aboutsummaryrefslogtreecommitdiff
path: root/drivers/block/drbd/drbd_req.c
blob: 11a75d32a2e27f0d78c26b511921780f4d336264 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
/*
   drbd_req.c

   This file is part of DRBD by Philipp Reisner and Lars Ellenberg.

   Copyright (C) 2001-2008, LINBIT Information Technologies GmbH.
   Copyright (C) 1999-2008, Philipp Reisner <philipp.reisner@linbit.com>.
   Copyright (C) 2002-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.

   drbd is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 2, or (at your option)
   any later version.

   drbd is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with drbd; see the file COPYING.  If not, write to
   the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.

 */

#include <linux/module.h>

#include <linux/slab.h>
#include <linux/drbd.h>
#include "drbd_int.h"
#include "drbd_req.h"


/* Update disk stats at start of I/O request */
static void _drbd_start_io_acct(struct drbd_conf *mdev, struct drbd_request *req, struct bio *bio)
{
	const int rw = bio_data_dir(bio);
	int cpu;
	cpu = part_stat_lock();
	part_stat_inc(cpu, &mdev->vdisk->part0, ios[rw]);
	part_stat_add(cpu, &mdev->vdisk->part0, sectors[rw], bio_sectors(bio));
	part_inc_in_flight(&mdev->vdisk->part0, rw);
	part_stat_unlock();
}

/* Update disk stats when completing request upwards */
static void _drbd_end_io_acct(struct drbd_conf *mdev, struct drbd_request *req)
{
	int rw = bio_data_dir(req->master_bio);
	unsigned long duration = jiffies - req->start_time;
	int cpu;
	cpu = part_stat_lock();
	part_stat_add(cpu, &mdev->vdisk->part0, ticks[rw], duration);
	part_round_stats(cpu, &mdev->vdisk->part0);
	part_dec_in_flight(&mdev->vdisk->part0, rw);
	part_stat_unlock();
}

static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const int rw)
{
	const unsigned long s = req->rq_state;

	/* remove it from the transfer log.
	 * well, only if it had been there in the first
	 * place... if it had not (local only or conflicting
	 * and never sent), it should still be "empty" as
	 * initialized in drbd_req_new(), so we can list_del() it
	 * here unconditionally */
	list_del(&req->tl_requests);

	/* if it was a write, we may have to set the corresponding
	 * bit(s) out-of-sync first. If it had a local part, we need to
	 * release the reference to the activity log. */
	if (rw == WRITE) {
		/* Set out-of-sync unless both OK flags are set
		 * (local only or remote failed).
		 * Other places where we set out-of-sync:
		 * READ with local io-error */
		if (!(s & RQ_NET_OK) || !(s & RQ_LOCAL_OK))
			drbd_set_out_of_sync(mdev, req->sector, req->size);

		if ((s & RQ_NET_OK) && (s & RQ_LOCAL_OK) && (s & RQ_NET_SIS))
			drbd_set_in_sync(mdev, req->sector, req->size);

		/* one might be tempted to move the drbd_al_complete_io
		 * to the local io completion callback drbd_endio_pri.
		 * but, if this was a mirror write, we may only
		 * drbd_al_complete_io after this is RQ_NET_DONE,
		 * otherwise the extent could be dropped from the al
		 * before it has actually been written on the peer.
		 * if we crash before our peer knows about the request,
		 * but after the extent has been dropped from the al,
		 * we would forget to resync the corresponding extent.
		 */
		if (s & RQ_LOCAL_MASK) {
			if (get_ldev_if_state(mdev, D_FAILED)) {
				if (s & RQ_IN_ACT_LOG)
					drbd_al_complete_io(mdev, req->sector);
				put_ldev(mdev);
			} else if (__ratelimit(&drbd_ratelimit_state)) {
				dev_warn(DEV, "Should have called drbd_al_complete_io(, %llu), "
				     "but my Disk seems to have failed :(\n",
				     (unsigned long long) req->sector);
			}
		}
	}

	drbd_req_free(req);
}

static void queue_barrier(struct drbd_conf *mdev)
{
	struct drbd_tl_epoch *b;

	/* We are within the req_lock. Once we queued the barrier for sending,
	 * we set the CREATE_BARRIER bit. It is cleared as soon as a new
	 * barrier/epoch object is added. This is the only place this bit is
	 * set. It indicates that the barrier for this epoch is already queued,
	 * and no new epoch has been created yet. */
	if (test_bit(CREATE_BARRIER, &mdev->flags))
		return;

	b = mdev->newest_tle;
	b->w.cb = w_send_barrier;
	/* inc_ap_pending done here, so we won't
	 * get imbalanced on connection loss.
	 * dec_ap_pending will be done in got_BarrierAck
	 * or (on connection loss) in tl_clear.  */
	inc_ap_pending(mdev);
	drbd_queue_work(&mdev->data.work, &b->w);
	set_bit(CREATE_BARRIER, &mdev->flags);
}

static void _about_to_complete_local_write(struct drbd_conf *mdev,
	struct drbd_request *req)
{
	const unsigned long s = req->rq_state;
	struct drbd_request *i;
	struct drbd_epoch_entry *e;
	struct hlist_node *n;
	struct hlist_head *slot;

	/* before we can signal completion to the upper layers,
	 * we may need to close the current epoch */
	if (mdev->state.conn >= C_CONNECTED &&
	    req->epoch == mdev->newest_tle->br_number)
		queue_barrier(mdev);

	/* we need to do the conflict detection stuff,
	 * if we have the ee_hash (two_primaries) and
	 * this has been on the network */
	if ((s & RQ_NET_DONE) && mdev->ee_hash != NULL) {
		const sector_t sector = req->sector;
		const int size = req->size;

		/* ASSERT:
		 * there must be no conflicting requests, since
		 * they must have been failed on the spot */
#define OVERLAPS overlaps(sector, size, i->sector, i->size)
		slot = tl_hash_slot(mdev, sector);
		hlist_for_each_entry(i, n, slot, colision) {
			if (OVERLAPS) {
				dev_alert(DEV, "LOGIC BUG: completed: %p %llus +%u; "
				      "other: %p %llus +%u\n",
				      req, (unsigned long long)sector, size,
				      i, (unsigned long long)i->sector, i->size);
			}
		}

		/* maybe "wake" those conflicting epoch entries
		 * that wait for this request to finish.
		 *
		 * currently, there can be only _one_ such ee
		 * (well, or some more, which would be pending
		 * P_DISCARD_ACK not yet sent by the asender...),
		 * since we block the receiver thread upon the
		 * first conflict detection, which will wait on
		 * misc_wait.  maybe we want to assert that?
		 *
		 * anyways, if we found one,
		 * we just have to do a wake_up.  */
#undef OVERLAPS
#define OVERLAPS overlaps(sector, size, e->sector, e->size)
		slot = ee_hash_slot(mdev, req->sector);
		hlist_for_each_entry(e, n, slot, colision) {
			if (OVERLAPS) {
				wake_up(&mdev->misc_wait);
				break;
			}
		}
	}
#undef OVERLAPS
}

void complete_master_bio(struct drbd_conf *mdev,
		struct bio_and_error *m)
{
	bio_endio(m->bio, m->error);
	dec_ap_bio(mdev);
}

/* Helper for __req_mod().
 * Set m->bio to the master bio, if it is fit to be completed,
 * or leave it alone (it is initialized to NULL in __req_mod),
 * if it has already been completed, or cannot be completed yet.
 * If m->bio is set, the error status to be returned is placed in m->error.
 */
void _req_may_be_done(struct drbd_request *req, struct bio_and_error *m)
{
	const unsigned long s = req->rq_state;
	struct drbd_conf *mdev = req->mdev;
	/* only WRITES may end up here without a master bio (on barrier ack) */
	int rw = req->master_bio ? bio_data_dir(req->master_bio) : WRITE;

	/* we must not complete the master bio, while it is
	 *	still being processed by _drbd_send_zc_bio (drbd_send_dblock)
	 *	not yet acknowledged by the peer
	 *	not yet completed by the local io subsystem
	 * these flags may get cleared in any order by
	 *	the worker,
	 *	the receiver,
	 *	the bio_endio completion callbacks.
	 */
	if (s & RQ_NET_QUEUED)
		return;
	if (s & RQ_NET_PENDING)
		return;
	if (s & RQ_LOCAL_PENDING)
		return;

	if (req->master_bio) {
		/* this is data_received (remote read)
		 * or protocol C P_WRITE_ACK
		 * or protocol B P_RECV_ACK
		 * or protocol A "handed_over_to_network" (SendAck)
		 * or canceled or failed,
		 * or killed from the transfer log due to connection loss.
		 */

		/*
		 * figure out whether to report success or failure.
		 *
		 * report success when at least one of the operations succeeded.
		 * or, to put the other way,
		 * only report failure, when both operations failed.
		 *
		 * what to do about the failures is handled elsewhere.
		 * what we need to do here is just: complete the master_bio.
		 *
		 * local completion error, if any, has been stored as ERR_PTR
		 * in private_bio within drbd_endio_pri.
		 */
		int ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK);
		int error = PTR_ERR(req->private_bio);

		/* remove the request from the conflict detection
		 * respective block_id verification hash */
		if (!hlist_unhashed(&req->colision))
			hlist_del(&req->colision);
		else
			D_ASSERT((s & (RQ_NET_MASK & ~RQ_NET_DONE)) == 0);

		/* for writes we need to do some extra housekeeping */
		if (rw == WRITE)
			_about_to_complete_local_write(mdev, req);

		/* Update disk stats */
		_drbd_end_io_acct(mdev, req);

		m->error = ok ? 0 : (error ?: -EIO);
		m->bio = req->master_bio;
		req->master_bio = NULL;
	}

	if ((s & RQ_NET_MASK) == 0 || (s & RQ_NET_DONE)) {
		/* this is disconnected (local only) operation,
		 * or protocol C P_WRITE_ACK,
		 * or protocol A or B P_BARRIER_ACK,
		 * or killed from the transfer log due to connection loss. */
		_req_is_done(mdev, req, rw);
	}
	/* else: network part and not DONE yet. that is
	 * protocol A or B, barrier ack still pending... */
}

static void _req_may_be_done_not_susp(struct drbd_request *req, struct bio_and_error *m)
{
	struct drbd_conf *mdev = req->mdev;

	if (!is_susp(mdev->state))
		_req_may_be_done(req, m);
}

/*
 * checks whether there was an overlapping request
 * or ee already registered.
 *
 * if so, return 1, in which case this request is completed on the spot,
 * without ever being submitted or send.
 *
 * return 0 if it is ok to submit this request.
 *
 * NOTE:
 * paranoia: assume something above us is broken, and issues different write
 * requests for the same block simultaneously...
 *
 * To ensure these won't be reordered differently on both nodes, resulting in
 * diverging data sets, we discard the later one(s). Not that this is supposed
 * to happen, but this is the rationale why we also have to check for
 * conflicting requests with local origin, and why we have to do so regardless
 * of whether we allowed multiple primaries.
 *
 * BTW, in case we only have one primary, the ee_hash is empty anyways, and the
 * second hlist_for_each_entry becomes a noop. This is even simpler than to
 * grab a reference on the net_conf, and check for the two_primaries flag...
 */
static int _req_conflicts(struct drbd_request *req)
{
	struct drbd_conf *mdev = req->mdev;
	const sector_t sector = req->sector;
	const int size = req->size;
	struct drbd_request *i;
	struct drbd_epoch_entry *e;
	struct hlist_node *n;
	struct hlist_head *slot;

	D_ASSERT(hlist_unhashed(&req->colision));

	if (!get_net_conf(mdev))
		return 0;

	/* BUG_ON */
	ERR_IF (mdev->tl_hash_s == 0)
		goto out_no_conflict;
	BUG_ON(mdev->tl_hash == NULL);

#define OVERLAPS overlaps(i->sector, i->size, sector, size)
	slot = tl_hash_slot(mdev, sector);
	hlist_for_each_entry(i, n, slot, colision) {
		if (OVERLAPS) {
			dev_alert(DEV, "%s[%u] Concurrent local write detected! "
			      "[DISCARD L] new: %llus +%u; "
			      "pending: %llus +%u\n",
			      current->comm, current->pid,
			      (unsigned long long)sector, size,
			      (unsigned long long)i->sector, i->size);
			goto out_conflict;
		}
	}

	if (mdev->ee_hash_s) {
		/* now, check for overlapping requests with remote origin */
		BUG_ON(mdev->ee_hash == NULL);
#undef OVERLAPS
#define OVERLAPS overlaps(e->sector, e->size, sector, size)
		slot = ee_hash_slot(mdev, sector);
		hlist_for_each_entry(e, n, slot, colision) {
			if (OVERLAPS) {
				dev_alert(DEV, "%s[%u] Concurrent remote write detected!"
				      " [DISCARD L] new: %llus +%u; "
				      "pending: %llus +%u\n",
				      current->comm, current->pid,
				      (unsigned long long)sector, size,
				      (unsigned long long)e->sector, e->size);
				goto out_conflict;
			}
		}
	}
#undef OVERLAPS

out_no_conflict:
	/* this is like it should be, and what we expected.
	 * our users do behave after all... */
	put_net_conf(mdev);
	return 0;

out_conflict:
	put_net_conf(mdev);
	return 1;
}

/* obviously this could be coded as many single functions
 * instead of one huge switch,
 * or by putting the code directly in the respective locations
 * (as it has been before).
 *
 * but having it this way
 *  enforces that it is all in this one place, where it is easier to audit,
 *  it makes it obvious that whatever "event" "happens" to a request should
 *  happen "atomically" within the req_lock,
 *  and it enforces that we have to think in a very structured manner
 *  about the "events" that may happen to a request during its life time ...
 */
int __req_mod(struct drbd_request *req, enum drbd_req_event what,
		struct bio_and_error *m)
{
	struct drbd_conf *mdev = req->mdev;
	int rv = 0;
	m->bio = NULL;

	switch (what) {
	default:
		dev_err(DEV, "LOGIC BUG in %s:%u\n", __FILE__ , __LINE__);
		break;

	/* does not happen...
	 * initialization done in drbd_req_new
	case created:
		break;
		*/

	case to_be_send: /* via network */
		/* reached via drbd_make_request_common
		 * and from w_read_retry_remote */
		D_ASSERT(!(req->rq_state & RQ_NET_MASK));
		req->rq_state |= RQ_NET_PENDING;
		inc_ap_pending(mdev);
		break;

	case to_be_submitted: /* locally */
		/* reached via drbd_make_request_common */
		D_ASSERT(!(req->rq_state & RQ_LOCAL_MASK));
		req->rq_state |= RQ_LOCAL_PENDING;
		break;

	case completed_ok:
		if (bio_data_dir(req->master_bio) == WRITE)
			mdev->writ_cnt += req->size>>9;
		else
			mdev->read_cnt += req->size>>9;

		req->rq_state |= (RQ_LOCAL_COMPLETED|RQ_LOCAL_OK);
		req->rq_state &= ~RQ_LOCAL_PENDING;

		_req_may_be_done_not_susp(req, m);
		put_ldev(mdev);
		break;

	case write_completed_with_error:
		req->rq_state |= RQ_LOCAL_COMPLETED;
		req->rq_state &= ~RQ_LOCAL_PENDING;

		__drbd_chk_io_error(mdev, FALSE);
		_req_may_be_done_not_susp(req, m);
		put_ldev(mdev);
		break;

	case read_ahead_completed_with_error:
		/* it is legal to fail READA */
		req->rq_state |= RQ_LOCAL_COMPLETED;
		req->rq_state &= ~RQ_LOCAL_PENDING;
		_req_may_be_done_not_susp(req, m);
		put_ldev(mdev);
		break;

	case read_completed_with_error:
		drbd_set_out_of_sync(mdev, req->sector, req->size);

		req->rq_state |= RQ_LOCAL_COMPLETED;
		req->rq_state &= ~RQ_LOCAL_PENDING;

		D_ASSERT(!(req->rq_state & RQ_NET_MASK));

		__drbd_chk_io_error(mdev, FALSE);
		put_ldev(mdev);

		/* no point in retrying if there is no good remote data,
		 * or we have no connection. */
		if (mdev->state.pdsk != D_UP_TO_DATE) {
			_req_may_be_done_not_susp(req, m);
			break;
		}

		/* _req_mod(req,to_be_send); oops, recursion... */
		req->rq_state |= RQ_NET_PENDING;
		inc_ap_pending(mdev);
		/* fall through: _req_mod(req,queue_for_net_read); */

	case queue_for_net_read:
		/* READ or READA, and
		 * no local disk,
		 * or target area marked as invalid,
		 * or just got an io-error. */
		/* from drbd_make_request_common
		 * or from bio_endio during read io-error recovery */

		/* so we can verify the handle in the answer packet
		 * corresponding hlist_del is in _req_may_be_done() */
		hlist_add_head(&req->colision, ar_hash_slot(mdev, req->sector));

		set_bit(UNPLUG_REMOTE, &mdev->flags);

		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		req->rq_state |= RQ_NET_QUEUED;
		req->w.cb = (req->rq_state & RQ_LOCAL_MASK)
			? w_read_retry_remote
			: w_send_read_req;
		drbd_queue_work(&mdev->data.work, &req->w);
		break;

	case queue_for_net_write:
		/* assert something? */
		/* from drbd_make_request_common only */

		hlist_add_head(&req->colision, tl_hash_slot(mdev, req->sector));
		/* corresponding hlist_del is in _req_may_be_done() */

		/* NOTE
		 * In case the req ended up on the transfer log before being
		 * queued on the worker, it could lead to this request being
		 * missed during cleanup after connection loss.
		 * So we have to do both operations here,
		 * within the same lock that protects the transfer log.
		 *
		 * _req_add_to_epoch(req); this has to be after the
		 * _maybe_start_new_epoch(req); which happened in
		 * drbd_make_request_common, because we now may set the bit
		 * again ourselves to close the current epoch.
		 *
		 * Add req to the (now) current epoch (barrier). */

		/* otherwise we may lose an unplug, which may cause some remote
		 * io-scheduler timeout to expire, increasing maximum latency,
		 * hurting performance. */
		set_bit(UNPLUG_REMOTE, &mdev->flags);

		/* see drbd_make_request_common,
		 * just after it grabs the req_lock */
		D_ASSERT(test_bit(CREATE_BARRIER, &mdev->flags) == 0);

		req->epoch = mdev->newest_tle->br_number;

		/* increment size of current epoch */
		mdev->newest_tle->n_writes++;

		/* queue work item to send data */
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		req->rq_state |= RQ_NET_QUEUED;
		req->w.cb =  w_send_dblock;
		drbd_queue_work(&mdev->data.work, &req->w);

		/* close the epoch, in case it outgrew the limit */
		if (mdev->newest_tle->n_writes >= mdev->net_conf->max_epoch_size)
			queue_barrier(mdev);

		break;

	case send_canceled:
		/* treat it the same */
	case send_failed:
		/* real cleanup will be done from tl_clear.  just update flags
		 * so it is no longer marked as on the worker queue */
		req->rq_state &= ~RQ_NET_QUEUED;
		/* if we did it right, tl_clear should be scheduled only after
		 * this, so this should not be necessary! */
		_req_may_be_done_not_susp(req, m);
		break;

	case handed_over_to_network:
		/* assert something? */
		if (bio_data_dir(req->master_bio) == WRITE &&
		    mdev->net_conf->wire_protocol == DRBD_PROT_A) {
			/* this is what is dangerous about protocol A:
			 * pretend it was successfully written on the peer. */
			if (req->rq_state & RQ_NET_PENDING) {
				dec_ap_pending(mdev);
				req->rq_state &= ~RQ_NET_PENDING;
				req->rq_state |= RQ_NET_OK;
			} /* else: neg-ack was faster... */
			/* it is still not yet RQ_NET_DONE until the
			 * corresponding epoch barrier got acked as well,
			 * so we know what to dirty on connection loss */
		}
		req->rq_state &= ~RQ_NET_QUEUED;
		req->rq_state |= RQ_NET_SENT;
		/* because _drbd_send_zc_bio could sleep, and may want to
		 * dereference the bio even after the "write_acked_by_peer" and
		 * "completed_ok" events came in, once we return from
		 * _drbd_send_zc_bio (drbd_send_dblock), we have to check
		 * whether it is done already, and end it.  */
		_req_may_be_done_not_susp(req, m);
		break;

	case read_retry_remote_canceled:
		req->rq_state &= ~RQ_NET_QUEUED;
		/* fall through, in case we raced with drbd_disconnect */
	case connection_lost_while_pending:
		/* transfer log cleanup after connection loss */
		/* assert something? */
		if (req->rq_state & RQ_NET_PENDING)
			dec_ap_pending(mdev);
		req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING);
		req->rq_state |= RQ_NET_DONE;
		/* if it is still queued, we may not complete it here.
		 * it will be canceled soon. */
		if (!(req->rq_state & RQ_NET_QUEUED))
			_req_may_be_done(req, m); /* Allowed while state.susp */
		break;

	case write_acked_by_peer_and_sis:
		req->rq_state |= RQ_NET_SIS;
	case conflict_discarded_by_peer:
		/* for discarded conflicting writes of multiple primaries,
		 * there is no need to keep anything in the tl, potential
		 * node crashes are covered by the activity log. */
		if (what == conflict_discarded_by_peer)
			dev_alert(DEV, "Got DiscardAck packet %llus +%u!"
			      " DRBD is not a random data generator!\n",
			      (unsigned long long)req->sector, req->size);
		req->rq_state |= RQ_NET_DONE;
		/* fall through */
	case write_acked_by_peer:
		/* protocol C; successfully written on peer.
		 * Nothing to do here.
		 * We want to keep the tl in place for all protocols, to cater
		 * for volatile write-back caches on lower level devices.
		 *
		 * A barrier request is expected to have forced all prior
		 * requests onto stable storage, so completion of a barrier
		 * request could set NET_DONE right here, and not wait for the
		 * P_BARRIER_ACK, but that is an unnecessary optimization. */

		/* this makes it effectively the same as for: */
	case recv_acked_by_peer:
		/* protocol B; pretends to be successfully written on peer.
		 * see also notes above in handed_over_to_network about
		 * protocol != C */
		req->rq_state |= RQ_NET_OK;
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		dec_ap_pending(mdev);
		req->rq_state &= ~RQ_NET_PENDING;
		_req_may_be_done_not_susp(req, m);
		break;

	case neg_acked:
		/* assert something? */
		if (req->rq_state & RQ_NET_PENDING)
			dec_ap_pending(mdev);
		req->rq_state &= ~(RQ_NET_OK|RQ_NET_PENDING);

		req->rq_state |= RQ_NET_DONE;
		_req_may_be_done_not_susp(req, m);
		/* else: done by handed_over_to_network */
		break;

	case fail_frozen_disk_io:
		if (!(req->rq_state & RQ_LOCAL_COMPLETED))
			break;

		_req_may_be_done(req, m); /* Allowed while state.susp */
		break;

	case restart_frozen_disk_io:
		if (!(req->rq_state & RQ_LOCAL_COMPLETED))
			break;

		req->rq_state &= ~RQ_LOCAL_COMPLETED;

		rv = MR_READ;
		if (bio_data_dir(req->master_bio) == WRITE)
			rv = MR_WRITE;

		get_ldev(mdev);
		req->w.cb = w_restart_disk_io;
		drbd_queue_work(&mdev->data.work, &req->w);
		break;

	case resend:
		/* If RQ_NET_OK is already set, we got a P_WRITE_ACK or P_RECV_ACK
		   before the connection loss (B&C only); only P_BARRIER_ACK was missing.
		   Trowing them out of the TL here by pretending we got a BARRIER_ACK
		   We ensure that the peer was not rebooted */
		if (!(req->rq_state & RQ_NET_OK)) {
			if (req->w.cb) {
				drbd_queue_work(&mdev->data.work, &req->w);
				rv = req->rq_state & RQ_WRITE ? MR_WRITE : MR_READ;
			}
			break;
		}
		/* else, fall through to barrier_acked */

	case barrier_acked:
		if (!(req->rq_state & RQ_WRITE))
			break;

		if (req->rq_state & RQ_NET_PENDING) {
			/* barrier came in before all requests have been acked.
			 * this is bad, because if the connection is lost now,
			 * we won't be able to clean them up... */
			dev_err(DEV, "FIXME (barrier_acked but pending)\n");
			list_move(&req->tl_requests, &mdev->out_of_sequence_requests);
		}
		D_ASSERT(req->rq_state & RQ_NET_SENT);
		req->rq_state |= RQ_NET_DONE;
		_req_may_be_done(req, m); /* Allowed while state.susp */
		break;

	case data_received:
		D_ASSERT(req->rq_state & RQ_NET_PENDING);
		dec_ap_pending(mdev);
		req->rq_state &= ~RQ_NET_PENDING;
		req->rq_state |= (RQ_NET_OK|RQ_NET_DONE);
		_req_may_be_done_not_susp(req, m);
		break;
	};

	return rv;
}

/* we may do a local read if:
 * - we are consistent (of course),
 * - or we are generally inconsistent,
 *   BUT we are still/already IN SYNC for this area.
 *   since size may be bigger than BM_BLOCK_SIZE,
 *   we may need to check several bits.
 */
static int drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int size)
{
	unsigned long sbnr, ebnr;
	sector_t esector, nr_sectors;

	if (mdev->state.disk == D_UP_TO_DATE)
		return 1;
	if (mdev->state.disk >= D_OUTDATED)
		return 0;
	if (mdev->state.disk <  D_INCONSISTENT)
		return 0;
	/* state.disk == D_INCONSISTENT   We will have a look at the BitMap */
	nr_sectors = drbd_get_capacity(mdev->this_bdev);
	esector = sector + (size >> 9) - 1;

	D_ASSERT(sector  < nr_sectors);
	D_ASSERT(esector < nr_sectors);

	sbnr = BM_SECT_TO_BIT(sector);
	ebnr = BM_SECT_TO_BIT(esector);

	return 0 == drbd_bm_count_bits(mdev, sbnr, ebnr);
}

static int drbd_make_request_common(struct drbd_conf *mdev, struct bio *bio)
{
	const int rw = bio_rw(bio);
	const int size = bio->bi_size;
	const sector_t sector = bio->bi_sector;
	struct drbd_tl_epoch *b = NULL;
	struct drbd_request *req;
	int local, remote;
	int err = -EIO;
	int ret = 0;

	/* allocate outside of all locks; */
	req = drbd_req_new(mdev, bio);
	if (!req) {
		dec_ap_bio(mdev);
		/* only pass the error to the upper layers.
		 * if user cannot handle io errors, that's not our business. */
		dev_err(DEV, "could not kmalloc() req\n");
		bio_endio(bio, -ENOMEM);
		return 0;
	}

	local = get_ldev(mdev);
	if (!local) {
		bio_put(req->private_bio); /* or we get a bio leak */
		req->private_bio = NULL;
	}
	if (rw == WRITE) {
		remote = 1;
	} else {
		/* READ || READA */
		if (local) {
			if (!drbd_may_do_local_read(mdev, sector, size)) {
				/* we could kick the syncer to
				 * sync this extent asap, wait for
				 * it, then continue locally.
				 * Or just issue the request remotely.
				 */
				local = 0;
				bio_put(req->private_bio);
				req->private_bio = NULL;
				put_ldev(mdev);
			}
		}
		remote = !local && mdev->state.pdsk >= D_UP_TO_DATE;
	}

	/* If we have a disk, but a READA request is mapped to remote,
	 * we are R_PRIMARY, D_INCONSISTENT, SyncTarget.
	 * Just fail that READA request right here.
	 *
	 * THINK: maybe fail all READA when not local?
	 *        or make this configurable...
	 *        if network is slow, READA won't do any good.
	 */
	if (rw == READA && mdev->state.disk >= D_INCONSISTENT && !local) {
		err = -EWOULDBLOCK;
		goto fail_and_free_req;
	}

	/* For WRITES going to the local disk, grab a reference on the target
	 * extent.  This waits for any resync activity in the corresponding
	 * resync extent to finish, and, if necessary, pulls in the target
	 * extent into the activity log, which involves further disk io because
	 * of transactional on-disk meta data updates. */
	if (rw == WRITE && local && !test_bit(AL_SUSPENDED, &mdev->flags)) {
		req->rq_state |= RQ_IN_ACT_LOG;
		drbd_al_begin_io(mdev, sector);
	}

	remote = remote && (mdev->state.pdsk == D_UP_TO_DATE ||
			    (mdev->state.pdsk == D_INCONSISTENT &&
			     mdev->state.conn >= C_CONNECTED));

	if (!(local || remote) && !is_susp(mdev->state)) {
		if (__ratelimit(&drbd_ratelimit_state))
			dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
		goto fail_free_complete;
	}

	/* For WRITE request, we have to make sure that we have an
	 * unused_spare_tle, in case we need to start a new epoch.
	 * I try to be smart and avoid to pre-allocate always "just in case",
	 * but there is a race between testing the bit and pointer outside the
	 * spinlock, and grabbing the spinlock.
	 * if we lost that race, we retry.  */
	if (rw == WRITE && remote &&
	    mdev->unused_spare_tle == NULL &&
	    test_bit(CREATE_BARRIER, &mdev->flags)) {
allocate_barrier:
		b = kmalloc(sizeof(struct drbd_tl_epoch), GFP_NOIO);
		if (!b) {
			dev_err(DEV, "Failed to alloc barrier.\n");
			err = -ENOMEM;
			goto fail_free_complete;
		}
	}

	/* GOOD, everything prepared, grab the spin_lock */
	spin_lock_irq(&mdev->req_lock);

	if (is_susp(mdev->state)) {
		/* If we got suspended, use the retry mechanism of
		   generic_make_request() to restart processing of this
		   bio. In the next call to drbd_make_request_26
		   we sleep in inc_ap_bio() */
		ret = 1;
		spin_unlock_irq(&mdev->req_lock);
		goto fail_free_complete;
	}

	if (remote) {
		remote = (mdev->state.pdsk == D_UP_TO_DATE ||
			    (mdev->state.pdsk == D_INCONSISTENT &&
			     mdev->state.conn >= C_CONNECTED));
		if (!remote)
			dev_warn(DEV, "lost connection while grabbing the req_lock!\n");
		if (!(local || remote)) {
			dev_err(DEV, "IO ERROR: neither local nor remote disk\n");
			spin_unlock_irq(&mdev->req_lock);
			goto fail_free_complete;
		}
	}

	if (b && mdev->unused_spare_tle == NULL) {
		mdev->unused_spare_tle = b;
		b = NULL;
	}
	if (rw == WRITE && remote &&
	    mdev->unused_spare_tle == NULL &&
	    test_bit(CREATE_BARRIER, &mdev->flags)) {
		/* someone closed the current epoch
		 * while we were grabbing the spinlock */
		spin_unlock_irq(&mdev->req_lock);
		goto allocate_barrier;
	}


	/* Update disk stats */
	_drbd_start_io_acct(mdev, req, bio);

	/* _maybe_start_new_epoch(mdev);
	 * If we need to generate a write barrier packet, we have to add the
	 * new epoch (barrier) object, and queue the barrier packet for sending,
	 * and queue the req's data after it _within the same lock_, otherwise
	 * we have race conditions were the reorder domains could be mixed up.
	 *
	 * Even read requests may start a new epoch and queue the corresponding
	 * barrier packet.  To get the write ordering right, we only have to
	 * make sure that, if this is a write request and it triggered a
	 * barrier packet, this request is queued within the same spinlock. */
	if (remote && mdev->unused_spare_tle &&
	    test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) {
		_tl_add_barrier(mdev, mdev->unused_spare_tle);
		mdev->unused_spare_tle = NULL;
	} else {
		D_ASSERT(!(remote && rw == WRITE &&
			   test_bit(CREATE_BARRIER, &mdev->flags)));
	}

	/* NOTE
	 * Actually, 'local' may be wrong here already, since we may have failed
	 * to write to the meta data, and may become wrong anytime because of
	 * local io-error for some other request, which would lead to us
	 * "detaching" the local disk.
	 *
	 * 'remote' may become wrong any time because the network could fail.
	 *
	 * This is a harmless race condition, though, since it is handled
	 * correctly at the appropriate places; so it just defers the failure
	 * of the respective operation.
	 */

	/* mark them early for readability.
	 * this just sets some state flags. */
	if (remote)
		_req_mod(req, to_be_send);
	if (local)
		_req_mod(req, to_be_submitted);

	/* check this request on the collision detection hash tables.
	 * if we have a conflict, just complete it here.
	 * THINK do we want to check reads, too? (I don't think so...) */
	if (rw == WRITE && _req_conflicts(req))
		goto fail_conflicting;

	list_add_tail(&req->tl_requests, &mdev->newest_tle->requests);

	/* NOTE remote first: to get the concurrent write detection right,
	 * we must register the request before start of local IO.  */
	if (remote) {
		/* either WRITE and C_CONNECTED,
		 * or READ, and no local disk,
		 * or READ, but not in sync.
		 */
		_req_mod(req, (rw == WRITE)
				? queue_for_net_write
				: queue_for_net_read);
	}
	spin_unlock_irq(&mdev->req_lock);
	kfree(b); /* if someone else has beaten us to it... */

	if (local) {
		req->private_bio->bi_bdev = mdev->ldev->backing_bdev;

		/* State may have changed since we grabbed our reference on the
		 * mdev->ldev member. Double check, and short-circuit to endio.
		 * In case the last activity log transaction failed to get on
		 * stable storage, and this is a WRITE, we may not even submit
		 * this bio. */
		if (get_ldev(mdev)) {
			if (FAULT_ACTIVE(mdev, rw == WRITE ? DRBD_FAULT_DT_WR
					     : rw == READ  ? DRBD_FAULT_DT_RD
					     :               DRBD_FAULT_DT_RA))
				bio_endio(req->private_bio, -EIO);
			else
				generic_make_request(req->private_bio);
			put_ldev(mdev);
		} else
			bio_endio(req->private_bio, -EIO);
	}

	/* we need to plug ALWAYS since we possibly need to kick lo_dev.
	 * we plug after submit, so we won't miss an unplug event */
	drbd_plug_device(mdev);

	return 0;

fail_conflicting:
	/* this is a conflicting request.
	 * even though it may have been only _partially_
	 * overlapping with one of the currently pending requests,
	 * without even submitting or sending it, we will
	 * pretend that it was successfully served right now.
	 */
	_drbd_end_io_acct(mdev, req);
	spin_unlock_irq(&mdev->req_lock);
	if (remote)
		dec_ap_pending(mdev);
	/* THINK: do we want to fail it (-EIO), or pretend success?
	 * this pretends success. */
	err = 0;

fail_free_complete:
	if (rw == WRITE && local)
		drbd_al_complete_io(mdev, sector);
fail_and_free_req:
	if (local) {
		bio_put(req->private_bio);
		req->private_bio = NULL;
		put_ldev(mdev);
	}
	if (!ret)
		bio_endio(bio, err);

	drbd_req_free(req);
	dec_ap_bio(mdev);
	kfree(b);

	return ret;
}

/* helper function for drbd_make_request
 * if we can determine just by the mdev (state) that this request will fail,
 * return 1
 * otherwise return 0
 */
static int drbd_fail_request_early(struct drbd_conf *mdev, int is_write)
{
	if (mdev->state.role != R_PRIMARY &&
		(!allow_oos || is_write)) {
		if (__ratelimit(&drbd_ratelimit_state)) {
			dev_err(DEV, "Process %s[%u] tried to %s; "
			    "since we are not in Primary state, "
			    "we cannot allow this\n",
			    current->comm, current->pid,
			    is_write ? "WRITE" : "READ");
		}
		return 1;
	}

	return 0;
}

int drbd_make_request_26(struct request_queue *q, struct bio *bio)
{
	unsigned int s_enr, e_enr;
	struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata;

	if (drbd_fail_request_early(mdev, bio_data_dir(bio) & WRITE)) {
		bio_endio(bio, -EPERM);
		return 0;
	}

	/*
	 * what we "blindly" assume:
	 */
	D_ASSERT(bio->bi_size > 0);
	D_ASSERT((bio->bi_size & 0x1ff) == 0);
	D_ASSERT(bio->bi_idx == 0);

	/* to make some things easier, force alignment of requests within the
	 * granularity of our hash tables */
	s_enr = bio->bi_sector >> HT_SHIFT;
	e_enr = (bio->bi_sector+(bio->bi_size>>9)-1) >> HT_SHIFT;

	if (likely(s_enr == e_enr)) {
		inc_ap_bio(mdev, 1);
		return drbd_make_request_common(mdev, bio);
	}

	/* can this bio be split generically?
	 * Maybe add our own split-arbitrary-bios function. */
	if (bio->bi_vcnt != 1 || bio->bi_idx != 0 || bio->bi_size > DRBD_MAX_SEGMENT_SIZE) {
		/* rather error out here than BUG in bio_split */
		dev_err(DEV, "bio would need to, but cannot, be split: "
		    "(vcnt=%u,idx=%u,size=%u,sector=%llu)\n",
		    bio->bi_vcnt, bio->bi_idx, bio->bi_size,
		    (unsigned long long)bio->bi_sector);
		bio_endio(bio, -EINVAL);
	} else {
		/* This bio crosses some boundary, so we have to split it. */
		struct bio_pair *bp;
		/* works for the "do not cross hash slot boundaries" case
		 * e.g. sector 262269, size 4096
		 * s_enr = 262269 >> 6 = 4097
		 * e_enr = (262269+8-1) >> 6 = 4098
		 * HT_SHIFT = 6
		 * sps = 64, mask = 63
		 * first_sectors = 64 - (262269 & 63) = 3
		 */
		const sector_t sect = bio->bi_sector;
		const int sps = 1 << HT_SHIFT; /* sectors per slot */
		const int mask = sps - 1;
		const sector_t first_sectors = sps - (sect & mask);
		bp = bio_split(bio,
#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
				bio_split_pool,
#endif
				first_sectors);

		/* we need to get a "reference count" (ap_bio_cnt)
		 * to avoid races with the disconnect/reconnect/suspend code.
		 * In case we need to split the bio here, we need to get three references
		 * atomically, otherwise we might deadlock when trying to submit the
		 * second one! */
		inc_ap_bio(mdev, 3);

		D_ASSERT(e_enr == s_enr + 1);

		while (drbd_make_request_common(mdev, &bp->bio1))
			inc_ap_bio(mdev, 1);

		while (drbd_make_request_common(mdev, &bp->bio2))
			inc_ap_bio(mdev, 1);

		dec_ap_bio(mdev);

		bio_pair_release(bp);
	}
	return 0;
}

/* This is called by bio_add_page().  With this function we reduce
 * the number of BIOs that span over multiple DRBD_MAX_SEGMENT_SIZEs
 * units (was AL_EXTENTs).
 *
 * we do the calculation within the lower 32bit of the byte offsets,
 * since we don't care for actual offset, but only check whether it
 * would cross "activity log extent" boundaries.
 *
 * As long as the BIO is empty we have to allow at least one bvec,
 * regardless of size and offset.  so the resulting bio may still
 * cross extent boundaries.  those are dealt with (bio_split) in
 * drbd_make_request_26.
 */
int drbd_merge_bvec(struct request_queue *q, struct bvec_merge_data *bvm, struct bio_vec *bvec)
{
	struct drbd_conf *mdev = (struct drbd_conf *) q->queuedata;
	unsigned int bio_offset =
		(unsigned int)bvm->bi_sector << 9; /* 32 bit */
	unsigned int bio_size = bvm->bi_size;
	int limit, backing_limit;

	limit = DRBD_MAX_SEGMENT_SIZE
	      - ((bio_offset & (DRBD_MAX_SEGMENT_SIZE-1)) + bio_size);
	if (limit < 0)
		limit = 0;
	if (bio_size == 0) {
		if (limit <= bvec->bv_len)
			limit = bvec->bv_len;
	} else if (limit && get_ldev(mdev)) {
		struct request_queue * const b =
			mdev->ldev->backing_bdev->bd_disk->queue;
		if (b->merge_bvec_fn) {
			backing_limit = b->merge_bvec_fn(b, bvm, bvec);
			limit = min(limit, backing_limit);
		}
		put_ldev(mdev);
	}
	return limit;
}