]>
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> | |
29 | #include <linux/smp_lock.h> | |
30 | #include <linux/wait.h> | |
31 | #include <linux/mm.h> | |
32 | #include <linux/memcontrol.h> | |
33 | #include <linux/mm_inline.h> | |
34 | #include <linux/slab.h> | |
35 | #include <linux/random.h> | |
b411b363 PR |
36 | #include <linux/string.h> |
37 | #include <linux/scatterlist.h> | |
38 | ||
39 | #include "drbd_int.h" | |
40 | #include "drbd_req.h" | |
b411b363 | 41 | |
b411b363 PR |
42 | static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel); |
43 | ||
44 | ||
45 | ||
46 | /* defined here: | |
47 | drbd_md_io_complete | |
45bb912b | 48 | drbd_endio_sec |
b411b363 PR |
49 | drbd_endio_pri |
50 | ||
51 | * more endio handlers: | |
52 | atodb_endio in drbd_actlog.c | |
53 | drbd_bm_async_io_complete in drbd_bitmap.c | |
54 | ||
55 | * For all these callbacks, note the following: | |
56 | * The callbacks will be called in irq context by the IDE drivers, | |
57 | * and in Softirqs/Tasklets/BH context by the SCSI drivers. | |
58 | * Try to get the locking right :) | |
59 | * | |
60 | */ | |
61 | ||
62 | ||
63 | /* About the global_state_lock | |
64 | Each state transition on an device holds a read lock. In case we have | |
65 | to evaluate the sync after dependencies, we grab a write lock, because | |
66 | we need stable states on all devices for that. */ | |
67 | rwlock_t global_state_lock; | |
68 | ||
69 | /* used for synchronous meta data and bitmap IO | |
70 | * submitted by drbd_md_sync_page_io() | |
71 | */ | |
72 | void drbd_md_io_complete(struct bio *bio, int error) | |
73 | { | |
74 | struct drbd_md_io *md_io; | |
75 | ||
76 | md_io = (struct drbd_md_io *)bio->bi_private; | |
77 | md_io->error = error; | |
78 | ||
b411b363 PR |
79 | complete(&md_io->event); |
80 | } | |
81 | ||
82 | /* reads on behalf of the partner, | |
83 | * "submitted" by the receiver | |
84 | */ | |
45bb912b | 85 | void drbd_endio_read_sec_final(struct drbd_epoch_entry *e) __releases(local) |
b411b363 PR |
86 | { |
87 | unsigned long flags = 0; | |
45bb912b | 88 | struct drbd_conf *mdev = e->mdev; |
b411b363 PR |
89 | |
90 | D_ASSERT(e->block_id != ID_VACANT); | |
91 | ||
b411b363 PR |
92 | spin_lock_irqsave(&mdev->req_lock, flags); |
93 | mdev->read_cnt += e->size >> 9; | |
94 | list_del(&e->w.list); | |
95 | if (list_empty(&mdev->read_ee)) | |
96 | wake_up(&mdev->ee_wait); | |
45bb912b LE |
97 | if (test_bit(__EE_WAS_ERROR, &e->flags)) |
98 | __drbd_chk_io_error(mdev, FALSE); | |
b411b363 PR |
99 | spin_unlock_irqrestore(&mdev->req_lock, flags); |
100 | ||
b411b363 PR |
101 | drbd_queue_work(&mdev->data.work, &e->w); |
102 | put_ldev(mdev); | |
b411b363 PR |
103 | } |
104 | ||
105 | /* writes on behalf of the partner, or resync writes, | |
45bb912b LE |
106 | * "submitted" by the receiver, final stage. */ |
107 | static void drbd_endio_write_sec_final(struct drbd_epoch_entry *e) __releases(local) | |
b411b363 PR |
108 | { |
109 | unsigned long flags = 0; | |
45bb912b | 110 | struct drbd_conf *mdev = e->mdev; |
b411b363 PR |
111 | sector_t e_sector; |
112 | int do_wake; | |
113 | int is_syncer_req; | |
114 | int do_al_complete_io; | |
b411b363 | 115 | |
b411b363 PR |
116 | D_ASSERT(e->block_id != ID_VACANT); |
117 | ||
b411b363 PR |
118 | /* after we moved e to done_ee, |
119 | * we may no longer access it, | |
120 | * it may be freed/reused already! | |
121 | * (as soon as we release the req_lock) */ | |
122 | e_sector = e->sector; | |
123 | do_al_complete_io = e->flags & EE_CALL_AL_COMPLETE_IO; | |
45bb912b | 124 | is_syncer_req = is_syncer_block_id(e->block_id); |
b411b363 | 125 | |
45bb912b LE |
126 | spin_lock_irqsave(&mdev->req_lock, flags); |
127 | mdev->writ_cnt += e->size >> 9; | |
b411b363 PR |
128 | list_del(&e->w.list); /* has been on active_ee or sync_ee */ |
129 | list_add_tail(&e->w.list, &mdev->done_ee); | |
130 | ||
b411b363 PR |
131 | /* No hlist_del_init(&e->colision) here, we did not send the Ack yet, |
132 | * neither did we wake possibly waiting conflicting requests. | |
133 | * done from "drbd_process_done_ee" within the appropriate w.cb | |
134 | * (e_end_block/e_end_resync_block) or from _drbd_clear_done_ee */ | |
135 | ||
136 | do_wake = is_syncer_req | |
137 | ? list_empty(&mdev->sync_ee) | |
138 | : list_empty(&mdev->active_ee); | |
139 | ||
45bb912b | 140 | if (test_bit(__EE_WAS_ERROR, &e->flags)) |
b411b363 PR |
141 | __drbd_chk_io_error(mdev, FALSE); |
142 | spin_unlock_irqrestore(&mdev->req_lock, flags); | |
143 | ||
144 | if (is_syncer_req) | |
145 | drbd_rs_complete_io(mdev, e_sector); | |
146 | ||
147 | if (do_wake) | |
148 | wake_up(&mdev->ee_wait); | |
149 | ||
150 | if (do_al_complete_io) | |
151 | drbd_al_complete_io(mdev, e_sector); | |
152 | ||
153 | wake_asender(mdev); | |
154 | put_ldev(mdev); | |
45bb912b | 155 | } |
b411b363 | 156 | |
45bb912b LE |
157 | /* writes on behalf of the partner, or resync writes, |
158 | * "submitted" by the receiver. | |
159 | */ | |
160 | void drbd_endio_sec(struct bio *bio, int error) | |
161 | { | |
162 | struct drbd_epoch_entry *e = bio->bi_private; | |
163 | struct drbd_conf *mdev = e->mdev; | |
164 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
165 | int is_write = bio_data_dir(bio) == WRITE; | |
166 | ||
167 | if (error) | |
168 | dev_warn(DEV, "%s: error=%d s=%llus\n", | |
169 | is_write ? "write" : "read", error, | |
170 | (unsigned long long)e->sector); | |
171 | if (!error && !uptodate) { | |
172 | dev_warn(DEV, "%s: setting error to -EIO s=%llus\n", | |
173 | is_write ? "write" : "read", | |
174 | (unsigned long long)e->sector); | |
175 | /* strange behavior of some lower level drivers... | |
176 | * fail the request by clearing the uptodate flag, | |
177 | * but do not return any error?! */ | |
178 | error = -EIO; | |
179 | } | |
180 | ||
181 | if (error) | |
182 | set_bit(__EE_WAS_ERROR, &e->flags); | |
183 | ||
184 | bio_put(bio); /* no need for the bio anymore */ | |
185 | if (atomic_dec_and_test(&e->pending_bios)) { | |
186 | if (is_write) | |
187 | drbd_endio_write_sec_final(e); | |
188 | else | |
189 | drbd_endio_read_sec_final(e); | |
190 | } | |
b411b363 PR |
191 | } |
192 | ||
193 | /* read, readA or write requests on R_PRIMARY coming from drbd_make_request | |
194 | */ | |
195 | void drbd_endio_pri(struct bio *bio, int error) | |
196 | { | |
b411b363 PR |
197 | struct drbd_request *req = bio->bi_private; |
198 | struct drbd_conf *mdev = req->mdev; | |
b411b363 PR |
199 | enum drbd_req_event what; |
200 | int uptodate = bio_flagged(bio, BIO_UPTODATE); | |
201 | ||
b411b363 PR |
202 | if (!error && !uptodate) { |
203 | dev_warn(DEV, "p %s: setting error to -EIO\n", | |
204 | bio_data_dir(bio) == WRITE ? "write" : "read"); | |
205 | /* strange behavior of some lower level drivers... | |
206 | * fail the request by clearing the uptodate flag, | |
207 | * but do not return any error?! */ | |
208 | error = -EIO; | |
209 | } | |
210 | ||
b411b363 PR |
211 | /* to avoid recursion in __req_mod */ |
212 | if (unlikely(error)) { | |
213 | what = (bio_data_dir(bio) == WRITE) | |
214 | ? write_completed_with_error | |
5c3c7e64 | 215 | : (bio_rw(bio) == READ) |
b411b363 PR |
216 | ? read_completed_with_error |
217 | : read_ahead_completed_with_error; | |
218 | } else | |
219 | what = completed_ok; | |
220 | ||
221 | bio_put(req->private_bio); | |
222 | req->private_bio = ERR_PTR(error); | |
223 | ||
0f0601f4 | 224 | req_mod(req, what); |
b411b363 PR |
225 | } |
226 | ||
b411b363 PR |
227 | int w_read_retry_remote(struct drbd_conf *mdev, struct drbd_work *w, int cancel) |
228 | { | |
229 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
230 | ||
231 | /* We should not detach for read io-error, | |
232 | * but try to WRITE the P_DATA_REPLY to the failed location, | |
233 | * to give the disk the chance to relocate that block */ | |
234 | ||
235 | spin_lock_irq(&mdev->req_lock); | |
d255e5ff LE |
236 | if (cancel || mdev->state.pdsk != D_UP_TO_DATE) { |
237 | _req_mod(req, read_retry_remote_canceled); | |
b411b363 | 238 | spin_unlock_irq(&mdev->req_lock); |
b411b363 PR |
239 | return 1; |
240 | } | |
241 | spin_unlock_irq(&mdev->req_lock); | |
242 | ||
243 | return w_send_read_req(mdev, w, 0); | |
244 | } | |
245 | ||
246 | int w_resync_inactive(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
247 | { | |
248 | ERR_IF(cancel) return 1; | |
249 | dev_err(DEV, "resync inactive, but callback triggered??\n"); | |
250 | return 1; /* Simply ignore this! */ | |
251 | } | |
252 | ||
45bb912b LE |
253 | void drbd_csum_ee(struct drbd_conf *mdev, struct crypto_hash *tfm, struct drbd_epoch_entry *e, void *digest) |
254 | { | |
255 | struct hash_desc desc; | |
256 | struct scatterlist sg; | |
257 | struct page *page = e->pages; | |
258 | struct page *tmp; | |
259 | unsigned len; | |
260 | ||
261 | desc.tfm = tfm; | |
262 | desc.flags = 0; | |
263 | ||
264 | sg_init_table(&sg, 1); | |
265 | crypto_hash_init(&desc); | |
266 | ||
267 | while ((tmp = page_chain_next(page))) { | |
268 | /* all but the last page will be fully used */ | |
269 | sg_set_page(&sg, page, PAGE_SIZE, 0); | |
270 | crypto_hash_update(&desc, &sg, sg.length); | |
271 | page = tmp; | |
272 | } | |
273 | /* and now the last, possibly only partially used page */ | |
274 | len = e->size & (PAGE_SIZE - 1); | |
275 | sg_set_page(&sg, page, len ?: PAGE_SIZE, 0); | |
276 | crypto_hash_update(&desc, &sg, sg.length); | |
277 | crypto_hash_final(&desc, digest); | |
278 | } | |
279 | ||
280 | void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest) | |
b411b363 PR |
281 | { |
282 | struct hash_desc desc; | |
283 | struct scatterlist sg; | |
284 | struct bio_vec *bvec; | |
285 | int i; | |
286 | ||
287 | desc.tfm = tfm; | |
288 | desc.flags = 0; | |
289 | ||
290 | sg_init_table(&sg, 1); | |
291 | crypto_hash_init(&desc); | |
292 | ||
293 | __bio_for_each_segment(bvec, bio, i, 0) { | |
294 | sg_set_page(&sg, bvec->bv_page, bvec->bv_len, bvec->bv_offset); | |
295 | crypto_hash_update(&desc, &sg, sg.length); | |
296 | } | |
297 | crypto_hash_final(&desc, digest); | |
298 | } | |
299 | ||
300 | static int w_e_send_csum(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
301 | { | |
302 | struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); | |
303 | int digest_size; | |
304 | void *digest; | |
305 | int ok; | |
306 | ||
307 | D_ASSERT(e->block_id == DRBD_MAGIC + 0xbeef); | |
308 | ||
309 | if (unlikely(cancel)) { | |
310 | drbd_free_ee(mdev, e); | |
311 | return 1; | |
312 | } | |
313 | ||
45bb912b | 314 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
315 | digest_size = crypto_hash_digestsize(mdev->csums_tfm); |
316 | digest = kmalloc(digest_size, GFP_NOIO); | |
317 | if (digest) { | |
45bb912b | 318 | drbd_csum_ee(mdev, mdev->csums_tfm, e, digest); |
b411b363 PR |
319 | |
320 | inc_rs_pending(mdev); | |
321 | ok = drbd_send_drequest_csum(mdev, | |
322 | e->sector, | |
323 | e->size, | |
324 | digest, | |
325 | digest_size, | |
326 | P_CSUM_RS_REQUEST); | |
327 | kfree(digest); | |
328 | } else { | |
329 | dev_err(DEV, "kmalloc() of digest failed.\n"); | |
330 | ok = 0; | |
331 | } | |
332 | } else | |
333 | ok = 1; | |
334 | ||
335 | drbd_free_ee(mdev, e); | |
336 | ||
337 | if (unlikely(!ok)) | |
338 | dev_err(DEV, "drbd_send_drequest(..., csum) failed\n"); | |
339 | return ok; | |
340 | } | |
341 | ||
342 | #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN) | |
343 | ||
344 | static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size) | |
345 | { | |
346 | struct drbd_epoch_entry *e; | |
347 | ||
348 | if (!get_ldev(mdev)) | |
80a40e43 | 349 | return -EIO; |
b411b363 | 350 | |
0f0601f4 LE |
351 | if (drbd_rs_should_slow_down(mdev)) |
352 | goto defer; | |
353 | ||
b411b363 PR |
354 | /* GFP_TRY, because if there is no memory available right now, this may |
355 | * be rescheduled for later. It is "only" background resync, after all. */ | |
356 | e = drbd_alloc_ee(mdev, DRBD_MAGIC+0xbeef, sector, size, GFP_TRY); | |
45bb912b | 357 | if (!e) |
80a40e43 | 358 | goto defer; |
b411b363 | 359 | |
80a40e43 | 360 | e->w.cb = w_e_send_csum; |
b411b363 PR |
361 | spin_lock_irq(&mdev->req_lock); |
362 | list_add(&e->w.list, &mdev->read_ee); | |
363 | spin_unlock_irq(&mdev->req_lock); | |
364 | ||
0f0601f4 | 365 | atomic_add(size >> 9, &mdev->rs_sect_ev); |
45bb912b | 366 | if (drbd_submit_ee(mdev, e, READ, DRBD_FAULT_RS_RD) == 0) |
80a40e43 | 367 | return 0; |
b411b363 | 368 | |
22cc37a9 LE |
369 | /* drbd_submit_ee currently fails for one reason only: |
370 | * not being able to allocate enough bios. | |
371 | * Is dropping the connection going to help? */ | |
372 | spin_lock_irq(&mdev->req_lock); | |
373 | list_del(&e->w.list); | |
374 | spin_unlock_irq(&mdev->req_lock); | |
375 | ||
45bb912b | 376 | drbd_free_ee(mdev, e); |
80a40e43 | 377 | defer: |
45bb912b | 378 | put_ldev(mdev); |
80a40e43 | 379 | return -EAGAIN; |
b411b363 PR |
380 | } |
381 | ||
382 | void resync_timer_fn(unsigned long data) | |
383 | { | |
b411b363 PR |
384 | struct drbd_conf *mdev = (struct drbd_conf *) data; |
385 | int queue; | |
386 | ||
63106d3c PR |
387 | queue = 1; |
388 | switch (mdev->state.conn) { | |
389 | case C_VERIFY_S: | |
390 | mdev->resync_work.cb = w_make_ov_request; | |
391 | break; | |
392 | case C_SYNC_TARGET: | |
393 | mdev->resync_work.cb = w_make_resync_request; | |
394 | break; | |
395 | default: | |
b411b363 PR |
396 | queue = 0; |
397 | mdev->resync_work.cb = w_resync_inactive; | |
398 | } | |
399 | ||
b411b363 PR |
400 | /* harmless race: list_empty outside data.work.q_lock */ |
401 | if (list_empty(&mdev->resync_work.list) && queue) | |
402 | drbd_queue_work(&mdev->data.work, &mdev->resync_work); | |
403 | } | |
404 | ||
778f271d PR |
405 | static void fifo_set(struct fifo_buffer *fb, int value) |
406 | { | |
407 | int i; | |
408 | ||
409 | for (i = 0; i < fb->size; i++) | |
f10f2623 | 410 | fb->values[i] = value; |
778f271d PR |
411 | } |
412 | ||
413 | static int fifo_push(struct fifo_buffer *fb, int value) | |
414 | { | |
415 | int ov; | |
416 | ||
417 | ov = fb->values[fb->head_index]; | |
418 | fb->values[fb->head_index++] = value; | |
419 | ||
420 | if (fb->head_index >= fb->size) | |
421 | fb->head_index = 0; | |
422 | ||
423 | return ov; | |
424 | } | |
425 | ||
426 | static void fifo_add_val(struct fifo_buffer *fb, int value) | |
427 | { | |
428 | int i; | |
429 | ||
430 | for (i = 0; i < fb->size; i++) | |
431 | fb->values[i] += value; | |
432 | } | |
433 | ||
434 | int drbd_rs_controller(struct drbd_conf *mdev) | |
435 | { | |
436 | unsigned int sect_in; /* Number of sectors that came in since the last turn */ | |
437 | unsigned int want; /* The number of sectors we want in the proxy */ | |
438 | int req_sect; /* Number of sectors to request in this turn */ | |
439 | int correction; /* Number of sectors more we need in the proxy*/ | |
440 | int cps; /* correction per invocation of drbd_rs_controller() */ | |
441 | int steps; /* Number of time steps to plan ahead */ | |
442 | int curr_corr; | |
443 | int max_sect; | |
444 | ||
445 | sect_in = atomic_xchg(&mdev->rs_sect_in, 0); /* Number of sectors that came in */ | |
446 | mdev->rs_in_flight -= sect_in; | |
447 | ||
448 | spin_lock(&mdev->peer_seq_lock); /* get an atomic view on mdev->rs_plan_s */ | |
449 | ||
450 | steps = mdev->rs_plan_s.size; /* (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; */ | |
451 | ||
452 | if (mdev->rs_in_flight + sect_in == 0) { /* At start of resync */ | |
453 | want = ((mdev->sync_conf.rate * 2 * SLEEP_TIME) / HZ) * steps; | |
454 | } else { /* normal path */ | |
455 | want = mdev->sync_conf.c_fill_target ? mdev->sync_conf.c_fill_target : | |
456 | sect_in * mdev->sync_conf.c_delay_target * HZ / (SLEEP_TIME * 10); | |
457 | } | |
458 | ||
459 | correction = want - mdev->rs_in_flight - mdev->rs_planed; | |
460 | ||
461 | /* Plan ahead */ | |
462 | cps = correction / steps; | |
463 | fifo_add_val(&mdev->rs_plan_s, cps); | |
464 | mdev->rs_planed += cps * steps; | |
465 | ||
466 | /* What we do in this step */ | |
467 | curr_corr = fifo_push(&mdev->rs_plan_s, 0); | |
468 | spin_unlock(&mdev->peer_seq_lock); | |
469 | mdev->rs_planed -= curr_corr; | |
470 | ||
471 | req_sect = sect_in + curr_corr; | |
472 | if (req_sect < 0) | |
473 | req_sect = 0; | |
474 | ||
475 | max_sect = (mdev->sync_conf.c_max_rate * 2 * SLEEP_TIME) / HZ; | |
476 | if (req_sect > max_sect) | |
477 | req_sect = max_sect; | |
478 | ||
479 | /* | |
480 | dev_warn(DEV, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n", | |
481 | sect_in, mdev->rs_in_flight, want, correction, | |
482 | steps, cps, mdev->rs_planed, curr_corr, req_sect); | |
483 | */ | |
484 | ||
485 | return req_sect; | |
486 | } | |
487 | ||
b411b363 PR |
488 | int w_make_resync_request(struct drbd_conf *mdev, |
489 | struct drbd_work *w, int cancel) | |
490 | { | |
491 | unsigned long bit; | |
492 | sector_t sector; | |
493 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
bb3d000c | 494 | int max_segment_size; |
0f0601f4 | 495 | int number, rollback_i, size, pe, mx; |
b411b363 | 496 | int align, queued, sndbuf; |
0f0601f4 | 497 | int i = 0; |
b411b363 PR |
498 | |
499 | if (unlikely(cancel)) | |
500 | return 1; | |
501 | ||
502 | if (unlikely(mdev->state.conn < C_CONNECTED)) { | |
503 | dev_err(DEV, "Confused in w_make_resync_request()! cstate < Connected"); | |
504 | return 0; | |
505 | } | |
506 | ||
507 | if (mdev->state.conn != C_SYNC_TARGET) | |
508 | dev_err(DEV, "%s in w_make_resync_request\n", | |
509 | drbd_conn_str(mdev->state.conn)); | |
510 | ||
af85e8e8 LE |
511 | if (mdev->rs_total == 0) { |
512 | /* empty resync? */ | |
513 | drbd_resync_finished(mdev); | |
514 | return 1; | |
515 | } | |
516 | ||
b411b363 PR |
517 | if (!get_ldev(mdev)) { |
518 | /* Since we only need to access mdev->rsync a | |
519 | get_ldev_if_state(mdev,D_FAILED) would be sufficient, but | |
520 | to continue resync with a broken disk makes no sense at | |
521 | all */ | |
522 | dev_err(DEV, "Disk broke down during resync!\n"); | |
523 | mdev->resync_work.cb = w_resync_inactive; | |
524 | return 1; | |
525 | } | |
526 | ||
bb3d000c LE |
527 | /* starting with drbd 8.3.8, we can handle multi-bio EEs, |
528 | * if it should be necessary */ | |
5a75cc7c PR |
529 | max_segment_size = |
530 | mdev->agreed_pro_version < 94 ? queue_max_segment_size(mdev->rq_queue) : | |
531 | mdev->agreed_pro_version < 95 ? DRBD_MAX_SIZE_H80_PACKET : DRBD_MAX_SEGMENT_SIZE; | |
bb3d000c | 532 | |
778f271d PR |
533 | if (mdev->rs_plan_s.size) { /* mdev->sync_conf.c_plan_ahead */ |
534 | number = drbd_rs_controller(mdev) >> (BM_BLOCK_SHIFT - 9); | |
535 | mdev->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME; | |
536 | } else { | |
537 | mdev->c_sync_rate = mdev->sync_conf.rate; | |
538 | number = SLEEP_TIME * mdev->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ); | |
539 | } | |
0f0601f4 LE |
540 | |
541 | /* Throttle resync on lower level disk activity, which may also be | |
542 | * caused by application IO on Primary/SyncTarget. | |
543 | * Keep this after the call to drbd_rs_controller, as that assumes | |
544 | * to be called as precisely as possible every SLEEP_TIME, | |
545 | * and would be confused otherwise. */ | |
546 | if (drbd_rs_should_slow_down(mdev)) | |
547 | goto requeue; | |
b411b363 PR |
548 | |
549 | mutex_lock(&mdev->data.mutex); | |
550 | if (mdev->data.socket) | |
551 | mx = mdev->data.socket->sk->sk_rcvbuf / sizeof(struct p_block_req); | |
552 | else | |
553 | mx = 1; | |
554 | mutex_unlock(&mdev->data.mutex); | |
555 | ||
556 | /* For resync rates >160MB/sec, allow more pending RS requests */ | |
557 | if (number > mx) | |
558 | mx = number; | |
559 | ||
560 | /* Limit the number of pending RS requests to no more than the peer's receive buffer */ | |
0f0601f4 | 561 | pe = atomic_read(&mdev->rs_pending_cnt); |
b411b363 PR |
562 | if ((pe + number) > mx) { |
563 | number = mx - pe; | |
564 | } | |
565 | ||
566 | for (i = 0; i < number; i++) { | |
567 | /* Stop generating RS requests, when half of the send buffer is filled */ | |
568 | mutex_lock(&mdev->data.mutex); | |
569 | if (mdev->data.socket) { | |
570 | queued = mdev->data.socket->sk->sk_wmem_queued; | |
571 | sndbuf = mdev->data.socket->sk->sk_sndbuf; | |
572 | } else { | |
573 | queued = 1; | |
574 | sndbuf = 0; | |
575 | } | |
576 | mutex_unlock(&mdev->data.mutex); | |
577 | if (queued > sndbuf / 2) | |
578 | goto requeue; | |
579 | ||
580 | next_sector: | |
581 | size = BM_BLOCK_SIZE; | |
582 | bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo); | |
583 | ||
584 | if (bit == -1UL) { | |
585 | mdev->bm_resync_fo = drbd_bm_bits(mdev); | |
586 | mdev->resync_work.cb = w_resync_inactive; | |
587 | put_ldev(mdev); | |
588 | return 1; | |
589 | } | |
590 | ||
591 | sector = BM_BIT_TO_SECT(bit); | |
592 | ||
593 | if (drbd_try_rs_begin_io(mdev, sector)) { | |
594 | mdev->bm_resync_fo = bit; | |
595 | goto requeue; | |
596 | } | |
597 | mdev->bm_resync_fo = bit + 1; | |
598 | ||
599 | if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) { | |
600 | drbd_rs_complete_io(mdev, sector); | |
601 | goto next_sector; | |
602 | } | |
603 | ||
604 | #if DRBD_MAX_SEGMENT_SIZE > BM_BLOCK_SIZE | |
605 | /* try to find some adjacent bits. | |
606 | * we stop if we have already the maximum req size. | |
607 | * | |
608 | * Additionally always align bigger requests, in order to | |
609 | * be prepared for all stripe sizes of software RAIDs. | |
b411b363 PR |
610 | */ |
611 | align = 1; | |
d207450c | 612 | rollback_i = i; |
b411b363 PR |
613 | for (;;) { |
614 | if (size + BM_BLOCK_SIZE > max_segment_size) | |
615 | break; | |
616 | ||
617 | /* Be always aligned */ | |
618 | if (sector & ((1<<(align+3))-1)) | |
619 | break; | |
620 | ||
621 | /* do not cross extent boundaries */ | |
622 | if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0) | |
623 | break; | |
624 | /* now, is it actually dirty, after all? | |
625 | * caution, drbd_bm_test_bit is tri-state for some | |
626 | * obscure reason; ( b == 0 ) would get the out-of-band | |
627 | * only accidentally right because of the "oddly sized" | |
628 | * adjustment below */ | |
629 | if (drbd_bm_test_bit(mdev, bit+1) != 1) | |
630 | break; | |
631 | bit++; | |
632 | size += BM_BLOCK_SIZE; | |
633 | if ((BM_BLOCK_SIZE << align) <= size) | |
634 | align++; | |
635 | i++; | |
636 | } | |
637 | /* if we merged some, | |
638 | * reset the offset to start the next drbd_bm_find_next from */ | |
639 | if (size > BM_BLOCK_SIZE) | |
640 | mdev->bm_resync_fo = bit + 1; | |
641 | #endif | |
642 | ||
643 | /* adjust very last sectors, in case we are oddly sized */ | |
644 | if (sector + (size>>9) > capacity) | |
645 | size = (capacity-sector)<<9; | |
646 | if (mdev->agreed_pro_version >= 89 && mdev->csums_tfm) { | |
647 | switch (read_for_csum(mdev, sector, size)) { | |
80a40e43 | 648 | case -EIO: /* Disk failure */ |
b411b363 PR |
649 | put_ldev(mdev); |
650 | return 0; | |
80a40e43 | 651 | case -EAGAIN: /* allocation failed, or ldev busy */ |
b411b363 PR |
652 | drbd_rs_complete_io(mdev, sector); |
653 | mdev->bm_resync_fo = BM_SECT_TO_BIT(sector); | |
d207450c | 654 | i = rollback_i; |
b411b363 | 655 | goto requeue; |
80a40e43 LE |
656 | case 0: |
657 | /* everything ok */ | |
658 | break; | |
659 | default: | |
660 | BUG(); | |
b411b363 PR |
661 | } |
662 | } else { | |
663 | inc_rs_pending(mdev); | |
664 | if (!drbd_send_drequest(mdev, P_RS_DATA_REQUEST, | |
665 | sector, size, ID_SYNCER)) { | |
666 | dev_err(DEV, "drbd_send_drequest() failed, aborting...\n"); | |
667 | dec_rs_pending(mdev); | |
668 | put_ldev(mdev); | |
669 | return 0; | |
670 | } | |
671 | } | |
672 | } | |
673 | ||
674 | if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) { | |
675 | /* last syncer _request_ was sent, | |
676 | * but the P_RS_DATA_REPLY not yet received. sync will end (and | |
677 | * next sync group will resume), as soon as we receive the last | |
678 | * resync data block, and the last bit is cleared. | |
679 | * until then resync "work" is "inactive" ... | |
680 | */ | |
681 | mdev->resync_work.cb = w_resync_inactive; | |
682 | put_ldev(mdev); | |
683 | return 1; | |
684 | } | |
685 | ||
686 | requeue: | |
778f271d | 687 | mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9)); |
b411b363 PR |
688 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); |
689 | put_ldev(mdev); | |
690 | return 1; | |
691 | } | |
692 | ||
693 | static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
694 | { | |
695 | int number, i, size; | |
696 | sector_t sector; | |
697 | const sector_t capacity = drbd_get_capacity(mdev->this_bdev); | |
698 | ||
699 | if (unlikely(cancel)) | |
700 | return 1; | |
701 | ||
702 | if (unlikely(mdev->state.conn < C_CONNECTED)) { | |
703 | dev_err(DEV, "Confused in w_make_ov_request()! cstate < Connected"); | |
704 | return 0; | |
705 | } | |
706 | ||
707 | number = SLEEP_TIME*mdev->sync_conf.rate / ((BM_BLOCK_SIZE/1024)*HZ); | |
708 | if (atomic_read(&mdev->rs_pending_cnt) > number) | |
709 | goto requeue; | |
710 | ||
711 | number -= atomic_read(&mdev->rs_pending_cnt); | |
712 | ||
713 | sector = mdev->ov_position; | |
714 | for (i = 0; i < number; i++) { | |
715 | if (sector >= capacity) { | |
716 | mdev->resync_work.cb = w_resync_inactive; | |
717 | return 1; | |
718 | } | |
719 | ||
720 | size = BM_BLOCK_SIZE; | |
721 | ||
722 | if (drbd_try_rs_begin_io(mdev, sector)) { | |
723 | mdev->ov_position = sector; | |
724 | goto requeue; | |
725 | } | |
726 | ||
727 | if (sector + (size>>9) > capacity) | |
728 | size = (capacity-sector)<<9; | |
729 | ||
730 | inc_rs_pending(mdev); | |
731 | if (!drbd_send_ov_request(mdev, sector, size)) { | |
732 | dec_rs_pending(mdev); | |
733 | return 0; | |
734 | } | |
735 | sector += BM_SECT_PER_BIT; | |
736 | } | |
737 | mdev->ov_position = sector; | |
738 | ||
739 | requeue: | |
740 | mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME); | |
741 | return 1; | |
742 | } | |
743 | ||
744 | ||
745 | int w_ov_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
746 | { | |
747 | kfree(w); | |
748 | ov_oos_print(mdev); | |
749 | drbd_resync_finished(mdev); | |
750 | ||
751 | return 1; | |
752 | } | |
753 | ||
754 | static int w_resync_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
755 | { | |
756 | kfree(w); | |
757 | ||
758 | drbd_resync_finished(mdev); | |
759 | ||
760 | return 1; | |
761 | } | |
762 | ||
af85e8e8 LE |
763 | static void ping_peer(struct drbd_conf *mdev) |
764 | { | |
765 | clear_bit(GOT_PING_ACK, &mdev->flags); | |
766 | request_ping(mdev); | |
767 | wait_event(mdev->misc_wait, | |
768 | test_bit(GOT_PING_ACK, &mdev->flags) || mdev->state.conn < C_CONNECTED); | |
769 | } | |
770 | ||
b411b363 PR |
771 | int drbd_resync_finished(struct drbd_conf *mdev) |
772 | { | |
773 | unsigned long db, dt, dbdt; | |
774 | unsigned long n_oos; | |
775 | union drbd_state os, ns; | |
776 | struct drbd_work *w; | |
777 | char *khelper_cmd = NULL; | |
778 | ||
779 | /* Remove all elements from the resync LRU. Since future actions | |
780 | * might set bits in the (main) bitmap, then the entries in the | |
781 | * resync LRU would be wrong. */ | |
782 | if (drbd_rs_del_all(mdev)) { | |
783 | /* In case this is not possible now, most probably because | |
784 | * there are P_RS_DATA_REPLY Packets lingering on the worker's | |
785 | * queue (or even the read operations for those packets | |
786 | * is not finished by now). Retry in 100ms. */ | |
787 | ||
788 | drbd_kick_lo(mdev); | |
789 | __set_current_state(TASK_INTERRUPTIBLE); | |
790 | schedule_timeout(HZ / 10); | |
791 | w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC); | |
792 | if (w) { | |
793 | w->cb = w_resync_finished; | |
794 | drbd_queue_work(&mdev->data.work, w); | |
795 | return 1; | |
796 | } | |
797 | dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n"); | |
798 | } | |
799 | ||
800 | dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ; | |
801 | if (dt <= 0) | |
802 | dt = 1; | |
803 | db = mdev->rs_total; | |
804 | dbdt = Bit2KB(db/dt); | |
805 | mdev->rs_paused /= HZ; | |
806 | ||
807 | if (!get_ldev(mdev)) | |
808 | goto out; | |
809 | ||
af85e8e8 LE |
810 | ping_peer(mdev); |
811 | ||
b411b363 PR |
812 | spin_lock_irq(&mdev->req_lock); |
813 | os = mdev->state; | |
814 | ||
815 | /* This protects us against multiple calls (that can happen in the presence | |
816 | of application IO), and against connectivity loss just before we arrive here. */ | |
817 | if (os.conn <= C_CONNECTED) | |
818 | goto out_unlock; | |
819 | ||
820 | ns = os; | |
821 | ns.conn = C_CONNECTED; | |
822 | ||
823 | dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n", | |
824 | (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) ? | |
825 | "Online verify " : "Resync", | |
826 | dt + mdev->rs_paused, mdev->rs_paused, dbdt); | |
827 | ||
828 | n_oos = drbd_bm_total_weight(mdev); | |
829 | ||
830 | if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) { | |
831 | if (n_oos) { | |
832 | dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n", | |
833 | n_oos, Bit2KB(1)); | |
834 | khelper_cmd = "out-of-sync"; | |
835 | } | |
836 | } else { | |
837 | D_ASSERT((n_oos - mdev->rs_failed) == 0); | |
838 | ||
839 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) | |
840 | khelper_cmd = "after-resync-target"; | |
841 | ||
842 | if (mdev->csums_tfm && mdev->rs_total) { | |
843 | const unsigned long s = mdev->rs_same_csum; | |
844 | const unsigned long t = mdev->rs_total; | |
845 | const int ratio = | |
846 | (t == 0) ? 0 : | |
847 | (t < 100000) ? ((s*100)/t) : (s/(t/100)); | |
848 | dev_info(DEV, "%u %% had equal check sums, eliminated: %luK; " | |
849 | "transferred %luK total %luK\n", | |
850 | ratio, | |
851 | Bit2KB(mdev->rs_same_csum), | |
852 | Bit2KB(mdev->rs_total - mdev->rs_same_csum), | |
853 | Bit2KB(mdev->rs_total)); | |
854 | } | |
855 | } | |
856 | ||
857 | if (mdev->rs_failed) { | |
858 | dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed); | |
859 | ||
860 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
861 | ns.disk = D_INCONSISTENT; | |
862 | ns.pdsk = D_UP_TO_DATE; | |
863 | } else { | |
864 | ns.disk = D_UP_TO_DATE; | |
865 | ns.pdsk = D_INCONSISTENT; | |
866 | } | |
867 | } else { | |
868 | ns.disk = D_UP_TO_DATE; | |
869 | ns.pdsk = D_UP_TO_DATE; | |
870 | ||
871 | if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) { | |
872 | if (mdev->p_uuid) { | |
873 | int i; | |
874 | for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++) | |
875 | _drbd_uuid_set(mdev, i, mdev->p_uuid[i]); | |
876 | drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]); | |
877 | _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]); | |
878 | } else { | |
879 | dev_err(DEV, "mdev->p_uuid is NULL! BUG\n"); | |
880 | } | |
881 | } | |
882 | ||
883 | drbd_uuid_set_bm(mdev, 0UL); | |
884 | ||
885 | if (mdev->p_uuid) { | |
886 | /* Now the two UUID sets are equal, update what we | |
887 | * know of the peer. */ | |
888 | int i; | |
889 | for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++) | |
890 | mdev->p_uuid[i] = mdev->ldev->md.uuid[i]; | |
891 | } | |
892 | } | |
893 | ||
894 | _drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
895 | out_unlock: | |
896 | spin_unlock_irq(&mdev->req_lock); | |
897 | put_ldev(mdev); | |
898 | out: | |
899 | mdev->rs_total = 0; | |
900 | mdev->rs_failed = 0; | |
901 | mdev->rs_paused = 0; | |
902 | mdev->ov_start_sector = 0; | |
903 | ||
13d42685 LE |
904 | drbd_md_sync(mdev); |
905 | ||
b411b363 | 906 | if (test_and_clear_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags)) { |
bc571b8c | 907 | dev_info(DEV, "Writing the whole bitmap\n"); |
b411b363 PR |
908 | drbd_queue_bitmap_io(mdev, &drbd_bm_write, NULL, "write from resync_finished"); |
909 | } | |
910 | ||
911 | if (khelper_cmd) | |
912 | drbd_khelper(mdev, khelper_cmd); | |
913 | ||
914 | return 1; | |
915 | } | |
916 | ||
917 | /* helper */ | |
918 | static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_epoch_entry *e) | |
919 | { | |
45bb912b | 920 | if (drbd_ee_has_active_page(e)) { |
b411b363 | 921 | /* This might happen if sendpage() has not finished */ |
78db8928 | 922 | int i = (e->size + PAGE_SIZE -1) >> PAGE_SHIFT; |
435f0740 LE |
923 | atomic_add(i, &mdev->pp_in_use_by_net); |
924 | atomic_sub(i, &mdev->pp_in_use); | |
b411b363 PR |
925 | spin_lock_irq(&mdev->req_lock); |
926 | list_add_tail(&e->w.list, &mdev->net_ee); | |
927 | spin_unlock_irq(&mdev->req_lock); | |
435f0740 | 928 | wake_up(&drbd_pp_wait); |
b411b363 PR |
929 | } else |
930 | drbd_free_ee(mdev, e); | |
931 | } | |
932 | ||
933 | /** | |
934 | * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST | |
935 | * @mdev: DRBD device. | |
936 | * @w: work object. | |
937 | * @cancel: The connection will be closed anyways | |
938 | */ | |
939 | int w_e_end_data_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
940 | { | |
941 | struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); | |
942 | int ok; | |
943 | ||
944 | if (unlikely(cancel)) { | |
945 | drbd_free_ee(mdev, e); | |
946 | dec_unacked(mdev); | |
947 | return 1; | |
948 | } | |
949 | ||
45bb912b | 950 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
951 | ok = drbd_send_block(mdev, P_DATA_REPLY, e); |
952 | } else { | |
953 | if (__ratelimit(&drbd_ratelimit_state)) | |
954 | dev_err(DEV, "Sending NegDReply. sector=%llus.\n", | |
955 | (unsigned long long)e->sector); | |
956 | ||
957 | ok = drbd_send_ack(mdev, P_NEG_DREPLY, e); | |
958 | } | |
959 | ||
960 | dec_unacked(mdev); | |
961 | ||
962 | move_to_net_ee_or_free(mdev, e); | |
963 | ||
964 | if (unlikely(!ok)) | |
965 | dev_err(DEV, "drbd_send_block() failed\n"); | |
966 | return ok; | |
967 | } | |
968 | ||
969 | /** | |
970 | * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS | |
971 | * @mdev: DRBD device. | |
972 | * @w: work object. | |
973 | * @cancel: The connection will be closed anyways | |
974 | */ | |
975 | int w_e_end_rsdata_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
976 | { | |
977 | struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); | |
978 | int ok; | |
979 | ||
980 | if (unlikely(cancel)) { | |
981 | drbd_free_ee(mdev, e); | |
982 | dec_unacked(mdev); | |
983 | return 1; | |
984 | } | |
985 | ||
986 | if (get_ldev_if_state(mdev, D_FAILED)) { | |
987 | drbd_rs_complete_io(mdev, e->sector); | |
988 | put_ldev(mdev); | |
989 | } | |
990 | ||
45bb912b | 991 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
992 | if (likely(mdev->state.pdsk >= D_INCONSISTENT)) { |
993 | inc_rs_pending(mdev); | |
994 | ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e); | |
995 | } else { | |
996 | if (__ratelimit(&drbd_ratelimit_state)) | |
997 | dev_err(DEV, "Not sending RSDataReply, " | |
998 | "partner DISKLESS!\n"); | |
999 | ok = 1; | |
1000 | } | |
1001 | } else { | |
1002 | if (__ratelimit(&drbd_ratelimit_state)) | |
1003 | dev_err(DEV, "Sending NegRSDReply. sector %llus.\n", | |
1004 | (unsigned long long)e->sector); | |
1005 | ||
1006 | ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); | |
1007 | ||
1008 | /* update resync data with failure */ | |
1009 | drbd_rs_failed_io(mdev, e->sector, e->size); | |
1010 | } | |
1011 | ||
1012 | dec_unacked(mdev); | |
1013 | ||
1014 | move_to_net_ee_or_free(mdev, e); | |
1015 | ||
1016 | if (unlikely(!ok)) | |
1017 | dev_err(DEV, "drbd_send_block() failed\n"); | |
1018 | return ok; | |
1019 | } | |
1020 | ||
1021 | int w_e_end_csum_rs_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1022 | { | |
1023 | struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); | |
1024 | struct digest_info *di; | |
1025 | int digest_size; | |
1026 | void *digest = NULL; | |
1027 | int ok, eq = 0; | |
1028 | ||
1029 | if (unlikely(cancel)) { | |
1030 | drbd_free_ee(mdev, e); | |
1031 | dec_unacked(mdev); | |
1032 | return 1; | |
1033 | } | |
1034 | ||
1d53f09e LE |
1035 | if (get_ldev(mdev)) { |
1036 | drbd_rs_complete_io(mdev, e->sector); | |
1037 | put_ldev(mdev); | |
1038 | } | |
b411b363 | 1039 | |
85719573 | 1040 | di = e->digest; |
b411b363 | 1041 | |
45bb912b | 1042 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1043 | /* quick hack to try to avoid a race against reconfiguration. |
1044 | * a real fix would be much more involved, | |
1045 | * introducing more locking mechanisms */ | |
1046 | if (mdev->csums_tfm) { | |
1047 | digest_size = crypto_hash_digestsize(mdev->csums_tfm); | |
1048 | D_ASSERT(digest_size == di->digest_size); | |
1049 | digest = kmalloc(digest_size, GFP_NOIO); | |
1050 | } | |
1051 | if (digest) { | |
45bb912b | 1052 | drbd_csum_ee(mdev, mdev->csums_tfm, e, digest); |
b411b363 PR |
1053 | eq = !memcmp(digest, di->digest, digest_size); |
1054 | kfree(digest); | |
1055 | } | |
1056 | ||
1057 | if (eq) { | |
1058 | drbd_set_in_sync(mdev, e->sector, e->size); | |
676396d5 LE |
1059 | /* rs_same_csums unit is BM_BLOCK_SIZE */ |
1060 | mdev->rs_same_csum += e->size >> BM_BLOCK_SHIFT; | |
b411b363 PR |
1061 | ok = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, e); |
1062 | } else { | |
1063 | inc_rs_pending(mdev); | |
204bba99 PR |
1064 | e->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */ |
1065 | e->flags &= ~EE_HAS_DIGEST; /* This e no longer has a digest pointer */ | |
1066 | kfree(di); | |
b411b363 PR |
1067 | ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e); |
1068 | } | |
1069 | } else { | |
1070 | ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); | |
1071 | if (__ratelimit(&drbd_ratelimit_state)) | |
1072 | dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); | |
1073 | } | |
1074 | ||
1075 | dec_unacked(mdev); | |
b411b363 PR |
1076 | move_to_net_ee_or_free(mdev, e); |
1077 | ||
1078 | if (unlikely(!ok)) | |
1079 | dev_err(DEV, "drbd_send_block/ack() failed\n"); | |
1080 | return ok; | |
1081 | } | |
1082 | ||
1083 | int w_e_end_ov_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1084 | { | |
1085 | struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); | |
1086 | int digest_size; | |
1087 | void *digest; | |
1088 | int ok = 1; | |
1089 | ||
1090 | if (unlikely(cancel)) | |
1091 | goto out; | |
1092 | ||
45bb912b | 1093 | if (unlikely((e->flags & EE_WAS_ERROR) != 0)) |
b411b363 PR |
1094 | goto out; |
1095 | ||
1096 | digest_size = crypto_hash_digestsize(mdev->verify_tfm); | |
1097 | /* FIXME if this allocation fails, online verify will not terminate! */ | |
1098 | digest = kmalloc(digest_size, GFP_NOIO); | |
1099 | if (digest) { | |
45bb912b | 1100 | drbd_csum_ee(mdev, mdev->verify_tfm, e, digest); |
b411b363 PR |
1101 | inc_rs_pending(mdev); |
1102 | ok = drbd_send_drequest_csum(mdev, e->sector, e->size, | |
1103 | digest, digest_size, P_OV_REPLY); | |
1104 | if (!ok) | |
1105 | dec_rs_pending(mdev); | |
1106 | kfree(digest); | |
1107 | } | |
1108 | ||
1109 | out: | |
1110 | drbd_free_ee(mdev, e); | |
1111 | ||
1112 | dec_unacked(mdev); | |
1113 | ||
1114 | return ok; | |
1115 | } | |
1116 | ||
1117 | void drbd_ov_oos_found(struct drbd_conf *mdev, sector_t sector, int size) | |
1118 | { | |
1119 | if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) { | |
1120 | mdev->ov_last_oos_size += size>>9; | |
1121 | } else { | |
1122 | mdev->ov_last_oos_start = sector; | |
1123 | mdev->ov_last_oos_size = size>>9; | |
1124 | } | |
1125 | drbd_set_out_of_sync(mdev, sector, size); | |
1126 | set_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags); | |
1127 | } | |
1128 | ||
1129 | int w_e_end_ov_reply(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1130 | { | |
1131 | struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w); | |
1132 | struct digest_info *di; | |
1133 | int digest_size; | |
1134 | void *digest; | |
1135 | int ok, eq = 0; | |
1136 | ||
1137 | if (unlikely(cancel)) { | |
1138 | drbd_free_ee(mdev, e); | |
1139 | dec_unacked(mdev); | |
1140 | return 1; | |
1141 | } | |
1142 | ||
1143 | /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all | |
1144 | * the resync lru has been cleaned up already */ | |
1d53f09e LE |
1145 | if (get_ldev(mdev)) { |
1146 | drbd_rs_complete_io(mdev, e->sector); | |
1147 | put_ldev(mdev); | |
1148 | } | |
b411b363 | 1149 | |
85719573 | 1150 | di = e->digest; |
b411b363 | 1151 | |
45bb912b | 1152 | if (likely((e->flags & EE_WAS_ERROR) == 0)) { |
b411b363 PR |
1153 | digest_size = crypto_hash_digestsize(mdev->verify_tfm); |
1154 | digest = kmalloc(digest_size, GFP_NOIO); | |
1155 | if (digest) { | |
45bb912b | 1156 | drbd_csum_ee(mdev, mdev->verify_tfm, e, digest); |
b411b363 PR |
1157 | |
1158 | D_ASSERT(digest_size == di->digest_size); | |
1159 | eq = !memcmp(digest, di->digest, digest_size); | |
1160 | kfree(digest); | |
1161 | } | |
1162 | } else { | |
1163 | ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e); | |
1164 | if (__ratelimit(&drbd_ratelimit_state)) | |
1165 | dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n"); | |
1166 | } | |
1167 | ||
1168 | dec_unacked(mdev); | |
b411b363 PR |
1169 | if (!eq) |
1170 | drbd_ov_oos_found(mdev, e->sector, e->size); | |
1171 | else | |
1172 | ov_oos_print(mdev); | |
1173 | ||
1174 | ok = drbd_send_ack_ex(mdev, P_OV_RESULT, e->sector, e->size, | |
1175 | eq ? ID_IN_SYNC : ID_OUT_OF_SYNC); | |
1176 | ||
1177 | drbd_free_ee(mdev, e); | |
1178 | ||
1179 | if (--mdev->ov_left == 0) { | |
1180 | ov_oos_print(mdev); | |
1181 | drbd_resync_finished(mdev); | |
1182 | } | |
1183 | ||
1184 | return ok; | |
1185 | } | |
1186 | ||
1187 | int w_prev_work_done(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1188 | { | |
1189 | struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w); | |
1190 | complete(&b->done); | |
1191 | return 1; | |
1192 | } | |
1193 | ||
1194 | int w_send_barrier(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1195 | { | |
1196 | struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w); | |
1197 | struct p_barrier *p = &mdev->data.sbuf.barrier; | |
1198 | int ok = 1; | |
1199 | ||
1200 | /* really avoid racing with tl_clear. w.cb may have been referenced | |
1201 | * just before it was reassigned and re-queued, so double check that. | |
1202 | * actually, this race was harmless, since we only try to send the | |
1203 | * barrier packet here, and otherwise do nothing with the object. | |
1204 | * but compare with the head of w_clear_epoch */ | |
1205 | spin_lock_irq(&mdev->req_lock); | |
1206 | if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED) | |
1207 | cancel = 1; | |
1208 | spin_unlock_irq(&mdev->req_lock); | |
1209 | if (cancel) | |
1210 | return 1; | |
1211 | ||
1212 | if (!drbd_get_data_sock(mdev)) | |
1213 | return 0; | |
1214 | p->barrier = b->br_number; | |
1215 | /* inc_ap_pending was done where this was queued. | |
1216 | * dec_ap_pending will be done in got_BarrierAck | |
1217 | * or (on connection loss) in w_clear_epoch. */ | |
1218 | ok = _drbd_send_cmd(mdev, mdev->data.socket, P_BARRIER, | |
0b70a13d | 1219 | (struct p_header80 *)p, sizeof(*p), 0); |
b411b363 PR |
1220 | drbd_put_data_sock(mdev); |
1221 | ||
1222 | return ok; | |
1223 | } | |
1224 | ||
1225 | int w_send_write_hint(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1226 | { | |
1227 | if (cancel) | |
1228 | return 1; | |
1229 | return drbd_send_short_cmd(mdev, P_UNPLUG_REMOTE); | |
1230 | } | |
1231 | ||
1232 | /** | |
1233 | * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request | |
1234 | * @mdev: DRBD device. | |
1235 | * @w: work object. | |
1236 | * @cancel: The connection will be closed anyways | |
1237 | */ | |
1238 | int w_send_dblock(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1239 | { | |
1240 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
1241 | int ok; | |
1242 | ||
1243 | if (unlikely(cancel)) { | |
1244 | req_mod(req, send_canceled); | |
1245 | return 1; | |
1246 | } | |
1247 | ||
1248 | ok = drbd_send_dblock(mdev, req); | |
1249 | req_mod(req, ok ? handed_over_to_network : send_failed); | |
1250 | ||
1251 | return ok; | |
1252 | } | |
1253 | ||
1254 | /** | |
1255 | * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet | |
1256 | * @mdev: DRBD device. | |
1257 | * @w: work object. | |
1258 | * @cancel: The connection will be closed anyways | |
1259 | */ | |
1260 | int w_send_read_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel) | |
1261 | { | |
1262 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
1263 | int ok; | |
1264 | ||
1265 | if (unlikely(cancel)) { | |
1266 | req_mod(req, send_canceled); | |
1267 | return 1; | |
1268 | } | |
1269 | ||
1270 | ok = drbd_send_drequest(mdev, P_DATA_REQUEST, req->sector, req->size, | |
1271 | (unsigned long)req); | |
1272 | ||
1273 | if (!ok) { | |
1274 | /* ?? we set C_TIMEOUT or C_BROKEN_PIPE in drbd_send(); | |
1275 | * so this is probably redundant */ | |
1276 | if (mdev->state.conn >= C_CONNECTED) | |
1277 | drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE)); | |
1278 | } | |
1279 | req_mod(req, ok ? handed_over_to_network : send_failed); | |
1280 | ||
1281 | return ok; | |
1282 | } | |
1283 | ||
265be2d0 PR |
1284 | int w_restart_disk_io(struct drbd_conf *mdev, struct drbd_work *w, int cancel) |
1285 | { | |
1286 | struct drbd_request *req = container_of(w, struct drbd_request, w); | |
1287 | ||
0778286a | 1288 | if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG) |
265be2d0 PR |
1289 | drbd_al_begin_io(mdev, req->sector); |
1290 | /* Calling drbd_al_begin_io() out of the worker might deadlocks | |
1291 | theoretically. Practically it can not deadlock, since this is | |
1292 | only used when unfreezing IOs. All the extents of the requests | |
1293 | that made it into the TL are already active */ | |
1294 | ||
1295 | drbd_req_make_private_bio(req, req->master_bio); | |
1296 | req->private_bio->bi_bdev = mdev->ldev->backing_bdev; | |
1297 | generic_make_request(req->private_bio); | |
1298 | ||
1299 | return 1; | |
1300 | } | |
1301 | ||
b411b363 PR |
1302 | static int _drbd_may_sync_now(struct drbd_conf *mdev) |
1303 | { | |
1304 | struct drbd_conf *odev = mdev; | |
1305 | ||
1306 | while (1) { | |
1307 | if (odev->sync_conf.after == -1) | |
1308 | return 1; | |
1309 | odev = minor_to_mdev(odev->sync_conf.after); | |
1310 | ERR_IF(!odev) return 1; | |
1311 | if ((odev->state.conn >= C_SYNC_SOURCE && | |
1312 | odev->state.conn <= C_PAUSED_SYNC_T) || | |
1313 | odev->state.aftr_isp || odev->state.peer_isp || | |
1314 | odev->state.user_isp) | |
1315 | return 0; | |
1316 | } | |
1317 | } | |
1318 | ||
1319 | /** | |
1320 | * _drbd_pause_after() - Pause resync on all devices that may not resync now | |
1321 | * @mdev: DRBD device. | |
1322 | * | |
1323 | * Called from process context only (admin command and after_state_ch). | |
1324 | */ | |
1325 | static int _drbd_pause_after(struct drbd_conf *mdev) | |
1326 | { | |
1327 | struct drbd_conf *odev; | |
1328 | int i, rv = 0; | |
1329 | ||
1330 | for (i = 0; i < minor_count; i++) { | |
1331 | odev = minor_to_mdev(i); | |
1332 | if (!odev) | |
1333 | continue; | |
1334 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) | |
1335 | continue; | |
1336 | if (!_drbd_may_sync_now(odev)) | |
1337 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 1), CS_HARD, NULL) | |
1338 | != SS_NOTHING_TO_DO); | |
1339 | } | |
1340 | ||
1341 | return rv; | |
1342 | } | |
1343 | ||
1344 | /** | |
1345 | * _drbd_resume_next() - Resume resync on all devices that may resync now | |
1346 | * @mdev: DRBD device. | |
1347 | * | |
1348 | * Called from process context only (admin command and worker). | |
1349 | */ | |
1350 | static int _drbd_resume_next(struct drbd_conf *mdev) | |
1351 | { | |
1352 | struct drbd_conf *odev; | |
1353 | int i, rv = 0; | |
1354 | ||
1355 | for (i = 0; i < minor_count; i++) { | |
1356 | odev = minor_to_mdev(i); | |
1357 | if (!odev) | |
1358 | continue; | |
1359 | if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS) | |
1360 | continue; | |
1361 | if (odev->state.aftr_isp) { | |
1362 | if (_drbd_may_sync_now(odev)) | |
1363 | rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0), | |
1364 | CS_HARD, NULL) | |
1365 | != SS_NOTHING_TO_DO) ; | |
1366 | } | |
1367 | } | |
1368 | return rv; | |
1369 | } | |
1370 | ||
1371 | void resume_next_sg(struct drbd_conf *mdev) | |
1372 | { | |
1373 | write_lock_irq(&global_state_lock); | |
1374 | _drbd_resume_next(mdev); | |
1375 | write_unlock_irq(&global_state_lock); | |
1376 | } | |
1377 | ||
1378 | void suspend_other_sg(struct drbd_conf *mdev) | |
1379 | { | |
1380 | write_lock_irq(&global_state_lock); | |
1381 | _drbd_pause_after(mdev); | |
1382 | write_unlock_irq(&global_state_lock); | |
1383 | } | |
1384 | ||
1385 | static int sync_after_error(struct drbd_conf *mdev, int o_minor) | |
1386 | { | |
1387 | struct drbd_conf *odev; | |
1388 | ||
1389 | if (o_minor == -1) | |
1390 | return NO_ERROR; | |
1391 | if (o_minor < -1 || minor_to_mdev(o_minor) == NULL) | |
1392 | return ERR_SYNC_AFTER; | |
1393 | ||
1394 | /* check for loops */ | |
1395 | odev = minor_to_mdev(o_minor); | |
1396 | while (1) { | |
1397 | if (odev == mdev) | |
1398 | return ERR_SYNC_AFTER_CYCLE; | |
1399 | ||
1400 | /* dependency chain ends here, no cycles. */ | |
1401 | if (odev->sync_conf.after == -1) | |
1402 | return NO_ERROR; | |
1403 | ||
1404 | /* follow the dependency chain */ | |
1405 | odev = minor_to_mdev(odev->sync_conf.after); | |
1406 | } | |
1407 | } | |
1408 | ||
1409 | int drbd_alter_sa(struct drbd_conf *mdev, int na) | |
1410 | { | |
1411 | int changes; | |
1412 | int retcode; | |
1413 | ||
1414 | write_lock_irq(&global_state_lock); | |
1415 | retcode = sync_after_error(mdev, na); | |
1416 | if (retcode == NO_ERROR) { | |
1417 | mdev->sync_conf.after = na; | |
1418 | do { | |
1419 | changes = _drbd_pause_after(mdev); | |
1420 | changes |= _drbd_resume_next(mdev); | |
1421 | } while (changes); | |
1422 | } | |
1423 | write_unlock_irq(&global_state_lock); | |
1424 | return retcode; | |
1425 | } | |
1426 | ||
1427 | /** | |
1428 | * drbd_start_resync() - Start the resync process | |
1429 | * @mdev: DRBD device. | |
1430 | * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET | |
1431 | * | |
1432 | * This function might bring you directly into one of the | |
1433 | * C_PAUSED_SYNC_* states. | |
1434 | */ | |
1435 | void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side) | |
1436 | { | |
1437 | union drbd_state ns; | |
1438 | int r; | |
1439 | ||
1440 | if (mdev->state.conn >= C_SYNC_SOURCE) { | |
1441 | dev_err(DEV, "Resync already running!\n"); | |
1442 | return; | |
1443 | } | |
1444 | ||
b411b363 PR |
1445 | /* In case a previous resync run was aborted by an IO error/detach on the peer. */ |
1446 | drbd_rs_cancel_all(mdev); | |
1447 | ||
1448 | if (side == C_SYNC_TARGET) { | |
1449 | /* Since application IO was locked out during C_WF_BITMAP_T and | |
1450 | C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET | |
1451 | we check that we might make the data inconsistent. */ | |
1452 | r = drbd_khelper(mdev, "before-resync-target"); | |
1453 | r = (r >> 8) & 0xff; | |
1454 | if (r > 0) { | |
1455 | dev_info(DEV, "before-resync-target handler returned %d, " | |
1456 | "dropping connection.\n", r); | |
1457 | drbd_force_state(mdev, NS(conn, C_DISCONNECTING)); | |
1458 | return; | |
1459 | } | |
1460 | } | |
1461 | ||
1462 | drbd_state_lock(mdev); | |
1463 | ||
1464 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) { | |
1465 | drbd_state_unlock(mdev); | |
1466 | return; | |
1467 | } | |
1468 | ||
1469 | if (side == C_SYNC_TARGET) { | |
1470 | mdev->bm_resync_fo = 0; | |
1471 | } else /* side == C_SYNC_SOURCE */ { | |
1472 | u64 uuid; | |
1473 | ||
1474 | get_random_bytes(&uuid, sizeof(u64)); | |
1475 | drbd_uuid_set(mdev, UI_BITMAP, uuid); | |
1476 | drbd_send_sync_uuid(mdev, uuid); | |
1477 | ||
1478 | D_ASSERT(mdev->state.disk == D_UP_TO_DATE); | |
1479 | } | |
1480 | ||
1481 | write_lock_irq(&global_state_lock); | |
1482 | ns = mdev->state; | |
1483 | ||
1484 | ns.aftr_isp = !_drbd_may_sync_now(mdev); | |
1485 | ||
1486 | ns.conn = side; | |
1487 | ||
1488 | if (side == C_SYNC_TARGET) | |
1489 | ns.disk = D_INCONSISTENT; | |
1490 | else /* side == C_SYNC_SOURCE */ | |
1491 | ns.pdsk = D_INCONSISTENT; | |
1492 | ||
1493 | r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL); | |
1494 | ns = mdev->state; | |
1495 | ||
1496 | if (ns.conn < C_CONNECTED) | |
1497 | r = SS_UNKNOWN_ERROR; | |
1498 | ||
1499 | if (r == SS_SUCCESS) { | |
1d7734a0 LE |
1500 | unsigned long tw = drbd_bm_total_weight(mdev); |
1501 | unsigned long now = jiffies; | |
1502 | int i; | |
1503 | ||
b411b363 PR |
1504 | mdev->rs_failed = 0; |
1505 | mdev->rs_paused = 0; | |
b411b363 | 1506 | mdev->rs_same_csum = 0; |
0f0601f4 LE |
1507 | mdev->rs_last_events = 0; |
1508 | mdev->rs_last_sect_ev = 0; | |
1d7734a0 LE |
1509 | mdev->rs_total = tw; |
1510 | mdev->rs_start = now; | |
1511 | for (i = 0; i < DRBD_SYNC_MARKS; i++) { | |
1512 | mdev->rs_mark_left[i] = tw; | |
1513 | mdev->rs_mark_time[i] = now; | |
1514 | } | |
b411b363 PR |
1515 | _drbd_pause_after(mdev); |
1516 | } | |
1517 | write_unlock_irq(&global_state_lock); | |
b411b363 PR |
1518 | put_ldev(mdev); |
1519 | ||
1520 | if (r == SS_SUCCESS) { | |
1521 | dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n", | |
1522 | drbd_conn_str(ns.conn), | |
1523 | (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10), | |
1524 | (unsigned long) mdev->rs_total); | |
1525 | ||
af85e8e8 LE |
1526 | if (mdev->agreed_pro_version < 95 && mdev->rs_total == 0) { |
1527 | /* This still has a race (about when exactly the peers | |
1528 | * detect connection loss) that can lead to a full sync | |
1529 | * on next handshake. In 8.3.9 we fixed this with explicit | |
1530 | * resync-finished notifications, but the fix | |
1531 | * introduces a protocol change. Sleeping for some | |
1532 | * time longer than the ping interval + timeout on the | |
1533 | * SyncSource, to give the SyncTarget the chance to | |
1534 | * detect connection loss, then waiting for a ping | |
1535 | * response (implicit in drbd_resync_finished) reduces | |
1536 | * the race considerably, but does not solve it. */ | |
1537 | if (side == C_SYNC_SOURCE) | |
1538 | schedule_timeout_interruptible( | |
1539 | mdev->net_conf->ping_int * HZ + | |
1540 | mdev->net_conf->ping_timeo*HZ/9); | |
b411b363 | 1541 | drbd_resync_finished(mdev); |
b411b363 PR |
1542 | } |
1543 | ||
778f271d | 1544 | atomic_set(&mdev->rs_sect_in, 0); |
0f0601f4 | 1545 | atomic_set(&mdev->rs_sect_ev, 0); |
778f271d PR |
1546 | mdev->rs_in_flight = 0; |
1547 | mdev->rs_planed = 0; | |
1548 | spin_lock(&mdev->peer_seq_lock); | |
1549 | fifo_set(&mdev->rs_plan_s, 0); | |
1550 | spin_unlock(&mdev->peer_seq_lock); | |
b411b363 PR |
1551 | /* ns.conn may already be != mdev->state.conn, |
1552 | * we may have been paused in between, or become paused until | |
1553 | * the timer triggers. | |
1554 | * No matter, that is handled in resync_timer_fn() */ | |
1555 | if (ns.conn == C_SYNC_TARGET) | |
1556 | mod_timer(&mdev->resync_timer, jiffies); | |
1557 | ||
1558 | drbd_md_sync(mdev); | |
1559 | } | |
d0c3f60f | 1560 | drbd_state_unlock(mdev); |
b411b363 PR |
1561 | } |
1562 | ||
1563 | int drbd_worker(struct drbd_thread *thi) | |
1564 | { | |
1565 | struct drbd_conf *mdev = thi->mdev; | |
1566 | struct drbd_work *w = NULL; | |
1567 | LIST_HEAD(work_list); | |
1568 | int intr = 0, i; | |
1569 | ||
1570 | sprintf(current->comm, "drbd%d_worker", mdev_to_minor(mdev)); | |
1571 | ||
1572 | while (get_t_state(thi) == Running) { | |
1573 | drbd_thread_current_set_cpu(mdev); | |
1574 | ||
1575 | if (down_trylock(&mdev->data.work.s)) { | |
1576 | mutex_lock(&mdev->data.mutex); | |
1577 | if (mdev->data.socket && !mdev->net_conf->no_cork) | |
1578 | drbd_tcp_uncork(mdev->data.socket); | |
1579 | mutex_unlock(&mdev->data.mutex); | |
1580 | ||
1581 | intr = down_interruptible(&mdev->data.work.s); | |
1582 | ||
1583 | mutex_lock(&mdev->data.mutex); | |
1584 | if (mdev->data.socket && !mdev->net_conf->no_cork) | |
1585 | drbd_tcp_cork(mdev->data.socket); | |
1586 | mutex_unlock(&mdev->data.mutex); | |
1587 | } | |
1588 | ||
1589 | if (intr) { | |
1590 | D_ASSERT(intr == -EINTR); | |
1591 | flush_signals(current); | |
1592 | ERR_IF (get_t_state(thi) == Running) | |
1593 | continue; | |
1594 | break; | |
1595 | } | |
1596 | ||
1597 | if (get_t_state(thi) != Running) | |
1598 | break; | |
1599 | /* With this break, we have done a down() but not consumed | |
1600 | the entry from the list. The cleanup code takes care of | |
1601 | this... */ | |
1602 | ||
1603 | w = NULL; | |
1604 | spin_lock_irq(&mdev->data.work.q_lock); | |
1605 | ERR_IF(list_empty(&mdev->data.work.q)) { | |
1606 | /* something terribly wrong in our logic. | |
1607 | * we were able to down() the semaphore, | |
1608 | * but the list is empty... doh. | |
1609 | * | |
1610 | * what is the best thing to do now? | |
1611 | * try again from scratch, restarting the receiver, | |
1612 | * asender, whatnot? could break even more ugly, | |
1613 | * e.g. when we are primary, but no good local data. | |
1614 | * | |
1615 | * I'll try to get away just starting over this loop. | |
1616 | */ | |
1617 | spin_unlock_irq(&mdev->data.work.q_lock); | |
1618 | continue; | |
1619 | } | |
1620 | w = list_entry(mdev->data.work.q.next, struct drbd_work, list); | |
1621 | list_del_init(&w->list); | |
1622 | spin_unlock_irq(&mdev->data.work.q_lock); | |
1623 | ||
1624 | if (!w->cb(mdev, w, mdev->state.conn < C_CONNECTED)) { | |
1625 | /* dev_warn(DEV, "worker: a callback failed! \n"); */ | |
1626 | if (mdev->state.conn >= C_CONNECTED) | |
1627 | drbd_force_state(mdev, | |
1628 | NS(conn, C_NETWORK_FAILURE)); | |
1629 | } | |
1630 | } | |
1631 | D_ASSERT(test_bit(DEVICE_DYING, &mdev->flags)); | |
1632 | D_ASSERT(test_bit(CONFIG_PENDING, &mdev->flags)); | |
1633 | ||
1634 | spin_lock_irq(&mdev->data.work.q_lock); | |
1635 | i = 0; | |
1636 | while (!list_empty(&mdev->data.work.q)) { | |
1637 | list_splice_init(&mdev->data.work.q, &work_list); | |
1638 | spin_unlock_irq(&mdev->data.work.q_lock); | |
1639 | ||
1640 | while (!list_empty(&work_list)) { | |
1641 | w = list_entry(work_list.next, struct drbd_work, list); | |
1642 | list_del_init(&w->list); | |
1643 | w->cb(mdev, w, 1); | |
1644 | i++; /* dead debugging code */ | |
1645 | } | |
1646 | ||
1647 | spin_lock_irq(&mdev->data.work.q_lock); | |
1648 | } | |
1649 | sema_init(&mdev->data.work.s, 0); | |
1650 | /* DANGEROUS race: if someone did queue his work within the spinlock, | |
1651 | * but up() ed outside the spinlock, we could get an up() on the | |
1652 | * semaphore without corresponding list entry. | |
1653 | * So don't do that. | |
1654 | */ | |
1655 | spin_unlock_irq(&mdev->data.work.q_lock); | |
1656 | ||
1657 | D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE); | |
1658 | /* _drbd_set_state only uses stop_nowait. | |
1659 | * wait here for the Exiting receiver. */ | |
1660 | drbd_thread_stop(&mdev->receiver); | |
1661 | drbd_mdev_cleanup(mdev); | |
1662 | ||
1663 | dev_info(DEV, "worker terminated\n"); | |
1664 | ||
1665 | clear_bit(DEVICE_DYING, &mdev->flags); | |
1666 | clear_bit(CONFIG_PENDING, &mdev->flags); | |
1667 | wake_up(&mdev->state_wait); | |
1668 | ||
1669 | return 0; | |
1670 | } |