1 /*-------------------------------------------------------------------------
4 * POSTGRES primary lock mechanism
6 * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
11 * src/backend/storage/lmgr/lock.c
14 * A lock table is a shared memory hash table. When
15 * a process tries to acquire a lock of a type that conflicts
16 * with existing locks, it is put to sleep using the routines
17 * in storage/lmgr/proc.c.
19 * For the most part, this code should be invoked via lmgr.c
20 * or another lock-management module, not directly.
24 * InitLocks(), GetLocksMethodTable(), GetLockTagsMethodTable(),
25 * LockAcquire(), LockRelease(), LockReleaseAll(),
26 * LockCheckConflicts(), GrantLock()
28 *-------------------------------------------------------------------------
35 #include "access/transam.h"
36 #include "access/twophase.h"
37 #include "access/twophase_rmgr.h"
38 #include "access/xact.h"
39 #include "access/xlog.h"
40 #include "miscadmin.h"
43 #include "storage/proc.h"
44 #include "storage/procarray.h"
45 #include "storage/sinvaladt.h"
46 #include "storage/spin.h"
47 #include "storage/standby.h"
48 #include "utils/memutils.h"
49 #include "utils/ps_status.h"
50 #include "utils/resowner_private.h"
53 /* This configuration variable is used to set the lock table size */
54 int max_locks_per_xact; /* set by guc.c */
57 mul_size(max_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
61 * Data structures defining the semantics of the standard lock methods.
63 * The conflict table defines the semantics of the various lock modes.
65 static const LOCKMASK LockConflicts[] = {
69 LOCKBIT_ON(AccessExclusiveLock),
72 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
74 /* RowExclusiveLock */
75 LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
76 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
78 /* ShareUpdateExclusiveLock */
79 LOCKBIT_ON(ShareUpdateExclusiveLock) |
80 LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
81 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
84 LOCKBIT_ON(RowExclusiveLock) | LOCKBIT_ON(ShareUpdateExclusiveLock) |
85 LOCKBIT_ON(ShareRowExclusiveLock) |
86 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
88 /* ShareRowExclusiveLock */
89 LOCKBIT_ON(RowExclusiveLock) | LOCKBIT_ON(ShareUpdateExclusiveLock) |
90 LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
91 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
94 LOCKBIT_ON(RowShareLock) |
95 LOCKBIT_ON(RowExclusiveLock) | LOCKBIT_ON(ShareUpdateExclusiveLock) |
96 LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
97 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
99 /* AccessExclusiveLock */
100 LOCKBIT_ON(AccessShareLock) | LOCKBIT_ON(RowShareLock) |
101 LOCKBIT_ON(RowExclusiveLock) | LOCKBIT_ON(ShareUpdateExclusiveLock) |
102 LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
103 LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock)
107 /* Names of lock modes, for debug printouts */
108 static const char *const lock_mode_names[] =
114 "ShareUpdateExclusiveLock",
116 "ShareRowExclusiveLock",
118 "AccessExclusiveLock"
122 static bool Dummy_trace = false;
125 static const LockMethodData default_lockmethod = {
126 AccessExclusiveLock, /* highest valid lock mode number */
136 static const LockMethodData user_lockmethod = {
137 AccessExclusiveLock, /* highest valid lock mode number */
148 * map from lock method id to the lock table data structures
150 static const LockMethod LockMethods[] = {
157 /* Record that's written to 2PC state file when a lock is persisted */
158 typedef struct TwoPhaseLockRecord
162 } TwoPhaseLockRecord;
166 * Count of the number of fast path lock slots we believe to be used. This
167 * might be higher than the real number if another backend has transferred
168 * our locks to the primary lock table, but it can never be lower than the
169 * real value, since only we can acquire locks on our own behalf.
171 static int FastPathLocalUseCount = 0;
173 /* Macros for manipulating proc->fpLockBits */
174 #define FAST_PATH_BITS_PER_SLOT 3
175 #define FAST_PATH_LOCKNUMBER_OFFSET 1
176 #define FAST_PATH_MASK ((1 << FAST_PATH_BITS_PER_SLOT) - 1)
177 #define FAST_PATH_GET_BITS(proc, n) \
178 (((proc)->fpLockBits >> (FAST_PATH_BITS_PER_SLOT * n)) & FAST_PATH_MASK)
179 #define FAST_PATH_BIT_POSITION(n, l) \
180 (AssertMacro((l) >= FAST_PATH_LOCKNUMBER_OFFSET), \
181 AssertMacro((l) < FAST_PATH_BITS_PER_SLOT+FAST_PATH_LOCKNUMBER_OFFSET), \
182 AssertMacro((n) < FP_LOCK_SLOTS_PER_BACKEND), \
183 ((l) - FAST_PATH_LOCKNUMBER_OFFSET + FAST_PATH_BITS_PER_SLOT * (n)))
184 #define FAST_PATH_SET_LOCKMODE(proc, n, l) \
185 (proc)->fpLockBits |= UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l)
186 #define FAST_PATH_CLEAR_LOCKMODE(proc, n, l) \
187 (proc)->fpLockBits &= ~(UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l))
188 #define FAST_PATH_CHECK_LOCKMODE(proc, n, l) \
189 ((proc)->fpLockBits & (UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l)))
192 * The fast-path lock mechanism is concerned only with relation locks on
193 * unshared relations by backends bound to a database. The fast-path
194 * mechanism exists mostly to accelerate acquisition and release of locks
195 * that rarely conflict. Because ShareUpdateExclusiveLock is
196 * self-conflicting, it can't use the fast-path mechanism; but it also does
197 * not conflict with any of the locks that do, so we can ignore it completely.
199 #define EligibleForRelationFastPath(locktag, mode) \
200 ((locktag)->locktag_lockmethodid == DEFAULT_LOCKMETHOD && \
201 (locktag)->locktag_type == LOCKTAG_RELATION && \
202 (locktag)->locktag_field1 == MyDatabaseId && \
203 MyDatabaseId != InvalidOid && \
204 (mode) < ShareUpdateExclusiveLock)
205 #define ConflictsWithRelationFastPath(locktag, mode) \
206 ((locktag)->locktag_lockmethodid == DEFAULT_LOCKMETHOD && \
207 (locktag)->locktag_type == LOCKTAG_RELATION && \
208 (locktag)->locktag_field1 != InvalidOid && \
209 (mode) > ShareUpdateExclusiveLock)
211 static bool FastPathGrantRelationLock(Oid relid, LOCKMODE lockmode);
212 static bool FastPathUnGrantRelationLock(Oid relid, LOCKMODE lockmode);
213 static bool FastPathTransferRelationLocks(LockMethod lockMethodTable,
214 const LOCKTAG *locktag, uint32 hashcode);
215 static PROCLOCK *FastPathGetRelationLockEntry(LOCALLOCK *locallock);
218 * To make the fast-path lock mechanism work, we must have some way of
219 * preventing the use of the fast-path when a conflicting lock might be present.
220 * We partition* the locktag space into FAST_PATH_STRONG_LOCK_HASH_PARTITIONS,
221 * and maintain an integer count of the number of "strong" lockers
222 * in each partition. When any "strong" lockers are present (which is
223 * hopefully not very often), the fast-path mechanism can't be used, and we
224 * must fall back to the slower method of pushing matching locks directly
225 * into the main lock tables.
227 * The deadlock detector does not know anything about the fast path mechanism,
228 * so any locks that might be involved in a deadlock must be transferred from
229 * the fast-path queues to the main lock table.
232 #define FAST_PATH_STRONG_LOCK_HASH_BITS 10
233 #define FAST_PATH_STRONG_LOCK_HASH_PARTITIONS \
234 (1 << FAST_PATH_STRONG_LOCK_HASH_BITS)
235 #define FastPathStrongLockHashPartition(hashcode) \
236 ((hashcode) % FAST_PATH_STRONG_LOCK_HASH_PARTITIONS)
241 uint32 count[FAST_PATH_STRONG_LOCK_HASH_PARTITIONS];
242 } FastPathStrongRelationLockData;
244 static volatile FastPathStrongRelationLockData *FastPathStrongRelationLocks;
248 * Pointers to hash tables containing lock state
250 * The LockMethodLockHash and LockMethodProcLockHash hash tables are in
251 * shared memory; LockMethodLocalHash is local to each backend.
253 static HTAB *LockMethodLockHash;
254 static HTAB *LockMethodProcLockHash;
255 static HTAB *LockMethodLocalHash;
258 /* private state for error cleanup */
259 static LOCALLOCK *StrongLockInProgress;
260 static LOCALLOCK *awaitedLock;
261 static ResourceOwner awaitedOwner;
267 * The following configuration options are available for lock debugging:
269 * TRACE_LOCKS -- give a bunch of output what's going on in this file
270 * TRACE_USERLOCKS -- same but for user locks
271 * TRACE_LOCK_OIDMIN-- do not trace locks for tables below this oid
272 * (use to avoid output on system tables)
273 * TRACE_LOCK_TABLE -- trace locks on this table (oid) unconditionally
274 * DEBUG_DEADLOCKS -- currently dumps locks at untimely occasions ;)
276 * Furthermore, but in storage/lmgr/lwlock.c:
277 * TRACE_LWLOCKS -- trace lightweight locks (pretty useless)
279 * Define LOCK_DEBUG at compile time to get all these enabled.
283 int Trace_lock_oidmin = FirstNormalObjectId;
284 bool Trace_locks = false;
285 bool Trace_userlocks = false;
286 int Trace_lock_table = 0;
287 bool Debug_deadlocks = false;
291 LOCK_DEBUG_ENABLED(const LOCKTAG *tag)
294 (*(LockMethods[tag->locktag_lockmethodid]->trace_flag) &&
295 ((Oid) tag->locktag_field2 >= (Oid) Trace_lock_oidmin))
296 || (Trace_lock_table &&
297 (tag->locktag_field2 == Trace_lock_table));
302 LOCK_PRINT(const char *where, const LOCK *lock, LOCKMODE type)
304 if (LOCK_DEBUG_ENABLED(&lock->tag))
306 "%s: lock(%p) id(%u,%u,%u,%u,%u,%u) grantMask(%x) "
307 "req(%d,%d,%d,%d,%d,%d,%d)=%d "
308 "grant(%d,%d,%d,%d,%d,%d,%d)=%d wait(%d) type(%s)",
310 lock->tag.locktag_field1, lock->tag.locktag_field2,
311 lock->tag.locktag_field3, lock->tag.locktag_field4,
312 lock->tag.locktag_type, lock->tag.locktag_lockmethodid,
314 lock->requested[1], lock->requested[2], lock->requested[3],
315 lock->requested[4], lock->requested[5], lock->requested[6],
316 lock->requested[7], lock->nRequested,
317 lock->granted[1], lock->granted[2], lock->granted[3],
318 lock->granted[4], lock->granted[5], lock->granted[6],
319 lock->granted[7], lock->nGranted,
320 lock->waitProcs.size,
321 LockMethods[LOCK_LOCKMETHOD(*lock)]->lockModeNames[type]);
326 PROCLOCK_PRINT(const char *where, const PROCLOCK *proclockP)
328 if (LOCK_DEBUG_ENABLED(&proclockP->tag.myLock->tag))
330 "%s: proclock(%p) lock(%p) method(%u) proc(%p) hold(%x)",
331 where, proclockP, proclockP->tag.myLock,
332 PROCLOCK_LOCKMETHOD(*(proclockP)),
333 proclockP->tag.myProc, (int) proclockP->holdMask);
335 #else /* not LOCK_DEBUG */
337 #define LOCK_PRINT(where, lock, type) ((void) 0)
338 #define PROCLOCK_PRINT(where, proclockP) ((void) 0)
339 #endif /* not LOCK_DEBUG */
342 static uint32 proclock_hash(const void *key, Size keysize);
343 static void RemoveLocalLock(LOCALLOCK *locallock);
344 static PROCLOCK *SetupLockInTable(LockMethod lockMethodTable, PGPROC *proc,
345 const LOCKTAG *locktag, uint32 hashcode, LOCKMODE lockmode);
346 static void GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner);
347 static void BeginStrongLockAcquire(LOCALLOCK *locallock, uint32 fasthashcode);
348 static void FinishStrongLockAcquire(void);
349 static void WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner);
350 static void ReleaseLockIfHeld(LOCALLOCK *locallock, bool sessionLock);
351 static void LockReassignOwner(LOCALLOCK *locallock, ResourceOwner parent);
352 static bool UnGrantLock(LOCK *lock, LOCKMODE lockmode,
353 PROCLOCK *proclock, LockMethod lockMethodTable);
354 static void CleanUpLock(LOCK *lock, PROCLOCK *proclock,
355 LockMethod lockMethodTable, uint32 hashcode,
357 static void LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
358 LOCKTAG *locktag, LOCKMODE lockmode,
359 bool decrement_strong_lock_count);
360 static void GetSingleProcBlockerStatusData(PGPROC *blocked_proc,
361 BlockedProcsData *data);
365 * InitLocks -- Initialize the lock manager's data structures.
367 * This is called from CreateSharedMemoryAndSemaphores(), which see for
368 * more comments. In the normal postmaster case, the shared hash tables
369 * are created here, as well as a locallock hash table that will remain
370 * unused and empty in the postmaster itself. Backends inherit the pointers
371 * to the shared tables via fork(), and also inherit an image of the locallock
372 * hash table, which they proceed to use. In the EXEC_BACKEND case, each
373 * backend re-executes this code to obtain pointers to the already existing
374 * shared hash tables and to create its locallock hash table.
380 long init_table_size,
385 * Compute init/max size to request for lock hashtables. Note these
386 * calculations must agree with LockShmemSize!
388 max_table_size = NLOCKENTS();
389 init_table_size = max_table_size / 2;
392 * Allocate hash table for LOCK structs. This stores per-locked-object
395 MemSet(&info, 0, sizeof(info));
396 info.keysize = sizeof(LOCKTAG);
397 info.entrysize = sizeof(LOCK);
398 info.num_partitions = NUM_LOCK_PARTITIONS;
400 LockMethodLockHash = ShmemInitHash("LOCK hash",
404 HASH_ELEM | HASH_BLOBS | HASH_PARTITION);
406 /* Assume an average of 2 holders per lock */
408 init_table_size *= 2;
411 * Allocate hash table for PROCLOCK structs. This stores
412 * per-lock-per-holder information.
414 info.keysize = sizeof(PROCLOCKTAG);
415 info.entrysize = sizeof(PROCLOCK);
416 info.hash = proclock_hash;
417 info.num_partitions = NUM_LOCK_PARTITIONS;
419 LockMethodProcLockHash = ShmemInitHash("PROCLOCK hash",
423 HASH_ELEM | HASH_FUNCTION | HASH_PARTITION);
426 * Allocate fast-path structures.
428 FastPathStrongRelationLocks =
429 ShmemInitStruct("Fast Path Strong Relation Lock Data",
430 sizeof(FastPathStrongRelationLockData), &found);
432 SpinLockInit(&FastPathStrongRelationLocks->mutex);
435 * Allocate non-shared hash table for LOCALLOCK structs. This stores lock
436 * counts and resource owner information.
438 * The non-shared table could already exist in this process (this occurs
439 * when the postmaster is recreating shared memory after a backend crash).
440 * If so, delete and recreate it. (We could simply leave it, since it
441 * ought to be empty in the postmaster, but for safety let's zap it.)
443 if (LockMethodLocalHash)
444 hash_destroy(LockMethodLocalHash);
446 info.keysize = sizeof(LOCALLOCKTAG);
447 info.entrysize = sizeof(LOCALLOCK);
449 LockMethodLocalHash = hash_create("LOCALLOCK hash",
452 HASH_ELEM | HASH_BLOBS);
457 * Fetch the lock method table associated with a given lock
460 GetLocksMethodTable(const LOCK *lock)
462 LOCKMETHODID lockmethodid = LOCK_LOCKMETHOD(*lock);
464 Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
465 return LockMethods[lockmethodid];
469 * Fetch the lock method table associated with a given locktag
472 GetLockTagsMethodTable(const LOCKTAG *locktag)
474 LOCKMETHODID lockmethodid = (LOCKMETHODID) locktag->locktag_lockmethodid;
476 Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
477 return LockMethods[lockmethodid];
482 * Compute the hash code associated with a LOCKTAG.
484 * To avoid unnecessary recomputations of the hash code, we try to do this
485 * just once per function, and then pass it around as needed. Aside from
486 * passing the hashcode to hash_search_with_hash_value(), we can extract
487 * the lock partition number from the hashcode.
490 LockTagHashCode(const LOCKTAG *locktag)
492 return get_hash_value(LockMethodLockHash, (const void *) locktag);
496 * Compute the hash code associated with a PROCLOCKTAG.
498 * Because we want to use just one set of partition locks for both the
499 * LOCK and PROCLOCK hash tables, we have to make sure that PROCLOCKs
500 * fall into the same partition number as their associated LOCKs.
501 * dynahash.c expects the partition number to be the low-order bits of
502 * the hash code, and therefore a PROCLOCKTAG's hash code must have the
503 * same low-order bits as the associated LOCKTAG's hash code. We achieve
504 * this with this specialized hash function.
507 proclock_hash(const void *key, Size keysize)
509 const PROCLOCKTAG *proclocktag = (const PROCLOCKTAG *) key;
513 Assert(keysize == sizeof(PROCLOCKTAG));
515 /* Look into the associated LOCK object, and compute its hash code */
516 lockhash = LockTagHashCode(&proclocktag->myLock->tag);
519 * To make the hash code also depend on the PGPROC, we xor the proc
520 * struct's address into the hash code, left-shifted so that the
521 * partition-number bits don't change. Since this is only a hash, we
522 * don't care if we lose high-order bits of the address; use an
523 * intermediate variable to suppress cast-pointer-to-int warnings.
525 procptr = PointerGetDatum(proclocktag->myProc);
526 lockhash ^= ((uint32) procptr) << LOG2_NUM_LOCK_PARTITIONS;
532 * Compute the hash code associated with a PROCLOCKTAG, given the hashcode
533 * for its underlying LOCK.
535 * We use this just to avoid redundant calls of LockTagHashCode().
538 ProcLockHashCode(const PROCLOCKTAG *proclocktag, uint32 hashcode)
540 uint32 lockhash = hashcode;
544 * This must match proclock_hash()!
546 procptr = PointerGetDatum(proclocktag->myProc);
547 lockhash ^= ((uint32) procptr) << LOG2_NUM_LOCK_PARTITIONS;
553 * Given two lock modes, return whether they would conflict.
556 DoLockModesConflict(LOCKMODE mode1, LOCKMODE mode2)
558 LockMethod lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
560 if (lockMethodTable->conflictTab[mode1] & LOCKBIT_ON(mode2))
567 * LockHeldByMe -- test whether lock 'locktag' is held with mode 'lockmode'
568 * by the current transaction
571 LockHeldByMe(const LOCKTAG *locktag, LOCKMODE lockmode)
573 LOCALLOCKTAG localtag;
574 LOCALLOCK *locallock;
577 * See if there is a LOCALLOCK entry for this lock and lockmode
579 MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
580 localtag.lock = *locktag;
581 localtag.mode = lockmode;
583 locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
587 return (locallock && locallock->nLocks > 0);
591 * LockHasWaiters -- look up 'locktag' and check if releasing this
592 * lock would wake up other processes waiting for it.
595 LockHasWaiters(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
597 LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
598 LockMethod lockMethodTable;
599 LOCALLOCKTAG localtag;
600 LOCALLOCK *locallock;
603 LWLock *partitionLock;
604 bool hasWaiters = false;
606 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
607 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
608 lockMethodTable = LockMethods[lockmethodid];
609 if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
610 elog(ERROR, "unrecognized lock mode: %d", lockmode);
613 if (LOCK_DEBUG_ENABLED(locktag))
614 elog(LOG, "LockHasWaiters: lock [%u,%u] %s",
615 locktag->locktag_field1, locktag->locktag_field2,
616 lockMethodTable->lockModeNames[lockmode]);
620 * Find the LOCALLOCK entry for this lock and lockmode
622 MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
623 localtag.lock = *locktag;
624 localtag.mode = lockmode;
626 locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
631 * let the caller print its own error message, too. Do not ereport(ERROR).
633 if (!locallock || locallock->nLocks <= 0)
635 elog(WARNING, "you don't own a lock of type %s",
636 lockMethodTable->lockModeNames[lockmode]);
641 * Check the shared lock table.
643 partitionLock = LockHashPartitionLock(locallock->hashcode);
645 LWLockAcquire(partitionLock, LW_SHARED);
648 * We don't need to re-find the lock or proclock, since we kept their
649 * addresses in the locallock table, and they couldn't have been removed
650 * while we were holding a lock on them.
652 lock = locallock->lock;
653 LOCK_PRINT("LockHasWaiters: found", lock, lockmode);
654 proclock = locallock->proclock;
655 PROCLOCK_PRINT("LockHasWaiters: found", proclock);
658 * Double-check that we are actually holding a lock of the type we want to
661 if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
663 PROCLOCK_PRINT("LockHasWaiters: WRONGTYPE", proclock);
664 LWLockRelease(partitionLock);
665 elog(WARNING, "you don't own a lock of type %s",
666 lockMethodTable->lockModeNames[lockmode]);
667 RemoveLocalLock(locallock);
674 if ((lockMethodTable->conflictTab[lockmode] & lock->waitMask) != 0)
677 LWLockRelease(partitionLock);
683 * LockAcquire -- Check for lock conflicts, sleep if conflict found,
684 * set lock if/when no conflicts.
687 * locktag: unique identifier for the lockable object
688 * lockmode: lock mode to acquire
689 * sessionLock: if true, acquire lock for session not current transaction
690 * dontWait: if true, don't wait to acquire lock
693 * LOCKACQUIRE_NOT_AVAIL lock not available, and dontWait=true
694 * LOCKACQUIRE_OK lock successfully acquired
695 * LOCKACQUIRE_ALREADY_HELD incremented count for lock already held
696 * LOCKACQUIRE_ALREADY_CLEAR incremented count for lock already clear
698 * In the normal case where dontWait=false and the caller doesn't need to
699 * distinguish a freshly acquired lock from one already taken earlier in
700 * this same transaction, there is no need to examine the return value.
702 * Side Effects: The lock is acquired and recorded in lock tables.
704 * NOTE: if we wait for the lock, there is no way to abort the wait
705 * short of aborting the transaction.
708 LockAcquire(const LOCKTAG *locktag,
713 return LockAcquireExtended(locktag, lockmode, sessionLock, dontWait,
718 * LockAcquireExtended - allows us to specify additional options
720 * reportMemoryError specifies whether a lock request that fills the lock
721 * table should generate an ERROR or not. Passing "false" allows the caller
722 * to attempt to recover from lock-table-full situations, perhaps by forcibly
723 * cancelling other lock holders and then retrying. Note, however, that the
724 * return code for that is LOCKACQUIRE_NOT_AVAIL, so that it's unsafe to use
725 * in combination with dontWait = true, as the cause of failure couldn't be
728 * If locallockp isn't NULL, *locallockp receives a pointer to the LOCALLOCK
729 * table entry if a lock is successfully acquired, or NULL if not.
732 LockAcquireExtended(const LOCKTAG *locktag,
736 bool reportMemoryError,
737 LOCALLOCK **locallockp)
739 LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
740 LockMethod lockMethodTable;
741 LOCALLOCKTAG localtag;
742 LOCALLOCK *locallock;
748 LWLock *partitionLock;
750 bool log_lock = false;
752 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
753 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
754 lockMethodTable = LockMethods[lockmethodid];
755 if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
756 elog(ERROR, "unrecognized lock mode: %d", lockmode);
758 if (RecoveryInProgress() && !InRecovery &&
759 (locktag->locktag_type == LOCKTAG_OBJECT ||
760 locktag->locktag_type == LOCKTAG_RELATION) &&
761 lockmode > RowExclusiveLock)
763 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
764 errmsg("cannot acquire lock mode %s on database objects while recovery is in progress",
765 lockMethodTable->lockModeNames[lockmode]),
766 errhint("Only RowExclusiveLock or less can be acquired on database objects during recovery.")));
769 if (LOCK_DEBUG_ENABLED(locktag))
770 elog(LOG, "LockAcquire: lock [%u,%u] %s",
771 locktag->locktag_field1, locktag->locktag_field2,
772 lockMethodTable->lockModeNames[lockmode]);
775 /* Identify owner for lock */
779 owner = CurrentResourceOwner;
782 * Find or create a LOCALLOCK entry for this lock and lockmode
784 MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
785 localtag.lock = *locktag;
786 localtag.mode = lockmode;
788 locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
793 * if it's a new locallock object, initialize it
797 locallock->lock = NULL;
798 locallock->proclock = NULL;
799 locallock->hashcode = LockTagHashCode(&(localtag.lock));
800 locallock->nLocks = 0;
801 locallock->holdsStrongLockCount = false;
802 locallock->lockCleared = false;
803 locallock->numLockOwners = 0;
804 locallock->maxLockOwners = 8;
805 locallock->lockOwners = NULL; /* in case next line fails */
806 locallock->lockOwners = (LOCALLOCKOWNER *)
807 MemoryContextAlloc(TopMemoryContext,
808 locallock->maxLockOwners * sizeof(LOCALLOCKOWNER));
812 /* Make sure there will be room to remember the lock */
813 if (locallock->numLockOwners >= locallock->maxLockOwners)
815 int newsize = locallock->maxLockOwners * 2;
817 locallock->lockOwners = (LOCALLOCKOWNER *)
818 repalloc(locallock->lockOwners,
819 newsize * sizeof(LOCALLOCKOWNER));
820 locallock->maxLockOwners = newsize;
823 hashcode = locallock->hashcode;
826 *locallockp = locallock;
829 * If we already hold the lock, we can just increase the count locally.
831 * If lockCleared is already set, caller need not worry about absorbing
832 * sinval messages related to the lock's object.
834 if (locallock->nLocks > 0)
836 GrantLockLocal(locallock, owner);
837 if (locallock->lockCleared)
838 return LOCKACQUIRE_ALREADY_CLEAR;
840 return LOCKACQUIRE_ALREADY_HELD;
844 * Prepare to emit a WAL record if acquisition of this lock needs to be
845 * replayed in a standby server.
847 * Here we prepare to log; after lock is acquired we'll issue log record.
848 * This arrangement simplifies error recovery in case the preparation step
851 * Only AccessExclusiveLocks can conflict with lock types that read-only
852 * transactions can acquire in a standby server. Make sure this definition
853 * matches the one in GetRunningTransactionLocks().
855 if (lockmode >= AccessExclusiveLock &&
856 locktag->locktag_type == LOCKTAG_RELATION &&
857 !RecoveryInProgress() &&
858 XLogStandbyInfoActive())
860 LogAccessExclusiveLockPrepare();
865 * Attempt to take lock via fast path, if eligible. But if we remember
866 * having filled up the fast path array, we don't attempt to make any
867 * further use of it until we release some locks. It's possible that some
868 * other backend has transferred some of those locks to the shared hash
869 * table, leaving space free, but it's not worth acquiring the LWLock just
870 * to check. It's also possible that we're acquiring a second or third
871 * lock type on a relation we have already locked using the fast-path, but
872 * for now we don't worry about that case either.
874 if (EligibleForRelationFastPath(locktag, lockmode) &&
875 FastPathLocalUseCount < FP_LOCK_SLOTS_PER_BACKEND)
877 uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
881 * LWLockAcquire acts as a memory sequencing point, so it's safe to
882 * assume that any strong locker whose increment to
883 * FastPathStrongRelationLocks->counts becomes visible after we test
884 * it has yet to begin to transfer fast-path locks.
886 LWLockAcquire(&MyProc->backendLock, LW_EXCLUSIVE);
887 if (FastPathStrongRelationLocks->count[fasthashcode] != 0)
890 acquired = FastPathGrantRelationLock(locktag->locktag_field2,
892 LWLockRelease(&MyProc->backendLock);
896 * The locallock might contain stale pointers to some old shared
897 * objects; we MUST reset these to null before considering the
898 * lock to be acquired via fast-path.
900 locallock->lock = NULL;
901 locallock->proclock = NULL;
902 GrantLockLocal(locallock, owner);
903 return LOCKACQUIRE_OK;
908 * If this lock could potentially have been taken via the fast-path by
909 * some other backend, we must (temporarily) disable further use of the
910 * fast-path for this lock tag, and migrate any locks already taken via
911 * this method to the main lock table.
913 if (ConflictsWithRelationFastPath(locktag, lockmode))
915 uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
917 BeginStrongLockAcquire(locallock, fasthashcode);
918 if (!FastPathTransferRelationLocks(lockMethodTable, locktag,
921 AbortStrongLockAcquire();
922 if (locallock->nLocks == 0)
923 RemoveLocalLock(locallock);
926 if (reportMemoryError)
928 (errcode(ERRCODE_OUT_OF_MEMORY),
929 errmsg("out of shared memory"),
930 errhint("You might need to increase max_locks_per_transaction.")));
932 return LOCKACQUIRE_NOT_AVAIL;
937 * We didn't find the lock in our LOCALLOCK table, and we didn't manage to
938 * take it via the fast-path, either, so we've got to mess with the shared
941 partitionLock = LockHashPartitionLock(hashcode);
943 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
946 * Find or create lock and proclock entries with this tag
948 * Note: if the locallock object already existed, it might have a pointer
949 * to the lock already ... but we should not assume that that pointer is
950 * valid, since a lock object with zero hold and request counts can go
951 * away anytime. So we have to use SetupLockInTable() to recompute the
952 * lock and proclock pointers, even if they're already set.
954 proclock = SetupLockInTable(lockMethodTable, MyProc, locktag,
958 AbortStrongLockAcquire();
959 LWLockRelease(partitionLock);
960 if (locallock->nLocks == 0)
961 RemoveLocalLock(locallock);
964 if (reportMemoryError)
966 (errcode(ERRCODE_OUT_OF_MEMORY),
967 errmsg("out of shared memory"),
968 errhint("You might need to increase max_locks_per_transaction.")));
970 return LOCKACQUIRE_NOT_AVAIL;
972 locallock->proclock = proclock;
973 lock = proclock->tag.myLock;
974 locallock->lock = lock;
977 * If lock requested conflicts with locks requested by waiters, must join
978 * wait queue. Otherwise, check for conflict with already-held locks.
979 * (That's last because most complex check.)
981 if (lockMethodTable->conflictTab[lockmode] & lock->waitMask)
982 status = STATUS_FOUND;
984 status = LockCheckConflicts(lockMethodTable, lockmode,
987 if (status == STATUS_OK)
989 /* No conflict with held or previously requested locks */
990 GrantLock(lock, proclock, lockmode);
991 GrantLockLocal(locallock, owner);
995 Assert(status == STATUS_FOUND);
998 * We can't acquire the lock immediately. If caller specified no
999 * blocking, remove useless table entries and return
1000 * LOCKACQUIRE_NOT_AVAIL without waiting.
1004 AbortStrongLockAcquire();
1005 if (proclock->holdMask == 0)
1007 uint32 proclock_hashcode;
1009 proclock_hashcode = ProcLockHashCode(&proclock->tag, hashcode);
1010 SHMQueueDelete(&proclock->lockLink);
1011 SHMQueueDelete(&proclock->procLink);
1012 if (!hash_search_with_hash_value(LockMethodProcLockHash,
1013 (void *) &(proclock->tag),
1017 elog(PANIC, "proclock table corrupted");
1020 PROCLOCK_PRINT("LockAcquire: NOWAIT", proclock);
1022 lock->requested[lockmode]--;
1023 LOCK_PRINT("LockAcquire: conditional lock failed", lock, lockmode);
1024 Assert((lock->nRequested > 0) && (lock->requested[lockmode] >= 0));
1025 Assert(lock->nGranted <= lock->nRequested);
1026 LWLockRelease(partitionLock);
1027 if (locallock->nLocks == 0)
1028 RemoveLocalLock(locallock);
1031 return LOCKACQUIRE_NOT_AVAIL;
1035 * Set bitmask of locks this process already holds on this object.
1037 MyProc->heldLocks = proclock->holdMask;
1040 * Sleep till someone wakes me up.
1043 TRACE_POSTGRESQL_LOCK_WAIT_START(locktag->locktag_field1,
1044 locktag->locktag_field2,
1045 locktag->locktag_field3,
1046 locktag->locktag_field4,
1047 locktag->locktag_type,
1050 WaitOnLock(locallock, owner);
1052 TRACE_POSTGRESQL_LOCK_WAIT_DONE(locktag->locktag_field1,
1053 locktag->locktag_field2,
1054 locktag->locktag_field3,
1055 locktag->locktag_field4,
1056 locktag->locktag_type,
1060 * NOTE: do not do any material change of state between here and
1061 * return. All required changes in locktable state must have been
1062 * done when the lock was granted to us --- see notes in WaitOnLock.
1066 * Check the proclock entry status, in case something in the ipc
1067 * communication doesn't work correctly.
1069 if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
1071 AbortStrongLockAcquire();
1072 PROCLOCK_PRINT("LockAcquire: INCONSISTENT", proclock);
1073 LOCK_PRINT("LockAcquire: INCONSISTENT", lock, lockmode);
1074 /* Should we retry ? */
1075 LWLockRelease(partitionLock);
1076 elog(ERROR, "LockAcquire failed");
1078 PROCLOCK_PRINT("LockAcquire: granted", proclock);
1079 LOCK_PRINT("LockAcquire: granted", lock, lockmode);
1083 * Lock state is fully up-to-date now; if we error out after this, no
1084 * special error cleanup is required.
1086 FinishStrongLockAcquire();
1088 LWLockRelease(partitionLock);
1091 * Emit a WAL record if acquisition of this lock needs to be replayed in a
1097 * Decode the locktag back to the original values, to avoid sending
1098 * lots of empty bytes with every message. See lock.h to check how a
1099 * locktag is defined for LOCKTAG_RELATION
1101 LogAccessExclusiveLock(locktag->locktag_field1,
1102 locktag->locktag_field2);
1105 return LOCKACQUIRE_OK;
1109 * Find or create LOCK and PROCLOCK objects as needed for a new lock
1112 * Returns the PROCLOCK object, or NULL if we failed to create the objects
1113 * for lack of shared memory.
1115 * The appropriate partition lock must be held at entry, and will be
1119 SetupLockInTable(LockMethod lockMethodTable, PGPROC *proc,
1120 const LOCKTAG *locktag, uint32 hashcode, LOCKMODE lockmode)
1124 PROCLOCKTAG proclocktag;
1125 uint32 proclock_hashcode;
1129 * Find or create a lock with this tag.
1131 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
1132 (const void *) locktag,
1140 * if it's a new lock object, initialize it
1144 lock->grantMask = 0;
1146 SHMQueueInit(&(lock->procLocks));
1147 ProcQueueInit(&(lock->waitProcs));
1148 lock->nRequested = 0;
1150 MemSet(lock->requested, 0, sizeof(int) * MAX_LOCKMODES);
1151 MemSet(lock->granted, 0, sizeof(int) * MAX_LOCKMODES);
1152 LOCK_PRINT("LockAcquire: new", lock, lockmode);
1156 LOCK_PRINT("LockAcquire: found", lock, lockmode);
1157 Assert((lock->nRequested >= 0) && (lock->requested[lockmode] >= 0));
1158 Assert((lock->nGranted >= 0) && (lock->granted[lockmode] >= 0));
1159 Assert(lock->nGranted <= lock->nRequested);
1163 * Create the hash key for the proclock table.
1165 proclocktag.myLock = lock;
1166 proclocktag.myProc = proc;
1168 proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
1171 * Find or create a proclock entry with this tag
1173 proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
1174 (void *) &proclocktag,
1180 /* Oops, not enough shmem for the proclock */
1181 if (lock->nRequested == 0)
1184 * There are no other requestors of this lock, so garbage-collect
1185 * the lock object. We *must* do this to avoid a permanent leak
1186 * of shared memory, because there won't be anything to cause
1187 * anyone to release the lock object later.
1189 Assert(SHMQueueEmpty(&(lock->procLocks)));
1190 if (!hash_search_with_hash_value(LockMethodLockHash,
1191 (void *) &(lock->tag),
1195 elog(PANIC, "lock table corrupted");
1201 * If new, initialize the new entry
1205 uint32 partition = LockHashPartition(hashcode);
1208 * It might seem unsafe to access proclock->groupLeader without a
1209 * lock, but it's not really. Either we are initializing a proclock
1210 * on our own behalf, in which case our group leader isn't changing
1211 * because the group leader for a process can only ever be changed by
1212 * the process itself; or else we are transferring a fast-path lock to
1213 * the main lock table, in which case that process can't change it's
1214 * lock group leader without first releasing all of its locks (and in
1215 * particular the one we are currently transferring).
1217 proclock->groupLeader = proc->lockGroupLeader != NULL ?
1218 proc->lockGroupLeader : proc;
1219 proclock->holdMask = 0;
1220 proclock->releaseMask = 0;
1221 /* Add proclock to appropriate lists */
1222 SHMQueueInsertBefore(&lock->procLocks, &proclock->lockLink);
1223 SHMQueueInsertBefore(&(proc->myProcLocks[partition]),
1224 &proclock->procLink);
1225 PROCLOCK_PRINT("LockAcquire: new", proclock);
1229 PROCLOCK_PRINT("LockAcquire: found", proclock);
1230 Assert((proclock->holdMask & ~lock->grantMask) == 0);
1232 #ifdef CHECK_DEADLOCK_RISK
1235 * Issue warning if we already hold a lower-level lock on this object
1236 * and do not hold a lock of the requested level or higher. This
1237 * indicates a deadlock-prone coding practice (eg, we'd have a
1238 * deadlock if another backend were following the same code path at
1239 * about the same time).
1241 * This is not enabled by default, because it may generate log entries
1242 * about user-level coding practices that are in fact safe in context.
1243 * It can be enabled to help find system-level problems.
1245 * XXX Doing numeric comparison on the lockmodes is a hack; it'd be
1246 * better to use a table. For now, though, this works.
1251 for (i = lockMethodTable->numLockModes; i > 0; i--)
1253 if (proclock->holdMask & LOCKBIT_ON(i))
1255 if (i >= (int) lockmode)
1256 break; /* safe: we have a lock >= req level */
1257 elog(LOG, "deadlock risk: raising lock level"
1258 " from %s to %s on object %u/%u/%u",
1259 lockMethodTable->lockModeNames[i],
1260 lockMethodTable->lockModeNames[lockmode],
1261 lock->tag.locktag_field1, lock->tag.locktag_field2,
1262 lock->tag.locktag_field3);
1267 #endif /* CHECK_DEADLOCK_RISK */
1271 * lock->nRequested and lock->requested[] count the total number of
1272 * requests, whether granted or waiting, so increment those immediately.
1273 * The other counts don't increment till we get the lock.
1276 lock->requested[lockmode]++;
1277 Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
1280 * We shouldn't already hold the desired lock; else locallock table is
1283 if (proclock->holdMask & LOCKBIT_ON(lockmode))
1284 elog(ERROR, "lock %s on object %u/%u/%u is already held",
1285 lockMethodTable->lockModeNames[lockmode],
1286 lock->tag.locktag_field1, lock->tag.locktag_field2,
1287 lock->tag.locktag_field3);
1293 * Subroutine to free a locallock entry
1296 RemoveLocalLock(LOCALLOCK *locallock)
1300 for (i = locallock->numLockOwners - 1; i >= 0; i--)
1302 if (locallock->lockOwners[i].owner != NULL)
1303 ResourceOwnerForgetLock(locallock->lockOwners[i].owner, locallock);
1305 locallock->numLockOwners = 0;
1306 if (locallock->lockOwners != NULL)
1307 pfree(locallock->lockOwners);
1308 locallock->lockOwners = NULL;
1310 if (locallock->holdsStrongLockCount)
1312 uint32 fasthashcode;
1314 fasthashcode = FastPathStrongLockHashPartition(locallock->hashcode);
1316 SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1317 Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
1318 FastPathStrongRelationLocks->count[fasthashcode]--;
1319 locallock->holdsStrongLockCount = false;
1320 SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1323 if (!hash_search(LockMethodLocalHash,
1324 (void *) &(locallock->tag),
1326 elog(WARNING, "locallock table corrupted");
1330 * LockCheckConflicts -- test whether requested lock conflicts
1331 * with those already granted
1333 * Returns STATUS_FOUND if conflict, STATUS_OK if no conflict.
1336 * Here's what makes this complicated: one process's locks don't
1337 * conflict with one another, no matter what purpose they are held for
1338 * (eg, session and transaction locks do not conflict). Nor do the locks
1339 * of one process in a lock group conflict with those of another process in
1340 * the same group. So, we must subtract off these locks when determining
1341 * whether the requested new lock conflicts with those already held.
1344 LockCheckConflicts(LockMethod lockMethodTable,
1349 int numLockModes = lockMethodTable->numLockModes;
1351 int conflictMask = lockMethodTable->conflictTab[lockmode];
1352 int conflictsRemaining[MAX_LOCKMODES];
1353 int totalConflictsRemaining = 0;
1355 SHM_QUEUE *procLocks;
1356 PROCLOCK *otherproclock;
1359 * first check for global conflicts: If no locks conflict with my request,
1360 * then I get the lock.
1362 * Checking for conflict: lock->grantMask represents the types of
1363 * currently held locks. conflictTable[lockmode] has a bit set for each
1364 * type of lock that conflicts with request. Bitwise compare tells if
1365 * there is a conflict.
1367 if (!(conflictMask & lock->grantMask))
1369 PROCLOCK_PRINT("LockCheckConflicts: no conflict", proclock);
1374 * Rats. Something conflicts. But it could still be my own lock, or a
1375 * lock held by another member of my locking group. First, figure out how
1376 * many conflicts remain after subtracting out any locks I hold myself.
1378 myLocks = proclock->holdMask;
1379 for (i = 1; i <= numLockModes; i++)
1381 if ((conflictMask & LOCKBIT_ON(i)) == 0)
1383 conflictsRemaining[i] = 0;
1386 conflictsRemaining[i] = lock->granted[i];
1387 if (myLocks & LOCKBIT_ON(i))
1388 --conflictsRemaining[i];
1389 totalConflictsRemaining += conflictsRemaining[i];
1392 /* If no conflicts remain, we get the lock. */
1393 if (totalConflictsRemaining == 0)
1395 PROCLOCK_PRINT("LockCheckConflicts: resolved (simple)", proclock);
1399 /* If no group locking, it's definitely a conflict. */
1400 if (proclock->groupLeader == MyProc && MyProc->lockGroupLeader == NULL)
1402 Assert(proclock->tag.myProc == MyProc);
1403 PROCLOCK_PRINT("LockCheckConflicts: conflicting (simple)",
1405 return STATUS_FOUND;
1409 * Locks held in conflicting modes by members of our own lock group are
1410 * not real conflicts; we can subtract those out and see if we still have
1411 * a conflict. This is O(N) in the number of processes holding or
1412 * awaiting locks on this object. We could improve that by making the
1413 * shared memory state more complex (and larger) but it doesn't seem worth
1416 procLocks = &(lock->procLocks);
1417 otherproclock = (PROCLOCK *)
1418 SHMQueueNext(procLocks, procLocks, offsetof(PROCLOCK, lockLink));
1419 while (otherproclock != NULL)
1421 if (proclock != otherproclock &&
1422 proclock->groupLeader == otherproclock->groupLeader &&
1423 (otherproclock->holdMask & conflictMask) != 0)
1425 int intersectMask = otherproclock->holdMask & conflictMask;
1427 for (i = 1; i <= numLockModes; i++)
1429 if ((intersectMask & LOCKBIT_ON(i)) != 0)
1431 if (conflictsRemaining[i] <= 0)
1432 elog(PANIC, "proclocks held do not match lock");
1433 conflictsRemaining[i]--;
1434 totalConflictsRemaining--;
1438 if (totalConflictsRemaining == 0)
1440 PROCLOCK_PRINT("LockCheckConflicts: resolved (group)",
1445 otherproclock = (PROCLOCK *)
1446 SHMQueueNext(procLocks, &otherproclock->lockLink,
1447 offsetof(PROCLOCK, lockLink));
1450 /* Nope, it's a real conflict. */
1451 PROCLOCK_PRINT("LockCheckConflicts: conflicting (group)", proclock);
1452 return STATUS_FOUND;
1456 * GrantLock -- update the lock and proclock data structures to show
1457 * the lock request has been granted.
1459 * NOTE: if proc was blocked, it also needs to be removed from the wait list
1460 * and have its waitLock/waitProcLock fields cleared. That's not done here.
1462 * NOTE: the lock grant also has to be recorded in the associated LOCALLOCK
1463 * table entry; but since we may be awaking some other process, we can't do
1464 * that here; it's done by GrantLockLocal, instead.
1467 GrantLock(LOCK *lock, PROCLOCK *proclock, LOCKMODE lockmode)
1470 lock->granted[lockmode]++;
1471 lock->grantMask |= LOCKBIT_ON(lockmode);
1472 if (lock->granted[lockmode] == lock->requested[lockmode])
1473 lock->waitMask &= LOCKBIT_OFF(lockmode);
1474 proclock->holdMask |= LOCKBIT_ON(lockmode);
1475 LOCK_PRINT("GrantLock", lock, lockmode);
1476 Assert((lock->nGranted > 0) && (lock->granted[lockmode] > 0));
1477 Assert(lock->nGranted <= lock->nRequested);
1481 * UnGrantLock -- opposite of GrantLock.
1483 * Updates the lock and proclock data structures to show that the lock
1484 * is no longer held nor requested by the current holder.
1486 * Returns true if there were any waiters waiting on the lock that
1487 * should now be woken up with ProcLockWakeup.
1490 UnGrantLock(LOCK *lock, LOCKMODE lockmode,
1491 PROCLOCK *proclock, LockMethod lockMethodTable)
1493 bool wakeupNeeded = false;
1495 Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
1496 Assert((lock->nGranted > 0) && (lock->granted[lockmode] > 0));
1497 Assert(lock->nGranted <= lock->nRequested);
1500 * fix the general lock stats
1503 lock->requested[lockmode]--;
1505 lock->granted[lockmode]--;
1507 if (lock->granted[lockmode] == 0)
1509 /* change the conflict mask. No more of this lock type. */
1510 lock->grantMask &= LOCKBIT_OFF(lockmode);
1513 LOCK_PRINT("UnGrantLock: updated", lock, lockmode);
1516 * We need only run ProcLockWakeup if the released lock conflicts with at
1517 * least one of the lock types requested by waiter(s). Otherwise whatever
1518 * conflict made them wait must still exist. NOTE: before MVCC, we could
1519 * skip wakeup if lock->granted[lockmode] was still positive. But that's
1520 * not true anymore, because the remaining granted locks might belong to
1521 * some waiter, who could now be awakened because he doesn't conflict with
1524 if (lockMethodTable->conflictTab[lockmode] & lock->waitMask)
1525 wakeupNeeded = true;
1528 * Now fix the per-proclock state.
1530 proclock->holdMask &= LOCKBIT_OFF(lockmode);
1531 PROCLOCK_PRINT("UnGrantLock: updated", proclock);
1533 return wakeupNeeded;
1537 * CleanUpLock -- clean up after releasing a lock. We garbage-collect the
1538 * proclock and lock objects if possible, and call ProcLockWakeup if there
1539 * are remaining requests and the caller says it's OK. (Normally, this
1540 * should be called after UnGrantLock, and wakeupNeeded is the result from
1543 * The appropriate partition lock must be held at entry, and will be
1547 CleanUpLock(LOCK *lock, PROCLOCK *proclock,
1548 LockMethod lockMethodTable, uint32 hashcode,
1552 * If this was my last hold on this lock, delete my entry in the proclock
1555 if (proclock->holdMask == 0)
1557 uint32 proclock_hashcode;
1559 PROCLOCK_PRINT("CleanUpLock: deleting", proclock);
1560 SHMQueueDelete(&proclock->lockLink);
1561 SHMQueueDelete(&proclock->procLink);
1562 proclock_hashcode = ProcLockHashCode(&proclock->tag, hashcode);
1563 if (!hash_search_with_hash_value(LockMethodProcLockHash,
1564 (void *) &(proclock->tag),
1568 elog(PANIC, "proclock table corrupted");
1571 if (lock->nRequested == 0)
1574 * The caller just released the last lock, so garbage-collect the lock
1577 LOCK_PRINT("CleanUpLock: deleting", lock, 0);
1578 Assert(SHMQueueEmpty(&(lock->procLocks)));
1579 if (!hash_search_with_hash_value(LockMethodLockHash,
1580 (void *) &(lock->tag),
1584 elog(PANIC, "lock table corrupted");
1586 else if (wakeupNeeded)
1588 /* There are waiters on this lock, so wake them up. */
1589 ProcLockWakeup(lockMethodTable, lock);
1594 * GrantLockLocal -- update the locallock data structures to show
1595 * the lock request has been granted.
1597 * We expect that LockAcquire made sure there is room to add a new
1598 * ResourceOwner entry.
1601 GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner)
1603 LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
1606 Assert(locallock->numLockOwners < locallock->maxLockOwners);
1607 /* Count the total */
1608 locallock->nLocks++;
1609 /* Count the per-owner lock */
1610 for (i = 0; i < locallock->numLockOwners; i++)
1612 if (lockOwners[i].owner == owner)
1614 lockOwners[i].nLocks++;
1618 lockOwners[i].owner = owner;
1619 lockOwners[i].nLocks = 1;
1620 locallock->numLockOwners++;
1622 ResourceOwnerRememberLock(owner, locallock);
1626 * BeginStrongLockAcquire - inhibit use of fastpath for a given LOCALLOCK,
1627 * and arrange for error cleanup if it fails
1630 BeginStrongLockAcquire(LOCALLOCK *locallock, uint32 fasthashcode)
1632 Assert(StrongLockInProgress == NULL);
1633 Assert(locallock->holdsStrongLockCount == false);
1636 * Adding to a memory location is not atomic, so we take a spinlock to
1637 * ensure we don't collide with someone else trying to bump the count at
1640 * XXX: It might be worth considering using an atomic fetch-and-add
1641 * instruction here, on architectures where that is supported.
1644 SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1645 FastPathStrongRelationLocks->count[fasthashcode]++;
1646 locallock->holdsStrongLockCount = true;
1647 StrongLockInProgress = locallock;
1648 SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1652 * FinishStrongLockAcquire - cancel pending cleanup for a strong lock
1653 * acquisition once it's no longer needed
1656 FinishStrongLockAcquire(void)
1658 StrongLockInProgress = NULL;
1662 * AbortStrongLockAcquire - undo strong lock state changes performed by
1663 * BeginStrongLockAcquire.
1666 AbortStrongLockAcquire(void)
1668 uint32 fasthashcode;
1669 LOCALLOCK *locallock = StrongLockInProgress;
1671 if (locallock == NULL)
1674 fasthashcode = FastPathStrongLockHashPartition(locallock->hashcode);
1675 Assert(locallock->holdsStrongLockCount == true);
1676 SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1677 Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
1678 FastPathStrongRelationLocks->count[fasthashcode]--;
1679 locallock->holdsStrongLockCount = false;
1680 StrongLockInProgress = NULL;
1681 SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1685 * GrantAwaitedLock -- call GrantLockLocal for the lock we are doing
1688 * proc.c needs this for the case where we are booted off the lock by
1689 * timeout, but discover that someone granted us the lock anyway.
1691 * We could just export GrantLockLocal, but that would require including
1692 * resowner.h in lock.h, which creates circularity.
1695 GrantAwaitedLock(void)
1697 GrantLockLocal(awaitedLock, awaitedOwner);
1701 * MarkLockClear -- mark an acquired lock as "clear"
1703 * This means that we know we have absorbed all sinval messages that other
1704 * sessions generated before we acquired this lock, and so we can confidently
1705 * assume we know about any catalog changes protected by this lock.
1708 MarkLockClear(LOCALLOCK *locallock)
1710 Assert(locallock->nLocks > 0);
1711 locallock->lockCleared = true;
1715 * WaitOnLock -- wait to acquire a lock
1717 * Caller must have set MyProc->heldLocks to reflect locks already held
1718 * on the lockable object by this process.
1720 * The appropriate partition lock must be held at entry.
1723 WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner)
1725 LOCKMETHODID lockmethodid = LOCALLOCK_LOCKMETHOD(*locallock);
1726 LockMethod lockMethodTable = LockMethods[lockmethodid];
1727 char *volatile new_status = NULL;
1729 LOCK_PRINT("WaitOnLock: sleeping on lock",
1730 locallock->lock, locallock->tag.mode);
1732 /* Report change to waiting status */
1733 if (update_process_title)
1735 const char *old_status;
1738 old_status = get_ps_display(&len);
1739 new_status = (char *) palloc(len + 8 + 1);
1740 memcpy(new_status, old_status, len);
1741 strcpy(new_status + len, " waiting");
1742 set_ps_display(new_status, false);
1743 new_status[len] = '\0'; /* truncate off " waiting" */
1746 awaitedLock = locallock;
1747 awaitedOwner = owner;
1750 * NOTE: Think not to put any shared-state cleanup after the call to
1751 * ProcSleep, in either the normal or failure path. The lock state must
1752 * be fully set by the lock grantor, or by CheckDeadLock if we give up
1753 * waiting for the lock. This is necessary because of the possibility
1754 * that a cancel/die interrupt will interrupt ProcSleep after someone else
1755 * grants us the lock, but before we've noticed it. Hence, after granting,
1756 * the locktable state must fully reflect the fact that we own the lock;
1757 * we can't do additional work on return.
1759 * We can and do use a PG_TRY block to try to clean up after failure, but
1760 * this still has a major limitation: elog(FATAL) can occur while waiting
1761 * (eg, a "die" interrupt), and then control won't come back here. So all
1762 * cleanup of essential state should happen in LockErrorCleanup, not here.
1763 * We can use PG_TRY to clear the "waiting" status flags, since doing that
1764 * is unimportant if the process exits.
1768 if (ProcSleep(locallock, lockMethodTable) != STATUS_OK)
1771 * We failed as a result of a deadlock, see CheckDeadLock(). Quit
1775 LOCK_PRINT("WaitOnLock: aborting on lock",
1776 locallock->lock, locallock->tag.mode);
1777 LWLockRelease(LockHashPartitionLock(locallock->hashcode));
1780 * Now that we aren't holding the partition lock, we can give an
1781 * error report including details about the detected deadlock.
1789 /* In this path, awaitedLock remains set until LockErrorCleanup */
1791 /* Report change to non-waiting status */
1792 if (update_process_title)
1794 set_ps_display(new_status, false);
1798 /* and propagate the error */
1805 /* Report change to non-waiting status */
1806 if (update_process_title)
1808 set_ps_display(new_status, false);
1812 LOCK_PRINT("WaitOnLock: wakeup on lock",
1813 locallock->lock, locallock->tag.mode);
1817 * Remove a proc from the wait-queue it is on (caller must know it is on one).
1818 * This is only used when the proc has failed to get the lock, so we set its
1819 * waitStatus to STATUS_ERROR.
1821 * Appropriate partition lock must be held by caller. Also, caller is
1822 * responsible for signaling the proc if needed.
1824 * NB: this does not clean up any locallock object that may exist for the lock.
1827 RemoveFromWaitQueue(PGPROC *proc, uint32 hashcode)
1829 LOCK *waitLock = proc->waitLock;
1830 PROCLOCK *proclock = proc->waitProcLock;
1831 LOCKMODE lockmode = proc->waitLockMode;
1832 LOCKMETHODID lockmethodid = LOCK_LOCKMETHOD(*waitLock);
1834 /* Make sure proc is waiting */
1835 Assert(proc->waitStatus == STATUS_WAITING);
1836 Assert(proc->links.next != NULL);
1838 Assert(waitLock->waitProcs.size > 0);
1839 Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
1841 /* Remove proc from lock's wait queue */
1842 SHMQueueDelete(&(proc->links));
1843 waitLock->waitProcs.size--;
1845 /* Undo increments of request counts by waiting process */
1846 Assert(waitLock->nRequested > 0);
1847 Assert(waitLock->nRequested > proc->waitLock->nGranted);
1848 waitLock->nRequested--;
1849 Assert(waitLock->requested[lockmode] > 0);
1850 waitLock->requested[lockmode]--;
1851 /* don't forget to clear waitMask bit if appropriate */
1852 if (waitLock->granted[lockmode] == waitLock->requested[lockmode])
1853 waitLock->waitMask &= LOCKBIT_OFF(lockmode);
1855 /* Clean up the proc's own state, and pass it the ok/fail signal */
1856 proc->waitLock = NULL;
1857 proc->waitProcLock = NULL;
1858 proc->waitStatus = STATUS_ERROR;
1861 * Delete the proclock immediately if it represents no already-held locks.
1862 * (This must happen now because if the owner of the lock decides to
1863 * release it, and the requested/granted counts then go to zero,
1864 * LockRelease expects there to be no remaining proclocks.) Then see if
1865 * any other waiters for the lock can be woken up now.
1867 CleanUpLock(waitLock, proclock,
1868 LockMethods[lockmethodid], hashcode,
1873 * LockRelease -- look up 'locktag' and release one 'lockmode' lock on it.
1874 * Release a session lock if 'sessionLock' is true, else release a
1875 * regular transaction lock.
1877 * Side Effects: find any waiting processes that are now wakable,
1878 * grant them their requested locks and awaken them.
1879 * (We have to grant the lock here to avoid a race between
1880 * the waking process and any new process to
1881 * come along and request the lock.)
1884 LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
1886 LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
1887 LockMethod lockMethodTable;
1888 LOCALLOCKTAG localtag;
1889 LOCALLOCK *locallock;
1892 LWLock *partitionLock;
1895 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
1896 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
1897 lockMethodTable = LockMethods[lockmethodid];
1898 if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
1899 elog(ERROR, "unrecognized lock mode: %d", lockmode);
1902 if (LOCK_DEBUG_ENABLED(locktag))
1903 elog(LOG, "LockRelease: lock [%u,%u] %s",
1904 locktag->locktag_field1, locktag->locktag_field2,
1905 lockMethodTable->lockModeNames[lockmode]);
1909 * Find the LOCALLOCK entry for this lock and lockmode
1911 MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
1912 localtag.lock = *locktag;
1913 localtag.mode = lockmode;
1915 locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
1920 * let the caller print its own error message, too. Do not ereport(ERROR).
1922 if (!locallock || locallock->nLocks <= 0)
1924 elog(WARNING, "you don't own a lock of type %s",
1925 lockMethodTable->lockModeNames[lockmode]);
1930 * Decrease the count for the resource owner.
1933 LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
1934 ResourceOwner owner;
1937 /* Identify owner for lock */
1941 owner = CurrentResourceOwner;
1943 for (i = locallock->numLockOwners - 1; i >= 0; i--)
1945 if (lockOwners[i].owner == owner)
1947 Assert(lockOwners[i].nLocks > 0);
1948 if (--lockOwners[i].nLocks == 0)
1951 ResourceOwnerForgetLock(owner, locallock);
1952 /* compact out unused slot */
1953 locallock->numLockOwners--;
1954 if (i < locallock->numLockOwners)
1955 lockOwners[i] = lockOwners[locallock->numLockOwners];
1962 /* don't release a lock belonging to another owner */
1963 elog(WARNING, "you don't own a lock of type %s",
1964 lockMethodTable->lockModeNames[lockmode]);
1970 * Decrease the total local count. If we're still holding the lock, we're
1973 locallock->nLocks--;
1975 if (locallock->nLocks > 0)
1979 * At this point we can no longer suppose we are clear of invalidation
1980 * messages related to this lock. Although we'll delete the LOCALLOCK
1981 * object before any intentional return from this routine, it seems worth
1982 * the trouble to explicitly reset lockCleared right now, just in case
1983 * some error prevents us from deleting the LOCALLOCK.
1985 locallock->lockCleared = false;
1987 /* Attempt fast release of any lock eligible for the fast path. */
1988 if (EligibleForRelationFastPath(locktag, lockmode) &&
1989 FastPathLocalUseCount > 0)
1994 * We might not find the lock here, even if we originally entered it
1995 * here. Another backend may have moved it to the main table.
1997 LWLockAcquire(&MyProc->backendLock, LW_EXCLUSIVE);
1998 released = FastPathUnGrantRelationLock(locktag->locktag_field2,
2000 LWLockRelease(&MyProc->backendLock);
2003 RemoveLocalLock(locallock);
2009 * Otherwise we've got to mess with the shared lock table.
2011 partitionLock = LockHashPartitionLock(locallock->hashcode);
2013 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2016 * Normally, we don't need to re-find the lock or proclock, since we kept
2017 * their addresses in the locallock table, and they couldn't have been
2018 * removed while we were holding a lock on them. But it's possible that
2019 * the lock was taken fast-path and has since been moved to the main hash
2020 * table by another backend, in which case we will need to look up the
2021 * objects here. We assume the lock field is NULL if so.
2023 lock = locallock->lock;
2026 PROCLOCKTAG proclocktag;
2028 Assert(EligibleForRelationFastPath(locktag, lockmode));
2029 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
2030 (const void *) locktag,
2031 locallock->hashcode,
2035 elog(ERROR, "failed to re-find shared lock object");
2036 locallock->lock = lock;
2038 proclocktag.myLock = lock;
2039 proclocktag.myProc = MyProc;
2040 locallock->proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash,
2041 (void *) &proclocktag,
2044 if (!locallock->proclock)
2045 elog(ERROR, "failed to re-find shared proclock object");
2047 LOCK_PRINT("LockRelease: found", lock, lockmode);
2048 proclock = locallock->proclock;
2049 PROCLOCK_PRINT("LockRelease: found", proclock);
2052 * Double-check that we are actually holding a lock of the type we want to
2055 if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
2057 PROCLOCK_PRINT("LockRelease: WRONGTYPE", proclock);
2058 LWLockRelease(partitionLock);
2059 elog(WARNING, "you don't own a lock of type %s",
2060 lockMethodTable->lockModeNames[lockmode]);
2061 RemoveLocalLock(locallock);
2066 * Do the releasing. CleanUpLock will waken any now-wakable waiters.
2068 wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
2070 CleanUpLock(lock, proclock,
2071 lockMethodTable, locallock->hashcode,
2074 LWLockRelease(partitionLock);
2076 RemoveLocalLock(locallock);
2081 * LockReleaseAll -- Release all locks of the specified lock method that
2082 * are held by the current process.
2084 * Well, not necessarily *all* locks. The available behaviors are:
2085 * allLocks == true: release all locks including session locks.
2086 * allLocks == false: release all non-session locks.
2089 LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
2091 HASH_SEQ_STATUS status;
2092 LockMethod lockMethodTable;
2095 LOCALLOCK *locallock;
2099 bool have_fast_path_lwlock = false;
2101 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2102 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2103 lockMethodTable = LockMethods[lockmethodid];
2106 if (*(lockMethodTable->trace_flag))
2107 elog(LOG, "LockReleaseAll: lockmethod=%d", lockmethodid);
2111 * Get rid of our fast-path VXID lock, if appropriate. Note that this is
2112 * the only way that the lock we hold on our own VXID can ever get
2113 * released: it is always and only released when a toplevel transaction
2116 if (lockmethodid == DEFAULT_LOCKMETHOD)
2117 VirtualXactLockTableCleanup();
2119 numLockModes = lockMethodTable->numLockModes;
2122 * First we run through the locallock table and get rid of unwanted
2123 * entries, then we scan the process's proclocks and get rid of those. We
2124 * do this separately because we may have multiple locallock entries
2125 * pointing to the same proclock, and we daren't end up with any dangling
2126 * pointers. Fast-path locks are cleaned up during the locallock table
2129 hash_seq_init(&status, LockMethodLocalHash);
2131 while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2134 * If the LOCALLOCK entry is unused, we must've run out of shared
2135 * memory while trying to set up this lock. Just forget the local
2138 if (locallock->nLocks == 0)
2140 RemoveLocalLock(locallock);
2144 /* Ignore items that are not of the lockmethod to be removed */
2145 if (LOCALLOCK_LOCKMETHOD(*locallock) != lockmethodid)
2149 * If we are asked to release all locks, we can just zap the entry.
2150 * Otherwise, must scan to see if there are session locks. We assume
2151 * there is at most one lockOwners entry for session locks.
2155 LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
2157 /* If session lock is above array position 0, move it down to 0 */
2158 for (i = 0; i < locallock->numLockOwners; i++)
2160 if (lockOwners[i].owner == NULL)
2161 lockOwners[0] = lockOwners[i];
2163 ResourceOwnerForgetLock(lockOwners[i].owner, locallock);
2166 if (locallock->numLockOwners > 0 &&
2167 lockOwners[0].owner == NULL &&
2168 lockOwners[0].nLocks > 0)
2170 /* Fix the locallock to show just the session locks */
2171 locallock->nLocks = lockOwners[0].nLocks;
2172 locallock->numLockOwners = 1;
2173 /* We aren't deleting this locallock, so done */
2177 locallock->numLockOwners = 0;
2181 * If the lock or proclock pointers are NULL, this lock was taken via
2182 * the relation fast-path (and is not known to have been transferred).
2184 if (locallock->proclock == NULL || locallock->lock == NULL)
2186 LOCKMODE lockmode = locallock->tag.mode;
2189 /* Verify that a fast-path lock is what we've got. */
2190 if (!EligibleForRelationFastPath(&locallock->tag.lock, lockmode))
2191 elog(PANIC, "locallock table corrupted");
2194 * If we don't currently hold the LWLock that protects our
2195 * fast-path data structures, we must acquire it before attempting
2196 * to release the lock via the fast-path. We will continue to
2197 * hold the LWLock until we're done scanning the locallock table,
2198 * unless we hit a transferred fast-path lock. (XXX is this
2199 * really such a good idea? There could be a lot of entries ...)
2201 if (!have_fast_path_lwlock)
2203 LWLockAcquire(&MyProc->backendLock, LW_EXCLUSIVE);
2204 have_fast_path_lwlock = true;
2207 /* Attempt fast-path release. */
2208 relid = locallock->tag.lock.locktag_field2;
2209 if (FastPathUnGrantRelationLock(relid, lockmode))
2211 RemoveLocalLock(locallock);
2216 * Our lock, originally taken via the fast path, has been
2217 * transferred to the main lock table. That's going to require
2218 * some extra work, so release our fast-path lock before starting.
2220 LWLockRelease(&MyProc->backendLock);
2221 have_fast_path_lwlock = false;
2224 * Now dump the lock. We haven't got a pointer to the LOCK or
2225 * PROCLOCK in this case, so we have to handle this a bit
2226 * differently than a normal lock release. Unfortunately, this
2227 * requires an extra LWLock acquire-and-release cycle on the
2228 * partitionLock, but hopefully it shouldn't happen often.
2230 LockRefindAndRelease(lockMethodTable, MyProc,
2231 &locallock->tag.lock, lockmode, false);
2232 RemoveLocalLock(locallock);
2236 /* Mark the proclock to show we need to release this lockmode */
2237 if (locallock->nLocks > 0)
2238 locallock->proclock->releaseMask |= LOCKBIT_ON(locallock->tag.mode);
2240 /* And remove the locallock hashtable entry */
2241 RemoveLocalLock(locallock);
2244 /* Done with the fast-path data structures */
2245 if (have_fast_path_lwlock)
2246 LWLockRelease(&MyProc->backendLock);
2249 * Now, scan each lock partition separately.
2251 for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
2253 LWLock *partitionLock;
2254 SHM_QUEUE *procLocks = &(MyProc->myProcLocks[partition]);
2255 PROCLOCK *nextplock;
2257 partitionLock = LockHashPartitionLockByIndex(partition);
2260 * If the proclock list for this partition is empty, we can skip
2261 * acquiring the partition lock. This optimization is trickier than
2262 * it looks, because another backend could be in process of adding
2263 * something to our proclock list due to promoting one of our
2264 * fast-path locks. However, any such lock must be one that we
2265 * decided not to delete above, so it's okay to skip it again now;
2266 * we'd just decide not to delete it again. We must, however, be
2267 * careful to re-fetch the list header once we've acquired the
2268 * partition lock, to be sure we have a valid, up-to-date pointer.
2269 * (There is probably no significant risk if pointer fetch/store is
2270 * atomic, but we don't wish to assume that.)
2272 * XXX This argument assumes that the locallock table correctly
2273 * represents all of our fast-path locks. While allLocks mode
2274 * guarantees to clean up all of our normal locks regardless of the
2275 * locallock situation, we lose that guarantee for fast-path locks.
2276 * This is not ideal.
2278 if (SHMQueueNext(procLocks, procLocks,
2279 offsetof(PROCLOCK, procLink)) == NULL)
2280 continue; /* needn't examine this partition */
2282 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2284 for (proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
2285 offsetof(PROCLOCK, procLink));
2287 proclock = nextplock)
2289 bool wakeupNeeded = false;
2291 /* Get link first, since we may unlink/delete this proclock */
2292 nextplock = (PROCLOCK *)
2293 SHMQueueNext(procLocks, &proclock->procLink,
2294 offsetof(PROCLOCK, procLink));
2296 Assert(proclock->tag.myProc == MyProc);
2298 lock = proclock->tag.myLock;
2300 /* Ignore items that are not of the lockmethod to be removed */
2301 if (LOCK_LOCKMETHOD(*lock) != lockmethodid)
2305 * In allLocks mode, force release of all locks even if locallock
2306 * table had problems
2309 proclock->releaseMask = proclock->holdMask;
2311 Assert((proclock->releaseMask & ~proclock->holdMask) == 0);
2314 * Ignore items that have nothing to be released, unless they have
2315 * holdMask == 0 and are therefore recyclable
2317 if (proclock->releaseMask == 0 && proclock->holdMask != 0)
2320 PROCLOCK_PRINT("LockReleaseAll", proclock);
2321 LOCK_PRINT("LockReleaseAll", lock, 0);
2322 Assert(lock->nRequested >= 0);
2323 Assert(lock->nGranted >= 0);
2324 Assert(lock->nGranted <= lock->nRequested);
2325 Assert((proclock->holdMask & ~lock->grantMask) == 0);
2328 * Release the previously-marked lock modes
2330 for (i = 1; i <= numLockModes; i++)
2332 if (proclock->releaseMask & LOCKBIT_ON(i))
2333 wakeupNeeded |= UnGrantLock(lock, i, proclock,
2336 Assert((lock->nRequested >= 0) && (lock->nGranted >= 0));
2337 Assert(lock->nGranted <= lock->nRequested);
2338 LOCK_PRINT("LockReleaseAll: updated", lock, 0);
2340 proclock->releaseMask = 0;
2342 /* CleanUpLock will wake up waiters if needed. */
2343 CleanUpLock(lock, proclock,
2345 LockTagHashCode(&lock->tag),
2347 } /* loop over PROCLOCKs within this partition */
2349 LWLockRelease(partitionLock);
2350 } /* loop over partitions */
2353 if (*(lockMethodTable->trace_flag))
2354 elog(LOG, "LockReleaseAll done");
2359 * LockReleaseSession -- Release all session locks of the specified lock method
2360 * that are held by the current process.
2363 LockReleaseSession(LOCKMETHODID lockmethodid)
2365 HASH_SEQ_STATUS status;
2366 LOCALLOCK *locallock;
2368 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2369 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2371 hash_seq_init(&status, LockMethodLocalHash);
2373 while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2375 /* Ignore items that are not of the specified lock method */
2376 if (LOCALLOCK_LOCKMETHOD(*locallock) != lockmethodid)
2379 ReleaseLockIfHeld(locallock, true);
2384 * LockReleaseCurrentOwner
2385 * Release all locks belonging to CurrentResourceOwner
2387 * If the caller knows what those locks are, it can pass them as an array.
2388 * That speeds up the call significantly, when a lot of locks are held.
2389 * Otherwise, pass NULL for locallocks, and we'll traverse through our hash
2390 * table to find them.
2393 LockReleaseCurrentOwner(LOCALLOCK **locallocks, int nlocks)
2395 if (locallocks == NULL)
2397 HASH_SEQ_STATUS status;
2398 LOCALLOCK *locallock;
2400 hash_seq_init(&status, LockMethodLocalHash);
2402 while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2403 ReleaseLockIfHeld(locallock, false);
2409 for (i = nlocks - 1; i >= 0; i--)
2410 ReleaseLockIfHeld(locallocks[i], false);
2416 * Release any session-level locks on this lockable object if sessionLock
2417 * is true; else, release any locks held by CurrentResourceOwner.
2419 * It is tempting to pass this a ResourceOwner pointer (or NULL for session
2420 * locks), but without refactoring LockRelease() we cannot support releasing
2421 * locks belonging to resource owners other than CurrentResourceOwner.
2422 * If we were to refactor, it'd be a good idea to fix it so we don't have to
2423 * do a hashtable lookup of the locallock, too. However, currently this
2424 * function isn't used heavily enough to justify refactoring for its
2428 ReleaseLockIfHeld(LOCALLOCK *locallock, bool sessionLock)
2430 ResourceOwner owner;
2431 LOCALLOCKOWNER *lockOwners;
2434 /* Identify owner for lock (must match LockRelease!) */
2438 owner = CurrentResourceOwner;
2440 /* Scan to see if there are any locks belonging to the target owner */
2441 lockOwners = locallock->lockOwners;
2442 for (i = locallock->numLockOwners - 1; i >= 0; i--)
2444 if (lockOwners[i].owner == owner)
2446 Assert(lockOwners[i].nLocks > 0);
2447 if (lockOwners[i].nLocks < locallock->nLocks)
2450 * We will still hold this lock after forgetting this
2453 locallock->nLocks -= lockOwners[i].nLocks;
2454 /* compact out unused slot */
2455 locallock->numLockOwners--;
2457 ResourceOwnerForgetLock(owner, locallock);
2458 if (i < locallock->numLockOwners)
2459 lockOwners[i] = lockOwners[locallock->numLockOwners];
2463 Assert(lockOwners[i].nLocks == locallock->nLocks);
2464 /* We want to call LockRelease just once */
2465 lockOwners[i].nLocks = 1;
2466 locallock->nLocks = 1;
2467 if (!LockRelease(&locallock->tag.lock,
2468 locallock->tag.mode,
2470 elog(WARNING, "ReleaseLockIfHeld: failed??");
2478 * LockReassignCurrentOwner
2479 * Reassign all locks belonging to CurrentResourceOwner to belong
2480 * to its parent resource owner.
2482 * If the caller knows what those locks are, it can pass them as an array.
2483 * That speeds up the call significantly, when a lot of locks are held
2484 * (e.g pg_dump with a large schema). Otherwise, pass NULL for locallocks,
2485 * and we'll traverse through our hash table to find them.
2488 LockReassignCurrentOwner(LOCALLOCK **locallocks, int nlocks)
2490 ResourceOwner parent = ResourceOwnerGetParent(CurrentResourceOwner);
2492 Assert(parent != NULL);
2494 if (locallocks == NULL)
2496 HASH_SEQ_STATUS status;
2497 LOCALLOCK *locallock;
2499 hash_seq_init(&status, LockMethodLocalHash);
2501 while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2502 LockReassignOwner(locallock, parent);
2508 for (i = nlocks - 1; i >= 0; i--)
2509 LockReassignOwner(locallocks[i], parent);
2514 * Subroutine of LockReassignCurrentOwner. Reassigns a given lock belonging to
2515 * CurrentResourceOwner to its parent.
2518 LockReassignOwner(LOCALLOCK *locallock, ResourceOwner parent)
2520 LOCALLOCKOWNER *lockOwners;
2526 * Scan to see if there are any locks belonging to current owner or its
2529 lockOwners = locallock->lockOwners;
2530 for (i = locallock->numLockOwners - 1; i >= 0; i--)
2532 if (lockOwners[i].owner == CurrentResourceOwner)
2534 else if (lockOwners[i].owner == parent)
2539 return; /* no current locks */
2543 /* Parent has no slot, so just give it the child's slot */
2544 lockOwners[ic].owner = parent;
2545 ResourceOwnerRememberLock(parent, locallock);
2549 /* Merge child's count with parent's */
2550 lockOwners[ip].nLocks += lockOwners[ic].nLocks;
2551 /* compact out unused slot */
2552 locallock->numLockOwners--;
2553 if (ic < locallock->numLockOwners)
2554 lockOwners[ic] = lockOwners[locallock->numLockOwners];
2556 ResourceOwnerForgetLock(CurrentResourceOwner, locallock);
2560 * FastPathGrantRelationLock
2561 * Grant lock using per-backend fast-path array, if there is space.
2564 FastPathGrantRelationLock(Oid relid, LOCKMODE lockmode)
2567 uint32 unused_slot = FP_LOCK_SLOTS_PER_BACKEND;
2569 /* Scan for existing entry for this relid, remembering empty slot. */
2570 for (f = 0; f < FP_LOCK_SLOTS_PER_BACKEND; f++)
2572 if (FAST_PATH_GET_BITS(MyProc, f) == 0)
2574 else if (MyProc->fpRelId[f] == relid)
2576 Assert(!FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode));
2577 FAST_PATH_SET_LOCKMODE(MyProc, f, lockmode);
2582 /* If no existing entry, use any empty slot. */
2583 if (unused_slot < FP_LOCK_SLOTS_PER_BACKEND)
2585 MyProc->fpRelId[unused_slot] = relid;
2586 FAST_PATH_SET_LOCKMODE(MyProc, unused_slot, lockmode);
2587 ++FastPathLocalUseCount;
2591 /* No existing entry, and no empty slot. */
2596 * FastPathUnGrantRelationLock
2597 * Release fast-path lock, if present. Update backend-private local
2598 * use count, while we're at it.
2601 FastPathUnGrantRelationLock(Oid relid, LOCKMODE lockmode)
2604 bool result = false;
2606 FastPathLocalUseCount = 0;
2607 for (f = 0; f < FP_LOCK_SLOTS_PER_BACKEND; f++)
2609 if (MyProc->fpRelId[f] == relid
2610 && FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode))
2613 FAST_PATH_CLEAR_LOCKMODE(MyProc, f, lockmode);
2615 /* we continue iterating so as to update FastPathLocalUseCount */
2617 if (FAST_PATH_GET_BITS(MyProc, f) != 0)
2618 ++FastPathLocalUseCount;
2624 * FastPathTransferRelationLocks
2625 * Transfer locks matching the given lock tag from per-backend fast-path
2626 * arrays to the shared hash table.
2628 * Returns true if successful, false if ran out of shared memory.
2631 FastPathTransferRelationLocks(LockMethod lockMethodTable, const LOCKTAG *locktag,
2634 LWLock *partitionLock = LockHashPartitionLock(hashcode);
2635 Oid relid = locktag->locktag_field2;
2639 * Every PGPROC that can potentially hold a fast-path lock is present in
2640 * ProcGlobal->allProcs. Prepared transactions are not, but any
2641 * outstanding fast-path locks held by prepared transactions are
2642 * transferred to the main lock table.
2644 for (i = 0; i < ProcGlobal->allProcCount; i++)
2646 PGPROC *proc = &ProcGlobal->allProcs[i];
2649 LWLockAcquire(&proc->backendLock, LW_EXCLUSIVE);
2652 * If the target backend isn't referencing the same database as the
2653 * lock, then we needn't examine the individual relation IDs at all;
2654 * none of them can be relevant.
2656 * proc->databaseId is set at backend startup time and never changes
2657 * thereafter, so it might be safe to perform this test before
2658 * acquiring &proc->backendLock. In particular, it's certainly safe
2659 * to assume that if the target backend holds any fast-path locks, it
2660 * must have performed a memory-fencing operation (in particular, an
2661 * LWLock acquisition) since setting proc->databaseId. However, it's
2662 * less clear that our backend is certain to have performed a memory
2663 * fencing operation since the other backend set proc->databaseId. So
2664 * for now, we test it after acquiring the LWLock just to be safe.
2666 if (proc->databaseId != locktag->locktag_field1)
2668 LWLockRelease(&proc->backendLock);
2672 for (f = 0; f < FP_LOCK_SLOTS_PER_BACKEND; f++)
2676 /* Look for an allocated slot matching the given relid. */
2677 if (relid != proc->fpRelId[f] || FAST_PATH_GET_BITS(proc, f) == 0)
2680 /* Find or create lock object. */
2681 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2682 for (lockmode = FAST_PATH_LOCKNUMBER_OFFSET;
2683 lockmode < FAST_PATH_LOCKNUMBER_OFFSET + FAST_PATH_BITS_PER_SLOT;
2688 if (!FAST_PATH_CHECK_LOCKMODE(proc, f, lockmode))
2690 proclock = SetupLockInTable(lockMethodTable, proc, locktag,
2691 hashcode, lockmode);
2694 LWLockRelease(partitionLock);
2695 LWLockRelease(&proc->backendLock);
2698 GrantLock(proclock->tag.myLock, proclock, lockmode);
2699 FAST_PATH_CLEAR_LOCKMODE(proc, f, lockmode);
2701 LWLockRelease(partitionLock);
2703 /* No need to examine remaining slots. */
2706 LWLockRelease(&proc->backendLock);
2712 * FastPathGetRelationLockEntry
2713 * Return the PROCLOCK for a lock originally taken via the fast-path,
2714 * transferring it to the primary lock table if necessary.
2716 * Note: caller takes care of updating the locallock object.
2719 FastPathGetRelationLockEntry(LOCALLOCK *locallock)
2721 LockMethod lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
2722 LOCKTAG *locktag = &locallock->tag.lock;
2723 PROCLOCK *proclock = NULL;
2724 LWLock *partitionLock = LockHashPartitionLock(locallock->hashcode);
2725 Oid relid = locktag->locktag_field2;
2728 LWLockAcquire(&MyProc->backendLock, LW_EXCLUSIVE);
2730 for (f = 0; f < FP_LOCK_SLOTS_PER_BACKEND; f++)
2734 /* Look for an allocated slot matching the given relid. */
2735 if (relid != MyProc->fpRelId[f] || FAST_PATH_GET_BITS(MyProc, f) == 0)
2738 /* If we don't have a lock of the given mode, forget it! */
2739 lockmode = locallock->tag.mode;
2740 if (!FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode))
2743 /* Find or create lock object. */
2744 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2746 proclock = SetupLockInTable(lockMethodTable, MyProc, locktag,
2747 locallock->hashcode, lockmode);
2750 LWLockRelease(partitionLock);
2751 LWLockRelease(&MyProc->backendLock);
2753 (errcode(ERRCODE_OUT_OF_MEMORY),
2754 errmsg("out of shared memory"),
2755 errhint("You might need to increase max_locks_per_transaction.")));
2757 GrantLock(proclock->tag.myLock, proclock, lockmode);
2758 FAST_PATH_CLEAR_LOCKMODE(MyProc, f, lockmode);
2760 LWLockRelease(partitionLock);
2762 /* No need to examine remaining slots. */
2766 LWLockRelease(&MyProc->backendLock);
2768 /* Lock may have already been transferred by some other backend. */
2769 if (proclock == NULL)
2772 PROCLOCKTAG proclocktag;
2773 uint32 proclock_hashcode;
2775 LWLockAcquire(partitionLock, LW_SHARED);
2777 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
2779 locallock->hashcode,
2783 elog(ERROR, "failed to re-find shared lock object");
2785 proclocktag.myLock = lock;
2786 proclocktag.myProc = MyProc;
2788 proclock_hashcode = ProcLockHashCode(&proclocktag, locallock->hashcode);
2789 proclock = (PROCLOCK *)
2790 hash_search_with_hash_value(LockMethodProcLockHash,
2791 (void *) &proclocktag,
2796 elog(ERROR, "failed to re-find shared proclock object");
2797 LWLockRelease(partitionLock);
2805 * Get an array of VirtualTransactionIds of xacts currently holding locks
2806 * that would conflict with the specified lock/lockmode.
2807 * xacts merely awaiting such a lock are NOT reported.
2809 * The result array is palloc'd and is terminated with an invalid VXID.
2810 * *countp, if not null, is updated to the number of items set.
2812 * Of course, the result could be out of date by the time it's returned,
2813 * so use of this function has to be thought about carefully.
2815 * Note we never include the current xact's vxid in the result array,
2816 * since an xact never blocks itself. Also, prepared transactions are
2817 * ignored, which is a bit more debatable but is appropriate for current
2818 * uses of the result.
2820 VirtualTransactionId *
2821 GetLockConflicts(const LOCKTAG *locktag, LOCKMODE lockmode, int *countp)
2823 static VirtualTransactionId *vxids;
2824 LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
2825 LockMethod lockMethodTable;
2827 LOCKMASK conflictMask;
2828 SHM_QUEUE *procLocks;
2831 LWLock *partitionLock;
2835 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2836 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2837 lockMethodTable = LockMethods[lockmethodid];
2838 if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
2839 elog(ERROR, "unrecognized lock mode: %d", lockmode);
2842 * Allocate memory to store results, and fill with InvalidVXID. We only
2843 * need enough space for MaxBackends + a terminator, since prepared xacts
2844 * don't count. InHotStandby allocate once in TopMemoryContext.
2849 vxids = (VirtualTransactionId *)
2850 MemoryContextAlloc(TopMemoryContext,
2851 sizeof(VirtualTransactionId) * (MaxBackends + 1));
2854 vxids = (VirtualTransactionId *)
2855 palloc0(sizeof(VirtualTransactionId) * (MaxBackends + 1));
2857 /* Compute hash code and partition lock, and look up conflicting modes. */
2858 hashcode = LockTagHashCode(locktag);
2859 partitionLock = LockHashPartitionLock(hashcode);
2860 conflictMask = lockMethodTable->conflictTab[lockmode];
2863 * Fast path locks might not have been entered in the primary lock table.
2864 * If the lock we're dealing with could conflict with such a lock, we must
2865 * examine each backend's fast-path array for conflicts.
2867 if (ConflictsWithRelationFastPath(locktag, lockmode))
2870 Oid relid = locktag->locktag_field2;
2871 VirtualTransactionId vxid;
2874 * Iterate over relevant PGPROCs. Anything held by a prepared
2875 * transaction will have been transferred to the primary lock table,
2876 * so we need not worry about those. This is all a bit fuzzy, because
2877 * new locks could be taken after we've visited a particular
2878 * partition, but the callers had better be prepared to deal with that
2879 * anyway, since the locks could equally well be taken between the
2880 * time we return the value and the time the caller does something
2883 for (i = 0; i < ProcGlobal->allProcCount; i++)
2885 PGPROC *proc = &ProcGlobal->allProcs[i];
2888 /* A backend never blocks itself */
2892 LWLockAcquire(&proc->backendLock, LW_SHARED);
2895 * If the target backend isn't referencing the same database as
2896 * the lock, then we needn't examine the individual relation IDs
2897 * at all; none of them can be relevant.
2899 * See FastPathTransferRelationLocks() for discussion of why we do
2900 * this test after acquiring the lock.
2902 if (proc->databaseId != locktag->locktag_field1)
2904 LWLockRelease(&proc->backendLock);
2908 for (f = 0; f < FP_LOCK_SLOTS_PER_BACKEND; f++)
2912 /* Look for an allocated slot matching the given relid. */
2913 if (relid != proc->fpRelId[f])
2915 lockmask = FAST_PATH_GET_BITS(proc, f);
2918 lockmask <<= FAST_PATH_LOCKNUMBER_OFFSET;
2921 * There can only be one entry per relation, so if we found it
2922 * and it doesn't conflict, we can skip the rest of the slots.
2924 if ((lockmask & conflictMask) == 0)
2928 GET_VXID_FROM_PGPROC(vxid, *proc);
2931 * If we see an invalid VXID, then either the xact has already
2932 * committed (or aborted), or it's a prepared xact. In either
2933 * case we may ignore it.
2935 if (VirtualTransactionIdIsValid(vxid))
2936 vxids[count++] = vxid;
2938 /* No need to examine remaining slots. */
2942 LWLockRelease(&proc->backendLock);
2946 /* Remember how many fast-path conflicts we found. */
2950 * Look up the lock object matching the tag.
2952 LWLockAcquire(partitionLock, LW_SHARED);
2954 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
2955 (const void *) locktag,
2962 * If the lock object doesn't exist, there is nothing holding a lock
2963 * on this lockable object.
2965 LWLockRelease(partitionLock);
2966 vxids[count].backendId = InvalidBackendId;
2967 vxids[count].localTransactionId = InvalidLocalTransactionId;
2974 * Examine each existing holder (or awaiter) of the lock.
2977 procLocks = &(lock->procLocks);
2979 proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
2980 offsetof(PROCLOCK, lockLink));
2984 if (conflictMask & proclock->holdMask)
2986 PGPROC *proc = proclock->tag.myProc;
2988 /* A backend never blocks itself */
2991 VirtualTransactionId vxid;
2993 GET_VXID_FROM_PGPROC(vxid, *proc);
2996 * If we see an invalid VXID, then either the xact has already
2997 * committed (or aborted), or it's a prepared xact. In either
2998 * case we may ignore it.
3000 if (VirtualTransactionIdIsValid(vxid))
3004 /* Avoid duplicate entries. */
3005 for (i = 0; i < fast_count; ++i)
3006 if (VirtualTransactionIdEquals(vxids[i], vxid))
3008 if (i >= fast_count)
3009 vxids[count++] = vxid;
3014 proclock = (PROCLOCK *) SHMQueueNext(procLocks, &proclock->lockLink,
3015 offsetof(PROCLOCK, lockLink));
3018 LWLockRelease(partitionLock);
3020 if (count > MaxBackends) /* should never happen */
3021 elog(PANIC, "too many conflicting locks found");
3023 vxids[count].backendId = InvalidBackendId;
3024 vxids[count].localTransactionId = InvalidLocalTransactionId;
3031 * Find a lock in the shared lock table and release it. It is the caller's
3032 * responsibility to verify that this is a sane thing to do. (For example, it
3033 * would be bad to release a lock here if there might still be a LOCALLOCK
3034 * object with pointers to it.)
3036 * We currently use this in two situations: first, to release locks held by
3037 * prepared transactions on commit (see lock_twophase_postcommit); and second,
3038 * to release locks taken via the fast-path, transferred to the main hash
3039 * table, and then released (see LockReleaseAll).
3042 LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
3043 LOCKTAG *locktag, LOCKMODE lockmode,
3044 bool decrement_strong_lock_count)
3048 PROCLOCKTAG proclocktag;
3050 uint32 proclock_hashcode;
3051 LWLock *partitionLock;
3054 hashcode = LockTagHashCode(locktag);
3055 partitionLock = LockHashPartitionLock(hashcode);
3057 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3060 * Re-find the lock object (it had better be there).
3062 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
3068 elog(PANIC, "failed to re-find shared lock object");
3071 * Re-find the proclock object (ditto).
3073 proclocktag.myLock = lock;
3074 proclocktag.myProc = proc;
3076 proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
3078 proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
3079 (void *) &proclocktag,
3084 elog(PANIC, "failed to re-find shared proclock object");
3087 * Double-check that we are actually holding a lock of the type we want to
3090 if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
3092 PROCLOCK_PRINT("lock_twophase_postcommit: WRONGTYPE", proclock);
3093 LWLockRelease(partitionLock);
3094 elog(WARNING, "you don't own a lock of type %s",
3095 lockMethodTable->lockModeNames[lockmode]);
3100 * Do the releasing. CleanUpLock will waken any now-wakable waiters.
3102 wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
3104 CleanUpLock(lock, proclock,
3105 lockMethodTable, hashcode,
3108 LWLockRelease(partitionLock);
3111 * Decrement strong lock count. This logic is needed only for 2PC.
3113 if (decrement_strong_lock_count
3114 && ConflictsWithRelationFastPath(locktag, lockmode))
3116 uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
3118 SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
3119 Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
3120 FastPathStrongRelationLocks->count[fasthashcode]--;
3121 SpinLockRelease(&FastPathStrongRelationLocks->mutex);
3127 * Do the preparatory work for a PREPARE: make 2PC state file records
3128 * for all locks currently held.
3130 * Session-level locks are ignored, as are VXID locks.
3132 * There are some special cases that we error out on: we can't be holding any
3133 * locks at both session and transaction level (since we must either keep or
3134 * give away the PROCLOCK object), and we can't be holding any locks on
3135 * temporary objects (since that would mess up the current backend if it tries
3136 * to exit before the prepared xact is committed).
3139 AtPrepare_Locks(void)
3141 HASH_SEQ_STATUS status;
3142 LOCALLOCK *locallock;
3145 * For the most part, we don't need to touch shared memory for this ---
3146 * all the necessary state information is in the locallock table.
3147 * Fast-path locks are an exception, however: we move any such locks to
3148 * the main table before allowing PREPARE TRANSACTION to succeed.
3150 hash_seq_init(&status, LockMethodLocalHash);
3152 while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3154 TwoPhaseLockRecord record;
3155 LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3156 bool haveSessionLock;
3161 * Ignore VXID locks. We don't want those to be held by prepared
3162 * transactions, since they aren't meaningful after a restart.
3164 if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3167 /* Ignore it if we don't actually hold the lock */
3168 if (locallock->nLocks <= 0)
3171 /* Scan to see whether we hold it at session or transaction level */
3172 haveSessionLock = haveXactLock = false;
3173 for (i = locallock->numLockOwners - 1; i >= 0; i--)
3175 if (lockOwners[i].owner == NULL)
3176 haveSessionLock = true;
3178 haveXactLock = true;
3181 /* Ignore it if we have only session lock */
3186 * If we have both session- and transaction-level locks, fail. This
3187 * should never happen with regular locks, since we only take those at
3188 * session level in some special operations like VACUUM. It's
3189 * possible to hit this with advisory locks, though.
3191 * It would be nice if we could keep the session hold and give away
3192 * the transactional hold to the prepared xact. However, that would
3193 * require two PROCLOCK objects, and we cannot be sure that another
3194 * PROCLOCK will be available when it comes time for PostPrepare_Locks
3195 * to do the deed. So for now, we error out while we can still do so
3198 if (haveSessionLock)
3200 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3201 errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3204 * If the local lock was taken via the fast-path, we need to move it
3205 * to the primary lock table, or just get a pointer to the existing
3206 * primary lock table entry if by chance it's already been
3209 if (locallock->proclock == NULL)
3211 locallock->proclock = FastPathGetRelationLockEntry(locallock);
3212 locallock->lock = locallock->proclock->tag.myLock;
3216 * Arrange to not release any strong lock count held by this lock
3217 * entry. We must retain the count until the prepared transaction is
3218 * committed or rolled back.
3220 locallock->holdsStrongLockCount = false;
3223 * Create a 2PC record.
3225 memcpy(&(record.locktag), &(locallock->tag.lock), sizeof(LOCKTAG));
3226 record.lockmode = locallock->tag.mode;
3228 RegisterTwoPhaseRecord(TWOPHASE_RM_LOCK_ID, 0,
3229 &record, sizeof(TwoPhaseLockRecord));
3235 * Clean up after successful PREPARE
3237 * Here, we want to transfer ownership of our locks to a dummy PGPROC
3238 * that's now associated with the prepared transaction, and we want to
3239 * clean out the corresponding entries in the LOCALLOCK table.
3241 * Note: by removing the LOCALLOCK entries, we are leaving dangling
3242 * pointers in the transaction's resource owner. This is OK at the
3243 * moment since resowner.c doesn't try to free locks retail at a toplevel
3244 * transaction commit or abort. We could alternatively zero out nLocks
3245 * and leave the LOCALLOCK entries to be garbage-collected by LockReleaseAll,
3246 * but that probably costs more cycles.
3249 PostPrepare_Locks(TransactionId xid)
3251 PGPROC *newproc = TwoPhaseGetDummyProc(xid, false);
3252 HASH_SEQ_STATUS status;
3253 LOCALLOCK *locallock;
3256 PROCLOCKTAG proclocktag;
3259 /* Can't prepare a lock group follower. */
3260 Assert(MyProc->lockGroupLeader == NULL ||
3261 MyProc->lockGroupLeader == MyProc);
3263 /* This is a critical section: any error means big trouble */
3264 START_CRIT_SECTION();
3267 * First we run through the locallock table and get rid of unwanted
3268 * entries, then we scan the process's proclocks and transfer them to the
3271 * We do this separately because we may have multiple locallock entries
3272 * pointing to the same proclock, and we daren't end up with any dangling
3275 hash_seq_init(&status, LockMethodLocalHash);
3277 while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3279 LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3280 bool haveSessionLock;
3284 if (locallock->proclock == NULL || locallock->lock == NULL)
3287 * We must've run out of shared memory while trying to set up this
3288 * lock. Just forget the local entry.
3290 Assert(locallock->nLocks == 0);
3291 RemoveLocalLock(locallock);
3295 /* Ignore VXID locks */
3296 if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3299 /* Scan to see whether we hold it at session or transaction level */
3300 haveSessionLock = haveXactLock = false;
3301 for (i = locallock->numLockOwners - 1; i >= 0; i--)
3303 if (lockOwners[i].owner == NULL)
3304 haveSessionLock = true;
3306 haveXactLock = true;
3309 /* Ignore it if we have only session lock */
3313 /* This can't happen, because we already checked it */
3314 if (haveSessionLock)
3316 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3317 errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3319 /* Mark the proclock to show we need to release this lockmode */
3320 if (locallock->nLocks > 0)
3321 locallock->proclock->releaseMask |= LOCKBIT_ON(locallock->tag.mode);
3323 /* And remove the locallock hashtable entry */
3324 RemoveLocalLock(locallock);
3328 * Now, scan each lock partition separately.
3330 for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
3332 LWLock *partitionLock;
3333 SHM_QUEUE *procLocks = &(MyProc->myProcLocks[partition]);
3334 PROCLOCK *nextplock;
3336 partitionLock = LockHashPartitionLockByIndex(partition);
3339 * If the proclock list for this partition is empty, we can skip
3340 * acquiring the partition lock. This optimization is safer than the
3341 * situation in LockReleaseAll, because we got rid of any fast-path
3342 * locks during AtPrepare_Locks, so there cannot be any case where
3343 * another backend is adding something to our lists now. For safety,
3344 * though, we code this the same way as in LockReleaseAll.
3346 if (SHMQueueNext(procLocks, procLocks,
3347 offsetof(PROCLOCK, procLink)) == NULL)
3348 continue; /* needn't examine this partition */
3350 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3352 for (proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
3353 offsetof(PROCLOCK, procLink));
3355 proclock = nextplock)
3357 /* Get link first, since we may unlink/relink this proclock */
3358 nextplock = (PROCLOCK *)
3359 SHMQueueNext(procLocks, &proclock->procLink,
3360 offsetof(PROCLOCK, procLink));
3362 Assert(proclock->tag.myProc == MyProc);
3364 lock = proclock->tag.myLock;
3366 /* Ignore VXID locks */
3367 if (lock->tag.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3370 PROCLOCK_PRINT("PostPrepare_Locks", proclock);
3371 LOCK_PRINT("PostPrepare_Locks", lock, 0);
3372 Assert(lock->nRequested >= 0);
3373 Assert(lock->nGranted >= 0);
3374 Assert(lock->nGranted <= lock->nRequested);
3375 Assert((proclock->holdMask & ~lock->grantMask) == 0);
3377 /* Ignore it if nothing to release (must be a session lock) */
3378 if (proclock->releaseMask == 0)
3381 /* Else we should be releasing all locks */
3382 if (proclock->releaseMask != proclock->holdMask)
3383 elog(PANIC, "we seem to have dropped a bit somewhere");
3386 * We cannot simply modify proclock->tag.myProc to reassign
3387 * ownership of the lock, because that's part of the hash key and
3388 * the proclock would then be in the wrong hash chain. Instead
3389 * use hash_update_hash_key. (We used to create a new hash entry,
3390 * but that risks out-of-memory failure if other processes are
3391 * busy making proclocks too.) We must unlink the proclock from
3392 * our procLink chain and put it into the new proc's chain, too.
3394 * Note: the updated proclock hash key will still belong to the
3395 * same hash partition, cf proclock_hash(). So the partition lock
3396 * we already hold is sufficient for this.
3398 SHMQueueDelete(&proclock->procLink);
3401 * Create the new hash key for the proclock.
3403 proclocktag.myLock = lock;
3404 proclocktag.myProc = newproc;
3407 * Update groupLeader pointer to point to the new proc. (We'd
3408 * better not be a member of somebody else's lock group!)
3410 Assert(proclock->groupLeader == proclock->tag.myProc);
3411 proclock->groupLeader = newproc;
3414 * Update the proclock. We should not find any existing entry for
3415 * the same hash key, since there can be only one entry for any
3416 * given lock with my own proc.
3418 if (!hash_update_hash_key(LockMethodProcLockHash,
3420 (void *) &proclocktag))
3421 elog(PANIC, "duplicate entry found while reassigning a prepared transaction's locks");
3423 /* Re-link into the new proc's proclock list */
3424 SHMQueueInsertBefore(&(newproc->myProcLocks[partition]),
3425 &proclock->procLink);
3427 PROCLOCK_PRINT("PostPrepare_Locks: updated", proclock);
3428 } /* loop over PROCLOCKs within this partition */
3430 LWLockRelease(partitionLock);
3431 } /* loop over partitions */
3438 * Estimate shared-memory space used for lock tables
3444 long max_table_size;
3446 /* lock hash table */
3447 max_table_size = NLOCKENTS();
3448 size = add_size(size, hash_estimate_size(max_table_size, sizeof(LOCK)));
3450 /* proclock hash table */
3451 max_table_size *= 2;
3452 size = add_size(size, hash_estimate_size(max_table_size, sizeof(PROCLOCK)));
3455 * Since NLOCKENTS is only an estimate, add 10% safety margin.
3457 size = add_size(size, size / 10);
3463 * GetLockStatusData - Return a summary of the lock manager's internal
3464 * status, for use in a user-level reporting function.
3466 * The return data consists of an array of LockInstanceData objects,
3467 * which are a lightly abstracted version of the PROCLOCK data structures,
3468 * i.e. there is one entry for each unique lock and interested PGPROC.
3469 * It is the caller's responsibility to match up related items (such as
3470 * references to the same lockable object or PGPROC) if wanted.
3472 * The design goal is to hold the LWLocks for as short a time as possible;
3473 * thus, this function simply makes a copy of the necessary data and releases
3474 * the locks, allowing the caller to contemplate and format the data for as
3475 * long as it pleases.
3478 GetLockStatusData(void)
3482 HASH_SEQ_STATUS seqstat;
3487 data = (LockData *) palloc(sizeof(LockData));
3489 /* Guess how much space we'll need. */
3492 data->locks = (LockInstanceData *) palloc(sizeof(LockInstanceData) * els);
3495 * First, we iterate through the per-backend fast-path arrays, locking
3496 * them one at a time. This might produce an inconsistent picture of the
3497 * system state, but taking all of those LWLocks at the same time seems
3498 * impractical (in particular, note MAX_SIMUL_LWLOCKS). It shouldn't
3499 * matter too much, because none of these locks can be involved in lock
3500 * conflicts anyway - anything that might must be present in the main lock
3501 * table. (For the same reason, we don't sweat about making leaderPid
3502 * completely valid. We cannot safely dereference another backend's
3503 * lockGroupLeader field without holding all lock partition locks, and
3504 * it's not worth that.)
3506 for (i = 0; i < ProcGlobal->allProcCount; ++i)
3508 PGPROC *proc = &ProcGlobal->allProcs[i];
3511 LWLockAcquire(&proc->backendLock, LW_SHARED);
3513 for (f = 0; f < FP_LOCK_SLOTS_PER_BACKEND; ++f)
3515 LockInstanceData *instance;
3516 uint32 lockbits = FAST_PATH_GET_BITS(proc, f);
3518 /* Skip unallocated slots. */
3525 data->locks = (LockInstanceData *)
3526 repalloc(data->locks, sizeof(LockInstanceData) * els);
3529 instance = &data->locks[el];
3530 SET_LOCKTAG_RELATION(instance->locktag, proc->databaseId,
3532 instance->holdMask = lockbits << FAST_PATH_LOCKNUMBER_OFFSET;
3533 instance->waitLockMode = NoLock;
3534 instance->backend = proc->backendId;
3535 instance->lxid = proc->lxid;
3536 instance->pid = proc->pid;
3537 instance->leaderPid = proc->pid;
3538 instance->fastpath = true;
3543 if (proc->fpVXIDLock)
3545 VirtualTransactionId vxid;
3546 LockInstanceData *instance;
3551 data->locks = (LockInstanceData *)
3552 repalloc(data->locks, sizeof(LockInstanceData) * els);
3555 vxid.backendId = proc->backendId;
3556 vxid.localTransactionId = proc->fpLocalTransactionId;
3558 instance = &data->locks[el];
3559 SET_LOCKTAG_VIRTUALTRANSACTION(instance->locktag, vxid);
3560 instance->holdMask = LOCKBIT_ON(ExclusiveLock);
3561 instance->waitLockMode = NoLock;
3562 instance->backend = proc->backendId;
3563 instance->lxid = proc->lxid;
3564 instance->pid = proc->pid;
3565 instance->leaderPid = proc->pid;
3566 instance->fastpath = true;
3571 LWLockRelease(&proc->backendLock);
3575 * Next, acquire lock on the entire shared lock data structure. We do
3576 * this so that, at least for locks in the primary lock table, the state
3577 * will be self-consistent.
3579 * Since this is a read-only operation, we take shared instead of
3580 * exclusive lock. There's not a whole lot of point to this, because all
3581 * the normal operations require exclusive lock, but it doesn't hurt
3582 * anything either. It will at least allow two backends to do
3583 * GetLockStatusData in parallel.
3585 * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
3587 for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
3588 LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
3590 /* Now we can safely count the number of proclocks */
3591 data->nelements = el + hash_get_num_entries(LockMethodProcLockHash);
3592 if (data->nelements > els)
3594 els = data->nelements;
3595 data->locks = (LockInstanceData *)
3596 repalloc(data->locks, sizeof(LockInstanceData) * els);
3599 /* Now scan the tables to copy the data */
3600 hash_seq_init(&seqstat, LockMethodProcLockHash);
3602 while ((proclock = (PROCLOCK *) hash_seq_search(&seqstat)))
3604 PGPROC *proc = proclock->tag.myProc;
3605 LOCK *lock = proclock->tag.myLock;
3606 LockInstanceData *instance = &data->locks[el];
3608 memcpy(&instance->locktag, &lock->tag, sizeof(LOCKTAG));
3609 instance->holdMask = proclock->holdMask;
3610 if (proc->waitLock == proclock->tag.myLock)
3611 instance->waitLockMode = proc->waitLockMode;
3613 instance->waitLockMode = NoLock;
3614 instance->backend = proc->backendId;
3615 instance->lxid = proc->lxid;
3616 instance->pid = proc->pid;
3617 instance->leaderPid = proclock->groupLeader->pid;
3618 instance->fastpath = false;
3624 * And release locks. We do this in reverse order for two reasons: (1)
3625 * Anyone else who needs more than one of the locks will be trying to lock
3626 * them in increasing order; we don't want to release the other process
3627 * until it can get all the locks it needs. (2) This avoids O(N^2)
3628 * behavior inside LWLockRelease.
3630 for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
3631 LWLockRelease(LockHashPartitionLockByIndex(i));
3633 Assert(el == data->nelements);
3639 * GetBlockerStatusData - Return a summary of the lock manager's state
3640 * concerning locks that are blocking the specified PID or any member of
3641 * the PID's lock group, for use in a user-level reporting function.
3643 * For each PID within the lock group that is awaiting some heavyweight lock,
3644 * the return data includes an array of LockInstanceData objects, which are
3645 * the same data structure used by GetLockStatusData; but unlike that function,
3646 * this one reports only the PROCLOCKs associated with the lock that that PID
3647 * is blocked on. (Hence, all the locktags should be the same for any one
3648 * blocked PID.) In addition, we return an array of the PIDs of those backends
3649 * that are ahead of the blocked PID in the lock's wait queue. These can be
3650 * compared with the PIDs in the LockInstanceData objects to determine which
3651 * waiters are ahead of or behind the blocked PID in the queue.
3653 * If blocked_pid isn't a valid backend PID or nothing in its lock group is
3654 * waiting on any heavyweight lock, return empty arrays.
3656 * The design goal is to hold the LWLocks for as short a time as possible;
3657 * thus, this function simply makes a copy of the necessary data and releases
3658 * the locks, allowing the caller to contemplate and format the data for as
3659 * long as it pleases.
3662 GetBlockerStatusData(int blocked_pid)
3664 BlockedProcsData *data;
3668 data = (BlockedProcsData *) palloc(sizeof(BlockedProcsData));
3671 * Guess how much space we'll need, and preallocate. Most of the time
3672 * this will avoid needing to do repalloc while holding the LWLocks. (We
3673 * assume, but check with an Assert, that MaxBackends is enough entries
3674 * for the procs[] array; the other two could need enlargement, though.)
3676 data->nprocs = data->nlocks = data->npids = 0;
3677 data->maxprocs = data->maxlocks = data->maxpids = MaxBackends;
3678 data->procs = (BlockedProcData *) palloc(sizeof(BlockedProcData) * data->maxprocs);
3679 data->locks = (LockInstanceData *) palloc(sizeof(LockInstanceData) * data->maxlocks);
3680 data->waiter_pids = (int *) palloc(sizeof(int) * data->maxpids);
3683 * In order to search the ProcArray for blocked_pid and assume that that
3684 * entry won't immediately disappear under us, we must hold ProcArrayLock.
3685 * In addition, to examine the lock grouping fields of any other backend,
3686 * we must hold all the hash partition locks. (Only one of those locks is
3687 * actually relevant for any one lock group, but we can't know which one
3688 * ahead of time.) It's fairly annoying to hold all those locks
3689 * throughout this, but it's no worse than GetLockStatusData(), and it
3690 * does have the advantage that we're guaranteed to return a
3691 * self-consistent instantaneous state.
3693 LWLockAcquire(ProcArrayLock, LW_SHARED);
3695 proc = BackendPidGetProcWithLock(blocked_pid);
3697 /* Nothing to do if it's gone */
3701 * Acquire lock on the entire shared lock data structure. See notes
3702 * in GetLockStatusData().
3704 for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
3705 LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
3707 if (proc->lockGroupLeader == NULL)
3709 /* Easy case, proc is not a lock group member */
3710 GetSingleProcBlockerStatusData(proc, data);
3714 /* Examine all procs in proc's lock group */
3717 dlist_foreach(iter, &proc->lockGroupLeader->lockGroupMembers)
3721 memberProc = dlist_container(PGPROC, lockGroupLink, iter.cur);
3722 GetSingleProcBlockerStatusData(memberProc, data);
3727 * And release locks. See notes in GetLockStatusData().
3729 for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
3730 LWLockRelease(LockHashPartitionLockByIndex(i));
3732 Assert(data->nprocs <= data->maxprocs);
3735 LWLockRelease(ProcArrayLock);
3740 /* Accumulate data about one possibly-blocked proc for GetBlockerStatusData */
3742 GetSingleProcBlockerStatusData(PGPROC *blocked_proc, BlockedProcsData *data)
3744 LOCK *theLock = blocked_proc->waitLock;
3745 BlockedProcData *bproc;
3746 SHM_QUEUE *procLocks;
3748 PROC_QUEUE *waitQueue;
3753 /* Nothing to do if this proc is not blocked */
3754 if (theLock == NULL)
3757 /* Set up a procs[] element */
3758 bproc = &data->procs[data->nprocs++];
3759 bproc->pid = blocked_proc->pid;
3760 bproc->first_lock = data->nlocks;
3761 bproc->first_waiter = data->npids;
3764 * We may ignore the proc's fast-path arrays, since nothing in those could
3765 * be related to a contended lock.
3768 /* Collect all PROCLOCKs associated with theLock */
3769 procLocks = &(theLock->procLocks);
3770 proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
3771 offsetof(PROCLOCK, lockLink));
3774 PGPROC *proc = proclock->tag.myProc;
3775 LOCK *lock = proclock->tag.myLock;
3776 LockInstanceData *instance;
3778 if (data->nlocks >= data->maxlocks)
3780 data->maxlocks += MaxBackends;
3781 data->locks = (LockInstanceData *)
3782 repalloc(data->locks, sizeof(LockInstanceData) * data->maxlocks);
3785 instance = &data->locks[data->nlocks];
3786 memcpy(&instance->locktag, &lock->tag, sizeof(LOCKTAG));
3787 instance->holdMask = proclock->holdMask;
3788 if (proc->waitLock == lock)
3789 instance->waitLockMode = proc->waitLockMode;
3791 instance->waitLockMode = NoLock;
3792 instance->backend = proc->backendId;
3793 instance->lxid = proc->lxid;
3794 instance->pid = proc->pid;
3795 instance->leaderPid = proclock->groupLeader->pid;
3796 instance->fastpath = false;
3799 proclock = (PROCLOCK *) SHMQueueNext(procLocks, &proclock->lockLink,
3800 offsetof(PROCLOCK, lockLink));
3803 /* Enlarge waiter_pids[] if it's too small to hold all wait queue PIDs */
3804 waitQueue = &(theLock->waitProcs);
3805 queue_size = waitQueue->size;
3807 if (queue_size > data->maxpids - data->npids)
3809 data->maxpids = Max(data->maxpids + MaxBackends,
3810 data->npids + queue_size);
3811 data->waiter_pids = (int *) repalloc(data->waiter_pids,
3812 sizeof(int) * data->maxpids);
3815 /* Collect PIDs from the lock's wait queue, stopping at blocked_proc */
3816 proc = (PGPROC *) waitQueue->links.next;
3817 for (i = 0; i < queue_size; i++)
3819 if (proc == blocked_proc)
3821 data->waiter_pids[data->npids++] = proc->pid;
3822 proc = (PGPROC *) proc->links.next;
3825 bproc->num_locks = data->nlocks - bproc->first_lock;
3826 bproc->num_waiters = data->npids - bproc->first_waiter;
3830 * Returns a list of currently held AccessExclusiveLocks, for use by
3831 * LogStandbySnapshot(). The result is a palloc'd array,
3832 * with the number of elements returned into *nlocks.
3834 * XXX This currently takes a lock on all partitions of the lock table,
3835 * but it's possible to do better. By reference counting locks and storing
3836 * the value in the ProcArray entry for each backend we could tell if any
3837 * locks need recording without having to acquire the partition locks and
3838 * scan the lock table. Whether that's worth the additional overhead
3839 * is pretty dubious though.
3842 GetRunningTransactionLocks(int *nlocks)
3844 xl_standby_lock *accessExclusiveLocks;
3846 HASH_SEQ_STATUS seqstat;
3852 * Acquire lock on the entire shared lock data structure.
3854 * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
3856 for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
3857 LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
3859 /* Now we can safely count the number of proclocks */
3860 els = hash_get_num_entries(LockMethodProcLockHash);
3863 * Allocating enough space for all locks in the lock table is overkill,
3864 * but it's more convenient and faster than having to enlarge the array.
3866 accessExclusiveLocks = palloc(els * sizeof(xl_standby_lock));
3868 /* Now scan the tables to copy the data */
3869 hash_seq_init(&seqstat, LockMethodProcLockHash);
3872 * If lock is a currently granted AccessExclusiveLock then it will have
3873 * just one proclock holder, so locks are never accessed twice in this
3874 * particular case. Don't copy this code for use elsewhere because in the
3875 * general case this will give you duplicate locks when looking at
3876 * non-exclusive lock types.
3879 while ((proclock = (PROCLOCK *) hash_seq_search(&seqstat)))
3881 /* make sure this definition matches the one used in LockAcquire */
3882 if ((proclock->holdMask & LOCKBIT_ON(AccessExclusiveLock)) &&
3883 proclock->tag.myLock->tag.locktag_type == LOCKTAG_RELATION)
3885 PGPROC *proc = proclock->tag.myProc;
3886 PGXACT *pgxact = &ProcGlobal->allPgXact[proc->pgprocno];
3887 LOCK *lock = proclock->tag.myLock;
3888 TransactionId xid = pgxact->xid;
3891 * Don't record locks for transactions if we know they have
3892 * already issued their WAL record for commit but not yet released
3893 * lock. It is still possible that we see locks held by already
3894 * complete transactions, if they haven't yet zeroed their xids.
3896 if (!TransactionIdIsValid(xid))
3899 accessExclusiveLocks[index].xid = xid;
3900 accessExclusiveLocks[index].dbOid = lock->tag.locktag_field1;
3901 accessExclusiveLocks[index].relOid = lock->tag.locktag_field2;
3907 Assert(index <= els);
3910 * And release locks. We do this in reverse order for two reasons: (1)
3911 * Anyone else who needs more than one of the locks will be trying to lock
3912 * them in increasing order; we don't want to release the other process
3913 * until it can get all the locks it needs. (2) This avoids O(N^2)
3914 * behavior inside LWLockRelease.
3916 for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
3917 LWLockRelease(LockHashPartitionLockByIndex(i));
3920 return accessExclusiveLocks;
3923 /* Provide the textual name of any lock mode */
3925 GetLockmodeName(LOCKMETHODID lockmethodid, LOCKMODE mode)
3927 Assert(lockmethodid > 0 && lockmethodid < lengthof(LockMethods));
3928 Assert(mode > 0 && mode <= LockMethods[lockmethodid]->numLockModes);
3929 return LockMethods[lockmethodid]->lockModeNames[mode];
3934 * Dump all locks in the given proc's myProcLocks lists.
3936 * Caller is responsible for having acquired appropriate LWLocks.
3939 DumpLocks(PGPROC *proc)
3941 SHM_QUEUE *procLocks;
3950 LOCK_PRINT("DumpLocks: waiting on", proc->waitLock, 0);
3952 for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
3954 procLocks = &(proc->myProcLocks[i]);
3956 proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
3957 offsetof(PROCLOCK, procLink));
3961 Assert(proclock->tag.myProc == proc);
3963 lock = proclock->tag.myLock;
3965 PROCLOCK_PRINT("DumpLocks", proclock);
3966 LOCK_PRINT("DumpLocks", lock, 0);
3968 proclock = (PROCLOCK *)
3969 SHMQueueNext(procLocks, &proclock->procLink,
3970 offsetof(PROCLOCK, procLink));
3976 * Dump all lmgr locks.
3978 * Caller is responsible for having acquired appropriate LWLocks.
3986 HASH_SEQ_STATUS status;
3990 if (proc && proc->waitLock)
3991 LOCK_PRINT("DumpAllLocks: waiting on", proc->waitLock, 0);
3993 hash_seq_init(&status, LockMethodProcLockHash);
3995 while ((proclock = (PROCLOCK *) hash_seq_search(&status)) != NULL)
3997 PROCLOCK_PRINT("DumpAllLocks", proclock);
3999 lock = proclock->tag.myLock;
4001 LOCK_PRINT("DumpAllLocks", lock, 0);
4003 elog(LOG, "DumpAllLocks: proclock->tag.myLock = NULL");
4006 #endif /* LOCK_DEBUG */
4009 * LOCK 2PC resource manager's routines
4013 * Re-acquire a lock belonging to a transaction that was prepared.
4015 * Because this function is run at db startup, re-acquiring the locks should
4016 * never conflict with running transactions because there are none. We
4017 * assume that the lock state represented by the stored 2PC files is legal.
4019 * When switching from Hot Standby mode to normal operation, the locks will
4020 * be already held by the startup process. The locks are acquired for the new
4021 * procs without checking for conflicts, so we don't get a conflict between the
4022 * startup process and the dummy procs, even though we will momentarily have
4023 * a situation where two procs are holding the same AccessExclusiveLock,
4024 * which isn't normally possible because the conflict. If we're in standby
4025 * mode, but a recovery snapshot hasn't been established yet, it's possible
4026 * that some but not all of the locks are already held by the startup process.
4028 * This approach is simple, but also a bit dangerous, because if there isn't
4029 * enough shared memory to acquire the locks, an error will be thrown, which
4030 * is promoted to FATAL and recovery will abort, bringing down postmaster.
4031 * A safer approach would be to transfer the locks like we do in
4032 * AtPrepare_Locks, but then again, in hot standby mode it's possible for
4033 * read-only backends to use up all the shared lock memory anyway, so that
4034 * replaying the WAL record that needs to acquire a lock will throw an error
4038 lock_twophase_recover(TransactionId xid, uint16 info,
4039 void *recdata, uint32 len)
4041 TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4042 PGPROC *proc = TwoPhaseGetDummyProc(xid, false);
4045 LOCKMETHODID lockmethodid;
4048 PROCLOCKTAG proclocktag;
4051 uint32 proclock_hashcode;
4053 LWLock *partitionLock;
4054 LockMethod lockMethodTable;
4056 Assert(len == sizeof(TwoPhaseLockRecord));
4057 locktag = &rec->locktag;
4058 lockmode = rec->lockmode;
4059 lockmethodid = locktag->locktag_lockmethodid;
4061 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4062 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4063 lockMethodTable = LockMethods[lockmethodid];
4065 hashcode = LockTagHashCode(locktag);
4066 partition = LockHashPartition(hashcode);
4067 partitionLock = LockHashPartitionLock(hashcode);
4069 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4072 * Find or create a lock with this tag.
4074 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
4081 LWLockRelease(partitionLock);
4083 (errcode(ERRCODE_OUT_OF_MEMORY),
4084 errmsg("out of shared memory"),
4085 errhint("You might need to increase max_locks_per_transaction.")));
4089 * if it's a new lock object, initialize it
4093 lock->grantMask = 0;
4095 SHMQueueInit(&(lock->procLocks));
4096 ProcQueueInit(&(lock->waitProcs));
4097 lock->nRequested = 0;
4099 MemSet(lock->requested, 0, sizeof(int) * MAX_LOCKMODES);
4100 MemSet(lock->granted, 0, sizeof(int) * MAX_LOCKMODES);
4101 LOCK_PRINT("lock_twophase_recover: new", lock, lockmode);
4105 LOCK_PRINT("lock_twophase_recover: found", lock, lockmode);
4106 Assert((lock->nRequested >= 0) && (lock->requested[lockmode] >= 0));
4107 Assert((lock->nGranted >= 0) && (lock->granted[lockmode] >= 0));
4108 Assert(lock->nGranted <= lock->nRequested);
4112 * Create the hash key for the proclock table.
4114 proclocktag.myLock = lock;
4115 proclocktag.myProc = proc;
4117 proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
4120 * Find or create a proclock entry with this tag
4122 proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
4123 (void *) &proclocktag,
4129 /* Oops, not enough shmem for the proclock */
4130 if (lock->nRequested == 0)
4133 * There are no other requestors of this lock, so garbage-collect
4134 * the lock object. We *must* do this to avoid a permanent leak
4135 * of shared memory, because there won't be anything to cause
4136 * anyone to release the lock object later.
4138 Assert(SHMQueueEmpty(&(lock->procLocks)));
4139 if (!hash_search_with_hash_value(LockMethodLockHash,
4140 (void *) &(lock->tag),
4144 elog(PANIC, "lock table corrupted");
4146 LWLockRelease(partitionLock);
4148 (errcode(ERRCODE_OUT_OF_MEMORY),
4149 errmsg("out of shared memory"),
4150 errhint("You might need to increase max_locks_per_transaction.")));
4154 * If new, initialize the new entry
4158 Assert(proc->lockGroupLeader == NULL);
4159 proclock->groupLeader = proc;
4160 proclock->holdMask = 0;
4161 proclock->releaseMask = 0;
4162 /* Add proclock to appropriate lists */
4163 SHMQueueInsertBefore(&lock->procLocks, &proclock->lockLink);
4164 SHMQueueInsertBefore(&(proc->myProcLocks[partition]),
4165 &proclock->procLink);
4166 PROCLOCK_PRINT("lock_twophase_recover: new", proclock);
4170 PROCLOCK_PRINT("lock_twophase_recover: found", proclock);
4171 Assert((proclock->holdMask & ~lock->grantMask) == 0);
4175 * lock->nRequested and lock->requested[] count the total number of
4176 * requests, whether granted or waiting, so increment those immediately.
4179 lock->requested[lockmode]++;
4180 Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
4183 * We shouldn't already hold the desired lock.
4185 if (proclock->holdMask & LOCKBIT_ON(lockmode))
4186 elog(ERROR, "lock %s on object %u/%u/%u is already held",
4187 lockMethodTable->lockModeNames[lockmode],
4188 lock->tag.locktag_field1, lock->tag.locktag_field2,
4189 lock->tag.locktag_field3);
4192 * We ignore any possible conflicts and just grant ourselves the lock. Not
4193 * only because we don't bother, but also to avoid deadlocks when
4194 * switching from standby to normal mode. See function comment.
4196 GrantLock(lock, proclock, lockmode);
4199 * Bump strong lock count, to make sure any fast-path lock requests won't
4200 * be granted without consulting the primary lock table.
4202 if (ConflictsWithRelationFastPath(&lock->tag, lockmode))
4204 uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
4206 SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
4207 FastPathStrongRelationLocks->count[fasthashcode]++;
4208 SpinLockRelease(&FastPathStrongRelationLocks->mutex);
4211 LWLockRelease(partitionLock);
4215 * Re-acquire a lock belonging to a transaction that was prepared, when
4216 * starting up into hot standby mode.
4219 lock_twophase_standby_recover(TransactionId xid, uint16 info,
4220 void *recdata, uint32 len)
4222 TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4225 LOCKMETHODID lockmethodid;
4227 Assert(len == sizeof(TwoPhaseLockRecord));
4228 locktag = &rec->locktag;
4229 lockmode = rec->lockmode;
4230 lockmethodid = locktag->locktag_lockmethodid;
4232 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4233 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4235 if (lockmode == AccessExclusiveLock &&
4236 locktag->locktag_type == LOCKTAG_RELATION)
4238 StandbyAcquireAccessExclusiveLock(xid,
4239 locktag->locktag_field1 /* dboid */ ,
4240 locktag->locktag_field2 /* reloid */ );
4246 * 2PC processing routine for COMMIT PREPARED case.
4248 * Find and release the lock indicated by the 2PC record.
4251 lock_twophase_postcommit(TransactionId xid, uint16 info,
4252 void *recdata, uint32 len)
4254 TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4255 PGPROC *proc = TwoPhaseGetDummyProc(xid, true);
4257 LOCKMETHODID lockmethodid;
4258 LockMethod lockMethodTable;
4260 Assert(len == sizeof(TwoPhaseLockRecord));
4261 locktag = &rec->locktag;
4262 lockmethodid = locktag->locktag_lockmethodid;
4264 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4265 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4266 lockMethodTable = LockMethods[lockmethodid];
4268 LockRefindAndRelease(lockMethodTable, proc, locktag, rec->lockmode, true);
4272 * 2PC processing routine for ROLLBACK PREPARED case.
4274 * This is actually just the same as the COMMIT case.
4277 lock_twophase_postabort(TransactionId xid, uint16 info,
4278 void *recdata, uint32 len)
4280 lock_twophase_postcommit(xid, info, recdata, len);
4284 * VirtualXactLockTableInsert
4286 * Take vxid lock via the fast-path. There can't be any pre-existing
4287 * lockers, as we haven't advertised this vxid via the ProcArray yet.
4289 * Since MyProc->fpLocalTransactionId will normally contain the same data
4290 * as MyProc->lxid, you might wonder if we really need both. The
4291 * difference is that MyProc->lxid is set and cleared unlocked, and
4292 * examined by procarray.c, while fpLocalTransactionId is protected by
4293 * backendLock and is used only by the locking subsystem. Doing it this
4294 * way makes it easier to verify that there are no funny race conditions.
4296 * We don't bother recording this lock in the local lock table, since it's
4297 * only ever released at the end of a transaction. Instead,
4298 * LockReleaseAll() calls VirtualXactLockTableCleanup().
4301 VirtualXactLockTableInsert(VirtualTransactionId vxid)
4303 Assert(VirtualTransactionIdIsValid(vxid));
4305 LWLockAcquire(&MyProc->backendLock, LW_EXCLUSIVE);
4307 Assert(MyProc->backendId == vxid.backendId);
4308 Assert(MyProc->fpLocalTransactionId == InvalidLocalTransactionId);
4309 Assert(MyProc->fpVXIDLock == false);
4311 MyProc->fpVXIDLock = true;
4312 MyProc->fpLocalTransactionId = vxid.localTransactionId;
4314 LWLockRelease(&MyProc->backendLock);
4318 * VirtualXactLockTableCleanup
4320 * Check whether a VXID lock has been materialized; if so, release it,
4321 * unblocking waiters.
4324 VirtualXactLockTableCleanup(void)
4327 LocalTransactionId lxid;
4329 Assert(MyProc->backendId != InvalidBackendId);
4332 * Clean up shared memory state.
4334 LWLockAcquire(&MyProc->backendLock, LW_EXCLUSIVE);
4336 fastpath = MyProc->fpVXIDLock;
4337 lxid = MyProc->fpLocalTransactionId;
4338 MyProc->fpVXIDLock = false;
4339 MyProc->fpLocalTransactionId = InvalidLocalTransactionId;
4341 LWLockRelease(&MyProc->backendLock);
4344 * If fpVXIDLock has been cleared without touching fpLocalTransactionId,
4345 * that means someone transferred the lock to the main lock table.
4347 if (!fastpath && LocalTransactionIdIsValid(lxid))
4349 VirtualTransactionId vxid;
4352 vxid.backendId = MyBackendId;
4353 vxid.localTransactionId = lxid;
4354 SET_LOCKTAG_VIRTUALTRANSACTION(locktag, vxid);
4356 LockRefindAndRelease(LockMethods[DEFAULT_LOCKMETHOD], MyProc,
4357 &locktag, ExclusiveLock, false);
4364 * If wait = true, wait until the given VXID has been released, and then
4367 * If wait = false, just check whether the VXID is still running, and return
4371 VirtualXactLock(VirtualTransactionId vxid, bool wait)
4376 Assert(VirtualTransactionIdIsValid(vxid));
4378 SET_LOCKTAG_VIRTUALTRANSACTION(tag, vxid);
4381 * If a lock table entry must be made, this is the PGPROC on whose behalf
4382 * it must be done. Note that the transaction might end or the PGPROC
4383 * might be reassigned to a new backend before we get around to examining
4384 * it, but it doesn't matter. If we find upon examination that the
4385 * relevant lxid is no longer running here, that's enough to prove that
4386 * it's no longer running anywhere.
4388 proc = BackendIdGetProc(vxid.backendId);
4393 * We must acquire this lock before checking the backendId and lxid
4394 * against the ones we're waiting for. The target backend will only set
4395 * or clear lxid while holding this lock.
4397 LWLockAcquire(&proc->backendLock, LW_EXCLUSIVE);
4399 /* If the transaction has ended, our work here is done. */
4400 if (proc->backendId != vxid.backendId
4401 || proc->fpLocalTransactionId != vxid.localTransactionId)
4403 LWLockRelease(&proc->backendLock);
4408 * If we aren't asked to wait, there's no need to set up a lock table
4409 * entry. The transaction is still in progress, so just return false.
4413 LWLockRelease(&proc->backendLock);
4418 * OK, we're going to need to sleep on the VXID. But first, we must set
4419 * up the primary lock table entry, if needed (ie, convert the proc's
4420 * fast-path lock on its VXID to a regular lock).
4422 if (proc->fpVXIDLock)
4426 LWLock *partitionLock;
4428 hashcode = LockTagHashCode(&tag);
4430 partitionLock = LockHashPartitionLock(hashcode);
4431 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4433 proclock = SetupLockInTable(LockMethods[DEFAULT_LOCKMETHOD], proc,
4434 &tag, hashcode, ExclusiveLock);
4437 LWLockRelease(partitionLock);
4438 LWLockRelease(&proc->backendLock);
4440 (errcode(ERRCODE_OUT_OF_MEMORY),
4441 errmsg("out of shared memory"),
4442 errhint("You might need to increase max_locks_per_transaction.")));
4444 GrantLock(proclock->tag.myLock, proclock, ExclusiveLock);
4446 LWLockRelease(partitionLock);
4448 proc->fpVXIDLock = false;
4451 /* Done with proc->fpLockBits */
4452 LWLockRelease(&proc->backendLock);
4455 (void) LockAcquire(&tag, ShareLock, false, false);
4457 LockRelease(&tag, ShareLock, false);
4464 * Find the number of lock requester on this locktag
4467 LockWaiterCount(const LOCKTAG *locktag)
4469 LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
4473 LWLock *partitionLock;
4476 if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4477 elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4479 hashcode = LockTagHashCode(locktag);
4480 partitionLock = LockHashPartitionLock(hashcode);
4481 LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4483 lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
4484 (const void *) locktag,
4490 Assert(lock != NULL);
4491 waiters = lock->nRequested;
4493 LWLockRelease(partitionLock);
projects / postgresql / blob