blob: f82147029636d87b30e990daa230dfb17ca69c50 [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>
Arun Sharma600634972011-07-26 16:09:06 -070013#include <linux/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 Kergon138728dc2006-03-27 01:17:50 -080043
Milan Broz373a3922007-05-09 02:33:02 -070044 struct dm_io_client *io_client;
45
Alasdair G Kergon138728dc2006-03-27 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 struct page_list *pages;
228
229 /*
230 * Set this to ensure you are notified when the job has
231 * completed. 'context' is for callback to use.
232 */
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100233 dm_kcopyd_notify_fn fn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 void *context;
235
236 /*
237 * These fields are only used if the job has been split
238 * into more manageable parts.
239 */
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100240 struct mutex lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 atomic_t sub_jobs;
242 sector_t progress;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100244 struct kcopyd_job *master_job;
245};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Christoph Lametere18b8902006-12-06 20:33:20 -0800247static struct kmem_cache *_job_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100249int __init dm_kcopyd_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100251 _job_cache = kmem_cache_create("kcopyd_job",
252 sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
253 __alignof__(struct kcopyd_job), 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (!_job_cache)
255 return -ENOMEM;
256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return 0;
258}
259
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100260void dm_kcopyd_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 kmem_cache_destroy(_job_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 _job_cache = NULL;
264}
265
266/*
267 * Functions to push and pop a job onto the head of a given job
268 * list.
269 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100270static struct kcopyd_job *pop(struct list_head *jobs,
271 struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 struct kcopyd_job *job = NULL;
274 unsigned long flags;
275
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100276 spin_lock_irqsave(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
278 if (!list_empty(jobs)) {
279 job = list_entry(jobs->next, struct kcopyd_job, list);
280 list_del(&job->list);
281 }
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100282 spin_unlock_irqrestore(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 return job;
285}
286
Alasdair G Kergon028867a2007-07-12 17:26:32 +0100287static void push(struct list_head *jobs, struct kcopyd_job *job)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
289 unsigned long flags;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100290 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100292 spin_lock_irqsave(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 list_add_tail(&job->list, jobs);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100294 spin_unlock_irqrestore(&kc->job_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
Kazuo Itob673c3a2008-10-21 17:44:50 +0100297
298static void push_head(struct list_head *jobs, struct kcopyd_job *job)
299{
300 unsigned long flags;
301 struct dm_kcopyd_client *kc = job->kc;
302
303 spin_lock_irqsave(&kc->job_lock, flags);
304 list_add(&job->list, jobs);
305 spin_unlock_irqrestore(&kc->job_lock, flags);
306}
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308/*
309 * These three functions process 1 item from the corresponding
310 * job list.
311 *
312 * They return:
313 * < 0: error
314 * 0: success
315 * > 0: can't process yet.
316 */
317static int run_complete_job(struct kcopyd_job *job)
318{
319 void *context = job->context;
320 int read_err = job->read_err;
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700321 unsigned long write_err = job->write_err;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100322 dm_kcopyd_notify_fn fn = job->fn;
323 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Mikulas Patocka73830852009-04-09 00:27:16 +0100325 if (job->pages)
326 kcopyd_put_pages(kc, job->pages);
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100327 /*
328 * If this is the master job, the sub jobs have already
329 * completed so we can free everything.
330 */
331 if (job->master_job == job)
332 mempool_free(job, kc->job_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 fn(read_err, write_err, context);
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800334
335 if (atomic_dec_and_test(&kc->nr_jobs))
336 wake_up(&kc->destroyq);
337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 return 0;
339}
340
341static void complete_io(unsigned long error, void *context)
342{
343 struct kcopyd_job *job = (struct kcopyd_job *) context;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100344 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 if (error) {
347 if (job->rw == WRITE)
Jonathan Brassowce503f52006-06-26 00:27:30 -0700348 job->write_err |= error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 else
350 job->read_err = 1;
351
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100352 if (!test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100353 push(&kc->complete_jobs, job);
354 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 return;
356 }
357 }
358
359 if (job->rw == WRITE)
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100360 push(&kc->complete_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
362 else {
363 job->rw = WRITE;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100364 push(&kc->io_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 }
366
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100367 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
369
370/*
371 * Request io on as many buffer heads as we can currently get for
372 * a particular job.
373 */
374static int run_io_job(struct kcopyd_job *job)
375{
376 int r;
Milan Broz373a3922007-05-09 02:33:02 -0700377 struct dm_io_request io_req = {
Mikulas Patocka8d35d3e2011-01-13 19:59:50 +0000378 .bi_rw = job->rw,
Milan Broz373a3922007-05-09 02:33:02 -0700379 .mem.type = DM_IO_PAGE_LIST,
380 .mem.ptr.pl = job->pages,
Mikulas Patocka4622afb2011-08-02 12:32:02 +0100381 .mem.offset = 0,
Milan Broz373a3922007-05-09 02:33:02 -0700382 .notify.fn = complete_io,
383 .notify.context = job,
384 .client = job->kc->io_client,
385 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Jens Axboe7eaceac2011-03-10 08:52:07 +0100387 if (job->rw == READ)
Milan Broz373a3922007-05-09 02:33:02 -0700388 r = dm_io(&io_req, 1, &job->source, NULL);
Jens Axboe721a9602011-03-09 11:56:30 +0100389 else
Milan Broz373a3922007-05-09 02:33:02 -0700390 r = dm_io(&io_req, job->num_dests, job->dests, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 return r;
393}
394
395static int run_pages_job(struct kcopyd_job *job)
396{
397 int r;
Mikulas Patocka5bf45a32011-08-02 12:32:02 +0100398 unsigned nr_pages = dm_div_up(job->dests[0].count, PAGE_SIZE >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Mikulas Patocka5bf45a32011-08-02 12:32:02 +0100400 r = kcopyd_get_pages(job->kc, nr_pages, &job->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 if (!r) {
402 /* this job is ready for io */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100403 push(&job->kc->io_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 return 0;
405 }
406
407 if (r == -ENOMEM)
408 /* can't complete now */
409 return 1;
410
411 return r;
412}
413
414/*
415 * Run through a list for as long as possible. Returns the count
416 * of successful jobs.
417 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100418static int process_jobs(struct list_head *jobs, struct dm_kcopyd_client *kc,
419 int (*fn) (struct kcopyd_job *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
421 struct kcopyd_job *job;
422 int r, count = 0;
423
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100424 while ((job = pop(jobs, kc))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 r = fn(job);
427
428 if (r < 0) {
429 /* error this rogue job */
430 if (job->rw == WRITE)
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700431 job->write_err = (unsigned long) -1L;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 else
433 job->read_err = 1;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100434 push(&kc->complete_jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 break;
436 }
437
438 if (r > 0) {
439 /*
440 * We couldn't service this job ATM, so
441 * push this job back onto the list.
442 */
Kazuo Itob673c3a2008-10-21 17:44:50 +0100443 push_head(jobs, job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 break;
445 }
446
447 count++;
448 }
449
450 return count;
451}
452
453/*
454 * kcopyd does this every time it's woken up.
455 */
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100456static void do_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100458 struct dm_kcopyd_client *kc = container_of(work,
459 struct dm_kcopyd_client, kcopyd_work);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100460 struct blk_plug plug;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 /*
463 * The order that these are called is *very* important.
464 * complete jobs can free some pages for pages jobs.
465 * Pages jobs when successful will jump onto the io jobs
466 * list. io jobs call wake when they complete and it all
467 * starts again.
468 */
Jens Axboe7eaceac2011-03-10 08:52:07 +0100469 blk_start_plug(&plug);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100470 process_jobs(&kc->complete_jobs, kc, run_complete_job);
471 process_jobs(&kc->pages_jobs, kc, run_pages_job);
472 process_jobs(&kc->io_jobs, kc, run_io_job);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100473 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
475
476/*
477 * If we are copying a small region we just dispatch a single job
478 * to do the copy, otherwise the io has to be split up into many
479 * jobs.
480 */
481static void dispatch_job(struct kcopyd_job *job)
482{
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100483 struct dm_kcopyd_client *kc = job->kc;
484 atomic_inc(&kc->nr_jobs);
Mikulas Patocka9ca170a2009-12-10 23:52:13 +0000485 if (unlikely(!job->source.count))
486 push(&kc->complete_jobs, job);
487 else
488 push(&kc->pages_jobs, job);
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100489 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
Alasdair G Kergon4cdc1d12008-03-28 14:16:10 -0700492static void segment_complete(int read_err, unsigned long write_err,
493 void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
495 /* FIXME: tidy this function */
496 sector_t progress = 0;
497 sector_t count = 0;
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100498 struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
499 struct kcopyd_job *job = sub_job->master_job;
Mikulas Patocka73830852009-04-09 00:27:16 +0100500 struct dm_kcopyd_client *kc = job->kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100502 mutex_lock(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 /* update the error */
505 if (read_err)
506 job->read_err = 1;
507
508 if (write_err)
Jonathan Brassowce503f52006-06-26 00:27:30 -0700509 job->write_err |= write_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
511 /*
512 * Only dispatch more work if there hasn't been an error.
513 */
514 if ((!job->read_err && !job->write_err) ||
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100515 test_bit(DM_KCOPYD_IGNORE_ERROR, &job->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 /* get the next chunk of work */
517 progress = job->progress;
518 count = job->source.count - progress;
519 if (count) {
520 if (count > SUB_JOB_SIZE)
521 count = SUB_JOB_SIZE;
522
523 job->progress += count;
524 }
525 }
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100526 mutex_unlock(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 if (count) {
529 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 *sub_job = *job;
532 sub_job->source.sector += progress;
533 sub_job->source.count = count;
534
535 for (i = 0; i < job->num_dests; i++) {
536 sub_job->dests[i].sector += progress;
537 sub_job->dests[i].count = count;
538 }
539
540 sub_job->fn = segment_complete;
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100541 sub_job->context = sub_job;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 dispatch_job(sub_job);
543
544 } else if (atomic_dec_and_test(&job->sub_jobs)) {
545
546 /*
Mikulas Patocka340cd442009-04-09 00:27:17 +0100547 * Queue the completion callback to the kcopyd thread.
548 *
549 * Some callers assume that all the completions are called
550 * from a single thread and don't race with each other.
551 *
552 * We must not call the callback directly here because this
553 * code may not be executing in the thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
Mikulas Patocka340cd442009-04-09 00:27:17 +0100555 push(&kc->complete_jobs, job);
556 wake(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 }
558}
559
560/*
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100561 * Create some sub jobs to share the work between them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 */
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100563static void split_job(struct kcopyd_job *master_job)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564{
565 int i;
566
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100567 atomic_inc(&master_job->kc->nr_jobs);
Mikulas Patocka340cd442009-04-09 00:27:17 +0100568
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100569 atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
570 for (i = 0; i < SPLIT_COUNT; i++) {
571 master_job[i + 1].master_job = master_job;
572 segment_complete(0, 0u, &master_job[i + 1]);
573 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100576int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
577 unsigned int num_dests, struct dm_io_region *dests,
578 unsigned int flags, dm_kcopyd_notify_fn fn, void *context)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
580 struct kcopyd_job *job;
581
582 /*
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100583 * Allocate an array of jobs consisting of one master job
584 * followed by SPLIT_COUNT sub jobs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 */
Mikulas Patocka08d87572008-04-24 21:43:46 +0100586 job = mempool_alloc(kc->job_pool, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 /*
589 * set up for the read.
590 */
591 job->kc = kc;
592 job->flags = flags;
593 job->read_err = 0;
594 job->write_err = 0;
595 job->rw = READ;
596
597 job->source = *from;
598
599 job->num_dests = num_dests;
600 memcpy(&job->dests, dests, sizeof(*dests) * num_dests);
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 job->pages = NULL;
603
604 job->fn = fn;
605 job->context = context;
Mikulas Patockac6ea41f2011-05-29 13:03:00 +0100606 job->master_job = job;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Mikulas Patockaa705a342011-05-29 13:02:58 +0100608 if (job->source.count <= SUB_JOB_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 dispatch_job(job);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 else {
Matthias Kaehlckedef5b5b2007-10-19 22:38:52 +0100611 mutex_init(&job->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 job->progress = 0;
613 split_job(job);
614 }
615
616 return 0;
617}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100618EXPORT_SYMBOL(dm_kcopyd_copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
Mikulas Patockaa6e50b42011-08-02 12:32:04 +0100620void *dm_kcopyd_prepare_callback(struct dm_kcopyd_client *kc,
621 dm_kcopyd_notify_fn fn, void *context)
622{
623 struct kcopyd_job *job;
624
625 job = mempool_alloc(kc->job_pool, GFP_NOIO);
626
627 memset(job, 0, sizeof(struct kcopyd_job));
628 job->kc = kc;
629 job->fn = fn;
630 job->context = context;
631
632 atomic_inc(&kc->nr_jobs);
633
634 return job;
635}
636EXPORT_SYMBOL(dm_kcopyd_prepare_callback);
637
638void dm_kcopyd_do_callback(void *j, int read_err, unsigned long write_err)
639{
640 struct kcopyd_job *job = j;
641 struct dm_kcopyd_client *kc = job->kc;
642
643 job->read_err = read_err;
644 job->write_err = write_err;
645
646 push(&kc->complete_jobs, job);
647 wake(kc);
648}
649EXPORT_SYMBOL(dm_kcopyd_do_callback);
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651/*
652 * Cancels a kcopyd job, eg. someone might be deactivating a
653 * mirror.
654 */
Adrian Bunk0b563062006-01-06 00:20:08 -0800655#if 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656int kcopyd_cancel(struct kcopyd_job *job, int block)
657{
658 /* FIXME: finish */
659 return -1;
660}
Adrian Bunk0b563062006-01-06 00:20:08 -0800661#endif /* 0 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
663/*-----------------------------------------------------------------
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100664 * Client setup
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 *---------------------------------------------------------------*/
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100666struct dm_kcopyd_client *dm_kcopyd_client_create(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100668 int r = -ENOMEM;
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100669 struct dm_kcopyd_client *kc;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 kc = kmalloc(sizeof(*kc), GFP_KERNEL);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100672 if (!kc)
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100673 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100675 spin_lock_init(&kc->job_lock);
676 INIT_LIST_HEAD(&kc->complete_jobs);
677 INIT_LIST_HEAD(&kc->io_jobs);
678 INIT_LIST_HEAD(&kc->pages_jobs);
679
Mikulas Patocka08d87572008-04-24 21:43:46 +0100680 kc->job_pool = mempool_create_slab_pool(MIN_JOBS, _job_cache);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100681 if (!kc->job_pool)
682 goto bad_slab;
Mikulas Patocka08d87572008-04-24 21:43:46 +0100683
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100684 INIT_WORK(&kc->kcopyd_work, do_work);
Tejun Heo9c4376d2011-01-13 19:59:58 +0000685 kc->kcopyd_wq = alloc_workqueue("kcopyd",
686 WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100687 if (!kc->kcopyd_wq)
688 goto bad_workqueue;
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 kc->pages = NULL;
Mikulas Patockad0471452011-05-29 13:03:07 +0100691 kc->nr_reserved_pages = kc->nr_free_pages = 0;
Mikulas Patocka5f43ba22011-05-29 13:03:11 +0100692 r = client_reserve_pages(kc, RESERVE_PAGES);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100693 if (r)
694 goto bad_client_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Mikulas Patockabda8efe2011-05-29 13:03:09 +0100696 kc->io_client = dm_io_client_create();
Milan Broz373a3922007-05-09 02:33:02 -0700697 if (IS_ERR(kc->io_client)) {
698 r = PTR_ERR(kc->io_client);
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100699 goto bad_io_client;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800702 init_waitqueue_head(&kc->destroyq);
703 atomic_set(&kc->nr_jobs, 0);
704
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100705 return kc;
Mikulas Patocka945fa4d2008-04-24 21:43:49 +0100706
707bad_io_client:
708 client_free_pages(kc);
709bad_client_pages:
710 destroy_workqueue(kc->kcopyd_wq);
711bad_workqueue:
712 mempool_destroy(kc->job_pool);
713bad_slab:
714 kfree(kc);
715
Mikulas Patockafa34ce72011-05-29 13:03:13 +0100716 return ERR_PTR(r);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100718EXPORT_SYMBOL(dm_kcopyd_client_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100720void dm_kcopyd_client_destroy(struct dm_kcopyd_client *kc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Alasdair G Kergon138728dc2006-03-27 01:17:50 -0800722 /* Wait for completion of all jobs submitted by this client. */
723 wait_event(kc->destroyq, !atomic_read(&kc->nr_jobs));
724
Mikulas Patocka8c0cbc22008-04-24 21:43:44 +0100725 BUG_ON(!list_empty(&kc->complete_jobs));
726 BUG_ON(!list_empty(&kc->io_jobs));
727 BUG_ON(!list_empty(&kc->pages_jobs));
728 destroy_workqueue(kc->kcopyd_wq);
Milan Broz373a3922007-05-09 02:33:02 -0700729 dm_io_client_destroy(kc->io_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 client_free_pages(kc);
Mikulas Patocka08d87572008-04-24 21:43:46 +0100731 mempool_destroy(kc->job_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 kfree(kc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
Heinz Mauelshageneb69aca2008-04-24 21:43:19 +0100734EXPORT_SYMBOL(dm_kcopyd_client_destroy);