xref: /openbmc/linux/drivers/md/dm-kcopyd.c (revision 63dc02bd)
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/device-mapper.h>
26 #include <linux/dm-kcopyd.h>
27 
28 #include "dm.h"
29 
30 #define SUB_JOB_SIZE	128
31 #define SPLIT_COUNT	8
32 #define MIN_JOBS	8
33 #define RESERVE_PAGES	(DIV_ROUND_UP(SUB_JOB_SIZE << SECTOR_SHIFT, PAGE_SIZE))
34 
35 /*-----------------------------------------------------------------
36  * Each kcopyd client has its own little pool of preallocated
37  * pages for kcopyd io.
38  *---------------------------------------------------------------*/
39 struct dm_kcopyd_client {
40 	struct page_list *pages;
41 	unsigned nr_reserved_pages;
42 	unsigned nr_free_pages;
43 
44 	struct dm_io_client *io_client;
45 
46 	wait_queue_head_t destroyq;
47 	atomic_t nr_jobs;
48 
49 	mempool_t *job_pool;
50 
51 	struct workqueue_struct *kcopyd_wq;
52 	struct work_struct kcopyd_work;
53 
54 /*
55  * We maintain three lists of jobs:
56  *
57  * i)   jobs waiting for pages
58  * ii)  jobs that have pages, and are waiting for the io to be issued.
59  * iii) jobs that have completed.
60  *
61  * All three of these are protected by job_lock.
62  */
63 	spinlock_t job_lock;
64 	struct list_head complete_jobs;
65 	struct list_head io_jobs;
66 	struct list_head pages_jobs;
67 };
68 
69 static struct page_list zero_page_list;
70 
71 static void wake(struct dm_kcopyd_client *kc)
72 {
73 	queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
74 }
75 
76 /*
77  * Obtain one page for the use of kcopyd.
78  */
79 static struct page_list *alloc_pl(gfp_t gfp)
80 {
81 	struct page_list *pl;
82 
83 	pl = kmalloc(sizeof(*pl), gfp);
84 	if (!pl)
85 		return NULL;
86 
87 	pl->page = alloc_page(gfp);
88 	if (!pl->page) {
89 		kfree(pl);
90 		return NULL;
91 	}
92 
93 	return pl;
94 }
95 
96 static void free_pl(struct page_list *pl)
97 {
98 	__free_page(pl->page);
99 	kfree(pl);
100 }
101 
102 /*
103  * Add the provided pages to a client's free page list, releasing
104  * back to the system any beyond the reserved_pages limit.
105  */
106 static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
107 {
108 	struct page_list *next;
109 
110 	do {
111 		next = pl->next;
112 
113 		if (kc->nr_free_pages >= kc->nr_reserved_pages)
114 			free_pl(pl);
115 		else {
116 			pl->next = kc->pages;
117 			kc->pages = pl;
118 			kc->nr_free_pages++;
119 		}
120 
121 		pl = next;
122 	} while (pl);
123 }
124 
125 static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
126 			    unsigned int nr, struct page_list **pages)
127 {
128 	struct page_list *pl;
129 
130 	*pages = NULL;
131 
132 	do {
133 		pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY);
134 		if (unlikely(!pl)) {
135 			/* Use reserved pages */
136 			pl = kc->pages;
137 			if (unlikely(!pl))
138 				goto out_of_memory;
139 			kc->pages = pl->next;
140 			kc->nr_free_pages--;
141 		}
142 		pl->next = *pages;
143 		*pages = pl;
144 	} while (--nr);
145 
146 	return 0;
147 
148 out_of_memory:
149 	if (*pages)
150 		kcopyd_put_pages(kc, *pages);
151 	return -ENOMEM;
152 }
153 
154 /*
155  * These three functions resize the page pool.
156  */
157 static void drop_pages(struct page_list *pl)
158 {
159 	struct page_list *next;
160 
161 	while (pl) {
162 		next = pl->next;
163 		free_pl(pl);
164 		pl = next;
165 	}
166 }
167 
168 /*
169  * Allocate and reserve nr_pages for the use of a specific client.
170  */
171 static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
172 {
173 	unsigned i;
174 	struct page_list *pl = NULL, *next;
175 
176 	for (i = 0; i < nr_pages; i++) {
177 		next = alloc_pl(GFP_KERNEL);
178 		if (!next) {
179 			if (pl)
180 				drop_pages(pl);
181 			return -ENOMEM;
182 		}
183 		next->next = pl;
184 		pl = next;
185 	}
186 
187 	kc->nr_reserved_pages += nr_pages;
188 	kcopyd_put_pages(kc, pl);
189 
190 	return 0;
191 }
192 
193 static void client_free_pages(struct dm_kcopyd_client *kc)
194 {
195 	BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
196 	drop_pages(kc->pages);
197 	kc->pages = NULL;
198 	kc->nr_free_pages = kc->nr_reserved_pages = 0;
199 }
200 
201 /*-----------------------------------------------------------------
202  * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
203  * for this reason we use a mempool to prevent the client from
204  * ever having to do io (which could cause a deadlock).
205  *---------------------------------------------------------------*/
206 struct kcopyd_job {
207 	struct dm_kcopyd_client *kc;
208 	struct list_head list;
209 	unsigned long flags;
210 
211 	/*
212 	 * Error state of the job.
213 	 */
214 	int read_err;
215 	unsigned long write_err;
216 
217 	/*
218 	 * Either READ or WRITE
219 	 */
220 	int rw;
221 	struct dm_io_region source;
222 
223 	/*
224 	 * The destinations for the transfer.
225 	 */
226 	unsigned int num_dests;
227 	struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
228 
229 	struct page_list *pages;
230 
231 	/*
232 	 * Set this to ensure you are notified when the job has
233 	 * completed.  'context' is for callback to use.
234 	 */
235 	dm_kcopyd_notify_fn fn;
236 	void *context;
237 
238 	/*
239 	 * These fields are only used if the job has been split
240 	 * into more manageable parts.
241 	 */
242 	struct mutex lock;
243 	atomic_t sub_jobs;
244 	sector_t progress;
245 
246 	struct kcopyd_job *master_job;
247 };
248 
249 static struct kmem_cache *_job_cache;
250 
251 int __init dm_kcopyd_init(void)
252 {
253 	_job_cache = kmem_cache_create("kcopyd_job",
254 				sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
255 				__alignof__(struct kcopyd_job), 0, NULL);
256 	if (!_job_cache)
257 		return -ENOMEM;
258 
259 	zero_page_list.next = &zero_page_list;
260 	zero_page_list.page = ZERO_PAGE(0);
261 
262 	return 0;
263 }
264 
265 void dm_kcopyd_exit(void)
266 {
267 	kmem_cache_destroy(_job_cache);
268 	_job_cache = NULL;
269 }
270 
271 /*
272  * Functions to push and pop a job onto the head of a given job
273  * list.
274  */
275 static struct kcopyd_job *pop(struct list_head *jobs,
276 			      struct dm_kcopyd_client *kc)
277 {
278 	struct kcopyd_job *job = NULL;
279 	unsigned long flags;
280 
281 	spin_lock_irqsave(&kc->job_lock, flags);
282 
283 	if (!list_empty(jobs)) {
284 		job = list_entry(jobs->next, struct kcopyd_job, list);
285 		list_del(&job->list);
286 	}
287 	spin_unlock_irqrestore(&kc->job_lock, flags);
288 
289 	return job;
290 }
291 
292 static void push(struct list_head *jobs, struct kcopyd_job *job)
293 {
294 	unsigned long flags;
295 	struct dm_kcopyd_client *kc = job->kc;
296 
297 	spin_lock_irqsave(&kc->job_lock, flags);
298 	list_add_tail(&job->list, jobs);
299 	spin_unlock_irqrestore(&kc->job_lock, flags);
300 }
301 
302 
303 static void push_head(struct list_head *jobs, struct kcopyd_job *job)
304 {
305 	unsigned long flags;
306 	struct dm_kcopyd_client *kc = job->kc;
307 
308 	spin_lock_irqsave(&kc->job_lock, flags);
309 	list_add(&job->list, jobs);
310 	spin_unlock_irqrestore(&kc->job_lock, flags);
311 }
312 
313 /*
314  * These three functions process 1 item from the corresponding
315  * job list.
316  *
317  * They return:
318  * < 0: error
319  *   0: success
320  * > 0: can't process yet.
321  */
322 static int run_complete_job(struct kcopyd_job *job)
323 {
324 	void *context = job->context;
325 	int read_err = job->read_err;
326 	unsigned long write_err = job->write_err;
327 	dm_kcopyd_notify_fn fn = job->fn;
328 	struct dm_kcopyd_client *kc = job->kc;
329 
330 	if (job->pages && job->pages != &zero_page_list)
331 		kcopyd_put_pages(kc, job->pages);
332 	/*
333 	 * If this is the master job, the sub jobs have already
334 	 * completed so we can free everything.
335 	 */
336 	if (job->master_job == job)
337 		mempool_free(job, kc->job_pool);
338 	fn(read_err, write_err, context);
339 
340 	if (atomic_dec_and_test(&kc->nr_jobs))
341 		wake_up(&kc->destroyq);
342 
343 	return 0;
344 }
345 
346 static void complete_io(unsigned long error, void *context)
347 {
348 	struct kcopyd_job *job = (struct kcopyd_job *) context;
349 	struct dm_kcopyd_client *kc = job->kc;
350 
351 	if (error) {
352 		if (job->rw == WRITE)
353 			job->write_err |= error;
354 		else
355 			job->read_err = 1;
356 
357 		if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
358 			push(&kc->complete_jobs, job);
359 			wake(kc);
360 			return;
361 		}
362 	}
363 
364 	if (job->rw == WRITE)
365 		push(&kc->complete_jobs, job);
366 
367 	else {
368 		job->rw = WRITE;
369 		push(&kc->io_jobs, job);
370 	}
371 
372 	wake(kc);
373 }
374 
375 /*
376  * Request io on as many buffer heads as we can currently get for
377  * a particular job.
378  */
379 static int run_io_job(struct kcopyd_job *job)
380 {
381 	int r;
382 	struct dm_io_request io_req = {
383 		.bi_rw = job->rw,
384 		.mem.type = DM_IO_PAGE_LIST,
385 		.mem.ptr.pl = job->pages,
386 		.mem.offset = 0,
387 		.notify.fn = complete_io,
388 		.notify.context = job,
389 		.client = job->kc->io_client,
390 	};
391 
392 	if (job->rw == READ)
393 		r = dm_io(&io_req, 1, &job->source, NULL);
394 	else
395 		r = dm_io(&io_req, job->num_dests, job->dests, NULL);
396 
397 	return r;
398 }
399 
400 static int run_pages_job(struct kcopyd_job *job)
401 {
402 	int r;
403 	unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
404 
405 	r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
406 	if (!r) {
407 		/* this job is ready for io */
408 		push(&job->kc->io_jobs, job);
409 		return 0;
410 	}
411 
412 	if (r == -ENOMEM)
413 		/* can't complete now */
414 		return 1;
415 
416 	return r;
417 }
418 
419 /*
420  * Run through a list for as long as possible.  Returns the count
421  * of successful jobs.
422  */
423 static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
424 			int (*fn) (struct kcopyd_job *))
425 {
426 	struct kcopyd_job *job;
427 	int r, count = 0;
428 
429 	while ((job = pop(jobs, kc))) {
430 
431 		r = fn(job);
432 
433 		if (r < 0) {
434 			/* error this rogue job */
435 			if (job->rw == WRITE)
436 				job->write_err = (unsigned long) -1L;
437 			else
438 				job->read_err = 1;
439 			push(&kc->complete_jobs, job);
440 			break;
441 		}
442 
443 		if (r > 0) {
444 			/*
445 			 * We couldn't service this job ATM, so
446 			 * push this job back onto the list.
447 			 */
448 			push_head(jobs, job);
449 			break;
450 		}
451 
452 		count++;
453 	}
454 
455 	return count;
456 }
457 
458 /*
459  * kcopyd does this every time it's woken up.
460  */
461 static void do_work(struct work_struct *work)
462 {
463 	struct dm_kcopyd_client *kc = container_of(work,
464 					struct dm_kcopyd_client, kcopyd_work);
465 	struct blk_plug plug;
466 
467 	/*
468 	 * The order that these are called is *very* important.
469 	 * complete jobs can free some pages for pages jobs.
470 	 * Pages jobs when successful will jump onto the io jobs
471 	 * list.  io jobs call wake when they complete and it all
472 	 * starts again.
473 	 */
474 	blk_start_plug(&plug);
475 	process_jobs(&kc->complete_jobs, kc, run_complete_job);
476 	process_jobs(&kc->pages_jobs, kc, run_pages_job);
477 	process_jobs(&kc->io_jobs, kc, run_io_job);
478 	blk_finish_plug(&plug);
479 }
480 
481 /*
482  * If we are copying a small region we just dispatch a single job
483  * to do the copy, otherwise the io has to be split up into many
484  * jobs.
485  */
486 static void dispatch_job(struct kcopyd_job *job)
487 {
488 	struct dm_kcopyd_client *kc = job->kc;
489 	atomic_inc(&kc->nr_jobs);
490 	if (unlikely(!job->source.count))
491 		push(&kc->complete_jobs, job);
492 	else if (job->pages == &zero_page_list)
493 		push(&kc->io_jobs, job);
494 	else
495 		push(&kc->pages_jobs, job);
496 	wake(kc);
497 }
498 
499 static void segment_complete(int read_err, unsigned long write_err,
500 			     void *context)
501 {
502 	/* FIXME: tidy this function */
503 	sector_t progress = 0;
504 	sector_t count = 0;
505 	struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
506 	struct kcopyd_job *job = sub_job->master_job;
507 	struct dm_kcopyd_client *kc = job->kc;
508 
509 	mutex_lock(&job->lock);
510 
511 	/* update the error */
512 	if (read_err)
513 		job->read_err = 1;
514 
515 	if (write_err)
516 		job->write_err |= write_err;
517 
518 	/*
519 	 * Only dispatch more work if there hasn't been an error.
520 	 */
521 	if ((!job->read_err && !job->write_err) ||
522 	    test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
523 		/* get the next chunk of work */
524 		progress = job->progress;
525 		count = job->source.count - progress;
526 		if (count) {
527 			if (count > SUB_JOB_SIZE)
528 				count = SUB_JOB_SIZE;
529 
530 			job->progress += count;
531 		}
532 	}
533 	mutex_unlock(&job->lock);
534 
535 	if (count) {
536 		int i;
537 
538 		*sub_job = *job;
539 		sub_job->source.sector += progress;
540 		sub_job->source.count = count;
541 
542 		for (i = 0; i < job->num_dests; i++) {
543 			sub_job->dests[i].sector += progress;
544 			sub_job->dests[i].count = count;
545 		}
546 
547 		sub_job->fn = segment_complete;
548 		sub_job->context = sub_job;
549 		dispatch_job(sub_job);
550 
551 	} else if (atomic_dec_and_test(&job->sub_jobs)) {
552 
553 		/*
554 		 * Queue the completion callback to the kcopyd thread.
555 		 *
556 		 * Some callers assume that all the completions are called
557 		 * from a single thread and don't race with each other.
558 		 *
559 		 * We must not call the callback directly here because this
560 		 * code may not be executing in the thread.
561 		 */
562 		push(&kc->complete_jobs, job);
563 		wake(kc);
564 	}
565 }
566 
567 /*
568  * Create some sub jobs to share the work between them.
569  */
570 static void split_job(struct kcopyd_job *master_job)
571 {
572 	int i;
573 
574 	atomic_inc(&master_job->kc->nr_jobs);
575 
576 	atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
577 	for (i = 0; i < SPLIT_COUNT; i++) {
578 		master_job[i + 1].master_job = master_job;
579 		segment_complete(0, 0u, &master_job[i + 1]);
580 	}
581 }
582 
583 int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
584 		   unsigned int num_dests, struct dm_io_region *dests,
585 		   unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
586 {
587 	struct kcopyd_job *job;
588 
589 	/*
590 	 * Allocate an array of jobs consisting of one master job
591 	 * followed by SPLIT_COUNT sub jobs.
592 	 */
593 	job = mempool_alloc(kc->job_pool, GFP_NOIO);
594 
595 	/*
596 	 * set up for the read.
597 	 */
598 	job->kc = kc;
599 	job->flags = flags;
600 	job->read_err = 0;
601 	job->write_err = 0;
602 
603 	job->num_dests = num_dests;
604 	memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
605 
606 	if (from) {
607 		job->source = *from;
608 		job->pages = NULL;
609 		job->rw = READ;
610 	} else {
611 		memset(&job->source, 0, sizeof job->source);
612 		job->source.count = job->dests[0].count;
613 		job->pages = &zero_page_list;
614 		job->rw = WRITE;
615 	}
616 
617 	job->fn = fn;
618 	job->context = context;
619 	job->master_job = job;
620 
621 	if (job->source.count <= SUB_JOB_SIZE)
622 		dispatch_job(job);
623 	else {
624 		mutex_init(&job->lock);
625 		job->progress = 0;
626 		split_job(job);
627 	}
628 
629 	return 0;
630 }
631 EXPORT_SYMBOL(dm_kcopyd_copy);
632 
633 int dm_kcopyd_zero(struct dm_kcopyd_client *kc,
634 		   unsigned num_dests, struct dm_io_region *dests,
635 		   unsigned flags, dm_kcopyd_notify_fn fn, void *context)
636 {
637 	return dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
638 }
639 EXPORT_SYMBOL(dm_kcopyd_zero);
640 
641 void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
642 				 dm_kcopyd_notify_fn fn, void *context)
643 {
644 	struct kcopyd_job *job;
645 
646 	job = mempool_alloc(kc->job_pool, GFP_NOIO);
647 
648 	memset(job, 0, sizeof(struct kcopyd_job));
649 	job->kc = kc;
650 	job->fn = fn;
651 	job->context = context;
652 	job->master_job = job;
653 
654 	atomic_inc(&kc->nr_jobs);
655 
656 	return job;
657 }
658 EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
659 
660 void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
661 {
662 	struct kcopyd_job *job = j;
663 	struct dm_kcopyd_client *kc = job->kc;
664 
665 	job->read_err = read_err;
666 	job->write_err = write_err;
667 
668 	push(&kc->complete_jobs, job);
669 	wake(kc);
670 }
671 EXPORT_SYMBOL(dm_kcopyd_do_callback);
672 
673 /*
674  * Cancels a kcopyd job, eg. someone might be deactivating a
675  * mirror.
676  */
677 #if 0
678 int kcopyd_cancel(struct kcopyd_job *job, int block)
679 {
680 	/* FIXME: finish */
681 	return -1;
682 }
683 #endif  /*  0  */
684 
685 /*-----------------------------------------------------------------
686  * Client setup
687  *---------------------------------------------------------------*/
688 struct dm_kcopyd_client *dm_kcopyd_client_create(void)
689 {
690 	int r = -ENOMEM;
691 	struct dm_kcopyd_client *kc;
692 
693 	kc = kmalloc(sizeof(*kc), GFP_KERNEL);
694 	if (!kc)
695 		return ERR_PTR(-ENOMEM);
696 
697 	spin_lock_init(&kc->job_lock);
698 	INIT_LIST_HEAD(&kc->complete_jobs);
699 	INIT_LIST_HEAD(&kc->io_jobs);
700 	INIT_LIST_HEAD(&kc->pages_jobs);
701 
702 	kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
703 	if (!kc->job_pool)
704 		goto bad_slab;
705 
706 	INIT_WORK(&kc->kcopyd_work, do_work);
707 	kc->kcopyd_wq = alloc_workqueue("kcopyd",
708 					WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
709 	if (!kc->kcopyd_wq)
710 		goto bad_workqueue;
711 
712 	kc->pages = NULL;
713 	kc->nr_reserved_pages = kc->nr_free_pages = 0;
714 	r = client_reserve_pages(kc, RESERVE_PAGES);
715 	if (r)
716 		goto bad_client_pages;
717 
718 	kc->io_client = dm_io_client_create();
719 	if (IS_ERR(kc->io_client)) {
720 		r = PTR_ERR(kc->io_client);
721 		goto bad_io_client;
722 	}
723 
724 	init_waitqueue_head(&kc->destroyq);
725 	atomic_set(&kc->nr_jobs, 0);
726 
727 	return kc;
728 
729 bad_io_client:
730 	client_free_pages(kc);
731 bad_client_pages:
732 	destroy_workqueue(kc->kcopyd_wq);
733 bad_workqueue:
734 	mempool_destroy(kc->job_pool);
735 bad_slab:
736 	kfree(kc);
737 
738 	return ERR_PTR(r);
739 }
740 EXPORT_SYMBOL(dm_kcopyd_client_create);
741 
742 void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
743 {
744 	/* Wait for completion of all jobs submitted by this client. */
745 	wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
746 
747 	BUG_ON(!list_empty(&kc->complete_jobs));
748 	BUG_ON(!list_empty(&kc->io_jobs));
749 	BUG_ON(!list_empty(&kc->pages_jobs));
750 	destroy_workqueue(kc->kcopyd_wq);
751 	dm_io_client_destroy(kc->io_client);
752 	client_free_pages(kc);
753 	mempool_destroy(kc->job_pool);
754 	kfree(kc);
755 }
756 EXPORT_SYMBOL(dm_kcopyd_client_destroy);
757