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/mempool.h>
18 #include <linux/slab.h>
19 #include <linux/idr.h>
20 #include <linux/hdreg.h>
21 #include <linux/delay.h>
22 #include <linux/wait.h>
23 #include <linux/kthread.h>
24 #include <linux/ktime.h>
25 #include <linux/elevator.h> /* for rq_end_sector() */
26 #include <linux/blk-mq.h>
29 #include <trace/events/block.h>
31 #define DM_MSG_PREFIX "core"
35 * ratelimit state to be used in DMXXX_LIMIT().
37 DEFINE_RATELIMIT_STATE(dm_ratelimit_state
,
38 DEFAULT_RATELIMIT_INTERVAL
,
39 DEFAULT_RATELIMIT_BURST
);
40 EXPORT_SYMBOL(dm_ratelimit_state
);
44 * Cookies are numeric values sent with CHANGE and REMOVE
45 * uevents while resuming, removing or renaming the device.
47 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
48 #define DM_COOKIE_LENGTH 24
50 static const char *_name
= DM_NAME
;
52 static unsigned int major
= 0;
53 static unsigned int _major
= 0;
55 static DEFINE_IDR(_minor_idr
);
57 static DEFINE_SPINLOCK(_minor_lock
);
59 static void do_deferred_remove(struct work_struct
*w
);
61 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
63 static struct workqueue_struct
*deferred_remove_workqueue
;
67 * One of these is allocated per bio.
70 struct mapped_device
*md
;
74 unsigned long start_time
;
75 spinlock_t endio_lock
;
76 struct dm_stats_aux stats_aux
;
80 * For request-based dm.
81 * One of these is allocated per request.
83 struct dm_rq_target_io
{
84 struct mapped_device
*md
;
86 struct request
*orig
, *clone
;
87 struct kthread_work work
;
90 struct dm_stats_aux stats_aux
;
91 unsigned long duration_jiffies
;
96 * For request-based dm - the bio clones we allocate are embedded in these
99 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
100 * the bioset is created - this means the bio has to come at the end of the
103 struct dm_rq_clone_bio_info
{
105 struct dm_rq_target_io
*tio
;
109 #define MINOR_ALLOCED ((void *)-1)
112 * Bits for the md->flags field.
114 #define DMF_BLOCK_IO_FOR_SUSPEND 0
115 #define DMF_SUSPENDED 1
117 #define DMF_FREEING 3
118 #define DMF_DELETING 4
119 #define DMF_NOFLUSH_SUSPENDING 5
120 #define DMF_DEFERRED_REMOVE 6
121 #define DMF_SUSPENDED_INTERNALLY 7
124 * Work processed by per-device workqueue.
126 struct mapped_device
{
127 struct srcu_struct io_barrier
;
128 struct mutex suspend_lock
;
131 * The current mapping (struct dm_table *).
132 * Use dm_get_live_table{_fast} or take suspend_lock for
137 struct list_head table_devices
;
138 struct mutex table_devices_lock
;
142 struct request_queue
*queue
;
146 /* Protect queue and type against concurrent access. */
147 struct mutex type_lock
;
152 struct dm_target
*immutable_target
;
153 struct target_type
*immutable_target_type
;
155 struct gendisk
*disk
;
161 * A list of ios that arrived while we were suspended.
164 wait_queue_head_t wait
;
165 struct work_struct work
;
166 spinlock_t deferred_lock
;
167 struct bio_list deferred
;
172 wait_queue_head_t eventq
;
175 struct list_head uevent_list
;
176 spinlock_t uevent_lock
; /* Protect access to uevent_list */
178 /* the number of internal suspends */
179 unsigned internal_suspend_count
;
182 * Processing queue (flush)
184 struct workqueue_struct
*wq
;
187 * io objects are allocated from here.
195 * freeze/thaw support require holding onto a super block
197 struct super_block
*frozen_sb
;
199 /* forced geometry settings */
200 struct hd_geometry geometry
;
202 struct block_device
*bdev
;
204 /* kobject and completion */
205 struct dm_kobject_holder kobj_holder
;
207 /* zero-length flush that will be cloned and submitted to targets */
208 struct bio flush_bio
;
210 struct dm_stats stats
;
212 struct kthread_worker kworker
;
213 struct task_struct
*kworker_task
;
215 /* for request-based merge heuristic in dm_request_fn() */
216 unsigned seq_rq_merge_deadline_usecs
;
218 sector_t last_rq_pos
;
219 ktime_t last_rq_start_time
;
221 /* for blk-mq request-based DM support */
222 struct blk_mq_tag_set
*tag_set
;
227 #ifdef CONFIG_DM_MQ_DEFAULT
228 static bool use_blk_mq
= true;
230 static bool use_blk_mq
= false;
233 #define DM_MQ_NR_HW_QUEUES 1
234 #define DM_MQ_QUEUE_DEPTH 2048
235 #define DM_NUMA_NODE NUMA_NO_NODE
237 static unsigned dm_mq_nr_hw_queues
= DM_MQ_NR_HW_QUEUES
;
238 static unsigned dm_mq_queue_depth
= DM_MQ_QUEUE_DEPTH
;
239 static int dm_numa_node
= DM_NUMA_NODE
;
241 bool dm_use_blk_mq(struct mapped_device
*md
)
243 return md
->use_blk_mq
;
245 EXPORT_SYMBOL_GPL(dm_use_blk_mq
);
248 * For mempools pre-allocation at the table loading time.
250 struct dm_md_mempools
{
256 struct table_device
{
257 struct list_head list
;
259 struct dm_dev dm_dev
;
262 #define RESERVED_BIO_BASED_IOS 16
263 #define RESERVED_REQUEST_BASED_IOS 256
264 #define RESERVED_MAX_IOS 1024
265 static struct kmem_cache
*_io_cache
;
266 static struct kmem_cache
*_rq_tio_cache
;
267 static struct kmem_cache
*_rq_cache
;
270 * Bio-based DM's mempools' reserved IOs set by the user.
272 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
275 * Request-based DM's mempools' reserved IOs set by the user.
277 static unsigned reserved_rq_based_ios
= RESERVED_REQUEST_BASED_IOS
;
279 static int __dm_get_module_param_int(int *module_param
, int min
, int max
)
281 int param
= ACCESS_ONCE(*module_param
);
282 int modified_param
= 0;
283 bool modified
= true;
286 modified_param
= min
;
287 else if (param
> max
)
288 modified_param
= max
;
293 (void)cmpxchg(module_param
, param
, modified_param
);
294 param
= modified_param
;
300 static unsigned __dm_get_module_param(unsigned *module_param
,
301 unsigned def
, unsigned max
)
303 unsigned param
= ACCESS_ONCE(*module_param
);
304 unsigned modified_param
= 0;
307 modified_param
= def
;
308 else if (param
> max
)
309 modified_param
= max
;
311 if (modified_param
) {
312 (void)cmpxchg(module_param
, param
, modified_param
);
313 param
= modified_param
;
319 unsigned dm_get_reserved_bio_based_ios(void)
321 return __dm_get_module_param(&reserved_bio_based_ios
,
322 RESERVED_BIO_BASED_IOS
, RESERVED_MAX_IOS
);
324 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
326 unsigned dm_get_reserved_rq_based_ios(void)
328 return __dm_get_module_param(&reserved_rq_based_ios
,
329 RESERVED_REQUEST_BASED_IOS
, RESERVED_MAX_IOS
);
331 EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios
);
333 static unsigned dm_get_blk_mq_nr_hw_queues(void)
335 return __dm_get_module_param(&dm_mq_nr_hw_queues
, 1, 32);
338 static unsigned dm_get_blk_mq_queue_depth(void)
340 return __dm_get_module_param(&dm_mq_queue_depth
,
341 DM_MQ_QUEUE_DEPTH
, BLK_MQ_MAX_DEPTH
);
344 static unsigned dm_get_numa_node(void)
346 return __dm_get_module_param_int(&dm_numa_node
,
347 DM_NUMA_NODE
, num_online_nodes() - 1);
350 static int __init
local_init(void)
354 /* allocate a slab for the dm_ios */
355 _io_cache
= KMEM_CACHE(dm_io
, 0);
359 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
361 goto out_free_io_cache
;
363 _rq_cache
= kmem_cache_create("dm_old_clone_request", sizeof(struct request
),
364 __alignof__(struct request
), 0, NULL
);
366 goto out_free_rq_tio_cache
;
368 r
= dm_uevent_init();
370 goto out_free_rq_cache
;
372 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
373 if (!deferred_remove_workqueue
) {
375 goto out_uevent_exit
;
379 r
= register_blkdev(_major
, _name
);
381 goto out_free_workqueue
;
389 destroy_workqueue(deferred_remove_workqueue
);
393 kmem_cache_destroy(_rq_cache
);
394 out_free_rq_tio_cache
:
395 kmem_cache_destroy(_rq_tio_cache
);
397 kmem_cache_destroy(_io_cache
);
402 static void local_exit(void)
404 flush_scheduled_work();
405 destroy_workqueue(deferred_remove_workqueue
);
407 kmem_cache_destroy(_rq_cache
);
408 kmem_cache_destroy(_rq_tio_cache
);
409 kmem_cache_destroy(_io_cache
);
410 unregister_blkdev(_major
, _name
);
415 DMINFO("cleaned up");
418 static int (*_inits
[])(void) __initdata
= {
429 static void (*_exits
[])(void) = {
440 static int __init
dm_init(void)
442 const int count
= ARRAY_SIZE(_inits
);
446 for (i
= 0; i
< count
; i
++) {
461 static void __exit
dm_exit(void)
463 int i
= ARRAY_SIZE(_exits
);
469 * Should be empty by this point.
471 idr_destroy(&_minor_idr
);
475 * Block device functions
477 int dm_deleting_md(struct mapped_device
*md
)
479 return test_bit(DMF_DELETING
, &md
->flags
);
482 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
484 struct mapped_device
*md
;
486 spin_lock(&_minor_lock
);
488 md
= bdev
->bd_disk
->private_data
;
492 if (test_bit(DMF_FREEING
, &md
->flags
) ||
493 dm_deleting_md(md
)) {
499 atomic_inc(&md
->open_count
);
501 spin_unlock(&_minor_lock
);
503 return md
? 0 : -ENXIO
;
506 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
508 struct mapped_device
*md
;
510 spin_lock(&_minor_lock
);
512 md
= disk
->private_data
;
516 if (atomic_dec_and_test(&md
->open_count
) &&
517 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
518 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
522 spin_unlock(&_minor_lock
);
525 int dm_open_count(struct mapped_device
*md
)
527 return atomic_read(&md
->open_count
);
531 * Guarantees nothing is using the device before it's deleted.
533 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
537 spin_lock(&_minor_lock
);
539 if (dm_open_count(md
)) {
542 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
543 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
546 set_bit(DMF_DELETING
, &md
->flags
);
548 spin_unlock(&_minor_lock
);
553 int dm_cancel_deferred_remove(struct mapped_device
*md
)
557 spin_lock(&_minor_lock
);
559 if (test_bit(DMF_DELETING
, &md
->flags
))
562 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
564 spin_unlock(&_minor_lock
);
569 static void do_deferred_remove(struct work_struct
*w
)
571 dm_deferred_remove();
574 sector_t
dm_get_size(struct mapped_device
*md
)
576 return get_capacity(md
->disk
);
579 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
584 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
589 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
591 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
593 return dm_get_geometry(md
, geo
);
596 static int dm_grab_bdev_for_ioctl(struct mapped_device
*md
,
597 struct block_device
**bdev
,
600 struct dm_target
*tgt
;
601 struct dm_table
*map
;
606 map
= dm_get_live_table(md
, &srcu_idx
);
607 if (!map
|| !dm_table_get_size(map
))
610 /* We only support devices that have a single target */
611 if (dm_table_get_num_targets(map
) != 1)
614 tgt
= dm_table_get_target(map
, 0);
615 if (!tgt
->type
->prepare_ioctl
)
618 if (dm_suspended_md(md
)) {
623 r
= tgt
->type
->prepare_ioctl(tgt
, bdev
, mode
);
628 dm_put_live_table(md
, srcu_idx
);
632 dm_put_live_table(md
, srcu_idx
);
633 if (r
== -ENOTCONN
&& !fatal_signal_pending(current
)) {
640 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
641 unsigned int cmd
, unsigned long arg
)
643 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
646 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
652 * Target determined this ioctl is being issued against
653 * a logical partition of the parent bdev; so extra
654 * validation is needed.
656 r
= scsi_verify_blk_ioctl(NULL
, cmd
);
661 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
667 static struct dm_io
*alloc_io(struct mapped_device
*md
)
669 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
672 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
674 mempool_free(io
, md
->io_pool
);
677 static void free_tio(struct mapped_device
*md
, struct dm_target_io
*tio
)
679 bio_put(&tio
->clone
);
682 static struct dm_rq_target_io
*alloc_old_rq_tio(struct mapped_device
*md
,
685 return mempool_alloc(md
->io_pool
, gfp_mask
);
688 static void free_old_rq_tio(struct dm_rq_target_io
*tio
)
690 mempool_free(tio
, tio
->md
->io_pool
);
693 static struct request
*alloc_old_clone_request(struct mapped_device
*md
,
696 return mempool_alloc(md
->rq_pool
, gfp_mask
);
699 static void free_old_clone_request(struct mapped_device
*md
, struct request
*rq
)
701 mempool_free(rq
, md
->rq_pool
);
704 static int md_in_flight(struct mapped_device
*md
)
706 return atomic_read(&md
->pending
[READ
]) +
707 atomic_read(&md
->pending
[WRITE
]);
710 static void start_io_acct(struct dm_io
*io
)
712 struct mapped_device
*md
= io
->md
;
713 struct bio
*bio
= io
->bio
;
715 int rw
= bio_data_dir(bio
);
717 io
->start_time
= jiffies
;
719 cpu
= part_stat_lock();
720 part_round_stats(cpu
, &dm_disk(md
)->part0
);
722 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
723 atomic_inc_return(&md
->pending
[rw
]));
725 if (unlikely(dm_stats_used(&md
->stats
)))
726 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
727 bio_sectors(bio
), false, 0, &io
->stats_aux
);
730 static void end_io_acct(struct dm_io
*io
)
732 struct mapped_device
*md
= io
->md
;
733 struct bio
*bio
= io
->bio
;
734 unsigned long duration
= jiffies
- io
->start_time
;
736 int rw
= bio_data_dir(bio
);
738 generic_end_io_acct(rw
, &dm_disk(md
)->part0
, io
->start_time
);
740 if (unlikely(dm_stats_used(&md
->stats
)))
741 dm_stats_account_io(&md
->stats
, bio
->bi_rw
, bio
->bi_iter
.bi_sector
,
742 bio_sectors(bio
), true, duration
, &io
->stats_aux
);
745 * After this is decremented the bio must not be touched if it is
748 pending
= atomic_dec_return(&md
->pending
[rw
]);
749 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
750 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
752 /* nudge anyone waiting on suspend queue */
758 * Add the bio to the list of deferred io.
760 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
764 spin_lock_irqsave(&md
->deferred_lock
, flags
);
765 bio_list_add(&md
->deferred
, bio
);
766 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
767 queue_work(md
->wq
, &md
->work
);
771 * Everyone (including functions in this file), should use this
772 * function to access the md->map field, and make sure they call
773 * dm_put_live_table() when finished.
775 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
777 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
779 return srcu_dereference(md
->map
, &md
->io_barrier
);
782 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
784 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
787 void dm_sync_table(struct mapped_device
*md
)
789 synchronize_srcu(&md
->io_barrier
);
790 synchronize_rcu_expedited();
794 * A fast alternative to dm_get_live_table/dm_put_live_table.
795 * The caller must not block between these two functions.
797 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
800 return rcu_dereference(md
->map
);
803 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
809 * Open a table device so we can use it as a map destination.
811 static int open_table_device(struct table_device
*td
, dev_t dev
,
812 struct mapped_device
*md
)
814 static char *_claim_ptr
= "I belong to device-mapper";
815 struct block_device
*bdev
;
819 BUG_ON(td
->dm_dev
.bdev
);
821 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
823 return PTR_ERR(bdev
);
825 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
827 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
831 td
->dm_dev
.bdev
= bdev
;
836 * Close a table device that we've been using.
838 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
840 if (!td
->dm_dev
.bdev
)
843 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
844 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
845 td
->dm_dev
.bdev
= NULL
;
848 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
850 struct table_device
*td
;
852 list_for_each_entry(td
, l
, list
)
853 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
859 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
860 struct dm_dev
**result
) {
862 struct table_device
*td
;
864 mutex_lock(&md
->table_devices_lock
);
865 td
= find_table_device(&md
->table_devices
, dev
, mode
);
867 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
869 mutex_unlock(&md
->table_devices_lock
);
873 td
->dm_dev
.mode
= mode
;
874 td
->dm_dev
.bdev
= NULL
;
876 if ((r
= open_table_device(td
, dev
, md
))) {
877 mutex_unlock(&md
->table_devices_lock
);
882 format_dev_t(td
->dm_dev
.name
, dev
);
884 atomic_set(&td
->count
, 0);
885 list_add(&td
->list
, &md
->table_devices
);
887 atomic_inc(&td
->count
);
888 mutex_unlock(&md
->table_devices_lock
);
890 *result
= &td
->dm_dev
;
893 EXPORT_SYMBOL_GPL(dm_get_table_device
);
895 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
897 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
899 mutex_lock(&md
->table_devices_lock
);
900 if (atomic_dec_and_test(&td
->count
)) {
901 close_table_device(td
, md
);
905 mutex_unlock(&md
->table_devices_lock
);
907 EXPORT_SYMBOL(dm_put_table_device
);
909 static void free_table_devices(struct list_head
*devices
)
911 struct list_head
*tmp
, *next
;
913 list_for_each_safe(tmp
, next
, devices
) {
914 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
916 DMWARN("dm_destroy: %s still exists with %d references",
917 td
->dm_dev
.name
, atomic_read(&td
->count
));
923 * Get the geometry associated with a dm device
925 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
933 * Set the geometry of a device.
935 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
937 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
939 if (geo
->start
> sz
) {
940 DMWARN("Start sector is beyond the geometry limits.");
949 /*-----------------------------------------------------------------
951 * A more elegant soln is in the works that uses the queue
952 * merge fn, unfortunately there are a couple of changes to
953 * the block layer that I want to make for this. So in the
954 * interests of getting something for people to use I give
955 * you this clearly demarcated crap.
956 *---------------------------------------------------------------*/
958 static int __noflush_suspending(struct mapped_device
*md
)
960 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
964 * Decrements the number of outstanding ios that a bio has been
965 * cloned into, completing the original io if necc.
967 static void dec_pending(struct dm_io
*io
, int error
)
972 struct mapped_device
*md
= io
->md
;
974 /* Push-back supersedes any I/O errors */
975 if (unlikely(error
)) {
976 spin_lock_irqsave(&io
->endio_lock
, flags
);
977 if (!(io
->error
> 0 && __noflush_suspending(md
)))
979 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
982 if (atomic_dec_and_test(&io
->io_count
)) {
983 if (io
->error
== DM_ENDIO_REQUEUE
) {
985 * Target requested pushing back the I/O.
987 spin_lock_irqsave(&md
->deferred_lock
, flags
);
988 if (__noflush_suspending(md
))
989 bio_list_add_head(&md
->deferred
, io
->bio
);
991 /* noflush suspend was interrupted. */
993 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
996 io_error
= io
->error
;
1001 if (io_error
== DM_ENDIO_REQUEUE
)
1004 if ((bio
->bi_rw
& REQ_FLUSH
) && bio
->bi_iter
.bi_size
) {
1006 * Preflush done for flush with data, reissue
1007 * without REQ_FLUSH.
1009 bio
->bi_rw
&= ~REQ_FLUSH
;
1012 /* done with normal IO or empty flush */
1013 trace_block_bio_complete(md
->queue
, bio
, io_error
);
1014 bio
->bi_error
= io_error
;
1020 static void disable_write_same(struct mapped_device
*md
)
1022 struct queue_limits
*limits
= dm_get_queue_limits(md
);
1024 /* device doesn't really support WRITE SAME, disable it */
1025 limits
->max_write_same_sectors
= 0;
1028 static void clone_endio(struct bio
*bio
)
1030 int error
= bio
->bi_error
;
1032 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1033 struct dm_io
*io
= tio
->io
;
1034 struct mapped_device
*md
= tio
->io
->md
;
1035 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
1038 r
= endio(tio
->ti
, bio
, error
);
1039 if (r
< 0 || r
== DM_ENDIO_REQUEUE
)
1041 * error and requeue request are handled
1045 else if (r
== DM_ENDIO_INCOMPLETE
)
1046 /* The target will handle the io */
1049 DMWARN("unimplemented target endio return value: %d", r
);
1054 if (unlikely(r
== -EREMOTEIO
&& (bio
->bi_rw
& REQ_WRITE_SAME
) &&
1055 !bdev_get_queue(bio
->bi_bdev
)->limits
.max_write_same_sectors
))
1056 disable_write_same(md
);
1059 dec_pending(io
, error
);
1063 * Partial completion handling for request-based dm
1065 static void end_clone_bio(struct bio
*clone
)
1067 struct dm_rq_clone_bio_info
*info
=
1068 container_of(clone
, struct dm_rq_clone_bio_info
, clone
);
1069 struct dm_rq_target_io
*tio
= info
->tio
;
1070 struct bio
*bio
= info
->orig
;
1071 unsigned int nr_bytes
= info
->orig
->bi_iter
.bi_size
;
1072 int error
= clone
->bi_error
;
1078 * An error has already been detected on the request.
1079 * Once error occurred, just let clone->end_io() handle
1085 * Don't notice the error to the upper layer yet.
1086 * The error handling decision is made by the target driver,
1087 * when the request is completed.
1094 * I/O for the bio successfully completed.
1095 * Notice the data completion to the upper layer.
1099 * bios are processed from the head of the list.
1100 * So the completing bio should always be rq->bio.
1101 * If it's not, something wrong is happening.
1103 if (tio
->orig
->bio
!= bio
)
1104 DMERR("bio completion is going in the middle of the request");
1107 * Update the original request.
1108 * Do not use blk_end_request() here, because it may complete
1109 * the original request before the clone, and break the ordering.
1111 blk_update_request(tio
->orig
, 0, nr_bytes
);
1114 static struct dm_rq_target_io
*tio_from_request(struct request
*rq
)
1116 return (rq
->q
->mq_ops
? blk_mq_rq_to_pdu(rq
) : rq
->special
);
1119 static void rq_end_stats(struct mapped_device
*md
, struct request
*orig
)
1121 if (unlikely(dm_stats_used(&md
->stats
))) {
1122 struct dm_rq_target_io
*tio
= tio_from_request(orig
);
1123 tio
->duration_jiffies
= jiffies
- tio
->duration_jiffies
;
1124 dm_stats_account_io(&md
->stats
, orig
->cmd_flags
, blk_rq_pos(orig
),
1125 tio
->n_sectors
, true, tio
->duration_jiffies
,
1131 * Don't touch any member of the md after calling this function because
1132 * the md may be freed in dm_put() at the end of this function.
1133 * Or do dm_get() before calling this function and dm_put() later.
1135 static void rq_completed(struct mapped_device
*md
, int rw
, bool run_queue
)
1137 atomic_dec(&md
->pending
[rw
]);
1139 /* nudge anyone waiting on suspend queue */
1140 if (!md_in_flight(md
))
1144 * Run this off this callpath, as drivers could invoke end_io while
1145 * inside their request_fn (and holding the queue lock). Calling
1146 * back into ->request_fn() could deadlock attempting to grab the
1149 if (!md
->queue
->mq_ops
&& run_queue
)
1150 blk_run_queue_async(md
->queue
);
1153 * dm_put() must be at the end of this function. See the comment above
1158 static void free_rq_clone(struct request
*clone
)
1160 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1161 struct mapped_device
*md
= tio
->md
;
1163 blk_rq_unprep_clone(clone
);
1165 if (md
->type
== DM_TYPE_MQ_REQUEST_BASED
)
1166 /* stacked on blk-mq queue(s) */
1167 tio
->ti
->type
->release_clone_rq(clone
);
1168 else if (!md
->queue
->mq_ops
)
1169 /* request_fn queue stacked on request_fn queue(s) */
1170 free_old_clone_request(md
, clone
);
1172 if (!md
->queue
->mq_ops
)
1173 free_old_rq_tio(tio
);
1177 * Complete the clone and the original request.
1178 * Must be called without clone's queue lock held,
1179 * see end_clone_request() for more details.
1181 static void dm_end_request(struct request
*clone
, int error
)
1183 int rw
= rq_data_dir(clone
);
1184 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1185 struct mapped_device
*md
= tio
->md
;
1186 struct request
*rq
= tio
->orig
;
1188 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
1189 rq
->errors
= clone
->errors
;
1190 rq
->resid_len
= clone
->resid_len
;
1194 * We are using the sense buffer of the original
1196 * So setting the length of the sense data is enough.
1198 rq
->sense_len
= clone
->sense_len
;
1201 free_rq_clone(clone
);
1202 rq_end_stats(md
, rq
);
1204 blk_end_request_all(rq
, error
);
1206 blk_mq_end_request(rq
, error
);
1207 rq_completed(md
, rw
, true);
1210 static void dm_unprep_request(struct request
*rq
)
1212 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1213 struct request
*clone
= tio
->clone
;
1215 if (!rq
->q
->mq_ops
) {
1217 rq
->cmd_flags
&= ~REQ_DONTPREP
;
1221 free_rq_clone(clone
);
1222 else if (!tio
->md
->queue
->mq_ops
)
1223 free_old_rq_tio(tio
);
1227 * Requeue the original request of a clone.
1229 static void dm_old_requeue_request(struct request
*rq
)
1231 struct request_queue
*q
= rq
->q
;
1232 unsigned long flags
;
1234 spin_lock_irqsave(q
->queue_lock
, flags
);
1235 blk_requeue_request(q
, rq
);
1236 blk_run_queue_async(q
);
1237 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1240 static void dm_mq_requeue_request(struct request
*rq
)
1242 struct request_queue
*q
= rq
->q
;
1243 unsigned long flags
;
1245 blk_mq_requeue_request(rq
);
1246 spin_lock_irqsave(q
->queue_lock
, flags
);
1247 if (!blk_queue_stopped(q
))
1248 blk_mq_kick_requeue_list(q
);
1249 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1252 static void dm_requeue_original_request(struct mapped_device
*md
,
1255 int rw
= rq_data_dir(rq
);
1257 rq_end_stats(md
, rq
);
1258 dm_unprep_request(rq
);
1261 dm_old_requeue_request(rq
);
1263 dm_mq_requeue_request(rq
);
1265 rq_completed(md
, rw
, false);
1268 static void dm_old_stop_queue(struct request_queue
*q
)
1270 unsigned long flags
;
1272 spin_lock_irqsave(q
->queue_lock
, flags
);
1273 if (blk_queue_stopped(q
)) {
1274 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1279 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1282 static void dm_stop_queue(struct request_queue
*q
)
1285 dm_old_stop_queue(q
);
1287 blk_mq_stop_hw_queues(q
);
1290 static void dm_old_start_queue(struct request_queue
*q
)
1292 unsigned long flags
;
1294 spin_lock_irqsave(q
->queue_lock
, flags
);
1295 if (blk_queue_stopped(q
))
1297 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1300 static void dm_start_queue(struct request_queue
*q
)
1303 dm_old_start_queue(q
);
1305 blk_mq_start_stopped_hw_queues(q
, true);
1306 blk_mq_kick_requeue_list(q
);
1310 static void dm_done(struct request
*clone
, int error
, bool mapped
)
1313 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1314 dm_request_endio_fn rq_end_io
= NULL
;
1317 rq_end_io
= tio
->ti
->type
->rq_end_io
;
1319 if (mapped
&& rq_end_io
)
1320 r
= rq_end_io(tio
->ti
, clone
, error
, &tio
->info
);
1323 if (unlikely(r
== -EREMOTEIO
&& (clone
->cmd_flags
& REQ_WRITE_SAME
) &&
1324 !clone
->q
->limits
.max_write_same_sectors
))
1325 disable_write_same(tio
->md
);
1328 /* The target wants to complete the I/O */
1329 dm_end_request(clone
, r
);
1330 else if (r
== DM_ENDIO_INCOMPLETE
)
1331 /* The target will handle the I/O */
1333 else if (r
== DM_ENDIO_REQUEUE
)
1334 /* The target wants to requeue the I/O */
1335 dm_requeue_original_request(tio
->md
, tio
->orig
);
1337 DMWARN("unimplemented target endio return value: %d", r
);
1343 * Request completion handler for request-based dm
1345 static void dm_softirq_done(struct request
*rq
)
1348 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1349 struct request
*clone
= tio
->clone
;
1353 rq_end_stats(tio
->md
, rq
);
1354 rw
= rq_data_dir(rq
);
1355 if (!rq
->q
->mq_ops
) {
1356 blk_end_request_all(rq
, tio
->error
);
1357 rq_completed(tio
->md
, rw
, false);
1358 free_old_rq_tio(tio
);
1360 blk_mq_end_request(rq
, tio
->error
);
1361 rq_completed(tio
->md
, rw
, false);
1366 if (rq
->cmd_flags
& REQ_FAILED
)
1369 dm_done(clone
, tio
->error
, mapped
);
1373 * Complete the clone and the original request with the error status
1374 * through softirq context.
1376 static void dm_complete_request(struct request
*rq
, int error
)
1378 struct dm_rq_target_io
*tio
= tio_from_request(rq
);
1382 blk_complete_request(rq
);
1384 blk_mq_complete_request(rq
, error
);
1388 * Complete the not-mapped clone and the original request with the error status
1389 * through softirq context.
1390 * Target's rq_end_io() function isn't called.
1391 * This may be used when the target's map_rq() or clone_and_map_rq() functions fail.
1393 static void dm_kill_unmapped_request(struct request
*rq
, int error
)
1395 rq
->cmd_flags
|= REQ_FAILED
;
1396 dm_complete_request(rq
, error
);
1400 * Called with the clone's queue lock held (in the case of .request_fn)
1402 static void end_clone_request(struct request
*clone
, int error
)
1404 struct dm_rq_target_io
*tio
= clone
->end_io_data
;
1406 if (!clone
->q
->mq_ops
) {
1408 * For just cleaning up the information of the queue in which
1409 * the clone was dispatched.
1410 * The clone is *NOT* freed actually here because it is alloced
1411 * from dm own mempool (REQ_ALLOCED isn't set).
1413 __blk_put_request(clone
->q
, clone
);
1417 * Actual request completion is done in a softirq context which doesn't
1418 * hold the clone's queue lock. Otherwise, deadlock could occur because:
1419 * - another request may be submitted by the upper level driver
1420 * of the stacking during the completion
1421 * - the submission which requires queue lock may be done
1422 * against this clone's queue
1424 dm_complete_request(tio
->orig
, error
);
1428 * Return maximum size of I/O possible at the supplied sector up to the current
1431 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
1433 sector_t target_offset
= dm_target_offset(ti
, sector
);
1435 return ti
->len
- target_offset
;
1438 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
1440 sector_t len
= max_io_len_target_boundary(sector
, ti
);
1441 sector_t offset
, max_len
;
1444 * Does the target need to split even further?
1446 if (ti
->max_io_len
) {
1447 offset
= dm_target_offset(ti
, sector
);
1448 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
1449 max_len
= sector_div(offset
, ti
->max_io_len
);
1451 max_len
= offset
& (ti
->max_io_len
- 1);
1452 max_len
= ti
->max_io_len
- max_len
;
1461 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
1463 if (len
> UINT_MAX
) {
1464 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1465 (unsigned long long)len
, UINT_MAX
);
1466 ti
->error
= "Maximum size of target IO is too large";
1470 ti
->max_io_len
= (uint32_t) len
;
1474 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
1477 * A target may call dm_accept_partial_bio only from the map routine. It is
1478 * allowed for all bio types except REQ_FLUSH.
1480 * dm_accept_partial_bio informs the dm that the target only wants to process
1481 * additional n_sectors sectors of the bio and the rest of the data should be
1482 * sent in a next bio.
1484 * A diagram that explains the arithmetics:
1485 * +--------------------+---------------+-------+
1487 * +--------------------+---------------+-------+
1489 * <-------------- *tio->len_ptr --------------->
1490 * <------- bi_size ------->
1493 * Region 1 was already iterated over with bio_advance or similar function.
1494 * (it may be empty if the target doesn't use bio_advance)
1495 * Region 2 is the remaining bio size that the target wants to process.
1496 * (it may be empty if region 1 is non-empty, although there is no reason
1498 * The target requires that region 3 is to be sent in the next bio.
1500 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1501 * the partially processed part (the sum of regions 1+2) must be the same for all
1502 * copies of the bio.
1504 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1506 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1507 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1508 BUG_ON(bio
->bi_rw
& REQ_FLUSH
);
1509 BUG_ON(bi_size
> *tio
->len_ptr
);
1510 BUG_ON(n_sectors
> bi_size
);
1511 *tio
->len_ptr
-= bi_size
- n_sectors
;
1512 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1514 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1516 static void __map_bio(struct dm_target_io
*tio
)
1520 struct mapped_device
*md
;
1521 struct bio
*clone
= &tio
->clone
;
1522 struct dm_target
*ti
= tio
->ti
;
1524 clone
->bi_end_io
= clone_endio
;
1527 * Map the clone. If r == 0 we don't need to do
1528 * anything, the target has assumed ownership of
1531 atomic_inc(&tio
->io
->io_count
);
1532 sector
= clone
->bi_iter
.bi_sector
;
1533 r
= ti
->type
->map(ti
, clone
);
1534 if (r
== DM_MAPIO_REMAPPED
) {
1535 /* the bio has been remapped so dispatch it */
1537 trace_block_bio_remap(bdev_get_queue(clone
->bi_bdev
), clone
,
1538 tio
->io
->bio
->bi_bdev
->bd_dev
, sector
);
1540 generic_make_request(clone
);
1541 } else if (r
< 0 || r
== DM_MAPIO_REQUEUE
) {
1542 /* error the io and bail out, or requeue it if needed */
1544 dec_pending(tio
->io
, r
);
1546 } else if (r
!= DM_MAPIO_SUBMITTED
) {
1547 DMWARN("unimplemented target map return value: %d", r
);
1553 struct mapped_device
*md
;
1554 struct dm_table
*map
;
1558 unsigned sector_count
;
1561 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1563 bio
->bi_iter
.bi_sector
= sector
;
1564 bio
->bi_iter
.bi_size
= to_bytes(len
);
1568 * Creates a bio that consists of range of complete bvecs.
1570 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1571 sector_t sector
, unsigned len
)
1573 struct bio
*clone
= &tio
->clone
;
1575 __bio_clone_fast(clone
, bio
);
1577 if (bio_integrity(bio
)) {
1578 int r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1583 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1584 clone
->bi_iter
.bi_size
= to_bytes(len
);
1586 if (bio_integrity(bio
))
1587 bio_integrity_trim(clone
, 0, len
);
1592 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1593 struct dm_target
*ti
,
1594 unsigned target_bio_nr
)
1596 struct dm_target_io
*tio
;
1599 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1600 tio
= container_of(clone
, struct dm_target_io
, clone
);
1604 tio
->target_bio_nr
= target_bio_nr
;
1609 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1610 struct dm_target
*ti
,
1611 unsigned target_bio_nr
, unsigned *len
)
1613 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1614 struct bio
*clone
= &tio
->clone
;
1618 __bio_clone_fast(clone
, ci
->bio
);
1620 bio_setup_sector(clone
, ci
->sector
, *len
);
1625 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1626 unsigned num_bios
, unsigned *len
)
1628 unsigned target_bio_nr
;
1630 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1631 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1634 static int __send_empty_flush(struct clone_info
*ci
)
1636 unsigned target_nr
= 0;
1637 struct dm_target
*ti
;
1639 BUG_ON(bio_has_data(ci
->bio
));
1640 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1641 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1646 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1647 sector_t sector
, unsigned *len
)
1649 struct bio
*bio
= ci
->bio
;
1650 struct dm_target_io
*tio
;
1651 unsigned target_bio_nr
;
1652 unsigned num_target_bios
= 1;
1656 * Does the target want to receive duplicate copies of the bio?
1658 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1659 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1661 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1662 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1664 r
= clone_bio(tio
, bio
, sector
, *len
);
1666 free_tio(ci
->md
, tio
);
1675 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1677 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1679 return ti
->num_discard_bios
;
1682 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1684 return ti
->num_write_same_bios
;
1687 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1689 static bool is_split_required_for_discard(struct dm_target
*ti
)
1691 return ti
->split_discard_bios
;
1694 static int __send_changing_extent_only(struct clone_info
*ci
,
1695 get_num_bios_fn get_num_bios
,
1696 is_split_required_fn is_split_required
)
1698 struct dm_target
*ti
;
1703 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1704 if (!dm_target_is_valid(ti
))
1708 * Even though the device advertised support for this type of
1709 * request, that does not mean every target supports it, and
1710 * reconfiguration might also have changed that since the
1711 * check was performed.
1713 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1717 if (is_split_required
&& !is_split_required(ti
))
1718 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1720 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1722 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1725 } while (ci
->sector_count
-= len
);
1730 static int __send_discard(struct clone_info
*ci
)
1732 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1733 is_split_required_for_discard
);
1736 static int __send_write_same(struct clone_info
*ci
)
1738 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1742 * Select the correct strategy for processing a non-flush bio.
1744 static int __split_and_process_non_flush(struct clone_info
*ci
)
1746 struct bio
*bio
= ci
->bio
;
1747 struct dm_target
*ti
;
1751 if (unlikely(bio
->bi_rw
& REQ_DISCARD
))
1752 return __send_discard(ci
);
1753 else if (unlikely(bio
->bi_rw
& REQ_WRITE_SAME
))
1754 return __send_write_same(ci
);
1756 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1757 if (!dm_target_is_valid(ti
))
1760 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
), ci
->sector_count
);
1762 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1767 ci
->sector_count
-= len
;
1773 * Entry point to split a bio into clones and submit them to the targets.
1775 static void __split_and_process_bio(struct mapped_device
*md
,
1776 struct dm_table
*map
, struct bio
*bio
)
1778 struct clone_info ci
;
1781 if (unlikely(!map
)) {
1788 ci
.io
= alloc_io(md
);
1790 atomic_set(&ci
.io
->io_count
, 1);
1793 spin_lock_init(&ci
.io
->endio_lock
);
1794 ci
.sector
= bio
->bi_iter
.bi_sector
;
1796 start_io_acct(ci
.io
);
1798 if (bio
->bi_rw
& REQ_FLUSH
) {
1799 ci
.bio
= &ci
.md
->flush_bio
;
1800 ci
.sector_count
= 0;
1801 error
= __send_empty_flush(&ci
);
1802 /* dec_pending submits any data associated with flush */
1805 ci
.sector_count
= bio_sectors(bio
);
1806 while (ci
.sector_count
&& !error
)
1807 error
= __split_and_process_non_flush(&ci
);
1810 /* drop the extra reference count */
1811 dec_pending(ci
.io
, error
);
1813 /*-----------------------------------------------------------------
1815 *---------------------------------------------------------------*/
1818 * The request function that just remaps the bio built up by
1821 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1823 int rw
= bio_data_dir(bio
);
1824 struct mapped_device
*md
= q
->queuedata
;
1826 struct dm_table
*map
;
1828 map
= dm_get_live_table(md
, &srcu_idx
);
1830 generic_start_io_acct(rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1832 /* if we're suspended, we have to queue this io for later */
1833 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1834 dm_put_live_table(md
, srcu_idx
);
1836 if (bio_rw(bio
) != READA
)
1840 return BLK_QC_T_NONE
;
1843 __split_and_process_bio(md
, map
, bio
);
1844 dm_put_live_table(md
, srcu_idx
);
1845 return BLK_QC_T_NONE
;
1848 int dm_request_based(struct mapped_device
*md
)
1850 return blk_queue_stackable(md
->queue
);
1853 static void dm_dispatch_clone_request(struct request
*clone
, struct request
*rq
)
1857 if (blk_queue_io_stat(clone
->q
))
1858 clone
->cmd_flags
|= REQ_IO_STAT
;
1860 clone
->start_time
= jiffies
;
1861 r
= blk_insert_cloned_request(clone
->q
, clone
);
1863 /* must complete clone in terms of original request */
1864 dm_complete_request(rq
, r
);
1867 static int dm_rq_bio_constructor(struct bio
*bio
, struct bio
*bio_orig
,
1870 struct dm_rq_target_io
*tio
= data
;
1871 struct dm_rq_clone_bio_info
*info
=
1872 container_of(bio
, struct dm_rq_clone_bio_info
, clone
);
1874 info
->orig
= bio_orig
;
1876 bio
->bi_end_io
= end_clone_bio
;
1881 static int setup_clone(struct request
*clone
, struct request
*rq
,
1882 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1886 r
= blk_rq_prep_clone(clone
, rq
, tio
->md
->bs
, gfp_mask
,
1887 dm_rq_bio_constructor
, tio
);
1891 clone
->cmd
= rq
->cmd
;
1892 clone
->cmd_len
= rq
->cmd_len
;
1893 clone
->sense
= rq
->sense
;
1894 clone
->end_io
= end_clone_request
;
1895 clone
->end_io_data
= tio
;
1902 static struct request
*clone_old_rq(struct request
*rq
, struct mapped_device
*md
,
1903 struct dm_rq_target_io
*tio
, gfp_t gfp_mask
)
1906 * Create clone for use with .request_fn request_queue
1908 struct request
*clone
;
1910 clone
= alloc_old_clone_request(md
, gfp_mask
);
1914 blk_rq_init(NULL
, clone
);
1915 if (setup_clone(clone
, rq
, tio
, gfp_mask
)) {
1917 free_old_clone_request(md
, clone
);
1924 static void map_tio_request(struct kthread_work
*work
);
1926 static void init_tio(struct dm_rq_target_io
*tio
, struct request
*rq
,
1927 struct mapped_device
*md
)
1935 * Avoid initializing info for blk-mq; it passes
1936 * target-specific data through info.ptr
1937 * (see: dm_mq_init_request)
1939 if (!md
->init_tio_pdu
)
1940 memset(&tio
->info
, 0, sizeof(tio
->info
));
1941 if (md
->kworker_task
)
1942 init_kthread_work(&tio
->work
, map_tio_request
);
1945 static struct dm_rq_target_io
*dm_old_prep_tio(struct request
*rq
,
1946 struct mapped_device
*md
,
1949 struct dm_rq_target_io
*tio
;
1951 struct dm_table
*table
;
1953 tio
= alloc_old_rq_tio(md
, gfp_mask
);
1957 init_tio(tio
, rq
, md
);
1959 table
= dm_get_live_table(md
, &srcu_idx
);
1961 * Must clone a request if this .request_fn DM device
1962 * is stacked on .request_fn device(s).
1964 if (!dm_table_mq_request_based(table
)) {
1965 if (!clone_old_rq(rq
, md
, tio
, gfp_mask
)) {
1966 dm_put_live_table(md
, srcu_idx
);
1967 free_old_rq_tio(tio
);
1971 dm_put_live_table(md
, srcu_idx
);
1977 * Called with the queue lock held.
1979 static int dm_old_prep_fn(struct request_queue
*q
, struct request
*rq
)
1981 struct mapped_device
*md
= q
->queuedata
;
1982 struct dm_rq_target_io
*tio
;
1984 if (unlikely(rq
->special
)) {
1985 DMWARN("Already has something in rq->special.");
1986 return BLKPREP_KILL
;
1989 tio
= dm_old_prep_tio(rq
, md
, GFP_ATOMIC
);
1991 return BLKPREP_DEFER
;
1994 rq
->cmd_flags
|= REQ_DONTPREP
;
2001 * 0 : the request has been processed
2002 * DM_MAPIO_REQUEUE : the original request needs to be requeued
2003 * < 0 : the request was completed due to failure
2005 static int map_request(struct dm_rq_target_io
*tio
, struct request
*rq
,
2006 struct mapped_device
*md
)
2009 struct dm_target
*ti
= tio
->ti
;
2010 struct request
*clone
= NULL
;
2014 r
= ti
->type
->map_rq(ti
, clone
, &tio
->info
);
2016 r
= ti
->type
->clone_and_map_rq(ti
, rq
, &tio
->info
, &clone
);
2018 /* The target wants to complete the I/O */
2019 dm_kill_unmapped_request(rq
, r
);
2022 if (r
!= DM_MAPIO_REMAPPED
)
2024 if (setup_clone(clone
, rq
, tio
, GFP_ATOMIC
)) {
2026 ti
->type
->release_clone_rq(clone
);
2027 return DM_MAPIO_REQUEUE
;
2032 case DM_MAPIO_SUBMITTED
:
2033 /* The target has taken the I/O to submit by itself later */
2035 case DM_MAPIO_REMAPPED
:
2036 /* The target has remapped the I/O so dispatch it */
2037 trace_block_rq_remap(clone
->q
, clone
, disk_devt(dm_disk(md
)),
2039 dm_dispatch_clone_request(clone
, rq
);
2041 case DM_MAPIO_REQUEUE
:
2042 /* The target wants to requeue the I/O */
2043 dm_requeue_original_request(md
, tio
->orig
);
2047 DMWARN("unimplemented target map return value: %d", r
);
2051 /* The target wants to complete the I/O */
2052 dm_kill_unmapped_request(rq
, r
);
2059 static void map_tio_request(struct kthread_work
*work
)
2061 struct dm_rq_target_io
*tio
= container_of(work
, struct dm_rq_target_io
, work
);
2062 struct request
*rq
= tio
->orig
;
2063 struct mapped_device
*md
= tio
->md
;
2065 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
)
2066 dm_requeue_original_request(md
, rq
);
2069 static void dm_start_request(struct mapped_device
*md
, struct request
*orig
)
2071 if (!orig
->q
->mq_ops
)
2072 blk_start_request(orig
);
2074 blk_mq_start_request(orig
);
2075 atomic_inc(&md
->pending
[rq_data_dir(orig
)]);
2077 if (md
->seq_rq_merge_deadline_usecs
) {
2078 md
->last_rq_pos
= rq_end_sector(orig
);
2079 md
->last_rq_rw
= rq_data_dir(orig
);
2080 md
->last_rq_start_time
= ktime_get();
2083 if (unlikely(dm_stats_used(&md
->stats
))) {
2084 struct dm_rq_target_io
*tio
= tio_from_request(orig
);
2085 tio
->duration_jiffies
= jiffies
;
2086 tio
->n_sectors
= blk_rq_sectors(orig
);
2087 dm_stats_account_io(&md
->stats
, orig
->cmd_flags
, blk_rq_pos(orig
),
2088 tio
->n_sectors
, false, 0, &tio
->stats_aux
);
2092 * Hold the md reference here for the in-flight I/O.
2093 * We can't rely on the reference count by device opener,
2094 * because the device may be closed during the request completion
2095 * when all bios are completed.
2096 * See the comment in rq_completed() too.
2101 #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000
2103 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device
*md
, char *buf
)
2105 return sprintf(buf
, "%u\n", md
->seq_rq_merge_deadline_usecs
);
2108 ssize_t
dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device
*md
,
2109 const char *buf
, size_t count
)
2113 if (!dm_request_based(md
) || md
->use_blk_mq
)
2116 if (kstrtouint(buf
, 10, &deadline
))
2119 if (deadline
> MAX_SEQ_RQ_MERGE_DEADLINE_USECS
)
2120 deadline
= MAX_SEQ_RQ_MERGE_DEADLINE_USECS
;
2122 md
->seq_rq_merge_deadline_usecs
= deadline
;
2127 static bool dm_request_peeked_before_merge_deadline(struct mapped_device
*md
)
2129 ktime_t kt_deadline
;
2131 if (!md
->seq_rq_merge_deadline_usecs
)
2134 kt_deadline
= ns_to_ktime((u64
)md
->seq_rq_merge_deadline_usecs
* NSEC_PER_USEC
);
2135 kt_deadline
= ktime_add_safe(md
->last_rq_start_time
, kt_deadline
);
2137 return !ktime_after(ktime_get(), kt_deadline
);
2141 * q->request_fn for request-based dm.
2142 * Called with the queue lock held.
2144 static void dm_request_fn(struct request_queue
*q
)
2146 struct mapped_device
*md
= q
->queuedata
;
2147 struct dm_target
*ti
= md
->immutable_target
;
2149 struct dm_rq_target_io
*tio
;
2152 if (unlikely(!ti
)) {
2154 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2156 ti
= dm_table_find_target(map
, pos
);
2157 dm_put_live_table(md
, srcu_idx
);
2161 * For suspend, check blk_queue_stopped() and increment
2162 * ->pending within a single queue_lock not to increment the
2163 * number of in-flight I/Os after the queue is stopped in
2166 while (!blk_queue_stopped(q
)) {
2167 rq
= blk_peek_request(q
);
2171 /* always use block 0 to find the target for flushes for now */
2173 if (!(rq
->cmd_flags
& REQ_FLUSH
))
2174 pos
= blk_rq_pos(rq
);
2176 if ((dm_request_peeked_before_merge_deadline(md
) &&
2177 md_in_flight(md
) && rq
->bio
&& rq
->bio
->bi_vcnt
== 1 &&
2178 md
->last_rq_pos
== pos
&& md
->last_rq_rw
== rq_data_dir(rq
)) ||
2179 (ti
->type
->busy
&& ti
->type
->busy(ti
))) {
2180 blk_delay_queue(q
, HZ
/ 100);
2184 dm_start_request(md
, rq
);
2186 tio
= tio_from_request(rq
);
2187 /* Establish tio->ti before queuing work (map_tio_request) */
2189 queue_kthread_work(&md
->kworker
, &tio
->work
);
2190 BUG_ON(!irqs_disabled());
2194 static int dm_any_congested(void *congested_data
, int bdi_bits
)
2197 struct mapped_device
*md
= congested_data
;
2198 struct dm_table
*map
;
2200 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2201 if (dm_request_based(md
)) {
2203 * With request-based DM we only need to check the
2204 * top-level queue for congestion.
2206 r
= md
->queue
->backing_dev_info
.wb
.state
& bdi_bits
;
2208 map
= dm_get_live_table_fast(md
);
2210 r
= dm_table_any_congested(map
, bdi_bits
);
2211 dm_put_live_table_fast(md
);
2218 /*-----------------------------------------------------------------
2219 * An IDR is used to keep track of allocated minor numbers.
2220 *---------------------------------------------------------------*/
2221 static void free_minor(int minor
)
2223 spin_lock(&_minor_lock
);
2224 idr_remove(&_minor_idr
, minor
);
2225 spin_unlock(&_minor_lock
);
2229 * See if the device with a specific minor # is free.
2231 static int specific_minor(int minor
)
2235 if (minor
>= (1 << MINORBITS
))
2238 idr_preload(GFP_KERNEL
);
2239 spin_lock(&_minor_lock
);
2241 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
2243 spin_unlock(&_minor_lock
);
2246 return r
== -ENOSPC
? -EBUSY
: r
;
2250 static int next_free_minor(int *minor
)
2254 idr_preload(GFP_KERNEL
);
2255 spin_lock(&_minor_lock
);
2257 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
2259 spin_unlock(&_minor_lock
);
2267 static const struct block_device_operations dm_blk_dops
;
2269 static void dm_wq_work(struct work_struct
*work
);
2271 static void dm_init_md_queue(struct mapped_device
*md
)
2274 * Request-based dm devices cannot be stacked on top of bio-based dm
2275 * devices. The type of this dm device may not have been decided yet.
2276 * The type is decided at the first table loading time.
2277 * To prevent problematic device stacking, clear the queue flag
2278 * for request stacking support until then.
2280 * This queue is new, so no concurrency on the queue_flags.
2282 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE
, md
->queue
);
2285 * Initialize data that will only be used by a non-blk-mq DM queue
2286 * - must do so here (in alloc_dev callchain) before queue is used
2288 md
->queue
->queuedata
= md
;
2289 md
->queue
->backing_dev_info
.congested_data
= md
;
2292 static void dm_init_normal_md_queue(struct mapped_device
*md
)
2294 md
->use_blk_mq
= false;
2295 dm_init_md_queue(md
);
2298 * Initialize aspects of queue that aren't relevant for blk-mq
2300 md
->queue
->backing_dev_info
.congested_fn
= dm_any_congested
;
2301 blk_queue_bounce_limit(md
->queue
, BLK_BOUNCE_ANY
);
2304 static void cleanup_mapped_device(struct mapped_device
*md
)
2307 destroy_workqueue(md
->wq
);
2308 if (md
->kworker_task
)
2309 kthread_stop(md
->kworker_task
);
2310 mempool_destroy(md
->io_pool
);
2311 mempool_destroy(md
->rq_pool
);
2313 bioset_free(md
->bs
);
2315 cleanup_srcu_struct(&md
->io_barrier
);
2318 spin_lock(&_minor_lock
);
2319 md
->disk
->private_data
= NULL
;
2320 spin_unlock(&_minor_lock
);
2321 del_gendisk(md
->disk
);
2326 blk_cleanup_queue(md
->queue
);
2335 * Allocate and initialise a blank device with a given minor.
2337 static struct mapped_device
*alloc_dev(int minor
)
2339 int r
, numa_node_id
= dm_get_numa_node();
2340 struct mapped_device
*md
;
2343 md
= kzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
2345 DMWARN("unable to allocate device, out of memory.");
2349 if (!try_module_get(THIS_MODULE
))
2350 goto bad_module_get
;
2352 /* get a minor number for the dev */
2353 if (minor
== DM_ANY_MINOR
)
2354 r
= next_free_minor(&minor
);
2356 r
= specific_minor(minor
);
2360 r
= init_srcu_struct(&md
->io_barrier
);
2362 goto bad_io_barrier
;
2364 md
->numa_node_id
= numa_node_id
;
2365 md
->use_blk_mq
= use_blk_mq
;
2366 md
->init_tio_pdu
= false;
2367 md
->type
= DM_TYPE_NONE
;
2368 mutex_init(&md
->suspend_lock
);
2369 mutex_init(&md
->type_lock
);
2370 mutex_init(&md
->table_devices_lock
);
2371 spin_lock_init(&md
->deferred_lock
);
2372 atomic_set(&md
->holders
, 1);
2373 atomic_set(&md
->open_count
, 0);
2374 atomic_set(&md
->event_nr
, 0);
2375 atomic_set(&md
->uevent_seq
, 0);
2376 INIT_LIST_HEAD(&md
->uevent_list
);
2377 INIT_LIST_HEAD(&md
->table_devices
);
2378 spin_lock_init(&md
->uevent_lock
);
2380 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
2384 dm_init_md_queue(md
);
2386 md
->disk
= alloc_disk_node(1, numa_node_id
);
2390 atomic_set(&md
->pending
[0], 0);
2391 atomic_set(&md
->pending
[1], 0);
2392 init_waitqueue_head(&md
->wait
);
2393 INIT_WORK(&md
->work
, dm_wq_work
);
2394 init_waitqueue_head(&md
->eventq
);
2395 init_completion(&md
->kobj_holder
.completion
);
2396 md
->kworker_task
= NULL
;
2398 md
->disk
->major
= _major
;
2399 md
->disk
->first_minor
= minor
;
2400 md
->disk
->fops
= &dm_blk_dops
;
2401 md
->disk
->queue
= md
->queue
;
2402 md
->disk
->private_data
= md
;
2403 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
2405 format_dev_t(md
->name
, MKDEV(_major
, minor
));
2407 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
2411 md
->bdev
= bdget_disk(md
->disk
, 0);
2415 bio_init(&md
->flush_bio
);
2416 md
->flush_bio
.bi_bdev
= md
->bdev
;
2417 md
->flush_bio
.bi_rw
= WRITE_FLUSH
;
2419 dm_stats_init(&md
->stats
);
2421 /* Populate the mapping, nobody knows we exist yet */
2422 spin_lock(&_minor_lock
);
2423 old_md
= idr_replace(&_minor_idr
, md
, minor
);
2424 spin_unlock(&_minor_lock
);
2426 BUG_ON(old_md
!= MINOR_ALLOCED
);
2431 cleanup_mapped_device(md
);
2435 module_put(THIS_MODULE
);
2441 static void unlock_fs(struct mapped_device
*md
);
2443 static void free_dev(struct mapped_device
*md
)
2445 int minor
= MINOR(disk_devt(md
->disk
));
2449 cleanup_mapped_device(md
);
2451 blk_mq_free_tag_set(md
->tag_set
);
2455 free_table_devices(&md
->table_devices
);
2456 dm_stats_cleanup(&md
->stats
);
2459 module_put(THIS_MODULE
);
2463 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
2465 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
2468 /* The md already has necessary mempools. */
2469 if (dm_table_get_type(t
) == DM_TYPE_BIO_BASED
) {
2471 * Reload bioset because front_pad may have changed
2472 * because a different table was loaded.
2474 bioset_free(md
->bs
);
2479 * There's no need to reload with request-based dm
2480 * because the size of front_pad doesn't change.
2481 * Note for future: If you are to reload bioset,
2482 * prep-ed requests in the queue may refer
2483 * to bio from the old bioset, so you must walk
2484 * through the queue to unprep.
2489 BUG_ON(!p
|| md
->io_pool
|| md
->rq_pool
|| md
->bs
);
2491 md
->io_pool
= p
->io_pool
;
2493 md
->rq_pool
= p
->rq_pool
;
2499 /* mempool bind completed, no longer need any mempools in the table */
2500 dm_table_free_md_mempools(t
);
2504 * Bind a table to the device.
2506 static void event_callback(void *context
)
2508 unsigned long flags
;
2510 struct mapped_device
*md
= (struct mapped_device
*) context
;
2512 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2513 list_splice_init(&md
->uevent_list
, &uevents
);
2514 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2516 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
2518 atomic_inc(&md
->event_nr
);
2519 wake_up(&md
->eventq
);
2523 * Protected by md->suspend_lock obtained by dm_swap_table().
2525 static void __set_size(struct mapped_device
*md
, sector_t size
)
2527 set_capacity(md
->disk
, size
);
2529 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
2533 * Returns old map, which caller must destroy.
2535 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
2536 struct queue_limits
*limits
)
2538 struct dm_table
*old_map
;
2539 struct request_queue
*q
= md
->queue
;
2542 size
= dm_table_get_size(t
);
2545 * Wipe any geometry if the size of the table changed.
2547 if (size
!= dm_get_size(md
))
2548 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
2550 __set_size(md
, size
);
2552 dm_table_event_callback(t
, event_callback
, md
);
2555 * The queue hasn't been stopped yet, if the old table type wasn't
2556 * for request-based during suspension. So stop it to prevent
2557 * I/O mapping before resume.
2558 * This must be done before setting the queue restrictions,
2559 * because request-based dm may be run just after the setting.
2561 if (dm_table_request_based(t
)) {
2564 * Leverage the fact that request-based DM targets are
2565 * immutable singletons and establish md->immutable_target
2566 * - used to optimize both dm_request_fn and dm_mq_queue_rq
2568 md
->immutable_target
= dm_table_get_immutable_target(t
);
2571 __bind_mempools(md
, t
);
2573 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2574 rcu_assign_pointer(md
->map
, (void *)t
);
2575 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
2577 dm_table_set_restrictions(t
, q
, limits
);
2585 * Returns unbound table for the caller to free.
2587 static struct dm_table
*__unbind(struct mapped_device
*md
)
2589 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
2594 dm_table_event_callback(map
, NULL
, NULL
);
2595 RCU_INIT_POINTER(md
->map
, NULL
);
2602 * Constructor for a new device.
2604 int dm_create(int minor
, struct mapped_device
**result
)
2606 struct mapped_device
*md
;
2608 md
= alloc_dev(minor
);
2619 * Functions to manage md->type.
2620 * All are required to hold md->type_lock.
2622 void dm_lock_md_type(struct mapped_device
*md
)
2624 mutex_lock(&md
->type_lock
);
2627 void dm_unlock_md_type(struct mapped_device
*md
)
2629 mutex_unlock(&md
->type_lock
);
2632 void dm_set_md_type(struct mapped_device
*md
, unsigned type
)
2634 BUG_ON(!mutex_is_locked(&md
->type_lock
));
2638 unsigned dm_get_md_type(struct mapped_device
*md
)
2643 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
2645 return md
->immutable_target_type
;
2649 * The queue_limits are only valid as long as you have a reference
2652 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2654 BUG_ON(!atomic_read(&md
->holders
));
2655 return &md
->queue
->limits
;
2657 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2659 static void dm_old_init_rq_based_worker_thread(struct mapped_device
*md
)
2661 /* Initialize the request-based DM worker thread */
2662 init_kthread_worker(&md
->kworker
);
2663 md
->kworker_task
= kthread_run(kthread_worker_fn
, &md
->kworker
,
2664 "kdmwork-%s", dm_device_name(md
));
2668 * Fully initialize a .request_fn request-based queue.
2670 static int dm_old_init_request_queue(struct mapped_device
*md
)
2672 /* Fully initialize the queue */
2673 if (!blk_init_allocated_queue(md
->queue
, dm_request_fn
, NULL
))
2676 /* disable dm_request_fn's merge heuristic by default */
2677 md
->seq_rq_merge_deadline_usecs
= 0;
2679 dm_init_normal_md_queue(md
);
2680 blk_queue_softirq_done(md
->queue
, dm_softirq_done
);
2681 blk_queue_prep_rq(md
->queue
, dm_old_prep_fn
);
2683 dm_old_init_rq_based_worker_thread(md
);
2685 elv_register_queue(md
->queue
);
2690 static int dm_mq_init_request(void *data
, struct request
*rq
,
2691 unsigned int hctx_idx
, unsigned int request_idx
,
2692 unsigned int numa_node
)
2694 struct mapped_device
*md
= data
;
2695 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2698 * Must initialize md member of tio, otherwise it won't
2699 * be available in dm_mq_queue_rq.
2703 if (md
->init_tio_pdu
) {
2704 /* target-specific per-io data is immediately after the tio */
2705 tio
->info
.ptr
= tio
+ 1;
2711 static int dm_mq_queue_rq(struct blk_mq_hw_ctx
*hctx
,
2712 const struct blk_mq_queue_data
*bd
)
2714 struct request
*rq
= bd
->rq
;
2715 struct dm_rq_target_io
*tio
= blk_mq_rq_to_pdu(rq
);
2716 struct mapped_device
*md
= tio
->md
;
2717 struct dm_target
*ti
= md
->immutable_target
;
2719 if (unlikely(!ti
)) {
2721 struct dm_table
*map
= dm_get_live_table(md
, &srcu_idx
);
2723 ti
= dm_table_find_target(map
, 0);
2724 dm_put_live_table(md
, srcu_idx
);
2727 if (ti
->type
->busy
&& ti
->type
->busy(ti
))
2728 return BLK_MQ_RQ_QUEUE_BUSY
;
2730 dm_start_request(md
, rq
);
2732 /* Init tio using md established in .init_request */
2733 init_tio(tio
, rq
, md
);
2736 * Establish tio->ti before queuing work (map_tio_request)
2737 * or making direct call to map_request().
2741 /* Direct call is fine since .queue_rq allows allocations */
2742 if (map_request(tio
, rq
, md
) == DM_MAPIO_REQUEUE
) {
2743 /* Undo dm_start_request() before requeuing */
2744 rq_end_stats(md
, rq
);
2745 rq_completed(md
, rq_data_dir(rq
), false);
2746 return BLK_MQ_RQ_QUEUE_BUSY
;
2749 return BLK_MQ_RQ_QUEUE_OK
;
2752 static struct blk_mq_ops dm_mq_ops
= {
2753 .queue_rq
= dm_mq_queue_rq
,
2754 .map_queue
= blk_mq_map_queue
,
2755 .complete
= dm_softirq_done
,
2756 .init_request
= dm_mq_init_request
,
2759 static int dm_mq_init_request_queue(struct mapped_device
*md
,
2760 struct dm_target
*immutable_tgt
)
2762 struct request_queue
*q
;
2765 if (dm_get_md_type(md
) == DM_TYPE_REQUEST_BASED
) {
2766 DMERR("request-based dm-mq may only be stacked on blk-mq device(s)");
2770 md
->tag_set
= kzalloc_node(sizeof(struct blk_mq_tag_set
), GFP_KERNEL
, md
->numa_node_id
);
2774 md
->tag_set
->ops
= &dm_mq_ops
;
2775 md
->tag_set
->queue_depth
= dm_get_blk_mq_queue_depth();
2776 md
->tag_set
->numa_node
= md
->numa_node_id
;
2777 md
->tag_set
->flags
= BLK_MQ_F_SHOULD_MERGE
| BLK_MQ_F_SG_MERGE
;
2778 md
->tag_set
->nr_hw_queues
= dm_get_blk_mq_nr_hw_queues();
2779 md
->tag_set
->driver_data
= md
;
2781 md
->tag_set
->cmd_size
= sizeof(struct dm_rq_target_io
);
2782 if (immutable_tgt
&& immutable_tgt
->per_io_data_size
) {
2783 /* any target-specific per-io data is immediately after the tio */
2784 md
->tag_set
->cmd_size
+= immutable_tgt
->per_io_data_size
;
2785 md
->init_tio_pdu
= true;
2788 err
= blk_mq_alloc_tag_set(md
->tag_set
);
2790 goto out_kfree_tag_set
;
2792 q
= blk_mq_init_allocated_queue(md
->tag_set
, md
->queue
);
2797 dm_init_md_queue(md
);
2799 /* backfill 'mq' sysfs registration normally done in blk_register_queue */
2800 blk_mq_register_disk(md
->disk
);
2805 blk_mq_free_tag_set(md
->tag_set
);
2812 static unsigned filter_md_type(unsigned type
, struct mapped_device
*md
)
2814 if (type
== DM_TYPE_BIO_BASED
)
2817 return !md
->use_blk_mq
? DM_TYPE_REQUEST_BASED
: DM_TYPE_MQ_REQUEST_BASED
;
2821 * Setup the DM device's queue based on md's type
2823 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
2826 unsigned md_type
= filter_md_type(dm_get_md_type(md
), md
);
2829 case DM_TYPE_REQUEST_BASED
:
2830 r
= dm_old_init_request_queue(md
);
2832 DMERR("Cannot initialize queue for request-based mapped device");
2836 case DM_TYPE_MQ_REQUEST_BASED
:
2837 r
= dm_mq_init_request_queue(md
, dm_table_get_immutable_target(t
));
2839 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2843 case DM_TYPE_BIO_BASED
:
2844 dm_init_normal_md_queue(md
);
2845 blk_queue_make_request(md
->queue
, dm_make_request
);
2847 * DM handles splitting bios as needed. Free the bio_split bioset
2848 * since it won't be used (saves 1 process per bio-based DM device).
2850 bioset_free(md
->queue
->bio_split
);
2851 md
->queue
->bio_split
= NULL
;
2858 struct mapped_device
*dm_get_md(dev_t dev
)
2860 struct mapped_device
*md
;
2861 unsigned minor
= MINOR(dev
);
2863 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2866 spin_lock(&_minor_lock
);
2868 md
= idr_find(&_minor_idr
, minor
);
2870 if ((md
== MINOR_ALLOCED
||
2871 (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2872 dm_deleting_md(md
) ||
2873 test_bit(DMF_FREEING
, &md
->flags
))) {
2881 spin_unlock(&_minor_lock
);
2885 EXPORT_SYMBOL_GPL(dm_get_md
);
2887 void *dm_get_mdptr(struct mapped_device
*md
)
2889 return md
->interface_ptr
;
2892 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2894 md
->interface_ptr
= ptr
;
2897 void dm_get(struct mapped_device
*md
)
2899 atomic_inc(&md
->holders
);
2900 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2903 int dm_hold(struct mapped_device
*md
)
2905 spin_lock(&_minor_lock
);
2906 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2907 spin_unlock(&_minor_lock
);
2911 spin_unlock(&_minor_lock
);
2914 EXPORT_SYMBOL_GPL(dm_hold
);
2916 const char *dm_device_name(struct mapped_device
*md
)
2920 EXPORT_SYMBOL_GPL(dm_device_name
);
2922 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2924 struct dm_table
*map
;
2929 spin_lock(&_minor_lock
);
2930 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2931 set_bit(DMF_FREEING
, &md
->flags
);
2932 spin_unlock(&_minor_lock
);
2934 if (dm_request_based(md
) && md
->kworker_task
)
2935 flush_kthread_worker(&md
->kworker
);
2938 * Take suspend_lock so that presuspend and postsuspend methods
2939 * do not race with internal suspend.
2941 mutex_lock(&md
->suspend_lock
);
2942 map
= dm_get_live_table(md
, &srcu_idx
);
2943 if (!dm_suspended_md(md
)) {
2944 dm_table_presuspend_targets(map
);
2945 dm_table_postsuspend_targets(map
);
2947 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2948 dm_put_live_table(md
, srcu_idx
);
2949 mutex_unlock(&md
->suspend_lock
);
2952 * Rare, but there may be I/O requests still going to complete,
2953 * for example. Wait for all references to disappear.
2954 * No one should increment the reference count of the mapped_device,
2955 * after the mapped_device state becomes DMF_FREEING.
2958 while (atomic_read(&md
->holders
))
2960 else if (atomic_read(&md
->holders
))
2961 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2962 dm_device_name(md
), atomic_read(&md
->holders
));
2965 dm_table_destroy(__unbind(md
));
2969 void dm_destroy(struct mapped_device
*md
)
2971 __dm_destroy(md
, true);
2974 void dm_destroy_immediate(struct mapped_device
*md
)
2976 __dm_destroy(md
, false);
2979 void dm_put(struct mapped_device
*md
)
2981 atomic_dec(&md
->holders
);
2983 EXPORT_SYMBOL_GPL(dm_put
);
2985 static int dm_wait_for_completion(struct mapped_device
*md
, int interruptible
)
2988 DECLARE_WAITQUEUE(wait
, current
);
2990 add_wait_queue(&md
->wait
, &wait
);
2993 set_current_state(interruptible
);
2995 if (!md_in_flight(md
))
2998 if (interruptible
== TASK_INTERRUPTIBLE
&&
2999 signal_pending(current
)) {
3006 set_current_state(TASK_RUNNING
);
3008 remove_wait_queue(&md
->wait
, &wait
);
3014 * Process the deferred bios
3016 static void dm_wq_work(struct work_struct
*work
)
3018 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
3022 struct dm_table
*map
;
3024 map
= dm_get_live_table(md
, &srcu_idx
);
3026 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
3027 spin_lock_irq(&md
->deferred_lock
);
3028 c
= bio_list_pop(&md
->deferred
);
3029 spin_unlock_irq(&md
->deferred_lock
);
3034 if (dm_request_based(md
))
3035 generic_make_request(c
);
3037 __split_and_process_bio(md
, map
, c
);
3040 dm_put_live_table(md
, srcu_idx
);
3043 static void dm_queue_flush(struct mapped_device
*md
)
3045 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3046 smp_mb__after_atomic();
3047 queue_work(md
->wq
, &md
->work
);
3051 * Swap in a new table, returning the old one for the caller to destroy.
3053 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
3055 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
3056 struct queue_limits limits
;
3059 mutex_lock(&md
->suspend_lock
);
3061 /* device must be suspended */
3062 if (!dm_suspended_md(md
))
3066 * If the new table has no data devices, retain the existing limits.
3067 * This helps multipath with queue_if_no_path if all paths disappear,
3068 * then new I/O is queued based on these limits, and then some paths
3071 if (dm_table_has_no_data_devices(table
)) {
3072 live_map
= dm_get_live_table_fast(md
);
3074 limits
= md
->queue
->limits
;
3075 dm_put_live_table_fast(md
);
3079 r
= dm_calculate_queue_limits(table
, &limits
);
3086 map
= __bind(md
, table
, &limits
);
3089 mutex_unlock(&md
->suspend_lock
);
3094 * Functions to lock and unlock any filesystem running on the
3097 static int lock_fs(struct mapped_device
*md
)
3101 WARN_ON(md
->frozen_sb
);
3103 md
->frozen_sb
= freeze_bdev(md
->bdev
);
3104 if (IS_ERR(md
->frozen_sb
)) {
3105 r
= PTR_ERR(md
->frozen_sb
);
3106 md
->frozen_sb
= NULL
;
3110 set_bit(DMF_FROZEN
, &md
->flags
);
3115 static void unlock_fs(struct mapped_device
*md
)
3117 if (!test_bit(DMF_FROZEN
, &md
->flags
))
3120 thaw_bdev(md
->bdev
, md
->frozen_sb
);
3121 md
->frozen_sb
= NULL
;
3122 clear_bit(DMF_FROZEN
, &md
->flags
);
3126 * If __dm_suspend returns 0, the device is completely quiescent
3127 * now. There is no request-processing activity. All new requests
3128 * are being added to md->deferred list.
3130 * Caller must hold md->suspend_lock
3132 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
3133 unsigned suspend_flags
, int interruptible
)
3135 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
3136 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
3140 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
3141 * This flag is cleared before dm_suspend returns.
3144 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3147 * This gets reverted if there's an error later and the targets
3148 * provide the .presuspend_undo hook.
3150 dm_table_presuspend_targets(map
);
3153 * Flush I/O to the device.
3154 * Any I/O submitted after lock_fs() may not be flushed.
3155 * noflush takes precedence over do_lockfs.
3156 * (lock_fs() flushes I/Os and waits for them to complete.)
3158 if (!noflush
&& do_lockfs
) {
3161 dm_table_presuspend_undo_targets(map
);
3167 * Here we must make sure that no processes are submitting requests
3168 * to target drivers i.e. no one may be executing
3169 * __split_and_process_bio. This is called from dm_request and
3172 * To get all processes out of __split_and_process_bio in dm_request,
3173 * we take the write lock. To prevent any process from reentering
3174 * __split_and_process_bio from dm_request and quiesce the thread
3175 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
3176 * flush_workqueue(md->wq).
3178 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3180 synchronize_srcu(&md
->io_barrier
);
3183 * Stop md->queue before flushing md->wq in case request-based
3184 * dm defers requests to md->wq from md->queue.
3186 if (dm_request_based(md
)) {
3187 dm_stop_queue(md
->queue
);
3188 if (md
->kworker_task
)
3189 flush_kthread_worker(&md
->kworker
);
3192 flush_workqueue(md
->wq
);
3195 * At this point no more requests are entering target request routines.
3196 * We call dm_wait_for_completion to wait for all existing requests
3199 r
= dm_wait_for_completion(md
, interruptible
);
3202 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
3204 synchronize_srcu(&md
->io_barrier
);
3206 /* were we interrupted ? */
3210 if (dm_request_based(md
))
3211 dm_start_queue(md
->queue
);
3214 dm_table_presuspend_undo_targets(map
);
3215 /* pushback list is already flushed, so skip flush */
3222 * We need to be able to change a mapping table under a mounted
3223 * filesystem. For example we might want to move some data in
3224 * the background. Before the table can be swapped with
3225 * dm_bind_table, dm_suspend must be called to flush any in
3226 * flight bios and ensure that any further io gets deferred.
3229 * Suspend mechanism in request-based dm.
3231 * 1. Flush all I/Os by lock_fs() if needed.
3232 * 2. Stop dispatching any I/O by stopping the request_queue.
3233 * 3. Wait for all in-flight I/Os to be completed or requeued.
3235 * To abort suspend, start the request_queue.
3237 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3239 struct dm_table
*map
= NULL
;
3243 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3245 if (dm_suspended_md(md
)) {
3250 if (dm_suspended_internally_md(md
)) {
3251 /* already internally suspended, wait for internal resume */
3252 mutex_unlock(&md
->suspend_lock
);
3253 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3259 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3261 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
);
3265 set_bit(DMF_SUSPENDED
, &md
->flags
);
3267 dm_table_postsuspend_targets(map
);
3270 mutex_unlock(&md
->suspend_lock
);
3274 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
3277 int r
= dm_table_resume_targets(map
);
3285 * Flushing deferred I/Os must be done after targets are resumed
3286 * so that mapping of targets can work correctly.
3287 * Request-based dm is queueing the deferred I/Os in its request_queue.
3289 if (dm_request_based(md
))
3290 dm_start_queue(md
->queue
);
3297 int dm_resume(struct mapped_device
*md
)
3300 struct dm_table
*map
= NULL
;
3303 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
3305 if (!dm_suspended_md(md
))
3308 if (dm_suspended_internally_md(md
)) {
3309 /* already internally suspended, wait for internal resume */
3310 mutex_unlock(&md
->suspend_lock
);
3311 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
3317 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3318 if (!map
|| !dm_table_get_size(map
))
3321 r
= __dm_resume(md
, map
);
3325 clear_bit(DMF_SUSPENDED
, &md
->flags
);
3329 mutex_unlock(&md
->suspend_lock
);
3335 * Internal suspend/resume works like userspace-driven suspend. It waits
3336 * until all bios finish and prevents issuing new bios to the target drivers.
3337 * It may be used only from the kernel.
3340 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
3342 struct dm_table
*map
= NULL
;
3344 if (md
->internal_suspend_count
++)
3345 return; /* nested internal suspend */
3347 if (dm_suspended_md(md
)) {
3348 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3349 return; /* nest suspend */
3352 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
3355 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
3356 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
3357 * would require changing .presuspend to return an error -- avoid this
3358 * until there is a need for more elaborate variants of internal suspend.
3360 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
);
3362 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3364 dm_table_postsuspend_targets(map
);
3367 static void __dm_internal_resume(struct mapped_device
*md
)
3369 BUG_ON(!md
->internal_suspend_count
);
3371 if (--md
->internal_suspend_count
)
3372 return; /* resume from nested internal suspend */
3374 if (dm_suspended_md(md
))
3375 goto done
; /* resume from nested suspend */
3378 * NOTE: existing callers don't need to call dm_table_resume_targets
3379 * (which may fail -- so best to avoid it for now by passing NULL map)
3381 (void) __dm_resume(md
, NULL
);
3384 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3385 smp_mb__after_atomic();
3386 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
3389 void dm_internal_suspend_noflush(struct mapped_device
*md
)
3391 mutex_lock(&md
->suspend_lock
);
3392 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
3393 mutex_unlock(&md
->suspend_lock
);
3395 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
3397 void dm_internal_resume(struct mapped_device
*md
)
3399 mutex_lock(&md
->suspend_lock
);
3400 __dm_internal_resume(md
);
3401 mutex_unlock(&md
->suspend_lock
);
3403 EXPORT_SYMBOL_GPL(dm_internal_resume
);
3406 * Fast variants of internal suspend/resume hold md->suspend_lock,
3407 * which prevents interaction with userspace-driven suspend.
3410 void dm_internal_suspend_fast(struct mapped_device
*md
)
3412 mutex_lock(&md
->suspend_lock
);
3413 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3416 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
3417 synchronize_srcu(&md
->io_barrier
);
3418 flush_workqueue(md
->wq
);
3419 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
3421 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
3423 void dm_internal_resume_fast(struct mapped_device
*md
)
3425 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
3431 mutex_unlock(&md
->suspend_lock
);
3433 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
3435 /*-----------------------------------------------------------------
3436 * Event notification.
3437 *---------------------------------------------------------------*/
3438 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
3441 char udev_cookie
[DM_COOKIE_LENGTH
];
3442 char *envp
[] = { udev_cookie
, NULL
};
3445 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
3447 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
3448 DM_COOKIE_ENV_VAR_NAME
, cookie
);
3449 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
3454 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
3456 return atomic_add_return(1, &md
->uevent_seq
);
3459 uint32_t dm_get_event_nr(struct mapped_device
*md
)
3461 return atomic_read(&md
->event_nr
);
3464 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
3466 return wait_event_interruptible(md
->eventq
,
3467 (event_nr
!= atomic_read(&md
->event_nr
)));
3470 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
3472 unsigned long flags
;
3474 spin_lock_irqsave(&md
->uevent_lock
, flags
);
3475 list_add(elist
, &md
->uevent_list
);
3476 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
3480 * The gendisk is only valid as long as you have a reference
3483 struct gendisk
*dm_disk(struct mapped_device
*md
)
3487 EXPORT_SYMBOL_GPL(dm_disk
);
3489 struct kobject
*dm_kobject(struct mapped_device
*md
)
3491 return &md
->kobj_holder
.kobj
;
3494 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
3496 struct mapped_device
*md
;
3498 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
3500 if (test_bit(DMF_FREEING
, &md
->flags
) ||
3508 int dm_suspended_md(struct mapped_device
*md
)
3510 return test_bit(DMF_SUSPENDED
, &md
->flags
);
3513 int dm_suspended_internally_md(struct mapped_device
*md
)
3515 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
3518 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
3520 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
3523 int dm_suspended(struct dm_target
*ti
)
3525 return dm_suspended_md(dm_table_get_md(ti
->table
));
3527 EXPORT_SYMBOL_GPL(dm_suspended
);
3529 int dm_noflush_suspending(struct dm_target
*ti
)
3531 return __noflush_suspending(dm_table_get_md(ti
->table
));
3533 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
3535 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, unsigned type
,
3536 unsigned integrity
, unsigned per_io_data_size
)
3538 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
3539 struct kmem_cache
*cachep
= NULL
;
3540 unsigned int pool_size
= 0;
3541 unsigned int front_pad
;
3546 type
= filter_md_type(type
, md
);
3549 case DM_TYPE_BIO_BASED
:
3551 pool_size
= dm_get_reserved_bio_based_ios();
3552 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
3554 case DM_TYPE_REQUEST_BASED
:
3555 cachep
= _rq_tio_cache
;
3556 pool_size
= dm_get_reserved_rq_based_ios();
3557 pools
->rq_pool
= mempool_create_slab_pool(pool_size
, _rq_cache
);
3558 if (!pools
->rq_pool
)
3560 /* fall through to setup remaining rq-based pools */
3561 case DM_TYPE_MQ_REQUEST_BASED
:
3563 pool_size
= dm_get_reserved_rq_based_ios();
3564 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
3565 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3572 pools
->io_pool
= mempool_create_slab_pool(pool_size
, cachep
);
3573 if (!pools
->io_pool
)
3577 pools
->bs
= bioset_create_nobvec(pool_size
, front_pad
);
3581 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
3587 dm_free_md_mempools(pools
);
3592 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
3597 mempool_destroy(pools
->io_pool
);
3598 mempool_destroy(pools
->rq_pool
);
3601 bioset_free(pools
->bs
);
3606 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
3609 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3610 const struct pr_ops
*ops
;
3614 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
3618 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3619 if (ops
&& ops
->pr_register
)
3620 r
= ops
->pr_register(bdev
, old_key
, new_key
, flags
);
3628 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
3631 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3632 const struct pr_ops
*ops
;
3636 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
3640 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3641 if (ops
&& ops
->pr_reserve
)
3642 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
3650 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
3652 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3653 const struct pr_ops
*ops
;
3657 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
3661 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3662 if (ops
&& ops
->pr_release
)
3663 r
= ops
->pr_release(bdev
, key
, type
);
3671 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
3672 enum pr_type type
, bool abort
)
3674 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3675 const struct pr_ops
*ops
;
3679 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
3683 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3684 if (ops
&& ops
->pr_preempt
)
3685 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
3693 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
3695 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
3696 const struct pr_ops
*ops
;
3700 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
3704 ops
= bdev
->bd_disk
->fops
->pr_ops
;
3705 if (ops
&& ops
->pr_clear
)
3706 r
= ops
->pr_clear(bdev
, key
);
3714 static const struct pr_ops dm_pr_ops
= {
3715 .pr_register
= dm_pr_register
,
3716 .pr_reserve
= dm_pr_reserve
,
3717 .pr_release
= dm_pr_release
,
3718 .pr_preempt
= dm_pr_preempt
,
3719 .pr_clear
= dm_pr_clear
,
3722 static const struct block_device_operations dm_blk_dops
= {
3723 .open
= dm_blk_open
,
3724 .release
= dm_blk_close
,
3725 .ioctl
= dm_blk_ioctl
,
3726 .getgeo
= dm_blk_getgeo
,
3727 .pr_ops
= &dm_pr_ops
,
3728 .owner
= THIS_MODULE
3734 module_init(dm_init
);
3735 module_exit(dm_exit
);
3737 module_param(major
, uint
, 0);
3738 MODULE_PARM_DESC(major
, "The major number of the device mapper");
3740 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3741 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
3743 module_param(reserved_rq_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
3744 MODULE_PARM_DESC(reserved_rq_based_ios
, "Reserved IOs in request-based mempools");
3746 module_param(use_blk_mq
, bool, S_IRUGO
| S_IWUSR
);
3747 MODULE_PARM_DESC(use_blk_mq
, "Use block multiqueue for request-based DM devices");
3749 module_param(dm_mq_nr_hw_queues
, uint
, S_IRUGO
| S_IWUSR
);
3750 MODULE_PARM_DESC(dm_mq_nr_hw_queues
, "Number of hardware queues for request-based dm-mq devices");
3752 module_param(dm_mq_queue_depth
, uint
, S_IRUGO
| S_IWUSR
);
3753 MODULE_PARM_DESC(dm_mq_queue_depth
, "Queue depth for request-based dm-mq devices");
3755 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
3756 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
3758 MODULE_DESCRIPTION(DM_NAME
" driver");
3759 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3760 MODULE_LICENSE("GPL");