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
26 #include <linux/module.h>
27 #include <linux/drbd.h>
28 #include <linux/sched.h>
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>
35 #include <linux/string.h>
36 #include <linux/scatterlist.h>
37
38 #include "drbd_int.h"
39 #include "drbd_req.h"
40
41 static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel);
42 static int w_make_resync_request(struct drbd_conf *mdev,
43 struct drbd_work *w, int cancel);
44
45
46
47 /* endio handlers:
48 * drbd_md_io_complete (defined here)
49 * drbd_endio_pri (defined here)
50 * drbd_endio_sec (defined here)
51 * bm_async_io_complete (defined in drbd_bitmap.c)
52 *
53 * For all these callbacks, note the following:
54 * The callbacks will be called in irq context by the IDE drivers,
55 * and in Softirqs/Tasklets/BH context by the SCSI drivers.
56 * Try to get the locking right :)
57 *
58 */
59
60
61 /* About the global_state_lock
62 Each state transition on an device holds a read lock. In case we have
63 to evaluate the sync after dependencies, we grab a write lock, because
64 we need stable states on all devices for that. */
65 rwlock_t global_state_lock;
66
67 /* used for synchronous meta data and bitmap IO
68 * submitted by drbd_md_sync_page_io()
69 */
drbd_md_io_complete(struct bio * bio,int error)70 void drbd_md_io_complete(struct bio *bio, int error)
71 {
72 struct drbd_md_io *md_io;
73
74 md_io = (struct drbd_md_io *)bio->bi_private;
75 md_io->error = error;
76
77 complete(&md_io->event);
78 }
79
80 /* reads on behalf of the partner,
81 * "submitted" by the receiver
82 */
drbd_endio_read_sec_final(struct drbd_epoch_entry * e)83 void drbd_endio_read_sec_final(struct drbd_epoch_entry *e) __releases(local)
84 {
85 unsigned long flags = 0;
86 struct drbd_conf *mdev = e->mdev;
87
88 D_ASSERT(e->block_id != ID_VACANT);
89
90 spin_lock_irqsave(&mdev->req_lock, flags);
91 mdev->read_cnt += e->size >> 9;
92 list_del(&e->w.list);
93 if (list_empty(&mdev->read_ee))
94 wake_up(&mdev->ee_wait);
95 if (test_bit(__EE_WAS_ERROR, &e->flags))
96 __drbd_chk_io_error(mdev, false);
97 spin_unlock_irqrestore(&mdev->req_lock, flags);
98
99 drbd_queue_work(&mdev->data.work, &e->w);
100 put_ldev(mdev);
101 }
102
103 /* writes on behalf of the partner, or resync writes,
104 * "submitted" by the receiver, final stage. */
drbd_endio_write_sec_final(struct drbd_epoch_entry * e)105 static void drbd_endio_write_sec_final(struct drbd_epoch_entry *e) __releases(local)
106 {
107 unsigned long flags = 0;
108 struct drbd_conf *mdev = e->mdev;
109 sector_t e_sector;
110 int do_wake;
111 int is_syncer_req;
112 int do_al_complete_io;
113
114 D_ASSERT(e->block_id != ID_VACANT);
115
116 /* after we moved e to done_ee,
117 * we may no longer access it,
118 * it may be freed/reused already!
119 * (as soon as we release the req_lock) */
120 e_sector = e->sector;
121 do_al_complete_io = e->flags & EE_CALL_AL_COMPLETE_IO;
122 is_syncer_req = is_syncer_block_id(e->block_id);
123
124 spin_lock_irqsave(&mdev->req_lock, flags);
125 mdev->writ_cnt += e->size >> 9;
126 list_del(&e->w.list); /* has been on active_ee or sync_ee */
127 list_add_tail(&e->w.list, &mdev->done_ee);
128
129 /* No hlist_del_init(&e->collision) here, we did not send the Ack yet,
130 * neither did we wake possibly waiting conflicting requests.
131 * done from "drbd_process_done_ee" within the appropriate w.cb
132 * (e_end_block/e_end_resync_block) or from _drbd_clear_done_ee */
133
134 do_wake = is_syncer_req
135 ? list_empty(&mdev->sync_ee)
136 : list_empty(&mdev->active_ee);
137
138 if (test_bit(__EE_WAS_ERROR, &e->flags))
139 __drbd_chk_io_error(mdev, false);
140 spin_unlock_irqrestore(&mdev->req_lock, flags);
141
142 if (is_syncer_req)
143 drbd_rs_complete_io(mdev, e_sector);
144
145 if (do_wake)
146 wake_up(&mdev->ee_wait);
147
148 if (do_al_complete_io)
149 drbd_al_complete_io(mdev, e_sector);
150
151 wake_asender(mdev);
152 put_ldev(mdev);
153 }
154
155 /* writes on behalf of the partner, or resync writes,
156 * "submitted" by the receiver.
157 */
drbd_endio_sec(struct bio * bio,int error)158 void drbd_endio_sec(struct bio *bio, int error)
159 {
160 struct drbd_epoch_entry *e = bio->bi_private;
161 struct drbd_conf *mdev = e->mdev;
162 int uptodate = bio_flagged(bio, BIO_UPTODATE);
163 int is_write = bio_data_dir(bio) == WRITE;
164
165 if (error && __ratelimit(&drbd_ratelimit_state))
166 dev_warn(DEV, "%s: error=%d s=%llus\n",
167 is_write ? "write" : "read", error,
168 (unsigned long long)e->sector);
169 if (!error && !uptodate) {
170 if (__ratelimit(&drbd_ratelimit_state))
171 dev_warn(DEV, "%s: setting error to -EIO s=%llus\n",
172 is_write ? "write" : "read",
173 (unsigned long long)e->sector);
174 /* strange behavior of some lower level drivers...
175 * fail the request by clearing the uptodate flag,
176 * but do not return any error?! */
177 error = -EIO;
178 }
179
180 if (error)
181 set_bit(__EE_WAS_ERROR, &e->flags);
182
183 bio_put(bio); /* no need for the bio anymore */
184 if (atomic_dec_and_test(&e->pending_bios)) {
185 if (is_write)
186 drbd_endio_write_sec_final(e);
187 else
188 drbd_endio_read_sec_final(e);
189 }
190 }
191
192 /* read, readA or write requests on R_PRIMARY coming from drbd_make_request
193 */
drbd_endio_pri(struct bio * bio,int error)194 void drbd_endio_pri(struct bio *bio, int error)
195 {
196 unsigned long flags;
197 struct drbd_request *req = bio->bi_private;
198 struct drbd_conf *mdev = req->mdev;
199 struct bio_and_error m;
200 enum drbd_req_event what;
201 int uptodate = bio_flagged(bio, BIO_UPTODATE);
202
203 if (!error && !uptodate) {
204 dev_warn(DEV, "p %s: setting error to -EIO\n",
205 bio_data_dir(bio) == WRITE ? "write" : "read");
206 /* strange behavior of some lower level drivers...
207 * fail the request by clearing the uptodate flag,
208 * but do not return any error?! */
209 error = -EIO;
210 }
211
212 /* to avoid recursion in __req_mod */
213 if (unlikely(error)) {
214 what = (bio_data_dir(bio) == WRITE)
215 ? write_completed_with_error
216 : (bio_rw(bio) == READ)
217 ? read_completed_with_error
218 : read_ahead_completed_with_error;
219 } else
220 what = completed_ok;
221
222 bio_put(req->private_bio);
223 req->private_bio = ERR_PTR(error);
224
225 /* not req_mod(), we need irqsave here! */
226 spin_lock_irqsave(&mdev->req_lock, flags);
227 __req_mod(req, what, &m);
228 spin_unlock_irqrestore(&mdev->req_lock, flags);
229
230 if (m.bio)
231 complete_master_bio(mdev, &m);
232 }
233
w_read_retry_remote(struct drbd_conf * mdev,struct drbd_work * w,int cancel)234 int w_read_retry_remote(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
235 {
236 struct drbd_request *req = container_of(w, struct drbd_request, w);
237
238 /* We should not detach for read io-error,
239 * but try to WRITE the P_DATA_REPLY to the failed location,
240 * to give the disk the chance to relocate that block */
241
242 spin_lock_irq(&mdev->req_lock);
243 if (cancel || mdev->state.pdsk != D_UP_TO_DATE) {
244 _req_mod(req, read_retry_remote_canceled);
245 spin_unlock_irq(&mdev->req_lock);
246 return 1;
247 }
248 spin_unlock_irq(&mdev->req_lock);
249
250 return w_send_read_req(mdev, w, 0);
251 }
252
drbd_csum_ee(struct drbd_conf * mdev,struct crypto_hash * tfm,struct drbd_epoch_entry * e,void * digest)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
drbd_csum_bio(struct drbd_conf * mdev,struct crypto_hash * tfm,struct bio * bio,void * digest)280 void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *bio, void *digest)
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 /* TODO merge common code with w_e_end_ov_req */
w_e_send_csum(struct drbd_conf * mdev,struct drbd_work * w,int cancel)301 int w_e_send_csum(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
302 {
303 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w);
304 int digest_size;
305 void *digest;
306 int ok = 1;
307
308 D_ASSERT(e->block_id == DRBD_MAGIC + 0xbeef);
309
310 if (unlikely(cancel))
311 goto out;
312
313 if (likely((e->flags & EE_WAS_ERROR) != 0))
314 goto out;
315
316 digest_size = crypto_hash_digestsize(mdev->csums_tfm);
317 digest = kmalloc(digest_size, GFP_NOIO);
318 if (digest) {
319 sector_t sector = e->sector;
320 unsigned int size = e->size;
321 drbd_csum_ee(mdev, mdev->csums_tfm, e, digest);
322 /* Free e and pages before send.
323 * In case we block on congestion, we could otherwise run into
324 * some distributed deadlock, if the other side blocks on
325 * congestion as well, because our receiver blocks in
326 * drbd_pp_alloc due to pp_in_use > max_buffers. */
327 drbd_free_ee(mdev, e);
328 e = NULL;
329 inc_rs_pending(mdev);
330 ok = drbd_send_drequest_csum(mdev, sector, size,
331 digest, digest_size,
332 P_CSUM_RS_REQUEST);
333 kfree(digest);
334 } else {
335 dev_err(DEV, "kmalloc() of digest failed.\n");
336 ok = 0;
337 }
338
339 out:
340 if (e)
341 drbd_free_ee(mdev, e);
342
343 if (unlikely(!ok))
344 dev_err(DEV, "drbd_send_drequest(..., csum) failed\n");
345 return ok;
346 }
347
348 #define GFP_TRY (__GFP_HIGHMEM | __GFP_NOWARN)
349
read_for_csum(struct drbd_conf * mdev,sector_t sector,int size)350 static int read_for_csum(struct drbd_conf *mdev, sector_t sector, int size)
351 {
352 struct drbd_epoch_entry *e;
353
354 if (!get_ldev(mdev))
355 return -EIO;
356
357 if (drbd_rs_should_slow_down(mdev, sector))
358 goto defer;
359
360 /* GFP_TRY, because if there is no memory available right now, this may
361 * be rescheduled for later. It is "only" background resync, after all. */
362 e = drbd_alloc_ee(mdev, DRBD_MAGIC+0xbeef, sector, size, GFP_TRY);
363 if (!e)
364 goto defer;
365
366 e->w.cb = w_e_send_csum;
367 spin_lock_irq(&mdev->req_lock);
368 list_add(&e->w.list, &mdev->read_ee);
369 spin_unlock_irq(&mdev->req_lock);
370
371 atomic_add(size >> 9, &mdev->rs_sect_ev);
372 if (drbd_submit_ee(mdev, e, READ, DRBD_FAULT_RS_RD) == 0)
373 return 0;
374
375 /* If it failed because of ENOMEM, retry should help. If it failed
376 * because bio_add_page failed (probably broken lower level driver),
377 * retry may or may not help.
378 * If it does not, you may need to force disconnect. */
379 spin_lock_irq(&mdev->req_lock);
380 list_del(&e->w.list);
381 spin_unlock_irq(&mdev->req_lock);
382
383 drbd_free_ee(mdev, e);
384 defer:
385 put_ldev(mdev);
386 return -EAGAIN;
387 }
388
w_resync_timer(struct drbd_conf * mdev,struct drbd_work * w,int cancel)389 int w_resync_timer(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
390 {
391 switch (mdev->state.conn) {
392 case C_VERIFY_S:
393 w_make_ov_request(mdev, w, cancel);
394 break;
395 case C_SYNC_TARGET:
396 w_make_resync_request(mdev, w, cancel);
397 break;
398 }
399
400 return 1;
401 }
402
resync_timer_fn(unsigned long data)403 void resync_timer_fn(unsigned long data)
404 {
405 struct drbd_conf *mdev = (struct drbd_conf *) data;
406
407 if (list_empty(&mdev->resync_work.list))
408 drbd_queue_work(&mdev->data.work, &mdev->resync_work);
409 }
410
fifo_set(struct fifo_buffer * fb,int value)411 static void fifo_set(struct fifo_buffer *fb, int value)
412 {
413 int i;
414
415 for (i = 0; i < fb->size; i++)
416 fb->values[i] = value;
417 }
418
fifo_push(struct fifo_buffer * fb,int value)419 static int fifo_push(struct fifo_buffer *fb, int value)
420 {
421 int ov;
422
423 ov = fb->values[fb->head_index];
424 fb->values[fb->head_index++] = value;
425
426 if (fb->head_index >= fb->size)
427 fb->head_index = 0;
428
429 return ov;
430 }
431
fifo_add_val(struct fifo_buffer * fb,int value)432 static void fifo_add_val(struct fifo_buffer *fb, int value)
433 {
434 int i;
435
436 for (i = 0; i < fb->size; i++)
437 fb->values[i] += value;
438 }
439
drbd_rs_controller(struct drbd_conf * mdev)440 static int drbd_rs_controller(struct drbd_conf *mdev)
441 {
442 unsigned int sect_in; /* Number of sectors that came in since the last turn */
443 unsigned int want; /* The number of sectors we want in the proxy */
444 int req_sect; /* Number of sectors to request in this turn */
445 int correction; /* Number of sectors more we need in the proxy*/
446 int cps; /* correction per invocation of drbd_rs_controller() */
447 int steps; /* Number of time steps to plan ahead */
448 int curr_corr;
449 int max_sect;
450
451 sect_in = atomic_xchg(&mdev->rs_sect_in, 0); /* Number of sectors that came in */
452 mdev->rs_in_flight -= sect_in;
453
454 spin_lock(&mdev->peer_seq_lock); /* get an atomic view on mdev->rs_plan_s */
455
456 steps = mdev->rs_plan_s.size; /* (mdev->sync_conf.c_plan_ahead * 10 * SLEEP_TIME) / HZ; */
457
458 if (mdev->rs_in_flight + sect_in == 0) { /* At start of resync */
459 want = ((mdev->sync_conf.rate * 2 * SLEEP_TIME) / HZ) * steps;
460 } else { /* normal path */
461 want = mdev->sync_conf.c_fill_target ? mdev->sync_conf.c_fill_target :
462 sect_in * mdev->sync_conf.c_delay_target * HZ / (SLEEP_TIME * 10);
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);
474 spin_unlock(&mdev->peer_seq_lock);
475 mdev->rs_planed -= curr_corr;
476
477 req_sect = sect_in + curr_corr;
478 if (req_sect < 0)
479 req_sect = 0;
480
481 max_sect = (mdev->sync_conf.c_max_rate * 2 * SLEEP_TIME) / HZ;
482 if (req_sect > max_sect)
483 req_sect = max_sect;
484
485 /*
486 dev_warn(DEV, "si=%u if=%d wa=%u co=%d st=%d cps=%d pl=%d cc=%d rs=%d\n",
487 sect_in, mdev->rs_in_flight, want, correction,
488 steps, cps, mdev->rs_planed, curr_corr, req_sect);
489 */
490
491 return req_sect;
492 }
493
drbd_rs_number_requests(struct drbd_conf * mdev)494 static int drbd_rs_number_requests(struct drbd_conf *mdev)
495 {
496 int number;
497 if (mdev->rs_plan_s.size) { /* mdev->sync_conf.c_plan_ahead */
498 number = drbd_rs_controller(mdev) >> (BM_BLOCK_SHIFT - 9);
499 mdev->c_sync_rate = number * HZ * (BM_BLOCK_SIZE / 1024) / SLEEP_TIME;
500 } else {
501 mdev->c_sync_rate = mdev->sync_conf.rate;
502 number = SLEEP_TIME * mdev->c_sync_rate / ((BM_BLOCK_SIZE / 1024) * HZ);
503 }
504
505 /* ignore the amount of pending requests, the resync controller should
506 * throttle down to incoming reply rate soon enough anyways. */
507 return number;
508 }
509
w_make_resync_request(struct drbd_conf * mdev,struct drbd_work * w,int cancel)510 static int w_make_resync_request(struct drbd_conf *mdev,
511 struct drbd_work *w, int cancel)
512 {
513 unsigned long bit;
514 sector_t sector;
515 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
516 int max_bio_size;
517 int number, rollback_i, size;
518 int align, queued, sndbuf;
519 int i = 0;
520
521 if (unlikely(cancel))
522 return 1;
523
524 if (mdev->rs_total == 0) {
525 /* empty resync? */
526 drbd_resync_finished(mdev);
527 return 1;
528 }
529
530 if (!get_ldev(mdev)) {
531 /* Since we only need to access mdev->rsync a
532 get_ldev_if_state(mdev,D_FAILED) would be sufficient, but
533 to continue resync with a broken disk makes no sense at
534 all */
535 dev_err(DEV, "Disk broke down during resync!\n");
536 return 1;
537 }
538
539 max_bio_size = queue_max_hw_sectors(mdev->rq_queue) << 9;
540 number = drbd_rs_number_requests(mdev);
541 if (number == 0)
542 goto requeue;
543
544 for (i = 0; i < number; i++) {
545 /* Stop generating RS requests, when half of the send buffer is filled */
546 mutex_lock(&mdev->data.mutex);
547 if (mdev->data.socket) {
548 queued = mdev->data.socket->sk->sk_wmem_queued;
549 sndbuf = mdev->data.socket->sk->sk_sndbuf;
550 } else {
551 queued = 1;
552 sndbuf = 0;
553 }
554 mutex_unlock(&mdev->data.mutex);
555 if (queued > sndbuf / 2)
556 goto requeue;
557
558 next_sector:
559 size = BM_BLOCK_SIZE;
560 bit = drbd_bm_find_next(mdev, mdev->bm_resync_fo);
561
562 if (bit == DRBD_END_OF_BITMAP) {
563 mdev->bm_resync_fo = drbd_bm_bits(mdev);
564 put_ldev(mdev);
565 return 1;
566 }
567
568 sector = BM_BIT_TO_SECT(bit);
569
570 if (drbd_rs_should_slow_down(mdev, sector) ||
571 drbd_try_rs_begin_io(mdev, sector)) {
572 mdev->bm_resync_fo = bit;
573 goto requeue;
574 }
575 mdev->bm_resync_fo = bit + 1;
576
577 if (unlikely(drbd_bm_test_bit(mdev, bit) == 0)) {
578 drbd_rs_complete_io(mdev, sector);
579 goto next_sector;
580 }
581
582 #if DRBD_MAX_BIO_SIZE > BM_BLOCK_SIZE
583 /* try to find some adjacent bits.
584 * we stop if we have already the maximum req size.
585 *
586 * Additionally always align bigger requests, in order to
587 * be prepared for all stripe sizes of software RAIDs.
588 */
589 align = 1;
590 rollback_i = i;
591 for (;;) {
592 if (size + BM_BLOCK_SIZE > max_bio_size)
593 break;
594
595 /* Be always aligned */
596 if (sector & ((1<<(align+3))-1))
597 break;
598
599 /* do not cross extent boundaries */
600 if (((bit+1) & BM_BLOCKS_PER_BM_EXT_MASK) == 0)
601 break;
602 /* now, is it actually dirty, after all?
603 * caution, drbd_bm_test_bit is tri-state for some
604 * obscure reason; ( b == 0 ) would get the out-of-band
605 * only accidentally right because of the "oddly sized"
606 * adjustment below */
607 if (drbd_bm_test_bit(mdev, bit+1) != 1)
608 break;
609 bit++;
610 size += BM_BLOCK_SIZE;
611 if ((BM_BLOCK_SIZE << align) <= size)
612 align++;
613 i++;
614 }
615 /* if we merged some,
616 * reset the offset to start the next drbd_bm_find_next from */
617 if (size > BM_BLOCK_SIZE)
618 mdev->bm_resync_fo = bit + 1;
619 #endif
620
621 /* adjust very last sectors, in case we are oddly sized */
622 if (sector + (size>>9) > capacity)
623 size = (capacity-sector)<<9;
624 if (mdev->agreed_pro_version >= 89 && mdev->csums_tfm) {
625 switch (read_for_csum(mdev, sector, size)) {
626 case -EIO: /* Disk failure */
627 put_ldev(mdev);
628 return 0;
629 case -EAGAIN: /* allocation failed, or ldev busy */
630 drbd_rs_complete_io(mdev, sector);
631 mdev->bm_resync_fo = BM_SECT_TO_BIT(sector);
632 i = rollback_i;
633 goto requeue;
634 case 0:
635 /* everything ok */
636 break;
637 default:
638 BUG();
639 }
640 } else {
641 inc_rs_pending(mdev);
642 if (!drbd_send_drequest(mdev, P_RS_DATA_REQUEST,
643 sector, size, ID_SYNCER)) {
644 dev_err(DEV, "drbd_send_drequest() failed, aborting...\n");
645 dec_rs_pending(mdev);
646 put_ldev(mdev);
647 return 0;
648 }
649 }
650 }
651
652 if (mdev->bm_resync_fo >= drbd_bm_bits(mdev)) {
653 /* last syncer _request_ was sent,
654 * but the P_RS_DATA_REPLY not yet received. sync will end (and
655 * next sync group will resume), as soon as we receive the last
656 * resync data block, and the last bit is cleared.
657 * until then resync "work" is "inactive" ...
658 */
659 put_ldev(mdev);
660 return 1;
661 }
662
663 requeue:
664 mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9));
665 mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME);
666 put_ldev(mdev);
667 return 1;
668 }
669
w_make_ov_request(struct drbd_conf * mdev,struct drbd_work * w,int cancel)670 static int w_make_ov_request(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
671 {
672 int number, i, size;
673 sector_t sector;
674 const sector_t capacity = drbd_get_capacity(mdev->this_bdev);
675
676 if (unlikely(cancel))
677 return 1;
678
679 number = drbd_rs_number_requests(mdev);
680
681 sector = mdev->ov_position;
682 for (i = 0; i < number; i++) {
683 if (sector >= capacity) {
684 return 1;
685 }
686
687 size = BM_BLOCK_SIZE;
688
689 if (drbd_rs_should_slow_down(mdev, sector) ||
690 drbd_try_rs_begin_io(mdev, sector)) {
691 mdev->ov_position = sector;
692 goto requeue;
693 }
694
695 if (sector + (size>>9) > capacity)
696 size = (capacity-sector)<<9;
697
698 inc_rs_pending(mdev);
699 if (!drbd_send_ov_request(mdev, sector, size)) {
700 dec_rs_pending(mdev);
701 return 0;
702 }
703 sector += BM_SECT_PER_BIT;
704 }
705 mdev->ov_position = sector;
706
707 requeue:
708 mdev->rs_in_flight += (i << (BM_BLOCK_SHIFT - 9));
709 mod_timer(&mdev->resync_timer, jiffies + SLEEP_TIME);
710 return 1;
711 }
712
713
start_resync_timer_fn(unsigned long data)714 void start_resync_timer_fn(unsigned long data)
715 {
716 struct drbd_conf *mdev = (struct drbd_conf *) data;
717
718 drbd_queue_work(&mdev->data.work, &mdev->start_resync_work);
719 }
720
w_start_resync(struct drbd_conf * mdev,struct drbd_work * w,int cancel)721 int w_start_resync(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
722 {
723 if (atomic_read(&mdev->unacked_cnt) || atomic_read(&mdev->rs_pending_cnt)) {
724 dev_warn(DEV, "w_start_resync later...\n");
725 mdev->start_resync_timer.expires = jiffies + HZ/10;
726 add_timer(&mdev->start_resync_timer);
727 return 1;
728 }
729
730 drbd_start_resync(mdev, C_SYNC_SOURCE);
731 clear_bit(AHEAD_TO_SYNC_SOURCE, &mdev->current_epoch->flags);
732 return 1;
733 }
734
w_ov_finished(struct drbd_conf * mdev,struct drbd_work * w,int cancel)735 int w_ov_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
736 {
737 kfree(w);
738 ov_oos_print(mdev);
739 drbd_resync_finished(mdev);
740
741 return 1;
742 }
743
w_resync_finished(struct drbd_conf * mdev,struct drbd_work * w,int cancel)744 static int w_resync_finished(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
745 {
746 kfree(w);
747
748 drbd_resync_finished(mdev);
749
750 return 1;
751 }
752
ping_peer(struct drbd_conf * mdev)753 static void ping_peer(struct drbd_conf *mdev)
754 {
755 clear_bit(GOT_PING_ACK, &mdev->flags);
756 request_ping(mdev);
757 wait_event(mdev->misc_wait,
758 test_bit(GOT_PING_ACK, &mdev->flags) || mdev->state.conn < C_CONNECTED);
759 }
760
drbd_resync_finished(struct drbd_conf * mdev)761 int drbd_resync_finished(struct drbd_conf *mdev)
762 {
763 unsigned long db, dt, dbdt;
764 unsigned long n_oos;
765 union drbd_state os, ns;
766 struct drbd_work *w;
767 char *khelper_cmd = NULL;
768 int verify_done = 0;
769
770 /* Remove all elements from the resync LRU. Since future actions
771 * might set bits in the (main) bitmap, then the entries in the
772 * resync LRU would be wrong. */
773 if (drbd_rs_del_all(mdev)) {
774 /* In case this is not possible now, most probably because
775 * there are P_RS_DATA_REPLY Packets lingering on the worker's
776 * queue (or even the read operations for those packets
777 * is not finished by now). Retry in 100ms. */
778
779 schedule_timeout_interruptible(HZ / 10);
780 w = kmalloc(sizeof(struct drbd_work), GFP_ATOMIC);
781 if (w) {
782 w->cb = w_resync_finished;
783 drbd_queue_work(&mdev->data.work, w);
784 return 1;
785 }
786 dev_err(DEV, "Warn failed to drbd_rs_del_all() and to kmalloc(w).\n");
787 }
788
789 dt = (jiffies - mdev->rs_start - mdev->rs_paused) / HZ;
790 if (dt <= 0)
791 dt = 1;
792 db = mdev->rs_total;
793 dbdt = Bit2KB(db/dt);
794 mdev->rs_paused /= HZ;
795
796 if (!get_ldev(mdev))
797 goto out;
798
799 ping_peer(mdev);
800
801 spin_lock_irq(&mdev->req_lock);
802 os = mdev->state;
803
804 verify_done = (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T);
805
806 /* This protects us against multiple calls (that can happen in the presence
807 of application IO), and against connectivity loss just before we arrive here. */
808 if (os.conn <= C_CONNECTED)
809 goto out_unlock;
810
811 ns = os;
812 ns.conn = C_CONNECTED;
813
814 dev_info(DEV, "%s done (total %lu sec; paused %lu sec; %lu K/sec)\n",
815 verify_done ? "Online verify " : "Resync",
816 dt + mdev->rs_paused, mdev->rs_paused, dbdt);
817
818 n_oos = drbd_bm_total_weight(mdev);
819
820 if (os.conn == C_VERIFY_S || os.conn == C_VERIFY_T) {
821 if (n_oos) {
822 dev_alert(DEV, "Online verify found %lu %dk block out of sync!\n",
823 n_oos, Bit2KB(1));
824 khelper_cmd = "out-of-sync";
825 }
826 } else {
827 D_ASSERT((n_oos - mdev->rs_failed) == 0);
828
829 if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T)
830 khelper_cmd = "after-resync-target";
831
832 if (mdev->csums_tfm && mdev->rs_total) {
833 const unsigned long s = mdev->rs_same_csum;
834 const unsigned long t = mdev->rs_total;
835 const int ratio =
836 (t == 0) ? 0 :
837 (t < 100000) ? ((s*100)/t) : (s/(t/100));
838 dev_info(DEV, "%u %% had equal checksums, eliminated: %luK; "
839 "transferred %luK total %luK\n",
840 ratio,
841 Bit2KB(mdev->rs_same_csum),
842 Bit2KB(mdev->rs_total - mdev->rs_same_csum),
843 Bit2KB(mdev->rs_total));
844 }
845 }
846
847 if (mdev->rs_failed) {
848 dev_info(DEV, " %lu failed blocks\n", mdev->rs_failed);
849
850 if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) {
851 ns.disk = D_INCONSISTENT;
852 ns.pdsk = D_UP_TO_DATE;
853 } else {
854 ns.disk = D_UP_TO_DATE;
855 ns.pdsk = D_INCONSISTENT;
856 }
857 } else {
858 ns.disk = D_UP_TO_DATE;
859 ns.pdsk = D_UP_TO_DATE;
860
861 if (os.conn == C_SYNC_TARGET || os.conn == C_PAUSED_SYNC_T) {
862 if (mdev->p_uuid) {
863 int i;
864 for (i = UI_BITMAP ; i <= UI_HISTORY_END ; i++)
865 _drbd_uuid_set(mdev, i, mdev->p_uuid[i]);
866 drbd_uuid_set(mdev, UI_BITMAP, mdev->ldev->md.uuid[UI_CURRENT]);
867 _drbd_uuid_set(mdev, UI_CURRENT, mdev->p_uuid[UI_CURRENT]);
868 } else {
869 dev_err(DEV, "mdev->p_uuid is NULL! BUG\n");
870 }
871 }
872
873 if (!(os.conn == C_VERIFY_S || os.conn == C_VERIFY_T)) {
874 /* for verify runs, we don't update uuids here,
875 * so there would be nothing to report. */
876 drbd_uuid_set_bm(mdev, 0UL);
877 drbd_print_uuids(mdev, "updated UUIDs");
878 if (mdev->p_uuid) {
879 /* Now the two UUID sets are equal, update what we
880 * know of the peer. */
881 int i;
882 for (i = UI_CURRENT ; i <= UI_HISTORY_END ; i++)
883 mdev->p_uuid[i] = mdev->ldev->md.uuid[i];
884 }
885 }
886 }
887
888 _drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
889 out_unlock:
890 spin_unlock_irq(&mdev->req_lock);
891 put_ldev(mdev);
892 out:
893 mdev->rs_total = 0;
894 mdev->rs_failed = 0;
895 mdev->rs_paused = 0;
896 if (verify_done)
897 mdev->ov_start_sector = 0;
898
899 drbd_md_sync(mdev);
900
901 if (khelper_cmd)
902 drbd_khelper(mdev, khelper_cmd);
903
904 return 1;
905 }
906
907 /* helper */
move_to_net_ee_or_free(struct drbd_conf * mdev,struct drbd_epoch_entry * e)908 static void move_to_net_ee_or_free(struct drbd_conf *mdev, struct drbd_epoch_entry *e)
909 {
910 if (drbd_ee_has_active_page(e)) {
911 /* This might happen if sendpage() has not finished */
912 int i = (e->size + PAGE_SIZE -1) >> PAGE_SHIFT;
913 atomic_add(i, &mdev->pp_in_use_by_net);
914 atomic_sub(i, &mdev->pp_in_use);
915 spin_lock_irq(&mdev->req_lock);
916 list_add_tail(&e->w.list, &mdev->net_ee);
917 spin_unlock_irq(&mdev->req_lock);
918 wake_up(&drbd_pp_wait);
919 } else
920 drbd_free_ee(mdev, e);
921 }
922
923 /**
924 * w_e_end_data_req() - Worker callback, to send a P_DATA_REPLY packet in response to a P_DATA_REQUEST
925 * @mdev: DRBD device.
926 * @w: work object.
927 * @cancel: The connection will be closed anyways
928 */
w_e_end_data_req(struct drbd_conf * mdev,struct drbd_work * w,int cancel)929 int w_e_end_data_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
930 {
931 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w);
932 int ok;
933
934 if (unlikely(cancel)) {
935 drbd_free_ee(mdev, e);
936 dec_unacked(mdev);
937 return 1;
938 }
939
940 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
941 ok = drbd_send_block(mdev, P_DATA_REPLY, e);
942 } else {
943 if (__ratelimit(&drbd_ratelimit_state))
944 dev_err(DEV, "Sending NegDReply. sector=%llus.\n",
945 (unsigned long long)e->sector);
946
947 ok = drbd_send_ack(mdev, P_NEG_DREPLY, e);
948 }
949
950 dec_unacked(mdev);
951
952 move_to_net_ee_or_free(mdev, e);
953
954 if (unlikely(!ok))
955 dev_err(DEV, "drbd_send_block() failed\n");
956 return ok;
957 }
958
959 /**
960 * w_e_end_rsdata_req() - Worker callback to send a P_RS_DATA_REPLY packet in response to a P_RS_DATA_REQUESTRS
961 * @mdev: DRBD device.
962 * @w: work object.
963 * @cancel: The connection will be closed anyways
964 */
w_e_end_rsdata_req(struct drbd_conf * mdev,struct drbd_work * w,int cancel)965 int w_e_end_rsdata_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
966 {
967 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w);
968 int ok;
969
970 if (unlikely(cancel)) {
971 drbd_free_ee(mdev, e);
972 dec_unacked(mdev);
973 return 1;
974 }
975
976 if (get_ldev_if_state(mdev, D_FAILED)) {
977 drbd_rs_complete_io(mdev, e->sector);
978 put_ldev(mdev);
979 }
980
981 if (mdev->state.conn == C_AHEAD) {
982 ok = drbd_send_ack(mdev, P_RS_CANCEL, e);
983 } else if (likely((e->flags & EE_WAS_ERROR) == 0)) {
984 if (likely(mdev->state.pdsk >= D_INCONSISTENT)) {
985 inc_rs_pending(mdev);
986 ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e);
987 } else {
988 if (__ratelimit(&drbd_ratelimit_state))
989 dev_err(DEV, "Not sending RSDataReply, "
990 "partner DISKLESS!\n");
991 ok = 1;
992 }
993 } else {
994 if (__ratelimit(&drbd_ratelimit_state))
995 dev_err(DEV, "Sending NegRSDReply. sector %llus.\n",
996 (unsigned long long)e->sector);
997
998 ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e);
999
1000 /* update resync data with failure */
1001 drbd_rs_failed_io(mdev, e->sector, e->size);
1002 }
1003
1004 dec_unacked(mdev);
1005
1006 move_to_net_ee_or_free(mdev, e);
1007
1008 if (unlikely(!ok))
1009 dev_err(DEV, "drbd_send_block() failed\n");
1010 return ok;
1011 }
1012
w_e_end_csum_rs_req(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1013 int w_e_end_csum_rs_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1014 {
1015 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w);
1016 struct digest_info *di;
1017 int digest_size;
1018 void *digest = NULL;
1019 int ok, eq = 0;
1020
1021 if (unlikely(cancel)) {
1022 drbd_free_ee(mdev, e);
1023 dec_unacked(mdev);
1024 return 1;
1025 }
1026
1027 if (get_ldev(mdev)) {
1028 drbd_rs_complete_io(mdev, e->sector);
1029 put_ldev(mdev);
1030 }
1031
1032 di = e->digest;
1033
1034 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1035 /* quick hack to try to avoid a race against reconfiguration.
1036 * a real fix would be much more involved,
1037 * introducing more locking mechanisms */
1038 if (mdev->csums_tfm) {
1039 digest_size = crypto_hash_digestsize(mdev->csums_tfm);
1040 D_ASSERT(digest_size == di->digest_size);
1041 digest = kmalloc(digest_size, GFP_NOIO);
1042 }
1043 if (digest) {
1044 drbd_csum_ee(mdev, mdev->csums_tfm, e, digest);
1045 eq = !memcmp(digest, di->digest, digest_size);
1046 kfree(digest);
1047 }
1048
1049 if (eq) {
1050 drbd_set_in_sync(mdev, e->sector, e->size);
1051 /* rs_same_csums unit is BM_BLOCK_SIZE */
1052 mdev->rs_same_csum += e->size >> BM_BLOCK_SHIFT;
1053 ok = drbd_send_ack(mdev, P_RS_IS_IN_SYNC, e);
1054 } else {
1055 inc_rs_pending(mdev);
1056 e->block_id = ID_SYNCER; /* By setting block_id, digest pointer becomes invalid! */
1057 e->flags &= ~EE_HAS_DIGEST; /* This e no longer has a digest pointer */
1058 kfree(di);
1059 ok = drbd_send_block(mdev, P_RS_DATA_REPLY, e);
1060 }
1061 } else {
1062 ok = drbd_send_ack(mdev, P_NEG_RS_DREPLY, e);
1063 if (__ratelimit(&drbd_ratelimit_state))
1064 dev_err(DEV, "Sending NegDReply. I guess it gets messy.\n");
1065 }
1066
1067 dec_unacked(mdev);
1068 move_to_net_ee_or_free(mdev, e);
1069
1070 if (unlikely(!ok))
1071 dev_err(DEV, "drbd_send_block/ack() failed\n");
1072 return ok;
1073 }
1074
1075 /* TODO merge common code with w_e_send_csum */
w_e_end_ov_req(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1076 int w_e_end_ov_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1077 {
1078 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w);
1079 sector_t sector = e->sector;
1080 unsigned int size = e->size;
1081 int digest_size;
1082 void *digest;
1083 int ok = 1;
1084
1085 if (unlikely(cancel))
1086 goto out;
1087
1088 digest_size = crypto_hash_digestsize(mdev->verify_tfm);
1089 digest = kmalloc(digest_size, GFP_NOIO);
1090 if (!digest) {
1091 ok = 0; /* terminate the connection in case the allocation failed */
1092 goto out;
1093 }
1094
1095 if (likely(!(e->flags & EE_WAS_ERROR)))
1096 drbd_csum_ee(mdev, mdev->verify_tfm, e, digest);
1097 else
1098 memset(digest, 0, digest_size);
1099
1100 /* Free e and pages before send.
1101 * In case we block on congestion, we could otherwise run into
1102 * some distributed deadlock, if the other side blocks on
1103 * congestion as well, because our receiver blocks in
1104 * drbd_pp_alloc due to pp_in_use > max_buffers. */
1105 drbd_free_ee(mdev, e);
1106 e = NULL;
1107 inc_rs_pending(mdev);
1108 ok = drbd_send_drequest_csum(mdev, sector, size,
1109 digest, digest_size,
1110 P_OV_REPLY);
1111 if (!ok)
1112 dec_rs_pending(mdev);
1113 kfree(digest);
1114
1115 out:
1116 if (e)
1117 drbd_free_ee(mdev, e);
1118 dec_unacked(mdev);
1119 return ok;
1120 }
1121
drbd_ov_oos_found(struct drbd_conf * mdev,sector_t sector,int size)1122 void drbd_ov_oos_found(struct drbd_conf *mdev, sector_t sector, int size)
1123 {
1124 if (mdev->ov_last_oos_start + mdev->ov_last_oos_size == sector) {
1125 mdev->ov_last_oos_size += size>>9;
1126 } else {
1127 mdev->ov_last_oos_start = sector;
1128 mdev->ov_last_oos_size = size>>9;
1129 }
1130 drbd_set_out_of_sync(mdev, sector, size);
1131 }
1132
w_e_end_ov_reply(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1133 int w_e_end_ov_reply(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1134 {
1135 struct drbd_epoch_entry *e = container_of(w, struct drbd_epoch_entry, w);
1136 struct digest_info *di;
1137 void *digest;
1138 sector_t sector = e->sector;
1139 unsigned int size = e->size;
1140 int digest_size;
1141 int ok, eq = 0;
1142
1143 if (unlikely(cancel)) {
1144 drbd_free_ee(mdev, e);
1145 dec_unacked(mdev);
1146 return 1;
1147 }
1148
1149 /* after "cancel", because after drbd_disconnect/drbd_rs_cancel_all
1150 * the resync lru has been cleaned up already */
1151 if (get_ldev(mdev)) {
1152 drbd_rs_complete_io(mdev, e->sector);
1153 put_ldev(mdev);
1154 }
1155
1156 di = e->digest;
1157
1158 if (likely((e->flags & EE_WAS_ERROR) == 0)) {
1159 digest_size = crypto_hash_digestsize(mdev->verify_tfm);
1160 digest = kmalloc(digest_size, GFP_NOIO);
1161 if (digest) {
1162 drbd_csum_ee(mdev, mdev->verify_tfm, e, digest);
1163
1164 D_ASSERT(digest_size == di->digest_size);
1165 eq = !memcmp(digest, di->digest, digest_size);
1166 kfree(digest);
1167 }
1168 }
1169
1170 /* Free e and pages before send.
1171 * In case we block on congestion, we could otherwise run into
1172 * some distributed deadlock, if the other side blocks on
1173 * congestion as well, because our receiver blocks in
1174 * drbd_pp_alloc due to pp_in_use > max_buffers. */
1175 drbd_free_ee(mdev, e);
1176 if (!eq)
1177 drbd_ov_oos_found(mdev, sector, size);
1178 else
1179 ov_oos_print(mdev);
1180
1181 ok = drbd_send_ack_ex(mdev, P_OV_RESULT, sector, size,
1182 eq ? ID_IN_SYNC : ID_OUT_OF_SYNC);
1183
1184 dec_unacked(mdev);
1185
1186 --mdev->ov_left;
1187
1188 /* let's advance progress step marks only for every other megabyte */
1189 if ((mdev->ov_left & 0x200) == 0x200)
1190 drbd_advance_rs_marks(mdev, mdev->ov_left);
1191
1192 if (mdev->ov_left == 0) {
1193 ov_oos_print(mdev);
1194 drbd_resync_finished(mdev);
1195 }
1196
1197 return ok;
1198 }
1199
w_prev_work_done(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1200 int w_prev_work_done(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1201 {
1202 struct drbd_wq_barrier *b = container_of(w, struct drbd_wq_barrier, w);
1203 complete(&b->done);
1204 return 1;
1205 }
1206
w_send_barrier(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1207 int w_send_barrier(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1208 {
1209 struct drbd_tl_epoch *b = container_of(w, struct drbd_tl_epoch, w);
1210 struct p_barrier *p = &mdev->data.sbuf.barrier;
1211 int ok = 1;
1212
1213 /* really avoid racing with tl_clear. w.cb may have been referenced
1214 * just before it was reassigned and re-queued, so double check that.
1215 * actually, this race was harmless, since we only try to send the
1216 * barrier packet here, and otherwise do nothing with the object.
1217 * but compare with the head of w_clear_epoch */
1218 spin_lock_irq(&mdev->req_lock);
1219 if (w->cb != w_send_barrier || mdev->state.conn < C_CONNECTED)
1220 cancel = 1;
1221 spin_unlock_irq(&mdev->req_lock);
1222 if (cancel)
1223 return 1;
1224
1225 if (!drbd_get_data_sock(mdev))
1226 return 0;
1227 p->barrier = b->br_number;
1228 /* inc_ap_pending was done where this was queued.
1229 * dec_ap_pending will be done in got_BarrierAck
1230 * or (on connection loss) in w_clear_epoch. */
1231 ok = _drbd_send_cmd(mdev, mdev->data.socket, P_BARRIER,
1232 (struct p_header80 *)p, sizeof(*p), 0);
1233 drbd_put_data_sock(mdev);
1234
1235 return ok;
1236 }
1237
w_send_write_hint(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1238 int w_send_write_hint(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1239 {
1240 if (cancel)
1241 return 1;
1242 return drbd_send_short_cmd(mdev, P_UNPLUG_REMOTE);
1243 }
1244
w_send_oos(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1245 int w_send_oos(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1246 {
1247 struct drbd_request *req = container_of(w, struct drbd_request, w);
1248 int ok;
1249
1250 if (unlikely(cancel)) {
1251 req_mod(req, send_canceled);
1252 return 1;
1253 }
1254
1255 ok = drbd_send_oos(mdev, req);
1256 req_mod(req, oos_handed_to_network);
1257
1258 return ok;
1259 }
1260
1261 /**
1262 * w_send_dblock() - Worker callback to send a P_DATA packet in order to mirror a write request
1263 * @mdev: DRBD device.
1264 * @w: work object.
1265 * @cancel: The connection will be closed anyways
1266 */
w_send_dblock(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1267 int w_send_dblock(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1268 {
1269 struct drbd_request *req = container_of(w, struct drbd_request, w);
1270 int ok;
1271
1272 if (unlikely(cancel)) {
1273 req_mod(req, send_canceled);
1274 return 1;
1275 }
1276
1277 ok = drbd_send_dblock(mdev, req);
1278 req_mod(req, ok ? handed_over_to_network : send_failed);
1279
1280 return ok;
1281 }
1282
1283 /**
1284 * w_send_read_req() - Worker callback to send a read request (P_DATA_REQUEST) packet
1285 * @mdev: DRBD device.
1286 * @w: work object.
1287 * @cancel: The connection will be closed anyways
1288 */
w_send_read_req(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1289 int w_send_read_req(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1290 {
1291 struct drbd_request *req = container_of(w, struct drbd_request, w);
1292 int ok;
1293
1294 if (unlikely(cancel)) {
1295 req_mod(req, send_canceled);
1296 return 1;
1297 }
1298
1299 ok = drbd_send_drequest(mdev, P_DATA_REQUEST, req->sector, req->size,
1300 (unsigned long)req);
1301
1302 if (!ok) {
1303 /* ?? we set C_TIMEOUT or C_BROKEN_PIPE in drbd_send();
1304 * so this is probably redundant */
1305 if (mdev->state.conn >= C_CONNECTED)
1306 drbd_force_state(mdev, NS(conn, C_NETWORK_FAILURE));
1307 }
1308 req_mod(req, ok ? handed_over_to_network : send_failed);
1309
1310 return ok;
1311 }
1312
w_restart_disk_io(struct drbd_conf * mdev,struct drbd_work * w,int cancel)1313 int w_restart_disk_io(struct drbd_conf *mdev, struct drbd_work *w, int cancel)
1314 {
1315 struct drbd_request *req = container_of(w, struct drbd_request, w);
1316
1317 if (bio_data_dir(req->master_bio) == WRITE && req->rq_state & RQ_IN_ACT_LOG)
1318 drbd_al_begin_io(mdev, req->sector);
1319 /* Calling drbd_al_begin_io() out of the worker might deadlocks
1320 theoretically. Practically it can not deadlock, since this is
1321 only used when unfreezing IOs. All the extents of the requests
1322 that made it into the TL are already active */
1323
1324 drbd_req_make_private_bio(req, req->master_bio);
1325 req->private_bio->bi_bdev = mdev->ldev->backing_bdev;
1326 generic_make_request(req->private_bio);
1327
1328 return 1;
1329 }
1330
_drbd_may_sync_now(struct drbd_conf * mdev)1331 static int _drbd_may_sync_now(struct drbd_conf *mdev)
1332 {
1333 struct drbd_conf *odev = mdev;
1334
1335 while (1) {
1336 if (odev->sync_conf.after == -1)
1337 return 1;
1338 odev = minor_to_mdev(odev->sync_conf.after);
1339 ERR_IF(!odev) return 1;
1340 if ((odev->state.conn >= C_SYNC_SOURCE &&
1341 odev->state.conn <= C_PAUSED_SYNC_T) ||
1342 odev->state.aftr_isp || odev->state.peer_isp ||
1343 odev->state.user_isp)
1344 return 0;
1345 }
1346 }
1347
1348 /**
1349 * _drbd_pause_after() - Pause resync on all devices that may not resync now
1350 * @mdev: DRBD device.
1351 *
1352 * Called from process context only (admin command and after_state_ch).
1353 */
_drbd_pause_after(struct drbd_conf * mdev)1354 static int _drbd_pause_after(struct drbd_conf *mdev)
1355 {
1356 struct drbd_conf *odev;
1357 int i, rv = 0;
1358
1359 for (i = 0; i < minor_count; i++) {
1360 odev = minor_to_mdev(i);
1361 if (!odev)
1362 continue;
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 }
1369
1370 return rv;
1371 }
1372
1373 /**
1374 * _drbd_resume_next() - Resume resync on all devices that may resync now
1375 * @mdev: DRBD device.
1376 *
1377 * Called from process context only (admin command and worker).
1378 */
_drbd_resume_next(struct drbd_conf * mdev)1379 static int _drbd_resume_next(struct drbd_conf *mdev)
1380 {
1381 struct drbd_conf *odev;
1382 int i, rv = 0;
1383
1384 for (i = 0; i < minor_count; i++) {
1385 odev = minor_to_mdev(i);
1386 if (!odev)
1387 continue;
1388 if (odev->state.conn == C_STANDALONE && odev->state.disk == D_DISKLESS)
1389 continue;
1390 if (odev->state.aftr_isp) {
1391 if (_drbd_may_sync_now(odev))
1392 rv |= (__drbd_set_state(_NS(odev, aftr_isp, 0),
1393 CS_HARD, NULL)
1394 != SS_NOTHING_TO_DO) ;
1395 }
1396 }
1397 return rv;
1398 }
1399
resume_next_sg(struct drbd_conf * mdev)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
suspend_other_sg(struct drbd_conf * mdev)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
sync_after_error(struct drbd_conf * mdev,int o_minor)1414 static int sync_after_error(struct drbd_conf *mdev, int o_minor)
1415 {
1416 struct drbd_conf *odev;
1417
1418 if (o_minor == -1)
1419 return NO_ERROR;
1420 if (o_minor < -1 || minor_to_mdev(o_minor) == NULL)
1421 return ERR_SYNC_AFTER;
1422
1423 /* check for loops */
1424 odev = minor_to_mdev(o_minor);
1425 while (1) {
1426 if (odev == mdev)
1427 return ERR_SYNC_AFTER_CYCLE;
1428
1429 /* dependency chain ends here, no cycles. */
1430 if (odev->sync_conf.after == -1)
1431 return NO_ERROR;
1432
1433 /* follow the dependency chain */
1434 odev = minor_to_mdev(odev->sync_conf.after);
1435 }
1436 }
1437
drbd_alter_sa(struct drbd_conf * mdev,int na)1438 int drbd_alter_sa(struct drbd_conf *mdev, int na)
1439 {
1440 int changes;
1441 int retcode;
1442
1443 write_lock_irq(&global_state_lock);
1444 retcode = sync_after_error(mdev, na);
1445 if (retcode == NO_ERROR) {
1446 mdev->sync_conf.after = na;
1447 do {
1448 changes = _drbd_pause_after(mdev);
1449 changes |= _drbd_resume_next(mdev);
1450 } while (changes);
1451 }
1452 write_unlock_irq(&global_state_lock);
1453 return retcode;
1454 }
1455
drbd_rs_controller_reset(struct drbd_conf * mdev)1456 void drbd_rs_controller_reset(struct drbd_conf *mdev)
1457 {
1458 atomic_set(&mdev->rs_sect_in, 0);
1459 atomic_set(&mdev->rs_sect_ev, 0);
1460 mdev->rs_in_flight = 0;
1461 mdev->rs_planed = 0;
1462 spin_lock(&mdev->peer_seq_lock);
1463 fifo_set(&mdev->rs_plan_s, 0);
1464 spin_unlock(&mdev->peer_seq_lock);
1465 }
1466
1467 /**
1468 * drbd_start_resync() - Start the resync process
1469 * @mdev: DRBD device.
1470 * @side: Either C_SYNC_SOURCE or C_SYNC_TARGET
1471 *
1472 * This function might bring you directly into one of the
1473 * C_PAUSED_SYNC_* states.
1474 */
drbd_start_resync(struct drbd_conf * mdev,enum drbd_conns side)1475 void drbd_start_resync(struct drbd_conf *mdev, enum drbd_conns side)
1476 {
1477 union drbd_state ns;
1478 int r;
1479
1480 if (mdev->state.conn >= C_SYNC_SOURCE && mdev->state.conn < C_AHEAD) {
1481 dev_err(DEV, "Resync already running!\n");
1482 return;
1483 }
1484
1485 if (mdev->state.conn < C_AHEAD) {
1486 /* In case a previous resync run was aborted by an IO error/detach on the peer. */
1487 drbd_rs_cancel_all(mdev);
1488 /* This should be done when we abort the resync. We definitely do not
1489 want to have this for connections going back and forth between
1490 Ahead/Behind and SyncSource/SyncTarget */
1491 }
1492
1493 if (side == C_SYNC_TARGET) {
1494 /* Since application IO was locked out during C_WF_BITMAP_T and
1495 C_WF_SYNC_UUID we are still unmodified. Before going to C_SYNC_TARGET
1496 we check that we might make the data inconsistent. */
1497 r = drbd_khelper(mdev, "before-resync-target");
1498 r = (r >> 8) & 0xff;
1499 if (r > 0) {
1500 dev_info(DEV, "before-resync-target handler returned %d, "
1501 "dropping connection.\n", r);
1502 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1503 return;
1504 }
1505 } else /* C_SYNC_SOURCE */ {
1506 r = drbd_khelper(mdev, "before-resync-source");
1507 r = (r >> 8) & 0xff;
1508 if (r > 0) {
1509 if (r == 3) {
1510 dev_info(DEV, "before-resync-source handler returned %d, "
1511 "ignoring. Old userland tools?", r);
1512 } else {
1513 dev_info(DEV, "before-resync-source handler returned %d, "
1514 "dropping connection.\n", r);
1515 drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
1516 return;
1517 }
1518 }
1519 }
1520
1521 drbd_state_lock(mdev);
1522
1523 if (!get_ldev_if_state(mdev, D_NEGOTIATING)) {
1524 drbd_state_unlock(mdev);
1525 return;
1526 }
1527
1528 write_lock_irq(&global_state_lock);
1529 ns = mdev->state;
1530
1531 ns.aftr_isp = !_drbd_may_sync_now(mdev);
1532
1533 ns.conn = side;
1534
1535 if (side == C_SYNC_TARGET)
1536 ns.disk = D_INCONSISTENT;
1537 else /* side == C_SYNC_SOURCE */
1538 ns.pdsk = D_INCONSISTENT;
1539
1540 r = __drbd_set_state(mdev, ns, CS_VERBOSE, NULL);
1541 ns = mdev->state;
1542
1543 if (ns.conn < C_CONNECTED)
1544 r = SS_UNKNOWN_ERROR;
1545
1546 if (r == SS_SUCCESS) {
1547 unsigned long tw = drbd_bm_total_weight(mdev);
1548 unsigned long now = jiffies;
1549 int i;
1550
1551 mdev->rs_failed = 0;
1552 mdev->rs_paused = 0;
1553 mdev->rs_same_csum = 0;
1554 mdev->rs_last_events = 0;
1555 mdev->rs_last_sect_ev = 0;
1556 mdev->rs_total = tw;
1557 mdev->rs_start = now;
1558 for (i = 0; i < DRBD_SYNC_MARKS; i++) {
1559 mdev->rs_mark_left[i] = tw;
1560 mdev->rs_mark_time[i] = now;
1561 }
1562 _drbd_pause_after(mdev);
1563 }
1564 write_unlock_irq(&global_state_lock);
1565
1566 if (r == SS_SUCCESS) {
1567 dev_info(DEV, "Began resync as %s (will sync %lu KB [%lu bits set]).\n",
1568 drbd_conn_str(ns.conn),
1569 (unsigned long) mdev->rs_total << (BM_BLOCK_SHIFT-10),
1570 (unsigned long) mdev->rs_total);
1571 if (side == C_SYNC_TARGET)
1572 mdev->bm_resync_fo = 0;
1573
1574 /* Since protocol 96, we must serialize drbd_gen_and_send_sync_uuid
1575 * with w_send_oos, or the sync target will get confused as to
1576 * how much bits to resync. We cannot do that always, because for an
1577 * empty resync and protocol < 95, we need to do it here, as we call
1578 * drbd_resync_finished from here in that case.
1579 * We drbd_gen_and_send_sync_uuid here for protocol < 96,
1580 * and from after_state_ch otherwise. */
1581 if (side == C_SYNC_SOURCE && mdev->agreed_pro_version < 96)
1582 drbd_gen_and_send_sync_uuid(mdev);
1583
1584 if (mdev->agreed_pro_version < 95 && mdev->rs_total == 0) {
1585 /* This still has a race (about when exactly the peers
1586 * detect connection loss) that can lead to a full sync
1587 * on next handshake. In 8.3.9 we fixed this with explicit
1588 * resync-finished notifications, but the fix
1589 * introduces a protocol change. Sleeping for some
1590 * time longer than the ping interval + timeout on the
1591 * SyncSource, to give the SyncTarget the chance to
1592 * detect connection loss, then waiting for a ping
1593 * response (implicit in drbd_resync_finished) reduces
1594 * the race considerably, but does not solve it. */
1595 if (side == C_SYNC_SOURCE)
1596 schedule_timeout_interruptible(
1597 mdev->net_conf->ping_int * HZ +
1598 mdev->net_conf->ping_timeo*HZ/9);
1599 drbd_resync_finished(mdev);
1600 }
1601
1602 drbd_rs_controller_reset(mdev);
1603 /* ns.conn may already be != mdev->state.conn,
1604 * we may have been paused in between, or become paused until
1605 * the timer triggers.
1606 * No matter, that is handled in resync_timer_fn() */
1607 if (ns.conn == C_SYNC_TARGET)
1608 mod_timer(&mdev->resync_timer, jiffies);
1609
1610 drbd_md_sync(mdev);
1611 }
1612 put_ldev(mdev);
1613 drbd_state_unlock(mdev);
1614 }
1615
drbd_worker(struct drbd_thread * thi)1616 int drbd_worker(struct drbd_thread *thi)
1617 {
1618 struct drbd_conf *mdev = thi->mdev;
1619 struct drbd_work *w = NULL;
1620 LIST_HEAD(work_list);
1621 int intr = 0, i;
1622
1623 sprintf(current->comm, "drbd%d_worker", mdev_to_minor(mdev));
1624
1625 while (get_t_state(thi) == Running) {
1626 drbd_thread_current_set_cpu(mdev);
1627
1628 if (down_trylock(&mdev->data.work.s)) {
1629 mutex_lock(&mdev->data.mutex);
1630 if (mdev->data.socket && !mdev->net_conf->no_cork)
1631 drbd_tcp_uncork(mdev->data.socket);
1632 mutex_unlock(&mdev->data.mutex);
1633
1634 intr = down_interruptible(&mdev->data.work.s);
1635
1636 mutex_lock(&mdev->data.mutex);
1637 if (mdev->data.socket && !mdev->net_conf->no_cork)
1638 drbd_tcp_cork(mdev->data.socket);
1639 mutex_unlock(&mdev->data.mutex);
1640 }
1641
1642 if (intr) {
1643 D_ASSERT(intr == -EINTR);
1644 flush_signals(current);
1645 ERR_IF (get_t_state(thi) == Running)
1646 continue;
1647 break;
1648 }
1649
1650 if (get_t_state(thi) != Running)
1651 break;
1652 /* With this break, we have done a down() but not consumed
1653 the entry from the list. The cleanup code takes care of
1654 this... */
1655
1656 w = NULL;
1657 spin_lock_irq(&mdev->data.work.q_lock);
1658 ERR_IF(list_empty(&mdev->data.work.q)) {
1659 /* something terribly wrong in our logic.
1660 * we were able to down() the semaphore,
1661 * but the list is empty... doh.
1662 *
1663 * what is the best thing to do now?
1664 * try again from scratch, restarting the receiver,
1665 * asender, whatnot? could break even more ugly,
1666 * e.g. when we are primary, but no good local data.
1667 *
1668 * I'll try to get away just starting over this loop.
1669 */
1670 spin_unlock_irq(&mdev->data.work.q_lock);
1671 continue;
1672 }
1673 w = list_entry(mdev->data.work.q.next, struct drbd_work, list);
1674 list_del_init(&w->list);
1675 spin_unlock_irq(&mdev->data.work.q_lock);
1676
1677 if (!w->cb(mdev, w, mdev->state.conn < C_CONNECTED)) {
1678 /* dev_warn(DEV, "worker: a callback failed! \n"); */
1679 if (mdev->state.conn >= C_CONNECTED)
1680 drbd_force_state(mdev,
1681 NS(conn, C_NETWORK_FAILURE));
1682 }
1683 }
1684 D_ASSERT(test_bit(DEVICE_DYING, &mdev->flags));
1685 D_ASSERT(test_bit(CONFIG_PENDING, &mdev->flags));
1686
1687 spin_lock_irq(&mdev->data.work.q_lock);
1688 i = 0;
1689 while (!list_empty(&mdev->data.work.q)) {
1690 list_splice_init(&mdev->data.work.q, &work_list);
1691 spin_unlock_irq(&mdev->data.work.q_lock);
1692
1693 while (!list_empty(&work_list)) {
1694 w = list_entry(work_list.next, struct drbd_work, list);
1695 list_del_init(&w->list);
1696 w->cb(mdev, w, 1);
1697 i++; /* dead debugging code */
1698 }
1699
1700 spin_lock_irq(&mdev->data.work.q_lock);
1701 }
1702 sema_init(&mdev->data.work.s, 0);
1703 /* DANGEROUS race: if someone did queue his work within the spinlock,
1704 * but up() ed outside the spinlock, we could get an up() on the
1705 * semaphore without corresponding list entry.
1706 * So don't do that.
1707 */
1708 spin_unlock_irq(&mdev->data.work.q_lock);
1709
1710 D_ASSERT(mdev->state.disk == D_DISKLESS && mdev->state.conn == C_STANDALONE);
1711 /* _drbd_set_state only uses stop_nowait.
1712 * wait here for the Exiting receiver. */
1713 drbd_thread_stop(&mdev->receiver);
1714 drbd_mdev_cleanup(mdev);
1715
1716 dev_info(DEV, "worker terminated\n");
1717
1718 clear_bit(DEVICE_DYING, &mdev->flags);
1719 clear_bit(CONFIG_PENDING, &mdev->flags);
1720 wake_up(&mdev->state_wait);
1721
1722 return 0;
1723 }
1724