blob: 1bb73a13ca4003d841446efbe70ffda93a62061d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2002 Sistina Software (UK) Limited.
Milan Broz373a3922007-05-09 02:33:02 -07003 * Copyright (C) 2006 Red Hat GmbH
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
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
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +010012#include <linux/types.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <asm/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#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>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080024#include <linux/mutex.h>
Mikulas Patocka586e80e2008-10-21 17:44:59 +010025#include <linux/device-mapper.h>
Alasdair G Kergona765e202008-04-24 22:02:01 +010026#include <linux/dm-kcopyd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +010028#include "dm.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Linus Torvalds1da177e2005-04-16 15:20:36 -070030/*-----------------------------------------------------------------
31 * Each kcopyd client has its own little pool of preallocated
32 * pages for kcopyd io.
33 *---------------------------------------------------------------*/
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +010034struct dm_kcopyd_client {
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 spinlock_t lock;
36 struct page_list *pages;
37 unsigned int nr_pages;
38 unsigned int nr_free_pages;
Alasdair G Kergon138728dc2006-03-27 01:17:50 -080039
Milan Broz373a3922007-05-09 02:33:02 -070040 struct dm_io_client *io_client;
41
Alasdair G Kergon138728dc2006-03-27 01:17:50 -080042 wait_queue_head_t destroyq;
43 atomic_t nr_jobs;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +010044
Mikulas Patocka08d87572008-04-24 21:43:46 +010045 mempool_t *job_pool;
46
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +010047 struct workqueue_struct *kcopyd_wq;
48 struct work_struct kcopyd_work;
49
50/*
51 * We maintain three lists of jobs:
52 *
53 * i) jobs waiting for pages
54 * ii) jobs that have pages, and are waiting for the io to be issued.
55 * iii) jobs that have completed.
56 *
57 * All three of these are protected by job_lock.
58 */
59 spinlock_t job_lock;
60 struct list_head complete_jobs;
61 struct list_head io_jobs;
62 struct list_head pages_jobs;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +010065static void wake(struct dm_kcopyd_client *kc)
66{
67 queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
68}
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070static struct page_list *alloc_pl(void)
71{
72 struct page_list *pl;
73
74 pl = kmalloc(sizeof(*pl), GFP_KERNEL);
75 if (!pl)
76 return NULL;
77
78 pl->page = alloc_page(GFP_KERNEL);
79 if (!pl->page) {
80 kfree(pl);
81 return NULL;
82 }
83
84 return pl;
85}
86
87static void free_pl(struct page_list *pl)
88{
89 __free_page(pl->page);
90 kfree(pl);
91}
92
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +010093static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 unsigned int nr, struct page_list **pages)
95{
96 struct page_list *pl;
97
98 spin_lock(&kc->lock);
99 if (kc->nr_free_pages < nr) {
100 spin_unlock(&kc->lock);
101 return -ENOMEM;
102 }
103
104 kc->nr_free_pages -= nr;
105 for (*pages = pl = kc->pages; --nr; pl = pl->next)
106 ;
107
108 kc->pages = pl->next;
109 pl->next = NULL;
110
111 spin_unlock(&kc->lock);
112
113 return 0;
114}
115
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100116static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
118 struct page_list *cursor;
119
120 spin_lock(&kc->lock);
121 for (cursor = pl; cursor->next; cursor = cursor->next)
122 kc->nr_free_pages++;
123
124 kc->nr_free_pages++;
125 cursor->next = kc->pages;
126 kc->pages = pl;
127 spin_unlock(&kc->lock);
128}
129
130/*
131 * These three functions resize the page pool.
132 */
133static void drop_pages(struct page_list *pl)
134{
135 struct page_list *next;
136
137 while (pl) {
138 next = pl->next;
139 free_pl(pl);
140 pl = next;
141 }
142}
143
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100144static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145{
146 unsigned int i;
147 struct page_list *pl = NULL, *next;
148
149 for (i = 0; i < nr; i++) {
150 next = alloc_pl();
151 if (!next) {
152 if (pl)
153 drop_pages(pl);
154 return -ENOMEM;
155 }
156 next->next = pl;
157 pl = next;
158 }
159
160 kcopyd_put_pages(kc, pl);
161 kc->nr_pages += nr;
162 return 0;
163}
164
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100165static void client_free_pages(struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
167 BUG_ON(kc->nr_free_pages != kc->nr_pages);
168 drop_pages(kc->pages);
169 kc->pages = NULL;
170 kc->nr_free_pages = kc->nr_pages = 0;
171}
172
173/*-----------------------------------------------------------------
174 * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
175 * for this reason we use a mempool to prevent the client from
176 * ever having to do io (which could cause a deadlock).
177 *---------------------------------------------------------------*/
178struct kcopyd_job {
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100179 struct dm_kcopyd_client *kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 struct list_head list;
181 unsigned long flags;
182
183 /*
184 * Error state of the job.
185 */
186 int read_err;
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700187 unsigned long write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 /*
190 * Either READ or WRITE
191 */
192 int rw;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100193 struct dm_io_region source;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 /*
196 * The destinations for the transfer.
197 */
198 unsigned int num_dests;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100199 struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201 sector_t offset;
202 unsigned int nr_pages;
203 struct page_list *pages;
204
205 /*
206 * Set this to ensure you are notified when the job has
207 * completed. 'context' is for callback to use.
208 */
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100209 dm_kcopyd_notify_fn fn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 void *context;
211
212 /*
213 * These fields are only used if the job has been split
214 * into more manageable parts.
215 */
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100216 struct mutex lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 atomic_t sub_jobs;
218 sector_t progress;
219};
220
221/* FIXME: this should scale with the number of pages */
222#define MIN_JOBS 512
223
Christoph Lametere18b8902006-12-06 20:33:20 -0800224static struct kmem_cache *_job_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100226int __init dm_kcopyd_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100228 _job_cache = KMEM_CACHE(kcopyd_job, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 if (!_job_cache)
230 return -ENOMEM;
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return 0;
233}
234
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100235void dm_kcopyd_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 kmem_cache_destroy(_job_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 _job_cache = NULL;
239}
240
241/*
242 * Functions to push and pop a job onto the head of a given job
243 * list.
244 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100245static struct kcopyd_job *pop(struct list_head *jobs,
246 struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 struct kcopyd_job *job = NULL;
249 unsigned long flags;
250
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100251 spin_lock_irqsave(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253 if (!list_empty(jobs)) {
254 job = list_entry(jobs->next, struct kcopyd_job, list);
255 list_del(&job->list);
256 }
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100257 spin_unlock_irqrestore(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 return job;
260}
261
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100262static void push(struct list_head *jobs, struct kcopyd_job *job)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
264 unsigned long flags;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100265 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100267 spin_lock_irqsave(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 list_add_tail(&job->list, jobs);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100269 spin_unlock_irqrestore(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271
Kazuo Itob673c3a2008-10-21 17:44:50 +0100272
273static void push_head(struct list_head *jobs, struct kcopyd_job *job)
274{
275 unsigned long flags;
276 struct dm_kcopyd_client *kc = job->kc;
277
278 spin_lock_irqsave(&kc->job_lock, flags);
279 list_add(&job->list, jobs);
280 spin_unlock_irqrestore(&kc->job_lock, flags);
281}
282
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283/*
284 * These three functions process 1 item from the corresponding
285 * job list.
286 *
287 * They return:
288 * < 0: error
289 * 0: success
290 * > 0: can't process yet.
291 */
292static int run_complete_job(struct kcopyd_job *job)
293{
294 void *context = job->context;
295 int read_err = job->read_err;
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700296 unsigned long write_err = job->write_err;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100297 dm_kcopyd_notify_fn fn = job->fn;
298 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Mikulas Patocka73830852009-04-09 00:27:16 +0100300 if (job->pages)
301 kcopyd_put_pages(kc, job->pages);
Mikulas Patocka08d87572008-04-24 21:43:46 +0100302 mempool_free(job, kc->job_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 fn(read_err, write_err, context);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800304
305 if (atomic_dec_and_test(&kc->nr_jobs))
306 wake_up(&kc->destroyq);
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 return 0;
309}
310
311static void complete_io(unsigned long error, void *context)
312{
313 struct kcopyd_job *job = (struct kcopyd_job *) context;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100314 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316 if (error) {
317 if (job->rw == WRITE)
Jonathan Brassowce503f52006-06-26 00:27:30 -0700318 job->write_err |= error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 else
320 job->read_err = 1;
321
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100322 if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100323 push(&kc->complete_jobs, job);
324 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 return;
326 }
327 }
328
329 if (job->rw == WRITE)
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100330 push(&kc->complete_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 else {
333 job->rw = WRITE;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100334 push(&kc->io_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 }
336
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100337 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
340/*
341 * Request io on as many buffer heads as we can currently get for
342 * a particular job.
343 */
344static int run_io_job(struct kcopyd_job *job)
345{
346 int r;
Milan Broz373a3922007-05-09 02:33:02 -0700347 struct dm_io_request io_req = {
Mikulas Patocka8d35d3e2011-01-13 19:59:50 +0000348 .bi_rw = job->rw,
Milan Broz373a3922007-05-09 02:33:02 -0700349 .mem.type = DM_IO_PAGE_LIST,
350 .mem.ptr.pl = job->pages,
351 .mem.offset = job->offset,
352 .notify.fn = complete_io,
353 .notify.context = job,
354 .client = job->kc->io_client,
355 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
Jens Axboe7eaceac2011-03-10 08:52:07 +0100357 if (job->rw == READ)
Milan Broz373a3922007-05-09 02:33:02 -0700358 r = dm_io(&io_req, 1, &job->source, NULL);
Jens Axboe721a9602011-03-09 11:56:30 +0100359 else
Milan Broz373a3922007-05-09 02:33:02 -0700360 r = dm_io(&io_req, job->num_dests, job->dests, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 return r;
363}
364
365static int run_pages_job(struct kcopyd_job *job)
366{
367 int r;
368
369 job->nr_pages = dm_div_up(job->dests[0].count + job->offset,
370 PAGE_SIZE >> 9);
371 r = kcopyd_get_pages(job->kc, job->nr_pages, &job->pages);
372 if (!r) {
373 /* this job is ready for io */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100374 push(&job->kc->io_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 return 0;
376 }
377
378 if (r == -ENOMEM)
379 /* can't complete now */
380 return 1;
381
382 return r;
383}
384
385/*
386 * Run through a list for as long as possible. Returns the count
387 * of successful jobs.
388 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100389static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
390 int (*fn) (struct kcopyd_job *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391{
392 struct kcopyd_job *job;
393 int r, count = 0;
394
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100395 while ((job = pop(jobs, kc))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397 r = fn(job);
398
399 if (r < 0) {
400 /* error this rogue job */
401 if (job->rw == WRITE)
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700402 job->write_err = (unsigned long) -1L;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 else
404 job->read_err = 1;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100405 push(&kc->complete_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 break;
407 }
408
409 if (r > 0) {
410 /*
411 * We couldn't service this job ATM, so
412 * push this job back onto the list.
413 */
Kazuo Itob673c3a2008-10-21 17:44:50 +0100414 push_head(jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 break;
416 }
417
418 count++;
419 }
420
421 return count;
422}
423
424/*
425 * kcopyd does this every time it's woken up.
426 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100427static void do_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100429 struct dm_kcopyd_client *kc = container_of(work,
430 struct dm_kcopyd_client, kcopyd_work);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100431 struct blk_plug plug;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 /*
434 * The order that these are called is *very* important.
435 * complete jobs can free some pages for pages jobs.
436 * Pages jobs when successful will jump onto the io jobs
437 * list. io jobs call wake when they complete and it all
438 * starts again.
439 */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100440 blk_start_plug(&plug);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100441 process_jobs(&kc->complete_jobs, kc, run_complete_job);
442 process_jobs(&kc->pages_jobs, kc, run_pages_job);
443 process_jobs(&kc->io_jobs, kc, run_io_job);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100444 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
447/*
448 * If we are copying a small region we just dispatch a single job
449 * to do the copy, otherwise the io has to be split up into many
450 * jobs.
451 */
452static void dispatch_job(struct kcopyd_job *job)
453{
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100454 struct dm_kcopyd_client *kc = job->kc;
455 atomic_inc(&kc->nr_jobs);
Mikulas Patocka9ca170a2009-12-10 23:52:13 +0000456 if (unlikely(!job->source.count))
457 push(&kc->complete_jobs, job);
458 else
459 push(&kc->pages_jobs, job);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100460 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461}
462
463#define SUB_JOB_SIZE 128
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700464static void segment_complete(int read_err, unsigned long write_err,
465 void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
467 /* FIXME: tidy this function */
468 sector_t progress = 0;
469 sector_t count = 0;
470 struct kcopyd_job *job = (struct kcopyd_job *) context;
Mikulas Patocka73830852009-04-09 00:27:16 +0100471 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100473 mutex_lock(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 /* update the error */
476 if (read_err)
477 job->read_err = 1;
478
479 if (write_err)
Jonathan Brassowce503f52006-06-26 00:27:30 -0700480 job->write_err |= write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 /*
483 * Only dispatch more work if there hasn't been an error.
484 */
485 if ((!job->read_err && !job->write_err) ||
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100486 test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 /* get the next chunk of work */
488 progress = job->progress;
489 count = job->source.count - progress;
490 if (count) {
491 if (count > SUB_JOB_SIZE)
492 count = SUB_JOB_SIZE;
493
494 job->progress += count;
495 }
496 }
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100497 mutex_unlock(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
499 if (count) {
500 int i;
Mikulas Patocka73830852009-04-09 00:27:16 +0100501 struct kcopyd_job *sub_job = mempool_alloc(kc->job_pool,
Mikulas Patocka08d87572008-04-24 21:43:46 +0100502 GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 *sub_job = *job;
505 sub_job->source.sector += progress;
506 sub_job->source.count = count;
507
508 for (i = 0; i < job->num_dests; i++) {
509 sub_job->dests[i].sector += progress;
510 sub_job->dests[i].count = count;
511 }
512
513 sub_job->fn = segment_complete;
514 sub_job->context = job;
515 dispatch_job(sub_job);
516
517 } else if (atomic_dec_and_test(&job->sub_jobs)) {
518
519 /*
Mikulas Patocka340cd442009-04-09 00:27:17 +0100520 * Queue the completion callback to the kcopyd thread.
521 *
522 * Some callers assume that all the completions are called
523 * from a single thread and don't race with each other.
524 *
525 * We must not call the callback directly here because this
526 * code may not be executing in the thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 */
Mikulas Patocka340cd442009-04-09 00:27:17 +0100528 push(&kc->complete_jobs, job);
529 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 }
531}
532
533/*
534 * Create some little jobs that will do the move between
535 * them.
536 */
537#define SPLIT_COUNT 8
538static void split_job(struct kcopyd_job *job)
539{
540 int i;
541
Mikulas Patocka340cd442009-04-09 00:27:17 +0100542 atomic_inc(&job->kc->nr_jobs);
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 atomic_set(&job->sub_jobs, SPLIT_COUNT);
545 for (i = 0; i < SPLIT_COUNT; i++)
546 segment_complete(0, 0u, job);
547}
548
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100549int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
550 unsigned int num_dests, struct dm_io_region *dests,
551 unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
553 struct kcopyd_job *job;
554
555 /*
556 * Allocate a new job.
557 */
Mikulas Patocka08d87572008-04-24 21:43:46 +0100558 job = mempool_alloc(kc->job_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560 /*
561 * set up for the read.
562 */
563 job->kc = kc;
564 job->flags = flags;
565 job->read_err = 0;
566 job->write_err = 0;
567 job->rw = READ;
568
569 job->source = *from;
570
571 job->num_dests = num_dests;
572 memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
573
574 job->offset = 0;
575 job->nr_pages = 0;
576 job->pages = NULL;
577
578 job->fn = fn;
579 job->context = context;
580
581 if (job->source.count < SUB_JOB_SIZE)
582 dispatch_job(job);
583
584 else {
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100585 mutex_init(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 job->progress = 0;
587 split_job(job);
588 }
589
590 return 0;
591}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100592EXPORT_SYMBOL(dm_kcopyd_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594/*
595 * Cancels a kcopyd job, eg. someone might be deactivating a
596 * mirror.
597 */
Adrian Bunk0b563062006-01-06 00:20:08 -0800598#if 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599int kcopyd_cancel(struct kcopyd_job *job, int block)
600{
601 /* FIXME: finish */
602 return -1;
603}
Adrian Bunk0b563062006-01-06 00:20:08 -0800604#endif /* 0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606/*-----------------------------------------------------------------
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100607 * Client setup
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 *---------------------------------------------------------------*/
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100609int dm_kcopyd_client_create(unsigned int nr_pages,
610 struct dm_kcopyd_client **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100612 int r = -ENOMEM;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100613 struct dm_kcopyd_client *kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 kc = kmalloc(sizeof(*kc), GFP_KERNEL);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100616 if (!kc)
617 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 spin_lock_init(&kc->lock);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100620 spin_lock_init(&kc->job_lock);
621 INIT_LIST_HEAD(&kc->complete_jobs);
622 INIT_LIST_HEAD(&kc->io_jobs);
623 INIT_LIST_HEAD(&kc->pages_jobs);
624
Mikulas Patocka08d87572008-04-24 21:43:46 +0100625 kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100626 if (!kc->job_pool)
627 goto bad_slab;
Mikulas Patocka08d87572008-04-24 21:43:46 +0100628
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100629 INIT_WORK(&kc->kcopyd_work, do_work);
Tejun Heo9c4376d2011-01-13 19:59:58 +0000630 kc->kcopyd_wq = alloc_workqueue("kcopyd",
631 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100632 if (!kc->kcopyd_wq)
633 goto bad_workqueue;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 kc->pages = NULL;
636 kc->nr_pages = kc->nr_free_pages = 0;
637 r = client_alloc_pages(kc, nr_pages);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100638 if (r)
639 goto bad_client_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Milan Broz373a3922007-05-09 02:33:02 -0700641 kc->io_client = dm_io_client_create(nr_pages);
642 if (IS_ERR(kc->io_client)) {
643 r = PTR_ERR(kc->io_client);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100644 goto bad_io_client;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800647 init_waitqueue_head(&kc->destroyq);
648 atomic_set(&kc->nr_jobs, 0);
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 *result = kc;
651 return 0;
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100652
653bad_io_client:
654 client_free_pages(kc);
655bad_client_pages:
656 destroy_workqueue(kc->kcopyd_wq);
657bad_workqueue:
658 mempool_destroy(kc->job_pool);
659bad_slab:
660 kfree(kc);
661
662 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100664EXPORT_SYMBOL(dm_kcopyd_client_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100666void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800668 /* Wait for completion of all jobs submitted by this client. */
669 wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
670
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100671 BUG_ON(!list_empty(&kc->complete_jobs));
672 BUG_ON(!list_empty(&kc->io_jobs));
673 BUG_ON(!list_empty(&kc->pages_jobs));
674 destroy_workqueue(kc->kcopyd_wq);
Milan Broz373a3922007-05-09 02:33:02 -0700675 dm_io_client_destroy(kc->io_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 client_free_pages(kc);
Mikulas Patocka08d87572008-04-24 21:43:46 +0100677 mempool_destroy(kc->job_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 kfree(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100680EXPORT_SYMBOL(dm_kcopyd_client_destroy);