108 struct dm_target *ti)
1110 struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO);
1112 tio->io = ci->io;
1113 tio->ti = ti;
1114 memset(&tio->info, 0, sizeof(tio->info));
1116 return tio;
1119 static void __issue_target_request(struct clone_info *ci, struct dm_target *ti,
1120 unsigned request_nr, sector_t len)
1122 struct dm_target_io *tio = alloc_tio(ci, ti);
1123 struct bio *clone;
1125 tio->info.target_request_nr = request_nr;
1128 * Discard requests require the bio's inline iovecs be initialized.
1129 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1130 * and discard, so no need for concern about wasted bvec allocations.
1132 clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs);
1133 __bio_clone(clone, ci->bio);
1134 clone->bi_destructor = dm_bio_destructor;
1135 if (len) {
1136 clone->bi_sector = ci->sector;
1137 clone->bi_size = to_bytes(len);
1140 __map_bio(ti, clone, tio);
1143 static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti,
1144 unsigned num_requests, sector_t len)
1146 unsigned request_nr;
1148 for (request_nr = 0; request_nr < num_requests; request_nr++)
1149 __issue_target_request(ci, ti, request_nr, len);
1152 static int __clone_and_map_empty_flush(struct clone_info *ci)
1154 unsigned target_nr = 0;
1155 struct dm_target *ti;
1157 BUG_ON(bio_has_data(ci->bio));
1158 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1159 __issue_target_requests(ci, ti, ti->num_flush_requests, 0);
1161 return 0;
1165 * Perform all io with a single clone.
1167 static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti)
1169 struct bio *clone, *bio = ci->bio;
1170 struct dm_target_io *tio;
1172 tio = alloc_tio(ci, ti);
1173 clone = clone_bio(bio, ci->sector, ci->idx,
1174 bio->bi_vcnt - ci->idx, ci->sector_count,
1175 ci->md->bs);
1176 __map_bio(ti, clone, tio);
1177 ci->sector_count = 0;
1180 static int __clone_and_map_discard(struct clone_info *ci)
1182 struct dm_target *ti;
1183 sector_t len;
1185 do {
1186 ti = dm_table_find_target(ci->map, ci->sector);
1187 if (!dm_target_is_valid(ti))
1188 return -EIO;
1191 * Even though the device advertised discard support,
1192 * that does not mean every target supports it, and
1193 * reconfiguration might also have changed that since the
1194 * check was performed.
1196 if (!ti->num_discard_requests)
1197 return -EOPNOTSUPP;
1199 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1201 __issue_target_requests(ci, ti, ti->num_discard_requests, len);
1203 ci->sector += len;
1204 } while (ci->sector_count -= len);
1206 return 0;
1209 static int __clone_and_map(struct clone_info *ci)
1211 struct bio *clone, *bio = ci->bio;
1212 struct dm_target *ti;
1213 sector_t len = 0, max;
1214 struct dm_target_io *tio;
1216 if (unlikely(bio->bi_rw & REQ_DISCARD))
1217 return __clone_and_map_discard(ci);
1219 ti = dm_table_find_target(ci->map, ci->sector);
1220 if (!dm_target_is_valid(ti))
1221 return -EIO;
1223 max = max_io_len(ci->sector, ti);
1225 if (ci->sector_count <= max) {
1227 * Optimise for the simple case where we can do all of
1228 * the remaining io with a single clone.
1230 __clone_and_map_simple(ci, ti);
1232 } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) {
1234 * There are some bvecs that don't span targets.
1235 * Do as many of these as possible.
1237 int i;
1238 sector_t remaining = max;
1239 sector_t bv_len;
1241 for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) {
1242 bv_len = to_sector(bio->bi_io_vec[i].bv_len);
1244 if (bv_len > remaining)
1245 break;
1247 remaining -= bv_len;
1248 len += bv_len;
1251 tio = alloc_tio(ci, ti);
1252 clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
1253 ci->md->bs);
1254 __map_bio(ti, clone, tio);
1256 ci->sector += len;
1257 ci->sector_count -= len;
1258 ci->idx = i;
1260 } else {
1262 * Handle a bvec that must be split between two or more targets.
1264 struct bio_vec *bv = bio->bi_io_vec + ci->idx;
1265 sector_t remaining = to_sector(bv->bv_len);
1266 unsigned int offset = 0;
1268 do {
1269 if (offset) {
1270 ti = dm_table_find_target(ci->map, ci->sector);
1271 if (!dm_target_is_valid(ti))
1272 return -EIO;
1274 max = max_io_len(ci->sector, ti);
1277 len = min(remaining, max);
1279 tio = alloc_tio(ci, ti);
1280 clone = split_bvec(bio, ci->sector, ci->idx,
1281 bv->bv_offset + offset, len,
1282 ci->md->bs);
1284 __map_bio(ti, clone, tio);
1286 ci->sector += len;
1287 ci->sector_count -= len;
1288 offset += to_bytes(len);
1289 } while (remaining -= len);
1291 ci->idx++;
1294 return 0;
1298 * Split the bio into several clones and submit it to targets.
1300 static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
1302 struct clone_info ci;
1303 int error = 0;
1305 ci.map = dm_get_live_table(md);
1306 if (unlikely(!ci.map)) {
1307 bio_io_error(bio);
1308 return;
1311 ci.md = md;
1312 ci.io = alloc_io(md);
1313 ci.io->error = 0;
1314 atomic_set(&ci.io->io_count, 1);
1315 ci.io->bio = bio;
1316 ci.io->md = md;
1317 spin_lock_init(&ci.io->endio_lock);
1318 ci.sector = bio->bi_sector;
1319 ci.idx = bio->bi_idx;
1321 start_io_acct(ci.io);
1322 if (bio->bi_rw & REQ_FLUSH) {
1323 ci.bio = &ci.md->flush_bio;
1324 ci.sector_count = 0;
1325 error = __clone_and_map_empty_flush(&ci);
1326 /* dec_pending submits any data associated with flush */
1327 } else {
1328 ci.bio = bio;
1329 ci.sector_count = bio_sectors(bio);
1330 while (ci.sector_count && !error)
1331 error = __clone_and_map(&ci);
1334 /* drop the extra reference count */
1335 dec_pending(ci.io, error);
1336 dm_table_put(ci.map);
1338 /*-----------------------------------------------------------------
1339 * CRUD END
1340 *---------------------------------------------------------------*/
1342 static int dm_merge_bvec(struct request_queue *q,
1343 struct bvec_merge_data *bvm,
1344 struct bio_vec *biovec)
1346 struct mapped_device *md = q->queuedata;
1347 struct dm_table *map = dm_get_live_table(md);
1348 struct dm_target *ti;
1349 sector_t max_sectors;
1350 int max_size = 0;
1352 if (unlikely(!map))
1353 goto out;
1355 ti = dm_table_find_target(map, bvm->bi_sector);
1356 if (!dm_target_is_valid(ti))
1357 goto out_table;
1360 * Find maximum amount of I/O that won't need splitting
1362 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1363 (sector_t) BIO_MAX_SECTORS);
1364 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1365 if (max_size < 0)
1366 max_size = 0;
1369 * merge_bvec_fn() returns number of bytes
1370 * it can accept at this offset
1371 * max is precomputed maximal io size
1373 if (max_size && ti->type->merge)
1374 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1376 * If the target doesn't support merge method and some of the devices
1377 * provided their merge_bvec method (we know this by looking at
1378 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1379 * entries. So always set max_size to 0, and the code below allows
1380 * just one page.
1382 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1384 max_size = 0;
1386 out_table:
1387 dm_table_put(map);
1389 out:
1391 * Always allow an entire first page
1393 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1394 max_size = biovec->bv_len;
1396 return max_size;
1400 * The request function that just remaps the bio built up by
1401 * dm_merge_bvec.
1403 static void _dm_request(struct request_queue *q, struct bio *bio)
1405 int rw = bio_data_dir(bio);
1406 struct mapped_device *md = q->queuedata;
1407 int cpu;
1409 down_read(&md->io_lock);
1411 cpu = part_stat_lock();
1412 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1413 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1414 part_stat_unlock();
1416 /* if we're suspended, we have to queue this io for later */
1417 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1418 up_read(&md->io_lock);
1420 if (bio_rw(bio) != READA)
1421 queue_io(md, bio);
1422 else
1423 bio_io_error(bio);
1424 return;
1427 __split_and_process_bio(md, bio);
1428 up_read(&md->io_lock);
1429 return;
1432 static int dm_request_based(struct mapped_device *md)
1434 return blk_queue_stackable(md->queue);
1437 static void dm_request(struct request_queue *q, struct bio *bio)
1439 struct mapped_device *md = q->queuedata;
1441 if (dm_request_based(md))
1442 blk_queue_bio(q, bio);
1443 else
1444 _dm_request(q, bio);
1447 void dm_dispatch_request(struct request *rq)
1449 int r;
1451 if (blk_queue_io_stat(rq->q))
1452 rq->cmd_flags |= REQ_IO_STAT;
1454 rq->start_time = jiffies;
1455 r = blk_insert_cloned_request(rq->q, rq);
1456 if (r)
1457 dm_complete_request(rq, r);
1459 EXPORT_SYMBOL_GPL(dm_dispatch_request);
1461 static void dm_rq_bio_destructor(struct bio *bio)
1463 struct dm_rq_clone_bio_info *info = bio->bi_private;
1464 struct mapped_device *md = info->tio->md;
1466 free_bio_info(info);
1467 bio_free(bio, md->bs);
1470 static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1471 void *data)
1473 struct dm_rq_target_io *tio = data;
1474 struct mapped_device *md = tio->md;
1475 struct dm_rq_clone_bio_info *info = alloc_bio_info(md);
1477 if (!info)
1478 return -ENOMEM;
1480 info->orig = bio_orig;
1481 info->tio = tio;
1482 bio->bi_end_io = end_clone_bio;
1483 bio->bi_private = info;
1484 bio->bi_destructor = dm_rq_bio_destructor;
1486 return 0;
1489 static int setup_clone(struct request *clone, struct request *rq,
1490 struct dm_rq_target_io *tio)
1492 int r;
1494 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1495 dm_rq_bio_constructor, tio);
1496 if (r)
1497 return r;
1499 clone->cmd = rq->cmd;
1500 clone->cmd_len = rq->cmd_len;
1501 clone->sense = rq->sense;
1502 clone->buffer = rq->buffer;
1503 clone->end_io = end_clone_request;
1504 clone->end_io_data = tio;
1506 return 0;
1509 static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1510 gfp_t gfp_mask)
1512 struct request *clone;
1513 struct dm_rq_target_io *tio;
1515 tio = alloc_rq_tio(md, gfp_mask);
1516 if (!tio)
1517 return NULL;
1519 tio->md = md;
1520 tio->ti = NULL;
1521 tio->orig = rq;
1522 tio->error = 0;
1523 memset(&tio->info, 0, sizeof(tio->info));
1525 clone = &tio->clone;
1526 if (setup_clone(clone, rq, tio)) {
1527 /* -ENOMEM */
1528 free_rq_tio(tio);
1529 return NULL;
1532 return clone;
1536 * Called with the queue lock held.
1538 static int dm_prep_fn(struct request_queue *q, struct request *rq)
1540 struct mapped_device *md = q->queuedata;
1541 struct request *clone;
1543 if (unlikely(rq->special)) {
1544 DMWARN("Already has something in rq->special.");
1545 return BLKPREP_KILL;
1548 clone = clone_rq(rq, md, GFP_ATOMIC);
1549 if (!clone)
1550 return BLKPREP_DEFER;
1552 rq->special = clone;
1553 rq->cmd_flags |= REQ_DONTPREP;
1555 return BLKPREP_OK;
1559 * Returns:
1560 * 0 : the request has been processed (not requeued)
1561 * !0 : the request has been requeued
1563 static int map_request(struct dm_target *ti, struct request *clone,
1564 struct mapped_device *md)
1566 int r, requeued = 0;
1567 struct dm_rq_target_io *tio = clone->end_io_data;
1570 * Hold the md reference here for the in-flight I/O.
1571 * We can't rely on the reference count by device opener,
1572 * because the device may be closed during the request completion
1573 * when all bios are completed.
1574 * See the comment in rq_completed() too.
1576 dm_get(md);
1578 tio->ti = ti;
1579 r = ti->type->map_rq(ti, clone, &tio->info);
1580 switch (r) {
1581 case DM_MAPIO_SUBMITTED:
1582 /* The target has taken the I/O to submit by itself later */
1583 break;
1584 case DM_MAPIO_REMAPPED:
1585 /* The target has remapped the I/O so dispatch it */
1586 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1587 blk_rq_pos(tio->orig));
1588 dm_dispatch_request(clone);
1589 break;
1590 case DM_MAPIO_REQUEUE:
1591 /* The target wants to requeue the I/O */
1592 dm_requeue_unmapped_request(clone);
1593 requeued = 1;
1594 break;
1595 default:
1596 if (r > 0) {
1597 DMWARN("unimplemented target map return value: %d", r);
1598 BUG();
1601 /* The target wants to complete the I/O */
1602 dm_kill_unmapped_request(clone, r);
1603 break;
1606 return requeued;
1610 * q->request_fn for request-based dm.
1611 * Called with the queue lock held.
1613 static void dm_request_fn(struct request_queue *q)
1615 struct mapped_device *md = q->queuedata;
1616 struct dm_table *map = dm_get_live_table(md);
1617 struct dm_target *ti;
1618 struct request *rq, *clone;
1619 sector_t pos;
1622 * For suspend, check blk_queue_stopped() and increment
1623 * ->pending within a single queue_lock not to increment the
1624 * number of in-flight I/Os after the queue is stopped in
1625 * dm_suspend().
1627 while (!blk_queue_stopped(q)) {
1628 rq = blk_peek_request(q);
1629 if (!rq)
1630 goto delay_and_out;
1632 /* always use block 0 to find the target for flushes for now */
1633 pos = 0;
1634 if (!(rq->cmd_flags & REQ_FLUSH))
1635 pos = blk_rq_pos(rq);
1637 ti = dm_table_find_target(map, pos);
1638 BUG_ON(!dm_target_is_valid(ti));
1640 if (ti->type->busy && ti->type->busy(ti))
1641 goto delay_and_out;
1643 blk_start_request(rq);
1644 clone = rq->special;
1645 atomic_inc(&md->pending[rq_data_dir(clone)]);
1647 spin_unlock(q->queue_lock);
1648 if (map_request(ti, clone, md))
1649 goto requeued;
1651 BUG_ON(!irqs_disabled());
1652 spin_lock(q->queue_lock);
1655 goto out;
1657 requeued:
1658 BUG_ON(!irqs_disabled());
1659 spin_lock(q->queue_lock);
1661 delay_and_out:
1662 blk_delay_queue(q, HZ / 10);
1663 out:
1664 dm_table_put(map);
1666 return;
1669 int dm_underlying_device_busy(struct request_queue *q)
1671 return blk_lld_busy(q);
1673 EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1675 static int dm_lld_busy(struct request_queue *q)
1677 int r;
1678 struct mapped_device *md = q->queuedata;
1679 struct dm_table *map = dm_get_live_table(md);
1681 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1682 r = 1;
1683 else
1684 r = dm_table_any_busy_target(map);
1686 dm_table_put(map);
1688 return r;
1691 static int dm_any_congested(void *congested_data, int bdi_bits)
1693 int r = bdi_bits;
1694 struct mapped_device *md = congested_data;
1695 struct dm_table *map;
1697 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1698 map = dm_get_live_table(md);
1699 if (map) {
1701 * Request-based dm cares about only own queue for
1702 * the query about congestion status of request_queue
1704 if (dm_request_based(md))
1705 r = md->queue->backing_dev_info.state &
1706 bdi_bits;
1707 else
1708 r = dm_table_any_congested(map, bdi_bits);
1710 dm_table_put(map);
1714 return r;
1717 /*-----------------------------------------------------------------
1718 * An IDR is used to keep track of allocated minor numbers.
1719 *---------------------------------------------------------------*/
1720 static void free_minor(int minor)
1722 spin_lock(&_minor_lock);
1723 idr_remove(&_minor_idr, minor);
1724 spin_unlock(&_minor_lock);
1728 * See if the device with a specific minor # is free.
1730 static int specific_minor(int minor)
1732 int r, m;
1734 if (minor >= (1 << MINORBITS))
1735 return -EINVAL;
1737 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1738 if (!r)
1739 return -ENOMEM;
1741 spin_lock(&_minor_lock);
1743 if (idr_find(&_minor_idr, minor)) {
1744 r = -EBUSY;
1745 goto out;
1748 r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m);
1749 if (r)
1750 goto out;
1752 if (m != minor) {
1753 idr_remove(&_minor_idr, m);
1754 r = -EBUSY;
1755 goto out;
1758 out:
1759 spin_unlock(&_minor_lock);
1760 return r;
1763 static int next_free_minor(int *minor)
1765 int r, m;
1767 r = idr_pre_get(&_minor_idr, GFP_KERNEL);
1768 if (!r)
1769 return -ENOMEM;
1771 spin_lock(&_minor_lock);
1773 r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m);
1774 if (r)
1775 goto out;
1777 if (m >= (1 << MINORBITS)) {
1778 idr_remove(&_minor_idr, m);
1779 r = -ENOSPC;
1780 goto out;
1783 *minor = m;
1785 out:
1786 spin_unlock(&_minor_lock);
1787 return r;
1790 static const struct block_device_operations dm_blk_dops;
1792 static void dm_wq_work(struct work_struct *work);
1794 static void dm_init_md_queue(struct mapped_device *md)
1797 * Request-based dm devices cannot be stacked on top of bio-based dm
1798 * devices. The type of this dm device has not been decided yet.
1799 * The type is decided at the first table loading time.
1800 * To prevent problematic device stacking, clear the queue flag
1801 * for request stacking support until then.
1803 * This queue is new, so no concurrency on the queue_flags.
1805 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1807 md->queue->queuedata = md;
1808 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1809 md->queue->backing_dev_info.congested_data = md;
1810 blk_queue_make_request(md->queue, dm_request);
1811 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1812 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1816 * Allocate and initialise a blank device with a given minor.
1818 static struct mapped_device *alloc_dev(int minor)
1820 int r;
1821 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1822 void *old_md;
1824 if (!md) {
1825 DMWARN("unable to allocate device, out of memory.");
1826 return NULL;
1829 if (!try_module_get(THIS_MODULE))
1830 goto bad_module_get;
1832 /* get a minor number for the dev */
1833 if (minor == DM_ANY_MINOR)
1834 r = next_free_minor(&minor);
1835 else
1836 r = specific_minor(minor);
1837 if (r < 0)
1838 goto bad_minor;
1840 md->type = DM_TYPE_NONE;
1841 init_rwsem(&md->io_lock);
1842 mutex_init(&md->suspend_lock);
1843 mutex_init(&md->type_lock);
1844 spin_lock_init(&md->deferred_lock);
1845 rwlock_init(&md->map_lock);
1846 atomic_set(&md->holders, 1);
1847 atomic_set(&md->open_count, 0);
1848 atomic_set(&md->event_nr, 0);
1849 atomic_set(&md->uevent_seq, 0);
1850 INIT_LIST_HEAD(&md->uevent_list);
1851 spin_lock_init(&md->uevent_lock);
1853 md->queue = blk_alloc_queue(GFP_KERNEL);
1854 if (!md->queue)
1855 goto bad_queue;
1857 dm_init_md_queue(md);
1859 md->disk = alloc_disk(1);
1860 if (!md->disk)
1861 goto bad_disk;
1863 atomic_set(&md->pending[0], 0);
1864 atomic_set(&md->pending[1], 0);
1865 init_waitqueue_head(&md->wait);
1866 INIT_WORK(&md->work, dm_wq_work);
1867 init_waitqueue_head(&md->eventq);
1869 md->disk->major = _major;
1870 md->disk->first_minor = minor;
1871 md->disk->fops = &dm_blk_dops;
1872 md->disk->queue = md->queue;
1873 md->disk->private_data = md;
1874 sprintf(md->disk->disk_name, "dm-%d", minor);
1875 add_disk(md->disk);
1876 format_dev_t(md->name, MKDEV(_major, minor));
1878 md->wq = alloc_workqueue("kdmflush",
1879 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
1880 if (!md->wq)
1881 goto bad_thread;
1883 md->bdev = bdget_disk(md->disk, 0);
1884 if (!md->bdev)
1885 goto bad_bdev;
1887 bio_init(&md->flush_bio);
1888 md->flush_bio.bi_bdev = md->bdev;
1889 md->flush_bio.bi_rw = WRITE_FLUSH;
1891 /* Populate the mapping, nobody knows we exist yet */
1892 spin_lock(&_minor_lock);
1893 old_md = idr_replace(&_minor_idr, md, minor);
1894 spin_unlock(&_minor_lock);
1896 BUG_ON(old_md != MINOR_ALLOCED);
1898 return md;
1900 bad_bdev:
1901 destroy_workqueue(md->wq);
1902 bad_thread:
1903 del_gendisk(md->disk);
1904 put_disk(md->disk);
1905 bad_disk:
1906 blk_cleanup_queue(md->queue);
1907 bad_queue:
1908 free_minor(minor);
1909 bad_minor:
1910 module_put(THIS_MODULE);
1911 bad_module_get:
1912 kfree(md);
1913 return NULL;
1916 static void unlock_fs(struct mapped_device *md);
1918 static void free_dev(struct mapped_device *md)
1920 int minor = MINOR(disk_devt(md->disk));
1922 unlock_fs(md);
1923 bdput(md->bdev);
1924 destroy_workqueue(md->wq);
1925 if (md->tio_pool)
1926 mempool_destroy(md->tio_pool);
1927 if (md->io_pool)
1928 mempool_destroy(md->io_pool);
1929 if (md->bs)
1930 bioset_free(md->bs);
1931 blk_integrity_unregister(md->disk);
1932 del_gendisk(md->disk);
1933 free_minor(minor);
1935 spin_lock(&_minor_lock);
1936 md->disk->private_data = NULL;
1937 spin_unlock(&_minor_lock);
1939 put_disk(md->disk);
1940 blk_cleanup_queue(md->queue);
1941 module_put(THIS_MODULE);
1942 kfree(md);
1945 static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1947 struct dm_md_mempools *p;
1949 if (md->io_pool && md->tio_pool && md->bs)
1950 /* the md already has necessary mempools */
1951 goto out;
1953 p = dm_table_get_md_mempools(t);
1954 BUG_ON(!p || md->io_pool || md->tio_pool || md->bs);
1956 md->io_pool = p->io_pool;
1957 p->io_pool = NULL;
1958 md->tio_pool = p->tio_pool;
1959 p->tio_pool = NULL;
1960 md->bs = p->bs;
1961 p->bs = NULL;
1963 out:
1964 /* mempool bind completed, now no need any mempools in the table */
1965 dm_table_free_md_mempools(t);
1969 * Bind a table to the device.
1971 static void event_callback(void *context)
1973 unsigned long flags;
1974 LIST_HEAD(uevents);
1975 struct mapped_device *md = (struct mapped_device *) context;
1977 spin_lock_irqsave(&md->uevent_lock, flags);
1978 list_splice_init(&md->uevent_list, &uevents);
1979 spin_unlock_irqrestore(&md->uevent_lock, flags);
1981 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1983 atomic_inc(&md->event_nr);
1984 wake_up(&md->eventq);
1988 * Protected by md->suspend_lock obtained by dm_swap_table().
1990 static void __set_size(struct mapped_device *md, sector_t size)
1992 set_capacity(md->disk, size);
1994 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
1998 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2000 * If this function returns 0, then the device is either a non-dm
2001 * device without a merge_bvec_fn, or it is a dm device that is
2002 * able to split any bios it receives that are too big.
2004 int dm_queue_merge_is_compulsory(struct request_queue *q)
2006 struct mapped_device *dev_md;
2008 if (!q->merge_bvec_fn)
2009 return 0;
2011 if (q->make_request_fn == dm_request) {
2012 dev_md = q->queuedata;
2013 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2014 return 0;
2017 return 1;
2020 static int dm_device_merge_is_compulsory(struct dm_target *ti,
2021 struct dm_dev *dev, sector_t start,
2022 sector_t len, void *data)
2024 struct block_device *bdev = dev->bdev;
2025 struct request_queue *q = bdev_get_queue(bdev);
2027 return dm_queue_merge_is_compulsory(q);
2031 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2032 * on the properties of the underlying devices.
2034 static int dm_table_merge_is_optional(struct dm_table *table)
2036 unsigned i = 0;
2037 struct dm_target *ti;
2039 while (i < dm_table_get_num_targets(table)) {
2040 ti = dm_table_get_target(table, i++);
2042 if (ti->type->iterate_devices &&
2043 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2044 return 0;
2047 return 1;
2051 * Returns old map, which caller must destroy.
2053 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2054 struct queue_limits *limits)
2056 struct dm_table *old_map;
2057 struct request_queue *q = md->queue;
2058 sector_t size;
2059 unsigned long flags;
2060 int merge_is_optional;
2062 size = dm_table_get_size(t);
2065 * Wipe any geometry if the size of the table changed.
2067 if (size != get_capacity(md->disk))
2068 memset(&md->geometry, 0, sizeof(md->geometry));
2070 __set_size(md, size);
2072 dm_table_event_callback(t, event_callback, md);
2075 * The queue hasn't been stopped yet, if the old table type wasn't
2076 * for request-based during suspension. So stop it to prevent
2077 * I/O mapping before resume.
2078 * This must be done before setting the queue restrictions,
2079 * because request-based dm may be run just after the setting.
2081 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2082 stop_queue(q);
2084 __bind_mempools(md, t);
2086 merge_is_optional = dm_table_merge_is_optional(t);
2088 write_lock_irqsave(&md->map_lock, flags);
2089 old_map = md->map;
2090 md->map = t;
2091 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2093 dm_table_set_restrictions(t, q, limits);
2094 if (merge_is_optional)
2095 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2096 else
2097 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2098 write_unlock_irqrestore(&md->map_lock, flags);
2100 return old_map;
2104 * Returns unbound table for the caller to free.
2106 static struct dm_table *__unbind(struct mapped_device *md)
2108 struct dm_table *map = md->map;
2109 unsigned long flags;
2111 if (!map)
2112 return NULL;
2114 dm_table_event_callback(map, NULL, NULL);
2115 write_lock_irqsave(&md->map_lock, flags);
2116 md->map = NULL;
2117 write_unlock_irqrestore(&md->map_lock, flags);
2119 return map;
2123 * Constructor for a new device.
2125 int dm_create(int minor, struct mapped_device **result)
2127 struct mapped_device *md;
2129 md = alloc_dev(minor);
2130 if (!md)
2131 return -ENXIO;
2133 dm_sysfs_init(md);
2135 *result = md;
2136 return 0;
2140 * Functions to manage md->type.
2141 * All are required to hold md->type_lock.
2143 void dm_lock_md_type(struct mapped_device *md)
2145 mutex_lock(&md->type_lock);
2148 void dm_unlock_md_type(struct mapped_device *md)
2150 mutex_unlock(&md->type_lock);
2153 void dm_set_md_type(struct mapped_device *md, unsigned type)
2155 md->type = type;
2158 unsigned dm_get_md_type(struct mapped_device *md)
2160 return md->type;
2163 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2165 return md->immutable_target_type;
2169 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2171 static int dm_init_request_based_queue(struct mapped_device *md)
2173 struct request_queue *q = NULL;
2175 if (md->queue->elevator)
2176 return 1;
2178 /* Fully initialize the queue */
2179 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2180 if (!q)
2181 return 0;
2183 md->queue = q;
2184 dm_init_md_queue(md);
2185 blk_queue_softirq_done(md->queue, dm_softirq_done);
2186 blk_queue_prep_rq(md->queue, dm_prep_fn);
2187 blk_queue_lld_busy(md->queue, dm_lld_busy);
2189 elv_register_queue(md->queue);
2191 return 1;
2195 * Setup the DM device's queue based on md's type
2197 int dm_setup_md_queue(struct mapped_device *md)
2199 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2200 !dm_init_request_based_queue(md)) {
2201 DMWARN("Cannot initialize queue for request-based mapped device");
2202 return -EINVAL;
2205 return 0;
2208 static struct mapped_device *dm_find_md(dev_t dev)
2210 struct mapped_device *md;
2211 unsigned minor = MINOR(dev);
2213 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2214 return NULL;
2216 spin_lock(&_minor_lock);
2218 md = idr_find(&_minor_idr, minor);
2219 if (md && (md == MINOR_ALLOCED ||
2220 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2221 dm_deleting_md(md) ||
2222 test_bit(DMF_FREEING, &md->flags))) {
2223 md = NULL;
2224 goto out;
2227 out:
2228 spin_unlock(&_minor_lock);
2230 return md;
2233 struct mapped_device *dm_get_md(dev_t dev)
2235 struct mapped_device *md = dm_find_md(dev);
2237 if (md)
2238 dm_get(md);
2240 return md;
2242 EXPORT_SYMBOL_GPL(dm_get_md);
2244 void *dm_get_mdptr(struct mapped_device *md)
2246 return md->interface_ptr;
2249 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2251 md->interface_ptr = ptr;
2254 void dm_get(struct mapped_device *md)
2256 atomic_inc(&md->holders);
2257 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2260 const char *dm_device_name(struct mapped_device *md)
2262 return md->name;
2264 EXPORT_SYMBOL_GPL(dm_device_name);
2266 static void __dm_destroy(struct mapped_device *md, bool wait)
2268 struct dm_table *map;
2270 might_sleep();
2272 spin_lock(&_minor_lock);
2273 map = dm_get_live_table(md);
2274 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2275 set_bit(DMF_FREEING, &md->flags);
2276 spin_unlock(&_minor_lock);
2278 if (!dm_suspended_md(md)) {
2279 dm_table_presuspend_targets(map);
2280 dm_table_postsuspend_targets(map);
2284 * Rare, but there may be I/O requests still going to complete,
2285 * for example. Wait for all references to disappear.
2286 * No one should increment the reference count of the mapped_device,
2287 * after the mapped_device state becomes DMF_FREEING.
2289 if (wait)
2290 while (atomic_read(&md->holders))
2291 msleep(1);
2292 else if (atomic_read(&md->holders))
2293 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2294 dm_device_name(md), atomic_read(&md->holders));
2296 dm_sysfs_exit(md);
2297 dm_table_put(map);
2298 dm_table_destroy(__unbind(md));
2299 free_dev(md);
2302 void dm_destroy(struct mapped_device *md)
2304 __dm_destroy(md, true);
2307 void dm_destroy_immediate(struct mapped_device *md)
2309 __dm_destroy(md, false);
2312 void dm_put(struct mapped_device *md)
2314 atomic_dec(&md->holders);
2316 EXPORT_SYMBOL_GPL(dm_put);
2318 static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2320 int r = 0;
2321 DECLARE_WAITQUEUE(wait, current);
2323 add_wait_queue(&md->wait, &wait);
2325 while (1) {
2326 set_current_state(interruptible);
2328 if (!md_in_flight(md))
2329 break;
2331 if (interruptible == TASK_INTERRUPTIBLE &&
2332 signal_pending(current)) {
2333 r = -EINTR;
2334 break;
2337 io_schedule();
2339 set_current_state(TASK_RUNNING);
2341 remove_wait_queue(&md->wait, &wait);
2343 return r;
2347 * Process the deferred bios
2349 static void dm_wq_work(struct work_struct *work)
2351 struct mapped_device *md = container_of(work, struct mapped_device,
2352 work);
2353 struct bio *c;
2355 down_read(&md->io_lock);
2357 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2358 spin_lock_irq(&md->deferred_lock);
2359 c = bio_list_pop(&md->deferred);
2360 spin_unlock_irq(&md->deferred_lock);
2362 if (!c)
2363 break;
2365 up_read(&md->io_lock);
2367 if (dm_request_based(md))
2368 generic_make_request(c);
2369 else
2370 __split_and_process_bio(md, c);
2372 down_read(&md->io_lock);
2375 up_read(&md->io_lock);
2378 static void dm_queue_flush(struct mapped_device *md)
2380 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2381 smp_mb__after_clear_bit();
2382 queue_work(md->wq, &md->work);
2386 * Swap in a new table, returning the old one for the caller to destroy.
2388 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2390 struct dm_table *map = ERR_PTR(-EINVAL);
2391 struct queue_limits limits;
2392 int r;
2394 mutex_lock(&md->suspend_lock);
2396 /* device must be suspended */
2397 if (!dm_suspended_md(md))
2398 goto out;
2400 r = dm_calculate_queue_limits(table, &limits);
2401 if (r) {
2402 map = ERR_PTR(r);
2403 goto out;
2406 map = __bind(md, table, &limits);
2408 out:
2409 mutex_unlock(&md->suspend_lock);
2410 return map;
2414 * Functions to lock and unlock any filesystem running on the
2415 * device.
2417 static int lock_fs(struct mapped_device *md)
2419 int r;
2421 WARN_ON(md->frozen_sb);
2423 md->frozen_sb = freeze_bdev(md->bdev);
2424 if (IS_ERR(md->frozen_sb)) {
2425 r = PTR_ERR(md->frozen_sb);
2426 md->frozen_sb = NULL;
2427 return r;
2430 set_bit(DMF_FROZEN, &md->flags);
2432 return 0;
2435 static void unlock_fs(struct mapped_device *md)
2437 if (!test_bit(DMF_FROZEN, &md->flags))
2438 return;
2440 thaw_bdev(md->bdev, md->frozen_sb);
2441 md->frozen_sb = NULL;
2442 clear_bit(DMF_FROZEN, &md->flags);
2446 * We need to be able to change a mapping table under a mounted
2447 * filesystem. For example we might want to move some data in
2448 * the background. Before the table can be swapped with
2449 * dm_bind_table, dm_suspend must be called to flush any in
2450 * flight bios and ensure that any further io gets deferred.
2453 * Suspend mechanism in request-based dm.
2455 * 1. Flush all I/Os by lock_fs() if needed.
2456 * 2. Stop dispatching any I/O by stopping the request_queue.
2457 * 3. Wait for all in-flight I/Os to be completed or requeued.
2459 * To abort suspend, start the request_queue.
2461 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2463 struct dm_table *map = NULL;
2464 int r = 0;
2465 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2466 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2468 mutex_lock(&md->suspend_lock);
2470 if (dm_suspended_md(md)) {
2471 r = -EINVAL;
2472 goto out_unlock;
2475 map = dm_get_live_table(md);
2478 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2479 * This flag is cleared before dm_suspend returns.
2481 if (noflush)
2482 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2484 /* This does not get reverted if there's an error later. */
2485 dm_table_presuspend_targets(map);
2488 * Flush I/O to the device.
2489 * Any I/O submitted after lock_fs() may not be flushed.
2490 * noflush takes precedence over do_lockfs.
2491 * (lock_fs() flushes I/Os and waits for them to complete.)
2493 if (!noflush && do_lockfs) {
2494 r = lock_fs(md);
2495 if (r)
2496 goto out;
2500 * Here we must make sure that no processes are submitting requests
2501 * to target drivers i.e. no one may be executing
2502 * __split_and_process_bio. This is called from dm_request and
2503 * dm_wq_work.
2505 * To get all processes out of __split_and_process_bio in dm_request,
2506 * we take the write lock. To prevent any process from reentering
2507 * __split_and_process_bio from dm_request and quiesce the thread
2508 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2509 * flush_workqueue(md->wq).
2511 down_write(&md->io_lock);
2512 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2513 up_write(&md->io_lock);
2516 * Stop md->queue before flushing md->wq in case request-based
2517 * dm defers requests to md->wq from md->queue.
2519 if (dm_request_based(md))
2520 stop_queue(md->queue);
2522 flush_workqueue(md->wq);
2525 * At this point no more requests are entering target request routines.
2526 * We call dm_wait_for_completion to wait for all existing requests
2527 * to finish.
2529 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2531 down_write(&md->io_lock);
2532 if (noflush)
2533 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2534 up_write(&md->io_lock);
2536 /* were we interrupted ? */
2537 if (r < 0) {
2538 dm_queue_flush(md);
2540 if (dm_request_based(md))
2541 start_queue(md->queue);
2543 unlock_fs(md);
2544 goto out; /* pushback list is already flushed, so skip flush */
2548 * If dm_wait_for_completion returned 0, the device is completely
2549 * quiescent now. There is no request-processing activity. All new
2550 * requests are being added to md->deferred list.
2553 set_bit(DMF_SUSPENDED, &md->flags);
2555 dm_table_postsuspend_targets(map);
2557 out:
2558 dm_table_put(map);
2560 out_unlock:
2561 mutex_unlock(&md->suspend_lock);
2562 return r;
2565 int dm_resume(struct mapped_device *md)
2567 int r = -EINVAL;
2568 struct dm_table *map = NULL;
2570 mutex_lock(&md->suspend_lock);
2571 if (!dm_suspended_md(md))
2572 goto out;
2574 map = dm_get_live_table(md);
2575 if (!map || !dm_table_get_size(map))
2576 goto out;
2578 r = dm_table_resume_targets(map);
2579 if (r)
2580 goto out;
2582 dm_queue_flush(md);
2585 * Flushing deferred I/Os must be done after targets are resumed
2586 * so that mapping of targets can work correctly.
2587 * Request-based dm is queueing the deferred I/Os in its request_queue.
2589 if (dm_request_based(md))
2590 start_queue(md->queue);
2592 unlock_fs(md);
2594 clear_bit(DMF_SUSPENDED, &md->flags);
2596 r = 0;
2597 out:
2598 dm_table_put(map);
2599 mutex_unlock(&md->suspend_lock);
2601 return r;
2604 /*-----------------------------------------------------------------
2605 * Event notification.
2606 *---------------------------------------------------------------*/
2607 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2608 unsigned cookie)
2610 char udev_cookie[DM_COOKIE_LENGTH];
2611 char *envp[] = { udev_cookie, NULL };
2613 if (!cookie)
2614 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2615 else {
2616 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2617 DM_COOKIE_ENV_VAR_NAME, cookie);
2618 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2619 action, envp);
2623 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2625 return atomic_add_return(1, &md->uevent_seq);
2628 uint32_t dm_get_event_nr(struct mapped_device *md)
2630 return atomic_read(&md->event_nr);
2633 int dm_wait_event(struct mapped_device *md, int event_nr)
2635 return wait_event_interruptible(md->eventq,
2636 (event_nr != atomic_read(&md->event_nr)));
2639 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2641 unsigned long flags;
2643 spin_lock_irqsave(&md->uevent_lock, flags);
2644 list_add(elist, &md->uevent_list);
2645 spin_unlock_irqrestore(&md->uevent_lock, flags);
2649 * The gendisk is only valid as long as you have a reference
2650 * count on 'md'.
2652 struct gendisk *dm_disk(struct mapped_device *md)
2654 return md->disk;
2657 struct kobject *dm_kobject(struct mapped_device *md)
2659 return &md->kobj;
2663 * struct mapped_device should not be exported outside of dm.c
2664 * so use this check to verify that kobj is part of md structure
2666 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2668 struct mapped_device *md;
2670 md = container_of(kobj, struct mapped_device, kobj);
2671 if (&md->kobj != kobj)
2672 return NULL;
2674 if (test_bit(DMF_FREEING, &md->flags) ||
2675 dm_deleting_md(md))
2676 return NULL;
2678 dm_get(md);
2679 return md;
2682 int dm_suspended_md(struct mapped_device *md)
2684 return test_bit(DMF_SUSPENDED, &md->flags);
2687 int dm_suspended(struct dm_target *ti)
2689 return dm_suspended_md(dm_table_get_md(ti->table));
2691 EXPORT_SYMBOL_GPL(dm_suspended);
2693 int dm_noflush_suspending(struct dm_target *ti)
2695 return __noflush_suspending(dm_table_get_md(ti->table));
2697 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2699 struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity)
2701 struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL);
2702 unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS;
2704 if (!pools)
2705 return NULL;
2707 pools->io_pool = (type == DM_TYPE_BIO_BASED) ?
2708 mempool_create_slab_pool(MIN_IOS, _io_cache) :
2709 mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache);
2710 if (!pools->io_pool)
2711 goto free_pools_and_out;
2713 pools->tio_pool = (type == DM_TYPE_BIO_BASED) ?
2714 mempool_create_slab_pool(MIN_IOS, _tio_cache) :
2715 mempool_create_slab_pool(MIN_IOS, _rq_tio_cache);
2716 if (!pools->tio_pool)
2717 goto free_io_pool_and_out;
2719 pools->bs = bioset_create(pool_size, 0);
2720 if (!pools->bs)
2721 goto free_tio_pool_and_out;
2723 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2724 goto free_bioset_and_out;
2726 return pools;
2728 free_bioset_and_out:
2729 bioset_free(pools->bs);
2731 free_tio_pool_and_out:
2732 mempool_destroy(pools->tio_pool);
2734 free_io_pool_and_out:
2735 mempool_destroy(pools->io_pool);
2737 free_pools_and_out:
2738 kfree(pools);
2740 return NULL;
2743 void dm_free_md_mempools(struct dm_md_mempools *pools)
2745 if (!pools)
2746 return;
2748 if (pools->io_pool)
2749 mempool_destroy(pools->io_pool);
2751 if (pools->tio_pool)
2752 mempool_destroy(pools->tio_pool);
2754 if (pools->bs)
2755 bioset_free(pools->bs);
2757 kfree(pools);
2760 static const struct block_device_operations dm_blk_dops = {
2761 .open = dm_blk_open,
2762 .release = dm_blk_close,
2763 .ioctl = dm_blk_ioctl,
2764 .getgeo = dm_blk_getgeo,
2765 .owner = THIS_MODULE
2768 EXPORT_SYMBOL(dm_get_mapinfo);
2771 * module hooks
2773 module_init(dm_init);
2774 module_exit(dm_exit);
2776 module_param(major, uint, 0);
2777 MODULE_PARM_DESC(major, "The major number of the device mapper");
2778 MODULE_DESCRIPTION(DM_NAME " driver");
2779 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2780 MODULE_LICENSE("GPL");