PageRenderTime 73ms CodeModel.GetById 19ms app.highlight 47ms RepoModel.GetById 1ms app.codeStats 1ms

/drivers/md/dm-kcopyd.c

http://github.com/mirrors/linux
C | 985 lines | 628 code | 174 blank | 183 comment | 87 complexity | 757622df107bea22d8e771044c00dcca MD5 | raw file
  1/*
  2 * Copyright (C) 2002 Sistina Software (UK) Limited.
  3 * Copyright (C) 2006 Red Hat GmbH
  4 *
  5 * This file is released under the GPL.
  6 *
  7 * Kcopyd provides a simple interface for copying an area of one
  8 * block-device to one or more other block-devices, with an asynchronous
  9 * completion notification.
 10 */
 11
 12#include <linux/types.h>
 13#include <linux/atomic.h>
 14#include <linux/blkdev.h>
 15#include <linux/fs.h>
 16#include <linux/init.h>
 17#include <linux/list.h>
 18#include <linux/mempool.h>
 19#include <linux/module.h>
 20#include <linux/pagemap.h>
 21#include <linux/slab.h>
 22#include <linux/vmalloc.h>
 23#include <linux/workqueue.h>
 24#include <linux/mutex.h>
 25#include <linux/delay.h>
 26#include <linux/device-mapper.h>
 27#include <linux/dm-kcopyd.h>
 28
 29#include "dm-core.h"
 30
 31#define SPLIT_COUNT	8
 32#define MIN_JOBS	8
 33
 34#define DEFAULT_SUB_JOB_SIZE_KB 512
 35#define MAX_SUB_JOB_SIZE_KB     1024
 36
 37static unsigned kcopyd_subjob_size_kb = DEFAULT_SUB_JOB_SIZE_KB;
 38
 39module_param(kcopyd_subjob_size_kb, uint, S_IRUGO | S_IWUSR);
 40MODULE_PARM_DESC(kcopyd_subjob_size_kb, "Sub-job size for dm-kcopyd clients");
 41
 42static unsigned dm_get_kcopyd_subjob_size(void)
 43{
 44	unsigned sub_job_size_kb;
 45
 46	sub_job_size_kb = __dm_get_module_param(&kcopyd_subjob_size_kb,
 47						DEFAULT_SUB_JOB_SIZE_KB,
 48						MAX_SUB_JOB_SIZE_KB);
 49
 50	return sub_job_size_kb << 1;
 51}
 52
 53/*-----------------------------------------------------------------
 54 * Each kcopyd client has its own little pool of preallocated
 55 * pages for kcopyd io.
 56 *---------------------------------------------------------------*/
 57struct dm_kcopyd_client {
 58	struct page_list *pages;
 59	unsigned nr_reserved_pages;
 60	unsigned nr_free_pages;
 61	unsigned sub_job_size;
 62
 63	struct dm_io_client *io_client;
 64
 65	wait_queue_head_t destroyq;
 66
 67	mempool_t job_pool;
 68
 69	struct workqueue_struct *kcopyd_wq;
 70	struct work_struct kcopyd_work;
 71
 72	struct dm_kcopyd_throttle *throttle;
 73
 74	atomic_t nr_jobs;
 75
 76/*
 77 * We maintain four lists of jobs:
 78 *
 79 * i)   jobs waiting for pages
 80 * ii)  jobs that have pages, and are waiting for the io to be issued.
 81 * iii) jobs that don't need to do any IO and just run a callback
 82 * iv) jobs that have completed.
 83 *
 84 * All four of these are protected by job_lock.
 85 */
 86	spinlock_t job_lock;
 87	struct list_head callback_jobs;
 88	struct list_head complete_jobs;
 89	struct list_head io_jobs;
 90	struct list_head pages_jobs;
 91};
 92
 93static struct page_list zero_page_list;
 94
 95static DEFINE_SPINLOCK(throttle_spinlock);
 96
 97/*
 98 * IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
 99 * When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
100 * by 2.
101 */
102#define ACCOUNT_INTERVAL_SHIFT		SHIFT_HZ
103
104/*
105 * Sleep this number of milliseconds.
106 *
107 * The value was decided experimentally.
108 * Smaller values seem to cause an increased copy rate above the limit.
109 * The reason for this is unknown but possibly due to jiffies rounding errors
110 * or read/write cache inside the disk.
111 */
112#define SLEEP_MSEC			100
113
114/*
115 * Maximum number of sleep events. There is a theoretical livelock if more
116 * kcopyd clients do work simultaneously which this limit avoids.
117 */
118#define MAX_SLEEPS			10
119
120static void io_job_start(struct dm_kcopyd_throttle *t)
121{
122	unsigned throttle, now, difference;
123	int slept = 0, skew;
124
125	if (unlikely(!t))
126		return;
127
128try_again:
129	spin_lock_irq(&throttle_spinlock);
130
131	throttle = READ_ONCE(t->throttle);
132
133	if (likely(throttle >= 100))
134		goto skip_limit;
135
136	now = jiffies;
137	difference = now - t->last_jiffies;
138	t->last_jiffies = now;
139	if (t->num_io_jobs)
140		t->io_period += difference;
141	t->total_period += difference;
142
143	/*
144	 * Maintain sane values if we got a temporary overflow.
145	 */
146	if (unlikely(t->io_period > t->total_period))
147		t->io_period = t->total_period;
148
149	if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
150		int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
151		t->total_period >>= shift;
152		t->io_period >>= shift;
153	}
154
155	skew = t->io_period - throttle * t->total_period / 100;
156
157	if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
158		slept++;
159		spin_unlock_irq(&throttle_spinlock);
160		msleep(SLEEP_MSEC);
161		goto try_again;
162	}
163
164skip_limit:
165	t->num_io_jobs++;
166
167	spin_unlock_irq(&throttle_spinlock);
168}
169
170static void io_job_finish(struct dm_kcopyd_throttle *t)
171{
172	unsigned long flags;
173
174	if (unlikely(!t))
175		return;
176
177	spin_lock_irqsave(&throttle_spinlock, flags);
178
179	t->num_io_jobs--;
180
181	if (likely(READ_ONCE(t->throttle) >= 100))
182		goto skip_limit;
183
184	if (!t->num_io_jobs) {
185		unsigned now, difference;
186
187		now = jiffies;
188		difference = now - t->last_jiffies;
189		t->last_jiffies = now;
190
191		t->io_period += difference;
192		t->total_period += difference;
193
194		/*
195		 * Maintain sane values if we got a temporary overflow.
196		 */
197		if (unlikely(t->io_period > t->total_period))
198			t->io_period = t->total_period;
199	}
200
201skip_limit:
202	spin_unlock_irqrestore(&throttle_spinlock, flags);
203}
204
205
206static void wake(struct dm_kcopyd_client *kc)
207{
208	queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
209}
210
211/*
212 * Obtain one page for the use of kcopyd.
213 */
214static struct page_list *alloc_pl(gfp_t gfp)
215{
216	struct page_list *pl;
217
218	pl = kmalloc(sizeof(*pl), gfp);
219	if (!pl)
220		return NULL;
221
222	pl->page = alloc_page(gfp);
223	if (!pl->page) {
224		kfree(pl);
225		return NULL;
226	}
227
228	return pl;
229}
230
231static void free_pl(struct page_list *pl)
232{
233	__free_page(pl->page);
234	kfree(pl);
235}
236
237/*
238 * Add the provided pages to a client's free page list, releasing
239 * back to the system any beyond the reserved_pages limit.
240 */
241static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
242{
243	struct page_list *next;
244
245	do {
246		next = pl->next;
247
248		if (kc->nr_free_pages >= kc->nr_reserved_pages)
249			free_pl(pl);
250		else {
251			pl->next = kc->pages;
252			kc->pages = pl;
253			kc->nr_free_pages++;
254		}
255
256		pl = next;
257	} while (pl);
258}
259
260static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
261			    unsigned int nr, struct page_list **pages)
262{
263	struct page_list *pl;
264
265	*pages = NULL;
266
267	do {
268		pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
269		if (unlikely(!pl)) {
270			/* Use reserved pages */
271			pl = kc->pages;
272			if (unlikely(!pl))
273				goto out_of_memory;
274			kc->pages = pl->next;
275			kc->nr_free_pages--;
276		}
277		pl->next = *pages;
278		*pages = pl;
279	} while (--nr);
280
281	return 0;
282
283out_of_memory:
284	if (*pages)
285		kcopyd_put_pages(kc, *pages);
286	return -ENOMEM;
287}
288
289/*
290 * These three functions resize the page pool.
291 */
292static void drop_pages(struct page_list *pl)
293{
294	struct page_list *next;
295
296	while (pl) {
297		next = pl->next;
298		free_pl(pl);
299		pl = next;
300	}
301}
302
303/*
304 * Allocate and reserve nr_pages for the use of a specific client.
305 */
306static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
307{
308	unsigned i;
309	struct page_list *pl = NULL, *next;
310
311	for (i = 0; i < nr_pages; i++) {
312		next = alloc_pl(GFP_KERNEL);
313		if (!next) {
314			if (pl)
315				drop_pages(pl);
316			return -ENOMEM;
317		}
318		next->next = pl;
319		pl = next;
320	}
321
322	kc->nr_reserved_pages += nr_pages;
323	kcopyd_put_pages(kc, pl);
324
325	return 0;
326}
327
328static void client_free_pages(struct dm_kcopyd_client *kc)
329{
330	BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
331	drop_pages(kc->pages);
332	kc->pages = NULL;
333	kc->nr_free_pages = kc->nr_reserved_pages = 0;
334}
335
336/*-----------------------------------------------------------------
337 * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
338 * for this reason we use a mempool to prevent the client from
339 * ever having to do io (which could cause a deadlock).
340 *---------------------------------------------------------------*/
341struct kcopyd_job {
342	struct dm_kcopyd_client *kc;
343	struct list_head list;
344	unsigned long flags;
345
346	/*
347	 * Error state of the job.
348	 */
349	int read_err;
350	unsigned long write_err;
351
352	/*
353	 * Either READ or WRITE
354	 */
355	int rw;
356	struct dm_io_region source;
357
358	/*
359	 * The destinations for the transfer.
360	 */
361	unsigned int num_dests;
362	struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
363
364	struct page_list *pages;
365
366	/*
367	 * Set this to ensure you are notified when the job has
368	 * completed.  'context' is for callback to use.
369	 */
370	dm_kcopyd_notify_fn fn;
371	void *context;
372
373	/*
374	 * These fields are only used if the job has been split
375	 * into more manageable parts.
376	 */
377	struct mutex lock;
378	atomic_t sub_jobs;
379	sector_t progress;
380	sector_t write_offset;
381
382	struct kcopyd_job *master_job;
383};
384
385static struct kmem_cache *_job_cache;
386
387int __init dm_kcopyd_init(void)
388{
389	_job_cache = kmem_cache_create("kcopyd_job",
390				sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
391				__alignof__(struct kcopyd_job), 0, NULL);
392	if (!_job_cache)
393		return -ENOMEM;
394
395	zero_page_list.next = &zero_page_list;
396	zero_page_list.page = ZERO_PAGE(0);
397
398	return 0;
399}
400
401void dm_kcopyd_exit(void)
402{
403	kmem_cache_destroy(_job_cache);
404	_job_cache = NULL;
405}
406
407/*
408 * Functions to push and pop a job onto the head of a given job
409 * list.
410 */
411static struct kcopyd_job *pop_io_job(struct list_head *jobs,
412				     struct dm_kcopyd_client *kc)
413{
414	struct kcopyd_job *job;
415
416	/*
417	 * For I/O jobs, pop any read, any write without sequential write
418	 * constraint and sequential writes that are at the right position.
419	 */
420	list_for_each_entry(job, jobs, list) {
421		if (job->rw == READ || !test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
422			list_del(&job->list);
423			return job;
424		}
425
426		if (job->write_offset == job->master_job->write_offset) {
427			job->master_job->write_offset += job->source.count;
428			list_del(&job->list);
429			return job;
430		}
431	}
432
433	return NULL;
434}
435
436static struct kcopyd_job *pop(struct list_head *jobs,
437			      struct dm_kcopyd_client *kc)
438{
439	struct kcopyd_job *job = NULL;
440	unsigned long flags;
441
442	spin_lock_irqsave(&kc->job_lock, flags);
443
444	if (!list_empty(jobs)) {
445		if (jobs == &kc->io_jobs)
446			job = pop_io_job(jobs, kc);
447		else {
448			job = list_entry(jobs->next, struct kcopyd_job, list);
449			list_del(&job->list);
450		}
451	}
452	spin_unlock_irqrestore(&kc->job_lock, flags);
453
454	return job;
455}
456
457static void push(struct list_head *jobs, struct kcopyd_job *job)
458{
459	unsigned long flags;
460	struct dm_kcopyd_client *kc = job->kc;
461
462	spin_lock_irqsave(&kc->job_lock, flags);
463	list_add_tail(&job->list, jobs);
464	spin_unlock_irqrestore(&kc->job_lock, flags);
465}
466
467
468static void push_head(struct list_head *jobs, struct kcopyd_job *job)
469{
470	unsigned long flags;
471	struct dm_kcopyd_client *kc = job->kc;
472
473	spin_lock_irqsave(&kc->job_lock, flags);
474	list_add(&job->list, jobs);
475	spin_unlock_irqrestore(&kc->job_lock, flags);
476}
477
478/*
479 * These three functions process 1 item from the corresponding
480 * job list.
481 *
482 * They return:
483 * < 0: error
484 *   0: success
485 * > 0: can't process yet.
486 */
487static int run_complete_job(struct kcopyd_job *job)
488{
489	void *context = job->context;
490	int read_err = job->read_err;
491	unsigned long write_err = job->write_err;
492	dm_kcopyd_notify_fn fn = job->fn;
493	struct dm_kcopyd_client *kc = job->kc;
494
495	if (job->pages && job->pages != &zero_page_list)
496		kcopyd_put_pages(kc, job->pages);
497	/*
498	 * If this is the master job, the sub jobs have already
499	 * completed so we can free everything.
500	 */
501	if (job->master_job == job) {
502		mutex_destroy(&job->lock);
503		mempool_free(job, &kc->job_pool);
504	}
505	fn(read_err, write_err, context);
506
507	if (atomic_dec_and_test(&kc->nr_jobs))
508		wake_up(&kc->destroyq);
509
510	cond_resched();
511
512	return 0;
513}
514
515static void complete_io(unsigned long error, void *context)
516{
517	struct kcopyd_job *job = (struct kcopyd_job *) context;
518	struct dm_kcopyd_client *kc = job->kc;
519
520	io_job_finish(kc->throttle);
521
522	if (error) {
523		if (op_is_write(job->rw))
524			job->write_err |= error;
525		else
526			job->read_err = 1;
527
528		if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
529			push(&kc->complete_jobs, job);
530			wake(kc);
531			return;
532		}
533	}
534
535	if (op_is_write(job->rw))
536		push(&kc->complete_jobs, job);
537
538	else {
539		job->rw = WRITE;
540		push(&kc->io_jobs, job);
541	}
542
543	wake(kc);
544}
545
546/*
547 * Request io on as many buffer heads as we can currently get for
548 * a particular job.
549 */
550static int run_io_job(struct kcopyd_job *job)
551{
552	int r;
553	struct dm_io_request io_req = {
554		.bi_op = job->rw,
555		.bi_op_flags = 0,
556		.mem.type = DM_IO_PAGE_LIST,
557		.mem.ptr.pl = job->pages,
558		.mem.offset = 0,
559		.notify.fn = complete_io,
560		.notify.context = job,
561		.client = job->kc->io_client,
562	};
563
564	/*
565	 * If we need to write sequentially and some reads or writes failed,
566	 * no point in continuing.
567	 */
568	if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
569	    job->master_job->write_err) {
570		job->write_err = job->master_job->write_err;
571		return -EIO;
572	}
573
574	io_job_start(job->kc->throttle);
575
576	if (job->rw == READ)
577		r = dm_io(&io_req, 1, &job->source, NULL);
578	else
579		r = dm_io(&io_req, job->num_dests, job->dests, NULL);
580
581	return r;
582}
583
584static int run_pages_job(struct kcopyd_job *job)
585{
586	int r;
587	unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
588
589	r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
590	if (!r) {
591		/* this job is ready for io */
592		push(&job->kc->io_jobs, job);
593		return 0;
594	}
595
596	if (r == -ENOMEM)
597		/* can't complete now */
598		return 1;
599
600	return r;
601}
602
603/*
604 * Run through a list for as long as possible.  Returns the count
605 * of successful jobs.
606 */
607static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
608			int (*fn) (struct kcopyd_job *))
609{
610	struct kcopyd_job *job;
611	int r, count = 0;
612
613	while ((job = pop(jobs, kc))) {
614
615		r = fn(job);
616
617		if (r < 0) {
618			/* error this rogue job */
619			if (op_is_write(job->rw))
620				job->write_err = (unsigned long) -1L;
621			else
622				job->read_err = 1;
623			push(&kc->complete_jobs, job);
624			wake(kc);
625			break;
626		}
627
628		if (r > 0) {
629			/*
630			 * We couldn't service this job ATM, so
631			 * push this job back onto the list.
632			 */
633			push_head(jobs, job);
634			break;
635		}
636
637		count++;
638	}
639
640	return count;
641}
642
643/*
644 * kcopyd does this every time it's woken up.
645 */
646static void do_work(struct work_struct *work)
647{
648	struct dm_kcopyd_client *kc = container_of(work,
649					struct dm_kcopyd_client, kcopyd_work);
650	struct blk_plug plug;
651	unsigned long flags;
652
653	/*
654	 * The order that these are called is *very* important.
655	 * complete jobs can free some pages for pages jobs.
656	 * Pages jobs when successful will jump onto the io jobs
657	 * list.  io jobs call wake when they complete and it all
658	 * starts again.
659	 */
660	spin_lock_irqsave(&kc->job_lock, flags);
661	list_splice_tail_init(&kc->callback_jobs, &kc->complete_jobs);
662	spin_unlock_irqrestore(&kc->job_lock, flags);
663
664	blk_start_plug(&plug);
665	process_jobs(&kc->complete_jobs, kc, run_complete_job);
666	process_jobs(&kc->pages_jobs, kc, run_pages_job);
667	process_jobs(&kc->io_jobs, kc, run_io_job);
668	blk_finish_plug(&plug);
669}
670
671/*
672 * If we are copying a small region we just dispatch a single job
673 * to do the copy, otherwise the io has to be split up into many
674 * jobs.
675 */
676static void dispatch_job(struct kcopyd_job *job)
677{
678	struct dm_kcopyd_client *kc = job->kc;
679	atomic_inc(&kc->nr_jobs);
680	if (unlikely(!job->source.count))
681		push(&kc->callback_jobs, job);
682	else if (job->pages == &zero_page_list)
683		push(&kc->io_jobs, job);
684	else
685		push(&kc->pages_jobs, job);
686	wake(kc);
687}
688
689static void segment_complete(int read_err, unsigned long write_err,
690			     void *context)
691{
692	/* FIXME: tidy this function */
693	sector_t progress = 0;
694	sector_t count = 0;
695	struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
696	struct kcopyd_job *job = sub_job->master_job;
697	struct dm_kcopyd_client *kc = job->kc;
698
699	mutex_lock(&job->lock);
700
701	/* update the error */
702	if (read_err)
703		job->read_err = 1;
704
705	if (write_err)
706		job->write_err |= write_err;
707
708	/*
709	 * Only dispatch more work if there hasn't been an error.
710	 */
711	if ((!job->read_err && !job->write_err) ||
712	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
713		/* get the next chunk of work */
714		progress = job->progress;
715		count = job->source.count - progress;
716		if (count) {
717			if (count > kc->sub_job_size)
718				count = kc->sub_job_size;
719
720			job->progress += count;
721		}
722	}
723	mutex_unlock(&job->lock);
724
725	if (count) {
726		int i;
727
728		*sub_job = *job;
729		sub_job->write_offset = progress;
730		sub_job->source.sector += progress;
731		sub_job->source.count = count;
732
733		for (i = 0; i < job->num_dests; i++) {
734			sub_job->dests[i].sector += progress;
735			sub_job->dests[i].count = count;
736		}
737
738		sub_job->fn = segment_complete;
739		sub_job->context = sub_job;
740		dispatch_job(sub_job);
741
742	} else if (atomic_dec_and_test(&job->sub_jobs)) {
743
744		/*
745		 * Queue the completion callback to the kcopyd thread.
746		 *
747		 * Some callers assume that all the completions are called
748		 * from a single thread and don't race with each other.
749		 *
750		 * We must not call the callback directly here because this
751		 * code may not be executing in the thread.
752		 */
753		push(&kc->complete_jobs, job);
754		wake(kc);
755	}
756}
757
758/*
759 * Create some sub jobs to share the work between them.
760 */
761static void split_job(struct kcopyd_job *master_job)
762{
763	int i;
764
765	atomic_inc(&master_job->kc->nr_jobs);
766
767	atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
768	for (i = 0; i < SPLIT_COUNT; i++) {
769		master_job[i + 1].master_job = master_job;
770		segment_complete(0, 0u, &master_job[i + 1]);
771	}
772}
773
774void dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
775		    unsigned int num_dests, struct dm_io_region *dests,
776		    unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
777{
778	struct kcopyd_job *job;
779	int i;
780
781	/*
782	 * Allocate an array of jobs consisting of one master job
783	 * followed by SPLIT_COUNT sub jobs.
784	 */
785	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
786	mutex_init(&job->lock);
787
788	/*
789	 * set up for the read.
790	 */
791	job->kc = kc;
792	job->flags = flags;
793	job->read_err = 0;
794	job->write_err = 0;
795
796	job->num_dests = num_dests;
797	memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
798
799	/*
800	 * If one of the destination is a host-managed zoned block device,
801	 * we need to write sequentially. If one of the destination is a
802	 * host-aware device, then leave it to the caller to choose what to do.
803	 */
804	if (!test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
805		for (i = 0; i < job->num_dests; i++) {
806			if (bdev_zoned_model(dests[i].bdev) == BLK_ZONED_HM) {
807				set_bit(DM_KCOPYD_WRITE_SEQ, &job->flags);
808				break;
809			}
810		}
811	}
812
813	/*
814	 * If we need to write sequentially, errors cannot be ignored.
815	 */
816	if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
817	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags))
818		clear_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags);
819
820	if (from) {
821		job->source = *from;
822		job->pages = NULL;
823		job->rw = READ;
824	} else {
825		memset(&job->source, 0, sizeof job->source);
826		job->source.count = job->dests[0].count;
827		job->pages = &zero_page_list;
828
829		/*
830		 * Use WRITE ZEROES to optimize zeroing if all dests support it.
831		 */
832		job->rw = REQ_OP_WRITE_ZEROES;
833		for (i = 0; i < job->num_dests; i++)
834			if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
835				job->rw = WRITE;
836				break;
837			}
838	}
839
840	job->fn = fn;
841	job->context = context;
842	job->master_job = job;
843	job->write_offset = 0;
844
845	if (job->source.count <= kc->sub_job_size)
846		dispatch_job(job);
847	else {
848		job->progress = 0;
849		split_job(job);
850	}
851}
852EXPORT_SYMBOL(dm_kcopyd_copy);
853
854void dm_kcopyd_zero(struct dm_kcopyd_client *kc,
855		    unsigned num_dests, struct dm_io_region *dests,
856		    unsigned flags, dm_kcopyd_notify_fn fn, void *context)
857{
858	dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
859}
860EXPORT_SYMBOL(dm_kcopyd_zero);
861
862void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
863				 dm_kcopyd_notify_fn fn, void *context)
864{
865	struct kcopyd_job *job;
866
867	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
868
869	memset(job, 0, sizeof(struct kcopyd_job));
870	job->kc = kc;
871	job->fn = fn;
872	job->context = context;
873	job->master_job = job;
874
875	atomic_inc(&kc->nr_jobs);
876
877	return job;
878}
879EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
880
881void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
882{
883	struct kcopyd_job *job = j;
884	struct dm_kcopyd_client *kc = job->kc;
885
886	job->read_err = read_err;
887	job->write_err = write_err;
888
889	push(&kc->callback_jobs, job);
890	wake(kc);
891}
892EXPORT_SYMBOL(dm_kcopyd_do_callback);
893
894/*
895 * Cancels a kcopyd job, eg. someone might be deactivating a
896 * mirror.
897 */
898#if 0
899int kcopyd_cancel(struct kcopyd_job *job, int block)
900{
901	/* FIXME: finish */
902	return -1;
903}
904#endif  /*  0  */
905
906/*-----------------------------------------------------------------
907 * Client setup
908 *---------------------------------------------------------------*/
909struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
910{
911	int r;
912	unsigned reserve_pages;
913	struct dm_kcopyd_client *kc;
914
915	kc = kzalloc(sizeof(*kc), GFP_KERNEL);
916	if (!kc)
917		return ERR_PTR(-ENOMEM);
918
919	spin_lock_init(&kc->job_lock);
920	INIT_LIST_HEAD(&kc->callback_jobs);
921	INIT_LIST_HEAD(&kc->complete_jobs);
922	INIT_LIST_HEAD(&kc->io_jobs);
923	INIT_LIST_HEAD(&kc->pages_jobs);
924	kc->throttle = throttle;
925
926	r = mempool_init_slab_pool(&kc->job_pool, MIN_JOBS, _job_cache);
927	if (r)
928		goto bad_slab;
929
930	INIT_WORK(&kc->kcopyd_work, do_work);
931	kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
932	if (!kc->kcopyd_wq) {
933		r = -ENOMEM;
934		goto bad_workqueue;
935	}
936
937	kc->sub_job_size = dm_get_kcopyd_subjob_size();
938	reserve_pages = DIV_ROUND_UP(kc->sub_job_size << SECTOR_SHIFT, PAGE_SIZE);
939
940	kc->pages = NULL;
941	kc->nr_reserved_pages = kc->nr_free_pages = 0;
942	r = client_reserve_pages(kc, reserve_pages);
943	if (r)
944		goto bad_client_pages;
945
946	kc->io_client = dm_io_client_create();
947	if (IS_ERR(kc->io_client)) {
948		r = PTR_ERR(kc->io_client);
949		goto bad_io_client;
950	}
951
952	init_waitqueue_head(&kc->destroyq);
953	atomic_set(&kc->nr_jobs, 0);
954
955	return kc;
956
957bad_io_client:
958	client_free_pages(kc);
959bad_client_pages:
960	destroy_workqueue(kc->kcopyd_wq);
961bad_workqueue:
962	mempool_exit(&kc->job_pool);
963bad_slab:
964	kfree(kc);
965
966	return ERR_PTR(r);
967}
968EXPORT_SYMBOL(dm_kcopyd_client_create);
969
970void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
971{
972	/* Wait for completion of all jobs submitted by this client. */
973	wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
974
975	BUG_ON(!list_empty(&kc->callback_jobs));
976	BUG_ON(!list_empty(&kc->complete_jobs));
977	BUG_ON(!list_empty(&kc->io_jobs));
978	BUG_ON(!list_empty(&kc->pages_jobs));
979	destroy_workqueue(kc->kcopyd_wq);
980	dm_io_client_destroy(kc->io_client);
981	client_free_pages(kc);
982	mempool_exit(&kc->job_pool);
983	kfree(kc);
984}
985EXPORT_SYMBOL(dm_kcopyd_client_destroy);