xref: /openbmc/linux/drivers/md/dm-kcopyd.c (revision f519f0be)
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 four 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 don't need to do any IO and just run a callback
64  * iv) jobs that have completed.
65  *
66  * All four of these are protected by job_lock.
67  */
68 	spinlock_t job_lock;
69 	struct list_head callback_jobs;
70 	struct list_head complete_jobs;
71 	struct list_head io_jobs;
72 	struct list_head pages_jobs;
73 };
74 
75 static struct page_list zero_page_list;
76 
77 static DEFINE_SPINLOCK(throttle_spinlock);
78 
79 /*
80  * IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
81  * When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
82  * by 2.
83  */
84 #define ACCOUNT_INTERVAL_SHIFT		SHIFT_HZ
85 
86 /*
87  * Sleep this number of milliseconds.
88  *
89  * The value was decided experimentally.
90  * Smaller values seem to cause an increased copy rate above the limit.
91  * The reason for this is unknown but possibly due to jiffies rounding errors
92  * or read/write cache inside the disk.
93  */
94 #define SLEEP_MSEC			100
95 
96 /*
97  * Maximum number of sleep events. There is a theoretical livelock if more
98  * kcopyd clients do work simultaneously which this limit avoids.
99  */
100 #define MAX_SLEEPS			10
101 
102 static void io_job_start(struct dm_kcopyd_throttle *t)
103 {
104 	unsigned throttle, now, difference;
105 	int slept = 0, skew;
106 
107 	if (unlikely(!t))
108 		return;
109 
110 try_again:
111 	spin_lock_irq(&throttle_spinlock);
112 
113 	throttle = READ_ONCE(t->throttle);
114 
115 	if (likely(throttle >= 100))
116 		goto skip_limit;
117 
118 	now = jiffies;
119 	difference = now - t->last_jiffies;
120 	t->last_jiffies = now;
121 	if (t->num_io_jobs)
122 		t->io_period += difference;
123 	t->total_period += difference;
124 
125 	/*
126 	 * Maintain sane values if we got a temporary overflow.
127 	 */
128 	if (unlikely(t->io_period > t->total_period))
129 		t->io_period = t->total_period;
130 
131 	if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
132 		int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
133 		t->total_period >>= shift;
134 		t->io_period >>= shift;
135 	}
136 
137 	skew = t->io_period - throttle * t->total_period / 100;
138 
139 	if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
140 		slept++;
141 		spin_unlock_irq(&throttle_spinlock);
142 		msleep(SLEEP_MSEC);
143 		goto try_again;
144 	}
145 
146 skip_limit:
147 	t->num_io_jobs++;
148 
149 	spin_unlock_irq(&throttle_spinlock);
150 }
151 
152 static void io_job_finish(struct dm_kcopyd_throttle *t)
153 {
154 	unsigned long flags;
155 
156 	if (unlikely(!t))
157 		return;
158 
159 	spin_lock_irqsave(&throttle_spinlock, flags);
160 
161 	t->num_io_jobs--;
162 
163 	if (likely(READ_ONCE(t->throttle) >= 100))
164 		goto skip_limit;
165 
166 	if (!t->num_io_jobs) {
167 		unsigned now, difference;
168 
169 		now = jiffies;
170 		difference = now - t->last_jiffies;
171 		t->last_jiffies = now;
172 
173 		t->io_period += difference;
174 		t->total_period += difference;
175 
176 		/*
177 		 * Maintain sane values if we got a temporary overflow.
178 		 */
179 		if (unlikely(t->io_period > t->total_period))
180 			t->io_period = t->total_period;
181 	}
182 
183 skip_limit:
184 	spin_unlock_irqrestore(&throttle_spinlock, flags);
185 }
186 
187 
188 static void wake(struct dm_kcopyd_client *kc)
189 {
190 	queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
191 }
192 
193 /*
194  * Obtain one page for the use of kcopyd.
195  */
196 static struct page_list *alloc_pl(gfp_t gfp)
197 {
198 	struct page_list *pl;
199 
200 	pl = kmalloc(sizeof(*pl), gfp);
201 	if (!pl)
202 		return NULL;
203 
204 	pl->page = alloc_page(gfp);
205 	if (!pl->page) {
206 		kfree(pl);
207 		return NULL;
208 	}
209 
210 	return pl;
211 }
212 
213 static void free_pl(struct page_list *pl)
214 {
215 	__free_page(pl->page);
216 	kfree(pl);
217 }
218 
219 /*
220  * Add the provided pages to a client's free page list, releasing
221  * back to the system any beyond the reserved_pages limit.
222  */
223 static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
224 {
225 	struct page_list *next;
226 
227 	do {
228 		next = pl->next;
229 
230 		if (kc->nr_free_pages >= kc->nr_reserved_pages)
231 			free_pl(pl);
232 		else {
233 			pl->next = kc->pages;
234 			kc->pages = pl;
235 			kc->nr_free_pages++;
236 		}
237 
238 		pl = next;
239 	} while (pl);
240 }
241 
242 static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
243 			    unsigned int nr, struct page_list **pages)
244 {
245 	struct page_list *pl;
246 
247 	*pages = NULL;
248 
249 	do {
250 		pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
251 		if (unlikely(!pl)) {
252 			/* Use reserved pages */
253 			pl = kc->pages;
254 			if (unlikely(!pl))
255 				goto out_of_memory;
256 			kc->pages = pl->next;
257 			kc->nr_free_pages--;
258 		}
259 		pl->next = *pages;
260 		*pages = pl;
261 	} while (--nr);
262 
263 	return 0;
264 
265 out_of_memory:
266 	if (*pages)
267 		kcopyd_put_pages(kc, *pages);
268 	return -ENOMEM;
269 }
270 
271 /*
272  * These three functions resize the page pool.
273  */
274 static void drop_pages(struct page_list *pl)
275 {
276 	struct page_list *next;
277 
278 	while (pl) {
279 		next = pl->next;
280 		free_pl(pl);
281 		pl = next;
282 	}
283 }
284 
285 /*
286  * Allocate and reserve nr_pages for the use of a specific client.
287  */
288 static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
289 {
290 	unsigned i;
291 	struct page_list *pl = NULL, *next;
292 
293 	for (i = 0; i < nr_pages; i++) {
294 		next = alloc_pl(GFP_KERNEL);
295 		if (!next) {
296 			if (pl)
297 				drop_pages(pl);
298 			return -ENOMEM;
299 		}
300 		next->next = pl;
301 		pl = next;
302 	}
303 
304 	kc->nr_reserved_pages += nr_pages;
305 	kcopyd_put_pages(kc, pl);
306 
307 	return 0;
308 }
309 
310 static void client_free_pages(struct dm_kcopyd_client *kc)
311 {
312 	BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
313 	drop_pages(kc->pages);
314 	kc->pages = NULL;
315 	kc->nr_free_pages = kc->nr_reserved_pages = 0;
316 }
317 
318 /*-----------------------------------------------------------------
319  * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
320  * for this reason we use a mempool to prevent the client from
321  * ever having to do io (which could cause a deadlock).
322  *---------------------------------------------------------------*/
323 struct kcopyd_job {
324 	struct dm_kcopyd_client *kc;
325 	struct list_head list;
326 	unsigned long flags;
327 
328 	/*
329 	 * Error state of the job.
330 	 */
331 	int read_err;
332 	unsigned long write_err;
333 
334 	/*
335 	 * Either READ or WRITE
336 	 */
337 	int rw;
338 	struct dm_io_region source;
339 
340 	/*
341 	 * The destinations for the transfer.
342 	 */
343 	unsigned int num_dests;
344 	struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
345 
346 	struct page_list *pages;
347 
348 	/*
349 	 * Set this to ensure you are notified when the job has
350 	 * completed.  'context' is for callback to use.
351 	 */
352 	dm_kcopyd_notify_fn fn;
353 	void *context;
354 
355 	/*
356 	 * These fields are only used if the job has been split
357 	 * into more manageable parts.
358 	 */
359 	struct mutex lock;
360 	atomic_t sub_jobs;
361 	sector_t progress;
362 	sector_t write_offset;
363 
364 	struct kcopyd_job *master_job;
365 };
366 
367 static struct kmem_cache *_job_cache;
368 
369 int __init dm_kcopyd_init(void)
370 {
371 	_job_cache = kmem_cache_create("kcopyd_job",
372 				sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
373 				__alignof__(struct kcopyd_job), 0, NULL);
374 	if (!_job_cache)
375 		return -ENOMEM;
376 
377 	zero_page_list.next = &zero_page_list;
378 	zero_page_list.page = ZERO_PAGE(0);
379 
380 	return 0;
381 }
382 
383 void dm_kcopyd_exit(void)
384 {
385 	kmem_cache_destroy(_job_cache);
386 	_job_cache = NULL;
387 }
388 
389 /*
390  * Functions to push and pop a job onto the head of a given job
391  * list.
392  */
393 static struct kcopyd_job *pop_io_job(struct list_head *jobs,
394 				     struct dm_kcopyd_client *kc)
395 {
396 	struct kcopyd_job *job;
397 
398 	/*
399 	 * For I/O jobs, pop any read, any write without sequential write
400 	 * constraint and sequential writes that are at the right position.
401 	 */
402 	list_for_each_entry(job, jobs, list) {
403 		if (job->rw == READ || !test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
404 			list_del(&job->list);
405 			return job;
406 		}
407 
408 		if (job->write_offset == job->master_job->write_offset) {
409 			job->master_job->write_offset += job->source.count;
410 			list_del(&job->list);
411 			return job;
412 		}
413 	}
414 
415 	return NULL;
416 }
417 
418 static struct kcopyd_job *pop(struct list_head *jobs,
419 			      struct dm_kcopyd_client *kc)
420 {
421 	struct kcopyd_job *job = NULL;
422 	unsigned long flags;
423 
424 	spin_lock_irqsave(&kc->job_lock, flags);
425 
426 	if (!list_empty(jobs)) {
427 		if (jobs == &kc->io_jobs)
428 			job = pop_io_job(jobs, kc);
429 		else {
430 			job = list_entry(jobs->next, struct kcopyd_job, list);
431 			list_del(&job->list);
432 		}
433 	}
434 	spin_unlock_irqrestore(&kc->job_lock, flags);
435 
436 	return job;
437 }
438 
439 static void push(struct list_head *jobs, struct kcopyd_job *job)
440 {
441 	unsigned long flags;
442 	struct dm_kcopyd_client *kc = job->kc;
443 
444 	spin_lock_irqsave(&kc->job_lock, flags);
445 	list_add_tail(&job->list, jobs);
446 	spin_unlock_irqrestore(&kc->job_lock, flags);
447 }
448 
449 
450 static void push_head(struct list_head *jobs, struct kcopyd_job *job)
451 {
452 	unsigned long flags;
453 	struct dm_kcopyd_client *kc = job->kc;
454 
455 	spin_lock_irqsave(&kc->job_lock, flags);
456 	list_add(&job->list, jobs);
457 	spin_unlock_irqrestore(&kc->job_lock, flags);
458 }
459 
460 /*
461  * These three functions process 1 item from the corresponding
462  * job list.
463  *
464  * They return:
465  * < 0: error
466  *   0: success
467  * > 0: can't process yet.
468  */
469 static int run_complete_job(struct kcopyd_job *job)
470 {
471 	void *context = job->context;
472 	int read_err = job->read_err;
473 	unsigned long write_err = job->write_err;
474 	dm_kcopyd_notify_fn fn = job->fn;
475 	struct dm_kcopyd_client *kc = job->kc;
476 
477 	if (job->pages && job->pages != &zero_page_list)
478 		kcopyd_put_pages(kc, job->pages);
479 	/*
480 	 * If this is the master job, the sub jobs have already
481 	 * completed so we can free everything.
482 	 */
483 	if (job->master_job == job) {
484 		mutex_destroy(&job->lock);
485 		mempool_free(job, &kc->job_pool);
486 	}
487 	fn(read_err, write_err, context);
488 
489 	if (atomic_dec_and_test(&kc->nr_jobs))
490 		wake_up(&kc->destroyq);
491 
492 	cond_resched();
493 
494 	return 0;
495 }
496 
497 static void complete_io(unsigned long error, void *context)
498 {
499 	struct kcopyd_job *job = (struct kcopyd_job *) context;
500 	struct dm_kcopyd_client *kc = job->kc;
501 
502 	io_job_finish(kc->throttle);
503 
504 	if (error) {
505 		if (op_is_write(job->rw))
506 			job->write_err |= error;
507 		else
508 			job->read_err = 1;
509 
510 		if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
511 			push(&kc->complete_jobs, job);
512 			wake(kc);
513 			return;
514 		}
515 	}
516 
517 	if (op_is_write(job->rw))
518 		push(&kc->complete_jobs, job);
519 
520 	else {
521 		job->rw = WRITE;
522 		push(&kc->io_jobs, job);
523 	}
524 
525 	wake(kc);
526 }
527 
528 /*
529  * Request io on as many buffer heads as we can currently get for
530  * a particular job.
531  */
532 static int run_io_job(struct kcopyd_job *job)
533 {
534 	int r;
535 	struct dm_io_request io_req = {
536 		.bi_op = job->rw,
537 		.bi_op_flags = 0,
538 		.mem.type = DM_IO_PAGE_LIST,
539 		.mem.ptr.pl = job->pages,
540 		.mem.offset = 0,
541 		.notify.fn = complete_io,
542 		.notify.context = job,
543 		.client = job->kc->io_client,
544 	};
545 
546 	/*
547 	 * If we need to write sequentially and some reads or writes failed,
548 	 * no point in continuing.
549 	 */
550 	if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
551 	    job->master_job->write_err)
552 		return -EIO;
553 
554 	io_job_start(job->kc->throttle);
555 
556 	if (job->rw == READ)
557 		r = dm_io(&io_req, 1, &job->source, NULL);
558 	else
559 		r = dm_io(&io_req, job->num_dests, job->dests, NULL);
560 
561 	return r;
562 }
563 
564 static int run_pages_job(struct kcopyd_job *job)
565 {
566 	int r;
567 	unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
568 
569 	r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
570 	if (!r) {
571 		/* this job is ready for io */
572 		push(&job->kc->io_jobs, job);
573 		return 0;
574 	}
575 
576 	if (r == -ENOMEM)
577 		/* can't complete now */
578 		return 1;
579 
580 	return r;
581 }
582 
583 /*
584  * Run through a list for as long as possible.  Returns the count
585  * of successful jobs.
586  */
587 static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
588 			int (*fn) (struct kcopyd_job *))
589 {
590 	struct kcopyd_job *job;
591 	int r, count = 0;
592 
593 	while ((job = pop(jobs, kc))) {
594 
595 		r = fn(job);
596 
597 		if (r < 0) {
598 			/* error this rogue job */
599 			if (op_is_write(job->rw))
600 				job->write_err = (unsigned long) -1L;
601 			else
602 				job->read_err = 1;
603 			push(&kc->complete_jobs, job);
604 			break;
605 		}
606 
607 		if (r > 0) {
608 			/*
609 			 * We couldn't service this job ATM, so
610 			 * push this job back onto the list.
611 			 */
612 			push_head(jobs, job);
613 			break;
614 		}
615 
616 		count++;
617 	}
618 
619 	return count;
620 }
621 
622 /*
623  * kcopyd does this every time it's woken up.
624  */
625 static void do_work(struct work_struct *work)
626 {
627 	struct dm_kcopyd_client *kc = container_of(work,
628 					struct dm_kcopyd_client, kcopyd_work);
629 	struct blk_plug plug;
630 	unsigned long flags;
631 
632 	/*
633 	 * The order that these are called is *very* important.
634 	 * complete jobs can free some pages for pages jobs.
635 	 * Pages jobs when successful will jump onto the io jobs
636 	 * list.  io jobs call wake when they complete and it all
637 	 * starts again.
638 	 */
639 	spin_lock_irqsave(&kc->job_lock, flags);
640 	list_splice_tail_init(&kc->callback_jobs, &kc->complete_jobs);
641 	spin_unlock_irqrestore(&kc->job_lock, flags);
642 
643 	blk_start_plug(&plug);
644 	process_jobs(&kc->complete_jobs, kc, run_complete_job);
645 	process_jobs(&kc->pages_jobs, kc, run_pages_job);
646 	process_jobs(&kc->io_jobs, kc, run_io_job);
647 	blk_finish_plug(&plug);
648 }
649 
650 /*
651  * If we are copying a small region we just dispatch a single job
652  * to do the copy, otherwise the io has to be split up into many
653  * jobs.
654  */
655 static void dispatch_job(struct kcopyd_job *job)
656 {
657 	struct dm_kcopyd_client *kc = job->kc;
658 	atomic_inc(&kc->nr_jobs);
659 	if (unlikely(!job->source.count))
660 		push(&kc->callback_jobs, job);
661 	else if (job->pages == &zero_page_list)
662 		push(&kc->io_jobs, job);
663 	else
664 		push(&kc->pages_jobs, job);
665 	wake(kc);
666 }
667 
668 static void segment_complete(int read_err, unsigned long write_err,
669 			     void *context)
670 {
671 	/* FIXME: tidy this function */
672 	sector_t progress = 0;
673 	sector_t count = 0;
674 	struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
675 	struct kcopyd_job *job = sub_job->master_job;
676 	struct dm_kcopyd_client *kc = job->kc;
677 
678 	mutex_lock(&job->lock);
679 
680 	/* update the error */
681 	if (read_err)
682 		job->read_err = 1;
683 
684 	if (write_err)
685 		job->write_err |= write_err;
686 
687 	/*
688 	 * Only dispatch more work if there hasn't been an error.
689 	 */
690 	if ((!job->read_err && !job->write_err) ||
691 	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
692 		/* get the next chunk of work */
693 		progress = job->progress;
694 		count = job->source.count - progress;
695 		if (count) {
696 			if (count > SUB_JOB_SIZE)
697 				count = SUB_JOB_SIZE;
698 
699 			job->progress += count;
700 		}
701 	}
702 	mutex_unlock(&job->lock);
703 
704 	if (count) {
705 		int i;
706 
707 		*sub_job = *job;
708 		sub_job->write_offset = progress;
709 		sub_job->source.sector += progress;
710 		sub_job->source.count = count;
711 
712 		for (i = 0; i < job->num_dests; i++) {
713 			sub_job->dests[i].sector += progress;
714 			sub_job->dests[i].count = count;
715 		}
716 
717 		sub_job->fn = segment_complete;
718 		sub_job->context = sub_job;
719 		dispatch_job(sub_job);
720 
721 	} else if (atomic_dec_and_test(&job->sub_jobs)) {
722 
723 		/*
724 		 * Queue the completion callback to the kcopyd thread.
725 		 *
726 		 * Some callers assume that all the completions are called
727 		 * from a single thread and don't race with each other.
728 		 *
729 		 * We must not call the callback directly here because this
730 		 * code may not be executing in the thread.
731 		 */
732 		push(&kc->complete_jobs, job);
733 		wake(kc);
734 	}
735 }
736 
737 /*
738  * Create some sub jobs to share the work between them.
739  */
740 static void split_job(struct kcopyd_job *master_job)
741 {
742 	int i;
743 
744 	atomic_inc(&master_job->kc->nr_jobs);
745 
746 	atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
747 	for (i = 0; i < SPLIT_COUNT; i++) {
748 		master_job[i + 1].master_job = master_job;
749 		segment_complete(0, 0u, &master_job[i + 1]);
750 	}
751 }
752 
753 void dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
754 		    unsigned int num_dests, struct dm_io_region *dests,
755 		    unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
756 {
757 	struct kcopyd_job *job;
758 	int i;
759 
760 	/*
761 	 * Allocate an array of jobs consisting of one master job
762 	 * followed by SPLIT_COUNT sub jobs.
763 	 */
764 	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
765 	mutex_init(&job->lock);
766 
767 	/*
768 	 * set up for the read.
769 	 */
770 	job->kc = kc;
771 	job->flags = flags;
772 	job->read_err = 0;
773 	job->write_err = 0;
774 
775 	job->num_dests = num_dests;
776 	memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
777 
778 	/*
779 	 * If one of the destination is a host-managed zoned block device,
780 	 * we need to write sequentially. If one of the destination is a
781 	 * host-aware device, then leave it to the caller to choose what to do.
782 	 */
783 	if (!test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
784 		for (i = 0; i < job->num_dests; i++) {
785 			if (bdev_zoned_model(dests[i].bdev) == BLK_ZONED_HM) {
786 				set_bit(DM_KCOPYD_WRITE_SEQ, &job->flags);
787 				break;
788 			}
789 		}
790 	}
791 
792 	/*
793 	 * If we need to write sequentially, errors cannot be ignored.
794 	 */
795 	if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
796 	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags))
797 		clear_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags);
798 
799 	if (from) {
800 		job->source = *from;
801 		job->pages = NULL;
802 		job->rw = READ;
803 	} else {
804 		memset(&job->source, 0, sizeof job->source);
805 		job->source.count = job->dests[0].count;
806 		job->pages = &zero_page_list;
807 
808 		/*
809 		 * Use WRITE ZEROES to optimize zeroing if all dests support it.
810 		 */
811 		job->rw = REQ_OP_WRITE_ZEROES;
812 		for (i = 0; i < job->num_dests; i++)
813 			if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
814 				job->rw = WRITE;
815 				break;
816 			}
817 	}
818 
819 	job->fn = fn;
820 	job->context = context;
821 	job->master_job = job;
822 	job->write_offset = 0;
823 
824 	if (job->source.count <= SUB_JOB_SIZE)
825 		dispatch_job(job);
826 	else {
827 		job->progress = 0;
828 		split_job(job);
829 	}
830 }
831 EXPORT_SYMBOL(dm_kcopyd_copy);
832 
833 void dm_kcopyd_zero(struct dm_kcopyd_client *kc,
834 		    unsigned num_dests, struct dm_io_region *dests,
835 		    unsigned flags, dm_kcopyd_notify_fn fn, void *context)
836 {
837 	dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
838 }
839 EXPORT_SYMBOL(dm_kcopyd_zero);
840 
841 void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
842 				 dm_kcopyd_notify_fn fn, void *context)
843 {
844 	struct kcopyd_job *job;
845 
846 	job = mempool_alloc(&kc->job_pool, GFP_NOIO);
847 
848 	memset(job, 0, sizeof(struct kcopyd_job));
849 	job->kc = kc;
850 	job->fn = fn;
851 	job->context = context;
852 	job->master_job = job;
853 
854 	atomic_inc(&kc->nr_jobs);
855 
856 	return job;
857 }
858 EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
859 
860 void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
861 {
862 	struct kcopyd_job *job = j;
863 	struct dm_kcopyd_client *kc = job->kc;
864 
865 	job->read_err = read_err;
866 	job->write_err = write_err;
867 
868 	push(&kc->callback_jobs, job);
869 	wake(kc);
870 }
871 EXPORT_SYMBOL(dm_kcopyd_do_callback);
872 
873 /*
874  * Cancels a kcopyd job, eg. someone might be deactivating a
875  * mirror.
876  */
877 #if 0
878 int kcopyd_cancel(struct kcopyd_job *job, int block)
879 {
880 	/* FIXME: finish */
881 	return -1;
882 }
883 #endif  /*  0  */
884 
885 /*-----------------------------------------------------------------
886  * Client setup
887  *---------------------------------------------------------------*/
888 struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
889 {
890 	int r;
891 	struct dm_kcopyd_client *kc;
892 
893 	kc = kzalloc(sizeof(*kc), GFP_KERNEL);
894 	if (!kc)
895 		return ERR_PTR(-ENOMEM);
896 
897 	spin_lock_init(&kc->job_lock);
898 	INIT_LIST_HEAD(&kc->callback_jobs);
899 	INIT_LIST_HEAD(&kc->complete_jobs);
900 	INIT_LIST_HEAD(&kc->io_jobs);
901 	INIT_LIST_HEAD(&kc->pages_jobs);
902 	kc->throttle = throttle;
903 
904 	r = mempool_init_slab_pool(&kc->job_pool, MIN_JOBS, _job_cache);
905 	if (r)
906 		goto bad_slab;
907 
908 	INIT_WORK(&kc->kcopyd_work, do_work);
909 	kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
910 	if (!kc->kcopyd_wq) {
911 		r = -ENOMEM;
912 		goto bad_workqueue;
913 	}
914 
915 	kc->pages = NULL;
916 	kc->nr_reserved_pages = kc->nr_free_pages = 0;
917 	r = client_reserve_pages(kc, RESERVE_PAGES);
918 	if (r)
919 		goto bad_client_pages;
920 
921 	kc->io_client = dm_io_client_create();
922 	if (IS_ERR(kc->io_client)) {
923 		r = PTR_ERR(kc->io_client);
924 		goto bad_io_client;
925 	}
926 
927 	init_waitqueue_head(&kc->destroyq);
928 	atomic_set(&kc->nr_jobs, 0);
929 
930 	return kc;
931 
932 bad_io_client:
933 	client_free_pages(kc);
934 bad_client_pages:
935 	destroy_workqueue(kc->kcopyd_wq);
936 bad_workqueue:
937 	mempool_exit(&kc->job_pool);
938 bad_slab:
939 	kfree(kc);
940 
941 	return ERR_PTR(r);
942 }
943 EXPORT_SYMBOL(dm_kcopyd_client_create);
944 
945 void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
946 {
947 	/* Wait for completion of all jobs submitted by this client. */
948 	wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
949 
950 	BUG_ON(!list_empty(&kc->callback_jobs));
951 	BUG_ON(!list_empty(&kc->complete_jobs));
952 	BUG_ON(!list_empty(&kc->io_jobs));
953 	BUG_ON(!list_empty(&kc->pages_jobs));
954 	destroy_workqueue(kc->kcopyd_wq);
955 	dm_io_client_destroy(kc->io_client);
956 	client_free_pages(kc);
957 	mempool_exit(&kc->job_pool);
958 	kfree(kc);
959 }
960 EXPORT_SYMBOL(dm_kcopyd_client_destroy);
961