summaryrefslogtreecommitdiff
path: root/src/mongo/db/concurrency/lock_state.cpp
blob: ea2b9b9c63ad94f5972ee0c8fdc91ea6ee0a3d44 (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
/**
 *    Copyright (C) 2014 MongoDB Inc.
 *
 *    This program is free software: you can redistribute it and/or  modify
 *    it under the terms of the GNU Affero General Public License, version 3,
 *    as published by the Free Software Foundation.
 *
 *    This program 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 Affero General Public License for more details.
 *
 *    You should have received a copy of the GNU Affero General Public License
 *    along with this program.  If not, see <http://www.gnu.org/licenses/>.
 *
 *    As a special exception, the copyright holders give permission to link the
 *    code of portions of this program with the OpenSSL library under certain
 *    conditions as described in each individual source file and distribute
 *    linked combinations including the program with the OpenSSL library. You
 *    must comply with the GNU Affero General Public License in all respects for
 *    all of the code used other than as permitted herein. If you modify file(s)
 *    with this exception, you may extend this exception to your version of the
 *    file(s), but you are not obligated to do so. If you do not wish to do so,
 *    delete this exception statement from your version. If you delete this
 *    exception statement from all source files in the program, then also delete
 *    it in the license file.
 */

#define MONGO_LOG_DEFAULT_COMPONENT ::mongo::logger::LogComponent::kDefault

#include "mongo/platform/basic.h"

#include "mongo/db/concurrency/lock_state.h"

#include <vector>

#include "mongo/db/service_context.h"
#include "mongo/db/namespace_string.h"
#include "mongo/platform/compiler.h"
#include "mongo/util/background.h"
#include "mongo/util/concurrency/synchronization.h"
#include "mongo/util/debug_util.h"
#include "mongo/util/log.h"
#include "mongo/util/mongoutils/str.h"

namespace mongo {
namespace {

/**
 * Partitioned global lock statistics, so we don't hit the same bucket.
 */
class PartitionedInstanceWideLockStats {
    MONGO_DISALLOW_COPYING(PartitionedInstanceWideLockStats);

public:
    PartitionedInstanceWideLockStats() {}

    void recordAcquisition(LockerId id, ResourceId resId, LockMode mode) {
        _get(id).recordAcquisition(resId, mode);
    }

    void recordWait(LockerId id, ResourceId resId, LockMode mode) {
        _get(id).recordWait(resId, mode);
    }

    void recordWaitTime(LockerId id, ResourceId resId, LockMode mode, uint64_t waitMicros) {
        _get(id).recordWaitTime(resId, mode, waitMicros);
    }

    void recordDeadlock(ResourceId resId, LockMode mode) {
        _get(resId).recordDeadlock(resId, mode);
    }

    void report(SingleThreadedLockStats* outStats) const {
        for (int i = 0; i < NumPartitions; i++) {
            outStats->append(_partitions[i].stats);
        }
    }

    void reset() {
        for (int i = 0; i < NumPartitions; i++) {
            _partitions[i].stats.reset();
        }
    }

private:
    // This alignment is a best effort approach to ensure that each partition falls on a
    // separate page/cache line in order to avoid false sharing.
    struct MONGO_COMPILER_ALIGN_TYPE(128) AlignedLockStats {
        AtomicLockStats stats;
    };

    enum { NumPartitions = 8 };


    AtomicLockStats& _get(LockerId id) {
        return _partitions[id % NumPartitions].stats;
    }


    AlignedLockStats _partitions[NumPartitions];
};


// Global lock manager instance.
LockManager globalLockManager;

// Global lock. Every server operation, which uses the Locker must acquire this lock at least
// once. See comments in the header file (begin/endTransaction) for more information.
const ResourceId resourceIdGlobal = ResourceId(RESOURCE_GLOBAL, ResourceId::SINGLETON_GLOBAL);

// Flush lock. This is only used for the MMAP V1 storage engine and synchronizes journal writes
// to the shared view and remaps. See the comments in the header for information on how MMAP V1
// concurrency control works.
const ResourceId resourceIdMMAPV1Flush =
    ResourceId(RESOURCE_MMAPV1_FLUSH, ResourceId::SINGLETON_MMAPV1_FLUSH);

// How often (in millis) to check for deadlock if a lock has not been granted for some time
const unsigned DeadlockTimeoutMs = 500;

// Dispenses unique LockerId identifiers
AtomicUInt64 idCounter(0);

// Partitioned global lock statistics, so we don't hit the same bucket
PartitionedInstanceWideLockStats globalStats;


/**
 * Whether the particular lock's release should be held until the end of the operation. We
 * delay release of exclusive locks (locks that are for write operations) in order to ensure
 * that the data they protect is committed successfully.
 */
bool shouldDelayUnlock(ResourceId resId, LockMode mode) {
    // Global and flush lock are not used to protect transactional resources and as such, they
    // need to be acquired and released when requested.
    if (resId.getType() == RESOURCE_GLOBAL) {
        return false;
    }

    if (resId == resourceIdMMAPV1Flush) {
        return false;
    }

    switch (mode) {
        case MODE_X:
        case MODE_IX:
            return true;

        case MODE_IS:
        case MODE_S:
            return false;

        default:
            invariant(false);
    }
}

}  // namespace


template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isW() const {
    return getLockMode(resourceIdGlobal) == MODE_X;
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isR() const {
    return getLockMode(resourceIdGlobal) == MODE_S;
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isLocked() const {
    return getLockMode(resourceIdGlobal) != MODE_NONE;
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isWriteLocked() const {
    return isLockHeldForMode(resourceIdGlobal, MODE_IX);
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isReadLocked() const {
    return isLockHeldForMode(resourceIdGlobal, MODE_IS);
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::assertEmptyAndReset() {
    invariant(!inAWriteUnitOfWork());
    invariant(_resourcesToUnlockAtEndOfUnitOfWork.empty());
    invariant(_requests.empty());

    // Reset the locking statistics so the object can be reused
    _stats.reset();
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::dump() const {
    StringBuilder ss;
    ss << "Locker id " << _id << " status: ";

    _lock.lock();
    LockRequestsMap::ConstIterator it = _requests.begin();
    while (!it.finished()) {
        ss << it.key().toString() << " " << lockRequestStatusName(it->status) << " in "
           << modeName(it->mode) << "; ";
        it.next();
    }
    _lock.unlock();

    log() << ss.str() << std::endl;
}


//
// CondVarLockGrantNotification
//

CondVarLockGrantNotification::CondVarLockGrantNotification() {
    clear();
}

void CondVarLockGrantNotification::clear() {
    _result = LOCK_INVALID;
}

LockResult CondVarLockGrantNotification::wait(unsigned timeoutMs) {
    stdx::unique_lock<stdx::mutex> lock(_mutex);
    while (_result == LOCK_INVALID) {
        if (stdx::cv_status::timeout == _cond.wait_for(lock, Milliseconds(timeoutMs))) {
            // Timeout
            return LOCK_TIMEOUT;
        }
    }

    return _result;
}

void CondVarLockGrantNotification::notify(ResourceId resId, LockResult result) {
    stdx::unique_lock<stdx::mutex> lock(_mutex);
    invariant(_result == LOCK_INVALID);
    _result = result;

    _cond.notify_all();
}


//
// Locker
//

template <bool IsForMMAPV1>
LockerImpl<IsForMMAPV1>::LockerImpl()
    : _id(idCounter.addAndFetch(1)), _wuowNestingLevel(0), _batchWriter(false) {}

template <bool IsForMMAPV1>
LockerImpl<IsForMMAPV1>::~LockerImpl() {
    // Cannot delete the Locker while there are still outstanding requests, because the
    // LockManager may attempt to access deleted memory. Besides it is probably incorrect
    // to delete with unaccounted locks anyways.
    assertEmptyAndReset();
}

template <bool IsForMMAPV1>
LockResult LockerImpl<IsForMMAPV1>::lockGlobal(LockMode mode, unsigned timeoutMs) {
    LockResult result = lockGlobalBegin(mode);
    if (result == LOCK_WAITING) {
        result = lockGlobalComplete(timeoutMs);
    }

    if (result == LOCK_OK) {
        lockMMAPV1Flush();
    }

    return result;
}

template <bool IsForMMAPV1>
LockResult LockerImpl<IsForMMAPV1>::lockGlobalBegin(LockMode mode) {
    const LockResult result = lockBegin(resourceIdGlobal, mode);
    if (result == LOCK_OK)
        return LOCK_OK;

    // Currently, deadlock detection does not happen inline with lock acquisition so the only
    // unsuccessful result that the lock manager would return is LOCK_WAITING.
    invariant(result == LOCK_WAITING);

    return result;
}

template <bool IsForMMAPV1>
LockResult LockerImpl<IsForMMAPV1>::lockGlobalComplete(unsigned timeoutMs) {
    return lockComplete(resourceIdGlobal, getLockMode(resourceIdGlobal), timeoutMs, false);
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::lockMMAPV1Flush() {
    if (!IsForMMAPV1)
        return;

    // The flush lock always has a reference count of 1, because it is dropped at the end of
    // each write unit of work in order to allow the flush thread to run. See the comments in
    // the header for information on how the MMAP V1 journaling system works.
    LockRequest* globalLockRequest = _requests.find(resourceIdGlobal).objAddr();
    if (globalLockRequest->recursiveCount == 1) {
        invariant(LOCK_OK == lock(resourceIdMMAPV1Flush, _getModeForMMAPV1FlushLock()));
    }

    dassert(getLockMode(resourceIdMMAPV1Flush) == _getModeForMMAPV1FlushLock());
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::downgradeGlobalXtoSForMMAPV1() {
    invariant(!inAWriteUnitOfWork());

    LockRequest* globalLockRequest = _requests.find(resourceIdGlobal).objAddr();
    invariant(globalLockRequest->mode == MODE_X);
    invariant(globalLockRequest->recursiveCount == 1);

    // Making this call here will record lock downgrades as acquisitions, which is acceptable
    globalStats.recordAcquisition(_id, resourceIdGlobal, MODE_S);
    _stats.recordAcquisition(resourceIdGlobal, MODE_S);

    globalLockManager.downgrade(globalLockRequest, MODE_S);

    if (IsForMMAPV1) {
        invariant(unlock(resourceIdMMAPV1Flush));
    }
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::unlockAll() {
    if (!unlock(resourceIdGlobal)) {
        return false;
    }

    LockRequestsMap::Iterator it = _requests.begin();
    while (!it.finished()) {
        // If we're here we should only have one reference to any lock. It is a programming
        // error for any lock to have more references than the global lock, because every
        // scope starts by calling lockGlobal.
        if (it.key().getType() == RESOURCE_GLOBAL) {
            it.next();
        } else {
            invariant(_unlockImpl(it));
        }
    }

    return true;
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::beginWriteUnitOfWork() {
    // Sanity check that write transactions under MMAP V1 have acquired the flush lock, so we
    // don't allow partial changes to be written.
    dassert(!IsForMMAPV1 || isLockHeldForMode(resourceIdMMAPV1Flush, MODE_IX));

    _wuowNestingLevel++;
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::endWriteUnitOfWork() {
    invariant(_wuowNestingLevel > 0);

    if (--_wuowNestingLevel > 0) {
        // Don't do anything unless leaving outermost WUOW.
        return;
    }

    while (!_resourcesToUnlockAtEndOfUnitOfWork.empty()) {
        unlock(_resourcesToUnlockAtEndOfUnitOfWork.front());
        _resourcesToUnlockAtEndOfUnitOfWork.pop();
    }

    // For MMAP V1, we need to yield the flush lock so that the flush thread can run
    if (IsForMMAPV1) {
        invariant(unlock(resourceIdMMAPV1Flush));
        invariant(LOCK_OK == lock(resourceIdMMAPV1Flush, _getModeForMMAPV1FlushLock()));
    }
}

template <bool IsForMMAPV1>
LockResult LockerImpl<IsForMMAPV1>::lock(ResourceId resId,
                                         LockMode mode,
                                         unsigned timeoutMs,
                                         bool checkDeadlock) {
    const LockResult result = lockBegin(resId, mode);

    // Fast, uncontended path
    if (result == LOCK_OK)
        return LOCK_OK;

    // Currently, deadlock detection does not happen inline with lock acquisition so the only
    // unsuccessful result that the lock manager would return is LOCK_WAITING.
    invariant(result == LOCK_WAITING);

    return lockComplete(resId, mode, timeoutMs, checkDeadlock);
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::downgrade(ResourceId resId, LockMode newMode) {
    LockRequestsMap::Iterator it = _requests.find(resId);
    globalLockManager.downgrade(it.objAddr(), newMode);
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::unlock(ResourceId resId) {
    LockRequestsMap::Iterator it = _requests.find(resId);
    return _unlockImpl(it);
}

template <bool IsForMMAPV1>
LockMode LockerImpl<IsForMMAPV1>::getLockMode(ResourceId resId) const {
    scoped_spinlock scopedLock(_lock);

    const LockRequestsMap::ConstIterator it = _requests.find(resId);
    if (!it)
        return MODE_NONE;

    return it->mode;
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isLockHeldForMode(ResourceId resId, LockMode mode) const {
    return isModeCovered(mode, getLockMode(resId));
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isDbLockedForMode(StringData dbName, LockMode mode) const {
    invariant(nsIsDbOnly(dbName));

    if (isW())
        return true;
    if (isR() && isSharedLockMode(mode))
        return true;

    const ResourceId resIdDb(RESOURCE_DATABASE, dbName);
    return isLockHeldForMode(resIdDb, mode);
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::isCollectionLockedForMode(StringData ns, LockMode mode) const {
    invariant(nsIsFull(ns));

    if (isW())
        return true;
    if (isR() && isSharedLockMode(mode))
        return true;

    const NamespaceString nss(ns);
    const ResourceId resIdDb(RESOURCE_DATABASE, nss.db());

    LockMode dbMode = getLockMode(resIdDb);

    switch (dbMode) {
        case MODE_NONE:
            return false;
        case MODE_X:
            return true;
        case MODE_S:
            return isSharedLockMode(mode);
        case MODE_IX:
        case MODE_IS: {
            const ResourceId resIdColl(RESOURCE_COLLECTION, ns);
            return isLockHeldForMode(resIdColl, mode);
        } break;
        case LockModesCount:
            break;
    }

    invariant(false);
    return false;
}

template <bool IsForMMAPV1>
ResourceId LockerImpl<IsForMMAPV1>::getWaitingResource() const {
    scoped_spinlock scopedLock(_lock);

    LockRequestsMap::ConstIterator it = _requests.begin();
    while (!it.finished()) {
        if (it->status != LockRequest::STATUS_GRANTED) {
            return it.key();
        }

        it.next();
    }

    return ResourceId();
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::getLockerInfo(LockerInfo* lockerInfo) const {
    invariant(lockerInfo);

    // Zero-out the contents
    lockerInfo->locks.clear();
    lockerInfo->waitingResource = ResourceId();
    lockerInfo->stats.reset();

    _lock.lock();
    LockRequestsMap::ConstIterator it = _requests.begin();
    while (!it.finished()) {
        OneLock info;
        info.resourceId = it.key();
        info.mode = it->mode;

        lockerInfo->locks.push_back(info);
        it.next();
    }
    _lock.unlock();

    std::sort(lockerInfo->locks.begin(), lockerInfo->locks.end());

    lockerInfo->waitingResource = getWaitingResource();
    lockerInfo->stats.append(_stats);
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::saveLockStateAndUnlock(Locker::LockSnapshot* stateOut) {
    // We shouldn't be saving and restoring lock state from inside a WriteUnitOfWork.
    invariant(!inAWriteUnitOfWork());

    // Clear out whatever is in stateOut.
    stateOut->locks.clear();
    stateOut->globalMode = MODE_NONE;

    // First, we look at the global lock.  There is special handling for this (as the flush
    // lock goes along with it) so we store it separately from the more pedestrian locks.
    LockRequestsMap::Iterator globalRequest = _requests.find(resourceIdGlobal);
    if (!globalRequest) {
        // If there's no global lock there isn't really anything to do.
        invariant(_requests.empty());
        return false;
    }

    // If the global lock has been acquired more than once, we're probably somewhere in a
    // DBDirectClient call.  It's not safe to release and reacquire locks -- the context using
    // the DBDirectClient is probably not prepared for lock release.
    if (globalRequest->recursiveCount > 1) {
        return false;
    }

    // The global lock must have been acquired just once
    stateOut->globalMode = globalRequest->mode;
    invariant(unlock(resourceIdGlobal));

    // Next, the non-global locks.
    for (LockRequestsMap::Iterator it = _requests.begin(); !it.finished(); it.next()) {
        const ResourceId resId = it.key();

        // We should never have to save and restore metadata locks.
        invariant((IsForMMAPV1 && (resourceIdMMAPV1Flush == resId)) ||
                  RESOURCE_DATABASE == resId.getType() || RESOURCE_COLLECTION == resId.getType() ||
                  (RESOURCE_GLOBAL == resId.getType() && isSharedLockMode(it->mode)));

        // And, stuff the info into the out parameter.
        OneLock info;
        info.resourceId = resId;
        info.mode = it->mode;

        stateOut->locks.push_back(info);

        invariant(unlock(resId));
    }

    // Sort locks by ResourceId. They'll later be acquired in this canonical locking order.
    std::sort(stateOut->locks.begin(), stateOut->locks.end());

    return true;
}

template <bool IsForMMAPV1>
void LockerImpl<IsForMMAPV1>::restoreLockState(const Locker::LockSnapshot& state) {
    // We shouldn't be saving and restoring lock state from inside a WriteUnitOfWork.
    invariant(!inAWriteUnitOfWork());

    std::vector<OneLock>::const_iterator it = state.locks.begin();
    // If we locked the PBWM, it must be locked before the resourceIdGlobal resource.
    if (it != state.locks.end() && it->resourceId == resourceIdParallelBatchWriterMode) {
        invariant(LOCK_OK == lock(it->resourceId, it->mode));
        it++;
    }

    invariant(LOCK_OK == lockGlobal(state.globalMode));
    for (; it != state.locks.end(); it++) {
        // This is a sanity check that lockGlobal restored the MMAP V1 flush lock in the
        // expected mode.
        if (IsForMMAPV1 && (it->resourceId == resourceIdMMAPV1Flush)) {
            invariant(it->mode == _getModeForMMAPV1FlushLock());
        } else {
            invariant(LOCK_OK == lock(it->resourceId, it->mode));
        }
    }
}

template <bool IsForMMAPV1>
LockResult LockerImpl<IsForMMAPV1>::lockBegin(ResourceId resId, LockMode mode) {
    dassert(!getWaitingResource().isValid());

    LockRequest* request;
    bool isNew = true;

    LockRequestsMap::Iterator it = _requests.find(resId);
    if (!it) {
        scoped_spinlock scopedLock(_lock);
        LockRequestsMap::Iterator itNew = _requests.insert(resId);
        itNew->initNew(this, &_notify);

        request = itNew.objAddr();
    } else {
        request = it.objAddr();
        isNew = false;
    }

    // Making this call here will record lock re-acquisitions and conversions as well.
    globalStats.recordAcquisition(_id, resId, mode);
    _stats.recordAcquisition(resId, mode);

    // Give priority to the full modes for global, parallel batch writer mode,
    // and flush lock so we don't stall global operations such as shutdown or flush.
    const ResourceType resType = resId.getType();
    if (resType == RESOURCE_GLOBAL || (IsForMMAPV1 && resId == resourceIdMMAPV1Flush)) {
        if (mode == MODE_S || mode == MODE_X) {
            request->enqueueAtFront = true;
            request->compatibleFirst = true;
        }
    } else {
        // This is all sanity checks that the global and flush locks are always be acquired
        // before any other lock has been acquired and they must be in sync with the nesting.
        DEV {
            const LockRequestsMap::Iterator itGlobal = _requests.find(resourceIdGlobal);
            invariant(itGlobal->recursiveCount > 0);
            invariant(itGlobal->mode != MODE_NONE);

            // Check the MMAP V1 flush lock is held in the appropriate mode
            invariant(!IsForMMAPV1 ||
                      isLockHeldForMode(resourceIdMMAPV1Flush, _getModeForMMAPV1FlushLock()));
        };
    }

    // The notification object must be cleared before we invoke the lock manager, because
    // otherwise we might reset state if the lock becomes granted very fast.
    _notify.clear();

    LockResult result = isNew ? globalLockManager.lock(resId, request, mode)
                              : globalLockManager.convert(resId, request, mode);

    if (result == LOCK_WAITING) {
        globalStats.recordWait(_id, resId, mode);
        _stats.recordWait(resId, mode);
    }

    return result;
}

template <bool IsForMMAPV1>
LockResult LockerImpl<IsForMMAPV1>::lockComplete(ResourceId resId,
                                                 LockMode mode,
                                                 unsigned timeoutMs,
                                                 bool checkDeadlock) {
    // Under MMAP V1 engine a deadlock can occur if a thread goes to sleep waiting on
    // DB lock, while holding the flush lock, so it has to be released. This is only
    // correct to do if not in a write unit of work.
    const bool yieldFlushLock = IsForMMAPV1 && !inAWriteUnitOfWork() &&
        resId.getType() != RESOURCE_GLOBAL && resId != resourceIdMMAPV1Flush;
    if (yieldFlushLock) {
        invariant(unlock(resourceIdMMAPV1Flush));
    }

    LockResult result;

    // Don't go sleeping without bound in order to be able to report long waits or wake up for
    // deadlock detection.
    unsigned waitTimeMs = std::min(timeoutMs, DeadlockTimeoutMs);
    const uint64_t startOfTotalWaitTime = curTimeMicros64();
    uint64_t startOfCurrentWaitTime = startOfTotalWaitTime;

    while (true) {
        // It is OK if this call wakes up spuriously, because we re-evaluate the remaining
        // wait time anyways.
        result = _notify.wait(waitTimeMs);

        // Account for the time spent waiting on the notification object
        const uint64_t curTimeMicros = curTimeMicros64();
        const uint64_t elapsedTimeMicros = curTimeMicros - startOfCurrentWaitTime;
        startOfCurrentWaitTime = curTimeMicros;

        globalStats.recordWaitTime(_id, resId, mode, elapsedTimeMicros);
        _stats.recordWaitTime(resId, mode, elapsedTimeMicros);

        if (result == LOCK_OK)
            break;

        if (checkDeadlock) {
            DeadlockDetector wfg(globalLockManager, this);
            if (wfg.check().hasCycle()) {
                warning() << "Deadlock found: " << wfg.toString();

                globalStats.recordDeadlock(resId, mode);
                _stats.recordDeadlock(resId, mode);

                result = LOCK_DEADLOCK;
                break;
            }
        }

        // If infinite timeout was requested, just keep waiting
        if (timeoutMs == UINT_MAX) {
            continue;
        }

        const unsigned totalBlockTimeMs = (curTimeMicros - startOfTotalWaitTime) / 1000;
        waitTimeMs = (totalBlockTimeMs < timeoutMs)
            ? std::min(timeoutMs - totalBlockTimeMs, DeadlockTimeoutMs)
            : 0;

        if (waitTimeMs == 0) {
            break;
        }
    }

    // Cleanup the state, since this is an unused lock now
    if (result != LOCK_OK) {
        LockRequestsMap::Iterator it = _requests.find(resId);
        if (globalLockManager.unlock(it.objAddr())) {
            scoped_spinlock scopedLock(_lock);
            it.remove();
        }
    }

    if (yieldFlushLock) {
        // We cannot obey the timeout here, because it is not correct to return from the lock
        // request with the flush lock released.
        invariant(LOCK_OK == lock(resourceIdMMAPV1Flush, _getModeForMMAPV1FlushLock()));
    }

    return result;
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::_unlockImpl(LockRequestsMap::Iterator& it) {
    if (inAWriteUnitOfWork() && shouldDelayUnlock(it.key(), it->mode)) {
        _resourcesToUnlockAtEndOfUnitOfWork.push(it.key());
        return false;
    }

    if (globalLockManager.unlock(it.objAddr())) {
        scoped_spinlock scopedLock(_lock);
        it.remove();

        return true;
    }

    return false;
}

template <bool IsForMMAPV1>
LockMode LockerImpl<IsForMMAPV1>::_getModeForMMAPV1FlushLock() const {
    invariant(IsForMMAPV1);

    LockMode mode = getLockMode(resourceIdGlobal);
    switch (mode) {
        case MODE_X:
        case MODE_IX:
            return MODE_IX;
        case MODE_S:
        case MODE_IS:
            return MODE_IS;
        default:
            invariant(false);
            return MODE_NONE;
    }
}

template <bool IsForMMAPV1>
bool LockerImpl<IsForMMAPV1>::hasStrongLocks() const {
    if (!isLocked())
        return false;

    stdx::lock_guard<SpinLock> lk(_lock);
    LockRequestsMap::ConstIterator it = _requests.begin();
    while (!it.finished()) {
        if (it->mode == MODE_X || it->mode == MODE_S) {
            return true;
        }

        it.next();
    }

    return false;
}


//
// Auto classes
//

AutoYieldFlushLockForMMAPV1Commit::AutoYieldFlushLockForMMAPV1Commit(Locker* locker)
    : _locker(static_cast<MMAPV1LockerImpl*>(locker)) {
    // Explicit yielding of the flush lock should happen only at global synchronization points
    // such as database drop. There should not be any active writes at these points.
    invariant(!_locker->inAWriteUnitOfWork());

    if (isMMAPV1()) {
        invariant(_locker->unlock(resourceIdMMAPV1Flush));
    }
}

AutoYieldFlushLockForMMAPV1Commit::~AutoYieldFlushLockForMMAPV1Commit() {
    if (isMMAPV1()) {
        invariant(LOCK_OK ==
                  _locker->lock(resourceIdMMAPV1Flush, _locker->_getModeForMMAPV1FlushLock()));
    }
}


AutoAcquireFlushLockForMMAPV1Commit::AutoAcquireFlushLockForMMAPV1Commit(Locker* locker)
    : _locker(locker), _released(false) {
    // The journal thread acquiring the journal lock in S-mode opens opportunity for deadlock
    // involving operations which do not acquire and release the Oplog collection's X lock
    // inside a WUOW (see SERVER-17416 for the sequence of events), therefore acquire it with
    // check for deadlock and back-off if one is encountered.
    //
    // This exposes theoretical chance that we might starve the journaling system, but given
    // that these deadlocks happen extremely rarely and are usually due to incorrect locking
    // policy, and we have the deadlock counters as part of the locking statistics, this is a
    // reasonable handling.
    //
    // In the worst case, if we are to starve the journaling system, the server will shut down
    // due to too much uncommitted in-memory journal, but won't have corruption.

    while (true) {
        LockResult result = _locker->lock(resourceIdMMAPV1Flush, MODE_S, UINT_MAX, true);
        if (result == LOCK_OK) {
            break;
        }

        invariant(result == LOCK_DEADLOCK);

        warning() << "Delayed journaling in order to avoid deadlock during MMAP V1 journal "
                  << "lock acquisition. See the previous messages for information on the "
                  << "involved threads.";
    }
}

void AutoAcquireFlushLockForMMAPV1Commit::upgradeFlushLockToExclusive() {
    // This should not be able to deadlock, since we already hold the S journal lock, which
    // means all writers are kicked out. Readers always yield the journal lock if they block
    // waiting on any other lock.
    invariant(LOCK_OK == _locker->lock(resourceIdMMAPV1Flush, MODE_X, UINT_MAX, false));

    // Lock bumps the recursive count. Drop it back down so that the destructor doesn't
    // complain.
    invariant(!_locker->unlock(resourceIdMMAPV1Flush));
}

void AutoAcquireFlushLockForMMAPV1Commit::release() {
    if (!_released) {
        invariant(_locker->unlock(resourceIdMMAPV1Flush));
        _released = true;
    }
}

AutoAcquireFlushLockForMMAPV1Commit::~AutoAcquireFlushLockForMMAPV1Commit() {
    release();
}


namespace {
/**
 *  Periodically purges unused lock buckets. The first time the lock is used again after
 *  cleanup it needs to be allocated, and similarly, every first use by a client for an intent
 *  mode may need to create a partitioned lock head. Cleanup is done roughtly once a minute.
 */
class UnusedLockCleaner : PeriodicTask {
public:
    std::string taskName() const {
        return "UnusedLockCleaner";
    }

    void taskDoWork() {
        LOG(2) << "cleaning up unused lock buckets of the global lock manager";
        getGlobalLockManager()->cleanupUnusedLocks();
    }
} unusedLockCleaner;
}  // namespace


//
// Standalone functions
//

LockManager* getGlobalLockManager() {
    return &globalLockManager;
}

void reportGlobalLockingStats(SingleThreadedLockStats* outStats) {
    globalStats.report(outStats);
}

void resetGlobalLockStats() {
    globalStats.reset();
}


// Ensures that there are two instances compiled for LockerImpl for the two values of the
// template argument.
template class LockerImpl<true>;
template class LockerImpl<false>;

// Definition for the hardcoded localdb and oplog collection info
const ResourceId resourceIdLocalDB = ResourceId(RESOURCE_DATABASE, StringData("local"));
const ResourceId resourceIdOplog = ResourceId(RESOURCE_COLLECTION, StringData("local.oplog.rs"));
const ResourceId resourceIdAdminDB = ResourceId(RESOURCE_DATABASE, StringData("admin"));
const ResourceId resourceIdParallelBatchWriterMode =
    ResourceId(RESOURCE_GLOBAL, ResourceId::SINGLETON_PARALLEL_BATCH_WRITER_MODE);
const ResourceId resourceCappedInFlightForLocalDb =
    ResourceId(RESOURCE_METADATA, ResourceId::SINGLETON_CAPPED_IN_FLIGHT_LOCAL_DB);
const ResourceId resourceCappedInFlightForOtherDb =
    ResourceId(RESOURCE_METADATA, ResourceId::SINGLETON_CAPPED_IN_FLIGHT_OTHER_DB);

}  // namespace mongo