summaryrefslogtreecommitdiffstats
path: root/drivers/md/dm-kcopyd.c
blob: fda51bd140ed3d44369911953689eba44e668248 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
/*
 * Copyright (C) 2002 Sistina Software (UK) Limited.
 * Copyright (C) 2006 Red Hat GmbH
 *
 * This file is released under the GPL.
 *
 * Kcopyd provides a simple interface for copying an area of one
 * block-device to one or more other block-devices, with an asynchronous
 * completion notification.
 */

#include <linux/types.h>
#include <linux/atomic.h>
#include <linux/blkdev.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/list.h>
#include <linux/mempool.h>
#include <linux/module.h>
#include <linux/pagemap.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/workqueue.h>
#include <linux/mutex.h>
#include <linux/delay.h>
#include <linux/device-mapper.h>
#include <linux/dm-kcopyd.h>

#include "dm-core.h"

#define SPLIT_COUNT	8
#define MIN_JOBS	8

#define DEFAULT_SUB_JOB_SIZE_KB 512
#define MAX_SUB_JOB_SIZE_KB     1024

static unsigned int kcopyd_subjob_size_kb = DEFAULT_SUB_JOB_SIZE_KB;

module_param(kcopyd_subjob_size_kb, uint, S_IRUGO | S_IWUSR);
MODULE_PARM_DESC(kcopyd_subjob_size_kb, "Sub-job size for dm-kcopyd clients");

static unsigned int dm_get_kcopyd_subjob_size(void)
{
	unsigned int sub_job_size_kb;

	sub_job_size_kb = __dm_get_module_param(&kcopyd_subjob_size_kb,
						DEFAULT_SUB_JOB_SIZE_KB,
						MAX_SUB_JOB_SIZE_KB);

	return sub_job_size_kb << 1;
}

/*-----------------------------------------------------------------
 * Each kcopyd client has its own little pool of preallocated
 * pages for kcopyd io.
 *---------------------------------------------------------------*/
struct dm_kcopyd_client {
	struct page_list *pages;
	unsigned int nr_reserved_pages;
	unsigned int nr_free_pages;
	unsigned int sub_job_size;

	struct dm_io_client *io_client;

	wait_queue_head_t destroyq;

	mempool_t job_pool;

	struct workqueue_struct *kcopyd_wq;
	struct work_struct kcopyd_work;

	struct dm_kcopyd_throttle *throttle;

	atomic_t nr_jobs;

/*
 * We maintain four lists of jobs:
 *
 * i)   jobs waiting for pages
 * ii)  jobs that have pages, and are waiting for the io to be issued.
 * iii) jobs that don't need to do any IO and just run a callback
 * iv) jobs that have completed.
 *
 * All four of these are protected by job_lock.
 */
	spinlock_t job_lock;
	struct list_head callback_jobs;
	struct list_head complete_jobs;
	struct list_head io_jobs;
	struct list_head pages_jobs;
};

static struct page_list zero_page_list;

static DEFINE_SPINLOCK(throttle_spinlock);

/*
 * IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
 * When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
 * by 2.
 */
#define ACCOUNT_INTERVAL_SHIFT		SHIFT_HZ

/*
 * Sleep this number of milliseconds.
 *
 * The value was decided experimentally.
 * Smaller values seem to cause an increased copy rate above the limit.
 * The reason for this is unknown but possibly due to jiffies rounding errors
 * or read/write cache inside the disk.
 */
#define SLEEP_MSEC			100

/*
 * Maximum number of sleep events. There is a theoretical livelock if more
 * kcopyd clients do work simultaneously which this limit avoids.
 */
#define MAX_SLEEPS			10

static void io_job_start(struct dm_kcopyd_throttle *t)
{
	unsigned int throttle, now, difference;
	int slept = 0, skew;

	if (unlikely(!t))
		return;

try_again:
	spin_lock_irq(&throttle_spinlock);

	throttle = READ_ONCE(t->throttle);

	if (likely(throttle >= 100))
		goto skip_limit;

	now = jiffies;
	difference = now - t->last_jiffies;
	t->last_jiffies = now;
	if (t->num_io_jobs)
		t->io_period += difference;
	t->total_period += difference;

	/*
	 * Maintain sane values if we got a temporary overflow.
	 */
	if (unlikely(t->io_period > t->total_period))
		t->io_period = t->total_period;

	if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
		int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
		t->total_period >>= shift;
		t->io_period >>= shift;
	}

	skew = t->io_period - throttle * t->total_period / 100;

	if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
		slept++;
		spin_unlock_irq(&throttle_spinlock);
		msleep(SLEEP_MSEC);
		goto try_again;
	}

skip_limit:
	t->num_io_jobs++;

	spin_unlock_irq(&throttle_spinlock);
}

static void io_job_finish(struct dm_kcopyd_throttle *t)
{
	unsigned long flags;

	if (unlikely(!t))
		return;

	spin_lock_irqsave(&throttle_spinlock, flags);

	t->num_io_jobs--;

	if (likely(READ_ONCE(t->throttle) >= 100))
		goto skip_limit;

	if (!t->num_io_jobs) {
		unsigned int now, difference;

		now = jiffies;
		difference = now - t->last_jiffies;
		t->last_jiffies = now;

		t->io_period += difference;
		t->total_period += difference;

		/*
		 * Maintain sane values if we got a temporary overflow.
		 */
		if (unlikely(t->io_period > t->total_period))
			t->io_period = t->total_period;
	}

skip_limit:
	spin_unlock_irqrestore(&throttle_spinlock, flags);
}


static void wake(struct dm_kcopyd_client *kc)
{
	queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
}

/*
 * Obtain one page for the use of kcopyd.
 */
static struct page_list *alloc_pl(gfp_t gfp)
{
	struct page_list *pl;

	pl = kmalloc(sizeof(*pl), gfp);
	if (!pl)
		return NULL;

	pl->page = alloc_page(gfp | __GFP_HIGHMEM);
	if (!pl->page) {
		kfree(pl);
		return NULL;
	}

	return pl;
}

static void free_pl(struct page_list *pl)
{
	__free_page(pl->page);
	kfree(pl);
}

/*
 * Add the provided pages to a client's free page list, releasing
 * back to the system any beyond the reserved_pages limit.
 */
static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
{
	struct page_list *next;

	do {
		next = pl->next;

		if (kc->nr_free_pages >= kc->nr_reserved_pages)
			free_pl(pl);
		else {
			pl->next = kc->pages;
			kc->pages = pl;
			kc->nr_free_pages++;
		}

		pl = next;
	} while (pl);
}

static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
			    unsigned int nr, struct page_list **pages)
{
	struct page_list *pl;

	*pages = NULL;

	do {
		pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
		if (unlikely(!pl)) {
			/* Use reserved pages */
			pl = kc->pages;
			if (unlikely(!pl))
				goto out_of_memory;
			kc->pages = pl->next;
			kc->nr_free_pages--;
		}
		pl->next = *pages;
		*pages = pl;
	} while (--nr);

	return 0;

out_of_memory:
	if (*pages)
		kcopyd_put_pages(kc, *pages);
	return -ENOMEM;
}

/*
 * These three functions resize the page pool.
 */
static void drop_pages(struct page_list *pl)
{
	struct page_list *next;

	while (pl) {
		next = pl->next;
		free_pl(pl);
		pl = next;
	}
}

/*
 * Allocate and reserve nr_pages for the use of a specific client.
 */
static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned int nr_pages)
{
	unsigned int i;
	struct page_list *pl = NULL, *next;

	for (i = 0; i < nr_pages; i++) {
		next = alloc_pl(GFP_KERNEL);
		if (!next) {
			if (pl)
				drop_pages(pl);
			return -ENOMEM;
		}
		next->next = pl;
		pl = next;
	}

	kc->nr_reserved_pages += nr_pages;
	kcopyd_put_pages(kc, pl);

	return 0;
}

static void client_free_pages(struct dm_kcopyd_client *kc)
{
	BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
	drop_pages(kc->pages);
	kc->pages = NULL;
	kc->nr_free_pages = kc->nr_reserved_pages = 0;
}

/*-----------------------------------------------------------------
 * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
 * for this reason we use a mempool to prevent the client from
 * ever having to do io (which could cause a deadlock).
 *---------------------------------------------------------------*/
struct kcopyd_job {
	struct dm_kcopyd_client *kc;
	struct list_head list;
	unsigned int flags;

	/*
	 * Error state of the job.
	 */
	int read_err;
	unsigned long write_err;

	/*
	 * REQ_OP_READ, REQ_OP_WRITE or REQ_OP_WRITE_ZEROES.
	 */
	enum req_op op;
	struct dm_io_region source;

	/*
	 * The destinations for the transfer.
	 */
	unsigned int num_dests;
	struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];

	struct page_list *pages;

	/*
	 * Set this to ensure you are notified when the job has
	 * completed.  'context' is for callback to use.
	 */
	dm_kcopyd_notify_fn fn;
	void *context;

	/*
	 * These fields are only used if the job has been split
	 * into more manageable parts.
	 */
	struct mutex lock;
	atomic_t sub_jobs;
	sector_t progress;
	sector_t write_offset;

	struct kcopyd_job *master_job;
};

static struct kmem_cache *_job_cache;

int __init dm_kcopyd_init(void)
{
	_job_cache = kmem_cache_create("kcopyd_job",
				sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
				__alignof__(struct kcopyd_job), 0, NULL);
	if (!_job_cache)
		return -ENOMEM;

	zero_page_list.next = &zero_page_list;
	zero_page_list.page = ZERO_PAGE(0);

	return 0;
}

void dm_kcopyd_exit(void)
{
	kmem_cache_destroy(_job_cache);
	_job_cache = NULL;
}

/*
 * Functions to push and pop a job onto the head of a given job
 * list.
 */
static struct kcopyd_job *pop_io_job(struct list_head *jobs,
				     struct dm_kcopyd_client *kc)
{
	struct kcopyd_job *job;

	/*
	 * For I/O jobs, pop any read, any write without sequential write
	 * constraint and sequential writes that are at the right position.
	 */
	list_for_each_entry(job, jobs, list) {
		if (job->op == REQ_OP_READ ||
		    !(job->flags & BIT(DM_KCOPYD_WRITE_SEQ))) {
			list_del(&job->list);
			return job;
		}

		if (job->write_offset == job->master_job->write_offset) {
			job->master_job->write_offset += job->source.count;
			list_del(&job->list);
			return job;
		}
	}

	return NULL;
}

static struct kcopyd_job *pop(struct list_head *jobs,
			      struct dm_kcopyd_client *kc)
{
	struct kcopyd_job *job = NULL;

	spin_lock_irq(&kc->job_lock);

	if (!list_empty(jobs)) {
		if (jobs == &kc->io_jobs)
			job = pop_io_job(jobs, kc);
		else {
			job = list_entry(jobs->next, struct kcopyd_job, list);
			list_del(&job->list);
		}
	}
	spin_unlock_irq(&kc->job_lock);

	return job;
}

static void push(struct list_head *jobs, struct kcopyd_job *job)
{
	unsigned long flags;
	struct dm_kcopyd_client *kc = job->kc;

	spin_lock_irqsave(&kc->job_lock, flags);
	list_add_tail(&job->list, jobs);
	spin_unlock_irqrestore(&kc->job_lock, flags);
}


static void push_head(struct list_head *jobs, struct kcopyd_job *job)
{
	struct dm_kcopyd_client *kc = job->kc;

	spin_lock_irq(&kc->job_lock);
	list_add(&job->list, jobs);
	spin_unlock_irq(&kc->job_lock);
}

/*
 * These three functions process 1 item from the corresponding
 * job list.
 *
 * They return:
 * < 0: error
 *   0: success
 * > 0: can't process yet.
 */
static int run_complete_job(struct kcopyd_job *job)
{
	void *context = job->context;
	int read_err = job->read_err;
	unsigned long write_err = job->write_err;
	dm_kcopyd_notify_fn fn = job->fn;
	struct dm_kcopyd_client *kc = job->kc;

	if (job->pages && job->pages != &zero_page_list)
		kcopyd_put_pages(kc, job->pages);
	/*
	 * If this is the master job, the sub jobs have already
	 * completed so we can free everything.
	 */
	if (job->master_job == job) {
		mutex_destroy(&job->lock);
		mempool_free(job, &kc->job_pool);
	}
	fn(read_err, write_err, context);

	if (atomic_dec_and_test(&kc->nr_jobs))
		wake_up(&kc->destroyq);

	cond_resched();

	return 0;
}

static void complete_io(unsigned long error, void *context)
{
	struct kcopyd_job *job = (struct kcopyd_job *) context;
	struct dm_kcopyd_client *kc = job->kc;

	io_job_finish(kc->throttle);

	if (error) {
		if (op_is_write(job->op))
			job->write_err |= error;
		else
			job->read_err = 1;

		if (!(job->flags & BIT(DM_KCOPYD_IGNORE_ERROR))) {
			push(&kc->complete_jobs, job);
			wake(kc);
			return;
		}
	}

	if (op_is_write(job->op))
		push(&kc->complete_jobs, job);

	else {
		job->op = REQ_OP_WRITE;
		push(&kc->io_jobs, job);
	}

	wake(kc);
}

/*
 * Request io on as many buffer heads as we can currently get for
 * a particular job.
 */
static int run_io_job(struct kcopyd_job *job)
{
	int r;
	struct dm_io_request io_req = {
		.bi_opf = job->op,
		.mem.type = DM_IO_PAGE_LIST,
		.mem.ptr.pl = job->pages,
		.mem.offset = 0,
		.notify.fn = complete_io,
		.notify.context = job,
		.client = job->kc->io_client,
	};

	/*
	 * If we need to write sequentially and some reads or writes failed,
	 * no point in continuing.
	 */
	if (job->flags & BIT(DM_KCOPYD_WRITE_SEQ) &&
	    job->master_job->write_err) {
		job->write_err = job->master_job->write_err;
		return -EIO;
	}

	io_job_start(job->kc->throttle);

	if (job->op == REQ_OP_READ)
		r = dm_io(&io_req, 1, &job->source, NULL, IOPRIO_DEFAULT);
	else
		r = dm_io(&io_req, job->num_dests, job->dests, NULL, IOPRIO_DEFAULT);

	return r;
}

static int run_pages_job(struct kcopyd_job *job)
{
	int r;
	unsigned int nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);

	r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
	if (!r) {
		/* this job is ready for io */
		push(&job->kc->io_jobs, job);
		return 0;
	}

	if (r == -ENOMEM)
		/* can't complete now */
		return 1;

	return r;
}

/*
 * Run through a list for as long as possible.  Returns the count
 * of successful jobs.
 */
static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
			int (*fn) (struct kcopyd_job *))
{
	struct kcopyd_job *job;
	int r, count = 0;

	while ((job = pop(jobs, kc))) {

		r = fn(job);

		if (r < 0) {
			/* error this rogue job */
			if (op_is_write(job->op))
				job->write_err = (unsigned long) -1L;
			else
				job->read_err = 1;
			push(&kc->complete_jobs, job);
			wake(kc);
			break;
		}

		if (r > 0) {
			/*
			 * We couldn't service this job ATM, so
			 * push this job back onto the list.
			 */
			push_head(jobs, job);
			break;
		}

		count++;
	}

	return count;
}

/*
 * kcopyd does this every time it's woken up.
 */
static void do_work(struct work_struct *work)
{
	struct dm_kcopyd_client *kc = container_of(work,
					struct dm_kcopyd_client, kcopyd_work);
	struct blk_plug plug;

	/*
	 * The order that these are called is *very* important.
	 * complete jobs can free some pages for pages jobs.
	 * Pages jobs when successful will jump onto the io jobs
	 * list.  io jobs call wake when they complete and it all
	 * starts again.
	 */
	spin_lock_irq(&kc->job_lock);
	list_splice_tail_init(&kc->callback_jobs, &kc->complete_jobs);
	spin_unlock_irq(&kc->job_lock);

	blk_start_plug(&plug);
	process_jobs(&kc->complete_jobs, kc, run_complete_job);
	process_jobs(&kc->pages_jobs, kc, run_pages_job);
	process_jobs(&kc->io_jobs, kc, run_io_job);
	blk_finish_plug(&plug);
}

/*
 * If we are copying a small region we just dispatch a single job
 * to do the copy, otherwise the io has to be split up into many
 * jobs.
 */
static void dispatch_job(struct kcopyd_job *job)
{
	struct dm_kcopyd_client *kc = job->kc;
	atomic_inc(&kc->nr_jobs);
	if (unlikely(!job->source.count))
		push(&kc->callback_jobs, job);
	else if (job->pages == &zero_page_list)
		push(&kc->io_jobs, job);
	else
		push(&kc->pages_jobs, job);
	wake(kc);
}

static void segment_complete(int read_err, unsigned long write_err,
			     void *context)
{
	/* FIXME: tidy this function */
	sector_t progress = 0;
	sector_t count = 0;
	struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
	struct kcopyd_job *job = sub_job->master_job;
	struct dm_kcopyd_client *kc = job->kc;

	mutex_lock(&job->lock);

	/* update the error */
	if (read_err)
		job->read_err = 1;

	if (write_err)
		job->write_err |= write_err;

	/*
	 * Only dispatch more work if there hasn't been an error.
	 */
	if ((!job->read_err && !job->write_err) ||
	    job->flags & BIT(DM_KCOPYD_IGNORE_ERROR)) {
		/* get the next chunk of work */
		progress = job->progress;
		count = job->source.count - progress;
		if (count) {
			if (count > kc->sub_job_size)
				count = kc->sub_job_size;

			job->progress += count;
		}
	}
	mutex_unlock(&job->lock);

	if (count) {
		int i;

		*sub_job = *job;
		sub_job->write_offset = progress;
		sub_job->source.sector += progress;
		sub_job->source.count = count;

		for (i = 0; i < job->num_dests; i++) {
			sub_job->dests[i].sector += progress;
			sub_job->dests[i].count = count;
		}

		sub_job->fn = segment_complete;
		sub_job->context = sub_job;
		dispatch_job(sub_job);

	} else if (atomic_dec_and_test(&job->sub_jobs)) {

		/*
		 * Queue the completion callback to the kcopyd thread.
		 *
		 * Some callers assume that all the completions are called
		 * from a single thread and don't race with each other.
		 *
		 * We must not call the callback directly here because this
		 * code may not be executing in the thread.
		 */
		push(&kc->complete_jobs, job);
		wake(kc);
	}
}

/*
 * Create some sub jobs to share the work between them.
 */
static void split_job(struct kcopyd_job *master_job)
{
	int i;

	atomic_inc(&master_job->kc->nr_jobs);

	atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
	for (i = 0; i < SPLIT_COUNT; i++) {
		master_job[i + 1].master_job = master_job;
		segment_complete(0, 0u, &master_job[i + 1]);
	}
}

void dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
		    unsigned int num_dests, struct dm_io_region *dests,
		    unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
{
	struct kcopyd_job *job;
	int i;

	/*
	 * Allocate an array of jobs consisting of one master job
	 * followed by SPLIT_COUNT sub jobs.
	 */
	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
	mutex_init(&job->lock);

	/*
	 * set up for the read.
	 */
	job->kc = kc;
	job->flags = flags;
	job->read_err = 0;
	job->write_err = 0;

	job->num_dests = num_dests;
	memcpy(&job->dests, dests, sizeof(*dests) * num_dests);

	/*
	 * If one of the destination is a host-managed zoned block device,
	 * we need to write sequentially. If one of the destination is a
	 * host-aware device, then leave it to the caller to choose what to do.
	 */
	if (!(job->flags & BIT(DM_KCOPYD_WRITE_SEQ))) {
		for (i = 0; i < job->num_dests; i++) {
			if (bdev_zoned_model(dests[i].bdev) == BLK_ZONED_HM) {
				job->flags |= BIT(DM_KCOPYD_WRITE_SEQ);
				break;
			}
		}
	}

	/*
	 * If we need to write sequentially, errors cannot be ignored.
	 */
	if (job->flags & BIT(DM_KCOPYD_WRITE_SEQ) &&
	    job->flags & BIT(DM_KCOPYD_IGNORE_ERROR))
		job->flags &= ~BIT(DM_KCOPYD_IGNORE_ERROR);

	if (from) {
		job->source = *from;
		job->pages = NULL;
		job->op = REQ_OP_READ;
	} else {
		memset(&job->source, 0, sizeof job->source);
		job->source.count = job->dests[0].count;
		job->pages = &zero_page_list;

		/*
		 * Use WRITE ZEROES to optimize zeroing if all dests support it.
		 */
		job->op = REQ_OP_WRITE_ZEROES;
		for (i = 0; i < job->num_dests; i++)
			if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
				job->op = REQ_OP_WRITE;
				break;
			}
	}

	job->fn = fn;
	job->context = context;
	job->master_job = job;
	job->write_offset = 0;

	if (job->source.count <= kc->sub_job_size)
		dispatch_job(job);
	else {
		job->progress = 0;
		split_job(job);
	}
}
EXPORT_SYMBOL(dm_kcopyd_copy);

void dm_kcopyd_zero(struct dm_kcopyd_client *kc,
		    unsigned int num_dests, struct dm_io_region *dests,
		    unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
{
	dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
}
EXPORT_SYMBOL(dm_kcopyd_zero);

void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
				 dm_kcopyd_notify_fn fn, void *context)
{
	struct kcopyd_job *job;

	job = mempool_alloc(&kc->job_pool, GFP_NOIO);

	memset(job, 0, sizeof(struct kcopyd_job));
	job->kc = kc;
	job->fn = fn;
	job->context = context;
	job->master_job = job;

	atomic_inc(&kc->nr_jobs);

	return job;
}
EXPORT_SYMBOL(dm_kcopyd_prepare_callback);

void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
{
	struct kcopyd_job *job = j;
	struct dm_kcopyd_client *kc = job->kc;

	job->read_err = read_err;
	job->write_err = write_err;

	push(&kc->callback_jobs, job);
	wake(kc);
}
EXPORT_SYMBOL(dm_kcopyd_do_callback);

/*
 * Cancels a kcopyd job, eg. someone might be deactivating a
 * mirror.
 */
#if 0
int kcopyd_cancel(struct kcopyd_job *job, int block)
{
	/* FIXME: finish */
	return -1;
}
#endif  /*  0  */

/*-----------------------------------------------------------------
 * Client setup
 *---------------------------------------------------------------*/
struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
{
	int r;
	unsigned int reserve_pages;
	struct dm_kcopyd_client *kc;

	kc = kzalloc(sizeof(*kc), GFP_KERNEL);
	if (!kc)
		return ERR_PTR(-ENOMEM);

	spin_lock_init(&kc->job_lock);
	INIT_LIST_HEAD(&kc->callback_jobs);
	INIT_LIST_HEAD(&kc->complete_jobs);
	INIT_LIST_HEAD(&kc->io_jobs);
	INIT_LIST_HEAD(&kc->pages_jobs);
	kc->throttle = throttle;

	r = mempool_init_slab_pool(&kc->job_pool, MIN_JOBS, _job_cache);
	if (r)
		goto bad_slab;

	INIT_WORK(&kc->kcopyd_work, do_work);
	kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
	if (!kc->kcopyd_wq) {
		r = -ENOMEM;
		goto bad_workqueue;
	}

	kc->sub_job_size = dm_get_kcopyd_subjob_size();
	reserve_pages = DIV_ROUND_UP(kc->sub_job_size << SECTOR_SHIFT, PAGE_SIZE);

	kc->pages = NULL;
	kc->nr_reserved_pages = kc->nr_free_pages = 0;
	r = client_reserve_pages(kc, reserve_pages);
	if (r)
		goto bad_client_pages;

	kc->io_client = dm_io_client_create();
	if (IS_ERR(kc->io_client)) {
		r = PTR_ERR(kc->io_client);
		goto bad_io_client;
	}

	init_waitqueue_head(&kc->destroyq);
	atomic_set(&kc->nr_jobs, 0);

	return kc;

bad_io_client:
	client_free_pages(kc);
bad_client_pages:
	destroy_workqueue(kc->kcopyd_wq);
bad_workqueue:
	mempool_exit(&kc->job_pool);
bad_slab:
	kfree(kc);

	return ERR_PTR(r);
}
EXPORT_SYMBOL(dm_kcopyd_client_create);

void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
{
	/* Wait for completion of all jobs submitted by this client. */
	wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));

	BUG_ON(!list_empty(&kc->callback_jobs));
	BUG_ON(!list_empty(&kc->complete_jobs));
	BUG_ON(!list_empty(&kc->io_jobs));
	BUG_ON(!list_empty(&kc->pages_jobs));
	destroy_workqueue(kc->kcopyd_wq);
	dm_io_client_destroy(kc->io_client);
	client_free_pages(kc);
	mempool_exit(&kc->job_pool);
	kfree(kc);
}
EXPORT_SYMBOL(dm_kcopyd_client_destroy);

void dm_kcopyd_client_flush(struct dm_kcopyd_client *kc)
{
	flush_workqueue(kc->kcopyd_wq);
}
EXPORT_SYMBOL(dm_kcopyd_client_flush);