blob: 819e37eaaeba17cc3acf8009432f7b99f3dea4dd [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
Mikulas Patockac6ea41f2011-05-29 13:03:00 +010030#define SUB_JOB_SIZE 128
31#define SPLIT_COUNT 8
32#define MIN_JOBS 8
Mikulas Patocka5f43ba22011-05-29 13:03:11 +010033#define RESERVE_PAGES (DIV_ROUND_UP(SUB_JOB_SIZE << SECTOR_SHIFT, PAGE_SIZE))
Mikulas Patockac6ea41f2011-05-29 13:03:00 +010034
Linus Torvalds1da177e2005-04-16 15:20:36 -070035/*-----------------------------------------------------------------
36 * Each kcopyd client has its own little pool of preallocated
37 * pages for kcopyd io.
38 *---------------------------------------------------------------*/
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +010039struct dm_kcopyd_client {
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 struct page_list *pages;
Mikulas Patockad0471452011-05-29 13:03:07 +010041 unsigned nr_reserved_pages;
42 unsigned nr_free_pages;
Alasdair G Kergon138728d2006-03-27 01:17:50 -080043
Milan Broz373a3922007-05-09 02:33:02 -070044 struct dm_io_client *io_client;
45
Alasdair G Kergon138728d2006-03-27 01:17:50 -080046 wait_queue_head_t destroyq;
47 atomic_t nr_jobs;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +010048
Mikulas Patocka08d87572008-04-24 21:43:46 +010049 mempool_t *job_pool;
50
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +010051 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067};
68
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +010069static void wake(struct dm_kcopyd_client *kc)
70{
71 queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
72}
73
Mikulas Patockad0471452011-05-29 13:03:07 +010074/*
75 * Obtain one page for the use of kcopyd.
76 */
Mikulas Patockaf99b55e2011-05-29 13:03:04 +010077static struct page_list *alloc_pl(gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078{
79 struct page_list *pl;
80
Mikulas Patockaf99b55e2011-05-29 13:03:04 +010081 pl = kmalloc(sizeof(*pl), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 if (!pl)
83 return NULL;
84
Mikulas Patockaf99b55e2011-05-29 13:03:04 +010085 pl->page = alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 if (!pl->page) {
87 kfree(pl);
88 return NULL;
89 }
90
91 return pl;
92}
93
94static void free_pl(struct page_list *pl)
95{
96 __free_page(pl->page);
97 kfree(pl);
98}
99
Mikulas Patockad0471452011-05-29 13:03:07 +0100100/*
101 * Add the provided pages to a client's free page list, releasing
102 * back to the system any beyond the reserved_pages limit.
103 */
104static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
105{
106 struct page_list *next;
107
108 do {
109 next = pl->next;
110
111 if (kc->nr_free_pages >= kc->nr_reserved_pages)
112 free_pl(pl);
113 else {
114 pl->next = kc->pages;
115 kc->pages = pl;
116 kc->nr_free_pages++;
117 }
118
119 pl = next;
120 } while (pl);
121}
122
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100123static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 unsigned int nr, struct page_list **pages)
125{
126 struct page_list *pl;
127
Mikulas Patockad0471452011-05-29 13:03:07 +0100128 *pages = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Mikulas Patockad0471452011-05-29 13:03:07 +0100130 do {
131 pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY);
132 if (unlikely(!pl)) {
133 /* Use reserved pages */
134 pl = kc->pages;
135 if (unlikely(!pl))
136 goto out_of_memory;
137 kc->pages = pl->next;
138 kc->nr_free_pages--;
139 }
140 pl->next = *pages;
141 *pages = pl;
142 } while (--nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Mikulas Patockad0471452011-05-29 13:03:07 +0100146out_of_memory:
147 if (*pages)
148 kcopyd_put_pages(kc, *pages);
149 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
151
152/*
153 * These three functions resize the page pool.
154 */
155static void drop_pages(struct page_list *pl)
156{
157 struct page_list *next;
158
159 while (pl) {
160 next = pl->next;
161 free_pl(pl);
162 pl = next;
163 }
164}
165
Mikulas Patockad0471452011-05-29 13:03:07 +0100166/*
167 * Allocate and reserve nr_pages for the use of a specific client.
168 */
169static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Mikulas Patockad0471452011-05-29 13:03:07 +0100171 unsigned i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 struct page_list *pl = NULL, *next;
173
Mikulas Patockad0471452011-05-29 13:03:07 +0100174 for (i = 0; i < nr_pages; i++) {
Mikulas Patockaf99b55e2011-05-29 13:03:04 +0100175 next = alloc_pl(GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 if (!next) {
177 if (pl)
178 drop_pages(pl);
179 return -ENOMEM;
180 }
181 next->next = pl;
182 pl = next;
183 }
184
Mikulas Patockad0471452011-05-29 13:03:07 +0100185 kc->nr_reserved_pages += nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 kcopyd_put_pages(kc, pl);
Mikulas Patockad0471452011-05-29 13:03:07 +0100187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 return 0;
189}
190
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100191static void client_free_pages(struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192{
Mikulas Patockad0471452011-05-29 13:03:07 +0100193 BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 drop_pages(kc->pages);
195 kc->pages = NULL;
Mikulas Patockad0471452011-05-29 13:03:07 +0100196 kc->nr_free_pages = kc->nr_reserved_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197}
198
199/*-----------------------------------------------------------------
200 * kcopyd_jobs need to be allocated by the *clients* of kcopyd,
201 * for this reason we use a mempool to prevent the client from
202 * ever having to do io (which could cause a deadlock).
203 *---------------------------------------------------------------*/
204struct kcopyd_job {
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100205 struct dm_kcopyd_client *kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 struct list_head list;
207 unsigned long flags;
208
209 /*
210 * Error state of the job.
211 */
212 int read_err;
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700213 unsigned long write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
215 /*
216 * Either READ or WRITE
217 */
218 int rw;
Heinz Mauelshagen22a1ceb2008-04-24 21:43:17 +0100219 struct dm_io_region source;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
221 /*
222 * The destinations for the transfer.
223 */
224 unsigned int num_dests;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100225 struct dm_io_region dests[DM_KCOPYD_MAX_REGIONS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
227 sector_t offset;
228 unsigned int nr_pages;
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 */
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100235 dm_kcopyd_notify_fn fn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 void *context;
237
238 /*
239 * These fields are only used if the job has been split
240 * into more manageable parts.
241 */
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100242 struct mutex lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 atomic_t sub_jobs;
244 sector_t progress;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100246 struct kcopyd_job *master_job;
247};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Christoph Lametere18b8902006-12-06 20:33:20 -0800249static struct kmem_cache *_job_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100251int __init dm_kcopyd_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100253 _job_cache = kmem_cache_create("kcopyd_job",
254 sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
255 __alignof__(struct kcopyd_job), 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if (!_job_cache)
257 return -ENOMEM;
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 return 0;
260}
261
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100262void dm_kcopyd_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 kmem_cache_destroy(_job_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 _job_cache = NULL;
266}
267
268/*
269 * Functions to push and pop a job onto the head of a given job
270 * list.
271 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100272static struct kcopyd_job *pop(struct list_head *jobs,
273 struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
275 struct kcopyd_job *job = NULL;
276 unsigned long flags;
277
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100278 spin_lock_irqsave(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
280 if (!list_empty(jobs)) {
281 job = list_entry(jobs->next, struct kcopyd_job, list);
282 list_del(&job->list);
283 }
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100284 spin_unlock_irqrestore(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286 return job;
287}
288
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100289static void push(struct list_head *jobs, struct kcopyd_job *job)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290{
291 unsigned long flags;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100292 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100294 spin_lock_irqsave(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 list_add_tail(&job->list, jobs);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100296 spin_unlock_irqrestore(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
Kazuo Itob673c3a2008-10-21 17:44:50 +0100299
300static void push_head(struct list_head *jobs, struct kcopyd_job *job)
301{
302 unsigned long flags;
303 struct dm_kcopyd_client *kc = job->kc;
304
305 spin_lock_irqsave(&kc->job_lock, flags);
306 list_add(&job->list, jobs);
307 spin_unlock_irqrestore(&kc->job_lock, flags);
308}
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310/*
311 * These three functions process 1 item from the corresponding
312 * job list.
313 *
314 * They return:
315 * < 0: error
316 * 0: success
317 * > 0: can't process yet.
318 */
319static int run_complete_job(struct kcopyd_job *job)
320{
321 void *context = job->context;
322 int read_err = job->read_err;
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700323 unsigned long write_err = job->write_err;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100324 dm_kcopyd_notify_fn fn = job->fn;
325 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Mikulas Patocka73830852009-04-09 00:27:16 +0100327 if (job->pages)
328 kcopyd_put_pages(kc, job->pages);
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100329 /*
330 * If this is the master job, the sub jobs have already
331 * completed so we can free everything.
332 */
333 if (job->master_job == job)
334 mempool_free(job, kc->job_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 fn(read_err, write_err, context);
Alasdair G Kergon138728d2006-03-27 01:17:50 -0800336
337 if (atomic_dec_and_test(&kc->nr_jobs))
338 wake_up(&kc->destroyq);
339
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 return 0;
341}
342
343static void complete_io(unsigned long error, void *context)
344{
345 struct kcopyd_job *job = (struct kcopyd_job *) context;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100346 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 if (error) {
349 if (job->rw == WRITE)
Jonathan Brassowce503f52006-06-26 00:27:30 -0700350 job->write_err |= error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 else
352 job->read_err = 1;
353
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100354 if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100355 push(&kc->complete_jobs, job);
356 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return;
358 }
359 }
360
361 if (job->rw == WRITE)
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100362 push(&kc->complete_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363
364 else {
365 job->rw = WRITE;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100366 push(&kc->io_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
368
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100369 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
372/*
373 * Request io on as many buffer heads as we can currently get for
374 * a particular job.
375 */
376static int run_io_job(struct kcopyd_job *job)
377{
378 int r;
Milan Broz373a3922007-05-09 02:33:02 -0700379 struct dm_io_request io_req = {
Mikulas Patocka8d35d3e2011-01-13 19:59:50 +0000380 .bi_rw = job->rw,
Milan Broz373a3922007-05-09 02:33:02 -0700381 .mem.type = DM_IO_PAGE_LIST,
382 .mem.ptr.pl = job->pages,
383 .mem.offset = job->offset,
384 .notify.fn = complete_io,
385 .notify.context = job,
386 .client = job->kc->io_client,
387 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Jens Axboe7eaceac2011-03-10 08:52:07 +0100389 if (job->rw == READ)
Milan Broz373a3922007-05-09 02:33:02 -0700390 r = dm_io(&io_req, 1, &job->source, NULL);
Jens Axboe721a9602011-03-09 11:56:30 +0100391 else
Milan Broz373a3922007-05-09 02:33:02 -0700392 r = dm_io(&io_req, job->num_dests, job->dests, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 return r;
395}
396
397static int run_pages_job(struct kcopyd_job *job)
398{
399 int r;
400
401 job->nr_pages = dm_div_up(job->dests[0].count + job->offset,
402 PAGE_SIZE >> 9);
403 r = kcopyd_get_pages(job->kc, job->nr_pages, &job->pages);
404 if (!r) {
405 /* this job is ready for io */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100406 push(&job->kc->io_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 return 0;
408 }
409
410 if (r == -ENOMEM)
411 /* can't complete now */
412 return 1;
413
414 return r;
415}
416
417/*
418 * Run through a list for as long as possible. Returns the count
419 * of successful jobs.
420 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100421static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
422 int (*fn) (struct kcopyd_job *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
424 struct kcopyd_job *job;
425 int r, count = 0;
426
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100427 while ((job = pop(jobs, kc))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
429 r = fn(job);
430
431 if (r < 0) {
432 /* error this rogue job */
433 if (job->rw == WRITE)
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700434 job->write_err = (unsigned long) -1L;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 else
436 job->read_err = 1;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100437 push(&kc->complete_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 break;
439 }
440
441 if (r > 0) {
442 /*
443 * We couldn't service this job ATM, so
444 * push this job back onto the list.
445 */
Kazuo Itob673c3a2008-10-21 17:44:50 +0100446 push_head(jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 break;
448 }
449
450 count++;
451 }
452
453 return count;
454}
455
456/*
457 * kcopyd does this every time it's woken up.
458 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100459static void do_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100461 struct dm_kcopyd_client *kc = container_of(work,
462 struct dm_kcopyd_client, kcopyd_work);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100463 struct blk_plug plug;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 /*
466 * The order that these are called is *very* important.
467 * complete jobs can free some pages for pages jobs.
468 * Pages jobs when successful will jump onto the io jobs
469 * list. io jobs call wake when they complete and it all
470 * starts again.
471 */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100472 blk_start_plug(&plug);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100473 process_jobs(&kc->complete_jobs, kc, run_complete_job);
474 process_jobs(&kc->pages_jobs, kc, run_pages_job);
475 process_jobs(&kc->io_jobs, kc, run_io_job);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100476 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
479/*
480 * If we are copying a small region we just dispatch a single job
481 * to do the copy, otherwise the io has to be split up into many
482 * jobs.
483 */
484static void dispatch_job(struct kcopyd_job *job)
485{
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100486 struct dm_kcopyd_client *kc = job->kc;
487 atomic_inc(&kc->nr_jobs);
Mikulas Patocka9ca170a2009-12-10 23:52:13 +0000488 if (unlikely(!job->source.count))
489 push(&kc->complete_jobs, job);
490 else
491 push(&kc->pages_jobs, job);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100492 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493}
494
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700495static void segment_complete(int read_err, unsigned long write_err,
496 void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
498 /* FIXME: tidy this function */
499 sector_t progress = 0;
500 sector_t count = 0;
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100501 struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
502 struct kcopyd_job *job = sub_job->master_job;
Mikulas Patocka73830852009-04-09 00:27:16 +0100503 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100505 mutex_lock(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507 /* update the error */
508 if (read_err)
509 job->read_err = 1;
510
511 if (write_err)
Jonathan Brassowce503f52006-06-26 00:27:30 -0700512 job->write_err |= write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
514 /*
515 * Only dispatch more work if there hasn't been an error.
516 */
517 if ((!job->read_err && !job->write_err) ||
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100518 test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 /* get the next chunk of work */
520 progress = job->progress;
521 count = job->source.count - progress;
522 if (count) {
523 if (count > SUB_JOB_SIZE)
524 count = SUB_JOB_SIZE;
525
526 job->progress += count;
527 }
528 }
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100529 mutex_unlock(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 if (count) {
532 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 *sub_job = *job;
535 sub_job->source.sector += progress;
536 sub_job->source.count = count;
537
538 for (i = 0; i < job->num_dests; i++) {
539 sub_job->dests[i].sector += progress;
540 sub_job->dests[i].count = count;
541 }
542
543 sub_job->fn = segment_complete;
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100544 sub_job->context = sub_job;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 dispatch_job(sub_job);
546
547 } else if (atomic_dec_and_test(&job->sub_jobs)) {
548
549 /*
Mikulas Patocka340cd442009-04-09 00:27:17 +0100550 * Queue the completion callback to the kcopyd thread.
551 *
552 * Some callers assume that all the completions are called
553 * from a single thread and don't race with each other.
554 *
555 * We must not call the callback directly here because this
556 * code may not be executing in the thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 */
Mikulas Patocka340cd442009-04-09 00:27:17 +0100558 push(&kc->complete_jobs, job);
559 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 }
561}
562
563/*
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100564 * Create some sub jobs to share the work between them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100566static void split_job(struct kcopyd_job *master_job)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
568 int i;
569
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100570 atomic_inc(&master_job->kc->nr_jobs);
Mikulas Patocka340cd442009-04-09 00:27:17 +0100571
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100572 atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
573 for (i = 0; i < SPLIT_COUNT; i++) {
574 master_job[i + 1].master_job = master_job;
575 segment_complete(0, 0u, &master_job[i + 1]);
576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100579int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
580 unsigned int num_dests, struct dm_io_region *dests,
581 unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
583 struct kcopyd_job *job;
584
585 /*
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100586 * Allocate an array of jobs consisting of one master job
587 * followed by SPLIT_COUNT sub jobs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 */
Mikulas Patocka08d87572008-04-24 21:43:46 +0100589 job = mempool_alloc(kc->job_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 /*
592 * set up for the read.
593 */
594 job->kc = kc;
595 job->flags = flags;
596 job->read_err = 0;
597 job->write_err = 0;
598 job->rw = READ;
599
600 job->source = *from;
601
602 job->num_dests = num_dests;
603 memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
604
605 job->offset = 0;
606 job->nr_pages = 0;
607 job->pages = NULL;
608
609 job->fn = fn;
610 job->context = context;
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100611 job->master_job = job;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Mikulas Patockaa705a342011-05-29 13:02:58 +0100613 if (job->source.count <= SUB_JOB_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 dispatch_job(job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 else {
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100616 mutex_init(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 job->progress = 0;
618 split_job(job);
619 }
620
621 return 0;
622}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100623EXPORT_SYMBOL(dm_kcopyd_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
625/*
626 * Cancels a kcopyd job, eg. someone might be deactivating a
627 * mirror.
628 */
Adrian Bunk0b563062006-01-06 00:20:08 -0800629#if 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630int kcopyd_cancel(struct kcopyd_job *job, int block)
631{
632 /* FIXME: finish */
633 return -1;
634}
Adrian Bunk0b563062006-01-06 00:20:08 -0800635#endif /* 0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
637/*-----------------------------------------------------------------
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100638 * Client setup
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 *---------------------------------------------------------------*/
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100640struct dm_kcopyd_client *dm_kcopyd_client_create(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641{
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100642 int r = -ENOMEM;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100643 struct dm_kcopyd_client *kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 kc = kmalloc(sizeof(*kc), GFP_KERNEL);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100646 if (!kc)
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100647 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100649 spin_lock_init(&kc->job_lock);
650 INIT_LIST_HEAD(&kc->complete_jobs);
651 INIT_LIST_HEAD(&kc->io_jobs);
652 INIT_LIST_HEAD(&kc->pages_jobs);
653
Mikulas Patocka08d87572008-04-24 21:43:46 +0100654 kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100655 if (!kc->job_pool)
656 goto bad_slab;
Mikulas Patocka08d87572008-04-24 21:43:46 +0100657
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100658 INIT_WORK(&kc->kcopyd_work, do_work);
Tejun Heo9c4376d2011-01-13 19:59:58 +0000659 kc->kcopyd_wq = alloc_workqueue("kcopyd",
660 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100661 if (!kc->kcopyd_wq)
662 goto bad_workqueue;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 kc->pages = NULL;
Mikulas Patockad0471452011-05-29 13:03:07 +0100665 kc->nr_reserved_pages = kc->nr_free_pages = 0;
Mikulas Patocka5f43ba22011-05-29 13:03:11 +0100666 r = client_reserve_pages(kc, RESERVE_PAGES);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100667 if (r)
668 goto bad_client_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100670 kc->io_client = dm_io_client_create();
Milan Broz373a3922007-05-09 02:33:02 -0700671 if (IS_ERR(kc->io_client)) {
672 r = PTR_ERR(kc->io_client);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100673 goto bad_io_client;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
675
Alasdair G Kergon138728d2006-03-27 01:17:50 -0800676 init_waitqueue_head(&kc->destroyq);
677 atomic_set(&kc->nr_jobs, 0);
678
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100679 return kc;
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100680
681bad_io_client:
682 client_free_pages(kc);
683bad_client_pages:
684 destroy_workqueue(kc->kcopyd_wq);
685bad_workqueue:
686 mempool_destroy(kc->job_pool);
687bad_slab:
688 kfree(kc);
689
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100690 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100692EXPORT_SYMBOL(dm_kcopyd_client_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100694void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Alasdair G Kergon138728d2006-03-27 01:17:50 -0800696 /* Wait for completion of all jobs submitted by this client. */
697 wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
698
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100699 BUG_ON(!list_empty(&kc->complete_jobs));
700 BUG_ON(!list_empty(&kc->io_jobs));
701 BUG_ON(!list_empty(&kc->pages_jobs));
702 destroy_workqueue(kc->kcopyd_wq);
Milan Broz373a3922007-05-09 02:33:02 -0700703 dm_io_client_destroy(kc->io_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 client_free_pages(kc);
Mikulas Patocka08d87572008-04-24 21:43:46 +0100705 mempool_destroy(kc->job_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 kfree(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100708EXPORT_SYMBOL(dm_kcopyd_client_destroy);