]>
Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
2 | drbd_worker.c | |
3 | ||
4 | This file is part of DRBD by Philipp Reisner and Lars Ellenberg. | |
5 | ||
6 | Copyright (C) 2001-2008, LINBIT Information Technologies GmbH. | |
7 | Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>. | |
8 | Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>. | |
9 | ||
10 | drbd is free software; you can redistribute it and/or modify | |
11 | it under the terms of the GNU General Public License as published by | |
12 | the Free Software Foundation; either version 2, or (at your option) | |
13 | any later version. | |
14 | ||
15 | drbd is distributed in the hope that it will be useful, | |
16 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
18 | GNU General Public License for more details. | |
19 | ||
20 | You should have received a copy of the GNU General Public License | |
21 | along with drbd; see the file COPYING. If not, write to | |
22 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. | |
23 | ||
24 | */ | |
25 | ||
b411b363 | 26 | #include <linux/module.h> |
b411b363 PR |
27 | #include <linux/drbd.h> |
28 | #include <linux/sched.h> | |
b411b363 PR |
29 | #include <linux/wait.h> |
30 | #include <linux/mm.h> | |
31 | #include <linux/memcontrol.h> | |
32 | #include <linux/mm_inline.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/random.h> | |
b411b363 PR |
35 | #include <linux/string.h> |
36 | #include <linux/scatterlist.h> | |
37 | ||
38 | #include "drbd_int.h" | |
39 | #include "drbd_req.h" | |
b411b363 | 40 | |
00d56944 | 41 | static int w_make_ov_request(struct drbd_work *w, int cancel); |
b411b363 PR |
42 | |
43 | ||
c5a91619 AG |
44 | /* endio handlers: |
45 | * drbd_md_io_complete (defined here) | |
fcefa62e AG |
46 | * drbd_request_endio (defined here) |
47 | * drbd_peer_request_endio (defined here) | |
c5a91619 AG |
48 | * bm_async_io_complete (defined in drbd_bitmap.c) |
49 | * | |
b411b363 PR |
50 | * For all these callbacks, note the following: |
51 | * The callbacks will be called in irq context by the IDE drivers, | |
52 | * and in Softirqs/Tasklets/BH context by the SCSI drivers. | |
53 | * Try to get the locking right :) | |
54 | * | |
55 | */ | |
56 | ||
57 | ||
58 | /* About the global_state_lock | |
59 | Each state transition on an device holds a read lock. In case we have | |
60 | to evaluate the sync after dependencies, we grab a write lock, because | |
61 | we need stable states on all devices for that. */ | |
62 | rwlock_t global_state_lock; | |
63 | ||
64 | /* used for synchronous meta data and bitmap IO | |
65 | * submitted by drbd_md_sync_page_io() | |
66 | */ | |
67 | void drbd_md_io_complete(struct bio *bio, int error) | |
68 | { | |
69 | struct drbd_md_io *md_io; | |
70 | ||
71 | md_io = (struct drbd_md_io *)bio->bi_private; | |
72 | md_io->error = error; | |
73 | ||
b411b363 PR |
74 | complete(&md_io->event); |
75 | } | |
76 | ||
77 | /* reads on behalf of the partner, | |
78 | * "submitted" by the receiver | |
79 | */ | |
db830c46 | 80 | void drbd_endio_read_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
81 | { |
82 | unsigned long flags = 0; | |
a21e9298 | 83 | struct drbd_conf *mdev = peer_req->w.mdev; |
b411b363 | 84 | |
87eeee41 | 85 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
db830c46 AG |
86 | mdev->read_cnt += peer_req->i.size >> 9; |
87 | list_del(&peer_req->w.list); | |
b411b363 PR |
88 | if (list_empty(&mdev->read_ee)) |
89 | wake_up(&mdev->ee_wait); | |
db830c46 | 90 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
81e84650 | 91 | __drbd_chk_io_error(mdev, false); |
87eeee41 | 92 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 | 93 | |
db830c46 | 94 | drbd_queue_work(&mdev->tconn->data.work, &peer_req->w); |
b411b363 | 95 | put_ldev(mdev); |
b411b363 PR |
96 | } |
97 | ||
98 | /* writes on behalf of the partner, or resync writes, | |
45bb912b | 99 | * "submitted" by the receiver, final stage. */ |
db830c46 | 100 | static void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(local) |
b411b363 PR |
101 | { |
102 | unsigned long flags = 0; | |
a21e9298 | 103 | struct drbd_conf *mdev = peer_req->w.mdev; |
181286ad | 104 | struct drbd_interval i; |
b411b363 | 105 | int do_wake; |
579b57ed | 106 | u64 block_id; |
b411b363 | 107 | int do_al_complete_io; |
b411b363 | 108 | |
db830c46 | 109 | /* after we moved peer_req to done_ee, |
b411b363 PR |
110 | * we may no longer access it, |
111 | * it may be freed/reused already! | |
112 | * (as soon as we release the req_lock) */ | |
181286ad | 113 | i = peer_req->i; |
db830c46 AG |
114 | do_al_complete_io = peer_req->flags & EE_CALL_AL_COMPLETE_IO; |
115 | block_id = peer_req->block_id; | |
b411b363 | 116 | |
87eeee41 | 117 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
db830c46 AG |
118 | mdev->writ_cnt += peer_req->i.size >> 9; |
119 | list_del(&peer_req->w.list); /* has been on active_ee or sync_ee */ | |
120 | list_add_tail(&peer_req->w.list, &mdev->done_ee); | |
b411b363 | 121 | |
bb3bfe96 | 122 | /* |
5e472264 | 123 | * Do not remove from the write_requests tree here: we did not send the |
bb3bfe96 AG |
124 | * Ack yet and did not wake possibly waiting conflicting requests. |
125 | * Removed from the tree from "drbd_process_done_ee" within the | |
126 | * appropriate w.cb (e_end_block/e_end_resync_block) or from | |
127 | * _drbd_clear_done_ee. | |
128 | */ | |
b411b363 | 129 | |
579b57ed | 130 | do_wake = list_empty(block_id == ID_SYNCER ? &mdev->sync_ee : &mdev->active_ee); |
b411b363 | 131 | |
db830c46 | 132 | if (test_bit(__EE_WAS_ERROR, &peer_req->flags)) |
81e84650 | 133 | __drbd_chk_io_error(mdev, false); |
87eeee41 | 134 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
b411b363 | 135 | |
579b57ed | 136 | if (block_id == ID_SYNCER) |
181286ad | 137 | drbd_rs_complete_io(mdev, i.sector); |
b411b363 PR |
138 | |
139 | if (do_wake) | |
140 | wake_up(&mdev->ee_wait); | |
141 | ||
142 | if (do_al_complete_io) | |
181286ad | 143 | drbd_al_complete_io(mdev, &i); |
b411b363 | 144 | |
0625ac19 | 145 | wake_asender(mdev->tconn); |
b411b363 | 146 | put_ldev(mdev); |
45bb912b | 147 | } |
b411b363 | 148 | |
45bb912b LE |
149 | /* writes on behalf of the partner, or resync writes, |
150 | * "submitted" by the receiver. | |
151 | */ | |
fcefa62e | 152 | void drbd_peer_request_endio(struct bio *bio, int error) |
45bb912b | 153 | { |
db830c46 | 154 | struct drbd_peer_request *peer_req = bio->bi_private; |
a21e9298 | 155 | struct drbd_conf *mdev = peer_req->w.mdev; |
45bb912b LE |
156 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
157 | int is_write = bio_data_dir(bio) == WRITE; | |
158 | ||
07194272 | 159 | if (error && __ratelimit(&drbd_ratelimit_state)) |
45bb912b LE |
160 | dev_warn(DEV, "%s: error=%d s=%llus\n", |
161 | is_write ? "write" : "read", error, | |
db830c46 | 162 | (unsigned long long)peer_req->i.sector); |
45bb912b | 163 | if (!error && !uptodate) { |
07194272 LE |
164 | if (__ratelimit(&drbd_ratelimit_state)) |
165 | dev_warn(DEV, "%s: setting error to -EIO s=%llus\n", | |
166 | is_write ? "write" : "read", | |
db830c46 | 167 | (unsigned long long)peer_req->i.sector); |
45bb912b LE |
168 | /* strange behavior of some lower level drivers... |
169 | * fail the request by clearing the uptodate flag, | |
170 | * but do not return any error?! */ | |
171 | error = -EIO; | |
172 | } | |
173 | ||
174 | if (error) | |
db830c46 | 175 | set_bit(__EE_WAS_ERROR, &peer_req->flags); |
45bb912b LE |
176 | |
177 | bio_put(bio); /* no need for the bio anymore */ | |
db830c46 | 178 | if (atomic_dec_and_test(&peer_req->pending_bios)) { |
45bb912b | 179 | if (is_write) |
db830c46 | 180 | drbd_endio_write_sec_final(peer_req); |
45bb912b | 181 | else |
db830c46 | 182 | drbd_endio_read_sec_final(peer_req); |
45bb912b | 183 | } |
b411b363 PR |
184 | } |
185 | ||
186 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
187 | */ | |
fcefa62e | 188 | void drbd_request_endio(struct bio *bio, int error) |
b411b363 | 189 | { |
a115413d | 190 | unsigned long flags; |
b411b363 | 191 | struct drbd_request *req = bio->bi_private; |
a21e9298 | 192 | struct drbd_conf *mdev = req->w.mdev; |
a115413d | 193 | struct bio_and_error m; |
b411b363 PR |
194 | enum drbd_req_event what; |
195 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
196 | ||
b411b363 PR |
197 | if (!error && !uptodate) { |
198 | dev_warn(DEV, "p %s: setting error to -EIO\n", | |
199 | bio_data_dir(bio) == WRITE ? "write" : "read"); | |
200 | /* strange behavior of some lower level drivers... | |
201 | * fail the request by clearing the uptodate flag, | |
202 | * but do not return any error?! */ | |
203 | error = -EIO; | |
204 | } | |
205 | ||
b411b363 PR |
206 | /* to avoid recursion in __req_mod */ |
207 | if (unlikely(error)) { | |
208 | what = (bio_data_dir(bio) == WRITE) | |
8554df1c | 209 | ? WRITE_COMPLETED_WITH_ERROR |
5c3c7e64 | 210 | : (bio_rw(bio) == READ) |
8554df1c AG |
211 | ? READ_COMPLETED_WITH_ERROR |
212 | : READ_AHEAD_COMPLETED_WITH_ERROR; | |
b411b363 | 213 | } else |
8554df1c | 214 | what = COMPLETED_OK; |
b411b363 PR |
215 | |
216 | bio_put(req->private_bio); | |
217 | req->private_bio = ERR_PTR(error); | |
218 | ||
a115413d | 219 | /* not req_mod(), we need irqsave here! */ |
87eeee41 | 220 | spin_lock_irqsave(&mdev->tconn->req_lock, flags); |
a115413d | 221 | __req_mod(req, what, &m); |
87eeee41 | 222 | spin_unlock_irqrestore(&mdev->tconn->req_lock, flags); |
a115413d LE |
223 | |
224 | if (m.bio) | |
225 | complete_master_bio(mdev, &m); | |
b411b363 PR |
226 | } |
227 | ||
99920dc5 | 228 | int w_read_retry_remote(struct drbd_work *w, int cancel) |
b411b363 PR |
229 | { |
230 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 231 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
232 | |
233 | /* We should not detach for read io-error, | |
234 | * but try to WRITE the P_DATA_REPLY to the failed location, | |
235 | * to give the disk the chance to relocate that block */ | |
236 | ||
87eeee41 | 237 | spin_lock_irq(&mdev->tconn->req_lock); |
d255e5ff | 238 | if (cancel || mdev->state.pdsk != D_UP_TO_DATE) { |
8554df1c | 239 | _req_mod(req, READ_RETRY_REMOTE_CANCELED); |
87eeee41 | 240 | spin_unlock_irq(&mdev->tconn->req_lock); |
99920dc5 | 241 | return 0; |
b411b363 | 242 | } |
87eeee41 | 243 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 244 | |
00d56944 | 245 | return w_send_read_req(w, 0); |
b411b363 PR |
246 | } |
247 | ||
f6ffca9f | 248 | void drbd_csum_ee(struct drbd_conf *mdev, struct crypto_hash *tfm, |
db830c46 | 249 | struct drbd_peer_request *peer_req, void *digest) |
45bb912b LE |
250 | { |
251 | struct hash_desc desc; | |
252 | struct scatterlist sg; | |
db830c46 | 253 | struct page *page = peer_req->pages; |
45bb912b LE |
254 | struct page *tmp; |
255 | unsigned len; | |
256 | ||
257 | desc.tfm = tfm; | |
258 | desc.flags = 0; | |
259 | ||
260 | sg_init_table(&sg, 1); | |
261 | crypto_hash_init(&desc); | |
262 | ||
263 | while ((tmp = page_chain_next(page))) { | |
264 | /* all but the last page will be fully used */ | |
265 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
266 | crypto_hash_update(&desc, &sg, sg.length); | |
267 | page = tmp; | |
268 | } | |
269 | /* and now the last, possibly only partially used page */ | |
db830c46 | 270 | len = peer_req->i.size & (PAGE_SIZE - 1); |
45bb912b LE |
271 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); |
272 | crypto_hash_update(&desc, &sg, sg.length); | |
273 | crypto_hash_final(&desc, digest); | |
274 | } | |
275 | ||
276 | void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest) | |
b411b363 PR |
277 | { |
278 | struct hash_desc desc; | |
279 | struct scatterlist sg; | |
280 | struct bio_vec *bvec; | |
281 | int i; | |
282 | ||
283 | desc.tfm = tfm; | |
284 | desc.flags = 0; | |
285 | ||
286 | sg_init_table(&sg, 1); | |
287 | crypto_hash_init(&desc); | |
288 | ||
289 | __bio_for_each_segment(bvec, bio, i, 0) { | |
290 | sg_set_page(&sg, bvec->bv_page, bvec->bv_len, bvec->bv_offset); | |
291 | crypto_hash_update(&desc, &sg, sg.length); | |
292 | } | |
293 | crypto_hash_final(&desc, digest); | |
294 | } | |
295 | ||
9676c760 | 296 | /* MAYBE merge common code with w_e_end_ov_req */ |
99920dc5 | 297 | static int w_e_send_csum(struct drbd_work *w, int cancel) |
b411b363 | 298 | { |
00d56944 PR |
299 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
300 | struct drbd_conf *mdev = w->mdev; | |
b411b363 PR |
301 | int digest_size; |
302 | void *digest; | |
99920dc5 | 303 | int err = 0; |
b411b363 | 304 | |
53ea4331 LE |
305 | if (unlikely(cancel)) |
306 | goto out; | |
b411b363 | 307 | |
9676c760 | 308 | if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0)) |
53ea4331 | 309 | goto out; |
b411b363 | 310 | |
f399002e | 311 | digest_size = crypto_hash_digestsize(mdev->tconn->csums_tfm); |
53ea4331 LE |
312 | digest = kmalloc(digest_size, GFP_NOIO); |
313 | if (digest) { | |
db830c46 AG |
314 | sector_t sector = peer_req->i.sector; |
315 | unsigned int size = peer_req->i.size; | |
f399002e | 316 | drbd_csum_ee(mdev, mdev->tconn->csums_tfm, peer_req, digest); |
9676c760 | 317 | /* Free peer_req and pages before send. |
53ea4331 LE |
318 | * In case we block on congestion, we could otherwise run into |
319 | * some distributed deadlock, if the other side blocks on | |
320 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 321 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 322 | drbd_free_peer_req(mdev, peer_req); |
db830c46 | 323 | peer_req = NULL; |
53ea4331 | 324 | inc_rs_pending(mdev); |
99920dc5 | 325 | err = drbd_send_drequest_csum(mdev, sector, size, |
db1b0b72 AG |
326 | digest, digest_size, |
327 | P_CSUM_RS_REQUEST); | |
53ea4331 LE |
328 | kfree(digest); |
329 | } else { | |
330 | dev_err(DEV, "kmalloc() of digest failed.\n"); | |
99920dc5 | 331 | err = -ENOMEM; |
53ea4331 | 332 | } |
b411b363 | 333 | |
53ea4331 | 334 | out: |
db830c46 | 335 | if (peer_req) |
3967deb1 | 336 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 337 | |
99920dc5 | 338 | if (unlikely(err)) |
b411b363 | 339 | dev_err(DEV, "drbd_send_drequest(..., csum) failed\n"); |
99920dc5 | 340 | return err; |
b411b363 PR |
341 | } |
342 | ||
343 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
344 | ||
345 | static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size) | |
346 | { | |
db830c46 | 347 | struct drbd_peer_request *peer_req; |
b411b363 PR |
348 | |
349 | if (!get_ldev(mdev)) | |
80a40e43 | 350 | return -EIO; |
b411b363 | 351 | |
e3555d85 | 352 | if (drbd_rs_should_slow_down(mdev, sector)) |
0f0601f4 LE |
353 | goto defer; |
354 | ||
b411b363 PR |
355 | /* GFP_TRY, because if there is no memory available right now, this may |
356 | * be rescheduled for later. It is "only" background resync, after all. */ | |
0db55363 AG |
357 | peer_req = drbd_alloc_peer_req(mdev, ID_SYNCER /* unused */, sector, |
358 | size, GFP_TRY); | |
db830c46 | 359 | if (!peer_req) |
80a40e43 | 360 | goto defer; |
b411b363 | 361 | |
db830c46 | 362 | peer_req->w.cb = w_e_send_csum; |
87eeee41 | 363 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 364 | list_add(&peer_req->w.list, &mdev->read_ee); |
87eeee41 | 365 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 366 | |
0f0601f4 | 367 | atomic_add(size >> 9, &mdev->rs_sect_ev); |
fbe29dec | 368 | if (drbd_submit_peer_request(mdev, peer_req, READ, DRBD_FAULT_RS_RD) == 0) |
80a40e43 | 369 | return 0; |
b411b363 | 370 | |
10f6d992 LE |
371 | /* If it failed because of ENOMEM, retry should help. If it failed |
372 | * because bio_add_page failed (probably broken lower level driver), | |
373 | * retry may or may not help. | |
374 | * If it does not, you may need to force disconnect. */ | |
87eeee41 | 375 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 376 | list_del(&peer_req->w.list); |
87eeee41 | 377 | spin_unlock_irq(&mdev->tconn->req_lock); |
22cc37a9 | 378 | |
3967deb1 | 379 | drbd_free_peer_req(mdev, peer_req); |
80a40e43 | 380 | defer: |
45bb912b | 381 | put_ldev(mdev); |
80a40e43 | 382 | return -EAGAIN; |
b411b363 PR |
383 | } |
384 | ||
99920dc5 | 385 | int w_resync_timer(struct drbd_work *w, int cancel) |
b411b363 | 386 | { |
00d56944 | 387 | struct drbd_conf *mdev = w->mdev; |
63106d3c PR |
388 | switch (mdev->state.conn) { |
389 | case C_VERIFY_S: | |
00d56944 | 390 | w_make_ov_request(w, cancel); |
63106d3c PR |
391 | break; |
392 | case C_SYNC_TARGET: | |
00d56944 | 393 | w_make_resync_request(w, cancel); |
63106d3c | 394 | break; |
b411b363 PR |
395 | } |
396 | ||
99920dc5 | 397 | return 0; |
794abb75 PR |
398 | } |
399 | ||
400 | void resync_timer_fn(unsigned long data) | |
401 | { | |
402 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
403 | ||
404 | if (list_empty(&mdev->resync_work.list)) | |
e42325a5 | 405 | drbd_queue_work(&mdev->tconn->data.work, &mdev->resync_work); |
b411b363 PR |
406 | } |
407 | ||
778f271d PR |
408 | static void fifo_set(struct fifo_buffer *fb, int value) |
409 | { | |
410 | int i; | |
411 | ||
412 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 413 | fb->values[i] = value; |
778f271d PR |
414 | } |
415 | ||
416 | static int fifo_push(struct fifo_buffer *fb, int value) | |
417 | { | |
418 | int ov; | |
419 | ||
420 | ov = fb->values[fb->head_index]; | |
421 | fb->values[fb->head_index++] = value; | |
422 | ||
423 | if (fb->head_index >= fb->size) | |
424 | fb->head_index = 0; | |
425 | ||
426 | return ov; | |
427 | } | |
428 | ||
429 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
430 | { | |
431 | int i; | |
432 | ||
433 | for (i = 0; i < fb->size; i++) | |
434 | fb->values[i] += value; | |
435 | } | |
436 | ||
9d77a5fe | 437 | static int drbd_rs_controller(struct drbd_conf *mdev) |
778f271d | 438 | { |
daeda1cc | 439 | struct disk_conf *dc; |
778f271d PR |
440 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ |
441 | unsigned int want; /* The number of sectors we want in the proxy */ | |
442 | int req_sect; /* Number of sectors to request in this turn */ | |
443 | int correction; /* Number of sectors more we need in the proxy*/ | |
444 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
445 | int steps; /* Number of time steps to plan ahead */ | |
446 | int curr_corr; | |
447 | int max_sect; | |
448 | ||
449 | sect_in = atomic_xchg(&mdev->rs_sect_in, 0); /* Number of sectors that came in */ | |
450 | mdev->rs_in_flight -= sect_in; | |
451 | ||
452 | spin_lock(&mdev->peer_seq_lock); /* get an atomic view on mdev->rs_plan_s */ | |
daeda1cc PR |
453 | rcu_read_lock(); |
454 | dc = rcu_dereference(mdev->ldev->disk_conf); | |
778f271d | 455 | |
daeda1cc | 456 | steps = mdev->rs_plan_s.size; /* (dc->c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ |
778f271d PR |
457 | |
458 | if (mdev->rs_in_flight + sect_in == 0) { /* At start of resync */ | |
daeda1cc | 459 | want = ((dc->resync_rate * 2 * SLEEP_TIME) / HZ) * steps; |
778f271d | 460 | } else { /* normal path */ |
daeda1cc PR |
461 | want = dc->c_fill_target ? dc->c_fill_target : |
462 | sect_in * dc->c_delay_target * HZ / (SLEEP_TIME * 10); | |
778f271d PR |
463 | } |
464 | ||
465 | correction = want - mdev->rs_in_flight - mdev->rs_planed; | |
466 | ||
467 | /* Plan ahead */ | |
468 | cps = correction / steps; | |
469 | fifo_add_val(&mdev->rs_plan_s, cps); | |
470 | mdev->rs_planed += cps * steps; | |
471 | ||
472 | /* What we do in this step */ | |
473 | curr_corr = fifo_push(&mdev->rs_plan_s, 0); | |
778f271d PR |
474 | mdev->rs_planed -= curr_corr; |
475 | ||
476 | req_sect = sect_in + curr_corr; | |
477 | if (req_sect < 0) | |
478 | req_sect = 0; | |
479 | ||
daeda1cc | 480 | max_sect = (dc->c_max_rate * 2 * SLEEP_TIME) / HZ; |
778f271d PR |
481 | if (req_sect > max_sect) |
482 | req_sect = max_sect; | |
483 | ||
484 | /* | |
485 | dev_warn(DEV, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", | |
486 | sect_in, mdev->rs_in_flight, want, correction, | |
487 | steps, cps, mdev->rs_planed, curr_corr, req_sect); | |
488 | */ | |
daeda1cc PR |
489 | rcu_read_unlock(); |
490 | spin_unlock(&mdev->peer_seq_lock); | |
778f271d PR |
491 | |
492 | return req_sect; | |
493 | } | |
494 | ||
9d77a5fe | 495 | static int drbd_rs_number_requests(struct drbd_conf *mdev) |
e65f440d LE |
496 | { |
497 | int number; | |
daeda1cc | 498 | if (mdev->rs_plan_s.size) { /* rcu_dereference(mdev->ldev->disk_conf)->c_plan_ahead */ |
e65f440d LE |
499 | number = drbd_rs_controller(mdev) >> (BM_BLOCK_SHIFT - 9); |
500 | mdev->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; | |
501 | } else { | |
daeda1cc PR |
502 | rcu_read_lock(); |
503 | mdev->c_sync_rate = rcu_dereference(mdev->ldev->disk_conf)->resync_rate; | |
504 | rcu_read_unlock(); | |
e65f440d LE |
505 | number = SLEEP_TIME * mdev->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); |
506 | } | |
507 | ||
e65f440d LE |
508 | /* ignore the amount of pending requests, the resync controller should |
509 | * throttle down to incoming reply rate soon enough anyways. */ | |
510 | return number; | |
511 | } | |
512 | ||
99920dc5 | 513 | int w_make_resync_request(struct drbd_work *w, int cancel) |
b411b363 | 514 | { |
00d56944 | 515 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
516 | unsigned long bit; |
517 | sector_t sector; | |
518 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
1816a2b4 | 519 | int max_bio_size; |
e65f440d | 520 | int number, rollback_i, size; |
b411b363 | 521 | int align, queued, sndbuf; |
0f0601f4 | 522 | int i = 0; |
b411b363 PR |
523 | |
524 | if (unlikely(cancel)) | |
99920dc5 | 525 | return 0; |
b411b363 | 526 | |
af85e8e8 LE |
527 | if (mdev->rs_total == 0) { |
528 | /* empty resync? */ | |
529 | drbd_resync_finished(mdev); | |
99920dc5 | 530 | return 0; |
af85e8e8 LE |
531 | } |
532 | ||
b411b363 PR |
533 | if (!get_ldev(mdev)) { |
534 | /* Since we only need to access mdev->rsync a | |
535 | get_ldev_if_state(mdev,D_FAILED) would be sufficient, but | |
536 | to continue resync with a broken disk makes no sense at | |
537 | all */ | |
538 | dev_err(DEV, "Disk broke down during resync!\n"); | |
99920dc5 | 539 | return 0; |
b411b363 PR |
540 | } |
541 | ||
0cfdd247 | 542 | max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9; |
e65f440d LE |
543 | number = drbd_rs_number_requests(mdev); |
544 | if (number == 0) | |
0f0601f4 | 545 | goto requeue; |
b411b363 | 546 | |
b411b363 PR |
547 | for (i = 0; i < number; i++) { |
548 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
e42325a5 PR |
549 | mutex_lock(&mdev->tconn->data.mutex); |
550 | if (mdev->tconn->data.socket) { | |
551 | queued = mdev->tconn->data.socket->sk->sk_wmem_queued; | |
552 | sndbuf = mdev->tconn->data.socket->sk->sk_sndbuf; | |
b411b363 PR |
553 | } else { |
554 | queued = 1; | |
555 | sndbuf = 0; | |
556 | } | |
e42325a5 | 557 | mutex_unlock(&mdev->tconn->data.mutex); |
b411b363 PR |
558 | if (queued > sndbuf / 2) |
559 | goto requeue; | |
560 | ||
561 | next_sector: | |
562 | size = BM_BLOCK_SIZE; | |
563 | bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo); | |
564 | ||
4b0715f0 | 565 | if (bit == DRBD_END_OF_BITMAP) { |
b411b363 | 566 | mdev->bm_resync_fo = drbd_bm_bits(mdev); |
b411b363 | 567 | put_ldev(mdev); |
99920dc5 | 568 | return 0; |
b411b363 PR |
569 | } |
570 | ||
571 | sector = BM_BIT_TO_SECT(bit); | |
572 | ||
e3555d85 PR |
573 | if (drbd_rs_should_slow_down(mdev, sector) || |
574 | drbd_try_rs_begin_io(mdev, sector)) { | |
b411b363 PR |
575 | mdev->bm_resync_fo = bit; |
576 | goto requeue; | |
577 | } | |
578 | mdev->bm_resync_fo = bit + 1; | |
579 | ||
580 | if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) { | |
581 | drbd_rs_complete_io(mdev, sector); | |
582 | goto next_sector; | |
583 | } | |
584 | ||
1816a2b4 | 585 | #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE |
b411b363 PR |
586 | /* try to find some adjacent bits. |
587 | * we stop if we have already the maximum req size. | |
588 | * | |
589 | * Additionally always align bigger requests, in order to | |
590 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
591 | */ |
592 | align = 1; | |
d207450c | 593 | rollback_i = i; |
b411b363 | 594 | for (;;) { |
1816a2b4 | 595 | if (size + BM_BLOCK_SIZE > max_bio_size) |
b411b363 PR |
596 | break; |
597 | ||
598 | /* Be always aligned */ | |
599 | if (sector & ((1<<(align+3))-1)) | |
600 | break; | |
601 | ||
602 | /* do not cross extent boundaries */ | |
603 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
604 | break; | |
605 | /* now, is it actually dirty, after all? | |
606 | * caution, drbd_bm_test_bit is tri-state for some | |
607 | * obscure reason; ( b == 0 ) would get the out-of-band | |
608 | * only accidentally right because of the "oddly sized" | |
609 | * adjustment below */ | |
610 | if (drbd_bm_test_bit(mdev, bit+1) != 1) | |
611 | break; | |
612 | bit++; | |
613 | size += BM_BLOCK_SIZE; | |
614 | if ((BM_BLOCK_SIZE << align) <= size) | |
615 | align++; | |
616 | i++; | |
617 | } | |
618 | /* if we merged some, | |
619 | * reset the offset to start the next drbd_bm_find_next from */ | |
620 | if (size > BM_BLOCK_SIZE) | |
621 | mdev->bm_resync_fo = bit + 1; | |
622 | #endif | |
623 | ||
624 | /* adjust very last sectors, in case we are oddly sized */ | |
625 | if (sector + (size>>9) > capacity) | |
626 | size = (capacity-sector)<<9; | |
f399002e | 627 | if (mdev->tconn->agreed_pro_version >= 89 && mdev->tconn->csums_tfm) { |
b411b363 | 628 | switch (read_for_csum(mdev, sector, size)) { |
80a40e43 | 629 | case -EIO: /* Disk failure */ |
b411b363 | 630 | put_ldev(mdev); |
99920dc5 | 631 | return -EIO; |
80a40e43 | 632 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b411b363 PR |
633 | drbd_rs_complete_io(mdev, sector); |
634 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 635 | i = rollback_i; |
b411b363 | 636 | goto requeue; |
80a40e43 LE |
637 | case 0: |
638 | /* everything ok */ | |
639 | break; | |
640 | default: | |
641 | BUG(); | |
b411b363 PR |
642 | } |
643 | } else { | |
99920dc5 AG |
644 | int err; |
645 | ||
b411b363 | 646 | inc_rs_pending(mdev); |
99920dc5 AG |
647 | err = drbd_send_drequest(mdev, P_RS_DATA_REQUEST, |
648 | sector, size, ID_SYNCER); | |
649 | if (err) { | |
b411b363 PR |
650 | dev_err(DEV, "drbd_send_drequest() failed, aborting...\n"); |
651 | dec_rs_pending(mdev); | |
652 | put_ldev(mdev); | |
99920dc5 | 653 | return err; |
b411b363 PR |
654 | } |
655 | } | |
656 | } | |
657 | ||
658 | if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) { | |
659 | /* last syncer _request_ was sent, | |
660 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
661 | * next sync group will resume), as soon as we receive the last | |
662 | * resync data block, and the last bit is cleared. | |
663 | * until then resync "work" is "inactive" ... | |
664 | */ | |
b411b363 | 665 | put_ldev(mdev); |
99920dc5 | 666 | return 0; |
b411b363 PR |
667 | } |
668 | ||
669 | requeue: | |
778f271d | 670 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
671 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
672 | put_ldev(mdev); | |
99920dc5 | 673 | return 0; |
b411b363 PR |
674 | } |
675 | ||
00d56944 | 676 | static int w_make_ov_request(struct drbd_work *w, int cancel) |
b411b363 | 677 | { |
00d56944 | 678 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
679 | int number, i, size; |
680 | sector_t sector; | |
681 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
682 | ||
683 | if (unlikely(cancel)) | |
684 | return 1; | |
685 | ||
2649f080 | 686 | number = drbd_rs_number_requests(mdev); |
b411b363 PR |
687 | |
688 | sector = mdev->ov_position; | |
689 | for (i = 0; i < number; i++) { | |
690 | if (sector >= capacity) { | |
b411b363 PR |
691 | return 1; |
692 | } | |
693 | ||
694 | size = BM_BLOCK_SIZE; | |
695 | ||
e3555d85 PR |
696 | if (drbd_rs_should_slow_down(mdev, sector) || |
697 | drbd_try_rs_begin_io(mdev, sector)) { | |
b411b363 PR |
698 | mdev->ov_position = sector; |
699 | goto requeue; | |
700 | } | |
701 | ||
702 | if (sector + (size>>9) > capacity) | |
703 | size = (capacity-sector)<<9; | |
704 | ||
705 | inc_rs_pending(mdev); | |
5b9f499c | 706 | if (drbd_send_ov_request(mdev, sector, size)) { |
b411b363 PR |
707 | dec_rs_pending(mdev); |
708 | return 0; | |
709 | } | |
710 | sector += BM_SECT_PER_BIT; | |
711 | } | |
712 | mdev->ov_position = sector; | |
713 | ||
714 | requeue: | |
2649f080 | 715 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
716 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
717 | return 1; | |
718 | } | |
719 | ||
99920dc5 | 720 | int w_ov_finished(struct drbd_work *w, int cancel) |
b411b363 | 721 | { |
00d56944 | 722 | struct drbd_conf *mdev = w->mdev; |
b411b363 | 723 | kfree(w); |
8f7bed77 | 724 | ov_out_of_sync_print(mdev); |
b411b363 PR |
725 | drbd_resync_finished(mdev); |
726 | ||
99920dc5 | 727 | return 0; |
b411b363 PR |
728 | } |
729 | ||
99920dc5 | 730 | static int w_resync_finished(struct drbd_work *w, int cancel) |
b411b363 | 731 | { |
00d56944 | 732 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
733 | kfree(w); |
734 | ||
735 | drbd_resync_finished(mdev); | |
736 | ||
99920dc5 | 737 | return 0; |
b411b363 PR |
738 | } |
739 | ||
af85e8e8 LE |
740 | static void ping_peer(struct drbd_conf *mdev) |
741 | { | |
2a67d8b9 PR |
742 | struct drbd_tconn *tconn = mdev->tconn; |
743 | ||
744 | clear_bit(GOT_PING_ACK, &tconn->flags); | |
745 | request_ping(tconn); | |
746 | wait_event(tconn->ping_wait, | |
747 | test_bit(GOT_PING_ACK, &tconn->flags) || mdev->state.conn < C_CONNECTED); | |
af85e8e8 LE |
748 | } |
749 | ||
b411b363 PR |
750 | int drbd_resync_finished(struct drbd_conf *mdev) |
751 | { | |
752 | unsigned long db, dt, dbdt; | |
753 | unsigned long n_oos; | |
754 | union drbd_state os, ns; | |
755 | struct drbd_work *w; | |
756 | char *khelper_cmd = NULL; | |
26525618 | 757 | int verify_done = 0; |
b411b363 PR |
758 | |
759 | /* Remove all elements from the resync LRU. Since future actions | |
760 | * might set bits in the (main) bitmap, then the entries in the | |
761 | * resync LRU would be wrong. */ | |
762 | if (drbd_rs_del_all(mdev)) { | |
763 | /* In case this is not possible now, most probably because | |
764 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
765 | * queue (or even the read operations for those packets | |
766 | * is not finished by now). Retry in 100ms. */ | |
767 | ||
20ee6390 | 768 | schedule_timeout_interruptible(HZ / 10); |
b411b363 PR |
769 | w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC); |
770 | if (w) { | |
771 | w->cb = w_resync_finished; | |
e42325a5 | 772 | drbd_queue_work(&mdev->tconn->data.work, w); |
b411b363 PR |
773 | return 1; |
774 | } | |
775 | dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n"); | |
776 | } | |
777 | ||
778 | dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ; | |
779 | if (dt <= 0) | |
780 | dt = 1; | |
781 | db = mdev->rs_total; | |
782 | dbdt = Bit2KB(db/dt); | |
783 | mdev->rs_paused /= HZ; | |
784 | ||
785 | if (!get_ldev(mdev)) | |
786 | goto out; | |
787 | ||
af85e8e8 LE |
788 | ping_peer(mdev); |
789 | ||
87eeee41 | 790 | spin_lock_irq(&mdev->tconn->req_lock); |
78bae59b | 791 | os = drbd_read_state(mdev); |
b411b363 | 792 | |
26525618 LE |
793 | verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T); |
794 | ||
b411b363 PR |
795 | /* This protects us against multiple calls (that can happen in the presence |
796 | of application IO), and against connectivity loss just before we arrive here. */ | |
797 | if (os.conn <= C_CONNECTED) | |
798 | goto out_unlock; | |
799 | ||
800 | ns = os; | |
801 | ns.conn = C_CONNECTED; | |
802 | ||
803 | dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", | |
26525618 | 804 | verify_done ? "Online verify " : "Resync", |
b411b363 PR |
805 | dt + mdev->rs_paused, mdev->rs_paused, dbdt); |
806 | ||
807 | n_oos = drbd_bm_total_weight(mdev); | |
808 | ||
809 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
810 | if (n_oos) { | |
811 | dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n", | |
812 | n_oos, Bit2KB(1)); | |
813 | khelper_cmd = "out-of-sync"; | |
814 | } | |
815 | } else { | |
816 | D_ASSERT((n_oos - mdev->rs_failed) == 0); | |
817 | ||
818 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
819 | khelper_cmd = "after-resync-target"; | |
820 | ||
f399002e | 821 | if (mdev->tconn->csums_tfm && mdev->rs_total) { |
b411b363 PR |
822 | const unsigned long s = mdev->rs_same_csum; |
823 | const unsigned long t = mdev->rs_total; | |
824 | const int ratio = | |
825 | (t == 0) ? 0 : | |
826 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
24c4830c | 827 | dev_info(DEV, "%u %% had equal checksums, eliminated: %luK; " |
b411b363 PR |
828 | "transferred %luK total %luK\n", |
829 | ratio, | |
830 | Bit2KB(mdev->rs_same_csum), | |
831 | Bit2KB(mdev->rs_total - mdev->rs_same_csum), | |
832 | Bit2KB(mdev->rs_total)); | |
833 | } | |
834 | } | |
835 | ||
836 | if (mdev->rs_failed) { | |
837 | dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed); | |
838 | ||
839 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
840 | ns.disk = D_INCONSISTENT; | |
841 | ns.pdsk = D_UP_TO_DATE; | |
842 | } else { | |
843 | ns.disk = D_UP_TO_DATE; | |
844 | ns.pdsk = D_INCONSISTENT; | |
845 | } | |
846 | } else { | |
847 | ns.disk = D_UP_TO_DATE; | |
848 | ns.pdsk = D_UP_TO_DATE; | |
849 | ||
850 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
851 | if (mdev->p_uuid) { | |
852 | int i; | |
853 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
854 | _drbd_uuid_set(mdev, i, mdev->p_uuid[i]); | |
855 | drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]); | |
856 | _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]); | |
857 | } else { | |
858 | dev_err(DEV, "mdev->p_uuid is NULL! BUG\n"); | |
859 | } | |
860 | } | |
861 | ||
62b0da3a LE |
862 | if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) { |
863 | /* for verify runs, we don't update uuids here, | |
864 | * so there would be nothing to report. */ | |
865 | drbd_uuid_set_bm(mdev, 0UL); | |
866 | drbd_print_uuids(mdev, "updated UUIDs"); | |
867 | if (mdev->p_uuid) { | |
868 | /* Now the two UUID sets are equal, update what we | |
869 | * know of the peer. */ | |
870 | int i; | |
871 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
872 | mdev->p_uuid[i] = mdev->ldev->md.uuid[i]; | |
873 | } | |
b411b363 PR |
874 | } |
875 | } | |
876 | ||
877 | _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
878 | out_unlock: | |
87eeee41 | 879 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
880 | put_ldev(mdev); |
881 | out: | |
882 | mdev->rs_total = 0; | |
883 | mdev->rs_failed = 0; | |
884 | mdev->rs_paused = 0; | |
26525618 LE |
885 | if (verify_done) |
886 | mdev->ov_start_sector = 0; | |
b411b363 | 887 | |
13d42685 LE |
888 | drbd_md_sync(mdev); |
889 | ||
b411b363 PR |
890 | if (khelper_cmd) |
891 | drbd_khelper(mdev, khelper_cmd); | |
892 | ||
893 | return 1; | |
894 | } | |
895 | ||
896 | /* helper */ | |
db830c46 | 897 | static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_peer_request *peer_req) |
b411b363 | 898 | { |
045417f7 | 899 | if (drbd_peer_req_has_active_page(peer_req)) { |
b411b363 | 900 | /* This might happen if sendpage() has not finished */ |
db830c46 | 901 | int i = (peer_req->i.size + PAGE_SIZE -1) >> PAGE_SHIFT; |
435f0740 LE |
902 | atomic_add(i, &mdev->pp_in_use_by_net); |
903 | atomic_sub(i, &mdev->pp_in_use); | |
87eeee41 | 904 | spin_lock_irq(&mdev->tconn->req_lock); |
db830c46 | 905 | list_add_tail(&peer_req->w.list, &mdev->net_ee); |
87eeee41 | 906 | spin_unlock_irq(&mdev->tconn->req_lock); |
435f0740 | 907 | wake_up(&drbd_pp_wait); |
b411b363 | 908 | } else |
3967deb1 | 909 | drbd_free_peer_req(mdev, peer_req); |
b411b363 PR |
910 | } |
911 | ||
912 | /** | |
913 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
914 | * @mdev: DRBD device. | |
915 | * @w: work object. | |
916 | * @cancel: The connection will be closed anyways | |
917 | */ | |
99920dc5 | 918 | int w_e_end_data_req(struct drbd_work *w, int cancel) |
b411b363 | 919 | { |
db830c46 | 920 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 921 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 922 | int err; |
b411b363 PR |
923 | |
924 | if (unlikely(cancel)) { | |
3967deb1 | 925 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 926 | dec_unacked(mdev); |
99920dc5 | 927 | return 0; |
b411b363 PR |
928 | } |
929 | ||
db830c46 | 930 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
99920dc5 | 931 | err = drbd_send_block(mdev, P_DATA_REPLY, peer_req); |
b411b363 PR |
932 | } else { |
933 | if (__ratelimit(&drbd_ratelimit_state)) | |
934 | dev_err(DEV, "Sending NegDReply. sector=%llus.\n", | |
db830c46 | 935 | (unsigned long long)peer_req->i.sector); |
b411b363 | 936 | |
99920dc5 | 937 | err = drbd_send_ack(mdev, P_NEG_DREPLY, peer_req); |
b411b363 PR |
938 | } |
939 | ||
940 | dec_unacked(mdev); | |
941 | ||
db830c46 | 942 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 943 | |
99920dc5 | 944 | if (unlikely(err)) |
b411b363 | 945 | dev_err(DEV, "drbd_send_block() failed\n"); |
99920dc5 | 946 | return err; |
b411b363 PR |
947 | } |
948 | ||
949 | /** | |
950 | * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS | |
951 | * @mdev: DRBD device. | |
952 | * @w: work object. | |
953 | * @cancel: The connection will be closed anyways | |
954 | */ | |
99920dc5 | 955 | int w_e_end_rsdata_req(struct drbd_work *w, int cancel) |
b411b363 | 956 | { |
db830c46 | 957 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 958 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 959 | int err; |
b411b363 PR |
960 | |
961 | if (unlikely(cancel)) { | |
3967deb1 | 962 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 963 | dec_unacked(mdev); |
99920dc5 | 964 | return 0; |
b411b363 PR |
965 | } |
966 | ||
967 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
db830c46 | 968 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
b411b363 PR |
969 | put_ldev(mdev); |
970 | } | |
971 | ||
d612d309 | 972 | if (mdev->state.conn == C_AHEAD) { |
99920dc5 | 973 | err = drbd_send_ack(mdev, P_RS_CANCEL, peer_req); |
db830c46 | 974 | } else if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
975 | if (likely(mdev->state.pdsk >= D_INCONSISTENT)) { |
976 | inc_rs_pending(mdev); | |
99920dc5 | 977 | err = drbd_send_block(mdev, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
978 | } else { |
979 | if (__ratelimit(&drbd_ratelimit_state)) | |
980 | dev_err(DEV, "Not sending RSDataReply, " | |
981 | "partner DISKLESS!\n"); | |
99920dc5 | 982 | err = 0; |
b411b363 PR |
983 | } |
984 | } else { | |
985 | if (__ratelimit(&drbd_ratelimit_state)) | |
986 | dev_err(DEV, "Sending NegRSDReply. sector %llus.\n", | |
db830c46 | 987 | (unsigned long long)peer_req->i.sector); |
b411b363 | 988 | |
99920dc5 | 989 | err = drbd_send_ack(mdev, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
990 | |
991 | /* update resync data with failure */ | |
db830c46 | 992 | drbd_rs_failed_io(mdev, peer_req->i.sector, peer_req->i.size); |
b411b363 PR |
993 | } |
994 | ||
995 | dec_unacked(mdev); | |
996 | ||
db830c46 | 997 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 998 | |
99920dc5 | 999 | if (unlikely(err)) |
b411b363 | 1000 | dev_err(DEV, "drbd_send_block() failed\n"); |
99920dc5 | 1001 | return err; |
b411b363 PR |
1002 | } |
1003 | ||
99920dc5 | 1004 | int w_e_end_csum_rs_req(struct drbd_work *w, int cancel) |
b411b363 | 1005 | { |
db830c46 | 1006 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1007 | struct drbd_conf *mdev = w->mdev; |
b411b363 PR |
1008 | struct digest_info *di; |
1009 | int digest_size; | |
1010 | void *digest = NULL; | |
99920dc5 | 1011 | int err, eq = 0; |
b411b363 PR |
1012 | |
1013 | if (unlikely(cancel)) { | |
3967deb1 | 1014 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1015 | dec_unacked(mdev); |
99920dc5 | 1016 | return 0; |
b411b363 PR |
1017 | } |
1018 | ||
1d53f09e | 1019 | if (get_ldev(mdev)) { |
db830c46 | 1020 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
1d53f09e LE |
1021 | put_ldev(mdev); |
1022 | } | |
b411b363 | 1023 | |
db830c46 | 1024 | di = peer_req->digest; |
b411b363 | 1025 | |
db830c46 | 1026 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1027 | /* quick hack to try to avoid a race against reconfiguration. |
1028 | * a real fix would be much more involved, | |
1029 | * introducing more locking mechanisms */ | |
f399002e LE |
1030 | if (mdev->tconn->csums_tfm) { |
1031 | digest_size = crypto_hash_digestsize(mdev->tconn->csums_tfm); | |
b411b363 PR |
1032 | D_ASSERT(digest_size == di->digest_size); |
1033 | digest = kmalloc(digest_size, GFP_NOIO); | |
1034 | } | |
1035 | if (digest) { | |
f399002e | 1036 | drbd_csum_ee(mdev, mdev->tconn->csums_tfm, peer_req, digest); |
b411b363 PR |
1037 | eq = !memcmp(digest, di->digest, digest_size); |
1038 | kfree(digest); | |
1039 | } | |
1040 | ||
1041 | if (eq) { | |
db830c46 | 1042 | drbd_set_in_sync(mdev, peer_req->i.sector, peer_req->i.size); |
676396d5 | 1043 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
db830c46 | 1044 | mdev->rs_same_csum += peer_req->i.size >> BM_BLOCK_SHIFT; |
99920dc5 | 1045 | err = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, peer_req); |
b411b363 PR |
1046 | } else { |
1047 | inc_rs_pending(mdev); | |
db830c46 AG |
1048 | peer_req->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1049 | peer_req->flags &= ~EE_HAS_DIGEST; /* This peer request no longer has a digest pointer */ | |
204bba99 | 1050 | kfree(di); |
99920dc5 | 1051 | err = drbd_send_block(mdev, P_RS_DATA_REPLY, peer_req); |
b411b363 PR |
1052 | } |
1053 | } else { | |
99920dc5 | 1054 | err = drbd_send_ack(mdev, P_NEG_RS_DREPLY, peer_req); |
b411b363 PR |
1055 | if (__ratelimit(&drbd_ratelimit_state)) |
1056 | dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); | |
1057 | } | |
1058 | ||
1059 | dec_unacked(mdev); | |
db830c46 | 1060 | move_to_net_ee_or_free(mdev, peer_req); |
b411b363 | 1061 | |
99920dc5 | 1062 | if (unlikely(err)) |
b411b363 | 1063 | dev_err(DEV, "drbd_send_block/ack() failed\n"); |
99920dc5 | 1064 | return err; |
b411b363 PR |
1065 | } |
1066 | ||
99920dc5 | 1067 | int w_e_end_ov_req(struct drbd_work *w, int cancel) |
b411b363 | 1068 | { |
db830c46 | 1069 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1070 | struct drbd_conf *mdev = w->mdev; |
db830c46 AG |
1071 | sector_t sector = peer_req->i.sector; |
1072 | unsigned int size = peer_req->i.size; | |
b411b363 PR |
1073 | int digest_size; |
1074 | void *digest; | |
99920dc5 | 1075 | int err = 0; |
b411b363 PR |
1076 | |
1077 | if (unlikely(cancel)) | |
1078 | goto out; | |
1079 | ||
f399002e | 1080 | digest_size = crypto_hash_digestsize(mdev->tconn->verify_tfm); |
b411b363 | 1081 | digest = kmalloc(digest_size, GFP_NOIO); |
8f21420e | 1082 | if (!digest) { |
99920dc5 | 1083 | err = 1; /* terminate the connection in case the allocation failed */ |
8f21420e | 1084 | goto out; |
b411b363 PR |
1085 | } |
1086 | ||
db830c46 | 1087 | if (likely(!(peer_req->flags & EE_WAS_ERROR))) |
f399002e | 1088 | drbd_csum_ee(mdev, mdev->tconn->verify_tfm, peer_req, digest); |
8f21420e PR |
1089 | else |
1090 | memset(digest, 0, digest_size); | |
1091 | ||
53ea4331 LE |
1092 | /* Free e and pages before send. |
1093 | * In case we block on congestion, we could otherwise run into | |
1094 | * some distributed deadlock, if the other side blocks on | |
1095 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1096 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 1097 | drbd_free_peer_req(mdev, peer_req); |
db830c46 | 1098 | peer_req = NULL; |
8f21420e | 1099 | inc_rs_pending(mdev); |
99920dc5 AG |
1100 | err = drbd_send_drequest_csum(mdev, sector, size, digest, digest_size, P_OV_REPLY); |
1101 | if (err) | |
8f21420e PR |
1102 | dec_rs_pending(mdev); |
1103 | kfree(digest); | |
1104 | ||
b411b363 | 1105 | out: |
db830c46 | 1106 | if (peer_req) |
3967deb1 | 1107 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1108 | dec_unacked(mdev); |
99920dc5 | 1109 | return err; |
b411b363 PR |
1110 | } |
1111 | ||
8f7bed77 | 1112 | void drbd_ov_out_of_sync_found(struct drbd_conf *mdev, sector_t sector, int size) |
b411b363 PR |
1113 | { |
1114 | if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) { | |
1115 | mdev->ov_last_oos_size += size>>9; | |
1116 | } else { | |
1117 | mdev->ov_last_oos_start = sector; | |
1118 | mdev->ov_last_oos_size = size>>9; | |
1119 | } | |
1120 | drbd_set_out_of_sync(mdev, sector, size); | |
b411b363 PR |
1121 | } |
1122 | ||
99920dc5 | 1123 | int w_e_end_ov_reply(struct drbd_work *w, int cancel) |
b411b363 | 1124 | { |
db830c46 | 1125 | struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w); |
00d56944 | 1126 | struct drbd_conf *mdev = w->mdev; |
b411b363 | 1127 | struct digest_info *di; |
b411b363 | 1128 | void *digest; |
db830c46 AG |
1129 | sector_t sector = peer_req->i.sector; |
1130 | unsigned int size = peer_req->i.size; | |
53ea4331 | 1131 | int digest_size; |
99920dc5 | 1132 | int err, eq = 0; |
b411b363 PR |
1133 | |
1134 | if (unlikely(cancel)) { | |
3967deb1 | 1135 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1136 | dec_unacked(mdev); |
99920dc5 | 1137 | return 0; |
b411b363 PR |
1138 | } |
1139 | ||
1140 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1141 | * the resync lru has been cleaned up already */ | |
1d53f09e | 1142 | if (get_ldev(mdev)) { |
db830c46 | 1143 | drbd_rs_complete_io(mdev, peer_req->i.sector); |
1d53f09e LE |
1144 | put_ldev(mdev); |
1145 | } | |
b411b363 | 1146 | |
db830c46 | 1147 | di = peer_req->digest; |
b411b363 | 1148 | |
db830c46 | 1149 | if (likely((peer_req->flags & EE_WAS_ERROR) == 0)) { |
f399002e | 1150 | digest_size = crypto_hash_digestsize(mdev->tconn->verify_tfm); |
b411b363 PR |
1151 | digest = kmalloc(digest_size, GFP_NOIO); |
1152 | if (digest) { | |
f399002e | 1153 | drbd_csum_ee(mdev, mdev->tconn->verify_tfm, peer_req, digest); |
b411b363 PR |
1154 | |
1155 | D_ASSERT(digest_size == di->digest_size); | |
1156 | eq = !memcmp(digest, di->digest, digest_size); | |
1157 | kfree(digest); | |
1158 | } | |
b411b363 PR |
1159 | } |
1160 | ||
9676c760 LE |
1161 | /* Free peer_req and pages before send. |
1162 | * In case we block on congestion, we could otherwise run into | |
1163 | * some distributed deadlock, if the other side blocks on | |
1164 | * congestion as well, because our receiver blocks in | |
c37c8ecf | 1165 | * drbd_alloc_pages due to pp_in_use > max_buffers. */ |
3967deb1 | 1166 | drbd_free_peer_req(mdev, peer_req); |
b411b363 | 1167 | if (!eq) |
8f7bed77 | 1168 | drbd_ov_out_of_sync_found(mdev, sector, size); |
b411b363 | 1169 | else |
8f7bed77 | 1170 | ov_out_of_sync_print(mdev); |
b411b363 | 1171 | |
99920dc5 | 1172 | err = drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size, |
fa79abd8 | 1173 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); |
b411b363 | 1174 | |
53ea4331 | 1175 | dec_unacked(mdev); |
b411b363 | 1176 | |
ea5442af LE |
1177 | --mdev->ov_left; |
1178 | ||
1179 | /* let's advance progress step marks only for every other megabyte */ | |
1180 | if ((mdev->ov_left & 0x200) == 0x200) | |
1181 | drbd_advance_rs_marks(mdev, mdev->ov_left); | |
1182 | ||
1183 | if (mdev->ov_left == 0) { | |
8f7bed77 | 1184 | ov_out_of_sync_print(mdev); |
b411b363 PR |
1185 | drbd_resync_finished(mdev); |
1186 | } | |
1187 | ||
99920dc5 | 1188 | return err; |
b411b363 PR |
1189 | } |
1190 | ||
99920dc5 | 1191 | int w_prev_work_done(struct drbd_work *w, int cancel) |
b411b363 PR |
1192 | { |
1193 | struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w); | |
00d56944 | 1194 | |
b411b363 | 1195 | complete(&b->done); |
99920dc5 | 1196 | return 0; |
b411b363 PR |
1197 | } |
1198 | ||
99920dc5 | 1199 | int w_send_barrier(struct drbd_work *w, int cancel) |
b411b363 | 1200 | { |
9f5bdc33 | 1201 | struct drbd_socket *sock; |
b411b363 | 1202 | struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w); |
00d56944 | 1203 | struct drbd_conf *mdev = w->mdev; |
9f5bdc33 | 1204 | struct p_barrier *p; |
b411b363 PR |
1205 | |
1206 | /* really avoid racing with tl_clear. w.cb may have been referenced | |
1207 | * just before it was reassigned and re-queued, so double check that. | |
1208 | * actually, this race was harmless, since we only try to send the | |
1209 | * barrier packet here, and otherwise do nothing with the object. | |
1210 | * but compare with the head of w_clear_epoch */ | |
87eeee41 | 1211 | spin_lock_irq(&mdev->tconn->req_lock); |
b411b363 PR |
1212 | if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED) |
1213 | cancel = 1; | |
87eeee41 | 1214 | spin_unlock_irq(&mdev->tconn->req_lock); |
b411b363 | 1215 | if (cancel) |
b411b363 | 1216 | return 0; |
99920dc5 | 1217 | |
9f5bdc33 AG |
1218 | sock = &mdev->tconn->data; |
1219 | p = drbd_prepare_command(mdev, sock); | |
1220 | if (!p) | |
1221 | return -EIO; | |
b411b363 PR |
1222 | p->barrier = b->br_number; |
1223 | /* inc_ap_pending was done where this was queued. | |
1224 | * dec_ap_pending will be done in got_BarrierAck | |
1225 | * or (on connection loss) in w_clear_epoch. */ | |
9f5bdc33 | 1226 | return drbd_send_command(mdev, sock, P_BARRIER, sizeof(*p), NULL, 0); |
b411b363 PR |
1227 | } |
1228 | ||
99920dc5 | 1229 | int w_send_write_hint(struct drbd_work *w, int cancel) |
b411b363 | 1230 | { |
00d56944 | 1231 | struct drbd_conf *mdev = w->mdev; |
9f5bdc33 AG |
1232 | struct drbd_socket *sock; |
1233 | ||
b411b363 | 1234 | if (cancel) |
99920dc5 | 1235 | return 0; |
9f5bdc33 AG |
1236 | sock = &mdev->tconn->data; |
1237 | if (!drbd_prepare_command(mdev, sock)) | |
1238 | return -EIO; | |
e658983a | 1239 | return drbd_send_command(mdev, sock, P_UNPLUG_REMOTE, 0, NULL, 0); |
b411b363 PR |
1240 | } |
1241 | ||
8f7bed77 | 1242 | int w_send_out_of_sync(struct drbd_work *w, int cancel) |
73a01a18 PR |
1243 | { |
1244 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1245 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1246 | int err; |
73a01a18 PR |
1247 | |
1248 | if (unlikely(cancel)) { | |
8554df1c | 1249 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1250 | return 0; |
73a01a18 PR |
1251 | } |
1252 | ||
8f7bed77 | 1253 | err = drbd_send_out_of_sync(mdev, req); |
8554df1c | 1254 | req_mod(req, OOS_HANDED_TO_NETWORK); |
73a01a18 | 1255 | |
99920dc5 | 1256 | return err; |
73a01a18 PR |
1257 | } |
1258 | ||
b411b363 PR |
1259 | /** |
1260 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
1261 | * @mdev: DRBD device. | |
1262 | * @w: work object. | |
1263 | * @cancel: The connection will be closed anyways | |
1264 | */ | |
99920dc5 | 1265 | int w_send_dblock(struct drbd_work *w, int cancel) |
b411b363 PR |
1266 | { |
1267 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1268 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1269 | int err; |
b411b363 PR |
1270 | |
1271 | if (unlikely(cancel)) { | |
8554df1c | 1272 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1273 | return 0; |
b411b363 PR |
1274 | } |
1275 | ||
99920dc5 AG |
1276 | err = drbd_send_dblock(mdev, req); |
1277 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); | |
b411b363 | 1278 | |
99920dc5 | 1279 | return err; |
b411b363 PR |
1280 | } |
1281 | ||
1282 | /** | |
1283 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
1284 | * @mdev: DRBD device. | |
1285 | * @w: work object. | |
1286 | * @cancel: The connection will be closed anyways | |
1287 | */ | |
99920dc5 | 1288 | int w_send_read_req(struct drbd_work *w, int cancel) |
b411b363 PR |
1289 | { |
1290 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1291 | struct drbd_conf *mdev = w->mdev; |
99920dc5 | 1292 | int err; |
b411b363 PR |
1293 | |
1294 | if (unlikely(cancel)) { | |
8554df1c | 1295 | req_mod(req, SEND_CANCELED); |
99920dc5 | 1296 | return 0; |
b411b363 PR |
1297 | } |
1298 | ||
99920dc5 | 1299 | err = drbd_send_drequest(mdev, P_DATA_REQUEST, req->i.sector, req->i.size, |
6c1005e7 | 1300 | (unsigned long)req); |
b411b363 | 1301 | |
99920dc5 | 1302 | req_mod(req, err ? SEND_FAILED : HANDED_OVER_TO_NETWORK); |
b411b363 | 1303 | |
99920dc5 | 1304 | return err; |
b411b363 PR |
1305 | } |
1306 | ||
99920dc5 | 1307 | int w_restart_disk_io(struct drbd_work *w, int cancel) |
265be2d0 PR |
1308 | { |
1309 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
00d56944 | 1310 | struct drbd_conf *mdev = w->mdev; |
265be2d0 | 1311 | |
0778286a | 1312 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
181286ad | 1313 | drbd_al_begin_io(mdev, &req->i); |
265be2d0 PR |
1314 | /* Calling drbd_al_begin_io() out of the worker might deadlocks |
1315 | theoretically. Practically it can not deadlock, since this is | |
1316 | only used when unfreezing IOs. All the extents of the requests | |
1317 | that made it into the TL are already active */ | |
1318 | ||
1319 | drbd_req_make_private_bio(req, req->master_bio); | |
1320 | req->private_bio->bi_bdev = mdev->ldev->backing_bdev; | |
1321 | generic_make_request(req->private_bio); | |
1322 | ||
99920dc5 | 1323 | return 0; |
265be2d0 PR |
1324 | } |
1325 | ||
b411b363 PR |
1326 | static int _drbd_may_sync_now(struct drbd_conf *mdev) |
1327 | { | |
1328 | struct drbd_conf *odev = mdev; | |
daeda1cc | 1329 | int ra; |
b411b363 PR |
1330 | |
1331 | while (1) { | |
438c8374 PR |
1332 | if (!odev->ldev) |
1333 | return 1; | |
daeda1cc PR |
1334 | rcu_read_lock(); |
1335 | ra = rcu_dereference(odev->ldev->disk_conf)->resync_after; | |
1336 | rcu_read_unlock(); | |
1337 | if (ra == -1) | |
b411b363 | 1338 | return 1; |
daeda1cc | 1339 | odev = minor_to_mdev(ra); |
841ce241 AG |
1340 | if (!expect(odev)) |
1341 | return 1; | |
b411b363 PR |
1342 | if ((odev->state.conn >= C_SYNC_SOURCE && |
1343 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1344 | odev->state.aftr_isp || odev->state.peer_isp || | |
1345 | odev->state.user_isp) | |
1346 | return 0; | |
1347 | } | |
1348 | } | |
1349 | ||
1350 | /** | |
1351 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
1352 | * @mdev: DRBD device. | |
1353 | * | |
1354 | * Called from process context only (admin command and after_state_ch). | |
1355 | */ | |
1356 | static int _drbd_pause_after(struct drbd_conf *mdev) | |
1357 | { | |
1358 | struct drbd_conf *odev; | |
1359 | int i, rv = 0; | |
1360 | ||
695d08fa | 1361 | rcu_read_lock(); |
81a5d60e | 1362 | idr_for_each_entry(&minors, odev, i) { |
b411b363 PR |
1363 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1364 | continue; | |
1365 | if (!_drbd_may_sync_now(odev)) | |
1366 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1367 | != SS_NOTHING_TO_DO); | |
1368 | } | |
695d08fa | 1369 | rcu_read_unlock(); |
b411b363 PR |
1370 | |
1371 | return rv; | |
1372 | } | |
1373 | ||
1374 | /** | |
1375 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
1376 | * @mdev: DRBD device. | |
1377 | * | |
1378 | * Called from process context only (admin command and worker). | |
1379 | */ | |
1380 | static int _drbd_resume_next(struct drbd_conf *mdev) | |
1381 | { | |
1382 | struct drbd_conf *odev; | |
1383 | int i, rv = 0; | |
1384 | ||
695d08fa | 1385 | rcu_read_lock(); |
81a5d60e | 1386 | idr_for_each_entry(&minors, odev, i) { |
b411b363 PR |
1387 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) |
1388 | continue; | |
1389 | if (odev->state.aftr_isp) { | |
1390 | if (_drbd_may_sync_now(odev)) | |
1391 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1392 | CS_HARD, NULL) | |
1393 | != SS_NOTHING_TO_DO) ; | |
1394 | } | |
1395 | } | |
695d08fa | 1396 | rcu_read_unlock(); |
b411b363 PR |
1397 | return rv; |
1398 | } | |
1399 | ||
1400 | void resume_next_sg(struct drbd_conf *mdev) | |
1401 | { | |
1402 | write_lock_irq(&global_state_lock); | |
1403 | _drbd_resume_next(mdev); | |
1404 | write_unlock_irq(&global_state_lock); | |
1405 | } | |
1406 | ||
1407 | void suspend_other_sg(struct drbd_conf *mdev) | |
1408 | { | |
1409 | write_lock_irq(&global_state_lock); | |
1410 | _drbd_pause_after(mdev); | |
1411 | write_unlock_irq(&global_state_lock); | |
1412 | } | |
1413 | ||
dc97b708 PR |
1414 | /* caller must hold global_state_lock */ |
1415 | enum drbd_ret_code drbd_sync_after_valid(struct drbd_conf *mdev, int o_minor) | |
b411b363 PR |
1416 | { |
1417 | struct drbd_conf *odev; | |
daeda1cc | 1418 | int ra; |
b411b363 PR |
1419 | |
1420 | if (o_minor == -1) | |
1421 | return NO_ERROR; | |
1422 | if (o_minor < -1 || minor_to_mdev(o_minor) == NULL) | |
1423 | return ERR_SYNC_AFTER; | |
1424 | ||
1425 | /* check for loops */ | |
1426 | odev = minor_to_mdev(o_minor); | |
1427 | while (1) { | |
1428 | if (odev == mdev) | |
1429 | return ERR_SYNC_AFTER_CYCLE; | |
1430 | ||
daeda1cc PR |
1431 | rcu_read_lock(); |
1432 | ra = rcu_dereference(odev->ldev->disk_conf)->resync_after; | |
1433 | rcu_read_unlock(); | |
b411b363 | 1434 | /* dependency chain ends here, no cycles. */ |
daeda1cc | 1435 | if (ra == -1) |
b411b363 PR |
1436 | return NO_ERROR; |
1437 | ||
1438 | /* follow the dependency chain */ | |
daeda1cc | 1439 | odev = minor_to_mdev(ra); |
b411b363 PR |
1440 | } |
1441 | } | |
1442 | ||
dc97b708 PR |
1443 | /* caller must hold global_state_lock */ |
1444 | void drbd_sync_after_changed(struct drbd_conf *mdev) | |
b411b363 PR |
1445 | { |
1446 | int changes; | |
b411b363 | 1447 | |
dc97b708 PR |
1448 | do { |
1449 | changes = _drbd_pause_after(mdev); | |
1450 | changes |= _drbd_resume_next(mdev); | |
1451 | } while (changes); | |
b411b363 PR |
1452 | } |
1453 | ||
9bd28d3c LE |
1454 | void drbd_rs_controller_reset(struct drbd_conf *mdev) |
1455 | { | |
1456 | atomic_set(&mdev->rs_sect_in, 0); | |
1457 | atomic_set(&mdev->rs_sect_ev, 0); | |
1458 | mdev->rs_in_flight = 0; | |
1459 | mdev->rs_planed = 0; | |
1460 | spin_lock(&mdev->peer_seq_lock); | |
1461 | fifo_set(&mdev->rs_plan_s, 0); | |
1462 | spin_unlock(&mdev->peer_seq_lock); | |
1463 | } | |
1464 | ||
1f04af33 PR |
1465 | void start_resync_timer_fn(unsigned long data) |
1466 | { | |
1467 | struct drbd_conf *mdev = (struct drbd_conf *) data; | |
1468 | ||
1469 | drbd_queue_work(&mdev->tconn->data.work, &mdev->start_resync_work); | |
1470 | } | |
1471 | ||
99920dc5 | 1472 | int w_start_resync(struct drbd_work *w, int cancel) |
1f04af33 | 1473 | { |
00d56944 PR |
1474 | struct drbd_conf *mdev = w->mdev; |
1475 | ||
1f04af33 PR |
1476 | if (atomic_read(&mdev->unacked_cnt) || atomic_read(&mdev->rs_pending_cnt)) { |
1477 | dev_warn(DEV, "w_start_resync later...\n"); | |
1478 | mdev->start_resync_timer.expires = jiffies + HZ/10; | |
1479 | add_timer(&mdev->start_resync_timer); | |
99920dc5 | 1480 | return 0; |
1f04af33 PR |
1481 | } |
1482 | ||
1483 | drbd_start_resync(mdev, C_SYNC_SOURCE); | |
1484 | clear_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags); | |
99920dc5 | 1485 | return 0; |
1f04af33 PR |
1486 | } |
1487 | ||
b411b363 PR |
1488 | /** |
1489 | * drbd_start_resync() - Start the resync process | |
1490 | * @mdev: DRBD device. | |
1491 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET | |
1492 | * | |
1493 | * This function might bring you directly into one of the | |
1494 | * C_PAUSED_SYNC_* states. | |
1495 | */ | |
1496 | void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side) | |
1497 | { | |
1498 | union drbd_state ns; | |
1499 | int r; | |
1500 | ||
c4752ef1 | 1501 | if (mdev->state.conn >= C_SYNC_SOURCE && mdev->state.conn < C_AHEAD) { |
b411b363 PR |
1502 | dev_err(DEV, "Resync already running!\n"); |
1503 | return; | |
1504 | } | |
1505 | ||
59817f4f PR |
1506 | if (mdev->state.conn < C_AHEAD) { |
1507 | /* In case a previous resync run was aborted by an IO error/detach on the peer. */ | |
1508 | drbd_rs_cancel_all(mdev); | |
1509 | /* This should be done when we abort the resync. We definitely do not | |
1510 | want to have this for connections going back and forth between | |
1511 | Ahead/Behind and SyncSource/SyncTarget */ | |
1512 | } | |
b411b363 | 1513 | |
e64a3294 PR |
1514 | if (!test_bit(B_RS_H_DONE, &mdev->flags)) { |
1515 | if (side == C_SYNC_TARGET) { | |
1516 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1517 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1518 | we check that we might make the data inconsistent. */ | |
1519 | r = drbd_khelper(mdev, "before-resync-target"); | |
1520 | r = (r >> 8) & 0xff; | |
1521 | if (r > 0) { | |
1522 | dev_info(DEV, "before-resync-target handler returned %d, " | |
09b9e797 | 1523 | "dropping connection.\n", r); |
38fa9988 | 1524 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
09b9e797 PR |
1525 | return; |
1526 | } | |
e64a3294 PR |
1527 | } else /* C_SYNC_SOURCE */ { |
1528 | r = drbd_khelper(mdev, "before-resync-source"); | |
1529 | r = (r >> 8) & 0xff; | |
1530 | if (r > 0) { | |
1531 | if (r == 3) { | |
1532 | dev_info(DEV, "before-resync-source handler returned %d, " | |
1533 | "ignoring. Old userland tools?", r); | |
1534 | } else { | |
1535 | dev_info(DEV, "before-resync-source handler returned %d, " | |
1536 | "dropping connection.\n", r); | |
38fa9988 | 1537 | conn_request_state(mdev->tconn, NS(conn, C_DISCONNECTING), CS_HARD); |
e64a3294 PR |
1538 | return; |
1539 | } | |
1540 | } | |
09b9e797 | 1541 | } |
b411b363 PR |
1542 | } |
1543 | ||
e64a3294 | 1544 | if (current == mdev->tconn->worker.task) { |
dad20554 | 1545 | /* The worker should not sleep waiting for state_mutex, |
e64a3294 | 1546 | that can take long */ |
8410da8f | 1547 | if (!mutex_trylock(mdev->state_mutex)) { |
e64a3294 PR |
1548 | set_bit(B_RS_H_DONE, &mdev->flags); |
1549 | mdev->start_resync_timer.expires = jiffies + HZ/5; | |
1550 | add_timer(&mdev->start_resync_timer); | |
1551 | return; | |
1552 | } | |
1553 | } else { | |
8410da8f | 1554 | mutex_lock(mdev->state_mutex); |
e64a3294 PR |
1555 | } |
1556 | clear_bit(B_RS_H_DONE, &mdev->flags); | |
b411b363 PR |
1557 | |
1558 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
8410da8f | 1559 | mutex_unlock(mdev->state_mutex); |
b411b363 PR |
1560 | return; |
1561 | } | |
1562 | ||
b411b363 | 1563 | write_lock_irq(&global_state_lock); |
78bae59b | 1564 | ns = drbd_read_state(mdev); |
b411b363 PR |
1565 | |
1566 | ns.aftr_isp = !_drbd_may_sync_now(mdev); | |
1567 | ||
1568 | ns.conn = side; | |
1569 | ||
1570 | if (side == C_SYNC_TARGET) | |
1571 | ns.disk = D_INCONSISTENT; | |
1572 | else /* side == C_SYNC_SOURCE */ | |
1573 | ns.pdsk = D_INCONSISTENT; | |
1574 | ||
1575 | r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
78bae59b | 1576 | ns = drbd_read_state(mdev); |
b411b363 PR |
1577 | |
1578 | if (ns.conn < C_CONNECTED) | |
1579 | r = SS_UNKNOWN_ERROR; | |
1580 | ||
1581 | if (r == SS_SUCCESS) { | |
1d7734a0 LE |
1582 | unsigned long tw = drbd_bm_total_weight(mdev); |
1583 | unsigned long now = jiffies; | |
1584 | int i; | |
1585 | ||
b411b363 PR |
1586 | mdev->rs_failed = 0; |
1587 | mdev->rs_paused = 0; | |
b411b363 | 1588 | mdev->rs_same_csum = 0; |
0f0601f4 LE |
1589 | mdev->rs_last_events = 0; |
1590 | mdev->rs_last_sect_ev = 0; | |
1d7734a0 LE |
1591 | mdev->rs_total = tw; |
1592 | mdev->rs_start = now; | |
1593 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
1594 | mdev->rs_mark_left[i] = tw; | |
1595 | mdev->rs_mark_time[i] = now; | |
1596 | } | |
b411b363 PR |
1597 | _drbd_pause_after(mdev); |
1598 | } | |
1599 | write_unlock_irq(&global_state_lock); | |
5a22db89 | 1600 | |
b411b363 PR |
1601 | if (r == SS_SUCCESS) { |
1602 | dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", | |
1603 | drbd_conn_str(ns.conn), | |
1604 | (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10), | |
1605 | (unsigned long) mdev->rs_total); | |
6c922ed5 LE |
1606 | if (side == C_SYNC_TARGET) |
1607 | mdev->bm_resync_fo = 0; | |
1608 | ||
1609 | /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid | |
1610 | * with w_send_oos, or the sync target will get confused as to | |
1611 | * how much bits to resync. We cannot do that always, because for an | |
1612 | * empty resync and protocol < 95, we need to do it here, as we call | |
1613 | * drbd_resync_finished from here in that case. | |
1614 | * We drbd_gen_and_send_sync_uuid here for protocol < 96, | |
1615 | * and from after_state_ch otherwise. */ | |
31890f4a | 1616 | if (side == C_SYNC_SOURCE && mdev->tconn->agreed_pro_version < 96) |
6c922ed5 | 1617 | drbd_gen_and_send_sync_uuid(mdev); |
b411b363 | 1618 | |
31890f4a | 1619 | if (mdev->tconn->agreed_pro_version < 95 && mdev->rs_total == 0) { |
af85e8e8 LE |
1620 | /* This still has a race (about when exactly the peers |
1621 | * detect connection loss) that can lead to a full sync | |
1622 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1623 | * resync-finished notifications, but the fix | |
1624 | * introduces a protocol change. Sleeping for some | |
1625 | * time longer than the ping interval + timeout on the | |
1626 | * SyncSource, to give the SyncTarget the chance to | |
1627 | * detect connection loss, then waiting for a ping | |
1628 | * response (implicit in drbd_resync_finished) reduces | |
1629 | * the race considerably, but does not solve it. */ | |
44ed167d PR |
1630 | if (side == C_SYNC_SOURCE) { |
1631 | struct net_conf *nc; | |
1632 | int timeo; | |
1633 | ||
1634 | rcu_read_lock(); | |
1635 | nc = rcu_dereference(mdev->tconn->net_conf); | |
1636 | timeo = nc->ping_int * HZ + nc->ping_timeo * HZ / 9; | |
1637 | rcu_read_unlock(); | |
1638 | schedule_timeout_interruptible(timeo); | |
1639 | } | |
b411b363 | 1640 | drbd_resync_finished(mdev); |
b411b363 PR |
1641 | } |
1642 | ||
9bd28d3c | 1643 | drbd_rs_controller_reset(mdev); |
b411b363 PR |
1644 | /* ns.conn may already be != mdev->state.conn, |
1645 | * we may have been paused in between, or become paused until | |
1646 | * the timer triggers. | |
1647 | * No matter, that is handled in resync_timer_fn() */ | |
1648 | if (ns.conn == C_SYNC_TARGET) | |
1649 | mod_timer(&mdev->resync_timer, jiffies); | |
1650 | ||
1651 | drbd_md_sync(mdev); | |
1652 | } | |
5a22db89 | 1653 | put_ldev(mdev); |
8410da8f | 1654 | mutex_unlock(mdev->state_mutex); |
b411b363 PR |
1655 | } |
1656 | ||
1657 | int drbd_worker(struct drbd_thread *thi) | |
1658 | { | |
392c8801 | 1659 | struct drbd_tconn *tconn = thi->tconn; |
b411b363 | 1660 | struct drbd_work *w = NULL; |
0e29d163 | 1661 | struct drbd_conf *mdev; |
44ed167d | 1662 | struct net_conf *nc; |
b411b363 | 1663 | LIST_HEAD(work_list); |
f399002e | 1664 | int vnr, intr = 0; |
44ed167d | 1665 | int cork; |
b411b363 | 1666 | |
e77a0a5c | 1667 | while (get_t_state(thi) == RUNNING) { |
80822284 | 1668 | drbd_thread_current_set_cpu(thi); |
b411b363 | 1669 | |
19393e10 PR |
1670 | if (down_trylock(&tconn->data.work.s)) { |
1671 | mutex_lock(&tconn->data.mutex); | |
44ed167d PR |
1672 | |
1673 | rcu_read_lock(); | |
1674 | nc = rcu_dereference(tconn->net_conf); | |
1675 | cork = nc ? !nc->no_cork : 0; | |
1676 | rcu_read_unlock(); | |
1677 | ||
1678 | if (tconn->data.socket && cork) | |
19393e10 PR |
1679 | drbd_tcp_uncork(tconn->data.socket); |
1680 | mutex_unlock(&tconn->data.mutex); | |
b411b363 | 1681 | |
19393e10 | 1682 | intr = down_interruptible(&tconn->data.work.s); |
b411b363 | 1683 | |
19393e10 | 1684 | mutex_lock(&tconn->data.mutex); |
44ed167d | 1685 | if (tconn->data.socket && cork) |
19393e10 PR |
1686 | drbd_tcp_cork(tconn->data.socket); |
1687 | mutex_unlock(&tconn->data.mutex); | |
b411b363 PR |
1688 | } |
1689 | ||
1690 | if (intr) { | |
b411b363 | 1691 | flush_signals(current); |
19393e10 PR |
1692 | if (get_t_state(thi) == RUNNING) { |
1693 | conn_warn(tconn, "Worker got an unexpected signal\n"); | |
b411b363 | 1694 | continue; |
19393e10 | 1695 | } |
b411b363 PR |
1696 | break; |
1697 | } | |
1698 | ||
e77a0a5c | 1699 | if (get_t_state(thi) != RUNNING) |
b411b363 PR |
1700 | break; |
1701 | /* With this break, we have done a down() but not consumed | |
1702 | the entry from the list. The cleanup code takes care of | |
1703 | this... */ | |
1704 | ||
1705 | w = NULL; | |
19393e10 PR |
1706 | spin_lock_irq(&tconn->data.work.q_lock); |
1707 | if (list_empty(&tconn->data.work.q)) { | |
b411b363 PR |
1708 | /* something terribly wrong in our logic. |
1709 | * we were able to down() the semaphore, | |
1710 | * but the list is empty... doh. | |
1711 | * | |
1712 | * what is the best thing to do now? | |
1713 | * try again from scratch, restarting the receiver, | |
1714 | * asender, whatnot? could break even more ugly, | |
1715 | * e.g. when we are primary, but no good local data. | |
1716 | * | |
1717 | * I'll try to get away just starting over this loop. | |
1718 | */ | |
19393e10 PR |
1719 | conn_warn(tconn, "Work list unexpectedly empty\n"); |
1720 | spin_unlock_irq(&tconn->data.work.q_lock); | |
b411b363 PR |
1721 | continue; |
1722 | } | |
19393e10 | 1723 | w = list_entry(tconn->data.work.q.next, struct drbd_work, list); |
b411b363 | 1724 | list_del_init(&w->list); |
19393e10 | 1725 | spin_unlock_irq(&tconn->data.work.q_lock); |
b411b363 | 1726 | |
99920dc5 | 1727 | if (w->cb(w, tconn->cstate < C_WF_REPORT_PARAMS)) { |
b411b363 | 1728 | /* dev_warn(DEV, "worker: a callback failed! \n"); */ |
bbeb641c PR |
1729 | if (tconn->cstate >= C_WF_REPORT_PARAMS) |
1730 | conn_request_state(tconn, NS(conn, C_NETWORK_FAILURE), CS_HARD); | |
b411b363 PR |
1731 | } |
1732 | } | |
b411b363 | 1733 | |
19393e10 PR |
1734 | spin_lock_irq(&tconn->data.work.q_lock); |
1735 | while (!list_empty(&tconn->data.work.q)) { | |
1736 | list_splice_init(&tconn->data.work.q, &work_list); | |
1737 | spin_unlock_irq(&tconn->data.work.q_lock); | |
b411b363 PR |
1738 | |
1739 | while (!list_empty(&work_list)) { | |
1740 | w = list_entry(work_list.next, struct drbd_work, list); | |
1741 | list_del_init(&w->list); | |
00d56944 | 1742 | w->cb(w, 1); |
b411b363 PR |
1743 | } |
1744 | ||
19393e10 | 1745 | spin_lock_irq(&tconn->data.work.q_lock); |
b411b363 | 1746 | } |
19393e10 | 1747 | sema_init(&tconn->data.work.s, 0); |
b411b363 PR |
1748 | /* DANGEROUS race: if someone did queue his work within the spinlock, |
1749 | * but up() ed outside the spinlock, we could get an up() on the | |
1750 | * semaphore without corresponding list entry. | |
1751 | * So don't do that. | |
1752 | */ | |
19393e10 | 1753 | spin_unlock_irq(&tconn->data.work.q_lock); |
b411b363 | 1754 | |
d3fcb490 | 1755 | down_read(&drbd_cfg_rwsem); |
f399002e | 1756 | idr_for_each_entry(&tconn->volumes, mdev, vnr) { |
0e29d163 | 1757 | D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE); |
0e29d163 PR |
1758 | drbd_mdev_cleanup(mdev); |
1759 | } | |
d3fcb490 | 1760 | up_read(&drbd_cfg_rwsem); |
b411b363 PR |
1761 | |
1762 | return 0; | |
1763 | } |