Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2002 Sistina Software (UK) Limited.
4 * Copyright (C) 2006 Red Hat GmbH
5 *
6 * This file is released under the GPL.
7 *
8 * Kcopyd provides a simple interface for copying an area of one
9 * block-device to one or more other block-devices, with an asynchronous
10 * completion notification.
11 */
12
13#include <linux/types.h>
14#include <linux/atomic.h>
15#include <linux/blkdev.h>
16#include <linux/fs.h>
17#include <linux/init.h>
18#include <linux/list.h>
19#include <linux/mempool.h>
20#include <linux/module.h>
21#include <linux/pagemap.h>
22#include <linux/slab.h>
23#include <linux/vmalloc.h>
24#include <linux/workqueue.h>
25#include <linux/mutex.h>
26#include <linux/delay.h>
27#include <linux/device-mapper.h>
28#include <linux/dm-kcopyd.h>
29
30#include "dm-core.h"
31
32#define SPLIT_COUNT 8
33#define MIN_JOBS 8
34
35#define DEFAULT_SUB_JOB_SIZE_KB 512
36#define MAX_SUB_JOB_SIZE_KB 1024
37
38static unsigned int kcopyd_subjob_size_kb = DEFAULT_SUB_JOB_SIZE_KB;
39
40module_param(kcopyd_subjob_size_kb, uint, 0644);
41MODULE_PARM_DESC(kcopyd_subjob_size_kb, "Sub-job size for dm-kcopyd clients");
42
43static unsigned int dm_get_kcopyd_subjob_size(void)
44{
45 unsigned int sub_job_size_kb;
46
47 sub_job_size_kb = __dm_get_module_param(&kcopyd_subjob_size_kb,
48 DEFAULT_SUB_JOB_SIZE_KB,
49 MAX_SUB_JOB_SIZE_KB);
50
51 return sub_job_size_kb << 1;
52}
53
54/*
55 *----------------------------------------------------------------
56 * Each kcopyd client has its own little pool of preallocated
57 * pages for kcopyd io.
58 *---------------------------------------------------------------
59 */
60struct dm_kcopyd_client {
61 struct page_list *pages;
62 unsigned int nr_reserved_pages;
63 unsigned int nr_free_pages;
64 unsigned int sub_job_size;
65
66 struct dm_io_client *io_client;
67
68 wait_queue_head_t destroyq;
69
70 mempool_t job_pool;
71
72 struct workqueue_struct *kcopyd_wq;
73 struct work_struct kcopyd_work;
74
75 struct dm_kcopyd_throttle *throttle;
76
77 atomic_t nr_jobs;
78
79/*
80 * We maintain four lists of jobs:
81 *
82 * i) jobs waiting for pages
83 * ii) jobs that have pages, and are waiting for the io to be issued.
84 * iii) jobs that don't need to do any IO and just run a callback
85 * iv) jobs that have completed.
86 *
87 * All four of these are protected by job_lock.
88 */
89 spinlock_t job_lock;
90 struct list_head callback_jobs;
91 struct list_head complete_jobs;
92 struct list_head io_jobs;
93 struct list_head pages_jobs;
94};
95
96static struct page_list zero_page_list;
97
98static DEFINE_SPINLOCK(throttle_spinlock);
99
100/*
101 * IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
102 * When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
103 * by 2.
104 */
105#define ACCOUNT_INTERVAL_SHIFT SHIFT_HZ
106
107/*
108 * Sleep this number of milliseconds.
109 *
110 * The value was decided experimentally.
111 * Smaller values seem to cause an increased copy rate above the limit.
112 * The reason for this is unknown but possibly due to jiffies rounding errors
113 * or read/write cache inside the disk.
114 */
115#define SLEEP_USEC 100000
116
117/*
118 * Maximum number of sleep events. There is a theoretical livelock if more
119 * kcopyd clients do work simultaneously which this limit avoids.
120 */
121#define MAX_SLEEPS 10
122
123static void io_job_start(struct dm_kcopyd_throttle *t)
124{
125 unsigned int throttle, now, difference;
126 int slept = 0, skew;
127
128 if (unlikely(!t))
129 return;
130
131try_again:
132 spin_lock_irq(&throttle_spinlock);
133
134 throttle = READ_ONCE(t->throttle);
135
136 if (likely(throttle >= 100))
137 goto skip_limit;
138
139 now = jiffies;
140 difference = now - t->last_jiffies;
141 t->last_jiffies = now;
142 if (t->num_io_jobs)
143 t->io_period += difference;
144 t->total_period += difference;
145
146 /*
147 * Maintain sane values if we got a temporary overflow.
148 */
149 if (unlikely(t->io_period > t->total_period))
150 t->io_period = t->total_period;
151
152 if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
153 int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
154
155 t->total_period >>= shift;
156 t->io_period >>= shift;
157 }
158
159 skew = t->io_period - throttle * t->total_period / 100;
160
161 if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
162 slept++;
163 spin_unlock_irq(&throttle_spinlock);
164 fsleep(SLEEP_USEC);
165 goto try_again;
166 }
167
168skip_limit:
169 t->num_io_jobs++;
170
171 spin_unlock_irq(&throttle_spinlock);
172}
173
174static void io_job_finish(struct dm_kcopyd_throttle *t)
175{
176 unsigned long flags;
177
178 if (unlikely(!t))
179 return;
180
181 spin_lock_irqsave(&throttle_spinlock, flags);
182
183 t->num_io_jobs--;
184
185 if (likely(READ_ONCE(t->throttle) >= 100))
186 goto skip_limit;
187
188 if (!t->num_io_jobs) {
189 unsigned int now, difference;
190
191 now = jiffies;
192 difference = now - t->last_jiffies;
193 t->last_jiffies = now;
194
195 t->io_period += difference;
196 t->total_period += difference;
197
198 /*
199 * Maintain sane values if we got a temporary overflow.
200 */
201 if (unlikely(t->io_period > t->total_period))
202 t->io_period = t->total_period;
203 }
204
205skip_limit:
206 spin_unlock_irqrestore(&throttle_spinlock, flags);
207}
208
209
210static void wake(struct dm_kcopyd_client *kc)
211{
212 queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
213}
214
215/*
216 * Obtain one page for the use of kcopyd.
217 */
218static struct page_list *alloc_pl(gfp_t gfp)
219{
220 struct page_list *pl;
221
222 pl = kmalloc(sizeof(*pl), gfp);
223 if (!pl)
224 return NULL;
225
226 pl->page = alloc_page(gfp | __GFP_HIGHMEM);
227 if (!pl->page) {
228 kfree(pl);
229 return NULL;
230 }
231
232 return pl;
233}
234
235static void free_pl(struct page_list *pl)
236{
237 __free_page(pl->page);
238 kfree(pl);
239}
240
241/*
242 * Add the provided pages to a client's free page list, releasing
243 * back to the system any beyond the reserved_pages limit.
244 */
245static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
246{
247 struct page_list *next;
248
249 do {
250 next = pl->next;
251
252 if (kc->nr_free_pages >= kc->nr_reserved_pages)
253 free_pl(pl);
254 else {
255 pl->next = kc->pages;
256 kc->pages = pl;
257 kc->nr_free_pages++;
258 }
259
260 pl = next;
261 } while (pl);
262}
263
264static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
265 unsigned int nr, struct page_list **pages)
266{
267 struct page_list *pl;
268
269 *pages = NULL;
270
271 do {
272 pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
273 if (unlikely(!pl)) {
274 /* Use reserved pages */
275 pl = kc->pages;
276 if (unlikely(!pl))
277 goto out_of_memory;
278 kc->pages = pl->next;
279 kc->nr_free_pages--;
280 }
281 pl->next = *pages;
282 *pages = pl;
283 } while (--nr);
284
285 return 0;
286
287out_of_memory:
288 if (*pages)
289 kcopyd_put_pages(kc, *pages);
290 return -ENOMEM;
291}
292
293/*
294 * These three functions resize the page pool.
295 */
296static void drop_pages(struct page_list *pl)
297{
298 struct page_list *next;
299
300 while (pl) {
301 next = pl->next;
302 free_pl(pl);
303 pl = next;
304 }
305}
306
307/*
308 * Allocate and reserve nr_pages for the use of a specific client.
309 */
310static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned int nr_pages)
311{
312 unsigned int i;
313 struct page_list *pl = NULL, *next;
314
315 for (i = 0; i < nr_pages; i++) {
316 next = alloc_pl(GFP_KERNEL);
317 if (!next) {
318 if (pl)
319 drop_pages(pl);
320 return -ENOMEM;
321 }
322 next->next = pl;
323 pl = next;
324 }
325
326 kc->nr_reserved_pages += nr_pages;
327 kcopyd_put_pages(kc, pl);
328
329 return 0;
330}
331
332static void client_free_pages(struct dm_kcopyd_client *kc)
333{
334 BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
335 drop_pages(kc->pages);
336 kc->pages = NULL;
337 kc->nr_free_pages = kc->nr_reserved_pages = 0;
338}
339
340/*
341 *---------------------------------------------------------------
342 * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
343 * for this reason we use a mempool to prevent the client from
344 * ever having to do io (which could cause a deadlock).
345 *---------------------------------------------------------------
346 */
347struct kcopyd_job {
348 struct dm_kcopyd_client *kc;
349 struct list_head list;
350 unsigned int flags;
351
352 /*
353 * Error state of the job.
354 */
355 int read_err;
356 unsigned long write_err;
357
358 /*
359 * REQ_OP_READ, REQ_OP_WRITE or REQ_OP_WRITE_ZEROES.
360 */
361 enum req_op op;
362 struct dm_io_region source;
363
364 /*
365 * The destinations for the transfer.
366 */
367 unsigned int num_dests;
368 struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
369
370 struct page_list *pages;
371
372 /*
373 * Set this to ensure you are notified when the job has
374 * completed. 'context' is for callback to use.
375 */
376 dm_kcopyd_notify_fn fn;
377 void *context;
378
379 /*
380 * These fields are only used if the job has been split
381 * into more manageable parts.
382 */
383 struct mutex lock;
384 atomic_t sub_jobs;
385 sector_t progress;
386 sector_t write_offset;
387
388 struct kcopyd_job *master_job;
389};
390
391static struct kmem_cache *_job_cache;
392
393int __init dm_kcopyd_init(void)
394{
395 _job_cache = kmem_cache_create("kcopyd_job",
396 sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
397 __alignof__(struct kcopyd_job), 0, NULL);
398 if (!_job_cache)
399 return -ENOMEM;
400
401 zero_page_list.next = &zero_page_list;
402 zero_page_list.page = ZERO_PAGE(0);
403
404 return 0;
405}
406
407void dm_kcopyd_exit(void)
408{
409 kmem_cache_destroy(_job_cache);
410 _job_cache = NULL;
411}
412
413/*
414 * Functions to push and pop a job onto the head of a given job
415 * list.
416 */
417static struct kcopyd_job *pop_io_job(struct list_head *jobs,
418 struct dm_kcopyd_client *kc)
419{
420 struct kcopyd_job *job;
421
422 /*
423 * For I/O jobs, pop any read, any write without sequential write
424 * constraint and sequential writes that are at the right position.
425 */
426 list_for_each_entry(job, jobs, list) {
427 if (job->op == REQ_OP_READ ||
428 !(job->flags & BIT(DM_KCOPYD_WRITE_SEQ))) {
429 list_del(&job->list);
430 return job;
431 }
432
433 if (job->write_offset == job->master_job->write_offset) {
434 job->master_job->write_offset += job->source.count;
435 list_del(&job->list);
436 return job;
437 }
438 }
439
440 return NULL;
441}
442
443static struct kcopyd_job *pop(struct list_head *jobs,
444 struct dm_kcopyd_client *kc)
445{
446 struct kcopyd_job *job = NULL;
447
448 spin_lock_irq(&kc->job_lock);
449
450 if (!list_empty(jobs)) {
451 if (jobs == &kc->io_jobs)
452 job = pop_io_job(jobs, kc);
453 else {
454 job = list_entry(jobs->next, struct kcopyd_job, list);
455 list_del(&job->list);
456 }
457 }
458 spin_unlock_irq(&kc->job_lock);
459
460 return job;
461}
462
463static void push(struct list_head *jobs, struct kcopyd_job *job)
464{
465 unsigned long flags;
466 struct dm_kcopyd_client *kc = job->kc;
467
468 spin_lock_irqsave(&kc->job_lock, flags);
469 list_add_tail(&job->list, jobs);
470 spin_unlock_irqrestore(&kc->job_lock, flags);
471}
472
473
474static void push_head(struct list_head *jobs, struct kcopyd_job *job)
475{
476 struct dm_kcopyd_client *kc = job->kc;
477
478 spin_lock_irq(&kc->job_lock);
479 list_add(&job->list, jobs);
480 spin_unlock_irq(&kc->job_lock);
481}
482
483/*
484 * These three functions process 1 item from the corresponding
485 * job list.
486 *
487 * They return:
488 * < 0: error
489 * 0: success
490 * > 0: can't process yet.
491 */
492static int run_complete_job(struct kcopyd_job *job)
493{
494 void *context = job->context;
495 int read_err = job->read_err;
496 unsigned long write_err = job->write_err;
497 dm_kcopyd_notify_fn fn = job->fn;
498 struct dm_kcopyd_client *kc = job->kc;
499
500 if (job->pages && job->pages != &zero_page_list)
501 kcopyd_put_pages(kc, job->pages);
502 /*
503 * If this is the master job, the sub jobs have already
504 * completed so we can free everything.
505 */
506 if (job->master_job == job) {
507 mutex_destroy(&job->lock);
508 mempool_free(job, &kc->job_pool);
509 }
510 fn(read_err, write_err, context);
511
512 if (atomic_dec_and_test(&kc->nr_jobs))
513 wake_up(&kc->destroyq);
514
515 cond_resched();
516
517 return 0;
518}
519
520static void complete_io(unsigned long error, void *context)
521{
522 struct kcopyd_job *job = context;
523 struct dm_kcopyd_client *kc = job->kc;
524
525 io_job_finish(kc->throttle);
526
527 if (error) {
528 if (op_is_write(job->op))
529 job->write_err |= error;
530 else
531 job->read_err = 1;
532
533 if (!(job->flags & BIT(DM_KCOPYD_IGNORE_ERROR))) {
534 push(&kc->complete_jobs, job);
535 wake(kc);
536 return;
537 }
538 }
539
540 if (op_is_write(job->op))
541 push(&kc->complete_jobs, job);
542
543 else {
544 job->op = REQ_OP_WRITE;
545 push(&kc->io_jobs, job);
546 }
547
548 wake(kc);
549}
550
551/*
552 * Request io on as many buffer heads as we can currently get for
553 * a particular job.
554 */
555static int run_io_job(struct kcopyd_job *job)
556{
557 int r;
558 struct dm_io_request io_req = {
559 .bi_opf = job->op,
560 .mem.type = DM_IO_PAGE_LIST,
561 .mem.ptr.pl = job->pages,
562 .mem.offset = 0,
563 .notify.fn = complete_io,
564 .notify.context = job,
565 .client = job->kc->io_client,
566 };
567
568 /*
569 * If we need to write sequentially and some reads or writes failed,
570 * no point in continuing.
571 */
572 if (job->flags & BIT(DM_KCOPYD_WRITE_SEQ) &&
573 job->master_job->write_err) {
574 job->write_err = job->master_job->write_err;
575 return -EIO;
576 }
577
578 io_job_start(job->kc->throttle);
579
580 if (job->op == REQ_OP_READ)
581 r = dm_io(&io_req, 1, &job->source, NULL, IOPRIO_DEFAULT);
582 else
583 r = dm_io(&io_req, job->num_dests, job->dests, NULL, IOPRIO_DEFAULT);
584
585 return r;
586}
587
588static int run_pages_job(struct kcopyd_job *job)
589{
590 int r;
591 unsigned int nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
592
593 r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
594 if (!r) {
595 /* this job is ready for io */
596 push(&job->kc->io_jobs, job);
597 return 0;
598 }
599
600 if (r == -ENOMEM)
601 /* can't complete now */
602 return 1;
603
604 return r;
605}
606
607/*
608 * Run through a list for as long as possible. Returns the count
609 * of successful jobs.
610 */
611static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
612 int (*fn)(struct kcopyd_job *))
613{
614 struct kcopyd_job *job;
615 int r, count = 0;
616
617 while ((job = pop(jobs, kc))) {
618
619 r = fn(job);
620
621 if (r < 0) {
622 /* error this rogue job */
623 if (op_is_write(job->op))
624 job->write_err = (unsigned long) -1L;
625 else
626 job->read_err = 1;
627 push(&kc->complete_jobs, job);
628 wake(kc);
629 break;
630 }
631
632 if (r > 0) {
633 /*
634 * We couldn't service this job ATM, so
635 * push this job back onto the list.
636 */
637 push_head(jobs, job);
638 break;
639 }
640
641 count++;
642 }
643
644 return count;
645}
646
647/*
648 * kcopyd does this every time it's woken up.
649 */
650static void do_work(struct work_struct *work)
651{
652 struct dm_kcopyd_client *kc = container_of(work,
653 struct dm_kcopyd_client, kcopyd_work);
654 struct blk_plug plug;
655
656 /*
657 * The order that these are called is *very* important.
658 * complete jobs can free some pages for pages jobs.
659 * Pages jobs when successful will jump onto the io jobs
660 * list. io jobs call wake when they complete and it all
661 * starts again.
662 */
663 spin_lock_irq(&kc->job_lock);
664 list_splice_tail_init(&kc->callback_jobs, &kc->complete_jobs);
665 spin_unlock_irq(&kc->job_lock);
666
667 blk_start_plug(&plug);
668 process_jobs(&kc->complete_jobs, kc, run_complete_job);
669 process_jobs(&kc->pages_jobs, kc, run_pages_job);
670 process_jobs(&kc->io_jobs, kc, run_io_job);
671 blk_finish_plug(&plug);
672}
673
674/*
675 * If we are copying a small region we just dispatch a single job
676 * to do the copy, otherwise the io has to be split up into many
677 * jobs.
678 */
679static void dispatch_job(struct kcopyd_job *job)
680{
681 struct dm_kcopyd_client *kc = job->kc;
682
683 atomic_inc(&kc->nr_jobs);
684 if (unlikely(!job->source.count))
685 push(&kc->callback_jobs, job);
686 else if (job->pages == &zero_page_list)
687 push(&kc->io_jobs, job);
688 else
689 push(&kc->pages_jobs, job);
690 wake(kc);
691}
692
693static void segment_complete(int read_err, unsigned long write_err,
694 void *context)
695{
696 /* FIXME: tidy this function */
697 sector_t progress = 0;
698 sector_t count = 0;
699 struct kcopyd_job *sub_job = context;
700 struct kcopyd_job *job = sub_job->master_job;
701 struct dm_kcopyd_client *kc = job->kc;
702
703 mutex_lock(&job->lock);
704
705 /* update the error */
706 if (read_err)
707 job->read_err = 1;
708
709 if (write_err)
710 job->write_err |= write_err;
711
712 /*
713 * Only dispatch more work if there hasn't been an error.
714 */
715 if ((!job->read_err && !job->write_err) ||
716 job->flags & BIT(DM_KCOPYD_IGNORE_ERROR)) {
717 /* get the next chunk of work */
718 progress = job->progress;
719 count = job->source.count - progress;
720 if (count) {
721 if (count > kc->sub_job_size)
722 count = kc->sub_job_size;
723
724 job->progress += count;
725 }
726 }
727 mutex_unlock(&job->lock);
728
729 if (count) {
730 int i;
731
732 *sub_job = *job;
733 sub_job->write_offset = progress;
734 sub_job->source.sector += progress;
735 sub_job->source.count = count;
736
737 for (i = 0; i < job->num_dests; i++) {
738 sub_job->dests[i].sector += progress;
739 sub_job->dests[i].count = count;
740 }
741
742 sub_job->fn = segment_complete;
743 sub_job->context = sub_job;
744 dispatch_job(sub_job);
745
746 } else if (atomic_dec_and_test(&job->sub_jobs)) {
747
748 /*
749 * Queue the completion callback to the kcopyd thread.
750 *
751 * Some callers assume that all the completions are called
752 * from a single thread and don't race with each other.
753 *
754 * We must not call the callback directly here because this
755 * code may not be executing in the thread.
756 */
757 push(&kc->complete_jobs, job);
758 wake(kc);
759 }
760}
761
762/*
763 * Create some sub jobs to share the work between them.
764 */
765static void split_job(struct kcopyd_job *master_job)
766{
767 int i;
768
769 atomic_inc(&master_job->kc->nr_jobs);
770
771 atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
772 for (i = 0; i < SPLIT_COUNT; i++) {
773 master_job[i + 1].master_job = master_job;
774 segment_complete(0, 0u, &master_job[i + 1]);
775 }
776}
777
778void dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
779 unsigned int num_dests, struct dm_io_region *dests,
780 unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
781{
782 struct kcopyd_job *job;
783 int i;
784
785 /*
786 * Allocate an array of jobs consisting of one master job
787 * followed by SPLIT_COUNT sub jobs.
788 */
789 job = mempool_alloc(&kc->job_pool, GFP_NOIO);
790 mutex_init(&job->lock);
791
792 /*
793 * set up for the read.
794 */
795 job->kc = kc;
796 job->flags = flags;
797 job->read_err = 0;
798 job->write_err = 0;
799
800 job->num_dests = num_dests;
801 memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
802
803 /*
804 * If one of the destination is a host-managed zoned block device,
805 * we need to write sequentially. If one of the destination is a
806 * host-aware device, then leave it to the caller to choose what to do.
807 */
808 if (!(job->flags & BIT(DM_KCOPYD_WRITE_SEQ))) {
809 for (i = 0; i < job->num_dests; i++) {
810 if (bdev_is_zoned(dests[i].bdev)) {
811 job->flags |= BIT(DM_KCOPYD_WRITE_SEQ);
812 break;
813 }
814 }
815 }
816
817 /*
818 * If we need to write sequentially, errors cannot be ignored.
819 */
820 if (job->flags & BIT(DM_KCOPYD_WRITE_SEQ) &&
821 job->flags & BIT(DM_KCOPYD_IGNORE_ERROR))
822 job->flags &= ~BIT(DM_KCOPYD_IGNORE_ERROR);
823
824 if (from) {
825 job->source = *from;
826 job->pages = NULL;
827 job->op = REQ_OP_READ;
828 } else {
829 memset(&job->source, 0, sizeof(job->source));
830 job->source.count = job->dests[0].count;
831 job->pages = &zero_page_list;
832
833 /*
834 * Use WRITE ZEROES to optimize zeroing if all dests support it.
835 */
836 job->op = REQ_OP_WRITE_ZEROES;
837 for (i = 0; i < job->num_dests; i++)
838 if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
839 job->op = REQ_OP_WRITE;
840 break;
841 }
842 }
843
844 job->fn = fn;
845 job->context = context;
846 job->master_job = job;
847 job->write_offset = 0;
848
849 if (job->source.count <= kc->sub_job_size)
850 dispatch_job(job);
851 else {
852 job->progress = 0;
853 split_job(job);
854 }
855}
856EXPORT_SYMBOL(dm_kcopyd_copy);
857
858void dm_kcopyd_zero(struct dm_kcopyd_client *kc,
859 unsigned int num_dests, struct dm_io_region *dests,
860 unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
861{
862 dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
863}
864EXPORT_SYMBOL(dm_kcopyd_zero);
865
866void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
867 dm_kcopyd_notify_fn fn, void *context)
868{
869 struct kcopyd_job *job;
870
871 job = mempool_alloc(&kc->job_pool, GFP_NOIO);
872
873 memset(job, 0, sizeof(struct kcopyd_job));
874 job->kc = kc;
875 job->fn = fn;
876 job->context = context;
877 job->master_job = job;
878
879 atomic_inc(&kc->nr_jobs);
880
881 return job;
882}
883EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
884
885void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
886{
887 struct kcopyd_job *job = j;
888 struct dm_kcopyd_client *kc = job->kc;
889
890 job->read_err = read_err;
891 job->write_err = write_err;
892
893 push(&kc->callback_jobs, job);
894 wake(kc);
895}
896EXPORT_SYMBOL(dm_kcopyd_do_callback);
897
898/*
899 * Cancels a kcopyd job, eg. someone might be deactivating a
900 * mirror.
901 */
902#if 0
903int kcopyd_cancel(struct kcopyd_job *job, int block)
904{
905 /* FIXME: finish */
906 return -1;
907}
908#endif /* 0 */
909
910/*
911 *---------------------------------------------------------------
912 * Client setup
913 *---------------------------------------------------------------
914 */
915struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
916{
917 int r;
918 unsigned int reserve_pages;
919 struct dm_kcopyd_client *kc;
920
921 kc = kzalloc(sizeof(*kc), GFP_KERNEL);
922 if (!kc)
923 return ERR_PTR(-ENOMEM);
924
925 spin_lock_init(&kc->job_lock);
926 INIT_LIST_HEAD(&kc->callback_jobs);
927 INIT_LIST_HEAD(&kc->complete_jobs);
928 INIT_LIST_HEAD(&kc->io_jobs);
929 INIT_LIST_HEAD(&kc->pages_jobs);
930 kc->throttle = throttle;
931
932 r = mempool_init_slab_pool(&kc->job_pool, MIN_JOBS, _job_cache);
933 if (r)
934 goto bad_slab;
935
936 INIT_WORK(&kc->kcopyd_work, do_work);
937 kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
938 if (!kc->kcopyd_wq) {
939 r = -ENOMEM;
940 goto bad_workqueue;
941 }
942
943 kc->sub_job_size = dm_get_kcopyd_subjob_size();
944 reserve_pages = DIV_ROUND_UP(kc->sub_job_size << SECTOR_SHIFT, PAGE_SIZE);
945
946 kc->pages = NULL;
947 kc->nr_reserved_pages = kc->nr_free_pages = 0;
948 r = client_reserve_pages(kc, reserve_pages);
949 if (r)
950 goto bad_client_pages;
951
952 kc->io_client = dm_io_client_create();
953 if (IS_ERR(kc->io_client)) {
954 r = PTR_ERR(kc->io_client);
955 goto bad_io_client;
956 }
957
958 init_waitqueue_head(&kc->destroyq);
959 atomic_set(&kc->nr_jobs, 0);
960
961 return kc;
962
963bad_io_client:
964 client_free_pages(kc);
965bad_client_pages:
966 destroy_workqueue(kc->kcopyd_wq);
967bad_workqueue:
968 mempool_exit(&kc->job_pool);
969bad_slab:
970 kfree(kc);
971
972 return ERR_PTR(r);
973}
974EXPORT_SYMBOL(dm_kcopyd_client_create);
975
976void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
977{
978 /* Wait for completion of all jobs submitted by this client. */
979 wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
980
981 BUG_ON(!list_empty(&kc->callback_jobs));
982 BUG_ON(!list_empty(&kc->complete_jobs));
983 BUG_ON(!list_empty(&kc->io_jobs));
984 BUG_ON(!list_empty(&kc->pages_jobs));
985 destroy_workqueue(kc->kcopyd_wq);
986 dm_io_client_destroy(kc->io_client);
987 client_free_pages(kc);
988 mempool_exit(&kc->job_pool);
989 kfree(kc);
990}
991EXPORT_SYMBOL(dm_kcopyd_client_destroy);
992
993void dm_kcopyd_client_flush(struct dm_kcopyd_client *kc)
994{
995 flush_workqueue(kc->kcopyd_wq);
996}
997EXPORT_SYMBOL(dm_kcopyd_client_flush);
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 *---------------------------------------------------------------*/
40struct 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 atomic_t nr_jobs;
49
50 mempool_t *job_pool;
51
52 struct workqueue_struct *kcopyd_wq;
53 struct work_struct kcopyd_work;
54
55 struct dm_kcopyd_throttle *throttle;
56
57/*
58 * We maintain three lists of jobs:
59 *
60 * i) jobs waiting for pages
61 * ii) jobs that have pages, and are waiting for the io to be issued.
62 * iii) jobs that have completed.
63 *
64 * All three of these are protected by job_lock.
65 */
66 spinlock_t job_lock;
67 struct list_head complete_jobs;
68 struct list_head io_jobs;
69 struct list_head pages_jobs;
70};
71
72static struct page_list zero_page_list;
73
74static DEFINE_SPINLOCK(throttle_spinlock);
75
76/*
77 * IO/IDLE accounting slowly decays after (1 << ACCOUNT_INTERVAL_SHIFT) period.
78 * When total_period >= (1 << ACCOUNT_INTERVAL_SHIFT) the counters are divided
79 * by 2.
80 */
81#define ACCOUNT_INTERVAL_SHIFT SHIFT_HZ
82
83/*
84 * Sleep this number of milliseconds.
85 *
86 * The value was decided experimentally.
87 * Smaller values seem to cause an increased copy rate above the limit.
88 * The reason for this is unknown but possibly due to jiffies rounding errors
89 * or read/write cache inside the disk.
90 */
91#define SLEEP_MSEC 100
92
93/*
94 * Maximum number of sleep events. There is a theoretical livelock if more
95 * kcopyd clients do work simultaneously which this limit avoids.
96 */
97#define MAX_SLEEPS 10
98
99static void io_job_start(struct dm_kcopyd_throttle *t)
100{
101 unsigned throttle, now, difference;
102 int slept = 0, skew;
103
104 if (unlikely(!t))
105 return;
106
107try_again:
108 spin_lock_irq(&throttle_spinlock);
109
110 throttle = READ_ONCE(t->throttle);
111
112 if (likely(throttle >= 100))
113 goto skip_limit;
114
115 now = jiffies;
116 difference = now - t->last_jiffies;
117 t->last_jiffies = now;
118 if (t->num_io_jobs)
119 t->io_period += difference;
120 t->total_period += difference;
121
122 /*
123 * Maintain sane values if we got a temporary overflow.
124 */
125 if (unlikely(t->io_period > t->total_period))
126 t->io_period = t->total_period;
127
128 if (unlikely(t->total_period >= (1 << ACCOUNT_INTERVAL_SHIFT))) {
129 int shift = fls(t->total_period >> ACCOUNT_INTERVAL_SHIFT);
130 t->total_period >>= shift;
131 t->io_period >>= shift;
132 }
133
134 skew = t->io_period - throttle * t->total_period / 100;
135
136 if (unlikely(skew > 0) && slept < MAX_SLEEPS) {
137 slept++;
138 spin_unlock_irq(&throttle_spinlock);
139 msleep(SLEEP_MSEC);
140 goto try_again;
141 }
142
143skip_limit:
144 t->num_io_jobs++;
145
146 spin_unlock_irq(&throttle_spinlock);
147}
148
149static void io_job_finish(struct dm_kcopyd_throttle *t)
150{
151 unsigned long flags;
152
153 if (unlikely(!t))
154 return;
155
156 spin_lock_irqsave(&throttle_spinlock, flags);
157
158 t->num_io_jobs--;
159
160 if (likely(READ_ONCE(t->throttle) >= 100))
161 goto skip_limit;
162
163 if (!t->num_io_jobs) {
164 unsigned now, difference;
165
166 now = jiffies;
167 difference = now - t->last_jiffies;
168 t->last_jiffies = now;
169
170 t->io_period += difference;
171 t->total_period += difference;
172
173 /*
174 * Maintain sane values if we got a temporary overflow.
175 */
176 if (unlikely(t->io_period > t->total_period))
177 t->io_period = t->total_period;
178 }
179
180skip_limit:
181 spin_unlock_irqrestore(&throttle_spinlock, flags);
182}
183
184
185static void wake(struct dm_kcopyd_client *kc)
186{
187 queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
188}
189
190/*
191 * Obtain one page for the use of kcopyd.
192 */
193static struct page_list *alloc_pl(gfp_t gfp)
194{
195 struct page_list *pl;
196
197 pl = kmalloc(sizeof(*pl), gfp);
198 if (!pl)
199 return NULL;
200
201 pl->page = alloc_page(gfp);
202 if (!pl->page) {
203 kfree(pl);
204 return NULL;
205 }
206
207 return pl;
208}
209
210static void free_pl(struct page_list *pl)
211{
212 __free_page(pl->page);
213 kfree(pl);
214}
215
216/*
217 * Add the provided pages to a client's free page list, releasing
218 * back to the system any beyond the reserved_pages limit.
219 */
220static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
221{
222 struct page_list *next;
223
224 do {
225 next = pl->next;
226
227 if (kc->nr_free_pages >= kc->nr_reserved_pages)
228 free_pl(pl);
229 else {
230 pl->next = kc->pages;
231 kc->pages = pl;
232 kc->nr_free_pages++;
233 }
234
235 pl = next;
236 } while (pl);
237}
238
239static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
240 unsigned int nr, struct page_list **pages)
241{
242 struct page_list *pl;
243
244 *pages = NULL;
245
246 do {
247 pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY | __GFP_KSWAPD_RECLAIM);
248 if (unlikely(!pl)) {
249 /* Use reserved pages */
250 pl = kc->pages;
251 if (unlikely(!pl))
252 goto out_of_memory;
253 kc->pages = pl->next;
254 kc->nr_free_pages--;
255 }
256 pl->next = *pages;
257 *pages = pl;
258 } while (--nr);
259
260 return 0;
261
262out_of_memory:
263 if (*pages)
264 kcopyd_put_pages(kc, *pages);
265 return -ENOMEM;
266}
267
268/*
269 * These three functions resize the page pool.
270 */
271static void drop_pages(struct page_list *pl)
272{
273 struct page_list *next;
274
275 while (pl) {
276 next = pl->next;
277 free_pl(pl);
278 pl = next;
279 }
280}
281
282/*
283 * Allocate and reserve nr_pages for the use of a specific client.
284 */
285static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
286{
287 unsigned i;
288 struct page_list *pl = NULL, *next;
289
290 for (i = 0; i < nr_pages; i++) {
291 next = alloc_pl(GFP_KERNEL);
292 if (!next) {
293 if (pl)
294 drop_pages(pl);
295 return -ENOMEM;
296 }
297 next->next = pl;
298 pl = next;
299 }
300
301 kc->nr_reserved_pages += nr_pages;
302 kcopyd_put_pages(kc, pl);
303
304 return 0;
305}
306
307static void client_free_pages(struct dm_kcopyd_client *kc)
308{
309 BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
310 drop_pages(kc->pages);
311 kc->pages = NULL;
312 kc->nr_free_pages = kc->nr_reserved_pages = 0;
313}
314
315/*-----------------------------------------------------------------
316 * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
317 * for this reason we use a mempool to prevent the client from
318 * ever having to do io (which could cause a deadlock).
319 *---------------------------------------------------------------*/
320struct kcopyd_job {
321 struct dm_kcopyd_client *kc;
322 struct list_head list;
323 unsigned long flags;
324
325 /*
326 * Error state of the job.
327 */
328 int read_err;
329 unsigned long write_err;
330
331 /*
332 * Either READ or WRITE
333 */
334 int rw;
335 struct dm_io_region source;
336
337 /*
338 * The destinations for the transfer.
339 */
340 unsigned int num_dests;
341 struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
342
343 struct page_list *pages;
344
345 /*
346 * Set this to ensure you are notified when the job has
347 * completed. 'context' is for callback to use.
348 */
349 dm_kcopyd_notify_fn fn;
350 void *context;
351
352 /*
353 * These fields are only used if the job has been split
354 * into more manageable parts.
355 */
356 struct mutex lock;
357 atomic_t sub_jobs;
358 sector_t progress;
359 sector_t write_offset;
360
361 struct kcopyd_job *master_job;
362};
363
364static struct kmem_cache *_job_cache;
365
366int __init dm_kcopyd_init(void)
367{
368 _job_cache = kmem_cache_create("kcopyd_job",
369 sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
370 __alignof__(struct kcopyd_job), 0, NULL);
371 if (!_job_cache)
372 return -ENOMEM;
373
374 zero_page_list.next = &zero_page_list;
375 zero_page_list.page = ZERO_PAGE(0);
376
377 return 0;
378}
379
380void dm_kcopyd_exit(void)
381{
382 kmem_cache_destroy(_job_cache);
383 _job_cache = NULL;
384}
385
386/*
387 * Functions to push and pop a job onto the head of a given job
388 * list.
389 */
390static struct kcopyd_job *pop_io_job(struct list_head *jobs,
391 struct dm_kcopyd_client *kc)
392{
393 struct kcopyd_job *job;
394
395 /*
396 * For I/O jobs, pop any read, any write without sequential write
397 * constraint and sequential writes that are at the right position.
398 */
399 list_for_each_entry(job, jobs, list) {
400 if (job->rw == READ || !test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
401 list_del(&job->list);
402 return job;
403 }
404
405 if (job->write_offset == job->master_job->write_offset) {
406 job->master_job->write_offset += job->source.count;
407 list_del(&job->list);
408 return job;
409 }
410 }
411
412 return NULL;
413}
414
415static struct kcopyd_job *pop(struct list_head *jobs,
416 struct dm_kcopyd_client *kc)
417{
418 struct kcopyd_job *job = NULL;
419 unsigned long flags;
420
421 spin_lock_irqsave(&kc->job_lock, flags);
422
423 if (!list_empty(jobs)) {
424 if (jobs == &kc->io_jobs)
425 job = pop_io_job(jobs, kc);
426 else {
427 job = list_entry(jobs->next, struct kcopyd_job, list);
428 list_del(&job->list);
429 }
430 }
431 spin_unlock_irqrestore(&kc->job_lock, flags);
432
433 return job;
434}
435
436static void push(struct list_head *jobs, struct kcopyd_job *job)
437{
438 unsigned long flags;
439 struct dm_kcopyd_client *kc = job->kc;
440
441 spin_lock_irqsave(&kc->job_lock, flags);
442 list_add_tail(&job->list, jobs);
443 spin_unlock_irqrestore(&kc->job_lock, flags);
444}
445
446
447static void push_head(struct list_head *jobs, struct kcopyd_job *job)
448{
449 unsigned long flags;
450 struct dm_kcopyd_client *kc = job->kc;
451
452 spin_lock_irqsave(&kc->job_lock, flags);
453 list_add(&job->list, jobs);
454 spin_unlock_irqrestore(&kc->job_lock, flags);
455}
456
457/*
458 * These three functions process 1 item from the corresponding
459 * job list.
460 *
461 * They return:
462 * < 0: error
463 * 0: success
464 * > 0: can't process yet.
465 */
466static int run_complete_job(struct kcopyd_job *job)
467{
468 void *context = job->context;
469 int read_err = job->read_err;
470 unsigned long write_err = job->write_err;
471 dm_kcopyd_notify_fn fn = job->fn;
472 struct dm_kcopyd_client *kc = job->kc;
473
474 if (job->pages && job->pages != &zero_page_list)
475 kcopyd_put_pages(kc, job->pages);
476 /*
477 * If this is the master job, the sub jobs have already
478 * completed so we can free everything.
479 */
480 if (job->master_job == job) {
481 mutex_destroy(&job->lock);
482 mempool_free(job, kc->job_pool);
483 }
484 fn(read_err, write_err, context);
485
486 if (atomic_dec_and_test(&kc->nr_jobs))
487 wake_up(&kc->destroyq);
488
489 return 0;
490}
491
492static void complete_io(unsigned long error, void *context)
493{
494 struct kcopyd_job *job = (struct kcopyd_job *) context;
495 struct dm_kcopyd_client *kc = job->kc;
496
497 io_job_finish(kc->throttle);
498
499 if (error) {
500 if (op_is_write(job->rw))
501 job->write_err |= error;
502 else
503 job->read_err = 1;
504
505 if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
506 push(&kc->complete_jobs, job);
507 wake(kc);
508 return;
509 }
510 }
511
512 if (op_is_write(job->rw))
513 push(&kc->complete_jobs, job);
514
515 else {
516 job->rw = WRITE;
517 push(&kc->io_jobs, job);
518 }
519
520 wake(kc);
521}
522
523/*
524 * Request io on as many buffer heads as we can currently get for
525 * a particular job.
526 */
527static int run_io_job(struct kcopyd_job *job)
528{
529 int r;
530 struct dm_io_request io_req = {
531 .bi_op = job->rw,
532 .bi_op_flags = 0,
533 .mem.type = DM_IO_PAGE_LIST,
534 .mem.ptr.pl = job->pages,
535 .mem.offset = 0,
536 .notify.fn = complete_io,
537 .notify.context = job,
538 .client = job->kc->io_client,
539 };
540
541 /*
542 * If we need to write sequentially and some reads or writes failed,
543 * no point in continuing.
544 */
545 if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
546 job->master_job->write_err)
547 return -EIO;
548
549 io_job_start(job->kc->throttle);
550
551 if (job->rw == READ)
552 r = dm_io(&io_req, 1, &job->source, NULL);
553 else
554 r = dm_io(&io_req, job->num_dests, job->dests, NULL);
555
556 return r;
557}
558
559static int run_pages_job(struct kcopyd_job *job)
560{
561 int r;
562 unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
563
564 r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
565 if (!r) {
566 /* this job is ready for io */
567 push(&job->kc->io_jobs, job);
568 return 0;
569 }
570
571 if (r == -ENOMEM)
572 /* can't complete now */
573 return 1;
574
575 return r;
576}
577
578/*
579 * Run through a list for as long as possible. Returns the count
580 * of successful jobs.
581 */
582static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
583 int (*fn) (struct kcopyd_job *))
584{
585 struct kcopyd_job *job;
586 int r, count = 0;
587
588 while ((job = pop(jobs, kc))) {
589
590 r = fn(job);
591
592 if (r < 0) {
593 /* error this rogue job */
594 if (op_is_write(job->rw))
595 job->write_err = (unsigned long) -1L;
596 else
597 job->read_err = 1;
598 push(&kc->complete_jobs, job);
599 break;
600 }
601
602 if (r > 0) {
603 /*
604 * We couldn't service this job ATM, so
605 * push this job back onto the list.
606 */
607 push_head(jobs, job);
608 break;
609 }
610
611 count++;
612 }
613
614 return count;
615}
616
617/*
618 * kcopyd does this every time it's woken up.
619 */
620static void do_work(struct work_struct *work)
621{
622 struct dm_kcopyd_client *kc = container_of(work,
623 struct dm_kcopyd_client, kcopyd_work);
624 struct blk_plug plug;
625
626 /*
627 * The order that these are called is *very* important.
628 * complete jobs can free some pages for pages jobs.
629 * Pages jobs when successful will jump onto the io jobs
630 * list. io jobs call wake when they complete and it all
631 * starts again.
632 */
633 blk_start_plug(&plug);
634 process_jobs(&kc->complete_jobs, kc, run_complete_job);
635 process_jobs(&kc->pages_jobs, kc, run_pages_job);
636 process_jobs(&kc->io_jobs, kc, run_io_job);
637 blk_finish_plug(&plug);
638}
639
640/*
641 * If we are copying a small region we just dispatch a single job
642 * to do the copy, otherwise the io has to be split up into many
643 * jobs.
644 */
645static void dispatch_job(struct kcopyd_job *job)
646{
647 struct dm_kcopyd_client *kc = job->kc;
648 atomic_inc(&kc->nr_jobs);
649 if (unlikely(!job->source.count))
650 push(&kc->complete_jobs, job);
651 else if (job->pages == &zero_page_list)
652 push(&kc->io_jobs, job);
653 else
654 push(&kc->pages_jobs, job);
655 wake(kc);
656}
657
658static void segment_complete(int read_err, unsigned long write_err,
659 void *context)
660{
661 /* FIXME: tidy this function */
662 sector_t progress = 0;
663 sector_t count = 0;
664 struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
665 struct kcopyd_job *job = sub_job->master_job;
666 struct dm_kcopyd_client *kc = job->kc;
667
668 mutex_lock(&job->lock);
669
670 /* update the error */
671 if (read_err)
672 job->read_err = 1;
673
674 if (write_err)
675 job->write_err |= write_err;
676
677 /*
678 * Only dispatch more work if there hasn't been an error.
679 */
680 if ((!job->read_err && !job->write_err) ||
681 test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
682 /* get the next chunk of work */
683 progress = job->progress;
684 count = job->source.count - progress;
685 if (count) {
686 if (count > SUB_JOB_SIZE)
687 count = SUB_JOB_SIZE;
688
689 job->progress += count;
690 }
691 }
692 mutex_unlock(&job->lock);
693
694 if (count) {
695 int i;
696
697 *sub_job = *job;
698 sub_job->write_offset = progress;
699 sub_job->source.sector += progress;
700 sub_job->source.count = count;
701
702 for (i = 0; i < job->num_dests; i++) {
703 sub_job->dests[i].sector += progress;
704 sub_job->dests[i].count = count;
705 }
706
707 sub_job->fn = segment_complete;
708 sub_job->context = sub_job;
709 dispatch_job(sub_job);
710
711 } else if (atomic_dec_and_test(&job->sub_jobs)) {
712
713 /*
714 * Queue the completion callback to the kcopyd thread.
715 *
716 * Some callers assume that all the completions are called
717 * from a single thread and don't race with each other.
718 *
719 * We must not call the callback directly here because this
720 * code may not be executing in the thread.
721 */
722 push(&kc->complete_jobs, job);
723 wake(kc);
724 }
725}
726
727/*
728 * Create some sub jobs to share the work between them.
729 */
730static void split_job(struct kcopyd_job *master_job)
731{
732 int i;
733
734 atomic_inc(&master_job->kc->nr_jobs);
735
736 atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
737 for (i = 0; i < SPLIT_COUNT; i++) {
738 master_job[i + 1].master_job = master_job;
739 segment_complete(0, 0u, &master_job[i + 1]);
740 }
741}
742
743int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
744 unsigned int num_dests, struct dm_io_region *dests,
745 unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
746{
747 struct kcopyd_job *job;
748 int i;
749
750 /*
751 * Allocate an array of jobs consisting of one master job
752 * followed by SPLIT_COUNT sub jobs.
753 */
754 job = mempool_alloc(kc->job_pool, GFP_NOIO);
755 mutex_init(&job->lock);
756
757 /*
758 * set up for the read.
759 */
760 job->kc = kc;
761 job->flags = flags;
762 job->read_err = 0;
763 job->write_err = 0;
764
765 job->num_dests = num_dests;
766 memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
767
768 /*
769 * If one of the destination is a host-managed zoned block device,
770 * we need to write sequentially. If one of the destination is a
771 * host-aware device, then leave it to the caller to choose what to do.
772 */
773 if (!test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags)) {
774 for (i = 0; i < job->num_dests; i++) {
775 if (bdev_zoned_model(dests[i].bdev) == BLK_ZONED_HM) {
776 set_bit(DM_KCOPYD_WRITE_SEQ, &job->flags);
777 break;
778 }
779 }
780 }
781
782 /*
783 * If we need to write sequentially, errors cannot be ignored.
784 */
785 if (test_bit(DM_KCOPYD_WRITE_SEQ, &job->flags) &&
786 test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags))
787 clear_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags);
788
789 if (from) {
790 job->source = *from;
791 job->pages = NULL;
792 job->rw = READ;
793 } else {
794 memset(&job->source, 0, sizeof job->source);
795 job->source.count = job->dests[0].count;
796 job->pages = &zero_page_list;
797
798 /*
799 * Use WRITE ZEROES to optimize zeroing if all dests support it.
800 */
801 job->rw = REQ_OP_WRITE_ZEROES;
802 for (i = 0; i < job->num_dests; i++)
803 if (!bdev_write_zeroes_sectors(job->dests[i].bdev)) {
804 job->rw = WRITE;
805 break;
806 }
807 }
808
809 job->fn = fn;
810 job->context = context;
811 job->master_job = job;
812 job->write_offset = 0;
813
814 if (job->source.count <= SUB_JOB_SIZE)
815 dispatch_job(job);
816 else {
817 job->progress = 0;
818 split_job(job);
819 }
820
821 return 0;
822}
823EXPORT_SYMBOL(dm_kcopyd_copy);
824
825int dm_kcopyd_zero(struct dm_kcopyd_client *kc,
826 unsigned num_dests, struct dm_io_region *dests,
827 unsigned flags, dm_kcopyd_notify_fn fn, void *context)
828{
829 return dm_kcopyd_copy(kc, NULL, num_dests, dests, flags, fn, context);
830}
831EXPORT_SYMBOL(dm_kcopyd_zero);
832
833void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
834 dm_kcopyd_notify_fn fn, void *context)
835{
836 struct kcopyd_job *job;
837
838 job = mempool_alloc(kc->job_pool, GFP_NOIO);
839
840 memset(job, 0, sizeof(struct kcopyd_job));
841 job->kc = kc;
842 job->fn = fn;
843 job->context = context;
844 job->master_job = job;
845
846 atomic_inc(&kc->nr_jobs);
847
848 return job;
849}
850EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
851
852void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
853{
854 struct kcopyd_job *job = j;
855 struct dm_kcopyd_client *kc = job->kc;
856
857 job->read_err = read_err;
858 job->write_err = write_err;
859
860 push(&kc->complete_jobs, job);
861 wake(kc);
862}
863EXPORT_SYMBOL(dm_kcopyd_do_callback);
864
865/*
866 * Cancels a kcopyd job, eg. someone might be deactivating a
867 * mirror.
868 */
869#if 0
870int kcopyd_cancel(struct kcopyd_job *job, int block)
871{
872 /* FIXME: finish */
873 return -1;
874}
875#endif /* 0 */
876
877/*-----------------------------------------------------------------
878 * Client setup
879 *---------------------------------------------------------------*/
880struct dm_kcopyd_client *dm_kcopyd_client_create(struct dm_kcopyd_throttle *throttle)
881{
882 int r = -ENOMEM;
883 struct dm_kcopyd_client *kc;
884
885 kc = kmalloc(sizeof(*kc), GFP_KERNEL);
886 if (!kc)
887 return ERR_PTR(-ENOMEM);
888
889 spin_lock_init(&kc->job_lock);
890 INIT_LIST_HEAD(&kc->complete_jobs);
891 INIT_LIST_HEAD(&kc->io_jobs);
892 INIT_LIST_HEAD(&kc->pages_jobs);
893 kc->throttle = throttle;
894
895 kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
896 if (!kc->job_pool)
897 goto bad_slab;
898
899 INIT_WORK(&kc->kcopyd_work, do_work);
900 kc->kcopyd_wq = alloc_workqueue("kcopyd", WQ_MEM_RECLAIM, 0);
901 if (!kc->kcopyd_wq)
902 goto bad_workqueue;
903
904 kc->pages = NULL;
905 kc->nr_reserved_pages = kc->nr_free_pages = 0;
906 r = client_reserve_pages(kc, RESERVE_PAGES);
907 if (r)
908 goto bad_client_pages;
909
910 kc->io_client = dm_io_client_create();
911 if (IS_ERR(kc->io_client)) {
912 r = PTR_ERR(kc->io_client);
913 goto bad_io_client;
914 }
915
916 init_waitqueue_head(&kc->destroyq);
917 atomic_set(&kc->nr_jobs, 0);
918
919 return kc;
920
921bad_io_client:
922 client_free_pages(kc);
923bad_client_pages:
924 destroy_workqueue(kc->kcopyd_wq);
925bad_workqueue:
926 mempool_destroy(kc->job_pool);
927bad_slab:
928 kfree(kc);
929
930 return ERR_PTR(r);
931}
932EXPORT_SYMBOL(dm_kcopyd_client_create);
933
934void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
935{
936 /* Wait for completion of all jobs submitted by this client. */
937 wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
938
939 BUG_ON(!list_empty(&kc->complete_jobs));
940 BUG_ON(!list_empty(&kc->io_jobs));
941 BUG_ON(!list_empty(&kc->pages_jobs));
942 destroy_workqueue(kc->kcopyd_wq);
943 dm_io_client_destroy(kc->io_client);
944 client_free_pages(kc);
945 mempool_destroy(kc->job_pool);
946 kfree(kc);
947}
948EXPORT_SYMBOL(dm_kcopyd_client_destroy);