xref: /openbmc/linux/drivers/md/dm-kcopyd.c (revision 93f5715e)
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 SUB_JOB_SIZE	128
32 #define SPLIT_COUNT	8
33 #define MIN_JOBS	8
34 #define RESERVE_PAGES	(DIV_ROUND_UP(SUB_JOB_SIZE << SECTOR_SHIFT, PAGE_SIZE))
35 
36 /*-----------------------------------------------------------------
37  * Each kcopyd client has its own little pool of preallocated
38  * pages for kcopyd io.
39  *---------------------------------------------------------------*/
40 struct dm_kcopyd_client {
41 	struct page_list *pages;
42 	unsigned nr_reserved_pages;
43 	unsigned nr_free_pages;
44 
45 	struct dm_io_client *io_client;
46 
47 	wait_queue_head_t destroyq;
48 
49 	mempool_t job_pool;
50 
51 	struct workqueue_struct *kcopyd_wq;
52 	struct work_struct kcopyd_work;
53 
54 	struct dm_kcopyd_throttle *throttle;
55 
56 	atomic_t nr_jobs;
57 
58 /*
59  * We maintain three lists of jobs:
60  *
61  * i)   jobs waiting for pages
62  * ii)  jobs that have pages, and are waiting for the io to be issued.
63  * iii) jobs that have completed.
64  *
65  * All three of these are protected by job_lock.
66  */
67 	spinlock_t job_lock;
68 	struct list_head complete_jobs;
69 	struct list_head io_jobs;
70 	struct list_head pages_jobs;
71 };
72 
73 static struct page_list zero_page_list;
74 
75 static DEFINE_SPINLOCK(throttle_spinlock);
76 
77 /*
78  * IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
79  * When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
80  * by 2.
81  */
82 #define ACCOUNT_INTERVAL_SHIFT		SHIFT_HZ
83 
84 /*
85  * Sleep this number of milliseconds.
86  *
87  * The value was decided experimentally.
88  * Smaller values seem to cause an increased copy rate above the limit.
89  * The reason for this is unknown but possibly due to jiffies rounding errors
90  * or read/write cache inside the disk.
91  */
92 #define SLEEP_MSEC			100
93 
94 /*
95  * Maximum number of sleep events. There is a theoretical livelock if more
96  * kcopyd clients do work simultaneously which this limit avoids.
97  */
98 #define MAX_SLEEPS			10
99 
100 static void io_job_start(struct dm_kcopyd_throttle *t)
101 {
102 	unsigned throttle, now, difference;
103 	int slept = 0, skew;
104 
105 	if (unlikely(!t))
106 		return;
107 
108 try_again:
109 	spin_lock_irq(&throttle_spinlock);
110 
111 	throttle = READ_ONCE(t->throttle);
112 
113 	if (likely(throttle >= 100))
114 		goto skip_limit;
115 
116 	now = jiffies;
117 	difference = now - t->last_jiffies;
118 	t->last_jiffies = now;
119 	if (t->num_io_jobs)
120 		t->io_period += difference;
121 	t->total_period += difference;
122 
123 	/*
124 	 * Maintain sane values if we got a temporary overflow.
125 	 */
126 	if (unlikely(t->io_period > t->total_period))
127 		t->io_period = t->total_period;
128 
129 	if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
130 		int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
131 		t->total_period >>= shift;
132 		t->io_period >>= shift;
133 	}
134 
135 	skew = t->io_period - throttle * t->total_period / 100;
136 
137 	if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
138 		slept++;
139 		spin_unlock_irq(&throttle_spinlock);
140 		msleep(SLEEP_MSEC);
141 		goto try_again;
142 	}
143 
144 skip_limit:
145 	t->num_io_jobs++;
146 
147 	spin_unlock_irq(&throttle_spinlock);
148 }
149 
150 static void io_job_finish(struct dm_kcopyd_throttle *t)
151 {
152 	unsigned long flags;
153 
154 	if (unlikely(!t))
155 		return;
156 
157 	spin_lock_irqsave(&throttle_spinlock, flags);
158 
159 	t->num_io_jobs--;
160 
161 	if (likely(READ_ONCE(t->throttle) >= 100))
162 		goto skip_limit;
163 
164 	if (!t->num_io_jobs) {
165 		unsigned now, difference;
166 
167 		now = jiffies;
168 		difference = now - t->last_jiffies;
169 		t->last_jiffies = now;
170 
171 		t->io_period += difference;
172 		t->total_period += difference;
173 
174 		/*
175 		 * Maintain sane values if we got a temporary overflow.
176 		 */
177 		if (unlikely(t->io_period > t->total_period))
178 			t->io_period = t->total_period;
179 	}
180 
181 skip_limit:
182 	spin_unlock_irqrestore(&throttle_spinlock, flags);
183 }
184 
185 
186 static void wake(struct dm_kcopyd_client *kc)
187 {
188 	queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
189 }
190 
191 /*
192  * Obtain one page for the use of kcopyd.
193  */
194 static struct page_list *alloc_pl(gfp_t gfp)
195 {
196 	struct page_list *pl;
197 
198 	pl = kmalloc(sizeof(*pl), gfp);
199 	if (!pl)
200 		return NULL;
201 
202 	pl->page = alloc_page(gfp);
203 	if (!pl->page) {
204 		kfree(pl);
205 		return NULL;
206 	}
207 
208 	return pl;
209 }
210 
211 static void free_pl(struct page_list *pl)
212 {
213 	__free_page(pl->page);
214 	kfree(pl);
215 }
216 
217 /*
218  * Add the provided pages to a client's free page list, releasing
219  * back to the system any beyond the reserved_pages limit.
220  */
221 static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
222 {
223 	struct page_list *next;
224 
225 	do {
226 		next = pl->next;
227 
228 		if (kc->nr_free_pages >= kc->nr_reserved_pages)
229 			free_pl(pl);
230 		else {
231 			pl->next = kc->pages;
232 			kc->pages = pl;
233 			kc->nr_free_pages++;
234 		}
235 
236 		pl = next;
237 	} while (pl);
238 }
239 
240 static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
241 			    unsigned int nr, struct page_list **pages)
242 {
243 	struct page_list *pl;
244 
245 	*pages = NULL;
246 
247 	do {
248 		pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
249 		if (unlikely(!pl)) {
250 			/* Use reserved pages */
251 			pl = kc->pages;
252 			if (unlikely(!pl))
253 				goto out_of_memory;
254 			kc->pages = pl->next;
255 			kc->nr_free_pages--;
256 		}
257 		pl->next = *pages;
258 		*pages = pl;
259 	} while (--nr);
260 
261 	return 0;
262 
263 out_of_memory:
264 	if (*pages)
265 		kcopyd_put_pages(kc, *pages);
266 	return -ENOMEM;
267 }
268 
269 /*
270  * These three functions resize the page pool.
271  */
272 static void drop_pages(struct page_list *pl)
273 {
274 	struct page_list *next;
275 
276 	while (pl) {
277 		next = pl->next;
278 		free_pl(pl);
279 		pl = next;
280 	}
281 }
282 
283 /*
284  * Allocate and reserve nr_pages for the use of a specific client.
285  */
286 static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
287 {
288 	unsigned i;
289 	struct page_list *pl = NULL, *next;
290 
291 	for (i = 0; i < nr_pages; i++) {
292 		next = alloc_pl(GFP_KERNEL);
293 		if (!next) {
294 			if (pl)
295 				drop_pages(pl);
296 			return -ENOMEM;
297 		}
298 		next->next = pl;
299 		pl = next;
300 	}
301 
302 	kc->nr_reserved_pages += nr_pages;
303 	kcopyd_put_pages(kc, pl);
304 
305 	return 0;
306 }
307 
308 static void client_free_pages(struct dm_kcopyd_client *kc)
309 {
310 	BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
311 	drop_pages(kc->pages);
312 	kc->pages = NULL;
313 	kc->nr_free_pages = kc->nr_reserved_pages = 0;
314 }
315 
316 /*-----------------------------------------------------------------
317  * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
318  * for this reason we use a mempool to prevent the client from
319  * ever having to do io (which could cause a deadlock).
320  *---------------------------------------------------------------*/
321 struct kcopyd_job {
322 	struct dm_kcopyd_client *kc;
323 	struct list_head list;
324 	unsigned long flags;
325 
326 	/*
327 	 * Error state of the job.
328 	 */
329 	int read_err;
330 	unsigned long write_err;
331 
332 	/*
333 	 * Either READ or WRITE
334 	 */
335 	int rw;
336 	struct dm_io_region source;
337 
338 	/*
339 	 * The destinations for the transfer.
340 	 */
341 	unsigned int num_dests;
342 	struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
343 
344 	struct page_list *pages;
345 
346 	/*
347 	 * Set this to ensure you are notified when the job has
348 	 * completed.  'context' is for callback to use.
349 	 */
350 	dm_kcopyd_notify_fn fn;
351 	void *context;
352 
353 	/*
354 	 * These fields are only used if the job has been split
355 	 * into more manageable parts.
356 	 */
357 	struct mutex lock;
358 	atomic_t sub_jobs;
359 	sector_t progress;
360 	sector_t write_offset;
361 
362 	struct kcopyd_job *master_job;
363 };
364 
365 static struct kmem_cache *_job_cache;
366 
367 int __init dm_kcopyd_init(void)
368 {
369 	_job_cache = kmem_cache_create("kcopyd_job",
370 				sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
371 				__alignof__(struct kcopyd_job), 0, NULL);
372 	if (!_job_cache)
373 		return -ENOMEM;
374 
375 	zero_page_list.next = &zero_page_list;
376 	zero_page_list.page = ZERO_PAGE(0);
377 
378 	return 0;
379 }
380 
381 void dm_kcopyd_exit(void)
382 {
383 	kmem_cache_destroy(_job_cache);
384 	_job_cache = NULL;
385 }
386 
387 /*
388  * Functions to push and pop a job onto the head of a given job
389  * list.
390  */
391 static struct kcopyd_job *pop_io_job(struct list_head *jobs,
392 				     struct dm_kcopyd_client *kc)
393 {
394 	struct kcopyd_job *job;
395 
396 	/*
397 	 * For I/O jobs, pop any read, any write without sequential write
398 	 * constraint and sequential writes that are at the right position.
399 	 */
400 	list_for_each_entry(job, jobs, list) {
401 		if (job->rw == READ || !test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
402 			list_del(&job->list);
403 			return job;
404 		}
405 
406 		if (job->write_offset == job->master_job->write_offset) {
407 			job->master_job->write_offset += job->source.count;
408 			list_del(&job->list);
409 			return job;
410 		}
411 	}
412 
413 	return NULL;
414 }
415 
416 static struct kcopyd_job *pop(struct list_head *jobs,
417 			      struct dm_kcopyd_client *kc)
418 {
419 	struct kcopyd_job *job = NULL;
420 	unsigned long flags;
421 
422 	spin_lock_irqsave(&kc->job_lock, flags);
423 
424 	if (!list_empty(jobs)) {
425 		if (jobs == &kc->io_jobs)
426 			job = pop_io_job(jobs, kc);
427 		else {
428 			job = list_entry(jobs->next, struct kcopyd_job, list);
429 			list_del(&job->list);
430 		}
431 	}
432 	spin_unlock_irqrestore(&kc->job_lock, flags);
433 
434 	return job;
435 }
436 
437 static void push(struct list_head *jobs, struct kcopyd_job *job)
438 {
439 	unsigned long flags;
440 	struct dm_kcopyd_client *kc = job->kc;
441 
442 	spin_lock_irqsave(&kc->job_lock, flags);
443 	list_add_tail(&job->list, jobs);
444 	spin_unlock_irqrestore(&kc->job_lock, flags);
445 }
446 
447 
448 static void push_head(struct list_head *jobs, struct kcopyd_job *job)
449 {
450 	unsigned long flags;
451 	struct dm_kcopyd_client *kc = job->kc;
452 
453 	spin_lock_irqsave(&kc->job_lock, flags);
454 	list_add(&job->list, jobs);
455 	spin_unlock_irqrestore(&kc->job_lock, flags);
456 }
457 
458 /*
459  * These three functions process 1 item from the corresponding
460  * job list.
461  *
462  * They return:
463  * < 0: error
464  *   0: success
465  * > 0: can't process yet.
466  */
467 static int run_complete_job(struct kcopyd_job *job)
468 {
469 	void *context = job->context;
470 	int read_err = job->read_err;
471 	unsigned long write_err = job->write_err;
472 	dm_kcopyd_notify_fn fn = job->fn;
473 	struct dm_kcopyd_client *kc = job->kc;
474 
475 	if (job->pages && job->pages != &zero_page_list)
476 		kcopyd_put_pages(kc, job->pages);
477 	/*
478 	 * If this is the master job, the sub jobs have already
479 	 * completed so we can free everything.
480 	 */
481 	if (job->master_job == job) {
482 		mutex_destroy(&job->lock);
483 		mempool_free(job, &kc->job_pool);
484 	}
485 	fn(read_err, write_err, context);
486 
487 	if (atomic_dec_and_test(&kc->nr_jobs))
488 		wake_up(&kc->destroyq);
489 
490 	cond_resched();
491 
492 	return 0;
493 }
494 
495 static void complete_io(unsigned long error, void *context)
496 {
497 	struct kcopyd_job *job = (struct kcopyd_job *) context;
498 	struct dm_kcopyd_client *kc = job->kc;
499 
500 	io_job_finish(kc->throttle);
501 
502 	if (error) {
503 		if (op_is_write(job->rw))
504 			job->write_err |= error;
505 		else
506 			job->read_err = 1;
507 
508 		if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
509 			push(&kc->complete_jobs, job);
510 			wake(kc);
511 			return;
512 		}
513 	}
514 
515 	if (op_is_write(job->rw))
516 		push(&kc->complete_jobs, job);
517 
518 	else {
519 		job->rw = WRITE;
520 		push(&kc->io_jobs, job);
521 	}
522 
523 	wake(kc);
524 }
525 
526 /*
527  * Request io on as many buffer heads as we can currently get for
528  * a particular job.
529  */
530 static int run_io_job(struct kcopyd_job *job)
531 {
532 	int r;
533 	struct dm_io_request io_req = {
534 		.bi_op = job->rw,
535 		.bi_op_flags = 0,
536 		.mem.type = DM_IO_PAGE_LIST,
537 		.mem.ptr.pl = job->pages,
538 		.mem.offset = 0,
539 		.notify.fn = complete_io,
540 		.notify.context = job,
541 		.client = job->kc->io_client,
542 	};
543 
544 	/*
545 	 * If we need to write sequentially and some reads or writes failed,
546 	 * no point in continuing.
547 	 */
548 	if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
549 	    job->master_job->write_err)
550 		return -EIO;
551 
552 	io_job_start(job->kc->throttle);
553 
554 	if (job->rw == READ)
555 		r = dm_io(&io_req, 1, &job->source, NULL);
556 	else
557 		r = dm_io(&io_req, job->num_dests, job->dests, NULL);
558 
559 	return r;
560 }
561 
562 static int run_pages_job(struct kcopyd_job *job)
563 {
564 	int r;
565 	unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
566 
567 	r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
568 	if (!r) {
569 		/* this job is ready for io */
570 		push(&job->kc->io_jobs, job);
571 		return 0;
572 	}
573 
574 	if (r == -ENOMEM)
575 		/* can't complete now */
576 		return 1;
577 
578 	return r;
579 }
580 
581 /*
582  * Run through a list for as long as possible.  Returns the count
583  * of successful jobs.
584  */
585 static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
586 			int (*fn) (struct kcopyd_job *))
587 {
588 	struct kcopyd_job *job;
589 	int r, count = 0;
590 
591 	while ((job = pop(jobs, kc))) {
592 
593 		r = fn(job);
594 
595 		if (r < 0) {
596 			/* error this rogue job */
597 			if (op_is_write(job->rw))
598 				job->write_err = (unsigned long) -1L;
599 			else
600 				job->read_err = 1;
601 			push(&kc->complete_jobs, job);
602 			break;
603 		}
604 
605 		if (r > 0) {
606 			/*
607 			 * We couldn't service this job ATM, so
608 			 * push this job back onto the list.
609 			 */
610 			push_head(jobs, job);
611 			break;
612 		}
613 
614 		count++;
615 	}
616 
617 	return count;
618 }
619 
620 /*
621  * kcopyd does this every time it's woken up.
622  */
623 static void do_work(struct work_struct *work)
624 {
625 	struct dm_kcopyd_client *kc = container_of(work,
626 					struct dm_kcopyd_client, kcopyd_work);
627 	struct blk_plug plug;
628 
629 	/*
630 	 * The order that these are called is *very* important.
631 	 * complete jobs can free some pages for pages jobs.
632 	 * Pages jobs when successful will jump onto the io jobs
633 	 * list.  io jobs call wake when they complete and it all
634 	 * starts again.
635 	 */
636 	blk_start_plug(&plug);
637 	process_jobs(&kc->complete_jobs, kc, run_complete_job);
638 	process_jobs(&kc->pages_jobs, kc, run_pages_job);
639 	process_jobs(&kc->io_jobs, kc, run_io_job);
640 	blk_finish_plug(&plug);
641 }
642 
643 /*
644  * If we are copying a small region we just dispatch a single job
645  * to do the copy, otherwise the io has to be split up into many
646  * jobs.
647  */
648 static void dispatch_job(struct kcopyd_job *job)
649 {
650 	struct dm_kcopyd_client *kc = job->kc;
651 	atomic_inc(&kc->nr_jobs);
652 	if (unlikely(!job->source.count))
653 		push(&kc->complete_jobs, job);
654 	else if (job->pages == &zero_page_list)
655 		push(&kc->io_jobs, job);
656 	else
657 		push(&kc->pages_jobs, job);
658 	wake(kc);
659 }
660 
661 static void segment_complete(int read_err, unsigned long write_err,
662 			     void *context)
663 {
664 	/* FIXME: tidy this function */
665 	sector_t progress = 0;
666 	sector_t count = 0;
667 	struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
668 	struct kcopyd_job *job = sub_job->master_job;
669 	struct dm_kcopyd_client *kc = job->kc;
670 
671 	mutex_lock(&job->lock);
672 
673 	/* update the error */
674 	if (read_err)
675 		job->read_err = 1;
676 
677 	if (write_err)
678 		job->write_err |= write_err;
679 
680 	/*
681 	 * Only dispatch more work if there hasn't been an error.
682 	 */
683 	if ((!job->read_err && !job->write_err) ||
684 	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
685 		/* get the next chunk of work */
686 		progress = job->progress;
687 		count = job->source.count - progress;
688 		if (count) {
689 			if (count > SUB_JOB_SIZE)
690 				count = SUB_JOB_SIZE;
691 
692 			job->progress += count;
693 		}
694 	}
695 	mutex_unlock(&job->lock);
696 
697 	if (count) {
698 		int i;
699 
700 		*sub_job = *job;
701 		sub_job->write_offset = progress;
702 		sub_job->source.sector += progress;
703 		sub_job->source.count = count;
704 
705 		for (i = 0; i < job->num_dests; i++) {
706 			sub_job->dests[i].sector += progress;
707 			sub_job->dests[i].count = count;
708 		}
709 
710 		sub_job->fn = segment_complete;
711 		sub_job->context = sub_job;
712 		dispatch_job(sub_job);
713 
714 	} else if (atomic_dec_and_test(&job->sub_jobs)) {
715 
716 		/*
717 		 * Queue the completion callback to the kcopyd thread.
718 		 *
719 		 * Some callers assume that all the completions are called
720 		 * from a single thread and don't race with each other.
721 		 *
722 		 * We must not call the callback directly here because this
723 		 * code may not be executing in the thread.
724 		 */
725 		push(&kc->complete_jobs, job);
726 		wake(kc);
727 	}
728 }
729 
730 /*
731  * Create some sub jobs to share the work between them.
732  */
733 static void split_job(struct kcopyd_job *master_job)
734 {
735 	int i;
736 
737 	atomic_inc(&master_job->kc->nr_jobs);
738 
739 	atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
740 	for (i = 0; i < SPLIT_COUNT; i++) {
741 		master_job[i + 1].master_job = master_job;
742 		segment_complete(0, 0u, &master_job[i + 1]);
743 	}
744 }
745 
746 void dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
747 		    unsigned int num_dests, struct dm_io_region *dests,
748 		    unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
749 {
750 	struct kcopyd_job *job;
751 	int i;
752 
753 	/*
754 	 * Allocate an array of jobs consisting of one master job
755 	 * followed by SPLIT_COUNT sub jobs.
756 	 */
757 	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
758 	mutex_init(&job->lock);
759 
760 	/*
761 	 * set up for the read.
762 	 */
763 	job->kc = kc;
764 	job->flags = flags;
765 	job->read_err = 0;
766 	job->write_err = 0;
767 
768 	job->num_dests = num_dests;
769 	memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
770 
771 	/*
772 	 * If one of the destination is a host-managed zoned block device,
773 	 * we need to write sequentially. If one of the destination is a
774 	 * host-aware device, then leave it to the caller to choose what to do.
775 	 */
776 	if (!test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
777 		for (i = 0; i < job->num_dests; i++) {
778 			if (bdev_zoned_model(dests[i].bdev) == BLK_ZONED_HM) {
779 				set_bit(DM_KCOPYD_WRITE_SEQ, &job->flags);
780 				break;
781 			}
782 		}
783 	}
784 
785 	/*
786 	 * If we need to write sequentially, errors cannot be ignored.
787 	 */
788 	if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
789 	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags))
790 		clear_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags);
791 
792 	if (from) {
793 		job->source = *from;
794 		job->pages = NULL;
795 		job->rw = READ;
796 	} else {
797 		memset(&job->source, 0, sizeof job->source);
798 		job->source.count = job->dests[0].count;
799 		job->pages = &zero_page_list;
800 
801 		/*
802 		 * Use WRITE ZEROES to optimize zeroing if all dests support it.
803 		 */
804 		job->rw = REQ_OP_WRITE_ZEROES;
805 		for (i = 0; i < job->num_dests; i++)
806 			if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
807 				job->rw = WRITE;
808 				break;
809 			}
810 	}
811 
812 	job->fn = fn;
813 	job->context = context;
814 	job->master_job = job;
815 	job->write_offset = 0;
816 
817 	if (job->source.count <= SUB_JOB_SIZE)
818 		dispatch_job(job);
819 	else {
820 		job->progress = 0;
821 		split_job(job);
822 	}
823 }
824 EXPORT_SYMBOL(dm_kcopyd_copy);
825 
826 void dm_kcopyd_zero(struct dm_kcopyd_client *kc,
827 		    unsigned num_dests, struct dm_io_region *dests,
828 		    unsigned flags, dm_kcopyd_notify_fn fn, void *context)
829 {
830 	dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
831 }
832 EXPORT_SYMBOL(dm_kcopyd_zero);
833 
834 void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
835 				 dm_kcopyd_notify_fn fn, void *context)
836 {
837 	struct kcopyd_job *job;
838 
839 	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
840 
841 	memset(job, 0, sizeof(struct kcopyd_job));
842 	job->kc = kc;
843 	job->fn = fn;
844 	job->context = context;
845 	job->master_job = job;
846 
847 	atomic_inc(&kc->nr_jobs);
848 
849 	return job;
850 }
851 EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
852 
853 void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
854 {
855 	struct kcopyd_job *job = j;
856 	struct dm_kcopyd_client *kc = job->kc;
857 
858 	job->read_err = read_err;
859 	job->write_err = write_err;
860 
861 	push(&kc->complete_jobs, job);
862 	wake(kc);
863 }
864 EXPORT_SYMBOL(dm_kcopyd_do_callback);
865 
866 /*
867  * Cancels a kcopyd job, eg. someone might be deactivating a
868  * mirror.
869  */
870 #if 0
871 int kcopyd_cancel(struct kcopyd_job *job, int block)
872 {
873 	/* FIXME: finish */
874 	return -1;
875 }
876 #endif  /*  0  */
877 
878 /*-----------------------------------------------------------------
879  * Client setup
880  *---------------------------------------------------------------*/
881 struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
882 {
883 	int r;
884 	struct dm_kcopyd_client *kc;
885 
886 	kc = kzalloc(sizeof(*kc), GFP_KERNEL);
887 	if (!kc)
888 		return ERR_PTR(-ENOMEM);
889 
890 	spin_lock_init(&kc->job_lock);
891 	INIT_LIST_HEAD(&kc->complete_jobs);
892 	INIT_LIST_HEAD(&kc->io_jobs);
893 	INIT_LIST_HEAD(&kc->pages_jobs);
894 	kc->throttle = throttle;
895 
896 	r = mempool_init_slab_pool(&kc->job_pool, MIN_JOBS, _job_cache);
897 	if (r)
898 		goto bad_slab;
899 
900 	INIT_WORK(&kc->kcopyd_work, do_work);
901 	kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
902 	if (!kc->kcopyd_wq) {
903 		r = -ENOMEM;
904 		goto bad_workqueue;
905 	}
906 
907 	kc->pages = NULL;
908 	kc->nr_reserved_pages = kc->nr_free_pages = 0;
909 	r = client_reserve_pages(kc, RESERVE_PAGES);
910 	if (r)
911 		goto bad_client_pages;
912 
913 	kc->io_client = dm_io_client_create();
914 	if (IS_ERR(kc->io_client)) {
915 		r = PTR_ERR(kc->io_client);
916 		goto bad_io_client;
917 	}
918 
919 	init_waitqueue_head(&kc->destroyq);
920 	atomic_set(&kc->nr_jobs, 0);
921 
922 	return kc;
923 
924 bad_io_client:
925 	client_free_pages(kc);
926 bad_client_pages:
927 	destroy_workqueue(kc->kcopyd_wq);
928 bad_workqueue:
929 	mempool_exit(&kc->job_pool);
930 bad_slab:
931 	kfree(kc);
932 
933 	return ERR_PTR(r);
934 }
935 EXPORT_SYMBOL(dm_kcopyd_client_create);
936 
937 void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
938 {
939 	/* Wait for completion of all jobs submitted by this client. */
940 	wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
941 
942 	BUG_ON(!list_empty(&kc->complete_jobs));
943 	BUG_ON(!list_empty(&kc->io_jobs));
944 	BUG_ON(!list_empty(&kc->pages_jobs));
945 	destroy_workqueue(kc->kcopyd_wq);
946 	dm_io_client_destroy(kc->io_client);
947 	client_free_pages(kc);
948 	mempool_exit(&kc->job_pool);
949 	kfree(kc);
950 }
951 EXPORT_SYMBOL(dm_kcopyd_client_destroy);
952