Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * lock.c
4 : * POSTGRES primary lock mechanism
5 : *
6 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/storage/lmgr/lock.c
12 : *
13 : * NOTES
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.
18 : *
19 : * For the most part, this code should be invoked via lmgr.c
20 : * or another lock-management module, not directly.
21 : *
22 : * Interface:
23 : *
24 : * LockManagerShmemInit(), GetLocksMethodTable(), GetLockTagsMethodTable(),
25 : * LockAcquire(), LockRelease(), LockReleaseAll(),
26 : * LockCheckConflicts(), GrantLock()
27 : *
28 : *-------------------------------------------------------------------------
29 : */
30 : #include "postgres.h"
31 :
32 : #include <signal.h>
33 : #include <unistd.h>
34 :
35 : #include "access/transam.h"
36 : #include "access/twophase.h"
37 : #include "access/twophase_rmgr.h"
38 : #include "access/xlog.h"
39 : #include "access/xlogutils.h"
40 : #include "miscadmin.h"
41 : #include "pg_trace.h"
42 : #include "storage/lmgr.h"
43 : #include "storage/proc.h"
44 : #include "storage/procarray.h"
45 : #include "storage/spin.h"
46 : #include "storage/standby.h"
47 : #include "utils/memutils.h"
48 : #include "utils/ps_status.h"
49 : #include "utils/resowner.h"
50 :
51 :
52 : /* GUC variables */
53 : int max_locks_per_xact; /* used to set the lock table size */
54 : bool log_lock_failure = false;
55 :
56 : #define NLOCKENTS() \
57 : mul_size(max_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
58 :
59 :
60 : /*
61 : * Data structures defining the semantics of the standard lock methods.
62 : *
63 : * The conflict table defines the semantics of the various lock modes.
64 : */
65 : static const LOCKMASK LockConflicts[] = {
66 : 0,
67 :
68 : /* AccessShareLock */
69 : LOCKBIT_ON(AccessExclusiveLock),
70 :
71 : /* RowShareLock */
72 : LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
73 :
74 : /* RowExclusiveLock */
75 : LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
76 : LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
77 :
78 : /* ShareUpdateExclusiveLock */
79 : LOCKBIT_ON(ShareUpdateExclusiveLock) |
80 : LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
81 : LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
82 :
83 : /* ShareLock */
84 : LOCKBIT_ON(RowExclusiveLock) | LOCKBIT_ON(ShareUpdateExclusiveLock) |
85 : LOCKBIT_ON(ShareRowExclusiveLock) |
86 : LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
87 :
88 : /* ShareRowExclusiveLock */
89 : LOCKBIT_ON(RowExclusiveLock) | LOCKBIT_ON(ShareUpdateExclusiveLock) |
90 : LOCKBIT_ON(ShareLock) | LOCKBIT_ON(ShareRowExclusiveLock) |
91 : LOCKBIT_ON(ExclusiveLock) | LOCKBIT_ON(AccessExclusiveLock),
92 :
93 : /* ExclusiveLock */
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),
98 :
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)
104 :
105 : };
106 :
107 : /* Names of lock modes, for debug printouts */
108 : static const char *const lock_mode_names[] =
109 : {
110 : "INVALID",
111 : "AccessShareLock",
112 : "RowShareLock",
113 : "RowExclusiveLock",
114 : "ShareUpdateExclusiveLock",
115 : "ShareLock",
116 : "ShareRowExclusiveLock",
117 : "ExclusiveLock",
118 : "AccessExclusiveLock"
119 : };
120 :
121 : #ifndef LOCK_DEBUG
122 : static bool Dummy_trace = false;
123 : #endif
124 :
125 : static const LockMethodData default_lockmethod = {
126 : MaxLockMode,
127 : LockConflicts,
128 : lock_mode_names,
129 : #ifdef LOCK_DEBUG
130 : &Trace_locks
131 : #else
132 : &Dummy_trace
133 : #endif
134 : };
135 :
136 : static const LockMethodData user_lockmethod = {
137 : MaxLockMode,
138 : LockConflicts,
139 : lock_mode_names,
140 : #ifdef LOCK_DEBUG
141 : &Trace_userlocks
142 : #else
143 : &Dummy_trace
144 : #endif
145 : };
146 :
147 : /*
148 : * map from lock method id to the lock table data structures
149 : */
150 : static const LockMethod LockMethods[] = {
151 : NULL,
152 : &default_lockmethod,
153 : &user_lockmethod
154 : };
155 :
156 :
157 : /* Record that's written to 2PC state file when a lock is persisted */
158 : typedef struct TwoPhaseLockRecord
159 : {
160 : LOCKTAG locktag;
161 : LOCKMODE lockmode;
162 : } TwoPhaseLockRecord;
163 :
164 :
165 : /*
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.
170 : *
171 : * XXX Allocate a static array of the maximum size. We could use a pointer
172 : * and then allocate just the right size to save a couple kB, but then we
173 : * would have to initialize that, while for the static array that happens
174 : * automatically. Doesn't seem worth the extra complexity.
175 : */
176 : static int FastPathLocalUseCounts[FP_LOCK_GROUPS_PER_BACKEND_MAX];
177 :
178 : /*
179 : * Flag to indicate if the relation extension lock is held by this backend.
180 : * This flag is used to ensure that while holding the relation extension lock
181 : * we don't try to acquire a heavyweight lock on any other object. This
182 : * restriction implies that the relation extension lock won't ever participate
183 : * in the deadlock cycle because we can never wait for any other heavyweight
184 : * lock after acquiring this lock.
185 : *
186 : * Such a restriction is okay for relation extension locks as unlike other
187 : * heavyweight locks these are not held till the transaction end. These are
188 : * taken for a short duration to extend a particular relation and then
189 : * released.
190 : */
191 : static bool IsRelationExtensionLockHeld PG_USED_FOR_ASSERTS_ONLY = false;
192 :
193 : /*
194 : * Number of fast-path locks per backend - size of the arrays in PGPROC.
195 : * This is set only once during start, before initializing shared memory,
196 : * and remains constant after that.
197 : *
198 : * We set the limit based on max_locks_per_transaction GUC, because that's
199 : * the best information about expected number of locks per backend we have.
200 : * See InitializeFastPathLocks() for details.
201 : */
202 : int FastPathLockGroupsPerBackend = 0;
203 :
204 : /*
205 : * Macros to calculate the fast-path group and index for a relation.
206 : *
207 : * The formula is a simple hash function, designed to spread the OIDs a bit,
208 : * so that even contiguous values end up in different groups. In most cases
209 : * there will be gaps anyway, but the multiplication should help a bit.
210 : *
211 : * The selected constant (49157) is a prime not too close to 2^k, and it's
212 : * small enough to not cause overflows (in 64-bit).
213 : */
214 : #define FAST_PATH_REL_GROUP(rel) \
215 : (((uint64) (rel) * 49157) % FastPathLockGroupsPerBackend)
216 :
217 : /*
218 : * Given the group/slot indexes, calculate the slot index in the whole array
219 : * of fast-path lock slots.
220 : */
221 : #define FAST_PATH_SLOT(group, index) \
222 : (AssertMacro((uint32) (group) < FastPathLockGroupsPerBackend), \
223 : AssertMacro((uint32) (index) < FP_LOCK_SLOTS_PER_GROUP), \
224 : ((group) * FP_LOCK_SLOTS_PER_GROUP + (index)))
225 :
226 : /*
227 : * Given a slot index (into the whole per-backend array), calculated using
228 : * the FAST_PATH_SLOT macro, split it into group and index (in the group).
229 : */
230 : #define FAST_PATH_GROUP(index) \
231 : (AssertMacro((uint32) (index) < FastPathLockSlotsPerBackend()), \
232 : ((index) / FP_LOCK_SLOTS_PER_GROUP))
233 : #define FAST_PATH_INDEX(index) \
234 : (AssertMacro((uint32) (index) < FastPathLockSlotsPerBackend()), \
235 : ((index) % FP_LOCK_SLOTS_PER_GROUP))
236 :
237 : /* Macros for manipulating proc->fpLockBits */
238 : #define FAST_PATH_BITS_PER_SLOT 3
239 : #define FAST_PATH_LOCKNUMBER_OFFSET 1
240 : #define FAST_PATH_MASK ((1 << FAST_PATH_BITS_PER_SLOT) - 1)
241 : #define FAST_PATH_BITS(proc, n) (proc)->fpLockBits[FAST_PATH_GROUP(n)]
242 : #define FAST_PATH_GET_BITS(proc, n) \
243 : ((FAST_PATH_BITS(proc, n) >> (FAST_PATH_BITS_PER_SLOT * FAST_PATH_INDEX(n))) & FAST_PATH_MASK)
244 : #define FAST_PATH_BIT_POSITION(n, l) \
245 : (AssertMacro((l) >= FAST_PATH_LOCKNUMBER_OFFSET), \
246 : AssertMacro((l) < FAST_PATH_BITS_PER_SLOT+FAST_PATH_LOCKNUMBER_OFFSET), \
247 : AssertMacro((n) < FastPathLockSlotsPerBackend()), \
248 : ((l) - FAST_PATH_LOCKNUMBER_OFFSET + FAST_PATH_BITS_PER_SLOT * (FAST_PATH_INDEX(n))))
249 : #define FAST_PATH_SET_LOCKMODE(proc, n, l) \
250 : FAST_PATH_BITS(proc, n) |= UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l)
251 : #define FAST_PATH_CLEAR_LOCKMODE(proc, n, l) \
252 : FAST_PATH_BITS(proc, n) &= ~(UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l))
253 : #define FAST_PATH_CHECK_LOCKMODE(proc, n, l) \
254 : (FAST_PATH_BITS(proc, n) & (UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l)))
255 :
256 : /*
257 : * The fast-path lock mechanism is concerned only with relation locks on
258 : * unshared relations by backends bound to a database. The fast-path
259 : * mechanism exists mostly to accelerate acquisition and release of locks
260 : * that rarely conflict. Because ShareUpdateExclusiveLock is
261 : * self-conflicting, it can't use the fast-path mechanism; but it also does
262 : * not conflict with any of the locks that do, so we can ignore it completely.
263 : */
264 : #define EligibleForRelationFastPath(locktag, mode) \
265 : ((locktag)->locktag_lockmethodid == DEFAULT_LOCKMETHOD && \
266 : (locktag)->locktag_type == LOCKTAG_RELATION && \
267 : (locktag)->locktag_field1 == MyDatabaseId && \
268 : MyDatabaseId != InvalidOid && \
269 : (mode) < ShareUpdateExclusiveLock)
270 : #define ConflictsWithRelationFastPath(locktag, mode) \
271 : ((locktag)->locktag_lockmethodid == DEFAULT_LOCKMETHOD && \
272 : (locktag)->locktag_type == LOCKTAG_RELATION && \
273 : (locktag)->locktag_field1 != InvalidOid && \
274 : (mode) > ShareUpdateExclusiveLock)
275 :
276 : static bool FastPathGrantRelationLock(Oid relid, LOCKMODE lockmode);
277 : static bool FastPathUnGrantRelationLock(Oid relid, LOCKMODE lockmode);
278 : static bool FastPathTransferRelationLocks(LockMethod lockMethodTable,
279 : const LOCKTAG *locktag, uint32 hashcode);
280 : static PROCLOCK *FastPathGetRelationLockEntry(LOCALLOCK *locallock);
281 :
282 : /*
283 : * To make the fast-path lock mechanism work, we must have some way of
284 : * preventing the use of the fast-path when a conflicting lock might be present.
285 : * We partition* the locktag space into FAST_PATH_STRONG_LOCK_HASH_PARTITIONS,
286 : * and maintain an integer count of the number of "strong" lockers
287 : * in each partition. When any "strong" lockers are present (which is
288 : * hopefully not very often), the fast-path mechanism can't be used, and we
289 : * must fall back to the slower method of pushing matching locks directly
290 : * into the main lock tables.
291 : *
292 : * The deadlock detector does not know anything about the fast path mechanism,
293 : * so any locks that might be involved in a deadlock must be transferred from
294 : * the fast-path queues to the main lock table.
295 : */
296 :
297 : #define FAST_PATH_STRONG_LOCK_HASH_BITS 10
298 : #define FAST_PATH_STRONG_LOCK_HASH_PARTITIONS \
299 : (1 << FAST_PATH_STRONG_LOCK_HASH_BITS)
300 : #define FastPathStrongLockHashPartition(hashcode) \
301 : ((hashcode) % FAST_PATH_STRONG_LOCK_HASH_PARTITIONS)
302 :
303 : typedef struct
304 : {
305 : slock_t mutex;
306 : uint32 count[FAST_PATH_STRONG_LOCK_HASH_PARTITIONS];
307 : } FastPathStrongRelationLockData;
308 :
309 : static volatile FastPathStrongRelationLockData *FastPathStrongRelationLocks;
310 :
311 :
312 : /*
313 : * Pointers to hash tables containing lock state
314 : *
315 : * The LockMethodLockHash and LockMethodProcLockHash hash tables are in
316 : * shared memory; LockMethodLocalHash is local to each backend.
317 : */
318 : static HTAB *LockMethodLockHash;
319 : static HTAB *LockMethodProcLockHash;
320 : static HTAB *LockMethodLocalHash;
321 :
322 :
323 : /* private state for error cleanup */
324 : static LOCALLOCK *StrongLockInProgress;
325 : static LOCALLOCK *awaitedLock;
326 : static ResourceOwner awaitedOwner;
327 :
328 :
329 : #ifdef LOCK_DEBUG
330 :
331 : /*------
332 : * The following configuration options are available for lock debugging:
333 : *
334 : * TRACE_LOCKS -- give a bunch of output what's going on in this file
335 : * TRACE_USERLOCKS -- same but for user locks
336 : * TRACE_LOCK_OIDMIN-- do not trace locks for tables below this oid
337 : * (use to avoid output on system tables)
338 : * TRACE_LOCK_TABLE -- trace locks on this table (oid) unconditionally
339 : * DEBUG_DEADLOCKS -- currently dumps locks at untimely occasions ;)
340 : *
341 : * Furthermore, but in storage/lmgr/lwlock.c:
342 : * TRACE_LWLOCKS -- trace lightweight locks (pretty useless)
343 : *
344 : * Define LOCK_DEBUG at compile time to get all these enabled.
345 : * --------
346 : */
347 :
348 : int Trace_lock_oidmin = FirstNormalObjectId;
349 : bool Trace_locks = false;
350 : bool Trace_userlocks = false;
351 : int Trace_lock_table = 0;
352 : bool Debug_deadlocks = false;
353 :
354 :
355 : inline static bool
356 : LOCK_DEBUG_ENABLED(const LOCKTAG *tag)
357 : {
358 : return
359 : (*(LockMethods[tag->locktag_lockmethodid]->trace_flag) &&
360 : ((Oid) tag->locktag_field2 >= (Oid) Trace_lock_oidmin))
361 : || (Trace_lock_table &&
362 : (tag->locktag_field2 == Trace_lock_table));
363 : }
364 :
365 :
366 : inline static void
367 : LOCK_PRINT(const char *where, const LOCK *lock, LOCKMODE type)
368 : {
369 : if (LOCK_DEBUG_ENABLED(&lock->tag))
370 : elog(LOG,
371 : "%s: lock(%p) id(%u,%u,%u,%u,%u,%u) grantMask(%x) "
372 : "req(%d,%d,%d,%d,%d,%d,%d)=%d "
373 : "grant(%d,%d,%d,%d,%d,%d,%d)=%d wait(%d) type(%s)",
374 : where, lock,
375 : lock->tag.locktag_field1, lock->tag.locktag_field2,
376 : lock->tag.locktag_field3, lock->tag.locktag_field4,
377 : lock->tag.locktag_type, lock->tag.locktag_lockmethodid,
378 : lock->grantMask,
379 : lock->requested[1], lock->requested[2], lock->requested[3],
380 : lock->requested[4], lock->requested[5], lock->requested[6],
381 : lock->requested[7], lock->nRequested,
382 : lock->granted[1], lock->granted[2], lock->granted[3],
383 : lock->granted[4], lock->granted[5], lock->granted[6],
384 : lock->granted[7], lock->nGranted,
385 : dclist_count(&lock->waitProcs),
386 : LockMethods[LOCK_LOCKMETHOD(*lock)]->lockModeNames[type]);
387 : }
388 :
389 :
390 : inline static void
391 : PROCLOCK_PRINT(const char *where, const PROCLOCK *proclockP)
392 : {
393 : if (LOCK_DEBUG_ENABLED(&proclockP->tag.myLock->tag))
394 : elog(LOG,
395 : "%s: proclock(%p) lock(%p) method(%u) proc(%p) hold(%x)",
396 : where, proclockP, proclockP->tag.myLock,
397 : PROCLOCK_LOCKMETHOD(*(proclockP)),
398 : proclockP->tag.myProc, (int) proclockP->holdMask);
399 : }
400 : #else /* not LOCK_DEBUG */
401 :
402 : #define LOCK_PRINT(where, lock, type) ((void) 0)
403 : #define PROCLOCK_PRINT(where, proclockP) ((void) 0)
404 : #endif /* not LOCK_DEBUG */
405 :
406 :
407 : static uint32 proclock_hash(const void *key, Size keysize);
408 : static void RemoveLocalLock(LOCALLOCK *locallock);
409 : static PROCLOCK *SetupLockInTable(LockMethod lockMethodTable, PGPROC *proc,
410 : const LOCKTAG *locktag, uint32 hashcode, LOCKMODE lockmode);
411 : static void GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner);
412 : static void BeginStrongLockAcquire(LOCALLOCK *locallock, uint32 fasthashcode);
413 : static void FinishStrongLockAcquire(void);
414 : static ProcWaitStatus WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner);
415 : static void ReleaseLockIfHeld(LOCALLOCK *locallock, bool sessionLock);
416 : static void LockReassignOwner(LOCALLOCK *locallock, ResourceOwner parent);
417 : static bool UnGrantLock(LOCK *lock, LOCKMODE lockmode,
418 : PROCLOCK *proclock, LockMethod lockMethodTable);
419 : static void CleanUpLock(LOCK *lock, PROCLOCK *proclock,
420 : LockMethod lockMethodTable, uint32 hashcode,
421 : bool wakeupNeeded);
422 : static void LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
423 : LOCKTAG *locktag, LOCKMODE lockmode,
424 : bool decrement_strong_lock_count);
425 : static void GetSingleProcBlockerStatusData(PGPROC *blocked_proc,
426 : BlockedProcsData *data);
427 :
428 :
429 : /*
430 : * Initialize the lock manager's shmem data structures.
431 : *
432 : * This is called from CreateSharedMemoryAndSemaphores(), which see for more
433 : * comments. In the normal postmaster case, the shared hash tables are
434 : * created here, and backends inherit pointers to them via fork(). In the
435 : * EXEC_BACKEND case, each backend re-executes this code to obtain pointers to
436 : * the already existing shared hash tables. In either case, each backend must
437 : * also call InitLockManagerAccess() to create the locallock hash table.
438 : */
439 : void
440 2032 : LockManagerShmemInit(void)
441 : {
442 : HASHCTL info;
443 : long init_table_size,
444 : max_table_size;
445 : bool found;
446 :
447 : /*
448 : * Compute init/max size to request for lock hashtables. Note these
449 : * calculations must agree with LockManagerShmemSize!
450 : */
451 2032 : max_table_size = NLOCKENTS();
452 2032 : init_table_size = max_table_size / 2;
453 :
454 : /*
455 : * Allocate hash table for LOCK structs. This stores per-locked-object
456 : * information.
457 : */
458 2032 : info.keysize = sizeof(LOCKTAG);
459 2032 : info.entrysize = sizeof(LOCK);
460 2032 : info.num_partitions = NUM_LOCK_PARTITIONS;
461 :
462 2032 : LockMethodLockHash = ShmemInitHash("LOCK hash",
463 : init_table_size,
464 : max_table_size,
465 : &info,
466 : HASH_ELEM | HASH_BLOBS | HASH_PARTITION);
467 :
468 : /* Assume an average of 2 holders per lock */
469 2032 : max_table_size *= 2;
470 2032 : init_table_size *= 2;
471 :
472 : /*
473 : * Allocate hash table for PROCLOCK structs. This stores
474 : * per-lock-per-holder information.
475 : */
476 2032 : info.keysize = sizeof(PROCLOCKTAG);
477 2032 : info.entrysize = sizeof(PROCLOCK);
478 2032 : info.hash = proclock_hash;
479 2032 : info.num_partitions = NUM_LOCK_PARTITIONS;
480 :
481 2032 : LockMethodProcLockHash = ShmemInitHash("PROCLOCK hash",
482 : init_table_size,
483 : max_table_size,
484 : &info,
485 : HASH_ELEM | HASH_FUNCTION | HASH_PARTITION);
486 :
487 : /*
488 : * Allocate fast-path structures.
489 : */
490 2032 : FastPathStrongRelationLocks =
491 2032 : ShmemInitStruct("Fast Path Strong Relation Lock Data",
492 : sizeof(FastPathStrongRelationLockData), &found);
493 2032 : if (!found)
494 2032 : SpinLockInit(&FastPathStrongRelationLocks->mutex);
495 2032 : }
496 :
497 : /*
498 : * Initialize the lock manager's backend-private data structures.
499 : */
500 : void
501 42318 : InitLockManagerAccess(void)
502 : {
503 : /*
504 : * Allocate non-shared hash table for LOCALLOCK structs. This stores lock
505 : * counts and resource owner information.
506 : */
507 : HASHCTL info;
508 :
509 42318 : info.keysize = sizeof(LOCALLOCKTAG);
510 42318 : info.entrysize = sizeof(LOCALLOCK);
511 :
512 42318 : LockMethodLocalHash = hash_create("LOCALLOCK hash",
513 : 16,
514 : &info,
515 : HASH_ELEM | HASH_BLOBS);
516 42318 : }
517 :
518 :
519 : /*
520 : * Fetch the lock method table associated with a given lock
521 : */
522 : LockMethod
523 186 : GetLocksMethodTable(const LOCK *lock)
524 : {
525 186 : LOCKMETHODID lockmethodid = LOCK_LOCKMETHOD(*lock);
526 :
527 : Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
528 186 : return LockMethods[lockmethodid];
529 : }
530 :
531 : /*
532 : * Fetch the lock method table associated with a given locktag
533 : */
534 : LockMethod
535 2254 : GetLockTagsMethodTable(const LOCKTAG *locktag)
536 : {
537 2254 : LOCKMETHODID lockmethodid = (LOCKMETHODID) locktag->locktag_lockmethodid;
538 :
539 : Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
540 2254 : return LockMethods[lockmethodid];
541 : }
542 :
543 :
544 : /*
545 : * Compute the hash code associated with a LOCKTAG.
546 : *
547 : * To avoid unnecessary recomputations of the hash code, we try to do this
548 : * just once per function, and then pass it around as needed. Aside from
549 : * passing the hashcode to hash_search_with_hash_value(), we can extract
550 : * the lock partition number from the hashcode.
551 : */
552 : uint32
553 34884914 : LockTagHashCode(const LOCKTAG *locktag)
554 : {
555 34884914 : return get_hash_value(LockMethodLockHash, locktag);
556 : }
557 :
558 : /*
559 : * Compute the hash code associated with a PROCLOCKTAG.
560 : *
561 : * Because we want to use just one set of partition locks for both the
562 : * LOCK and PROCLOCK hash tables, we have to make sure that PROCLOCKs
563 : * fall into the same partition number as their associated LOCKs.
564 : * dynahash.c expects the partition number to be the low-order bits of
565 : * the hash code, and therefore a PROCLOCKTAG's hash code must have the
566 : * same low-order bits as the associated LOCKTAG's hash code. We achieve
567 : * this with this specialized hash function.
568 : */
569 : static uint32
570 1356 : proclock_hash(const void *key, Size keysize)
571 : {
572 1356 : const PROCLOCKTAG *proclocktag = (const PROCLOCKTAG *) key;
573 : uint32 lockhash;
574 : Datum procptr;
575 :
576 : Assert(keysize == sizeof(PROCLOCKTAG));
577 :
578 : /* Look into the associated LOCK object, and compute its hash code */
579 1356 : lockhash = LockTagHashCode(&proclocktag->myLock->tag);
580 :
581 : /*
582 : * To make the hash code also depend on the PGPROC, we xor the proc
583 : * struct's address into the hash code, left-shifted so that the
584 : * partition-number bits don't change. Since this is only a hash, we
585 : * don't care if we lose high-order bits of the address; use an
586 : * intermediate variable to suppress cast-pointer-to-int warnings.
587 : */
588 1356 : procptr = PointerGetDatum(proclocktag->myProc);
589 1356 : lockhash ^= ((uint32) procptr) << LOG2_NUM_LOCK_PARTITIONS;
590 :
591 1356 : return lockhash;
592 : }
593 :
594 : /*
595 : * Compute the hash code associated with a PROCLOCKTAG, given the hashcode
596 : * for its underlying LOCK.
597 : *
598 : * We use this just to avoid redundant calls of LockTagHashCode().
599 : */
600 : static inline uint32
601 8316796 : ProcLockHashCode(const PROCLOCKTAG *proclocktag, uint32 hashcode)
602 : {
603 8316796 : uint32 lockhash = hashcode;
604 : Datum procptr;
605 :
606 : /*
607 : * This must match proclock_hash()!
608 : */
609 8316796 : procptr = PointerGetDatum(proclocktag->myProc);
610 8316796 : lockhash ^= ((uint32) procptr) << LOG2_NUM_LOCK_PARTITIONS;
611 :
612 8316796 : return lockhash;
613 : }
614 :
615 : /*
616 : * Given two lock modes, return whether they would conflict.
617 : */
618 : bool
619 488 : DoLockModesConflict(LOCKMODE mode1, LOCKMODE mode2)
620 : {
621 488 : LockMethod lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
622 :
623 488 : if (lockMethodTable->conflictTab[mode1] & LOCKBIT_ON(mode2))
624 284 : return true;
625 :
626 204 : return false;
627 : }
628 :
629 : /*
630 : * LockHeldByMe -- test whether lock 'locktag' is held by the current
631 : * transaction
632 : *
633 : * Returns true if current transaction holds a lock on 'tag' of mode
634 : * 'lockmode'. If 'orstronger' is true, a stronger lockmode is also OK.
635 : * ("Stronger" is defined as "numerically higher", which is a bit
636 : * semantically dubious but is OK for the purposes we use this for.)
637 : */
638 : bool
639 0 : LockHeldByMe(const LOCKTAG *locktag,
640 : LOCKMODE lockmode, bool orstronger)
641 : {
642 : LOCALLOCKTAG localtag;
643 : LOCALLOCK *locallock;
644 :
645 : /*
646 : * See if there is a LOCALLOCK entry for this lock and lockmode
647 : */
648 0 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
649 0 : localtag.lock = *locktag;
650 0 : localtag.mode = lockmode;
651 :
652 0 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
653 : &localtag,
654 : HASH_FIND, NULL);
655 :
656 0 : if (locallock && locallock->nLocks > 0)
657 0 : return true;
658 :
659 0 : if (orstronger)
660 : {
661 : LOCKMODE slockmode;
662 :
663 0 : for (slockmode = lockmode + 1;
664 : slockmode <= MaxLockMode;
665 0 : slockmode++)
666 : {
667 0 : if (LockHeldByMe(locktag, slockmode, false))
668 0 : return true;
669 : }
670 : }
671 :
672 0 : return false;
673 : }
674 :
675 : #ifdef USE_ASSERT_CHECKING
676 : /*
677 : * GetLockMethodLocalHash -- return the hash of local locks, for modules that
678 : * evaluate assertions based on all locks held.
679 : */
680 : HTAB *
681 : GetLockMethodLocalHash(void)
682 : {
683 : return LockMethodLocalHash;
684 : }
685 : #endif
686 :
687 : /*
688 : * LockHasWaiters -- look up 'locktag' and check if releasing this
689 : * lock would wake up other processes waiting for it.
690 : */
691 : bool
692 0 : LockHasWaiters(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
693 : {
694 0 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
695 : LockMethod lockMethodTable;
696 : LOCALLOCKTAG localtag;
697 : LOCALLOCK *locallock;
698 : LOCK *lock;
699 : PROCLOCK *proclock;
700 : LWLock *partitionLock;
701 0 : bool hasWaiters = false;
702 :
703 0 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
704 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
705 0 : lockMethodTable = LockMethods[lockmethodid];
706 0 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
707 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
708 :
709 : #ifdef LOCK_DEBUG
710 : if (LOCK_DEBUG_ENABLED(locktag))
711 : elog(LOG, "LockHasWaiters: lock [%u,%u] %s",
712 : locktag->locktag_field1, locktag->locktag_field2,
713 : lockMethodTable->lockModeNames[lockmode]);
714 : #endif
715 :
716 : /*
717 : * Find the LOCALLOCK entry for this lock and lockmode
718 : */
719 0 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
720 0 : localtag.lock = *locktag;
721 0 : localtag.mode = lockmode;
722 :
723 0 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
724 : &localtag,
725 : HASH_FIND, NULL);
726 :
727 : /*
728 : * let the caller print its own error message, too. Do not ereport(ERROR).
729 : */
730 0 : if (!locallock || locallock->nLocks <= 0)
731 : {
732 0 : elog(WARNING, "you don't own a lock of type %s",
733 : lockMethodTable->lockModeNames[lockmode]);
734 0 : return false;
735 : }
736 :
737 : /*
738 : * Check the shared lock table.
739 : */
740 0 : partitionLock = LockHashPartitionLock(locallock->hashcode);
741 :
742 0 : LWLockAcquire(partitionLock, LW_SHARED);
743 :
744 : /*
745 : * We don't need to re-find the lock or proclock, since we kept their
746 : * addresses in the locallock table, and they couldn't have been removed
747 : * while we were holding a lock on them.
748 : */
749 0 : lock = locallock->lock;
750 : LOCK_PRINT("LockHasWaiters: found", lock, lockmode);
751 0 : proclock = locallock->proclock;
752 : PROCLOCK_PRINT("LockHasWaiters: found", proclock);
753 :
754 : /*
755 : * Double-check that we are actually holding a lock of the type we want to
756 : * release.
757 : */
758 0 : if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
759 : {
760 : PROCLOCK_PRINT("LockHasWaiters: WRONGTYPE", proclock);
761 0 : LWLockRelease(partitionLock);
762 0 : elog(WARNING, "you don't own a lock of type %s",
763 : lockMethodTable->lockModeNames[lockmode]);
764 0 : RemoveLocalLock(locallock);
765 0 : return false;
766 : }
767 :
768 : /*
769 : * Do the checking.
770 : */
771 0 : if ((lockMethodTable->conflictTab[lockmode] & lock->waitMask) != 0)
772 0 : hasWaiters = true;
773 :
774 0 : LWLockRelease(partitionLock);
775 :
776 0 : return hasWaiters;
777 : }
778 :
779 : /*
780 : * LockAcquire -- Check for lock conflicts, sleep if conflict found,
781 : * set lock if/when no conflicts.
782 : *
783 : * Inputs:
784 : * locktag: unique identifier for the lockable object
785 : * lockmode: lock mode to acquire
786 : * sessionLock: if true, acquire lock for session not current transaction
787 : * dontWait: if true, don't wait to acquire lock
788 : *
789 : * Returns one of:
790 : * LOCKACQUIRE_NOT_AVAIL lock not available, and dontWait=true
791 : * LOCKACQUIRE_OK lock successfully acquired
792 : * LOCKACQUIRE_ALREADY_HELD incremented count for lock already held
793 : * LOCKACQUIRE_ALREADY_CLEAR incremented count for lock already clear
794 : *
795 : * In the normal case where dontWait=false and the caller doesn't need to
796 : * distinguish a freshly acquired lock from one already taken earlier in
797 : * this same transaction, there is no need to examine the return value.
798 : *
799 : * Side Effects: The lock is acquired and recorded in lock tables.
800 : *
801 : * NOTE: if we wait for the lock, there is no way to abort the wait
802 : * short of aborting the transaction.
803 : */
804 : LockAcquireResult
805 1501062 : LockAcquire(const LOCKTAG *locktag,
806 : LOCKMODE lockmode,
807 : bool sessionLock,
808 : bool dontWait)
809 : {
810 1501062 : return LockAcquireExtended(locktag, lockmode, sessionLock, dontWait,
811 : true, NULL, false);
812 : }
813 :
814 : /*
815 : * LockAcquireExtended - allows us to specify additional options
816 : *
817 : * reportMemoryError specifies whether a lock request that fills the lock
818 : * table should generate an ERROR or not. Passing "false" allows the caller
819 : * to attempt to recover from lock-table-full situations, perhaps by forcibly
820 : * canceling other lock holders and then retrying. Note, however, that the
821 : * return code for that is LOCKACQUIRE_NOT_AVAIL, so that it's unsafe to use
822 : * in combination with dontWait = true, as the cause of failure couldn't be
823 : * distinguished.
824 : *
825 : * If locallockp isn't NULL, *locallockp receives a pointer to the LOCALLOCK
826 : * table entry if a lock is successfully acquired, or NULL if not.
827 : *
828 : * logLockFailure indicates whether to log details when a lock acquisition
829 : * fails with dontWait = true.
830 : */
831 : LockAcquireResult
832 37959594 : LockAcquireExtended(const LOCKTAG *locktag,
833 : LOCKMODE lockmode,
834 : bool sessionLock,
835 : bool dontWait,
836 : bool reportMemoryError,
837 : LOCALLOCK **locallockp,
838 : bool logLockFailure)
839 : {
840 37959594 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
841 : LockMethod lockMethodTable;
842 : LOCALLOCKTAG localtag;
843 : LOCALLOCK *locallock;
844 : LOCK *lock;
845 : PROCLOCK *proclock;
846 : bool found;
847 : ResourceOwner owner;
848 : uint32 hashcode;
849 : LWLock *partitionLock;
850 : bool found_conflict;
851 : ProcWaitStatus waitResult;
852 37959594 : bool log_lock = false;
853 :
854 37959594 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
855 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
856 37959594 : lockMethodTable = LockMethods[lockmethodid];
857 37959594 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
858 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
859 :
860 37959594 : if (RecoveryInProgress() && !InRecovery &&
861 562706 : (locktag->locktag_type == LOCKTAG_OBJECT ||
862 562706 : locktag->locktag_type == LOCKTAG_RELATION) &&
863 : lockmode > RowExclusiveLock)
864 0 : ereport(ERROR,
865 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
866 : errmsg("cannot acquire lock mode %s on database objects while recovery is in progress",
867 : lockMethodTable->lockModeNames[lockmode]),
868 : errhint("Only RowExclusiveLock or less can be acquired on database objects during recovery.")));
869 :
870 : #ifdef LOCK_DEBUG
871 : if (LOCK_DEBUG_ENABLED(locktag))
872 : elog(LOG, "LockAcquire: lock [%u,%u] %s",
873 : locktag->locktag_field1, locktag->locktag_field2,
874 : lockMethodTable->lockModeNames[lockmode]);
875 : #endif
876 :
877 : /* Identify owner for lock */
878 37959594 : if (sessionLock)
879 173718 : owner = NULL;
880 : else
881 37785876 : owner = CurrentResourceOwner;
882 :
883 : /*
884 : * Find or create a LOCALLOCK entry for this lock and lockmode
885 : */
886 37959594 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
887 37959594 : localtag.lock = *locktag;
888 37959594 : localtag.mode = lockmode;
889 :
890 37959594 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
891 : &localtag,
892 : HASH_ENTER, &found);
893 :
894 : /*
895 : * if it's a new locallock object, initialize it
896 : */
897 37959594 : if (!found)
898 : {
899 33799406 : locallock->lock = NULL;
900 33799406 : locallock->proclock = NULL;
901 33799406 : locallock->hashcode = LockTagHashCode(&(localtag.lock));
902 33799406 : locallock->nLocks = 0;
903 33799406 : locallock->holdsStrongLockCount = false;
904 33799406 : locallock->lockCleared = false;
905 33799406 : locallock->numLockOwners = 0;
906 33799406 : locallock->maxLockOwners = 8;
907 33799406 : locallock->lockOwners = NULL; /* in case next line fails */
908 33799406 : locallock->lockOwners = (LOCALLOCKOWNER *)
909 33799406 : MemoryContextAlloc(TopMemoryContext,
910 33799406 : locallock->maxLockOwners * sizeof(LOCALLOCKOWNER));
911 : }
912 : else
913 : {
914 : /* Make sure there will be room to remember the lock */
915 4160188 : if (locallock->numLockOwners >= locallock->maxLockOwners)
916 : {
917 38 : int newsize = locallock->maxLockOwners * 2;
918 :
919 38 : locallock->lockOwners = (LOCALLOCKOWNER *)
920 38 : repalloc(locallock->lockOwners,
921 : newsize * sizeof(LOCALLOCKOWNER));
922 38 : locallock->maxLockOwners = newsize;
923 : }
924 : }
925 37959594 : hashcode = locallock->hashcode;
926 :
927 37959594 : if (locallockp)
928 36458360 : *locallockp = locallock;
929 :
930 : /*
931 : * If we already hold the lock, we can just increase the count locally.
932 : *
933 : * If lockCleared is already set, caller need not worry about absorbing
934 : * sinval messages related to the lock's object.
935 : */
936 37959594 : if (locallock->nLocks > 0)
937 : {
938 4160188 : GrantLockLocal(locallock, owner);
939 4160188 : if (locallock->lockCleared)
940 4011248 : return LOCKACQUIRE_ALREADY_CLEAR;
941 : else
942 148940 : return LOCKACQUIRE_ALREADY_HELD;
943 : }
944 :
945 : /*
946 : * We don't acquire any other heavyweight lock while holding the relation
947 : * extension lock. We do allow to acquire the same relation extension
948 : * lock more than once but that case won't reach here.
949 : */
950 : Assert(!IsRelationExtensionLockHeld);
951 :
952 : /*
953 : * Prepare to emit a WAL record if acquisition of this lock needs to be
954 : * replayed in a standby server.
955 : *
956 : * Here we prepare to log; after lock is acquired we'll issue log record.
957 : * This arrangement simplifies error recovery in case the preparation step
958 : * fails.
959 : *
960 : * Only AccessExclusiveLocks can conflict with lock types that read-only
961 : * transactions can acquire in a standby server. Make sure this definition
962 : * matches the one in GetRunningTransactionLocks().
963 : */
964 33799406 : if (lockmode >= AccessExclusiveLock &&
965 451434 : locktag->locktag_type == LOCKTAG_RELATION &&
966 303378 : !RecoveryInProgress() &&
967 255688 : XLogStandbyInfoActive())
968 : {
969 195226 : LogAccessExclusiveLockPrepare();
970 195226 : log_lock = true;
971 : }
972 :
973 : /*
974 : * Attempt to take lock via fast path, if eligible. But if we remember
975 : * having filled up the fast path array, we don't attempt to make any
976 : * further use of it until we release some locks. It's possible that some
977 : * other backend has transferred some of those locks to the shared hash
978 : * table, leaving space free, but it's not worth acquiring the LWLock just
979 : * to check. It's also possible that we're acquiring a second or third
980 : * lock type on a relation we have already locked using the fast-path, but
981 : * for now we don't worry about that case either.
982 : */
983 33799406 : if (EligibleForRelationFastPath(locktag, lockmode) &&
984 30288928 : FastPathLocalUseCounts[FAST_PATH_REL_GROUP(locktag->locktag_field2)] < FP_LOCK_SLOTS_PER_GROUP)
985 : {
986 29937602 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
987 : bool acquired;
988 :
989 : /*
990 : * LWLockAcquire acts as a memory sequencing point, so it's safe to
991 : * assume that any strong locker whose increment to
992 : * FastPathStrongRelationLocks->counts becomes visible after we test
993 : * it has yet to begin to transfer fast-path locks.
994 : */
995 29937602 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
996 29937602 : if (FastPathStrongRelationLocks->count[fasthashcode] != 0)
997 487954 : acquired = false;
998 : else
999 29449648 : acquired = FastPathGrantRelationLock(locktag->locktag_field2,
1000 : lockmode);
1001 29937602 : LWLockRelease(&MyProc->fpInfoLock);
1002 29937602 : if (acquired)
1003 : {
1004 : /*
1005 : * The locallock might contain stale pointers to some old shared
1006 : * objects; we MUST reset these to null before considering the
1007 : * lock to be acquired via fast-path.
1008 : */
1009 29449648 : locallock->lock = NULL;
1010 29449648 : locallock->proclock = NULL;
1011 29449648 : GrantLockLocal(locallock, owner);
1012 29449648 : return LOCKACQUIRE_OK;
1013 : }
1014 : }
1015 :
1016 : /*
1017 : * If this lock could potentially have been taken via the fast-path by
1018 : * some other backend, we must (temporarily) disable further use of the
1019 : * fast-path for this lock tag, and migrate any locks already taken via
1020 : * this method to the main lock table.
1021 : */
1022 4349758 : if (ConflictsWithRelationFastPath(locktag, lockmode))
1023 : {
1024 359710 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
1025 :
1026 359710 : BeginStrongLockAcquire(locallock, fasthashcode);
1027 359710 : if (!FastPathTransferRelationLocks(lockMethodTable, locktag,
1028 : hashcode))
1029 : {
1030 0 : AbortStrongLockAcquire();
1031 0 : if (locallock->nLocks == 0)
1032 0 : RemoveLocalLock(locallock);
1033 0 : if (locallockp)
1034 0 : *locallockp = NULL;
1035 0 : if (reportMemoryError)
1036 0 : ereport(ERROR,
1037 : (errcode(ERRCODE_OUT_OF_MEMORY),
1038 : errmsg("out of shared memory"),
1039 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
1040 : else
1041 0 : return LOCKACQUIRE_NOT_AVAIL;
1042 : }
1043 : }
1044 :
1045 : /*
1046 : * We didn't find the lock in our LOCALLOCK table, and we didn't manage to
1047 : * take it via the fast-path, either, so we've got to mess with the shared
1048 : * lock table.
1049 : */
1050 4349758 : partitionLock = LockHashPartitionLock(hashcode);
1051 :
1052 4349758 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
1053 :
1054 : /*
1055 : * Find or create lock and proclock entries with this tag
1056 : *
1057 : * Note: if the locallock object already existed, it might have a pointer
1058 : * to the lock already ... but we should not assume that that pointer is
1059 : * valid, since a lock object with zero hold and request counts can go
1060 : * away anytime. So we have to use SetupLockInTable() to recompute the
1061 : * lock and proclock pointers, even if they're already set.
1062 : */
1063 4349758 : proclock = SetupLockInTable(lockMethodTable, MyProc, locktag,
1064 : hashcode, lockmode);
1065 4349758 : if (!proclock)
1066 : {
1067 0 : AbortStrongLockAcquire();
1068 0 : LWLockRelease(partitionLock);
1069 0 : if (locallock->nLocks == 0)
1070 0 : RemoveLocalLock(locallock);
1071 0 : if (locallockp)
1072 0 : *locallockp = NULL;
1073 0 : if (reportMemoryError)
1074 0 : ereport(ERROR,
1075 : (errcode(ERRCODE_OUT_OF_MEMORY),
1076 : errmsg("out of shared memory"),
1077 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
1078 : else
1079 0 : return LOCKACQUIRE_NOT_AVAIL;
1080 : }
1081 4349758 : locallock->proclock = proclock;
1082 4349758 : lock = proclock->tag.myLock;
1083 4349758 : locallock->lock = lock;
1084 :
1085 : /*
1086 : * If lock requested conflicts with locks requested by waiters, must join
1087 : * wait queue. Otherwise, check for conflict with already-held locks.
1088 : * (That's last because most complex check.)
1089 : */
1090 4349758 : if (lockMethodTable->conflictTab[lockmode] & lock->waitMask)
1091 370 : found_conflict = true;
1092 : else
1093 4349388 : found_conflict = LockCheckConflicts(lockMethodTable, lockmode,
1094 : lock, proclock);
1095 :
1096 4349758 : if (!found_conflict)
1097 : {
1098 : /* No conflict with held or previously requested locks */
1099 4345736 : GrantLock(lock, proclock, lockmode);
1100 4345736 : waitResult = PROC_WAIT_STATUS_OK;
1101 : }
1102 : else
1103 : {
1104 : /*
1105 : * Join the lock's wait queue. We call this even in the dontWait
1106 : * case, because JoinWaitQueue() may discover that we can acquire the
1107 : * lock immediately after all.
1108 : */
1109 4022 : waitResult = JoinWaitQueue(locallock, lockMethodTable, dontWait);
1110 : }
1111 :
1112 4349758 : if (waitResult == PROC_WAIT_STATUS_ERROR)
1113 : {
1114 : /*
1115 : * We're not getting the lock because a deadlock was detected already
1116 : * while trying to join the wait queue, or because we would have to
1117 : * wait but the caller requested no blocking.
1118 : *
1119 : * Undo the changes to shared entries before releasing the partition
1120 : * lock.
1121 : */
1122 1462 : AbortStrongLockAcquire();
1123 :
1124 1462 : if (proclock->holdMask == 0)
1125 : {
1126 : uint32 proclock_hashcode;
1127 :
1128 1056 : proclock_hashcode = ProcLockHashCode(&proclock->tag,
1129 : hashcode);
1130 1056 : dlist_delete(&proclock->lockLink);
1131 1056 : dlist_delete(&proclock->procLink);
1132 1056 : if (!hash_search_with_hash_value(LockMethodProcLockHash,
1133 1056 : &(proclock->tag),
1134 : proclock_hashcode,
1135 : HASH_REMOVE,
1136 : NULL))
1137 0 : elog(PANIC, "proclock table corrupted");
1138 : }
1139 : else
1140 : PROCLOCK_PRINT("LockAcquire: did not join wait queue", proclock);
1141 1462 : lock->nRequested--;
1142 1462 : lock->requested[lockmode]--;
1143 : LOCK_PRINT("LockAcquire: did not join wait queue",
1144 : lock, lockmode);
1145 : Assert((lock->nRequested > 0) &&
1146 : (lock->requested[lockmode] >= 0));
1147 : Assert(lock->nGranted <= lock->nRequested);
1148 1462 : LWLockRelease(partitionLock);
1149 1462 : if (locallock->nLocks == 0)
1150 1462 : RemoveLocalLock(locallock);
1151 :
1152 1462 : if (dontWait)
1153 : {
1154 : /*
1155 : * Log lock holders and waiters as a detail log message if
1156 : * logLockFailure = true and lock acquisition fails with dontWait
1157 : * = true
1158 : */
1159 1460 : if (logLockFailure)
1160 : {
1161 : StringInfoData buf,
1162 : lock_waiters_sbuf,
1163 : lock_holders_sbuf;
1164 : const char *modename;
1165 0 : int lockHoldersNum = 0;
1166 :
1167 0 : initStringInfo(&buf);
1168 0 : initStringInfo(&lock_waiters_sbuf);
1169 0 : initStringInfo(&lock_holders_sbuf);
1170 :
1171 0 : DescribeLockTag(&buf, &locallock->tag.lock);
1172 0 : modename = GetLockmodeName(locallock->tag.lock.locktag_lockmethodid,
1173 : lockmode);
1174 :
1175 : /* Gather a list of all lock holders and waiters */
1176 0 : LWLockAcquire(partitionLock, LW_SHARED);
1177 0 : GetLockHoldersAndWaiters(locallock, &lock_holders_sbuf,
1178 : &lock_waiters_sbuf, &lockHoldersNum);
1179 0 : LWLockRelease(partitionLock);
1180 :
1181 0 : ereport(LOG,
1182 : (errmsg("process %d could not obtain %s on %s",
1183 : MyProcPid, modename, buf.data),
1184 : errdetail_log_plural(
1185 : "Process holding the lock: %s, Wait queue: %s.",
1186 : "Processes holding the lock: %s, Wait queue: %s.",
1187 : lockHoldersNum,
1188 : lock_holders_sbuf.data,
1189 : lock_waiters_sbuf.data)));
1190 :
1191 0 : pfree(buf.data);
1192 0 : pfree(lock_holders_sbuf.data);
1193 0 : pfree(lock_waiters_sbuf.data);
1194 : }
1195 1460 : if (locallockp)
1196 440 : *locallockp = NULL;
1197 1460 : return LOCKACQUIRE_NOT_AVAIL;
1198 : }
1199 : else
1200 : {
1201 2 : DeadLockReport();
1202 : /* DeadLockReport() will not return */
1203 : }
1204 : }
1205 :
1206 : /*
1207 : * We are now in the lock queue, or the lock was already granted. If
1208 : * queued, go to sleep.
1209 : */
1210 4348296 : if (waitResult == PROC_WAIT_STATUS_WAITING)
1211 : {
1212 : Assert(!dontWait);
1213 : PROCLOCK_PRINT("LockAcquire: sleeping on lock", proclock);
1214 : LOCK_PRINT("LockAcquire: sleeping on lock", lock, lockmode);
1215 2556 : LWLockRelease(partitionLock);
1216 :
1217 2556 : waitResult = WaitOnLock(locallock, owner);
1218 :
1219 : /*
1220 : * NOTE: do not do any material change of state between here and
1221 : * return. All required changes in locktable state must have been
1222 : * done when the lock was granted to us --- see notes in WaitOnLock.
1223 : */
1224 :
1225 2472 : if (waitResult == PROC_WAIT_STATUS_ERROR)
1226 : {
1227 : /*
1228 : * We failed as a result of a deadlock, see CheckDeadLock(). Quit
1229 : * now.
1230 : */
1231 : Assert(!dontWait);
1232 10 : DeadLockReport();
1233 : /* DeadLockReport() will not return */
1234 : }
1235 : }
1236 : else
1237 4345740 : LWLockRelease(partitionLock);
1238 : Assert(waitResult == PROC_WAIT_STATUS_OK);
1239 :
1240 : /* The lock was granted to us. Update the local lock entry accordingly */
1241 : Assert((proclock->holdMask & LOCKBIT_ON(lockmode)) != 0);
1242 4348202 : GrantLockLocal(locallock, owner);
1243 :
1244 : /*
1245 : * Lock state is fully up-to-date now; if we error out after this, no
1246 : * special error cleanup is required.
1247 : */
1248 4348202 : FinishStrongLockAcquire();
1249 :
1250 : /*
1251 : * Emit a WAL record if acquisition of this lock needs to be replayed in a
1252 : * standby server.
1253 : */
1254 4348202 : if (log_lock)
1255 : {
1256 : /*
1257 : * Decode the locktag back to the original values, to avoid sending
1258 : * lots of empty bytes with every message. See lock.h to check how a
1259 : * locktag is defined for LOCKTAG_RELATION
1260 : */
1261 194800 : LogAccessExclusiveLock(locktag->locktag_field1,
1262 : locktag->locktag_field2);
1263 : }
1264 :
1265 4348202 : return LOCKACQUIRE_OK;
1266 : }
1267 :
1268 : /*
1269 : * Find or create LOCK and PROCLOCK objects as needed for a new lock
1270 : * request.
1271 : *
1272 : * Returns the PROCLOCK object, or NULL if we failed to create the objects
1273 : * for lack of shared memory.
1274 : *
1275 : * The appropriate partition lock must be held at entry, and will be
1276 : * held at exit.
1277 : */
1278 : static PROCLOCK *
1279 4353034 : SetupLockInTable(LockMethod lockMethodTable, PGPROC *proc,
1280 : const LOCKTAG *locktag, uint32 hashcode, LOCKMODE lockmode)
1281 : {
1282 : LOCK *lock;
1283 : PROCLOCK *proclock;
1284 : PROCLOCKTAG proclocktag;
1285 : uint32 proclock_hashcode;
1286 : bool found;
1287 :
1288 : /*
1289 : * Find or create a lock with this tag.
1290 : */
1291 4353034 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
1292 : locktag,
1293 : hashcode,
1294 : HASH_ENTER_NULL,
1295 : &found);
1296 4353034 : if (!lock)
1297 0 : return NULL;
1298 :
1299 : /*
1300 : * if it's a new lock object, initialize it
1301 : */
1302 4353034 : if (!found)
1303 : {
1304 3929118 : lock->grantMask = 0;
1305 3929118 : lock->waitMask = 0;
1306 3929118 : dlist_init(&lock->procLocks);
1307 3929118 : dclist_init(&lock->waitProcs);
1308 3929118 : lock->nRequested = 0;
1309 3929118 : lock->nGranted = 0;
1310 23574708 : MemSet(lock->requested, 0, sizeof(int) * MAX_LOCKMODES);
1311 3929118 : MemSet(lock->granted, 0, sizeof(int) * MAX_LOCKMODES);
1312 : LOCK_PRINT("LockAcquire: new", lock, lockmode);
1313 : }
1314 : else
1315 : {
1316 : LOCK_PRINT("LockAcquire: found", lock, lockmode);
1317 : Assert((lock->nRequested >= 0) && (lock->requested[lockmode] >= 0));
1318 : Assert((lock->nGranted >= 0) && (lock->granted[lockmode] >= 0));
1319 : Assert(lock->nGranted <= lock->nRequested);
1320 : }
1321 :
1322 : /*
1323 : * Create the hash key for the proclock table.
1324 : */
1325 4353034 : proclocktag.myLock = lock;
1326 4353034 : proclocktag.myProc = proc;
1327 :
1328 4353034 : proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
1329 :
1330 : /*
1331 : * Find or create a proclock entry with this tag
1332 : */
1333 4353034 : proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
1334 : &proclocktag,
1335 : proclock_hashcode,
1336 : HASH_ENTER_NULL,
1337 : &found);
1338 4353034 : if (!proclock)
1339 : {
1340 : /* Oops, not enough shmem for the proclock */
1341 0 : if (lock->nRequested == 0)
1342 : {
1343 : /*
1344 : * There are no other requestors of this lock, so garbage-collect
1345 : * the lock object. We *must* do this to avoid a permanent leak
1346 : * of shared memory, because there won't be anything to cause
1347 : * anyone to release the lock object later.
1348 : */
1349 : Assert(dlist_is_empty(&(lock->procLocks)));
1350 0 : if (!hash_search_with_hash_value(LockMethodLockHash,
1351 0 : &(lock->tag),
1352 : hashcode,
1353 : HASH_REMOVE,
1354 : NULL))
1355 0 : elog(PANIC, "lock table corrupted");
1356 : }
1357 0 : return NULL;
1358 : }
1359 :
1360 : /*
1361 : * If new, initialize the new entry
1362 : */
1363 4353034 : if (!found)
1364 : {
1365 3959370 : uint32 partition = LockHashPartition(hashcode);
1366 :
1367 : /*
1368 : * It might seem unsafe to access proclock->groupLeader without a
1369 : * lock, but it's not really. Either we are initializing a proclock
1370 : * on our own behalf, in which case our group leader isn't changing
1371 : * because the group leader for a process can only ever be changed by
1372 : * the process itself; or else we are transferring a fast-path lock to
1373 : * the main lock table, in which case that process can't change its
1374 : * lock group leader without first releasing all of its locks (and in
1375 : * particular the one we are currently transferring).
1376 : */
1377 7918740 : proclock->groupLeader = proc->lockGroupLeader != NULL ?
1378 3959370 : proc->lockGroupLeader : proc;
1379 3959370 : proclock->holdMask = 0;
1380 3959370 : proclock->releaseMask = 0;
1381 : /* Add proclock to appropriate lists */
1382 3959370 : dlist_push_tail(&lock->procLocks, &proclock->lockLink);
1383 3959370 : dlist_push_tail(&proc->myProcLocks[partition], &proclock->procLink);
1384 : PROCLOCK_PRINT("LockAcquire: new", proclock);
1385 : }
1386 : else
1387 : {
1388 : PROCLOCK_PRINT("LockAcquire: found", proclock);
1389 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
1390 :
1391 : #ifdef CHECK_DEADLOCK_RISK
1392 :
1393 : /*
1394 : * Issue warning if we already hold a lower-level lock on this object
1395 : * and do not hold a lock of the requested level or higher. This
1396 : * indicates a deadlock-prone coding practice (eg, we'd have a
1397 : * deadlock if another backend were following the same code path at
1398 : * about the same time).
1399 : *
1400 : * This is not enabled by default, because it may generate log entries
1401 : * about user-level coding practices that are in fact safe in context.
1402 : * It can be enabled to help find system-level problems.
1403 : *
1404 : * XXX Doing numeric comparison on the lockmodes is a hack; it'd be
1405 : * better to use a table. For now, though, this works.
1406 : */
1407 : {
1408 : int i;
1409 :
1410 : for (i = lockMethodTable->numLockModes; i > 0; i--)
1411 : {
1412 : if (proclock->holdMask & LOCKBIT_ON(i))
1413 : {
1414 : if (i >= (int) lockmode)
1415 : break; /* safe: we have a lock >= req level */
1416 : elog(LOG, "deadlock risk: raising lock level"
1417 : " from %s to %s on object %u/%u/%u",
1418 : lockMethodTable->lockModeNames[i],
1419 : lockMethodTable->lockModeNames[lockmode],
1420 : lock->tag.locktag_field1, lock->tag.locktag_field2,
1421 : lock->tag.locktag_field3);
1422 : break;
1423 : }
1424 : }
1425 : }
1426 : #endif /* CHECK_DEADLOCK_RISK */
1427 : }
1428 :
1429 : /*
1430 : * lock->nRequested and lock->requested[] count the total number of
1431 : * requests, whether granted or waiting, so increment those immediately.
1432 : * The other counts don't increment till we get the lock.
1433 : */
1434 4353034 : lock->nRequested++;
1435 4353034 : lock->requested[lockmode]++;
1436 : Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
1437 :
1438 : /*
1439 : * We shouldn't already hold the desired lock; else locallock table is
1440 : * broken.
1441 : */
1442 4353034 : if (proclock->holdMask & LOCKBIT_ON(lockmode))
1443 0 : elog(ERROR, "lock %s on object %u/%u/%u is already held",
1444 : lockMethodTable->lockModeNames[lockmode],
1445 : lock->tag.locktag_field1, lock->tag.locktag_field2,
1446 : lock->tag.locktag_field3);
1447 :
1448 4353034 : return proclock;
1449 : }
1450 :
1451 : /*
1452 : * Check and set/reset the flag that we hold the relation extension lock.
1453 : *
1454 : * It is callers responsibility that this function is called after
1455 : * acquiring/releasing the relation extension lock.
1456 : *
1457 : * Pass acquired as true if lock is acquired, false otherwise.
1458 : */
1459 : static inline void
1460 68788954 : CheckAndSetLockHeld(LOCALLOCK *locallock, bool acquired)
1461 : {
1462 : #ifdef USE_ASSERT_CHECKING
1463 : if (LOCALLOCK_LOCKTAG(*locallock) == LOCKTAG_RELATION_EXTEND)
1464 : IsRelationExtensionLockHeld = acquired;
1465 : #endif
1466 68788954 : }
1467 :
1468 : /*
1469 : * Subroutine to free a locallock entry
1470 : */
1471 : static void
1472 33799406 : RemoveLocalLock(LOCALLOCK *locallock)
1473 : {
1474 : int i;
1475 :
1476 33954622 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
1477 : {
1478 155216 : if (locallock->lockOwners[i].owner != NULL)
1479 155142 : ResourceOwnerForgetLock(locallock->lockOwners[i].owner, locallock);
1480 : }
1481 33799406 : locallock->numLockOwners = 0;
1482 33799406 : if (locallock->lockOwners != NULL)
1483 33799406 : pfree(locallock->lockOwners);
1484 33799406 : locallock->lockOwners = NULL;
1485 :
1486 33799406 : if (locallock->holdsStrongLockCount)
1487 : {
1488 : uint32 fasthashcode;
1489 :
1490 359134 : fasthashcode = FastPathStrongLockHashPartition(locallock->hashcode);
1491 :
1492 359134 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1493 : Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
1494 359134 : FastPathStrongRelationLocks->count[fasthashcode]--;
1495 359134 : locallock->holdsStrongLockCount = false;
1496 359134 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1497 : }
1498 :
1499 33799406 : if (!hash_search(LockMethodLocalHash,
1500 33799406 : &(locallock->tag),
1501 : HASH_REMOVE, NULL))
1502 0 : elog(WARNING, "locallock table corrupted");
1503 :
1504 : /*
1505 : * Indicate that the lock is released for certain types of locks
1506 : */
1507 33799406 : CheckAndSetLockHeld(locallock, false);
1508 33799406 : }
1509 :
1510 : /*
1511 : * LockCheckConflicts -- test whether requested lock conflicts
1512 : * with those already granted
1513 : *
1514 : * Returns true if conflict, false if no conflict.
1515 : *
1516 : * NOTES:
1517 : * Here's what makes this complicated: one process's locks don't
1518 : * conflict with one another, no matter what purpose they are held for
1519 : * (eg, session and transaction locks do not conflict). Nor do the locks
1520 : * of one process in a lock group conflict with those of another process in
1521 : * the same group. So, we must subtract off these locks when determining
1522 : * whether the requested new lock conflicts with those already held.
1523 : */
1524 : bool
1525 4352260 : LockCheckConflicts(LockMethod lockMethodTable,
1526 : LOCKMODE lockmode,
1527 : LOCK *lock,
1528 : PROCLOCK *proclock)
1529 : {
1530 4352260 : int numLockModes = lockMethodTable->numLockModes;
1531 : LOCKMASK myLocks;
1532 4352260 : int conflictMask = lockMethodTable->conflictTab[lockmode];
1533 : int conflictsRemaining[MAX_LOCKMODES];
1534 4352260 : int totalConflictsRemaining = 0;
1535 : dlist_iter proclock_iter;
1536 : int i;
1537 :
1538 : /*
1539 : * first check for global conflicts: If no locks conflict with my request,
1540 : * then I get the lock.
1541 : *
1542 : * Checking for conflict: lock->grantMask represents the types of
1543 : * currently held locks. conflictTable[lockmode] has a bit set for each
1544 : * type of lock that conflicts with request. Bitwise compare tells if
1545 : * there is a conflict.
1546 : */
1547 4352260 : if (!(conflictMask & lock->grantMask))
1548 : {
1549 : PROCLOCK_PRINT("LockCheckConflicts: no conflict", proclock);
1550 4170362 : return false;
1551 : }
1552 :
1553 : /*
1554 : * Rats. Something conflicts. But it could still be my own lock, or a
1555 : * lock held by another member of my locking group. First, figure out how
1556 : * many conflicts remain after subtracting out any locks I hold myself.
1557 : */
1558 181898 : myLocks = proclock->holdMask;
1559 1637082 : for (i = 1; i <= numLockModes; i++)
1560 : {
1561 1455184 : if ((conflictMask & LOCKBIT_ON(i)) == 0)
1562 : {
1563 779350 : conflictsRemaining[i] = 0;
1564 779350 : continue;
1565 : }
1566 675834 : conflictsRemaining[i] = lock->granted[i];
1567 675834 : if (myLocks & LOCKBIT_ON(i))
1568 196928 : --conflictsRemaining[i];
1569 675834 : totalConflictsRemaining += conflictsRemaining[i];
1570 : }
1571 :
1572 : /* If no conflicts remain, we get the lock. */
1573 181898 : if (totalConflictsRemaining == 0)
1574 : {
1575 : PROCLOCK_PRINT("LockCheckConflicts: resolved (simple)", proclock);
1576 176900 : return false;
1577 : }
1578 :
1579 : /* If no group locking, it's definitely a conflict. */
1580 4998 : if (proclock->groupLeader == MyProc && MyProc->lockGroupLeader == NULL)
1581 : {
1582 : Assert(proclock->tag.myProc == MyProc);
1583 : PROCLOCK_PRINT("LockCheckConflicts: conflicting (simple)",
1584 : proclock);
1585 3644 : return true;
1586 : }
1587 :
1588 : /*
1589 : * The relation extension lock conflict even between the group members.
1590 : */
1591 1354 : if (LOCK_LOCKTAG(*lock) == LOCKTAG_RELATION_EXTEND)
1592 : {
1593 : PROCLOCK_PRINT("LockCheckConflicts: conflicting (group)",
1594 : proclock);
1595 76 : return true;
1596 : }
1597 :
1598 : /*
1599 : * Locks held in conflicting modes by members of our own lock group are
1600 : * not real conflicts; we can subtract those out and see if we still have
1601 : * a conflict. This is O(N) in the number of processes holding or
1602 : * awaiting locks on this object. We could improve that by making the
1603 : * shared memory state more complex (and larger) but it doesn't seem worth
1604 : * it.
1605 : */
1606 2204 : dlist_foreach(proclock_iter, &lock->procLocks)
1607 : {
1608 1894 : PROCLOCK *otherproclock =
1609 1894 : dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
1610 :
1611 1894 : if (proclock != otherproclock &&
1612 1584 : proclock->groupLeader == otherproclock->groupLeader &&
1613 972 : (otherproclock->holdMask & conflictMask) != 0)
1614 : {
1615 968 : int intersectMask = otherproclock->holdMask & conflictMask;
1616 :
1617 8712 : for (i = 1; i <= numLockModes; i++)
1618 : {
1619 7744 : if ((intersectMask & LOCKBIT_ON(i)) != 0)
1620 : {
1621 988 : if (conflictsRemaining[i] <= 0)
1622 0 : elog(PANIC, "proclocks held do not match lock");
1623 988 : conflictsRemaining[i]--;
1624 988 : totalConflictsRemaining--;
1625 : }
1626 : }
1627 :
1628 968 : if (totalConflictsRemaining == 0)
1629 : {
1630 : PROCLOCK_PRINT("LockCheckConflicts: resolved (group)",
1631 : proclock);
1632 968 : return false;
1633 : }
1634 : }
1635 : }
1636 :
1637 : /* Nope, it's a real conflict. */
1638 : PROCLOCK_PRINT("LockCheckConflicts: conflicting (group)", proclock);
1639 310 : return true;
1640 : }
1641 :
1642 : /*
1643 : * GrantLock -- update the lock and proclock data structures to show
1644 : * the lock request has been granted.
1645 : *
1646 : * NOTE: if proc was blocked, it also needs to be removed from the wait list
1647 : * and have its waitLock/waitProcLock fields cleared. That's not done here.
1648 : *
1649 : * NOTE: the lock grant also has to be recorded in the associated LOCALLOCK
1650 : * table entry; but since we may be awaking some other process, we can't do
1651 : * that here; it's done by GrantLockLocal, instead.
1652 : */
1653 : void
1654 4351682 : GrantLock(LOCK *lock, PROCLOCK *proclock, LOCKMODE lockmode)
1655 : {
1656 4351682 : lock->nGranted++;
1657 4351682 : lock->granted[lockmode]++;
1658 4351682 : lock->grantMask |= LOCKBIT_ON(lockmode);
1659 4351682 : if (lock->granted[lockmode] == lock->requested[lockmode])
1660 4351116 : lock->waitMask &= LOCKBIT_OFF(lockmode);
1661 4351682 : proclock->holdMask |= LOCKBIT_ON(lockmode);
1662 : LOCK_PRINT("GrantLock", lock, lockmode);
1663 : Assert((lock->nGranted > 0) && (lock->granted[lockmode] > 0));
1664 : Assert(lock->nGranted <= lock->nRequested);
1665 4351682 : }
1666 :
1667 : /*
1668 : * UnGrantLock -- opposite of GrantLock.
1669 : *
1670 : * Updates the lock and proclock data structures to show that the lock
1671 : * is no longer held nor requested by the current holder.
1672 : *
1673 : * Returns true if there were any waiters waiting on the lock that
1674 : * should now be woken up with ProcLockWakeup.
1675 : */
1676 : static bool
1677 4351530 : UnGrantLock(LOCK *lock, LOCKMODE lockmode,
1678 : PROCLOCK *proclock, LockMethod lockMethodTable)
1679 : {
1680 4351530 : bool wakeupNeeded = false;
1681 :
1682 : Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
1683 : Assert((lock->nGranted > 0) && (lock->granted[lockmode] > 0));
1684 : Assert(lock->nGranted <= lock->nRequested);
1685 :
1686 : /*
1687 : * fix the general lock stats
1688 : */
1689 4351530 : lock->nRequested--;
1690 4351530 : lock->requested[lockmode]--;
1691 4351530 : lock->nGranted--;
1692 4351530 : lock->granted[lockmode]--;
1693 :
1694 4351530 : if (lock->granted[lockmode] == 0)
1695 : {
1696 : /* change the conflict mask. No more of this lock type. */
1697 4331032 : lock->grantMask &= LOCKBIT_OFF(lockmode);
1698 : }
1699 :
1700 : LOCK_PRINT("UnGrantLock: updated", lock, lockmode);
1701 :
1702 : /*
1703 : * We need only run ProcLockWakeup if the released lock conflicts with at
1704 : * least one of the lock types requested by waiter(s). Otherwise whatever
1705 : * conflict made them wait must still exist. NOTE: before MVCC, we could
1706 : * skip wakeup if lock->granted[lockmode] was still positive. But that's
1707 : * not true anymore, because the remaining granted locks might belong to
1708 : * some waiter, who could now be awakened because he doesn't conflict with
1709 : * his own locks.
1710 : */
1711 4351530 : if (lockMethodTable->conflictTab[lockmode] & lock->waitMask)
1712 2438 : wakeupNeeded = true;
1713 :
1714 : /*
1715 : * Now fix the per-proclock state.
1716 : */
1717 4351530 : proclock->holdMask &= LOCKBIT_OFF(lockmode);
1718 : PROCLOCK_PRINT("UnGrantLock: updated", proclock);
1719 :
1720 4351530 : return wakeupNeeded;
1721 : }
1722 :
1723 : /*
1724 : * CleanUpLock -- clean up after releasing a lock. We garbage-collect the
1725 : * proclock and lock objects if possible, and call ProcLockWakeup if there
1726 : * are remaining requests and the caller says it's OK. (Normally, this
1727 : * should be called after UnGrantLock, and wakeupNeeded is the result from
1728 : * UnGrantLock.)
1729 : *
1730 : * The appropriate partition lock must be held at entry, and will be
1731 : * held at exit.
1732 : */
1733 : static void
1734 4282342 : CleanUpLock(LOCK *lock, PROCLOCK *proclock,
1735 : LockMethod lockMethodTable, uint32 hashcode,
1736 : bool wakeupNeeded)
1737 : {
1738 : /*
1739 : * If this was my last hold on this lock, delete my entry in the proclock
1740 : * table.
1741 : */
1742 4282342 : if (proclock->holdMask == 0)
1743 : {
1744 : uint32 proclock_hashcode;
1745 :
1746 : PROCLOCK_PRINT("CleanUpLock: deleting", proclock);
1747 3958356 : dlist_delete(&proclock->lockLink);
1748 3958356 : dlist_delete(&proclock->procLink);
1749 3958356 : proclock_hashcode = ProcLockHashCode(&proclock->tag, hashcode);
1750 3958356 : if (!hash_search_with_hash_value(LockMethodProcLockHash,
1751 3958356 : &(proclock->tag),
1752 : proclock_hashcode,
1753 : HASH_REMOVE,
1754 : NULL))
1755 0 : elog(PANIC, "proclock table corrupted");
1756 : }
1757 :
1758 4282342 : if (lock->nRequested == 0)
1759 : {
1760 : /*
1761 : * The caller just released the last lock, so garbage-collect the lock
1762 : * object.
1763 : */
1764 : LOCK_PRINT("CleanUpLock: deleting", lock, 0);
1765 : Assert(dlist_is_empty(&lock->procLocks));
1766 3929130 : if (!hash_search_with_hash_value(LockMethodLockHash,
1767 3929130 : &(lock->tag),
1768 : hashcode,
1769 : HASH_REMOVE,
1770 : NULL))
1771 0 : elog(PANIC, "lock table corrupted");
1772 : }
1773 353212 : else if (wakeupNeeded)
1774 : {
1775 : /* There are waiters on this lock, so wake them up. */
1776 2524 : ProcLockWakeup(lockMethodTable, lock);
1777 : }
1778 4282342 : }
1779 :
1780 : /*
1781 : * GrantLockLocal -- update the locallock data structures to show
1782 : * the lock request has been granted.
1783 : *
1784 : * We expect that LockAcquire made sure there is room to add a new
1785 : * ResourceOwner entry.
1786 : */
1787 : static void
1788 37958038 : GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner)
1789 : {
1790 37958038 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
1791 : int i;
1792 :
1793 : Assert(locallock->numLockOwners < locallock->maxLockOwners);
1794 : /* Count the total */
1795 37958038 : locallock->nLocks++;
1796 : /* Count the per-owner lock */
1797 39720810 : for (i = 0; i < locallock->numLockOwners; i++)
1798 : {
1799 4731262 : if (lockOwners[i].owner == owner)
1800 : {
1801 2968490 : lockOwners[i].nLocks++;
1802 2968490 : return;
1803 : }
1804 : }
1805 34989548 : lockOwners[i].owner = owner;
1806 34989548 : lockOwners[i].nLocks = 1;
1807 34989548 : locallock->numLockOwners++;
1808 34989548 : if (owner != NULL)
1809 34816868 : ResourceOwnerRememberLock(owner, locallock);
1810 :
1811 : /* Indicate that the lock is acquired for certain types of locks. */
1812 34989548 : CheckAndSetLockHeld(locallock, true);
1813 : }
1814 :
1815 : /*
1816 : * BeginStrongLockAcquire - inhibit use of fastpath for a given LOCALLOCK,
1817 : * and arrange for error cleanup if it fails
1818 : */
1819 : static void
1820 359710 : BeginStrongLockAcquire(LOCALLOCK *locallock, uint32 fasthashcode)
1821 : {
1822 : Assert(StrongLockInProgress == NULL);
1823 : Assert(locallock->holdsStrongLockCount == false);
1824 :
1825 : /*
1826 : * Adding to a memory location is not atomic, so we take a spinlock to
1827 : * ensure we don't collide with someone else trying to bump the count at
1828 : * the same time.
1829 : *
1830 : * XXX: It might be worth considering using an atomic fetch-and-add
1831 : * instruction here, on architectures where that is supported.
1832 : */
1833 :
1834 359710 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1835 359710 : FastPathStrongRelationLocks->count[fasthashcode]++;
1836 359710 : locallock->holdsStrongLockCount = true;
1837 359710 : StrongLockInProgress = locallock;
1838 359710 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1839 359710 : }
1840 :
1841 : /*
1842 : * FinishStrongLockAcquire - cancel pending cleanup for a strong lock
1843 : * acquisition once it's no longer needed
1844 : */
1845 : static void
1846 4348202 : FinishStrongLockAcquire(void)
1847 : {
1848 4348202 : StrongLockInProgress = NULL;
1849 4348202 : }
1850 :
1851 : /*
1852 : * AbortStrongLockAcquire - undo strong lock state changes performed by
1853 : * BeginStrongLockAcquire.
1854 : */
1855 : void
1856 884252 : AbortStrongLockAcquire(void)
1857 : {
1858 : uint32 fasthashcode;
1859 884252 : LOCALLOCK *locallock = StrongLockInProgress;
1860 :
1861 884252 : if (locallock == NULL)
1862 883826 : return;
1863 :
1864 426 : fasthashcode = FastPathStrongLockHashPartition(locallock->hashcode);
1865 : Assert(locallock->holdsStrongLockCount == true);
1866 426 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1867 : Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
1868 426 : FastPathStrongRelationLocks->count[fasthashcode]--;
1869 426 : locallock->holdsStrongLockCount = false;
1870 426 : StrongLockInProgress = NULL;
1871 426 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1872 : }
1873 :
1874 : /*
1875 : * GrantAwaitedLock -- call GrantLockLocal for the lock we are doing
1876 : * WaitOnLock on.
1877 : *
1878 : * proc.c needs this for the case where we are booted off the lock by
1879 : * timeout, but discover that someone granted us the lock anyway.
1880 : *
1881 : * We could just export GrantLockLocal, but that would require including
1882 : * resowner.h in lock.h, which creates circularity.
1883 : */
1884 : void
1885 0 : GrantAwaitedLock(void)
1886 : {
1887 0 : GrantLockLocal(awaitedLock, awaitedOwner);
1888 0 : }
1889 :
1890 : /*
1891 : * GetAwaitedLock -- Return the lock we're currently doing WaitOnLock on.
1892 : */
1893 : LOCALLOCK *
1894 882806 : GetAwaitedLock(void)
1895 : {
1896 882806 : return awaitedLock;
1897 : }
1898 :
1899 : /*
1900 : * ResetAwaitedLock -- Forget that we are waiting on a lock.
1901 : */
1902 : void
1903 82 : ResetAwaitedLock(void)
1904 : {
1905 82 : awaitedLock = NULL;
1906 82 : }
1907 :
1908 : /*
1909 : * MarkLockClear -- mark an acquired lock as "clear"
1910 : *
1911 : * This means that we know we have absorbed all sinval messages that other
1912 : * sessions generated before we acquired this lock, and so we can confidently
1913 : * assume we know about any catalog changes protected by this lock.
1914 : */
1915 : void
1916 32570364 : MarkLockClear(LOCALLOCK *locallock)
1917 : {
1918 : Assert(locallock->nLocks > 0);
1919 32570364 : locallock->lockCleared = true;
1920 32570364 : }
1921 :
1922 : /*
1923 : * WaitOnLock -- wait to acquire a lock
1924 : *
1925 : * This is a wrapper around ProcSleep, with extra tracing and bookkeeping.
1926 : */
1927 : static ProcWaitStatus
1928 2556 : WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner)
1929 : {
1930 : ProcWaitStatus result;
1931 :
1932 : TRACE_POSTGRESQL_LOCK_WAIT_START(locallock->tag.lock.locktag_field1,
1933 : locallock->tag.lock.locktag_field2,
1934 : locallock->tag.lock.locktag_field3,
1935 : locallock->tag.lock.locktag_field4,
1936 : locallock->tag.lock.locktag_type,
1937 : locallock->tag.mode);
1938 :
1939 : /* adjust the process title to indicate that it's waiting */
1940 2556 : set_ps_display_suffix("waiting");
1941 :
1942 : /*
1943 : * Record the fact that we are waiting for a lock, so that
1944 : * LockErrorCleanup will clean up if cancel/die happens.
1945 : */
1946 2556 : awaitedLock = locallock;
1947 2556 : awaitedOwner = owner;
1948 :
1949 : /*
1950 : * NOTE: Think not to put any shared-state cleanup after the call to
1951 : * ProcSleep, in either the normal or failure path. The lock state must
1952 : * be fully set by the lock grantor, or by CheckDeadLock if we give up
1953 : * waiting for the lock. This is necessary because of the possibility
1954 : * that a cancel/die interrupt will interrupt ProcSleep after someone else
1955 : * grants us the lock, but before we've noticed it. Hence, after granting,
1956 : * the locktable state must fully reflect the fact that we own the lock;
1957 : * we can't do additional work on return.
1958 : *
1959 : * We can and do use a PG_TRY block to try to clean up after failure, but
1960 : * this still has a major limitation: elog(FATAL) can occur while waiting
1961 : * (eg, a "die" interrupt), and then control won't come back here. So all
1962 : * cleanup of essential state should happen in LockErrorCleanup, not here.
1963 : * We can use PG_TRY to clear the "waiting" status flags, since doing that
1964 : * is unimportant if the process exits.
1965 : */
1966 2556 : PG_TRY();
1967 : {
1968 2556 : result = ProcSleep(locallock);
1969 : }
1970 74 : PG_CATCH();
1971 : {
1972 : /* In this path, awaitedLock remains set until LockErrorCleanup */
1973 :
1974 : /* reset ps display to remove the suffix */
1975 74 : set_ps_display_remove_suffix();
1976 :
1977 : /* and propagate the error */
1978 74 : PG_RE_THROW();
1979 : }
1980 2472 : PG_END_TRY();
1981 :
1982 : /*
1983 : * We no longer want LockErrorCleanup to do anything.
1984 : */
1985 2472 : awaitedLock = NULL;
1986 :
1987 : /* reset ps display to remove the suffix */
1988 2472 : set_ps_display_remove_suffix();
1989 :
1990 : TRACE_POSTGRESQL_LOCK_WAIT_DONE(locallock->tag.lock.locktag_field1,
1991 : locallock->tag.lock.locktag_field2,
1992 : locallock->tag.lock.locktag_field3,
1993 : locallock->tag.lock.locktag_field4,
1994 : locallock->tag.lock.locktag_type,
1995 : locallock->tag.mode);
1996 :
1997 2472 : return result;
1998 : }
1999 :
2000 : /*
2001 : * Remove a proc from the wait-queue it is on (caller must know it is on one).
2002 : * This is only used when the proc has failed to get the lock, so we set its
2003 : * waitStatus to PROC_WAIT_STATUS_ERROR.
2004 : *
2005 : * Appropriate partition lock must be held by caller. Also, caller is
2006 : * responsible for signaling the proc if needed.
2007 : *
2008 : * NB: this does not clean up any locallock object that may exist for the lock.
2009 : */
2010 : void
2011 92 : RemoveFromWaitQueue(PGPROC *proc, uint32 hashcode)
2012 : {
2013 92 : LOCK *waitLock = proc->waitLock;
2014 92 : PROCLOCK *proclock = proc->waitProcLock;
2015 92 : LOCKMODE lockmode = proc->waitLockMode;
2016 92 : LOCKMETHODID lockmethodid = LOCK_LOCKMETHOD(*waitLock);
2017 :
2018 : /* Make sure proc is waiting */
2019 : Assert(proc->waitStatus == PROC_WAIT_STATUS_WAITING);
2020 : Assert(proc->links.next != NULL);
2021 : Assert(waitLock);
2022 : Assert(!dclist_is_empty(&waitLock->waitProcs));
2023 : Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
2024 :
2025 : /* Remove proc from lock's wait queue */
2026 92 : dclist_delete_from_thoroughly(&waitLock->waitProcs, &proc->links);
2027 :
2028 : /* Undo increments of request counts by waiting process */
2029 : Assert(waitLock->nRequested > 0);
2030 : Assert(waitLock->nRequested > proc->waitLock->nGranted);
2031 92 : waitLock->nRequested--;
2032 : Assert(waitLock->requested[lockmode] > 0);
2033 92 : waitLock->requested[lockmode]--;
2034 : /* don't forget to clear waitMask bit if appropriate */
2035 92 : if (waitLock->granted[lockmode] == waitLock->requested[lockmode])
2036 90 : waitLock->waitMask &= LOCKBIT_OFF(lockmode);
2037 :
2038 : /* Clean up the proc's own state, and pass it the ok/fail signal */
2039 92 : proc->waitLock = NULL;
2040 92 : proc->waitProcLock = NULL;
2041 92 : proc->waitStatus = PROC_WAIT_STATUS_ERROR;
2042 :
2043 : /*
2044 : * Delete the proclock immediately if it represents no already-held locks.
2045 : * (This must happen now because if the owner of the lock decides to
2046 : * release it, and the requested/granted counts then go to zero,
2047 : * LockRelease expects there to be no remaining proclocks.) Then see if
2048 : * any other waiters for the lock can be woken up now.
2049 : */
2050 92 : CleanUpLock(waitLock, proclock,
2051 : LockMethods[lockmethodid], hashcode,
2052 : true);
2053 92 : }
2054 :
2055 : /*
2056 : * LockRelease -- look up 'locktag' and release one 'lockmode' lock on it.
2057 : * Release a session lock if 'sessionLock' is true, else release a
2058 : * regular transaction lock.
2059 : *
2060 : * Side Effects: find any waiting processes that are now wakable,
2061 : * grant them their requested locks and awaken them.
2062 : * (We have to grant the lock here to avoid a race between
2063 : * the waking process and any new process to
2064 : * come along and request the lock.)
2065 : */
2066 : bool
2067 33636036 : LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
2068 : {
2069 33636036 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
2070 : LockMethod lockMethodTable;
2071 : LOCALLOCKTAG localtag;
2072 : LOCALLOCK *locallock;
2073 : LOCK *lock;
2074 : PROCLOCK *proclock;
2075 : LWLock *partitionLock;
2076 : bool wakeupNeeded;
2077 :
2078 33636036 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2079 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2080 33636036 : lockMethodTable = LockMethods[lockmethodid];
2081 33636036 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
2082 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
2083 :
2084 : #ifdef LOCK_DEBUG
2085 : if (LOCK_DEBUG_ENABLED(locktag))
2086 : elog(LOG, "LockRelease: lock [%u,%u] %s",
2087 : locktag->locktag_field1, locktag->locktag_field2,
2088 : lockMethodTable->lockModeNames[lockmode]);
2089 : #endif
2090 :
2091 : /*
2092 : * Find the LOCALLOCK entry for this lock and lockmode
2093 : */
2094 33636036 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
2095 33636036 : localtag.lock = *locktag;
2096 33636036 : localtag.mode = lockmode;
2097 :
2098 33636036 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
2099 : &localtag,
2100 : HASH_FIND, NULL);
2101 :
2102 : /*
2103 : * let the caller print its own error message, too. Do not ereport(ERROR).
2104 : */
2105 33636036 : if (!locallock || locallock->nLocks <= 0)
2106 : {
2107 26 : elog(WARNING, "you don't own a lock of type %s",
2108 : lockMethodTable->lockModeNames[lockmode]);
2109 26 : return false;
2110 : }
2111 :
2112 : /*
2113 : * Decrease the count for the resource owner.
2114 : */
2115 : {
2116 33636010 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
2117 : ResourceOwner owner;
2118 : int i;
2119 :
2120 : /* Identify owner for lock */
2121 33636010 : if (sessionLock)
2122 172658 : owner = NULL;
2123 : else
2124 33463352 : owner = CurrentResourceOwner;
2125 :
2126 33637728 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
2127 : {
2128 33637704 : if (lockOwners[i].owner == owner)
2129 : {
2130 : Assert(lockOwners[i].nLocks > 0);
2131 33635986 : if (--lockOwners[i].nLocks == 0)
2132 : {
2133 32459360 : if (owner != NULL)
2134 32286754 : ResourceOwnerForgetLock(owner, locallock);
2135 : /* compact out unused slot */
2136 32459360 : locallock->numLockOwners--;
2137 32459360 : if (i < locallock->numLockOwners)
2138 120 : lockOwners[i] = lockOwners[locallock->numLockOwners];
2139 : }
2140 33635986 : break;
2141 : }
2142 : }
2143 33636010 : if (i < 0)
2144 : {
2145 : /* don't release a lock belonging to another owner */
2146 24 : elog(WARNING, "you don't own a lock of type %s",
2147 : lockMethodTable->lockModeNames[lockmode]);
2148 24 : return false;
2149 : }
2150 : }
2151 :
2152 : /*
2153 : * Decrease the total local count. If we're still holding the lock, we're
2154 : * done.
2155 : */
2156 33635986 : locallock->nLocks--;
2157 :
2158 33635986 : if (locallock->nLocks > 0)
2159 1898916 : return true;
2160 :
2161 : /*
2162 : * At this point we can no longer suppose we are clear of invalidation
2163 : * messages related to this lock. Although we'll delete the LOCALLOCK
2164 : * object before any intentional return from this routine, it seems worth
2165 : * the trouble to explicitly reset lockCleared right now, just in case
2166 : * some error prevents us from deleting the LOCALLOCK.
2167 : */
2168 31737070 : locallock->lockCleared = false;
2169 :
2170 : /* Attempt fast release of any lock eligible for the fast path. */
2171 31737070 : if (EligibleForRelationFastPath(locktag, lockmode) &&
2172 29181272 : FastPathLocalUseCounts[FAST_PATH_REL_GROUP(locktag->locktag_field2)] > 0)
2173 : {
2174 : bool released;
2175 :
2176 : /*
2177 : * We might not find the lock here, even if we originally entered it
2178 : * here. Another backend may have moved it to the main table.
2179 : */
2180 28780466 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
2181 28780466 : released = FastPathUnGrantRelationLock(locktag->locktag_field2,
2182 : lockmode);
2183 28780466 : LWLockRelease(&MyProc->fpInfoLock);
2184 28780466 : if (released)
2185 : {
2186 28405720 : RemoveLocalLock(locallock);
2187 28405720 : return true;
2188 : }
2189 : }
2190 :
2191 : /*
2192 : * Otherwise we've got to mess with the shared lock table.
2193 : */
2194 3331350 : partitionLock = LockHashPartitionLock(locallock->hashcode);
2195 :
2196 3331350 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2197 :
2198 : /*
2199 : * Normally, we don't need to re-find the lock or proclock, since we kept
2200 : * their addresses in the locallock table, and they couldn't have been
2201 : * removed while we were holding a lock on them. But it's possible that
2202 : * the lock was taken fast-path and has since been moved to the main hash
2203 : * table by another backend, in which case we will need to look up the
2204 : * objects here. We assume the lock field is NULL if so.
2205 : */
2206 3331350 : lock = locallock->lock;
2207 3331350 : if (!lock)
2208 : {
2209 : PROCLOCKTAG proclocktag;
2210 :
2211 : Assert(EligibleForRelationFastPath(locktag, lockmode));
2212 10 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
2213 : locktag,
2214 : locallock->hashcode,
2215 : HASH_FIND,
2216 : NULL);
2217 10 : if (!lock)
2218 0 : elog(ERROR, "failed to re-find shared lock object");
2219 10 : locallock->lock = lock;
2220 :
2221 10 : proclocktag.myLock = lock;
2222 10 : proclocktag.myProc = MyProc;
2223 10 : locallock->proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash,
2224 : &proclocktag,
2225 : HASH_FIND,
2226 : NULL);
2227 10 : if (!locallock->proclock)
2228 0 : elog(ERROR, "failed to re-find shared proclock object");
2229 : }
2230 : LOCK_PRINT("LockRelease: found", lock, lockmode);
2231 3331350 : proclock = locallock->proclock;
2232 : PROCLOCK_PRINT("LockRelease: found", proclock);
2233 :
2234 : /*
2235 : * Double-check that we are actually holding a lock of the type we want to
2236 : * release.
2237 : */
2238 3331350 : if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
2239 : {
2240 : PROCLOCK_PRINT("LockRelease: WRONGTYPE", proclock);
2241 0 : LWLockRelease(partitionLock);
2242 0 : elog(WARNING, "you don't own a lock of type %s",
2243 : lockMethodTable->lockModeNames[lockmode]);
2244 0 : RemoveLocalLock(locallock);
2245 0 : return false;
2246 : }
2247 :
2248 : /*
2249 : * Do the releasing. CleanUpLock will waken any now-wakable waiters.
2250 : */
2251 3331350 : wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
2252 :
2253 3331350 : CleanUpLock(lock, proclock,
2254 : lockMethodTable, locallock->hashcode,
2255 : wakeupNeeded);
2256 :
2257 3331350 : LWLockRelease(partitionLock);
2258 :
2259 3331350 : RemoveLocalLock(locallock);
2260 3331350 : return true;
2261 : }
2262 :
2263 : /*
2264 : * LockReleaseAll -- Release all locks of the specified lock method that
2265 : * are held by the current process.
2266 : *
2267 : * Well, not necessarily *all* locks. The available behaviors are:
2268 : * allLocks == true: release all locks including session locks.
2269 : * allLocks == false: release all non-session locks.
2270 : */
2271 : void
2272 1683152 : LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
2273 : {
2274 : HASH_SEQ_STATUS status;
2275 : LockMethod lockMethodTable;
2276 : int i,
2277 : numLockModes;
2278 : LOCALLOCK *locallock;
2279 : LOCK *lock;
2280 : int partition;
2281 1683152 : bool have_fast_path_lwlock = false;
2282 :
2283 1683152 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2284 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2285 1683152 : lockMethodTable = LockMethods[lockmethodid];
2286 :
2287 : #ifdef LOCK_DEBUG
2288 : if (*(lockMethodTable->trace_flag))
2289 : elog(LOG, "LockReleaseAll: lockmethod=%d", lockmethodid);
2290 : #endif
2291 :
2292 : /*
2293 : * Get rid of our fast-path VXID lock, if appropriate. Note that this is
2294 : * the only way that the lock we hold on our own VXID can ever get
2295 : * released: it is always and only released when a toplevel transaction
2296 : * ends.
2297 : */
2298 1683152 : if (lockmethodid == DEFAULT_LOCKMETHOD)
2299 824760 : VirtualXactLockTableCleanup();
2300 :
2301 1683152 : numLockModes = lockMethodTable->numLockModes;
2302 :
2303 : /*
2304 : * First we run through the locallock table and get rid of unwanted
2305 : * entries, then we scan the process's proclocks and get rid of those. We
2306 : * do this separately because we may have multiple locallock entries
2307 : * pointing to the same proclock, and we daren't end up with any dangling
2308 : * pointers. Fast-path locks are cleaned up during the locallock table
2309 : * scan, though.
2310 : */
2311 1683152 : hash_seq_init(&status, LockMethodLocalHash);
2312 :
2313 4042828 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2314 : {
2315 : /*
2316 : * If the LOCALLOCK entry is unused, something must've gone wrong
2317 : * while trying to acquire this lock. Just forget the local entry.
2318 : */
2319 2359676 : if (locallock->nLocks == 0)
2320 : {
2321 94 : RemoveLocalLock(locallock);
2322 94 : continue;
2323 : }
2324 :
2325 : /* Ignore items that are not of the lockmethod to be removed */
2326 2359582 : if (LOCALLOCK_LOCKMETHOD(*locallock) != lockmethodid)
2327 150436 : continue;
2328 :
2329 : /*
2330 : * If we are asked to release all locks, we can just zap the entry.
2331 : * Otherwise, must scan to see if there are session locks. We assume
2332 : * there is at most one lockOwners entry for session locks.
2333 : */
2334 2209146 : if (!allLocks)
2335 : {
2336 2056824 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
2337 :
2338 : /* If session lock is above array position 0, move it down to 0 */
2339 4243464 : for (i = 0; i < locallock->numLockOwners; i++)
2340 : {
2341 2186640 : if (lockOwners[i].owner == NULL)
2342 149826 : lockOwners[0] = lockOwners[i];
2343 : else
2344 2036814 : ResourceOwnerForgetLock(lockOwners[i].owner, locallock);
2345 : }
2346 :
2347 2056824 : if (locallock->numLockOwners > 0 &&
2348 2056824 : lockOwners[0].owner == NULL &&
2349 149826 : lockOwners[0].nLocks > 0)
2350 : {
2351 : /* Fix the locallock to show just the session locks */
2352 149826 : locallock->nLocks = lockOwners[0].nLocks;
2353 149826 : locallock->numLockOwners = 1;
2354 : /* We aren't deleting this locallock, so done */
2355 149826 : continue;
2356 : }
2357 : else
2358 1906998 : locallock->numLockOwners = 0;
2359 : }
2360 :
2361 : #ifdef USE_ASSERT_CHECKING
2362 :
2363 : /*
2364 : * Tuple locks are currently held only for short durations within a
2365 : * transaction. Check that we didn't forget to release one.
2366 : */
2367 : if (LOCALLOCK_LOCKTAG(*locallock) == LOCKTAG_TUPLE && !allLocks)
2368 : elog(WARNING, "tuple lock held at commit");
2369 : #endif
2370 :
2371 : /*
2372 : * If the lock or proclock pointers are NULL, this lock was taken via
2373 : * the relation fast-path (and is not known to have been transferred).
2374 : */
2375 2059320 : if (locallock->proclock == NULL || locallock->lock == NULL)
2376 : {
2377 1043300 : LOCKMODE lockmode = locallock->tag.mode;
2378 : Oid relid;
2379 :
2380 : /* Verify that a fast-path lock is what we've got. */
2381 1043300 : if (!EligibleForRelationFastPath(&locallock->tag.lock, lockmode))
2382 0 : elog(PANIC, "locallock table corrupted");
2383 :
2384 : /*
2385 : * If we don't currently hold the LWLock that protects our
2386 : * fast-path data structures, we must acquire it before attempting
2387 : * to release the lock via the fast-path. We will continue to
2388 : * hold the LWLock until we're done scanning the locallock table,
2389 : * unless we hit a transferred fast-path lock. (XXX is this
2390 : * really such a good idea? There could be a lot of entries ...)
2391 : */
2392 1043300 : if (!have_fast_path_lwlock)
2393 : {
2394 358252 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
2395 358252 : have_fast_path_lwlock = true;
2396 : }
2397 :
2398 : /* Attempt fast-path release. */
2399 1043300 : relid = locallock->tag.lock.locktag_field2;
2400 1043300 : if (FastPathUnGrantRelationLock(relid, lockmode))
2401 : {
2402 1041146 : RemoveLocalLock(locallock);
2403 1041146 : continue;
2404 : }
2405 :
2406 : /*
2407 : * Our lock, originally taken via the fast path, has been
2408 : * transferred to the main lock table. That's going to require
2409 : * some extra work, so release our fast-path lock before starting.
2410 : */
2411 2154 : LWLockRelease(&MyProc->fpInfoLock);
2412 2154 : have_fast_path_lwlock = false;
2413 :
2414 : /*
2415 : * Now dump the lock. We haven't got a pointer to the LOCK or
2416 : * PROCLOCK in this case, so we have to handle this a bit
2417 : * differently than a normal lock release. Unfortunately, this
2418 : * requires an extra LWLock acquire-and-release cycle on the
2419 : * partitionLock, but hopefully it shouldn't happen often.
2420 : */
2421 2154 : LockRefindAndRelease(lockMethodTable, MyProc,
2422 : &locallock->tag.lock, lockmode, false);
2423 2154 : RemoveLocalLock(locallock);
2424 2154 : continue;
2425 : }
2426 :
2427 : /* Mark the proclock to show we need to release this lockmode */
2428 1016020 : if (locallock->nLocks > 0)
2429 1016020 : locallock->proclock->releaseMask |= LOCKBIT_ON(locallock->tag.mode);
2430 :
2431 : /* And remove the locallock hashtable entry */
2432 1016020 : RemoveLocalLock(locallock);
2433 : }
2434 :
2435 : /* Done with the fast-path data structures */
2436 1683152 : if (have_fast_path_lwlock)
2437 356098 : LWLockRelease(&MyProc->fpInfoLock);
2438 :
2439 : /*
2440 : * Now, scan each lock partition separately.
2441 : */
2442 28613584 : for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
2443 : {
2444 : LWLock *partitionLock;
2445 26930432 : dlist_head *procLocks = &MyProc->myProcLocks[partition];
2446 : dlist_mutable_iter proclock_iter;
2447 :
2448 26930432 : partitionLock = LockHashPartitionLockByIndex(partition);
2449 :
2450 : /*
2451 : * If the proclock list for this partition is empty, we can skip
2452 : * acquiring the partition lock. This optimization is trickier than
2453 : * it looks, because another backend could be in process of adding
2454 : * something to our proclock list due to promoting one of our
2455 : * fast-path locks. However, any such lock must be one that we
2456 : * decided not to delete above, so it's okay to skip it again now;
2457 : * we'd just decide not to delete it again. We must, however, be
2458 : * careful to re-fetch the list header once we've acquired the
2459 : * partition lock, to be sure we have a valid, up-to-date pointer.
2460 : * (There is probably no significant risk if pointer fetch/store is
2461 : * atomic, but we don't wish to assume that.)
2462 : *
2463 : * XXX This argument assumes that the locallock table correctly
2464 : * represents all of our fast-path locks. While allLocks mode
2465 : * guarantees to clean up all of our normal locks regardless of the
2466 : * locallock situation, we lose that guarantee for fast-path locks.
2467 : * This is not ideal.
2468 : */
2469 26930432 : if (dlist_is_empty(procLocks))
2470 25901544 : continue; /* needn't examine this partition */
2471 :
2472 1028888 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2473 :
2474 2274488 : dlist_foreach_modify(proclock_iter, procLocks)
2475 : {
2476 1245600 : PROCLOCK *proclock = dlist_container(PROCLOCK, procLink, proclock_iter.cur);
2477 1245600 : bool wakeupNeeded = false;
2478 :
2479 : Assert(proclock->tag.myProc == MyProc);
2480 :
2481 1245600 : lock = proclock->tag.myLock;
2482 :
2483 : /* Ignore items that are not of the lockmethod to be removed */
2484 1245600 : if (LOCK_LOCKMETHOD(*lock) != lockmethodid)
2485 150432 : continue;
2486 :
2487 : /*
2488 : * In allLocks mode, force release of all locks even if locallock
2489 : * table had problems
2490 : */
2491 1095168 : if (allLocks)
2492 83174 : proclock->releaseMask = proclock->holdMask;
2493 : else
2494 : Assert((proclock->releaseMask & ~proclock->holdMask) == 0);
2495 :
2496 : /*
2497 : * Ignore items that have nothing to be released, unless they have
2498 : * holdMask == 0 and are therefore recyclable
2499 : */
2500 1095168 : if (proclock->releaseMask == 0 && proclock->holdMask != 0)
2501 148426 : continue;
2502 :
2503 : PROCLOCK_PRINT("LockReleaseAll", proclock);
2504 : LOCK_PRINT("LockReleaseAll", lock, 0);
2505 : Assert(lock->nRequested >= 0);
2506 : Assert(lock->nGranted >= 0);
2507 : Assert(lock->nGranted <= lock->nRequested);
2508 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
2509 :
2510 : /*
2511 : * Release the previously-marked lock modes
2512 : */
2513 8520678 : for (i = 1; i <= numLockModes; i++)
2514 : {
2515 7573936 : if (proclock->releaseMask & LOCKBIT_ON(i))
2516 1016022 : wakeupNeeded |= UnGrantLock(lock, i, proclock,
2517 : lockMethodTable);
2518 : }
2519 : Assert((lock->nRequested >= 0) && (lock->nGranted >= 0));
2520 : Assert(lock->nGranted <= lock->nRequested);
2521 : LOCK_PRINT("LockReleaseAll: updated", lock, 0);
2522 :
2523 946742 : proclock->releaseMask = 0;
2524 :
2525 : /* CleanUpLock will wake up waiters if needed. */
2526 946742 : CleanUpLock(lock, proclock,
2527 : lockMethodTable,
2528 946742 : LockTagHashCode(&lock->tag),
2529 : wakeupNeeded);
2530 : } /* loop over PROCLOCKs within this partition */
2531 :
2532 1028888 : LWLockRelease(partitionLock);
2533 : } /* loop over partitions */
2534 :
2535 : #ifdef LOCK_DEBUG
2536 : if (*(lockMethodTable->trace_flag))
2537 : elog(LOG, "LockReleaseAll done");
2538 : #endif
2539 1683152 : }
2540 :
2541 : /*
2542 : * LockReleaseSession -- Release all session locks of the specified lock method
2543 : * that are held by the current process.
2544 : */
2545 : void
2546 238 : LockReleaseSession(LOCKMETHODID lockmethodid)
2547 : {
2548 : HASH_SEQ_STATUS status;
2549 : LOCALLOCK *locallock;
2550 :
2551 238 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2552 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2553 :
2554 238 : hash_seq_init(&status, LockMethodLocalHash);
2555 :
2556 452 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2557 : {
2558 : /* Ignore items that are not of the specified lock method */
2559 214 : if (LOCALLOCK_LOCKMETHOD(*locallock) != lockmethodid)
2560 20 : continue;
2561 :
2562 194 : ReleaseLockIfHeld(locallock, true);
2563 : }
2564 238 : }
2565 :
2566 : /*
2567 : * LockReleaseCurrentOwner
2568 : * Release all locks belonging to CurrentResourceOwner
2569 : *
2570 : * If the caller knows what those locks are, it can pass them as an array.
2571 : * That speeds up the call significantly, when a lot of locks are held.
2572 : * Otherwise, pass NULL for locallocks, and we'll traverse through our hash
2573 : * table to find them.
2574 : */
2575 : void
2576 10622 : LockReleaseCurrentOwner(LOCALLOCK **locallocks, int nlocks)
2577 : {
2578 10622 : if (locallocks == NULL)
2579 : {
2580 : HASH_SEQ_STATUS status;
2581 : LOCALLOCK *locallock;
2582 :
2583 8 : hash_seq_init(&status, LockMethodLocalHash);
2584 :
2585 544 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2586 536 : ReleaseLockIfHeld(locallock, false);
2587 : }
2588 : else
2589 : {
2590 : int i;
2591 :
2592 15704 : for (i = nlocks - 1; i >= 0; i--)
2593 5090 : ReleaseLockIfHeld(locallocks[i], false);
2594 : }
2595 10622 : }
2596 :
2597 : /*
2598 : * ReleaseLockIfHeld
2599 : * Release any session-level locks on this lockable object if sessionLock
2600 : * is true; else, release any locks held by CurrentResourceOwner.
2601 : *
2602 : * It is tempting to pass this a ResourceOwner pointer (or NULL for session
2603 : * locks), but without refactoring LockRelease() we cannot support releasing
2604 : * locks belonging to resource owners other than CurrentResourceOwner.
2605 : * If we were to refactor, it'd be a good idea to fix it so we don't have to
2606 : * do a hashtable lookup of the locallock, too. However, currently this
2607 : * function isn't used heavily enough to justify refactoring for its
2608 : * convenience.
2609 : */
2610 : static void
2611 5820 : ReleaseLockIfHeld(LOCALLOCK *locallock, bool sessionLock)
2612 : {
2613 : ResourceOwner owner;
2614 : LOCALLOCKOWNER *lockOwners;
2615 : int i;
2616 :
2617 : /* Identify owner for lock (must match LockRelease!) */
2618 5820 : if (sessionLock)
2619 194 : owner = NULL;
2620 : else
2621 5626 : owner = CurrentResourceOwner;
2622 :
2623 : /* Scan to see if there are any locks belonging to the target owner */
2624 5820 : lockOwners = locallock->lockOwners;
2625 6206 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
2626 : {
2627 5820 : if (lockOwners[i].owner == owner)
2628 : {
2629 : Assert(lockOwners[i].nLocks > 0);
2630 5434 : if (lockOwners[i].nLocks < locallock->nLocks)
2631 : {
2632 : /*
2633 : * We will still hold this lock after forgetting this
2634 : * ResourceOwner.
2635 : */
2636 1406 : locallock->nLocks -= lockOwners[i].nLocks;
2637 : /* compact out unused slot */
2638 1406 : locallock->numLockOwners--;
2639 1406 : if (owner != NULL)
2640 1406 : ResourceOwnerForgetLock(owner, locallock);
2641 1406 : if (i < locallock->numLockOwners)
2642 0 : lockOwners[i] = lockOwners[locallock->numLockOwners];
2643 : }
2644 : else
2645 : {
2646 : Assert(lockOwners[i].nLocks == locallock->nLocks);
2647 : /* We want to call LockRelease just once */
2648 4028 : lockOwners[i].nLocks = 1;
2649 4028 : locallock->nLocks = 1;
2650 4028 : if (!LockRelease(&locallock->tag.lock,
2651 : locallock->tag.mode,
2652 : sessionLock))
2653 0 : elog(WARNING, "ReleaseLockIfHeld: failed??");
2654 : }
2655 5434 : break;
2656 : }
2657 : }
2658 5820 : }
2659 :
2660 : /*
2661 : * LockReassignCurrentOwner
2662 : * Reassign all locks belonging to CurrentResourceOwner to belong
2663 : * to its parent resource owner.
2664 : *
2665 : * If the caller knows what those locks are, it can pass them as an array.
2666 : * That speeds up the call significantly, when a lot of locks are held
2667 : * (e.g pg_dump with a large schema). Otherwise, pass NULL for locallocks,
2668 : * and we'll traverse through our hash table to find them.
2669 : */
2670 : void
2671 697384 : LockReassignCurrentOwner(LOCALLOCK **locallocks, int nlocks)
2672 : {
2673 697384 : ResourceOwner parent = ResourceOwnerGetParent(CurrentResourceOwner);
2674 :
2675 : Assert(parent != NULL);
2676 :
2677 697384 : if (locallocks == NULL)
2678 : {
2679 : HASH_SEQ_STATUS status;
2680 : LOCALLOCK *locallock;
2681 :
2682 6900 : hash_seq_init(&status, LockMethodLocalHash);
2683 :
2684 208206 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2685 201306 : LockReassignOwner(locallock, parent);
2686 : }
2687 : else
2688 : {
2689 : int i;
2690 :
2691 1551204 : for (i = nlocks - 1; i >= 0; i--)
2692 860720 : LockReassignOwner(locallocks[i], parent);
2693 : }
2694 697384 : }
2695 :
2696 : /*
2697 : * Subroutine of LockReassignCurrentOwner. Reassigns a given lock belonging to
2698 : * CurrentResourceOwner to its parent.
2699 : */
2700 : static void
2701 1062026 : LockReassignOwner(LOCALLOCK *locallock, ResourceOwner parent)
2702 : {
2703 : LOCALLOCKOWNER *lockOwners;
2704 : int i;
2705 1062026 : int ic = -1;
2706 1062026 : int ip = -1;
2707 :
2708 : /*
2709 : * Scan to see if there are any locks belonging to current owner or its
2710 : * parent
2711 : */
2712 1062026 : lockOwners = locallock->lockOwners;
2713 2547572 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
2714 : {
2715 1485546 : if (lockOwners[i].owner == CurrentResourceOwner)
2716 1027072 : ic = i;
2717 458474 : else if (lockOwners[i].owner == parent)
2718 371646 : ip = i;
2719 : }
2720 :
2721 1062026 : if (ic < 0)
2722 34954 : return; /* no current locks */
2723 :
2724 1027072 : if (ip < 0)
2725 : {
2726 : /* Parent has no slot, so just give it the child's slot */
2727 690320 : lockOwners[ic].owner = parent;
2728 690320 : ResourceOwnerRememberLock(parent, locallock);
2729 : }
2730 : else
2731 : {
2732 : /* Merge child's count with parent's */
2733 336752 : lockOwners[ip].nLocks += lockOwners[ic].nLocks;
2734 : /* compact out unused slot */
2735 336752 : locallock->numLockOwners--;
2736 336752 : if (ic < locallock->numLockOwners)
2737 1644 : lockOwners[ic] = lockOwners[locallock->numLockOwners];
2738 : }
2739 1027072 : ResourceOwnerForgetLock(CurrentResourceOwner, locallock);
2740 : }
2741 :
2742 : /*
2743 : * FastPathGrantRelationLock
2744 : * Grant lock using per-backend fast-path array, if there is space.
2745 : */
2746 : static bool
2747 29449648 : FastPathGrantRelationLock(Oid relid, LOCKMODE lockmode)
2748 : {
2749 : uint32 i;
2750 29449648 : uint32 unused_slot = FastPathLockSlotsPerBackend();
2751 :
2752 : /* fast-path group the lock belongs to */
2753 29449648 : uint32 group = FAST_PATH_REL_GROUP(relid);
2754 :
2755 : /* Scan for existing entry for this relid, remembering empty slot. */
2756 499514108 : for (i = 0; i < FP_LOCK_SLOTS_PER_GROUP; i++)
2757 : {
2758 : /* index into the whole per-backend array */
2759 470855924 : uint32 f = FAST_PATH_SLOT(group, i);
2760 :
2761 470855924 : if (FAST_PATH_GET_BITS(MyProc, f) == 0)
2762 454831278 : unused_slot = f;
2763 16024646 : else if (MyProc->fpRelId[f] == relid)
2764 : {
2765 : Assert(!FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode));
2766 791464 : FAST_PATH_SET_LOCKMODE(MyProc, f, lockmode);
2767 791464 : return true;
2768 : }
2769 : }
2770 :
2771 : /* If no existing entry, use any empty slot. */
2772 28658184 : if (unused_slot < FastPathLockSlotsPerBackend())
2773 : {
2774 28658184 : MyProc->fpRelId[unused_slot] = relid;
2775 28658184 : FAST_PATH_SET_LOCKMODE(MyProc, unused_slot, lockmode);
2776 28658184 : ++FastPathLocalUseCounts[group];
2777 28658184 : return true;
2778 : }
2779 :
2780 : /* No existing entry, and no empty slot. */
2781 0 : return false;
2782 : }
2783 :
2784 : /*
2785 : * FastPathUnGrantRelationLock
2786 : * Release fast-path lock, if present. Update backend-private local
2787 : * use count, while we're at it.
2788 : */
2789 : static bool
2790 29823766 : FastPathUnGrantRelationLock(Oid relid, LOCKMODE lockmode)
2791 : {
2792 : uint32 i;
2793 29823766 : bool result = false;
2794 :
2795 : /* fast-path group the lock belongs to */
2796 29823766 : uint32 group = FAST_PATH_REL_GROUP(relid);
2797 :
2798 29823766 : FastPathLocalUseCounts[group] = 0;
2799 507004022 : for (i = 0; i < FP_LOCK_SLOTS_PER_GROUP; i++)
2800 : {
2801 : /* index into the whole per-backend array */
2802 477180256 : uint32 f = FAST_PATH_SLOT(group, i);
2803 :
2804 477180256 : if (MyProc->fpRelId[f] == relid
2805 40729870 : && FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode))
2806 : {
2807 : Assert(!result);
2808 29446866 : FAST_PATH_CLEAR_LOCKMODE(MyProc, f, lockmode);
2809 29446866 : result = true;
2810 : /* we continue iterating so as to update FastPathLocalUseCount */
2811 : }
2812 477180256 : if (FAST_PATH_GET_BITS(MyProc, f) != 0)
2813 21503626 : ++FastPathLocalUseCounts[group];
2814 : }
2815 29823766 : return result;
2816 : }
2817 :
2818 : /*
2819 : * FastPathTransferRelationLocks
2820 : * Transfer locks matching the given lock tag from per-backend fast-path
2821 : * arrays to the shared hash table.
2822 : *
2823 : * Returns true if successful, false if ran out of shared memory.
2824 : */
2825 : static bool
2826 359710 : FastPathTransferRelationLocks(LockMethod lockMethodTable, const LOCKTAG *locktag,
2827 : uint32 hashcode)
2828 : {
2829 359710 : LWLock *partitionLock = LockHashPartitionLock(hashcode);
2830 359710 : Oid relid = locktag->locktag_field2;
2831 : uint32 i;
2832 :
2833 : /* fast-path group the lock belongs to */
2834 359710 : uint32 group = FAST_PATH_REL_GROUP(relid);
2835 :
2836 : /*
2837 : * Every PGPROC that can potentially hold a fast-path lock is present in
2838 : * ProcGlobal->allProcs. Prepared transactions are not, but any
2839 : * outstanding fast-path locks held by prepared transactions are
2840 : * transferred to the main lock table.
2841 : */
2842 52344292 : for (i = 0; i < ProcGlobal->allProcCount; i++)
2843 : {
2844 51984582 : PGPROC *proc = &ProcGlobal->allProcs[i];
2845 : uint32 j;
2846 :
2847 51984582 : LWLockAcquire(&proc->fpInfoLock, LW_EXCLUSIVE);
2848 :
2849 : /*
2850 : * If the target backend isn't referencing the same database as the
2851 : * lock, then we needn't examine the individual relation IDs at all;
2852 : * none of them can be relevant.
2853 : *
2854 : * proc->databaseId is set at backend startup time and never changes
2855 : * thereafter, so it might be safe to perform this test before
2856 : * acquiring &proc->fpInfoLock. In particular, it's certainly safe to
2857 : * assume that if the target backend holds any fast-path locks, it
2858 : * must have performed a memory-fencing operation (in particular, an
2859 : * LWLock acquisition) since setting proc->databaseId. However, it's
2860 : * less clear that our backend is certain to have performed a memory
2861 : * fencing operation since the other backend set proc->databaseId. So
2862 : * for now, we test it after acquiring the LWLock just to be safe.
2863 : *
2864 : * Also skip groups without any registered fast-path locks.
2865 : */
2866 51984582 : if (proc->databaseId != locktag->locktag_field1 ||
2867 18333348 : proc->fpLockBits[group] == 0)
2868 : {
2869 51663744 : LWLockRelease(&proc->fpInfoLock);
2870 51663744 : continue;
2871 : }
2872 :
2873 5451914 : for (j = 0; j < FP_LOCK_SLOTS_PER_GROUP; j++)
2874 : {
2875 : uint32 lockmode;
2876 :
2877 : /* index into the whole per-backend array */
2878 5133146 : uint32 f = FAST_PATH_SLOT(group, j);
2879 :
2880 : /* Look for an allocated slot matching the given relid. */
2881 5133146 : if (relid != proc->fpRelId[f] || FAST_PATH_GET_BITS(proc, f) == 0)
2882 5131076 : continue;
2883 :
2884 : /* Find or create lock object. */
2885 2070 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2886 8280 : for (lockmode = FAST_PATH_LOCKNUMBER_OFFSET;
2887 : lockmode < FAST_PATH_LOCKNUMBER_OFFSET + FAST_PATH_BITS_PER_SLOT;
2888 6210 : ++lockmode)
2889 : {
2890 : PROCLOCK *proclock;
2891 :
2892 6210 : if (!FAST_PATH_CHECK_LOCKMODE(proc, f, lockmode))
2893 4030 : continue;
2894 2180 : proclock = SetupLockInTable(lockMethodTable, proc, locktag,
2895 : hashcode, lockmode);
2896 2180 : if (!proclock)
2897 : {
2898 0 : LWLockRelease(partitionLock);
2899 0 : LWLockRelease(&proc->fpInfoLock);
2900 0 : return false;
2901 : }
2902 2180 : GrantLock(proclock->tag.myLock, proclock, lockmode);
2903 2180 : FAST_PATH_CLEAR_LOCKMODE(proc, f, lockmode);
2904 : }
2905 2070 : LWLockRelease(partitionLock);
2906 :
2907 : /* No need to examine remaining slots. */
2908 2070 : break;
2909 : }
2910 320838 : LWLockRelease(&proc->fpInfoLock);
2911 : }
2912 359710 : return true;
2913 : }
2914 :
2915 : /*
2916 : * FastPathGetRelationLockEntry
2917 : * Return the PROCLOCK for a lock originally taken via the fast-path,
2918 : * transferring it to the primary lock table if necessary.
2919 : *
2920 : * Note: caller takes care of updating the locallock object.
2921 : */
2922 : static PROCLOCK *
2923 618 : FastPathGetRelationLockEntry(LOCALLOCK *locallock)
2924 : {
2925 618 : LockMethod lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
2926 618 : LOCKTAG *locktag = &locallock->tag.lock;
2927 618 : PROCLOCK *proclock = NULL;
2928 618 : LWLock *partitionLock = LockHashPartitionLock(locallock->hashcode);
2929 618 : Oid relid = locktag->locktag_field2;
2930 : uint32 i,
2931 : group;
2932 :
2933 : /* fast-path group the lock belongs to */
2934 618 : group = FAST_PATH_REL_GROUP(relid);
2935 :
2936 618 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
2937 :
2938 9892 : for (i = 0; i < FP_LOCK_SLOTS_PER_GROUP; i++)
2939 : {
2940 : uint32 lockmode;
2941 :
2942 : /* index into the whole per-backend array */
2943 9876 : uint32 f = FAST_PATH_SLOT(group, i);
2944 :
2945 : /* Look for an allocated slot matching the given relid. */
2946 9876 : if (relid != MyProc->fpRelId[f] || FAST_PATH_GET_BITS(MyProc, f) == 0)
2947 9274 : continue;
2948 :
2949 : /* If we don't have a lock of the given mode, forget it! */
2950 602 : lockmode = locallock->tag.mode;
2951 602 : if (!FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode))
2952 0 : break;
2953 :
2954 : /* Find or create lock object. */
2955 602 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2956 :
2957 602 : proclock = SetupLockInTable(lockMethodTable, MyProc, locktag,
2958 : locallock->hashcode, lockmode);
2959 602 : if (!proclock)
2960 : {
2961 0 : LWLockRelease(partitionLock);
2962 0 : LWLockRelease(&MyProc->fpInfoLock);
2963 0 : ereport(ERROR,
2964 : (errcode(ERRCODE_OUT_OF_MEMORY),
2965 : errmsg("out of shared memory"),
2966 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
2967 : }
2968 602 : GrantLock(proclock->tag.myLock, proclock, lockmode);
2969 602 : FAST_PATH_CLEAR_LOCKMODE(MyProc, f, lockmode);
2970 :
2971 602 : LWLockRelease(partitionLock);
2972 :
2973 : /* No need to examine remaining slots. */
2974 602 : break;
2975 : }
2976 :
2977 618 : LWLockRelease(&MyProc->fpInfoLock);
2978 :
2979 : /* Lock may have already been transferred by some other backend. */
2980 618 : if (proclock == NULL)
2981 : {
2982 : LOCK *lock;
2983 : PROCLOCKTAG proclocktag;
2984 : uint32 proclock_hashcode;
2985 :
2986 16 : LWLockAcquire(partitionLock, LW_SHARED);
2987 :
2988 16 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
2989 : locktag,
2990 : locallock->hashcode,
2991 : HASH_FIND,
2992 : NULL);
2993 16 : if (!lock)
2994 0 : elog(ERROR, "failed to re-find shared lock object");
2995 :
2996 16 : proclocktag.myLock = lock;
2997 16 : proclocktag.myProc = MyProc;
2998 :
2999 16 : proclock_hashcode = ProcLockHashCode(&proclocktag, locallock->hashcode);
3000 : proclock = (PROCLOCK *)
3001 16 : hash_search_with_hash_value(LockMethodProcLockHash,
3002 : &proclocktag,
3003 : proclock_hashcode,
3004 : HASH_FIND,
3005 : NULL);
3006 16 : if (!proclock)
3007 0 : elog(ERROR, "failed to re-find shared proclock object");
3008 16 : LWLockRelease(partitionLock);
3009 : }
3010 :
3011 618 : return proclock;
3012 : }
3013 :
3014 : /*
3015 : * GetLockConflicts
3016 : * Get an array of VirtualTransactionIds of xacts currently holding locks
3017 : * that would conflict with the specified lock/lockmode.
3018 : * xacts merely awaiting such a lock are NOT reported.
3019 : *
3020 : * The result array is palloc'd and is terminated with an invalid VXID.
3021 : * *countp, if not null, is updated to the number of items set.
3022 : *
3023 : * Of course, the result could be out of date by the time it's returned, so
3024 : * use of this function has to be thought about carefully. Similarly, a
3025 : * PGPROC with no "lxid" will be considered non-conflicting regardless of any
3026 : * lock it holds. Existing callers don't care about a locker after that
3027 : * locker's pg_xact updates complete. CommitTransaction() clears "lxid" after
3028 : * pg_xact updates and before releasing locks.
3029 : *
3030 : * Note we never include the current xact's vxid in the result array,
3031 : * since an xact never blocks itself.
3032 : */
3033 : VirtualTransactionId *
3034 2700 : GetLockConflicts(const LOCKTAG *locktag, LOCKMODE lockmode, int *countp)
3035 : {
3036 : static VirtualTransactionId *vxids;
3037 2700 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
3038 : LockMethod lockMethodTable;
3039 : LOCK *lock;
3040 : LOCKMASK conflictMask;
3041 : dlist_iter proclock_iter;
3042 : PROCLOCK *proclock;
3043 : uint32 hashcode;
3044 : LWLock *partitionLock;
3045 2700 : int count = 0;
3046 2700 : int fast_count = 0;
3047 :
3048 2700 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
3049 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
3050 2700 : lockMethodTable = LockMethods[lockmethodid];
3051 2700 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
3052 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
3053 :
3054 : /*
3055 : * Allocate memory to store results, and fill with InvalidVXID. We only
3056 : * need enough space for MaxBackends + max_prepared_xacts + a terminator.
3057 : * InHotStandby allocate once in TopMemoryContext.
3058 : */
3059 2700 : if (InHotStandby)
3060 : {
3061 8 : if (vxids == NULL)
3062 2 : vxids = (VirtualTransactionId *)
3063 2 : MemoryContextAlloc(TopMemoryContext,
3064 : sizeof(VirtualTransactionId) *
3065 2 : (MaxBackends + max_prepared_xacts + 1));
3066 : }
3067 : else
3068 2692 : vxids = (VirtualTransactionId *)
3069 2692 : palloc0(sizeof(VirtualTransactionId) *
3070 2692 : (MaxBackends + max_prepared_xacts + 1));
3071 :
3072 : /* Compute hash code and partition lock, and look up conflicting modes. */
3073 2700 : hashcode = LockTagHashCode(locktag);
3074 2700 : partitionLock = LockHashPartitionLock(hashcode);
3075 2700 : conflictMask = lockMethodTable->conflictTab[lockmode];
3076 :
3077 : /*
3078 : * Fast path locks might not have been entered in the primary lock table.
3079 : * If the lock we're dealing with could conflict with such a lock, we must
3080 : * examine each backend's fast-path array for conflicts.
3081 : */
3082 2700 : if (ConflictsWithRelationFastPath(locktag, lockmode))
3083 : {
3084 : int i;
3085 2700 : Oid relid = locktag->locktag_field2;
3086 : VirtualTransactionId vxid;
3087 :
3088 : /* fast-path group the lock belongs to */
3089 2700 : uint32 group = FAST_PATH_REL_GROUP(relid);
3090 :
3091 : /*
3092 : * Iterate over relevant PGPROCs. Anything held by a prepared
3093 : * transaction will have been transferred to the primary lock table,
3094 : * so we need not worry about those. This is all a bit fuzzy, because
3095 : * new locks could be taken after we've visited a particular
3096 : * partition, but the callers had better be prepared to deal with that
3097 : * anyway, since the locks could equally well be taken between the
3098 : * time we return the value and the time the caller does something
3099 : * with it.
3100 : */
3101 422772 : for (i = 0; i < ProcGlobal->allProcCount; i++)
3102 : {
3103 420072 : PGPROC *proc = &ProcGlobal->allProcs[i];
3104 : uint32 j;
3105 :
3106 : /* A backend never blocks itself */
3107 420072 : if (proc == MyProc)
3108 2700 : continue;
3109 :
3110 417372 : LWLockAcquire(&proc->fpInfoLock, LW_SHARED);
3111 :
3112 : /*
3113 : * If the target backend isn't referencing the same database as
3114 : * the lock, then we needn't examine the individual relation IDs
3115 : * at all; none of them can be relevant.
3116 : *
3117 : * See FastPathTransferRelationLocks() for discussion of why we do
3118 : * this test after acquiring the lock.
3119 : *
3120 : * Also skip groups without any registered fast-path locks.
3121 : */
3122 417372 : if (proc->databaseId != locktag->locktag_field1 ||
3123 173430 : proc->fpLockBits[group] == 0)
3124 : {
3125 416616 : LWLockRelease(&proc->fpInfoLock);
3126 416616 : continue;
3127 : }
3128 :
3129 12434 : for (j = 0; j < FP_LOCK_SLOTS_PER_GROUP; j++)
3130 : {
3131 : uint32 lockmask;
3132 :
3133 : /* index into the whole per-backend array */
3134 12072 : uint32 f = FAST_PATH_SLOT(group, j);
3135 :
3136 : /* Look for an allocated slot matching the given relid. */
3137 12072 : if (relid != proc->fpRelId[f])
3138 11678 : continue;
3139 394 : lockmask = FAST_PATH_GET_BITS(proc, f);
3140 394 : if (!lockmask)
3141 0 : continue;
3142 394 : lockmask <<= FAST_PATH_LOCKNUMBER_OFFSET;
3143 :
3144 : /*
3145 : * There can only be one entry per relation, so if we found it
3146 : * and it doesn't conflict, we can skip the rest of the slots.
3147 : */
3148 394 : if ((lockmask & conflictMask) == 0)
3149 10 : break;
3150 :
3151 : /* Conflict! */
3152 384 : GET_VXID_FROM_PGPROC(vxid, *proc);
3153 :
3154 384 : if (VirtualTransactionIdIsValid(vxid))
3155 384 : vxids[count++] = vxid;
3156 : /* else, xact already committed or aborted */
3157 :
3158 : /* No need to examine remaining slots. */
3159 384 : break;
3160 : }
3161 :
3162 756 : LWLockRelease(&proc->fpInfoLock);
3163 : }
3164 : }
3165 :
3166 : /* Remember how many fast-path conflicts we found. */
3167 2700 : fast_count = count;
3168 :
3169 : /*
3170 : * Look up the lock object matching the tag.
3171 : */
3172 2700 : LWLockAcquire(partitionLock, LW_SHARED);
3173 :
3174 2700 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
3175 : locktag,
3176 : hashcode,
3177 : HASH_FIND,
3178 : NULL);
3179 2700 : if (!lock)
3180 : {
3181 : /*
3182 : * If the lock object doesn't exist, there is nothing holding a lock
3183 : * on this lockable object.
3184 : */
3185 140 : LWLockRelease(partitionLock);
3186 140 : vxids[count].procNumber = INVALID_PROC_NUMBER;
3187 140 : vxids[count].localTransactionId = InvalidLocalTransactionId;
3188 140 : if (countp)
3189 0 : *countp = count;
3190 140 : return vxids;
3191 : }
3192 :
3193 : /*
3194 : * Examine each existing holder (or awaiter) of the lock.
3195 : */
3196 5162 : dlist_foreach(proclock_iter, &lock->procLocks)
3197 : {
3198 2602 : proclock = dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
3199 :
3200 2602 : if (conflictMask & proclock->holdMask)
3201 : {
3202 2594 : PGPROC *proc = proclock->tag.myProc;
3203 :
3204 : /* A backend never blocks itself */
3205 2594 : if (proc != MyProc)
3206 : {
3207 : VirtualTransactionId vxid;
3208 :
3209 42 : GET_VXID_FROM_PGPROC(vxid, *proc);
3210 :
3211 42 : if (VirtualTransactionIdIsValid(vxid))
3212 : {
3213 : int i;
3214 :
3215 : /* Avoid duplicate entries. */
3216 72 : for (i = 0; i < fast_count; ++i)
3217 32 : if (VirtualTransactionIdEquals(vxids[i], vxid))
3218 2 : break;
3219 42 : if (i >= fast_count)
3220 40 : vxids[count++] = vxid;
3221 : }
3222 : /* else, xact already committed or aborted */
3223 : }
3224 : }
3225 : }
3226 :
3227 2560 : LWLockRelease(partitionLock);
3228 :
3229 2560 : if (count > MaxBackends + max_prepared_xacts) /* should never happen */
3230 0 : elog(PANIC, "too many conflicting locks found");
3231 :
3232 2560 : vxids[count].procNumber = INVALID_PROC_NUMBER;
3233 2560 : vxids[count].localTransactionId = InvalidLocalTransactionId;
3234 2560 : if (countp)
3235 2554 : *countp = count;
3236 2560 : return vxids;
3237 : }
3238 :
3239 : /*
3240 : * Find a lock in the shared lock table and release it. It is the caller's
3241 : * responsibility to verify that this is a sane thing to do. (For example, it
3242 : * would be bad to release a lock here if there might still be a LOCALLOCK
3243 : * object with pointers to it.)
3244 : *
3245 : * We currently use this in two situations: first, to release locks held by
3246 : * prepared transactions on commit (see lock_twophase_postcommit); and second,
3247 : * to release locks taken via the fast-path, transferred to the main hash
3248 : * table, and then released (see LockReleaseAll).
3249 : */
3250 : static void
3251 4158 : LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
3252 : LOCKTAG *locktag, LOCKMODE lockmode,
3253 : bool decrement_strong_lock_count)
3254 : {
3255 : LOCK *lock;
3256 : PROCLOCK *proclock;
3257 : PROCLOCKTAG proclocktag;
3258 : uint32 hashcode;
3259 : uint32 proclock_hashcode;
3260 : LWLock *partitionLock;
3261 : bool wakeupNeeded;
3262 :
3263 4158 : hashcode = LockTagHashCode(locktag);
3264 4158 : partitionLock = LockHashPartitionLock(hashcode);
3265 :
3266 4158 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3267 :
3268 : /*
3269 : * Re-find the lock object (it had better be there).
3270 : */
3271 4158 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
3272 : locktag,
3273 : hashcode,
3274 : HASH_FIND,
3275 : NULL);
3276 4158 : if (!lock)
3277 0 : elog(PANIC, "failed to re-find shared lock object");
3278 :
3279 : /*
3280 : * Re-find the proclock object (ditto).
3281 : */
3282 4158 : proclocktag.myLock = lock;
3283 4158 : proclocktag.myProc = proc;
3284 :
3285 4158 : proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
3286 :
3287 4158 : proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
3288 : &proclocktag,
3289 : proclock_hashcode,
3290 : HASH_FIND,
3291 : NULL);
3292 4158 : if (!proclock)
3293 0 : elog(PANIC, "failed to re-find shared proclock object");
3294 :
3295 : /*
3296 : * Double-check that we are actually holding a lock of the type we want to
3297 : * release.
3298 : */
3299 4158 : if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
3300 : {
3301 : PROCLOCK_PRINT("lock_twophase_postcommit: WRONGTYPE", proclock);
3302 0 : LWLockRelease(partitionLock);
3303 0 : elog(WARNING, "you don't own a lock of type %s",
3304 : lockMethodTable->lockModeNames[lockmode]);
3305 0 : return;
3306 : }
3307 :
3308 : /*
3309 : * Do the releasing. CleanUpLock will waken any now-wakable waiters.
3310 : */
3311 4158 : wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
3312 :
3313 4158 : CleanUpLock(lock, proclock,
3314 : lockMethodTable, hashcode,
3315 : wakeupNeeded);
3316 :
3317 4158 : LWLockRelease(partitionLock);
3318 :
3319 : /*
3320 : * Decrement strong lock count. This logic is needed only for 2PC.
3321 : */
3322 4158 : if (decrement_strong_lock_count
3323 1510 : && ConflictsWithRelationFastPath(locktag, lockmode))
3324 : {
3325 162 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
3326 :
3327 162 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
3328 : Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
3329 162 : FastPathStrongRelationLocks->count[fasthashcode]--;
3330 162 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
3331 : }
3332 : }
3333 :
3334 : /*
3335 : * CheckForSessionAndXactLocks
3336 : * Check to see if transaction holds both session-level and xact-level
3337 : * locks on the same object; if so, throw an error.
3338 : *
3339 : * If we have both session- and transaction-level locks on the same object,
3340 : * PREPARE TRANSACTION must fail. This should never happen with regular
3341 : * locks, since we only take those at session level in some special operations
3342 : * like VACUUM. It's possible to hit this with advisory locks, though.
3343 : *
3344 : * It would be nice if we could keep the session hold and give away the
3345 : * transactional hold to the prepared xact. However, that would require two
3346 : * PROCLOCK objects, and we cannot be sure that another PROCLOCK will be
3347 : * available when it comes time for PostPrepare_Locks to do the deed.
3348 : * So for now, we error out while we can still do so safely.
3349 : *
3350 : * Since the LOCALLOCK table stores a separate entry for each lockmode,
3351 : * we can't implement this check by examining LOCALLOCK entries in isolation.
3352 : * We must build a transient hashtable that is indexed by locktag only.
3353 : */
3354 : static void
3355 604 : CheckForSessionAndXactLocks(void)
3356 : {
3357 : typedef struct
3358 : {
3359 : LOCKTAG lock; /* identifies the lockable object */
3360 : bool sessLock; /* is any lockmode held at session level? */
3361 : bool xactLock; /* is any lockmode held at xact level? */
3362 : } PerLockTagEntry;
3363 :
3364 : HASHCTL hash_ctl;
3365 : HTAB *lockhtab;
3366 : HASH_SEQ_STATUS status;
3367 : LOCALLOCK *locallock;
3368 :
3369 : /* Create a local hash table keyed by LOCKTAG only */
3370 604 : hash_ctl.keysize = sizeof(LOCKTAG);
3371 604 : hash_ctl.entrysize = sizeof(PerLockTagEntry);
3372 604 : hash_ctl.hcxt = CurrentMemoryContext;
3373 :
3374 604 : lockhtab = hash_create("CheckForSessionAndXactLocks table",
3375 : 256, /* arbitrary initial size */
3376 : &hash_ctl,
3377 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
3378 :
3379 : /* Scan local lock table to find entries for each LOCKTAG */
3380 604 : hash_seq_init(&status, LockMethodLocalHash);
3381 :
3382 2084 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3383 : {
3384 1484 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3385 : PerLockTagEntry *hentry;
3386 : bool found;
3387 : int i;
3388 :
3389 : /*
3390 : * Ignore VXID locks. We don't want those to be held by prepared
3391 : * transactions, since they aren't meaningful after a restart.
3392 : */
3393 1484 : if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3394 0 : continue;
3395 :
3396 : /* Ignore it if we don't actually hold the lock */
3397 1484 : if (locallock->nLocks <= 0)
3398 0 : continue;
3399 :
3400 : /* Otherwise, find or make an entry in lockhtab */
3401 1484 : hentry = (PerLockTagEntry *) hash_search(lockhtab,
3402 1484 : &locallock->tag.lock,
3403 : HASH_ENTER, &found);
3404 1484 : if (!found) /* initialize, if newly created */
3405 1366 : hentry->sessLock = hentry->xactLock = false;
3406 :
3407 : /* Scan to see if we hold lock at session or xact level or both */
3408 2968 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
3409 : {
3410 1484 : if (lockOwners[i].owner == NULL)
3411 18 : hentry->sessLock = true;
3412 : else
3413 1466 : hentry->xactLock = true;
3414 : }
3415 :
3416 : /*
3417 : * We can throw error immediately when we see both types of locks; no
3418 : * need to wait around to see if there are more violations.
3419 : */
3420 1484 : if (hentry->sessLock && hentry->xactLock)
3421 4 : ereport(ERROR,
3422 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3423 : errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3424 : }
3425 :
3426 : /* Success, so clean up */
3427 600 : hash_destroy(lockhtab);
3428 600 : }
3429 :
3430 : /*
3431 : * AtPrepare_Locks
3432 : * Do the preparatory work for a PREPARE: make 2PC state file records
3433 : * for all locks currently held.
3434 : *
3435 : * Session-level locks are ignored, as are VXID locks.
3436 : *
3437 : * For the most part, we don't need to touch shared memory for this ---
3438 : * all the necessary state information is in the locallock table.
3439 : * Fast-path locks are an exception, however: we move any such locks to
3440 : * the main table before allowing PREPARE TRANSACTION to succeed.
3441 : */
3442 : void
3443 604 : AtPrepare_Locks(void)
3444 : {
3445 : HASH_SEQ_STATUS status;
3446 : LOCALLOCK *locallock;
3447 :
3448 : /* First, verify there aren't locks of both xact and session level */
3449 604 : CheckForSessionAndXactLocks();
3450 :
3451 : /* Now do the per-locallock cleanup work */
3452 600 : hash_seq_init(&status, LockMethodLocalHash);
3453 :
3454 2074 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3455 : {
3456 : TwoPhaseLockRecord record;
3457 1474 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3458 : bool haveSessionLock;
3459 : bool haveXactLock;
3460 : int i;
3461 :
3462 : /*
3463 : * Ignore VXID locks. We don't want those to be held by prepared
3464 : * transactions, since they aren't meaningful after a restart.
3465 : */
3466 1474 : if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3467 14 : continue;
3468 :
3469 : /* Ignore it if we don't actually hold the lock */
3470 1474 : if (locallock->nLocks <= 0)
3471 0 : continue;
3472 :
3473 : /* Scan to see whether we hold it at session or transaction level */
3474 1474 : haveSessionLock = haveXactLock = false;
3475 2948 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
3476 : {
3477 1474 : if (lockOwners[i].owner == NULL)
3478 14 : haveSessionLock = true;
3479 : else
3480 1460 : haveXactLock = true;
3481 : }
3482 :
3483 : /* Ignore it if we have only session lock */
3484 1474 : if (!haveXactLock)
3485 14 : continue;
3486 :
3487 : /* This can't happen, because we already checked it */
3488 1460 : if (haveSessionLock)
3489 0 : ereport(ERROR,
3490 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3491 : errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3492 :
3493 : /*
3494 : * If the local lock was taken via the fast-path, we need to move it
3495 : * to the primary lock table, or just get a pointer to the existing
3496 : * primary lock table entry if by chance it's already been
3497 : * transferred.
3498 : */
3499 1460 : if (locallock->proclock == NULL)
3500 : {
3501 618 : locallock->proclock = FastPathGetRelationLockEntry(locallock);
3502 618 : locallock->lock = locallock->proclock->tag.myLock;
3503 : }
3504 :
3505 : /*
3506 : * Arrange to not release any strong lock count held by this lock
3507 : * entry. We must retain the count until the prepared transaction is
3508 : * committed or rolled back.
3509 : */
3510 1460 : locallock->holdsStrongLockCount = false;
3511 :
3512 : /*
3513 : * Create a 2PC record.
3514 : */
3515 1460 : memcpy(&(record.locktag), &(locallock->tag.lock), sizeof(LOCKTAG));
3516 1460 : record.lockmode = locallock->tag.mode;
3517 :
3518 1460 : RegisterTwoPhaseRecord(TWOPHASE_RM_LOCK_ID, 0,
3519 : &record, sizeof(TwoPhaseLockRecord));
3520 : }
3521 600 : }
3522 :
3523 : /*
3524 : * PostPrepare_Locks
3525 : * Clean up after successful PREPARE
3526 : *
3527 : * Here, we want to transfer ownership of our locks to a dummy PGPROC
3528 : * that's now associated with the prepared transaction, and we want to
3529 : * clean out the corresponding entries in the LOCALLOCK table.
3530 : *
3531 : * Note: by removing the LOCALLOCK entries, we are leaving dangling
3532 : * pointers in the transaction's resource owner. This is OK at the
3533 : * moment since resowner.c doesn't try to free locks retail at a toplevel
3534 : * transaction commit or abort. We could alternatively zero out nLocks
3535 : * and leave the LOCALLOCK entries to be garbage-collected by LockReleaseAll,
3536 : * but that probably costs more cycles.
3537 : */
3538 : void
3539 600 : PostPrepare_Locks(TransactionId xid)
3540 : {
3541 600 : PGPROC *newproc = TwoPhaseGetDummyProc(xid, false);
3542 : HASH_SEQ_STATUS status;
3543 : LOCALLOCK *locallock;
3544 : LOCK *lock;
3545 : PROCLOCK *proclock;
3546 : PROCLOCKTAG proclocktag;
3547 : int partition;
3548 :
3549 : /* Can't prepare a lock group follower. */
3550 : Assert(MyProc->lockGroupLeader == NULL ||
3551 : MyProc->lockGroupLeader == MyProc);
3552 :
3553 : /* This is a critical section: any error means big trouble */
3554 600 : START_CRIT_SECTION();
3555 :
3556 : /*
3557 : * First we run through the locallock table and get rid of unwanted
3558 : * entries, then we scan the process's proclocks and transfer them to the
3559 : * target proc.
3560 : *
3561 : * We do this separately because we may have multiple locallock entries
3562 : * pointing to the same proclock, and we daren't end up with any dangling
3563 : * pointers.
3564 : */
3565 600 : hash_seq_init(&status, LockMethodLocalHash);
3566 :
3567 2074 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3568 : {
3569 1474 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3570 : bool haveSessionLock;
3571 : bool haveXactLock;
3572 : int i;
3573 :
3574 1474 : if (locallock->proclock == NULL || locallock->lock == NULL)
3575 : {
3576 : /*
3577 : * We must've run out of shared memory while trying to set up this
3578 : * lock. Just forget the local entry.
3579 : */
3580 : Assert(locallock->nLocks == 0);
3581 0 : RemoveLocalLock(locallock);
3582 0 : continue;
3583 : }
3584 :
3585 : /* Ignore VXID locks */
3586 1474 : if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3587 0 : continue;
3588 :
3589 : /* Scan to see whether we hold it at session or transaction level */
3590 1474 : haveSessionLock = haveXactLock = false;
3591 2948 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
3592 : {
3593 1474 : if (lockOwners[i].owner == NULL)
3594 14 : haveSessionLock = true;
3595 : else
3596 1460 : haveXactLock = true;
3597 : }
3598 :
3599 : /* Ignore it if we have only session lock */
3600 1474 : if (!haveXactLock)
3601 14 : continue;
3602 :
3603 : /* This can't happen, because we already checked it */
3604 1460 : if (haveSessionLock)
3605 0 : ereport(PANIC,
3606 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3607 : errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3608 :
3609 : /* Mark the proclock to show we need to release this lockmode */
3610 1460 : if (locallock->nLocks > 0)
3611 1460 : locallock->proclock->releaseMask |= LOCKBIT_ON(locallock->tag.mode);
3612 :
3613 : /* And remove the locallock hashtable entry */
3614 1460 : RemoveLocalLock(locallock);
3615 : }
3616 :
3617 : /*
3618 : * Now, scan each lock partition separately.
3619 : */
3620 10200 : for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
3621 : {
3622 : LWLock *partitionLock;
3623 9600 : dlist_head *procLocks = &(MyProc->myProcLocks[partition]);
3624 : dlist_mutable_iter proclock_iter;
3625 :
3626 9600 : partitionLock = LockHashPartitionLockByIndex(partition);
3627 :
3628 : /*
3629 : * If the proclock list for this partition is empty, we can skip
3630 : * acquiring the partition lock. This optimization is safer than the
3631 : * situation in LockReleaseAll, because we got rid of any fast-path
3632 : * locks during AtPrepare_Locks, so there cannot be any case where
3633 : * another backend is adding something to our lists now. For safety,
3634 : * though, we code this the same way as in LockReleaseAll.
3635 : */
3636 9600 : if (dlist_is_empty(procLocks))
3637 8228 : continue; /* needn't examine this partition */
3638 :
3639 1372 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3640 :
3641 2792 : dlist_foreach_modify(proclock_iter, procLocks)
3642 : {
3643 1420 : proclock = dlist_container(PROCLOCK, procLink, proclock_iter.cur);
3644 :
3645 : Assert(proclock->tag.myProc == MyProc);
3646 :
3647 1420 : lock = proclock->tag.myLock;
3648 :
3649 : /* Ignore VXID locks */
3650 1420 : if (lock->tag.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3651 60 : continue;
3652 :
3653 : PROCLOCK_PRINT("PostPrepare_Locks", proclock);
3654 : LOCK_PRINT("PostPrepare_Locks", lock, 0);
3655 : Assert(lock->nRequested >= 0);
3656 : Assert(lock->nGranted >= 0);
3657 : Assert(lock->nGranted <= lock->nRequested);
3658 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
3659 :
3660 : /* Ignore it if nothing to release (must be a session lock) */
3661 1360 : if (proclock->releaseMask == 0)
3662 14 : continue;
3663 :
3664 : /* Else we should be releasing all locks */
3665 1346 : if (proclock->releaseMask != proclock->holdMask)
3666 0 : elog(PANIC, "we seem to have dropped a bit somewhere");
3667 :
3668 : /*
3669 : * We cannot simply modify proclock->tag.myProc to reassign
3670 : * ownership of the lock, because that's part of the hash key and
3671 : * the proclock would then be in the wrong hash chain. Instead
3672 : * use hash_update_hash_key. (We used to create a new hash entry,
3673 : * but that risks out-of-memory failure if other processes are
3674 : * busy making proclocks too.) We must unlink the proclock from
3675 : * our procLink chain and put it into the new proc's chain, too.
3676 : *
3677 : * Note: the updated proclock hash key will still belong to the
3678 : * same hash partition, cf proclock_hash(). So the partition lock
3679 : * we already hold is sufficient for this.
3680 : */
3681 1346 : dlist_delete(&proclock->procLink);
3682 :
3683 : /*
3684 : * Create the new hash key for the proclock.
3685 : */
3686 1346 : proclocktag.myLock = lock;
3687 1346 : proclocktag.myProc = newproc;
3688 :
3689 : /*
3690 : * Update groupLeader pointer to point to the new proc. (We'd
3691 : * better not be a member of somebody else's lock group!)
3692 : */
3693 : Assert(proclock->groupLeader == proclock->tag.myProc);
3694 1346 : proclock->groupLeader = newproc;
3695 :
3696 : /*
3697 : * Update the proclock. We should not find any existing entry for
3698 : * the same hash key, since there can be only one entry for any
3699 : * given lock with my own proc.
3700 : */
3701 1346 : if (!hash_update_hash_key(LockMethodProcLockHash,
3702 : proclock,
3703 : &proclocktag))
3704 0 : elog(PANIC, "duplicate entry found while reassigning a prepared transaction's locks");
3705 :
3706 : /* Re-link into the new proc's proclock list */
3707 1346 : dlist_push_tail(&newproc->myProcLocks[partition], &proclock->procLink);
3708 :
3709 : PROCLOCK_PRINT("PostPrepare_Locks: updated", proclock);
3710 : } /* loop over PROCLOCKs within this partition */
3711 :
3712 1372 : LWLockRelease(partitionLock);
3713 : } /* loop over partitions */
3714 :
3715 600 : END_CRIT_SECTION();
3716 600 : }
3717 :
3718 :
3719 : /*
3720 : * Estimate shared-memory space used for lock tables
3721 : */
3722 : Size
3723 3794 : LockManagerShmemSize(void)
3724 : {
3725 3794 : Size size = 0;
3726 : long max_table_size;
3727 :
3728 : /* lock hash table */
3729 3794 : max_table_size = NLOCKENTS();
3730 3794 : size = add_size(size, hash_estimate_size(max_table_size, sizeof(LOCK)));
3731 :
3732 : /* proclock hash table */
3733 3794 : max_table_size *= 2;
3734 3794 : size = add_size(size, hash_estimate_size(max_table_size, sizeof(PROCLOCK)));
3735 :
3736 : /*
3737 : * Since NLOCKENTS is only an estimate, add 10% safety margin.
3738 : */
3739 3794 : size = add_size(size, size / 10);
3740 :
3741 3794 : return size;
3742 : }
3743 :
3744 : /*
3745 : * GetLockStatusData - Return a summary of the lock manager's internal
3746 : * status, for use in a user-level reporting function.
3747 : *
3748 : * The return data consists of an array of LockInstanceData objects,
3749 : * which are a lightly abstracted version of the PROCLOCK data structures,
3750 : * i.e. there is one entry for each unique lock and interested PGPROC.
3751 : * It is the caller's responsibility to match up related items (such as
3752 : * references to the same lockable object or PGPROC) if wanted.
3753 : *
3754 : * The design goal is to hold the LWLocks for as short a time as possible;
3755 : * thus, this function simply makes a copy of the necessary data and releases
3756 : * the locks, allowing the caller to contemplate and format the data for as
3757 : * long as it pleases.
3758 : */
3759 : LockData *
3760 736 : GetLockStatusData(void)
3761 : {
3762 : LockData *data;
3763 : PROCLOCK *proclock;
3764 : HASH_SEQ_STATUS seqstat;
3765 : int els;
3766 : int el;
3767 : int i;
3768 :
3769 736 : data = (LockData *) palloc(sizeof(LockData));
3770 :
3771 : /* Guess how much space we'll need. */
3772 736 : els = MaxBackends;
3773 736 : el = 0;
3774 736 : data->locks = (LockInstanceData *) palloc(sizeof(LockInstanceData) * els);
3775 :
3776 : /*
3777 : * First, we iterate through the per-backend fast-path arrays, locking
3778 : * them one at a time. This might produce an inconsistent picture of the
3779 : * system state, but taking all of those LWLocks at the same time seems
3780 : * impractical (in particular, note MAX_SIMUL_LWLOCKS). It shouldn't
3781 : * matter too much, because none of these locks can be involved in lock
3782 : * conflicts anyway - anything that might must be present in the main lock
3783 : * table. (For the same reason, we don't sweat about making leaderPid
3784 : * completely valid. We cannot safely dereference another backend's
3785 : * lockGroupLeader field without holding all lock partition locks, and
3786 : * it's not worth that.)
3787 : */
3788 112206 : for (i = 0; i < ProcGlobal->allProcCount; ++i)
3789 : {
3790 111470 : PGPROC *proc = &ProcGlobal->allProcs[i];
3791 :
3792 : /* Skip backends with pid=0, as they don't hold fast-path locks */
3793 111470 : if (proc->pid == 0)
3794 102398 : continue;
3795 :
3796 9072 : LWLockAcquire(&proc->fpInfoLock, LW_SHARED);
3797 :
3798 45360 : for (uint32 g = 0; g < FastPathLockGroupsPerBackend; g++)
3799 : {
3800 : /* Skip groups without registered fast-path locks */
3801 36288 : if (proc->fpLockBits[g] == 0)
3802 31466 : continue;
3803 :
3804 81974 : for (int j = 0; j < FP_LOCK_SLOTS_PER_GROUP; j++)
3805 : {
3806 : LockInstanceData *instance;
3807 77152 : uint32 f = FAST_PATH_SLOT(g, j);
3808 77152 : uint32 lockbits = FAST_PATH_GET_BITS(proc, f);
3809 :
3810 : /* Skip unallocated slots */
3811 77152 : if (!lockbits)
3812 69620 : continue;
3813 :
3814 7532 : if (el >= els)
3815 : {
3816 20 : els += MaxBackends;
3817 20 : data->locks = (LockInstanceData *)
3818 20 : repalloc(data->locks, sizeof(LockInstanceData) * els);
3819 : }
3820 :
3821 7532 : instance = &data->locks[el];
3822 7532 : SET_LOCKTAG_RELATION(instance->locktag, proc->databaseId,
3823 : proc->fpRelId[f]);
3824 7532 : instance->holdMask = lockbits << FAST_PATH_LOCKNUMBER_OFFSET;
3825 7532 : instance->waitLockMode = NoLock;
3826 7532 : instance->vxid.procNumber = proc->vxid.procNumber;
3827 7532 : instance->vxid.localTransactionId = proc->vxid.lxid;
3828 7532 : instance->pid = proc->pid;
3829 7532 : instance->leaderPid = proc->pid;
3830 7532 : instance->fastpath = true;
3831 :
3832 : /*
3833 : * Successfully taking fast path lock means there were no
3834 : * conflicting locks.
3835 : */
3836 7532 : instance->waitStart = 0;
3837 :
3838 7532 : el++;
3839 : }
3840 : }
3841 :
3842 9072 : if (proc->fpVXIDLock)
3843 : {
3844 : VirtualTransactionId vxid;
3845 : LockInstanceData *instance;
3846 :
3847 2532 : if (el >= els)
3848 : {
3849 4 : els += MaxBackends;
3850 4 : data->locks = (LockInstanceData *)
3851 4 : repalloc(data->locks, sizeof(LockInstanceData) * els);
3852 : }
3853 :
3854 2532 : vxid.procNumber = proc->vxid.procNumber;
3855 2532 : vxid.localTransactionId = proc->fpLocalTransactionId;
3856 :
3857 2532 : instance = &data->locks[el];
3858 2532 : SET_LOCKTAG_VIRTUALTRANSACTION(instance->locktag, vxid);
3859 2532 : instance->holdMask = LOCKBIT_ON(ExclusiveLock);
3860 2532 : instance->waitLockMode = NoLock;
3861 2532 : instance->vxid.procNumber = proc->vxid.procNumber;
3862 2532 : instance->vxid.localTransactionId = proc->vxid.lxid;
3863 2532 : instance->pid = proc->pid;
3864 2532 : instance->leaderPid = proc->pid;
3865 2532 : instance->fastpath = true;
3866 2532 : instance->waitStart = 0;
3867 :
3868 2532 : el++;
3869 : }
3870 :
3871 9072 : LWLockRelease(&proc->fpInfoLock);
3872 : }
3873 :
3874 : /*
3875 : * Next, acquire lock on the entire shared lock data structure. We do
3876 : * this so that, at least for locks in the primary lock table, the state
3877 : * will be self-consistent.
3878 : *
3879 : * Since this is a read-only operation, we take shared instead of
3880 : * exclusive lock. There's not a whole lot of point to this, because all
3881 : * the normal operations require exclusive lock, but it doesn't hurt
3882 : * anything either. It will at least allow two backends to do
3883 : * GetLockStatusData in parallel.
3884 : *
3885 : * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
3886 : */
3887 12512 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
3888 11776 : LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
3889 :
3890 : /* Now we can safely count the number of proclocks */
3891 736 : data->nelements = el + hash_get_num_entries(LockMethodProcLockHash);
3892 736 : if (data->nelements > els)
3893 : {
3894 48 : els = data->nelements;
3895 48 : data->locks = (LockInstanceData *)
3896 48 : repalloc(data->locks, sizeof(LockInstanceData) * els);
3897 : }
3898 :
3899 : /* Now scan the tables to copy the data */
3900 736 : hash_seq_init(&seqstat, LockMethodProcLockHash);
3901 :
3902 6130 : while ((proclock = (PROCLOCK *) hash_seq_search(&seqstat)))
3903 : {
3904 5394 : PGPROC *proc = proclock->tag.myProc;
3905 5394 : LOCK *lock = proclock->tag.myLock;
3906 5394 : LockInstanceData *instance = &data->locks[el];
3907 :
3908 5394 : memcpy(&instance->locktag, &lock->tag, sizeof(LOCKTAG));
3909 5394 : instance->holdMask = proclock->holdMask;
3910 5394 : if (proc->waitLock == proclock->tag.myLock)
3911 18 : instance->waitLockMode = proc->waitLockMode;
3912 : else
3913 5376 : instance->waitLockMode = NoLock;
3914 5394 : instance->vxid.procNumber = proc->vxid.procNumber;
3915 5394 : instance->vxid.localTransactionId = proc->vxid.lxid;
3916 5394 : instance->pid = proc->pid;
3917 5394 : instance->leaderPid = proclock->groupLeader->pid;
3918 5394 : instance->fastpath = false;
3919 5394 : instance->waitStart = (TimestampTz) pg_atomic_read_u64(&proc->waitStart);
3920 :
3921 5394 : el++;
3922 : }
3923 :
3924 : /*
3925 : * And release locks. We do this in reverse order for two reasons: (1)
3926 : * Anyone else who needs more than one of the locks will be trying to lock
3927 : * them in increasing order; we don't want to release the other process
3928 : * until it can get all the locks it needs. (2) This avoids O(N^2)
3929 : * behavior inside LWLockRelease.
3930 : */
3931 12512 : for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
3932 11776 : LWLockRelease(LockHashPartitionLockByIndex(i));
3933 :
3934 : Assert(el == data->nelements);
3935 :
3936 736 : return data;
3937 : }
3938 :
3939 : /*
3940 : * GetBlockerStatusData - Return a summary of the lock manager's state
3941 : * concerning locks that are blocking the specified PID or any member of
3942 : * the PID's lock group, for use in a user-level reporting function.
3943 : *
3944 : * For each PID within the lock group that is awaiting some heavyweight lock,
3945 : * the return data includes an array of LockInstanceData objects, which are
3946 : * the same data structure used by GetLockStatusData; but unlike that function,
3947 : * this one reports only the PROCLOCKs associated with the lock that that PID
3948 : * is blocked on. (Hence, all the locktags should be the same for any one
3949 : * blocked PID.) In addition, we return an array of the PIDs of those backends
3950 : * that are ahead of the blocked PID in the lock's wait queue. These can be
3951 : * compared with the PIDs in the LockInstanceData objects to determine which
3952 : * waiters are ahead of or behind the blocked PID in the queue.
3953 : *
3954 : * If blocked_pid isn't a valid backend PID or nothing in its lock group is
3955 : * waiting on any heavyweight lock, return empty arrays.
3956 : *
3957 : * The design goal is to hold the LWLocks for as short a time as possible;
3958 : * thus, this function simply makes a copy of the necessary data and releases
3959 : * the locks, allowing the caller to contemplate and format the data for as
3960 : * long as it pleases.
3961 : */
3962 : BlockedProcsData *
3963 3126 : GetBlockerStatusData(int blocked_pid)
3964 : {
3965 : BlockedProcsData *data;
3966 : PGPROC *proc;
3967 : int i;
3968 :
3969 3126 : data = (BlockedProcsData *) palloc(sizeof(BlockedProcsData));
3970 :
3971 : /*
3972 : * Guess how much space we'll need, and preallocate. Most of the time
3973 : * this will avoid needing to do repalloc while holding the LWLocks. (We
3974 : * assume, but check with an Assert, that MaxBackends is enough entries
3975 : * for the procs[] array; the other two could need enlargement, though.)
3976 : */
3977 3126 : data->nprocs = data->nlocks = data->npids = 0;
3978 3126 : data->maxprocs = data->maxlocks = data->maxpids = MaxBackends;
3979 3126 : data->procs = (BlockedProcData *) palloc(sizeof(BlockedProcData) * data->maxprocs);
3980 3126 : data->locks = (LockInstanceData *) palloc(sizeof(LockInstanceData) * data->maxlocks);
3981 3126 : data->waiter_pids = (int *) palloc(sizeof(int) * data->maxpids);
3982 :
3983 : /*
3984 : * In order to search the ProcArray for blocked_pid and assume that that
3985 : * entry won't immediately disappear under us, we must hold ProcArrayLock.
3986 : * In addition, to examine the lock grouping fields of any other backend,
3987 : * we must hold all the hash partition locks. (Only one of those locks is
3988 : * actually relevant for any one lock group, but we can't know which one
3989 : * ahead of time.) It's fairly annoying to hold all those locks
3990 : * throughout this, but it's no worse than GetLockStatusData(), and it
3991 : * does have the advantage that we're guaranteed to return a
3992 : * self-consistent instantaneous state.
3993 : */
3994 3126 : LWLockAcquire(ProcArrayLock, LW_SHARED);
3995 :
3996 3126 : proc = BackendPidGetProcWithLock(blocked_pid);
3997 :
3998 : /* Nothing to do if it's gone */
3999 3126 : if (proc != NULL)
4000 : {
4001 : /*
4002 : * Acquire lock on the entire shared lock data structure. See notes
4003 : * in GetLockStatusData().
4004 : */
4005 53142 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
4006 50016 : LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
4007 :
4008 3126 : if (proc->lockGroupLeader == NULL)
4009 : {
4010 : /* Easy case, proc is not a lock group member */
4011 2674 : GetSingleProcBlockerStatusData(proc, data);
4012 : }
4013 : else
4014 : {
4015 : /* Examine all procs in proc's lock group */
4016 : dlist_iter iter;
4017 :
4018 1046 : dlist_foreach(iter, &proc->lockGroupLeader->lockGroupMembers)
4019 : {
4020 : PGPROC *memberProc;
4021 :
4022 594 : memberProc = dlist_container(PGPROC, lockGroupLink, iter.cur);
4023 594 : GetSingleProcBlockerStatusData(memberProc, data);
4024 : }
4025 : }
4026 :
4027 : /*
4028 : * And release locks. See notes in GetLockStatusData().
4029 : */
4030 53142 : for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
4031 50016 : LWLockRelease(LockHashPartitionLockByIndex(i));
4032 :
4033 : Assert(data->nprocs <= data->maxprocs);
4034 : }
4035 :
4036 3126 : LWLockRelease(ProcArrayLock);
4037 :
4038 3126 : return data;
4039 : }
4040 :
4041 : /* Accumulate data about one possibly-blocked proc for GetBlockerStatusData */
4042 : static void
4043 3268 : GetSingleProcBlockerStatusData(PGPROC *blocked_proc, BlockedProcsData *data)
4044 : {
4045 3268 : LOCK *theLock = blocked_proc->waitLock;
4046 : BlockedProcData *bproc;
4047 : dlist_iter proclock_iter;
4048 : dlist_iter proc_iter;
4049 : dclist_head *waitQueue;
4050 : int queue_size;
4051 :
4052 : /* Nothing to do if this proc is not blocked */
4053 3268 : if (theLock == NULL)
4054 1014 : return;
4055 :
4056 : /* Set up a procs[] element */
4057 2254 : bproc = &data->procs[data->nprocs++];
4058 2254 : bproc->pid = blocked_proc->pid;
4059 2254 : bproc->first_lock = data->nlocks;
4060 2254 : bproc->first_waiter = data->npids;
4061 :
4062 : /*
4063 : * We may ignore the proc's fast-path arrays, since nothing in those could
4064 : * be related to a contended lock.
4065 : */
4066 :
4067 : /* Collect all PROCLOCKs associated with theLock */
4068 6850 : dlist_foreach(proclock_iter, &theLock->procLocks)
4069 : {
4070 4596 : PROCLOCK *proclock =
4071 4596 : dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
4072 4596 : PGPROC *proc = proclock->tag.myProc;
4073 4596 : LOCK *lock = proclock->tag.myLock;
4074 : LockInstanceData *instance;
4075 :
4076 4596 : if (data->nlocks >= data->maxlocks)
4077 : {
4078 0 : data->maxlocks += MaxBackends;
4079 0 : data->locks = (LockInstanceData *)
4080 0 : repalloc(data->locks, sizeof(LockInstanceData) * data->maxlocks);
4081 : }
4082 :
4083 4596 : instance = &data->locks[data->nlocks];
4084 4596 : memcpy(&instance->locktag, &lock->tag, sizeof(LOCKTAG));
4085 4596 : instance->holdMask = proclock->holdMask;
4086 4596 : if (proc->waitLock == lock)
4087 2332 : instance->waitLockMode = proc->waitLockMode;
4088 : else
4089 2264 : instance->waitLockMode = NoLock;
4090 4596 : instance->vxid.procNumber = proc->vxid.procNumber;
4091 4596 : instance->vxid.localTransactionId = proc->vxid.lxid;
4092 4596 : instance->pid = proc->pid;
4093 4596 : instance->leaderPid = proclock->groupLeader->pid;
4094 4596 : instance->fastpath = false;
4095 4596 : data->nlocks++;
4096 : }
4097 :
4098 : /* Enlarge waiter_pids[] if it's too small to hold all wait queue PIDs */
4099 2254 : waitQueue = &(theLock->waitProcs);
4100 2254 : queue_size = dclist_count(waitQueue);
4101 :
4102 2254 : if (queue_size > data->maxpids - data->npids)
4103 : {
4104 0 : data->maxpids = Max(data->maxpids + MaxBackends,
4105 : data->npids + queue_size);
4106 0 : data->waiter_pids = (int *) repalloc(data->waiter_pids,
4107 0 : sizeof(int) * data->maxpids);
4108 : }
4109 :
4110 : /* Collect PIDs from the lock's wait queue, stopping at blocked_proc */
4111 2290 : dclist_foreach(proc_iter, waitQueue)
4112 : {
4113 2290 : PGPROC *queued_proc = dlist_container(PGPROC, links, proc_iter.cur);
4114 :
4115 2290 : if (queued_proc == blocked_proc)
4116 2254 : break;
4117 36 : data->waiter_pids[data->npids++] = queued_proc->pid;
4118 36 : queued_proc = (PGPROC *) queued_proc->links.next;
4119 : }
4120 :
4121 2254 : bproc->num_locks = data->nlocks - bproc->first_lock;
4122 2254 : bproc->num_waiters = data->npids - bproc->first_waiter;
4123 : }
4124 :
4125 : /*
4126 : * Returns a list of currently held AccessExclusiveLocks, for use by
4127 : * LogStandbySnapshot(). The result is a palloc'd array,
4128 : * with the number of elements returned into *nlocks.
4129 : *
4130 : * XXX This currently takes a lock on all partitions of the lock table,
4131 : * but it's possible to do better. By reference counting locks and storing
4132 : * the value in the ProcArray entry for each backend we could tell if any
4133 : * locks need recording without having to acquire the partition locks and
4134 : * scan the lock table. Whether that's worth the additional overhead
4135 : * is pretty dubious though.
4136 : */
4137 : xl_standby_lock *
4138 1976 : GetRunningTransactionLocks(int *nlocks)
4139 : {
4140 : xl_standby_lock *accessExclusiveLocks;
4141 : PROCLOCK *proclock;
4142 : HASH_SEQ_STATUS seqstat;
4143 : int i;
4144 : int index;
4145 : int els;
4146 :
4147 : /*
4148 : * Acquire lock on the entire shared lock data structure.
4149 : *
4150 : * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
4151 : */
4152 33592 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
4153 31616 : LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
4154 :
4155 : /* Now we can safely count the number of proclocks */
4156 1976 : els = hash_get_num_entries(LockMethodProcLockHash);
4157 :
4158 : /*
4159 : * Allocating enough space for all locks in the lock table is overkill,
4160 : * but it's more convenient and faster than having to enlarge the array.
4161 : */
4162 1976 : accessExclusiveLocks = palloc(els * sizeof(xl_standby_lock));
4163 :
4164 : /* Now scan the tables to copy the data */
4165 1976 : hash_seq_init(&seqstat, LockMethodProcLockHash);
4166 :
4167 : /*
4168 : * If lock is a currently granted AccessExclusiveLock then it will have
4169 : * just one proclock holder, so locks are never accessed twice in this
4170 : * particular case. Don't copy this code for use elsewhere because in the
4171 : * general case this will give you duplicate locks when looking at
4172 : * non-exclusive lock types.
4173 : */
4174 1976 : index = 0;
4175 10496 : while ((proclock = (PROCLOCK *) hash_seq_search(&seqstat)))
4176 : {
4177 : /* make sure this definition matches the one used in LockAcquire */
4178 8520 : if ((proclock->holdMask & LOCKBIT_ON(AccessExclusiveLock)) &&
4179 4532 : proclock->tag.myLock->tag.locktag_type == LOCKTAG_RELATION)
4180 : {
4181 2830 : PGPROC *proc = proclock->tag.myProc;
4182 2830 : LOCK *lock = proclock->tag.myLock;
4183 2830 : TransactionId xid = proc->xid;
4184 :
4185 : /*
4186 : * Don't record locks for transactions if we know they have
4187 : * already issued their WAL record for commit but not yet released
4188 : * lock. It is still possible that we see locks held by already
4189 : * complete transactions, if they haven't yet zeroed their xids.
4190 : */
4191 2830 : if (!TransactionIdIsValid(xid))
4192 18 : continue;
4193 :
4194 2812 : accessExclusiveLocks[index].xid = xid;
4195 2812 : accessExclusiveLocks[index].dbOid = lock->tag.locktag_field1;
4196 2812 : accessExclusiveLocks[index].relOid = lock->tag.locktag_field2;
4197 :
4198 2812 : index++;
4199 : }
4200 : }
4201 :
4202 : Assert(index <= els);
4203 :
4204 : /*
4205 : * And release locks. We do this in reverse order for two reasons: (1)
4206 : * Anyone else who needs more than one of the locks will be trying to lock
4207 : * them in increasing order; we don't want to release the other process
4208 : * until it can get all the locks it needs. (2) This avoids O(N^2)
4209 : * behavior inside LWLockRelease.
4210 : */
4211 33592 : for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
4212 31616 : LWLockRelease(LockHashPartitionLockByIndex(i));
4213 :
4214 1976 : *nlocks = index;
4215 1976 : return accessExclusiveLocks;
4216 : }
4217 :
4218 : /* Provide the textual name of any lock mode */
4219 : const char *
4220 16466 : GetLockmodeName(LOCKMETHODID lockmethodid, LOCKMODE mode)
4221 : {
4222 : Assert(lockmethodid > 0 && lockmethodid < lengthof(LockMethods));
4223 : Assert(mode > 0 && mode <= LockMethods[lockmethodid]->numLockModes);
4224 16466 : return LockMethods[lockmethodid]->lockModeNames[mode];
4225 : }
4226 :
4227 : #ifdef LOCK_DEBUG
4228 : /*
4229 : * Dump all locks in the given proc's myProcLocks lists.
4230 : *
4231 : * Caller is responsible for having acquired appropriate LWLocks.
4232 : */
4233 : void
4234 : DumpLocks(PGPROC *proc)
4235 : {
4236 : int i;
4237 :
4238 : if (proc == NULL)
4239 : return;
4240 :
4241 : if (proc->waitLock)
4242 : LOCK_PRINT("DumpLocks: waiting on", proc->waitLock, 0);
4243 :
4244 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
4245 : {
4246 : dlist_head *procLocks = &proc->myProcLocks[i];
4247 : dlist_iter iter;
4248 :
4249 : dlist_foreach(iter, procLocks)
4250 : {
4251 : PROCLOCK *proclock = dlist_container(PROCLOCK, procLink, iter.cur);
4252 : LOCK *lock = proclock->tag.myLock;
4253 :
4254 : Assert(proclock->tag.myProc == proc);
4255 : PROCLOCK_PRINT("DumpLocks", proclock);
4256 : LOCK_PRINT("DumpLocks", lock, 0);
4257 : }
4258 : }
4259 : }
4260 :
4261 : /*
4262 : * Dump all lmgr locks.
4263 : *
4264 : * Caller is responsible for having acquired appropriate LWLocks.
4265 : */
4266 : void
4267 : DumpAllLocks(void)
4268 : {
4269 : PGPROC *proc;
4270 : PROCLOCK *proclock;
4271 : LOCK *lock;
4272 : HASH_SEQ_STATUS status;
4273 :
4274 : proc = MyProc;
4275 :
4276 : if (proc && proc->waitLock)
4277 : LOCK_PRINT("DumpAllLocks: waiting on", proc->waitLock, 0);
4278 :
4279 : hash_seq_init(&status, LockMethodProcLockHash);
4280 :
4281 : while ((proclock = (PROCLOCK *) hash_seq_search(&status)) != NULL)
4282 : {
4283 : PROCLOCK_PRINT("DumpAllLocks", proclock);
4284 :
4285 : lock = proclock->tag.myLock;
4286 : if (lock)
4287 : LOCK_PRINT("DumpAllLocks", lock, 0);
4288 : else
4289 : elog(LOG, "DumpAllLocks: proclock->tag.myLock = NULL");
4290 : }
4291 : }
4292 : #endif /* LOCK_DEBUG */
4293 :
4294 : /*
4295 : * LOCK 2PC resource manager's routines
4296 : */
4297 :
4298 : /*
4299 : * Re-acquire a lock belonging to a transaction that was prepared.
4300 : *
4301 : * Because this function is run at db startup, re-acquiring the locks should
4302 : * never conflict with running transactions because there are none. We
4303 : * assume that the lock state represented by the stored 2PC files is legal.
4304 : *
4305 : * When switching from Hot Standby mode to normal operation, the locks will
4306 : * be already held by the startup process. The locks are acquired for the new
4307 : * procs without checking for conflicts, so we don't get a conflict between the
4308 : * startup process and the dummy procs, even though we will momentarily have
4309 : * a situation where two procs are holding the same AccessExclusiveLock,
4310 : * which isn't normally possible because the conflict. If we're in standby
4311 : * mode, but a recovery snapshot hasn't been established yet, it's possible
4312 : * that some but not all of the locks are already held by the startup process.
4313 : *
4314 : * This approach is simple, but also a bit dangerous, because if there isn't
4315 : * enough shared memory to acquire the locks, an error will be thrown, which
4316 : * is promoted to FATAL and recovery will abort, bringing down postmaster.
4317 : * A safer approach would be to transfer the locks like we do in
4318 : * AtPrepare_Locks, but then again, in hot standby mode it's possible for
4319 : * read-only backends to use up all the shared lock memory anyway, so that
4320 : * replaying the WAL record that needs to acquire a lock will throw an error
4321 : * and PANIC anyway.
4322 : */
4323 : void
4324 176 : lock_twophase_recover(TransactionId xid, uint16 info,
4325 : void *recdata, uint32 len)
4326 : {
4327 176 : TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4328 176 : PGPROC *proc = TwoPhaseGetDummyProc(xid, false);
4329 : LOCKTAG *locktag;
4330 : LOCKMODE lockmode;
4331 : LOCKMETHODID lockmethodid;
4332 : LOCK *lock;
4333 : PROCLOCK *proclock;
4334 : PROCLOCKTAG proclocktag;
4335 : bool found;
4336 : uint32 hashcode;
4337 : uint32 proclock_hashcode;
4338 : int partition;
4339 : LWLock *partitionLock;
4340 : LockMethod lockMethodTable;
4341 :
4342 : Assert(len == sizeof(TwoPhaseLockRecord));
4343 176 : locktag = &rec->locktag;
4344 176 : lockmode = rec->lockmode;
4345 176 : lockmethodid = locktag->locktag_lockmethodid;
4346 :
4347 176 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4348 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4349 176 : lockMethodTable = LockMethods[lockmethodid];
4350 :
4351 176 : hashcode = LockTagHashCode(locktag);
4352 176 : partition = LockHashPartition(hashcode);
4353 176 : partitionLock = LockHashPartitionLock(hashcode);
4354 :
4355 176 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4356 :
4357 : /*
4358 : * Find or create a lock with this tag.
4359 : */
4360 176 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
4361 : locktag,
4362 : hashcode,
4363 : HASH_ENTER_NULL,
4364 : &found);
4365 176 : if (!lock)
4366 : {
4367 0 : LWLockRelease(partitionLock);
4368 0 : ereport(ERROR,
4369 : (errcode(ERRCODE_OUT_OF_MEMORY),
4370 : errmsg("out of shared memory"),
4371 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
4372 : }
4373 :
4374 : /*
4375 : * if it's a new lock object, initialize it
4376 : */
4377 176 : if (!found)
4378 : {
4379 152 : lock->grantMask = 0;
4380 152 : lock->waitMask = 0;
4381 152 : dlist_init(&lock->procLocks);
4382 152 : dclist_init(&lock->waitProcs);
4383 152 : lock->nRequested = 0;
4384 152 : lock->nGranted = 0;
4385 912 : MemSet(lock->requested, 0, sizeof(int) * MAX_LOCKMODES);
4386 152 : MemSet(lock->granted, 0, sizeof(int) * MAX_LOCKMODES);
4387 : LOCK_PRINT("lock_twophase_recover: new", lock, lockmode);
4388 : }
4389 : else
4390 : {
4391 : LOCK_PRINT("lock_twophase_recover: found", lock, lockmode);
4392 : Assert((lock->nRequested >= 0) && (lock->requested[lockmode] >= 0));
4393 : Assert((lock->nGranted >= 0) && (lock->granted[lockmode] >= 0));
4394 : Assert(lock->nGranted <= lock->nRequested);
4395 : }
4396 :
4397 : /*
4398 : * Create the hash key for the proclock table.
4399 : */
4400 176 : proclocktag.myLock = lock;
4401 176 : proclocktag.myProc = proc;
4402 :
4403 176 : proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
4404 :
4405 : /*
4406 : * Find or create a proclock entry with this tag
4407 : */
4408 176 : proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
4409 : &proclocktag,
4410 : proclock_hashcode,
4411 : HASH_ENTER_NULL,
4412 : &found);
4413 176 : if (!proclock)
4414 : {
4415 : /* Oops, not enough shmem for the proclock */
4416 0 : if (lock->nRequested == 0)
4417 : {
4418 : /*
4419 : * There are no other requestors of this lock, so garbage-collect
4420 : * the lock object. We *must* do this to avoid a permanent leak
4421 : * of shared memory, because there won't be anything to cause
4422 : * anyone to release the lock object later.
4423 : */
4424 : Assert(dlist_is_empty(&lock->procLocks));
4425 0 : if (!hash_search_with_hash_value(LockMethodLockHash,
4426 0 : &(lock->tag),
4427 : hashcode,
4428 : HASH_REMOVE,
4429 : NULL))
4430 0 : elog(PANIC, "lock table corrupted");
4431 : }
4432 0 : LWLockRelease(partitionLock);
4433 0 : ereport(ERROR,
4434 : (errcode(ERRCODE_OUT_OF_MEMORY),
4435 : errmsg("out of shared memory"),
4436 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
4437 : }
4438 :
4439 : /*
4440 : * If new, initialize the new entry
4441 : */
4442 176 : if (!found)
4443 : {
4444 : Assert(proc->lockGroupLeader == NULL);
4445 160 : proclock->groupLeader = proc;
4446 160 : proclock->holdMask = 0;
4447 160 : proclock->releaseMask = 0;
4448 : /* Add proclock to appropriate lists */
4449 160 : dlist_push_tail(&lock->procLocks, &proclock->lockLink);
4450 160 : dlist_push_tail(&proc->myProcLocks[partition],
4451 : &proclock->procLink);
4452 : PROCLOCK_PRINT("lock_twophase_recover: new", proclock);
4453 : }
4454 : else
4455 : {
4456 : PROCLOCK_PRINT("lock_twophase_recover: found", proclock);
4457 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
4458 : }
4459 :
4460 : /*
4461 : * lock->nRequested and lock->requested[] count the total number of
4462 : * requests, whether granted or waiting, so increment those immediately.
4463 : */
4464 176 : lock->nRequested++;
4465 176 : lock->requested[lockmode]++;
4466 : Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
4467 :
4468 : /*
4469 : * We shouldn't already hold the desired lock.
4470 : */
4471 176 : if (proclock->holdMask & LOCKBIT_ON(lockmode))
4472 0 : elog(ERROR, "lock %s on object %u/%u/%u is already held",
4473 : lockMethodTable->lockModeNames[lockmode],
4474 : lock->tag.locktag_field1, lock->tag.locktag_field2,
4475 : lock->tag.locktag_field3);
4476 :
4477 : /*
4478 : * We ignore any possible conflicts and just grant ourselves the lock. Not
4479 : * only because we don't bother, but also to avoid deadlocks when
4480 : * switching from standby to normal mode. See function comment.
4481 : */
4482 176 : GrantLock(lock, proclock, lockmode);
4483 :
4484 : /*
4485 : * Bump strong lock count, to make sure any fast-path lock requests won't
4486 : * be granted without consulting the primary lock table.
4487 : */
4488 176 : if (ConflictsWithRelationFastPath(&lock->tag, lockmode))
4489 : {
4490 36 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
4491 :
4492 36 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
4493 36 : FastPathStrongRelationLocks->count[fasthashcode]++;
4494 36 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
4495 : }
4496 :
4497 176 : LWLockRelease(partitionLock);
4498 176 : }
4499 :
4500 : /*
4501 : * Re-acquire a lock belonging to a transaction that was prepared, when
4502 : * starting up into hot standby mode.
4503 : */
4504 : void
4505 0 : lock_twophase_standby_recover(TransactionId xid, uint16 info,
4506 : void *recdata, uint32 len)
4507 : {
4508 0 : TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4509 : LOCKTAG *locktag;
4510 : LOCKMODE lockmode;
4511 : LOCKMETHODID lockmethodid;
4512 :
4513 : Assert(len == sizeof(TwoPhaseLockRecord));
4514 0 : locktag = &rec->locktag;
4515 0 : lockmode = rec->lockmode;
4516 0 : lockmethodid = locktag->locktag_lockmethodid;
4517 :
4518 0 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4519 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4520 :
4521 0 : if (lockmode == AccessExclusiveLock &&
4522 0 : locktag->locktag_type == LOCKTAG_RELATION)
4523 : {
4524 0 : StandbyAcquireAccessExclusiveLock(xid,
4525 : locktag->locktag_field1 /* dboid */ ,
4526 : locktag->locktag_field2 /* reloid */ );
4527 : }
4528 0 : }
4529 :
4530 :
4531 : /*
4532 : * 2PC processing routine for COMMIT PREPARED case.
4533 : *
4534 : * Find and release the lock indicated by the 2PC record.
4535 : */
4536 : void
4537 1510 : lock_twophase_postcommit(TransactionId xid, uint16 info,
4538 : void *recdata, uint32 len)
4539 : {
4540 1510 : TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4541 1510 : PGPROC *proc = TwoPhaseGetDummyProc(xid, true);
4542 : LOCKTAG *locktag;
4543 : LOCKMETHODID lockmethodid;
4544 : LockMethod lockMethodTable;
4545 :
4546 : Assert(len == sizeof(TwoPhaseLockRecord));
4547 1510 : locktag = &rec->locktag;
4548 1510 : lockmethodid = locktag->locktag_lockmethodid;
4549 :
4550 1510 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4551 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4552 1510 : lockMethodTable = LockMethods[lockmethodid];
4553 :
4554 1510 : LockRefindAndRelease(lockMethodTable, proc, locktag, rec->lockmode, true);
4555 1510 : }
4556 :
4557 : /*
4558 : * 2PC processing routine for ROLLBACK PREPARED case.
4559 : *
4560 : * This is actually just the same as the COMMIT case.
4561 : */
4562 : void
4563 252 : lock_twophase_postabort(TransactionId xid, uint16 info,
4564 : void *recdata, uint32 len)
4565 : {
4566 252 : lock_twophase_postcommit(xid, info, recdata, len);
4567 252 : }
4568 :
4569 : /*
4570 : * VirtualXactLockTableInsert
4571 : *
4572 : * Take vxid lock via the fast-path. There can't be any pre-existing
4573 : * lockers, as we haven't advertised this vxid via the ProcArray yet.
4574 : *
4575 : * Since MyProc->fpLocalTransactionId will normally contain the same data
4576 : * as MyProc->vxid.lxid, you might wonder if we really need both. The
4577 : * difference is that MyProc->vxid.lxid is set and cleared unlocked, and
4578 : * examined by procarray.c, while fpLocalTransactionId is protected by
4579 : * fpInfoLock and is used only by the locking subsystem. Doing it this
4580 : * way makes it easier to verify that there are no funny race conditions.
4581 : *
4582 : * We don't bother recording this lock in the local lock table, since it's
4583 : * only ever released at the end of a transaction. Instead,
4584 : * LockReleaseAll() calls VirtualXactLockTableCleanup().
4585 : */
4586 : void
4587 824148 : VirtualXactLockTableInsert(VirtualTransactionId vxid)
4588 : {
4589 : Assert(VirtualTransactionIdIsValid(vxid));
4590 :
4591 824148 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
4592 :
4593 : Assert(MyProc->vxid.procNumber == vxid.procNumber);
4594 : Assert(MyProc->fpLocalTransactionId == InvalidLocalTransactionId);
4595 : Assert(MyProc->fpVXIDLock == false);
4596 :
4597 824148 : MyProc->fpVXIDLock = true;
4598 824148 : MyProc->fpLocalTransactionId = vxid.localTransactionId;
4599 :
4600 824148 : LWLockRelease(&MyProc->fpInfoLock);
4601 824148 : }
4602 :
4603 : /*
4604 : * VirtualXactLockTableCleanup
4605 : *
4606 : * Check whether a VXID lock has been materialized; if so, release it,
4607 : * unblocking waiters.
4608 : */
4609 : void
4610 824966 : VirtualXactLockTableCleanup(void)
4611 : {
4612 : bool fastpath;
4613 : LocalTransactionId lxid;
4614 :
4615 : Assert(MyProc->vxid.procNumber != INVALID_PROC_NUMBER);
4616 :
4617 : /*
4618 : * Clean up shared memory state.
4619 : */
4620 824966 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
4621 :
4622 824966 : fastpath = MyProc->fpVXIDLock;
4623 824966 : lxid = MyProc->fpLocalTransactionId;
4624 824966 : MyProc->fpVXIDLock = false;
4625 824966 : MyProc->fpLocalTransactionId = InvalidLocalTransactionId;
4626 :
4627 824966 : LWLockRelease(&MyProc->fpInfoLock);
4628 :
4629 : /*
4630 : * If fpVXIDLock has been cleared without touching fpLocalTransactionId,
4631 : * that means someone transferred the lock to the main lock table.
4632 : */
4633 824966 : if (!fastpath && LocalTransactionIdIsValid(lxid))
4634 : {
4635 : VirtualTransactionId vxid;
4636 : LOCKTAG locktag;
4637 :
4638 494 : vxid.procNumber = MyProcNumber;
4639 494 : vxid.localTransactionId = lxid;
4640 494 : SET_LOCKTAG_VIRTUALTRANSACTION(locktag, vxid);
4641 :
4642 494 : LockRefindAndRelease(LockMethods[DEFAULT_LOCKMETHOD], MyProc,
4643 : &locktag, ExclusiveLock, false);
4644 : }
4645 824966 : }
4646 :
4647 : /*
4648 : * XactLockForVirtualXact
4649 : *
4650 : * If TransactionIdIsValid(xid), this is essentially XactLockTableWait(xid,
4651 : * NULL, NULL, XLTW_None) or ConditionalXactLockTableWait(xid). Unlike those
4652 : * functions, it assumes "xid" is never a subtransaction and that "xid" is
4653 : * prepared, committed, or aborted.
4654 : *
4655 : * If !TransactionIdIsValid(xid), this locks every prepared XID having been
4656 : * known as "vxid" before its PREPARE TRANSACTION.
4657 : */
4658 : static bool
4659 552 : XactLockForVirtualXact(VirtualTransactionId vxid,
4660 : TransactionId xid, bool wait)
4661 : {
4662 552 : bool more = false;
4663 :
4664 : /* There is no point to wait for 2PCs if you have no 2PCs. */
4665 552 : if (max_prepared_xacts == 0)
4666 160 : return true;
4667 :
4668 : do
4669 : {
4670 : LockAcquireResult lar;
4671 : LOCKTAG tag;
4672 :
4673 : /* Clear state from previous iterations. */
4674 392 : if (more)
4675 : {
4676 0 : xid = InvalidTransactionId;
4677 0 : more = false;
4678 : }
4679 :
4680 : /* If we have no xid, try to find one. */
4681 392 : if (!TransactionIdIsValid(xid))
4682 192 : xid = TwoPhaseGetXidByVirtualXID(vxid, &more);
4683 392 : if (!TransactionIdIsValid(xid))
4684 : {
4685 : Assert(!more);
4686 180 : return true;
4687 : }
4688 :
4689 : /* Check or wait for XID completion. */
4690 212 : SET_LOCKTAG_TRANSACTION(tag, xid);
4691 212 : lar = LockAcquire(&tag, ShareLock, false, !wait);
4692 212 : if (lar == LOCKACQUIRE_NOT_AVAIL)
4693 0 : return false;
4694 212 : LockRelease(&tag, ShareLock, false);
4695 212 : } while (more);
4696 :
4697 212 : return true;
4698 : }
4699 :
4700 : /*
4701 : * VirtualXactLock
4702 : *
4703 : * If wait = true, wait as long as the given VXID or any XID acquired by the
4704 : * same transaction is still running. Then, return true.
4705 : *
4706 : * If wait = false, just check whether that VXID or one of those XIDs is still
4707 : * running, and return true or false.
4708 : */
4709 : bool
4710 632 : VirtualXactLock(VirtualTransactionId vxid, bool wait)
4711 : {
4712 : LOCKTAG tag;
4713 : PGPROC *proc;
4714 632 : TransactionId xid = InvalidTransactionId;
4715 :
4716 : Assert(VirtualTransactionIdIsValid(vxid));
4717 :
4718 632 : if (VirtualTransactionIdIsRecoveredPreparedXact(vxid))
4719 : /* no vxid lock; localTransactionId is a normal, locked XID */
4720 2 : return XactLockForVirtualXact(vxid, vxid.localTransactionId, wait);
4721 :
4722 630 : SET_LOCKTAG_VIRTUALTRANSACTION(tag, vxid);
4723 :
4724 : /*
4725 : * If a lock table entry must be made, this is the PGPROC on whose behalf
4726 : * it must be done. Note that the transaction might end or the PGPROC
4727 : * might be reassigned to a new backend before we get around to examining
4728 : * it, but it doesn't matter. If we find upon examination that the
4729 : * relevant lxid is no longer running here, that's enough to prove that
4730 : * it's no longer running anywhere.
4731 : */
4732 630 : proc = ProcNumberGetProc(vxid.procNumber);
4733 630 : if (proc == NULL)
4734 8 : return XactLockForVirtualXact(vxid, InvalidTransactionId, wait);
4735 :
4736 : /*
4737 : * We must acquire this lock before checking the procNumber and lxid
4738 : * against the ones we're waiting for. The target backend will only set
4739 : * or clear lxid while holding this lock.
4740 : */
4741 622 : LWLockAcquire(&proc->fpInfoLock, LW_EXCLUSIVE);
4742 :
4743 622 : if (proc->vxid.procNumber != vxid.procNumber
4744 622 : || proc->fpLocalTransactionId != vxid.localTransactionId)
4745 : {
4746 : /* VXID ended */
4747 84 : LWLockRelease(&proc->fpInfoLock);
4748 84 : return XactLockForVirtualXact(vxid, InvalidTransactionId, wait);
4749 : }
4750 :
4751 : /*
4752 : * If we aren't asked to wait, there's no need to set up a lock table
4753 : * entry. The transaction is still in progress, so just return false.
4754 : */
4755 538 : if (!wait)
4756 : {
4757 30 : LWLockRelease(&proc->fpInfoLock);
4758 30 : return false;
4759 : }
4760 :
4761 : /*
4762 : * OK, we're going to need to sleep on the VXID. But first, we must set
4763 : * up the primary lock table entry, if needed (ie, convert the proc's
4764 : * fast-path lock on its VXID to a regular lock).
4765 : */
4766 508 : if (proc->fpVXIDLock)
4767 : {
4768 : PROCLOCK *proclock;
4769 : uint32 hashcode;
4770 : LWLock *partitionLock;
4771 :
4772 494 : hashcode = LockTagHashCode(&tag);
4773 :
4774 494 : partitionLock = LockHashPartitionLock(hashcode);
4775 494 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4776 :
4777 494 : proclock = SetupLockInTable(LockMethods[DEFAULT_LOCKMETHOD], proc,
4778 : &tag, hashcode, ExclusiveLock);
4779 494 : if (!proclock)
4780 : {
4781 0 : LWLockRelease(partitionLock);
4782 0 : LWLockRelease(&proc->fpInfoLock);
4783 0 : ereport(ERROR,
4784 : (errcode(ERRCODE_OUT_OF_MEMORY),
4785 : errmsg("out of shared memory"),
4786 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
4787 : }
4788 494 : GrantLock(proclock->tag.myLock, proclock, ExclusiveLock);
4789 :
4790 494 : LWLockRelease(partitionLock);
4791 :
4792 494 : proc->fpVXIDLock = false;
4793 : }
4794 :
4795 : /*
4796 : * If the proc has an XID now, we'll avoid a TwoPhaseGetXidByVirtualXID()
4797 : * search. The proc might have assigned this XID but not yet locked it,
4798 : * in which case the proc will lock this XID before releasing the VXID.
4799 : * The fpInfoLock critical section excludes VirtualXactLockTableCleanup(),
4800 : * so we won't save an XID of a different VXID. It doesn't matter whether
4801 : * we save this before or after setting up the primary lock table entry.
4802 : */
4803 508 : xid = proc->xid;
4804 :
4805 : /* Done with proc->fpLockBits */
4806 508 : LWLockRelease(&proc->fpInfoLock);
4807 :
4808 : /* Time to wait. */
4809 508 : (void) LockAcquire(&tag, ShareLock, false, false);
4810 :
4811 458 : LockRelease(&tag, ShareLock, false);
4812 458 : return XactLockForVirtualXact(vxid, xid, wait);
4813 : }
4814 :
4815 : /*
4816 : * LockWaiterCount
4817 : *
4818 : * Find the number of lock requester on this locktag
4819 : */
4820 : int
4821 129872 : LockWaiterCount(const LOCKTAG *locktag)
4822 : {
4823 129872 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
4824 : LOCK *lock;
4825 : bool found;
4826 : uint32 hashcode;
4827 : LWLock *partitionLock;
4828 129872 : int waiters = 0;
4829 :
4830 129872 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4831 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4832 :
4833 129872 : hashcode = LockTagHashCode(locktag);
4834 129872 : partitionLock = LockHashPartitionLock(hashcode);
4835 129872 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4836 :
4837 129872 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
4838 : locktag,
4839 : hashcode,
4840 : HASH_FIND,
4841 : &found);
4842 129872 : if (found)
4843 : {
4844 : Assert(lock != NULL);
4845 48 : waiters = lock->nRequested;
4846 : }
4847 129872 : LWLockRelease(partitionLock);
4848 :
4849 129872 : return waiters;
4850 : }
|
