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_failures = 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 : * We can assume that FastPathLockGroupsPerBackend is a power-of-two per
215 : * InitializeFastPathLocks().
216 : */
217 : #define FAST_PATH_REL_GROUP(rel) \
218 : (((uint64) (rel) * 49157) & (FastPathLockGroupsPerBackend - 1))
219 :
220 : /*
221 : * Given the group/slot indexes, calculate the slot index in the whole array
222 : * of fast-path lock slots.
223 : */
224 : #define FAST_PATH_SLOT(group, index) \
225 : (AssertMacro((uint32) (group) < FastPathLockGroupsPerBackend), \
226 : AssertMacro((uint32) (index) < FP_LOCK_SLOTS_PER_GROUP), \
227 : ((group) * FP_LOCK_SLOTS_PER_GROUP + (index)))
228 :
229 : /*
230 : * Given a slot index (into the whole per-backend array), calculated using
231 : * the FAST_PATH_SLOT macro, split it into group and index (in the group).
232 : */
233 : #define FAST_PATH_GROUP(index) \
234 : (AssertMacro((uint32) (index) < FastPathLockSlotsPerBackend()), \
235 : ((index) / FP_LOCK_SLOTS_PER_GROUP))
236 : #define FAST_PATH_INDEX(index) \
237 : (AssertMacro((uint32) (index) < FastPathLockSlotsPerBackend()), \
238 : ((index) % FP_LOCK_SLOTS_PER_GROUP))
239 :
240 : /* Macros for manipulating proc->fpLockBits */
241 : #define FAST_PATH_BITS_PER_SLOT 3
242 : #define FAST_PATH_LOCKNUMBER_OFFSET 1
243 : #define FAST_PATH_MASK ((1 << FAST_PATH_BITS_PER_SLOT) - 1)
244 : #define FAST_PATH_BITS(proc, n) (proc)->fpLockBits[FAST_PATH_GROUP(n)]
245 : #define FAST_PATH_GET_BITS(proc, n) \
246 : ((FAST_PATH_BITS(proc, n) >> (FAST_PATH_BITS_PER_SLOT * FAST_PATH_INDEX(n))) & FAST_PATH_MASK)
247 : #define FAST_PATH_BIT_POSITION(n, l) \
248 : (AssertMacro((l) >= FAST_PATH_LOCKNUMBER_OFFSET), \
249 : AssertMacro((l) < FAST_PATH_BITS_PER_SLOT+FAST_PATH_LOCKNUMBER_OFFSET), \
250 : AssertMacro((n) < FastPathLockSlotsPerBackend()), \
251 : ((l) - FAST_PATH_LOCKNUMBER_OFFSET + FAST_PATH_BITS_PER_SLOT * (FAST_PATH_INDEX(n))))
252 : #define FAST_PATH_SET_LOCKMODE(proc, n, l) \
253 : FAST_PATH_BITS(proc, n) |= UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l)
254 : #define FAST_PATH_CLEAR_LOCKMODE(proc, n, l) \
255 : FAST_PATH_BITS(proc, n) &= ~(UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l))
256 : #define FAST_PATH_CHECK_LOCKMODE(proc, n, l) \
257 : (FAST_PATH_BITS(proc, n) & (UINT64CONST(1) << FAST_PATH_BIT_POSITION(n, l)))
258 :
259 : /*
260 : * The fast-path lock mechanism is concerned only with relation locks on
261 : * unshared relations by backends bound to a database. The fast-path
262 : * mechanism exists mostly to accelerate acquisition and release of locks
263 : * that rarely conflict. Because ShareUpdateExclusiveLock is
264 : * self-conflicting, it can't use the fast-path mechanism; but it also does
265 : * not conflict with any of the locks that do, so we can ignore it completely.
266 : */
267 : #define EligibleForRelationFastPath(locktag, mode) \
268 : ((locktag)->locktag_lockmethodid == DEFAULT_LOCKMETHOD && \
269 : (locktag)->locktag_type == LOCKTAG_RELATION && \
270 : (locktag)->locktag_field1 == MyDatabaseId && \
271 : MyDatabaseId != InvalidOid && \
272 : (mode) < ShareUpdateExclusiveLock)
273 : #define ConflictsWithRelationFastPath(locktag, mode) \
274 : ((locktag)->locktag_lockmethodid == DEFAULT_LOCKMETHOD && \
275 : (locktag)->locktag_type == LOCKTAG_RELATION && \
276 : (locktag)->locktag_field1 != InvalidOid && \
277 : (mode) > ShareUpdateExclusiveLock)
278 :
279 : static bool FastPathGrantRelationLock(Oid relid, LOCKMODE lockmode);
280 : static bool FastPathUnGrantRelationLock(Oid relid, LOCKMODE lockmode);
281 : static bool FastPathTransferRelationLocks(LockMethod lockMethodTable,
282 : const LOCKTAG *locktag, uint32 hashcode);
283 : static PROCLOCK *FastPathGetRelationLockEntry(LOCALLOCK *locallock);
284 :
285 : /*
286 : * To make the fast-path lock mechanism work, we must have some way of
287 : * preventing the use of the fast-path when a conflicting lock might be present.
288 : * We partition* the locktag space into FAST_PATH_STRONG_LOCK_HASH_PARTITIONS,
289 : * and maintain an integer count of the number of "strong" lockers
290 : * in each partition. When any "strong" lockers are present (which is
291 : * hopefully not very often), the fast-path mechanism can't be used, and we
292 : * must fall back to the slower method of pushing matching locks directly
293 : * into the main lock tables.
294 : *
295 : * The deadlock detector does not know anything about the fast path mechanism,
296 : * so any locks that might be involved in a deadlock must be transferred from
297 : * the fast-path queues to the main lock table.
298 : */
299 :
300 : #define FAST_PATH_STRONG_LOCK_HASH_BITS 10
301 : #define FAST_PATH_STRONG_LOCK_HASH_PARTITIONS \
302 : (1 << FAST_PATH_STRONG_LOCK_HASH_BITS)
303 : #define FastPathStrongLockHashPartition(hashcode) \
304 : ((hashcode) % FAST_PATH_STRONG_LOCK_HASH_PARTITIONS)
305 :
306 : typedef struct
307 : {
308 : slock_t mutex;
309 : uint32 count[FAST_PATH_STRONG_LOCK_HASH_PARTITIONS];
310 : } FastPathStrongRelationLockData;
311 :
312 : static volatile FastPathStrongRelationLockData *FastPathStrongRelationLocks;
313 :
314 :
315 : /*
316 : * Pointers to hash tables containing lock state
317 : *
318 : * The LockMethodLockHash and LockMethodProcLockHash hash tables are in
319 : * shared memory; LockMethodLocalHash is local to each backend.
320 : */
321 : static HTAB *LockMethodLockHash;
322 : static HTAB *LockMethodProcLockHash;
323 : static HTAB *LockMethodLocalHash;
324 :
325 :
326 : /* private state for error cleanup */
327 : static LOCALLOCK *StrongLockInProgress;
328 : static LOCALLOCK *awaitedLock;
329 : static ResourceOwner awaitedOwner;
330 :
331 :
332 : #ifdef LOCK_DEBUG
333 :
334 : /*------
335 : * The following configuration options are available for lock debugging:
336 : *
337 : * TRACE_LOCKS -- give a bunch of output what's going on in this file
338 : * TRACE_USERLOCKS -- same but for user locks
339 : * TRACE_LOCK_OIDMIN-- do not trace locks for tables below this oid
340 : * (use to avoid output on system tables)
341 : * TRACE_LOCK_TABLE -- trace locks on this table (oid) unconditionally
342 : * DEBUG_DEADLOCKS -- currently dumps locks at untimely occasions ;)
343 : *
344 : * Furthermore, but in storage/lmgr/lwlock.c:
345 : * TRACE_LWLOCKS -- trace lightweight locks (pretty useless)
346 : *
347 : * Define LOCK_DEBUG at compile time to get all these enabled.
348 : * --------
349 : */
350 :
351 : int Trace_lock_oidmin = FirstNormalObjectId;
352 : bool Trace_locks = false;
353 : bool Trace_userlocks = false;
354 : int Trace_lock_table = 0;
355 : bool Debug_deadlocks = false;
356 :
357 :
358 : inline static bool
359 : LOCK_DEBUG_ENABLED(const LOCKTAG *tag)
360 : {
361 : return
362 : (*(LockMethods[tag->locktag_lockmethodid]->trace_flag) &&
363 : ((Oid) tag->locktag_field2 >= (Oid) Trace_lock_oidmin))
364 : || (Trace_lock_table &&
365 : (tag->locktag_field2 == Trace_lock_table));
366 : }
367 :
368 :
369 : inline static void
370 : LOCK_PRINT(const char *where, const LOCK *lock, LOCKMODE type)
371 : {
372 : if (LOCK_DEBUG_ENABLED(&lock->tag))
373 : elog(LOG,
374 : "%s: lock(%p) id(%u,%u,%u,%u,%u,%u) grantMask(%x) "
375 : "req(%d,%d,%d,%d,%d,%d,%d)=%d "
376 : "grant(%d,%d,%d,%d,%d,%d,%d)=%d wait(%d) type(%s)",
377 : where, lock,
378 : lock->tag.locktag_field1, lock->tag.locktag_field2,
379 : lock->tag.locktag_field3, lock->tag.locktag_field4,
380 : lock->tag.locktag_type, lock->tag.locktag_lockmethodid,
381 : lock->grantMask,
382 : lock->requested[1], lock->requested[2], lock->requested[3],
383 : lock->requested[4], lock->requested[5], lock->requested[6],
384 : lock->requested[7], lock->nRequested,
385 : lock->granted[1], lock->granted[2], lock->granted[3],
386 : lock->granted[4], lock->granted[5], lock->granted[6],
387 : lock->granted[7], lock->nGranted,
388 : dclist_count(&lock->waitProcs),
389 : LockMethods[LOCK_LOCKMETHOD(*lock)]->lockModeNames[type]);
390 : }
391 :
392 :
393 : inline static void
394 : PROCLOCK_PRINT(const char *where, const PROCLOCK *proclockP)
395 : {
396 : if (LOCK_DEBUG_ENABLED(&proclockP->tag.myLock->tag))
397 : elog(LOG,
398 : "%s: proclock(%p) lock(%p) method(%u) proc(%p) hold(%x)",
399 : where, proclockP, proclockP->tag.myLock,
400 : PROCLOCK_LOCKMETHOD(*(proclockP)),
401 : proclockP->tag.myProc, (int) proclockP->holdMask);
402 : }
403 : #else /* not LOCK_DEBUG */
404 :
405 : #define LOCK_PRINT(where, lock, type) ((void) 0)
406 : #define PROCLOCK_PRINT(where, proclockP) ((void) 0)
407 : #endif /* not LOCK_DEBUG */
408 :
409 :
410 : static uint32 proclock_hash(const void *key, Size keysize);
411 : static void RemoveLocalLock(LOCALLOCK *locallock);
412 : static PROCLOCK *SetupLockInTable(LockMethod lockMethodTable, PGPROC *proc,
413 : const LOCKTAG *locktag, uint32 hashcode, LOCKMODE lockmode);
414 : static void GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner);
415 : static void BeginStrongLockAcquire(LOCALLOCK *locallock, uint32 fasthashcode);
416 : static void FinishStrongLockAcquire(void);
417 : static ProcWaitStatus WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner);
418 : static void waitonlock_error_callback(void *arg);
419 : static void ReleaseLockIfHeld(LOCALLOCK *locallock, bool sessionLock);
420 : static void LockReassignOwner(LOCALLOCK *locallock, ResourceOwner parent);
421 : static bool UnGrantLock(LOCK *lock, LOCKMODE lockmode,
422 : PROCLOCK *proclock, LockMethod lockMethodTable);
423 : static void CleanUpLock(LOCK *lock, PROCLOCK *proclock,
424 : LockMethod lockMethodTable, uint32 hashcode,
425 : bool wakeupNeeded);
426 : static void LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
427 : LOCKTAG *locktag, LOCKMODE lockmode,
428 : bool decrement_strong_lock_count);
429 : static void GetSingleProcBlockerStatusData(PGPROC *blocked_proc,
430 : BlockedProcsData *data);
431 :
432 :
433 : /*
434 : * Initialize the lock manager's shmem data structures.
435 : *
436 : * This is called from CreateSharedMemoryAndSemaphores(), which see for more
437 : * comments. In the normal postmaster case, the shared hash tables are
438 : * created here, and backends inherit pointers to them via fork(). In the
439 : * EXEC_BACKEND case, each backend re-executes this code to obtain pointers to
440 : * the already existing shared hash tables. In either case, each backend must
441 : * also call InitLockManagerAccess() to create the locallock hash table.
442 : */
443 : void
444 2194 : LockManagerShmemInit(void)
445 : {
446 : HASHCTL info;
447 : int64 init_table_size,
448 : max_table_size;
449 : bool found;
450 :
451 : /*
452 : * Compute init/max size to request for lock hashtables. Note these
453 : * calculations must agree with LockManagerShmemSize!
454 : */
455 2194 : max_table_size = NLOCKENTS();
456 2194 : init_table_size = max_table_size / 2;
457 :
458 : /*
459 : * Allocate hash table for LOCK structs. This stores per-locked-object
460 : * information.
461 : */
462 2194 : info.keysize = sizeof(LOCKTAG);
463 2194 : info.entrysize = sizeof(LOCK);
464 2194 : info.num_partitions = NUM_LOCK_PARTITIONS;
465 :
466 2194 : LockMethodLockHash = ShmemInitHash("LOCK hash",
467 : init_table_size,
468 : max_table_size,
469 : &info,
470 : HASH_ELEM | HASH_BLOBS | HASH_PARTITION);
471 :
472 : /* Assume an average of 2 holders per lock */
473 2194 : max_table_size *= 2;
474 2194 : init_table_size *= 2;
475 :
476 : /*
477 : * Allocate hash table for PROCLOCK structs. This stores
478 : * per-lock-per-holder information.
479 : */
480 2194 : info.keysize = sizeof(PROCLOCKTAG);
481 2194 : info.entrysize = sizeof(PROCLOCK);
482 2194 : info.hash = proclock_hash;
483 2194 : info.num_partitions = NUM_LOCK_PARTITIONS;
484 :
485 2194 : LockMethodProcLockHash = ShmemInitHash("PROCLOCK hash",
486 : init_table_size,
487 : max_table_size,
488 : &info,
489 : HASH_ELEM | HASH_FUNCTION | HASH_PARTITION);
490 :
491 : /*
492 : * Allocate fast-path structures.
493 : */
494 2194 : FastPathStrongRelationLocks =
495 2194 : ShmemInitStruct("Fast Path Strong Relation Lock Data",
496 : sizeof(FastPathStrongRelationLockData), &found);
497 2194 : if (!found)
498 2194 : SpinLockInit(&FastPathStrongRelationLocks->mutex);
499 2194 : }
500 :
501 : /*
502 : * Initialize the lock manager's backend-private data structures.
503 : */
504 : void
505 44916 : InitLockManagerAccess(void)
506 : {
507 : /*
508 : * Allocate non-shared hash table for LOCALLOCK structs. This stores lock
509 : * counts and resource owner information.
510 : */
511 : HASHCTL info;
512 :
513 44916 : info.keysize = sizeof(LOCALLOCKTAG);
514 44916 : info.entrysize = sizeof(LOCALLOCK);
515 :
516 44916 : LockMethodLocalHash = hash_create("LOCALLOCK hash",
517 : 16,
518 : &info,
519 : HASH_ELEM | HASH_BLOBS);
520 44916 : }
521 :
522 :
523 : /*
524 : * Fetch the lock method table associated with a given lock
525 : */
526 : LockMethod
527 210 : GetLocksMethodTable(const LOCK *lock)
528 : {
529 210 : LOCKMETHODID lockmethodid = LOCK_LOCKMETHOD(*lock);
530 :
531 : Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
532 210 : return LockMethods[lockmethodid];
533 : }
534 :
535 : /*
536 : * Fetch the lock method table associated with a given locktag
537 : */
538 : LockMethod
539 2332 : GetLockTagsMethodTable(const LOCKTAG *locktag)
540 : {
541 2332 : LOCKMETHODID lockmethodid = (LOCKMETHODID) locktag->locktag_lockmethodid;
542 :
543 : Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
544 2332 : return LockMethods[lockmethodid];
545 : }
546 :
547 :
548 : /*
549 : * Compute the hash code associated with a LOCKTAG.
550 : *
551 : * To avoid unnecessary recomputations of the hash code, we try to do this
552 : * just once per function, and then pass it around as needed. Aside from
553 : * passing the hashcode to hash_search_with_hash_value(), we can extract
554 : * the lock partition number from the hashcode.
555 : */
556 : uint32
557 40256992 : LockTagHashCode(const LOCKTAG *locktag)
558 : {
559 40256992 : return get_hash_value(LockMethodLockHash, locktag);
560 : }
561 :
562 : /*
563 : * Compute the hash code associated with a PROCLOCKTAG.
564 : *
565 : * Because we want to use just one set of partition locks for both the
566 : * LOCK and PROCLOCK hash tables, we have to make sure that PROCLOCKs
567 : * fall into the same partition number as their associated LOCKs.
568 : * dynahash.c expects the partition number to be the low-order bits of
569 : * the hash code, and therefore a PROCLOCKTAG's hash code must have the
570 : * same low-order bits as the associated LOCKTAG's hash code. We achieve
571 : * this with this specialized hash function.
572 : */
573 : static uint32
574 1314 : proclock_hash(const void *key, Size keysize)
575 : {
576 1314 : const PROCLOCKTAG *proclocktag = (const PROCLOCKTAG *) key;
577 : uint32 lockhash;
578 : Datum procptr;
579 :
580 : Assert(keysize == sizeof(PROCLOCKTAG));
581 :
582 : /* Look into the associated LOCK object, and compute its hash code */
583 1314 : lockhash = LockTagHashCode(&proclocktag->myLock->tag);
584 :
585 : /*
586 : * To make the hash code also depend on the PGPROC, we xor the proc
587 : * struct's address into the hash code, left-shifted so that the
588 : * partition-number bits don't change. Since this is only a hash, we
589 : * don't care if we lose high-order bits of the address; use an
590 : * intermediate variable to suppress cast-pointer-to-int warnings.
591 : */
592 1314 : procptr = PointerGetDatum(proclocktag->myProc);
593 1314 : lockhash ^= DatumGetUInt32(procptr) << LOG2_NUM_LOCK_PARTITIONS;
594 :
595 1314 : return lockhash;
596 : }
597 :
598 : /*
599 : * Compute the hash code associated with a PROCLOCKTAG, given the hashcode
600 : * for its underlying LOCK.
601 : *
602 : * We use this just to avoid redundant calls of LockTagHashCode().
603 : */
604 : static inline uint32
605 9482384 : ProcLockHashCode(const PROCLOCKTAG *proclocktag, uint32 hashcode)
606 : {
607 9482384 : uint32 lockhash = hashcode;
608 : Datum procptr;
609 :
610 : /*
611 : * This must match proclock_hash()!
612 : */
613 9482384 : procptr = PointerGetDatum(proclocktag->myProc);
614 9482384 : lockhash ^= DatumGetUInt32(procptr) << LOG2_NUM_LOCK_PARTITIONS;
615 :
616 9482384 : return lockhash;
617 : }
618 :
619 : /*
620 : * Given two lock modes, return whether they would conflict.
621 : */
622 : bool
623 488 : DoLockModesConflict(LOCKMODE mode1, LOCKMODE mode2)
624 : {
625 488 : LockMethod lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
626 :
627 488 : if (lockMethodTable->conflictTab[mode1] & LOCKBIT_ON(mode2))
628 284 : return true;
629 :
630 204 : return false;
631 : }
632 :
633 : /*
634 : * LockHeldByMe -- test whether lock 'locktag' is held by the current
635 : * transaction
636 : *
637 : * Returns true if current transaction holds a lock on 'tag' of mode
638 : * 'lockmode'. If 'orstronger' is true, a stronger lockmode is also OK.
639 : * ("Stronger" is defined as "numerically higher", which is a bit
640 : * semantically dubious but is OK for the purposes we use this for.)
641 : */
642 : bool
643 0 : LockHeldByMe(const LOCKTAG *locktag,
644 : LOCKMODE lockmode, bool orstronger)
645 : {
646 : LOCALLOCKTAG localtag;
647 : LOCALLOCK *locallock;
648 :
649 : /*
650 : * See if there is a LOCALLOCK entry for this lock and lockmode
651 : */
652 0 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
653 0 : localtag.lock = *locktag;
654 0 : localtag.mode = lockmode;
655 :
656 0 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
657 : &localtag,
658 : HASH_FIND, NULL);
659 :
660 0 : if (locallock && locallock->nLocks > 0)
661 0 : return true;
662 :
663 0 : if (orstronger)
664 : {
665 : LOCKMODE slockmode;
666 :
667 0 : for (slockmode = lockmode + 1;
668 0 : slockmode <= MaxLockMode;
669 0 : slockmode++)
670 : {
671 0 : if (LockHeldByMe(locktag, slockmode, false))
672 0 : return true;
673 : }
674 : }
675 :
676 0 : return false;
677 : }
678 :
679 : #ifdef USE_ASSERT_CHECKING
680 : /*
681 : * GetLockMethodLocalHash -- return the hash of local locks, for modules that
682 : * evaluate assertions based on all locks held.
683 : */
684 : HTAB *
685 : GetLockMethodLocalHash(void)
686 : {
687 : return LockMethodLocalHash;
688 : }
689 : #endif
690 :
691 : /*
692 : * LockHasWaiters -- look up 'locktag' and check if releasing this
693 : * lock would wake up other processes waiting for it.
694 : */
695 : bool
696 0 : LockHasWaiters(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
697 : {
698 0 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
699 : LockMethod lockMethodTable;
700 : LOCALLOCKTAG localtag;
701 : LOCALLOCK *locallock;
702 : LOCK *lock;
703 : PROCLOCK *proclock;
704 : LWLock *partitionLock;
705 0 : bool hasWaiters = false;
706 :
707 0 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
708 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
709 0 : lockMethodTable = LockMethods[lockmethodid];
710 0 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
711 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
712 :
713 : #ifdef LOCK_DEBUG
714 : if (LOCK_DEBUG_ENABLED(locktag))
715 : elog(LOG, "LockHasWaiters: lock [%u,%u] %s",
716 : locktag->locktag_field1, locktag->locktag_field2,
717 : lockMethodTable->lockModeNames[lockmode]);
718 : #endif
719 :
720 : /*
721 : * Find the LOCALLOCK entry for this lock and lockmode
722 : */
723 0 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
724 0 : localtag.lock = *locktag;
725 0 : localtag.mode = lockmode;
726 :
727 0 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
728 : &localtag,
729 : HASH_FIND, NULL);
730 :
731 : /*
732 : * let the caller print its own error message, too. Do not ereport(ERROR).
733 : */
734 0 : if (!locallock || locallock->nLocks <= 0)
735 : {
736 0 : elog(WARNING, "you don't own a lock of type %s",
737 : lockMethodTable->lockModeNames[lockmode]);
738 0 : return false;
739 : }
740 :
741 : /*
742 : * Check the shared lock table.
743 : */
744 0 : partitionLock = LockHashPartitionLock(locallock->hashcode);
745 :
746 0 : LWLockAcquire(partitionLock, LW_SHARED);
747 :
748 : /*
749 : * We don't need to re-find the lock or proclock, since we kept their
750 : * addresses in the locallock table, and they couldn't have been removed
751 : * while we were holding a lock on them.
752 : */
753 0 : lock = locallock->lock;
754 : LOCK_PRINT("LockHasWaiters: found", lock, lockmode);
755 0 : proclock = locallock->proclock;
756 : PROCLOCK_PRINT("LockHasWaiters: found", proclock);
757 :
758 : /*
759 : * Double-check that we are actually holding a lock of the type we want to
760 : * release.
761 : */
762 0 : if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
763 : {
764 : PROCLOCK_PRINT("LockHasWaiters: WRONGTYPE", proclock);
765 0 : LWLockRelease(partitionLock);
766 0 : elog(WARNING, "you don't own a lock of type %s",
767 : lockMethodTable->lockModeNames[lockmode]);
768 0 : RemoveLocalLock(locallock);
769 0 : return false;
770 : }
771 :
772 : /*
773 : * Do the checking.
774 : */
775 0 : if ((lockMethodTable->conflictTab[lockmode] & lock->waitMask) != 0)
776 0 : hasWaiters = true;
777 :
778 0 : LWLockRelease(partitionLock);
779 :
780 0 : return hasWaiters;
781 : }
782 :
783 : /*
784 : * LockAcquire -- Check for lock conflicts, sleep if conflict found,
785 : * set lock if/when no conflicts.
786 : *
787 : * Inputs:
788 : * locktag: unique identifier for the lockable object
789 : * lockmode: lock mode to acquire
790 : * sessionLock: if true, acquire lock for session not current transaction
791 : * dontWait: if true, don't wait to acquire lock
792 : *
793 : * Returns one of:
794 : * LOCKACQUIRE_NOT_AVAIL lock not available, and dontWait=true
795 : * LOCKACQUIRE_OK lock successfully acquired
796 : * LOCKACQUIRE_ALREADY_HELD incremented count for lock already held
797 : * LOCKACQUIRE_ALREADY_CLEAR incremented count for lock already clear
798 : *
799 : * In the normal case where dontWait=false and the caller doesn't need to
800 : * distinguish a freshly acquired lock from one already taken earlier in
801 : * this same transaction, there is no need to examine the return value.
802 : *
803 : * Side Effects: The lock is acquired and recorded in lock tables.
804 : *
805 : * NOTE: if we wait for the lock, there is no way to abort the wait
806 : * short of aborting the transaction.
807 : */
808 : LockAcquireResult
809 1870952 : LockAcquire(const LOCKTAG *locktag,
810 : LOCKMODE lockmode,
811 : bool sessionLock,
812 : bool dontWait)
813 : {
814 1870952 : return LockAcquireExtended(locktag, lockmode, sessionLock, dontWait,
815 : true, NULL, false);
816 : }
817 :
818 : /*
819 : * LockAcquireExtended - allows us to specify additional options
820 : *
821 : * reportMemoryError specifies whether a lock request that fills the lock
822 : * table should generate an ERROR or not. Passing "false" allows the caller
823 : * to attempt to recover from lock-table-full situations, perhaps by forcibly
824 : * canceling other lock holders and then retrying. Note, however, that the
825 : * return code for that is LOCKACQUIRE_NOT_AVAIL, so that it's unsafe to use
826 : * in combination with dontWait = true, as the cause of failure couldn't be
827 : * distinguished.
828 : *
829 : * If locallockp isn't NULL, *locallockp receives a pointer to the LOCALLOCK
830 : * table entry if a lock is successfully acquired, or NULL if not.
831 : *
832 : * logLockFailure indicates whether to log details when a lock acquisition
833 : * fails with dontWait = true.
834 : */
835 : LockAcquireResult
836 43431834 : LockAcquireExtended(const LOCKTAG *locktag,
837 : LOCKMODE lockmode,
838 : bool sessionLock,
839 : bool dontWait,
840 : bool reportMemoryError,
841 : LOCALLOCK **locallockp,
842 : bool logLockFailure)
843 : {
844 43431834 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
845 : LockMethod lockMethodTable;
846 : LOCALLOCKTAG localtag;
847 : LOCALLOCK *locallock;
848 : LOCK *lock;
849 : PROCLOCK *proclock;
850 : bool found;
851 : ResourceOwner owner;
852 : uint32 hashcode;
853 : LWLock *partitionLock;
854 : bool found_conflict;
855 : ProcWaitStatus waitResult;
856 43431834 : bool log_lock = false;
857 :
858 43431834 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
859 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
860 43431834 : lockMethodTable = LockMethods[lockmethodid];
861 43431834 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
862 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
863 :
864 43431834 : if (RecoveryInProgress() && !InRecovery &&
865 570194 : (locktag->locktag_type == LOCKTAG_OBJECT ||
866 570194 : locktag->locktag_type == LOCKTAG_RELATION) &&
867 : lockmode > RowExclusiveLock)
868 0 : ereport(ERROR,
869 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
870 : errmsg("cannot acquire lock mode %s on database objects while recovery is in progress",
871 : lockMethodTable->lockModeNames[lockmode]),
872 : errhint("Only RowExclusiveLock or less can be acquired on database objects during recovery.")));
873 :
874 : #ifdef LOCK_DEBUG
875 : if (LOCK_DEBUG_ENABLED(locktag))
876 : elog(LOG, "LockAcquire: lock [%u,%u] %s",
877 : locktag->locktag_field1, locktag->locktag_field2,
878 : lockMethodTable->lockModeNames[lockmode]);
879 : #endif
880 :
881 : /* Identify owner for lock */
882 43431834 : if (sessionLock)
883 325706 : owner = NULL;
884 : else
885 43106128 : owner = CurrentResourceOwner;
886 :
887 : /*
888 : * Find or create a LOCALLOCK entry for this lock and lockmode
889 : */
890 43431834 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
891 43431834 : localtag.lock = *locktag;
892 43431834 : localtag.mode = lockmode;
893 :
894 43431834 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
895 : &localtag,
896 : HASH_ENTER, &found);
897 :
898 : /*
899 : * if it's a new locallock object, initialize it
900 : */
901 43431834 : if (!found)
902 : {
903 39073536 : locallock->lock = NULL;
904 39073536 : locallock->proclock = NULL;
905 39073536 : locallock->hashcode = LockTagHashCode(&(localtag.lock));
906 39073536 : locallock->nLocks = 0;
907 39073536 : locallock->holdsStrongLockCount = false;
908 39073536 : locallock->lockCleared = false;
909 39073536 : locallock->numLockOwners = 0;
910 39073536 : locallock->maxLockOwners = 8;
911 39073536 : locallock->lockOwners = NULL; /* in case next line fails */
912 39073536 : locallock->lockOwners = (LOCALLOCKOWNER *)
913 39073536 : MemoryContextAlloc(TopMemoryContext,
914 39073536 : locallock->maxLockOwners * sizeof(LOCALLOCKOWNER));
915 : }
916 : else
917 : {
918 : /* Make sure there will be room to remember the lock */
919 4358298 : if (locallock->numLockOwners >= locallock->maxLockOwners)
920 : {
921 38 : int newsize = locallock->maxLockOwners * 2;
922 :
923 38 : locallock->lockOwners = (LOCALLOCKOWNER *)
924 38 : repalloc(locallock->lockOwners,
925 : newsize * sizeof(LOCALLOCKOWNER));
926 38 : locallock->maxLockOwners = newsize;
927 : }
928 : }
929 43431834 : hashcode = locallock->hashcode;
930 :
931 43431834 : if (locallockp)
932 41560710 : *locallockp = locallock;
933 :
934 : /*
935 : * If we already hold the lock, we can just increase the count locally.
936 : *
937 : * If lockCleared is already set, caller need not worry about absorbing
938 : * sinval messages related to the lock's object.
939 : */
940 43431834 : if (locallock->nLocks > 0)
941 : {
942 4358298 : GrantLockLocal(locallock, owner);
943 4358298 : if (locallock->lockCleared)
944 4202400 : return LOCKACQUIRE_ALREADY_CLEAR;
945 : else
946 155898 : return LOCKACQUIRE_ALREADY_HELD;
947 : }
948 :
949 : /*
950 : * We don't acquire any other heavyweight lock while holding the relation
951 : * extension lock. We do allow to acquire the same relation extension
952 : * lock more than once but that case won't reach here.
953 : */
954 : Assert(!IsRelationExtensionLockHeld);
955 :
956 : /*
957 : * Prepare to emit a WAL record if acquisition of this lock needs to be
958 : * replayed in a standby server.
959 : *
960 : * Here we prepare to log; after lock is acquired we'll issue log record.
961 : * This arrangement simplifies error recovery in case the preparation step
962 : * fails.
963 : *
964 : * Only AccessExclusiveLocks can conflict with lock types that read-only
965 : * transactions can acquire in a standby server. Make sure this definition
966 : * matches the one in GetRunningTransactionLocks().
967 : */
968 39073536 : if (lockmode >= AccessExclusiveLock &&
969 473954 : locktag->locktag_type == LOCKTAG_RELATION &&
970 316816 : !RecoveryInProgress() &&
971 267902 : XLogStandbyInfoActive())
972 : {
973 205258 : LogAccessExclusiveLockPrepare();
974 205258 : log_lock = true;
975 : }
976 :
977 : /*
978 : * Attempt to take lock via fast path, if eligible. But if we remember
979 : * having filled up the fast path array, we don't attempt to make any
980 : * further use of it until we release some locks. It's possible that some
981 : * other backend has transferred some of those locks to the shared hash
982 : * table, leaving space free, but it's not worth acquiring the LWLock just
983 : * to check. It's also possible that we're acquiring a second or third
984 : * lock type on a relation we have already locked using the fast-path, but
985 : * for now we don't worry about that case either.
986 : */
987 39073536 : if (EligibleForRelationFastPath(locktag, lockmode) &&
988 35035142 : FastPathLocalUseCounts[FAST_PATH_REL_GROUP(locktag->locktag_field2)] < FP_LOCK_SLOTS_PER_GROUP)
989 : {
990 34679252 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
991 : bool acquired;
992 :
993 : /*
994 : * LWLockAcquire acts as a memory sequencing point, so it's safe to
995 : * assume that any strong locker whose increment to
996 : * FastPathStrongRelationLocks->counts becomes visible after we test
997 : * it has yet to begin to transfer fast-path locks.
998 : */
999 34679252 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
1000 34679252 : if (FastPathStrongRelationLocks->count[fasthashcode] != 0)
1001 559142 : acquired = false;
1002 : else
1003 34120110 : acquired = FastPathGrantRelationLock(locktag->locktag_field2,
1004 : lockmode);
1005 34679252 : LWLockRelease(&MyProc->fpInfoLock);
1006 34679252 : if (acquired)
1007 : {
1008 : /*
1009 : * The locallock might contain stale pointers to some old shared
1010 : * objects; we MUST reset these to null before considering the
1011 : * lock to be acquired via fast-path.
1012 : */
1013 34120110 : locallock->lock = NULL;
1014 34120110 : locallock->proclock = NULL;
1015 34120110 : GrantLockLocal(locallock, owner);
1016 34120110 : return LOCKACQUIRE_OK;
1017 : }
1018 : }
1019 :
1020 : /*
1021 : * If this lock could potentially have been taken via the fast-path by
1022 : * some other backend, we must (temporarily) disable further use of the
1023 : * fast-path for this lock tag, and migrate any locks already taken via
1024 : * this method to the main lock table.
1025 : */
1026 4953426 : if (ConflictsWithRelationFastPath(locktag, lockmode))
1027 : {
1028 375308 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
1029 :
1030 375308 : BeginStrongLockAcquire(locallock, fasthashcode);
1031 375308 : if (!FastPathTransferRelationLocks(lockMethodTable, locktag,
1032 : hashcode))
1033 : {
1034 0 : AbortStrongLockAcquire();
1035 0 : if (locallock->nLocks == 0)
1036 0 : RemoveLocalLock(locallock);
1037 0 : if (locallockp)
1038 0 : *locallockp = NULL;
1039 0 : if (reportMemoryError)
1040 0 : ereport(ERROR,
1041 : (errcode(ERRCODE_OUT_OF_MEMORY),
1042 : errmsg("out of shared memory"),
1043 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
1044 : else
1045 0 : return LOCKACQUIRE_NOT_AVAIL;
1046 : }
1047 : }
1048 :
1049 : /*
1050 : * We didn't find the lock in our LOCALLOCK table, and we didn't manage to
1051 : * take it via the fast-path, either, so we've got to mess with the shared
1052 : * lock table.
1053 : */
1054 4953426 : partitionLock = LockHashPartitionLock(hashcode);
1055 :
1056 4953426 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
1057 :
1058 : /*
1059 : * Find or create lock and proclock entries with this tag
1060 : *
1061 : * Note: if the locallock object already existed, it might have a pointer
1062 : * to the lock already ... but we should not assume that that pointer is
1063 : * valid, since a lock object with zero hold and request counts can go
1064 : * away anytime. So we have to use SetupLockInTable() to recompute the
1065 : * lock and proclock pointers, even if they're already set.
1066 : */
1067 4953426 : proclock = SetupLockInTable(lockMethodTable, MyProc, locktag,
1068 : hashcode, lockmode);
1069 4953426 : if (!proclock)
1070 : {
1071 0 : AbortStrongLockAcquire();
1072 0 : LWLockRelease(partitionLock);
1073 0 : if (locallock->nLocks == 0)
1074 0 : RemoveLocalLock(locallock);
1075 0 : if (locallockp)
1076 0 : *locallockp = NULL;
1077 0 : if (reportMemoryError)
1078 0 : ereport(ERROR,
1079 : (errcode(ERRCODE_OUT_OF_MEMORY),
1080 : errmsg("out of shared memory"),
1081 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
1082 : else
1083 0 : return LOCKACQUIRE_NOT_AVAIL;
1084 : }
1085 4953426 : locallock->proclock = proclock;
1086 4953426 : lock = proclock->tag.myLock;
1087 4953426 : locallock->lock = lock;
1088 :
1089 : /*
1090 : * If lock requested conflicts with locks requested by waiters, must join
1091 : * wait queue. Otherwise, check for conflict with already-held locks.
1092 : * (That's last because most complex check.)
1093 : */
1094 4953426 : if (lockMethodTable->conflictTab[lockmode] & lock->waitMask)
1095 382 : found_conflict = true;
1096 : else
1097 4953044 : found_conflict = LockCheckConflicts(lockMethodTable, lockmode,
1098 : lock, proclock);
1099 :
1100 4953426 : if (!found_conflict)
1101 : {
1102 : /* No conflict with held or previously requested locks */
1103 4949462 : GrantLock(lock, proclock, lockmode);
1104 4949462 : waitResult = PROC_WAIT_STATUS_OK;
1105 : }
1106 : else
1107 : {
1108 : /*
1109 : * Join the lock's wait queue. We call this even in the dontWait
1110 : * case, because JoinWaitQueue() may discover that we can acquire the
1111 : * lock immediately after all.
1112 : */
1113 3964 : waitResult = JoinWaitQueue(locallock, lockMethodTable, dontWait);
1114 : }
1115 :
1116 4953426 : if (waitResult == PROC_WAIT_STATUS_ERROR)
1117 : {
1118 : /*
1119 : * We're not getting the lock because a deadlock was detected already
1120 : * while trying to join the wait queue, or because we would have to
1121 : * wait but the caller requested no blocking.
1122 : *
1123 : * Undo the changes to shared entries before releasing the partition
1124 : * lock.
1125 : */
1126 1470 : AbortStrongLockAcquire();
1127 :
1128 1470 : if (proclock->holdMask == 0)
1129 : {
1130 : uint32 proclock_hashcode;
1131 :
1132 1064 : proclock_hashcode = ProcLockHashCode(&proclock->tag,
1133 : hashcode);
1134 1064 : dlist_delete(&proclock->lockLink);
1135 1064 : dlist_delete(&proclock->procLink);
1136 1064 : if (!hash_search_with_hash_value(LockMethodProcLockHash,
1137 1064 : &(proclock->tag),
1138 : proclock_hashcode,
1139 : HASH_REMOVE,
1140 : NULL))
1141 0 : elog(PANIC, "proclock table corrupted");
1142 : }
1143 : else
1144 : PROCLOCK_PRINT("LockAcquire: did not join wait queue", proclock);
1145 1470 : lock->nRequested--;
1146 1470 : lock->requested[lockmode]--;
1147 : LOCK_PRINT("LockAcquire: did not join wait queue",
1148 : lock, lockmode);
1149 : Assert((lock->nRequested > 0) &&
1150 : (lock->requested[lockmode] >= 0));
1151 : Assert(lock->nGranted <= lock->nRequested);
1152 1470 : LWLockRelease(partitionLock);
1153 1470 : if (locallock->nLocks == 0)
1154 1470 : RemoveLocalLock(locallock);
1155 :
1156 1470 : if (dontWait)
1157 : {
1158 : /*
1159 : * Log lock holders and waiters as a detail log message if
1160 : * logLockFailure = true and lock acquisition fails with dontWait
1161 : * = true
1162 : */
1163 1468 : if (logLockFailure)
1164 : {
1165 : StringInfoData buf,
1166 : lock_waiters_sbuf,
1167 : lock_holders_sbuf;
1168 : const char *modename;
1169 0 : int lockHoldersNum = 0;
1170 :
1171 0 : initStringInfo(&buf);
1172 0 : initStringInfo(&lock_waiters_sbuf);
1173 0 : initStringInfo(&lock_holders_sbuf);
1174 :
1175 0 : DescribeLockTag(&buf, &locallock->tag.lock);
1176 0 : modename = GetLockmodeName(locallock->tag.lock.locktag_lockmethodid,
1177 : lockmode);
1178 :
1179 : /* Gather a list of all lock holders and waiters */
1180 0 : LWLockAcquire(partitionLock, LW_SHARED);
1181 0 : GetLockHoldersAndWaiters(locallock, &lock_holders_sbuf,
1182 : &lock_waiters_sbuf, &lockHoldersNum);
1183 0 : LWLockRelease(partitionLock);
1184 :
1185 0 : ereport(LOG,
1186 : (errmsg("process %d could not obtain %s on %s",
1187 : MyProcPid, modename, buf.data),
1188 : errdetail_log_plural(
1189 : "Process holding the lock: %s, Wait queue: %s.",
1190 : "Processes holding the lock: %s, Wait queue: %s.",
1191 : lockHoldersNum,
1192 : lock_holders_sbuf.data,
1193 : lock_waiters_sbuf.data)));
1194 :
1195 0 : pfree(buf.data);
1196 0 : pfree(lock_holders_sbuf.data);
1197 0 : pfree(lock_waiters_sbuf.data);
1198 : }
1199 1468 : if (locallockp)
1200 442 : *locallockp = NULL;
1201 1468 : return LOCKACQUIRE_NOT_AVAIL;
1202 : }
1203 : else
1204 : {
1205 2 : DeadLockReport();
1206 : /* DeadLockReport() will not return */
1207 : }
1208 : }
1209 :
1210 : /*
1211 : * We are now in the lock queue, or the lock was already granted. If
1212 : * queued, go to sleep.
1213 : */
1214 4951956 : if (waitResult == PROC_WAIT_STATUS_WAITING)
1215 : {
1216 : Assert(!dontWait);
1217 : PROCLOCK_PRINT("LockAcquire: sleeping on lock", proclock);
1218 : LOCK_PRINT("LockAcquire: sleeping on lock", lock, lockmode);
1219 2484 : LWLockRelease(partitionLock);
1220 :
1221 2484 : waitResult = WaitOnLock(locallock, owner);
1222 :
1223 : /*
1224 : * NOTE: do not do any material change of state between here and
1225 : * return. All required changes in locktable state must have been
1226 : * done when the lock was granted to us --- see notes in WaitOnLock.
1227 : */
1228 :
1229 2404 : if (waitResult == PROC_WAIT_STATUS_ERROR)
1230 : {
1231 : /*
1232 : * We failed as a result of a deadlock, see CheckDeadLock(). Quit
1233 : * now.
1234 : */
1235 : Assert(!dontWait);
1236 10 : DeadLockReport();
1237 : /* DeadLockReport() will not return */
1238 : }
1239 : }
1240 : else
1241 4949472 : LWLockRelease(partitionLock);
1242 : Assert(waitResult == PROC_WAIT_STATUS_OK);
1243 :
1244 : /* The lock was granted to us. Update the local lock entry accordingly */
1245 : Assert((proclock->holdMask & LOCKBIT_ON(lockmode)) != 0);
1246 4951866 : GrantLockLocal(locallock, owner);
1247 :
1248 : /*
1249 : * Lock state is fully up-to-date now; if we error out after this, no
1250 : * special error cleanup is required.
1251 : */
1252 4951866 : FinishStrongLockAcquire();
1253 :
1254 : /*
1255 : * Emit a WAL record if acquisition of this lock needs to be replayed in a
1256 : * standby server.
1257 : */
1258 4951866 : if (log_lock)
1259 : {
1260 : /*
1261 : * Decode the locktag back to the original values, to avoid sending
1262 : * lots of empty bytes with every message. See lock.h to check how a
1263 : * locktag is defined for LOCKTAG_RELATION
1264 : */
1265 204832 : LogAccessExclusiveLock(locktag->locktag_field1,
1266 204832 : locktag->locktag_field2);
1267 : }
1268 :
1269 4951866 : return LOCKACQUIRE_OK;
1270 : }
1271 :
1272 : /*
1273 : * Find or create LOCK and PROCLOCK objects as needed for a new lock
1274 : * request.
1275 : *
1276 : * Returns the PROCLOCK object, or NULL if we failed to create the objects
1277 : * for lack of shared memory.
1278 : *
1279 : * The appropriate partition lock must be held at entry, and will be
1280 : * held at exit.
1281 : */
1282 : static PROCLOCK *
1283 4956768 : SetupLockInTable(LockMethod lockMethodTable, PGPROC *proc,
1284 : const LOCKTAG *locktag, uint32 hashcode, LOCKMODE lockmode)
1285 : {
1286 : LOCK *lock;
1287 : PROCLOCK *proclock;
1288 : PROCLOCKTAG proclocktag;
1289 : uint32 proclock_hashcode;
1290 : bool found;
1291 :
1292 : /*
1293 : * Find or create a lock with this tag.
1294 : */
1295 4956768 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
1296 : locktag,
1297 : hashcode,
1298 : HASH_ENTER_NULL,
1299 : &found);
1300 4956768 : if (!lock)
1301 0 : return NULL;
1302 :
1303 : /*
1304 : * if it's a new lock object, initialize it
1305 : */
1306 4956768 : if (!found)
1307 : {
1308 4470316 : lock->grantMask = 0;
1309 4470316 : lock->waitMask = 0;
1310 4470316 : dlist_init(&lock->procLocks);
1311 4470316 : dclist_init(&lock->waitProcs);
1312 4470316 : lock->nRequested = 0;
1313 4470316 : lock->nGranted = 0;
1314 26821896 : MemSet(lock->requested, 0, sizeof(int) * MAX_LOCKMODES);
1315 4470316 : MemSet(lock->granted, 0, sizeof(int) * MAX_LOCKMODES);
1316 : LOCK_PRINT("LockAcquire: new", lock, lockmode);
1317 : }
1318 : else
1319 : {
1320 : LOCK_PRINT("LockAcquire: found", lock, lockmode);
1321 : Assert((lock->nRequested >= 0) && (lock->requested[lockmode] >= 0));
1322 : Assert((lock->nGranted >= 0) && (lock->granted[lockmode] >= 0));
1323 : Assert(lock->nGranted <= lock->nRequested);
1324 : }
1325 :
1326 : /*
1327 : * Create the hash key for the proclock table.
1328 : */
1329 4956768 : proclocktag.myLock = lock;
1330 4956768 : proclocktag.myProc = proc;
1331 :
1332 4956768 : proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
1333 :
1334 : /*
1335 : * Find or create a proclock entry with this tag
1336 : */
1337 4956768 : proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
1338 : &proclocktag,
1339 : proclock_hashcode,
1340 : HASH_ENTER_NULL,
1341 : &found);
1342 4956768 : if (!proclock)
1343 : {
1344 : /* Oops, not enough shmem for the proclock */
1345 0 : if (lock->nRequested == 0)
1346 : {
1347 : /*
1348 : * There are no other requestors of this lock, so garbage-collect
1349 : * the lock object. We *must* do this to avoid a permanent leak
1350 : * of shared memory, because there won't be anything to cause
1351 : * anyone to release the lock object later.
1352 : */
1353 : Assert(dlist_is_empty(&(lock->procLocks)));
1354 0 : if (!hash_search_with_hash_value(LockMethodLockHash,
1355 0 : &(lock->tag),
1356 : hashcode,
1357 : HASH_REMOVE,
1358 : NULL))
1359 0 : elog(PANIC, "lock table corrupted");
1360 : }
1361 0 : return NULL;
1362 : }
1363 :
1364 : /*
1365 : * If new, initialize the new entry
1366 : */
1367 4956768 : if (!found)
1368 : {
1369 4521218 : uint32 partition = LockHashPartition(hashcode);
1370 :
1371 : /*
1372 : * It might seem unsafe to access proclock->groupLeader without a
1373 : * lock, but it's not really. Either we are initializing a proclock
1374 : * on our own behalf, in which case our group leader isn't changing
1375 : * because the group leader for a process can only ever be changed by
1376 : * the process itself; or else we are transferring a fast-path lock to
1377 : * the main lock table, in which case that process can't change its
1378 : * lock group leader without first releasing all of its locks (and in
1379 : * particular the one we are currently transferring).
1380 : */
1381 9042436 : proclock->groupLeader = proc->lockGroupLeader != NULL ?
1382 4521218 : proc->lockGroupLeader : proc;
1383 4521218 : proclock->holdMask = 0;
1384 4521218 : proclock->releaseMask = 0;
1385 : /* Add proclock to appropriate lists */
1386 4521218 : dlist_push_tail(&lock->procLocks, &proclock->lockLink);
1387 4521218 : dlist_push_tail(&proc->myProcLocks[partition], &proclock->procLink);
1388 : PROCLOCK_PRINT("LockAcquire: new", proclock);
1389 : }
1390 : else
1391 : {
1392 : PROCLOCK_PRINT("LockAcquire: found", proclock);
1393 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
1394 :
1395 : #ifdef CHECK_DEADLOCK_RISK
1396 :
1397 : /*
1398 : * Issue warning if we already hold a lower-level lock on this object
1399 : * and do not hold a lock of the requested level or higher. This
1400 : * indicates a deadlock-prone coding practice (eg, we'd have a
1401 : * deadlock if another backend were following the same code path at
1402 : * about the same time).
1403 : *
1404 : * This is not enabled by default, because it may generate log entries
1405 : * about user-level coding practices that are in fact safe in context.
1406 : * It can be enabled to help find system-level problems.
1407 : *
1408 : * XXX Doing numeric comparison on the lockmodes is a hack; it'd be
1409 : * better to use a table. For now, though, this works.
1410 : */
1411 : {
1412 : int i;
1413 :
1414 : for (i = lockMethodTable->numLockModes; i > 0; i--)
1415 : {
1416 : if (proclock->holdMask & LOCKBIT_ON(i))
1417 : {
1418 : if (i >= (int) lockmode)
1419 : break; /* safe: we have a lock >= req level */
1420 : elog(LOG, "deadlock risk: raising lock level"
1421 : " from %s to %s on object %u/%u/%u",
1422 : lockMethodTable->lockModeNames[i],
1423 : lockMethodTable->lockModeNames[lockmode],
1424 : lock->tag.locktag_field1, lock->tag.locktag_field2,
1425 : lock->tag.locktag_field3);
1426 : break;
1427 : }
1428 : }
1429 : }
1430 : #endif /* CHECK_DEADLOCK_RISK */
1431 : }
1432 :
1433 : /*
1434 : * lock->nRequested and lock->requested[] count the total number of
1435 : * requests, whether granted or waiting, so increment those immediately.
1436 : * The other counts don't increment till we get the lock.
1437 : */
1438 4956768 : lock->nRequested++;
1439 4956768 : lock->requested[lockmode]++;
1440 : Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
1441 :
1442 : /*
1443 : * We shouldn't already hold the desired lock; else locallock table is
1444 : * broken.
1445 : */
1446 4956768 : if (proclock->holdMask & LOCKBIT_ON(lockmode))
1447 0 : elog(ERROR, "lock %s on object %u/%u/%u is already held",
1448 : lockMethodTable->lockModeNames[lockmode],
1449 : lock->tag.locktag_field1, lock->tag.locktag_field2,
1450 : lock->tag.locktag_field3);
1451 :
1452 4956768 : return proclock;
1453 : }
1454 :
1455 : /*
1456 : * Check and set/reset the flag that we hold the relation extension lock.
1457 : *
1458 : * It is callers responsibility that this function is called after
1459 : * acquiring/releasing the relation extension lock.
1460 : *
1461 : * Pass acquired as true if lock is acquired, false otherwise.
1462 : */
1463 : static inline void
1464 79344582 : CheckAndSetLockHeld(LOCALLOCK *locallock, bool acquired)
1465 : {
1466 : #ifdef USE_ASSERT_CHECKING
1467 : if (LOCALLOCK_LOCKTAG(*locallock) == LOCKTAG_RELATION_EXTEND)
1468 : IsRelationExtensionLockHeld = acquired;
1469 : #endif
1470 79344582 : }
1471 :
1472 : /*
1473 : * Subroutine to free a locallock entry
1474 : */
1475 : static void
1476 39073536 : RemoveLocalLock(LOCALLOCK *locallock)
1477 : {
1478 : int i;
1479 :
1480 39236498 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
1481 : {
1482 162962 : if (locallock->lockOwners[i].owner != NULL)
1483 162886 : ResourceOwnerForgetLock(locallock->lockOwners[i].owner, locallock);
1484 : }
1485 39073536 : locallock->numLockOwners = 0;
1486 39073536 : if (locallock->lockOwners != NULL)
1487 39073536 : pfree(locallock->lockOwners);
1488 39073536 : locallock->lockOwners = NULL;
1489 :
1490 39073536 : if (locallock->holdsStrongLockCount)
1491 : {
1492 : uint32 fasthashcode;
1493 :
1494 374752 : fasthashcode = FastPathStrongLockHashPartition(locallock->hashcode);
1495 :
1496 374752 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1497 : Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
1498 374752 : FastPathStrongRelationLocks->count[fasthashcode]--;
1499 374752 : locallock->holdsStrongLockCount = false;
1500 374752 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1501 : }
1502 :
1503 39073536 : if (!hash_search(LockMethodLocalHash,
1504 39073536 : &(locallock->tag),
1505 : HASH_REMOVE, NULL))
1506 0 : elog(WARNING, "locallock table corrupted");
1507 :
1508 : /*
1509 : * Indicate that the lock is released for certain types of locks
1510 : */
1511 39073536 : CheckAndSetLockHeld(locallock, false);
1512 39073536 : }
1513 :
1514 : /*
1515 : * LockCheckConflicts -- test whether requested lock conflicts
1516 : * with those already granted
1517 : *
1518 : * Returns true if conflict, false if no conflict.
1519 : *
1520 : * NOTES:
1521 : * Here's what makes this complicated: one process's locks don't
1522 : * conflict with one another, no matter what purpose they are held for
1523 : * (eg, session and transaction locks do not conflict). Nor do the locks
1524 : * of one process in a lock group conflict with those of another process in
1525 : * the same group. So, we must subtract off these locks when determining
1526 : * whether the requested new lock conflicts with those already held.
1527 : */
1528 : bool
1529 4955866 : LockCheckConflicts(LockMethod lockMethodTable,
1530 : LOCKMODE lockmode,
1531 : LOCK *lock,
1532 : PROCLOCK *proclock)
1533 : {
1534 4955866 : int numLockModes = lockMethodTable->numLockModes;
1535 : LOCKMASK myLocks;
1536 4955866 : int conflictMask = lockMethodTable->conflictTab[lockmode];
1537 : int conflictsRemaining[MAX_LOCKMODES];
1538 4955866 : int totalConflictsRemaining = 0;
1539 : dlist_iter proclock_iter;
1540 : int i;
1541 :
1542 : /*
1543 : * first check for global conflicts: If no locks conflict with my request,
1544 : * then I get the lock.
1545 : *
1546 : * Checking for conflict: lock->grantMask represents the types of
1547 : * currently held locks. conflictTable[lockmode] has a bit set for each
1548 : * type of lock that conflicts with request. Bitwise compare tells if
1549 : * there is a conflict.
1550 : */
1551 4955866 : if (!(conflictMask & lock->grantMask))
1552 : {
1553 : PROCLOCK_PRINT("LockCheckConflicts: no conflict", proclock);
1554 4762456 : return false;
1555 : }
1556 :
1557 : /*
1558 : * Rats. Something conflicts. But it could still be my own lock, or a
1559 : * lock held by another member of my locking group. First, figure out how
1560 : * many conflicts remain after subtracting out any locks I hold myself.
1561 : */
1562 193410 : myLocks = proclock->holdMask;
1563 1740690 : for (i = 1; i <= numLockModes; i++)
1564 : {
1565 1547280 : if ((conflictMask & LOCKBIT_ON(i)) == 0)
1566 : {
1567 829680 : conflictsRemaining[i] = 0;
1568 829680 : continue;
1569 : }
1570 717600 : conflictsRemaining[i] = lock->granted[i];
1571 717600 : if (myLocks & LOCKBIT_ON(i))
1572 209562 : --conflictsRemaining[i];
1573 717600 : totalConflictsRemaining += conflictsRemaining[i];
1574 : }
1575 :
1576 : /* If no conflicts remain, we get the lock. */
1577 193410 : if (totalConflictsRemaining == 0)
1578 : {
1579 : PROCLOCK_PRINT("LockCheckConflicts: resolved (simple)", proclock);
1580 188444 : return false;
1581 : }
1582 :
1583 : /* If no group locking, it's definitely a conflict. */
1584 4966 : if (proclock->groupLeader == MyProc && MyProc->lockGroupLeader == NULL)
1585 : {
1586 : Assert(proclock->tag.myProc == MyProc);
1587 : PROCLOCK_PRINT("LockCheckConflicts: conflicting (simple)",
1588 : proclock);
1589 3574 : return true;
1590 : }
1591 :
1592 : /*
1593 : * The relation extension lock conflict even between the group members.
1594 : */
1595 1392 : if (LOCK_LOCKTAG(*lock) == LOCKTAG_RELATION_EXTEND)
1596 : {
1597 : PROCLOCK_PRINT("LockCheckConflicts: conflicting (group)",
1598 : proclock);
1599 42 : return true;
1600 : }
1601 :
1602 : /*
1603 : * Locks held in conflicting modes by members of our own lock group are
1604 : * not real conflicts; we can subtract those out and see if we still have
1605 : * a conflict. This is O(N) in the number of processes holding or
1606 : * awaiting locks on this object. We could improve that by making the
1607 : * shared memory state more complex (and larger) but it doesn't seem worth
1608 : * it.
1609 : */
1610 2604 : dlist_foreach(proclock_iter, &lock->procLocks)
1611 : {
1612 2224 : PROCLOCK *otherproclock =
1613 2224 : dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
1614 :
1615 2224 : if (proclock != otherproclock &&
1616 1844 : proclock->groupLeader == otherproclock->groupLeader &&
1617 980 : (otherproclock->holdMask & conflictMask) != 0)
1618 : {
1619 976 : int intersectMask = otherproclock->holdMask & conflictMask;
1620 :
1621 8784 : for (i = 1; i <= numLockModes; i++)
1622 : {
1623 7808 : if ((intersectMask & LOCKBIT_ON(i)) != 0)
1624 : {
1625 996 : if (conflictsRemaining[i] <= 0)
1626 0 : elog(PANIC, "proclocks held do not match lock");
1627 996 : conflictsRemaining[i]--;
1628 996 : totalConflictsRemaining--;
1629 : }
1630 : }
1631 :
1632 976 : if (totalConflictsRemaining == 0)
1633 : {
1634 : PROCLOCK_PRINT("LockCheckConflicts: resolved (group)",
1635 : proclock);
1636 970 : return false;
1637 : }
1638 : }
1639 : }
1640 :
1641 : /* Nope, it's a real conflict. */
1642 : PROCLOCK_PRINT("LockCheckConflicts: conflicting (group)", proclock);
1643 380 : return true;
1644 : }
1645 :
1646 : /*
1647 : * GrantLock -- update the lock and proclock data structures to show
1648 : * the lock request has been granted.
1649 : *
1650 : * NOTE: if proc was blocked, it also needs to be removed from the wait list
1651 : * and have its waitLock/waitProcLock fields cleared. That's not done here.
1652 : *
1653 : * NOTE: the lock grant also has to be recorded in the associated LOCALLOCK
1654 : * table entry; but since we may be awaking some other process, we can't do
1655 : * that here; it's done by GrantLockLocal, instead.
1656 : */
1657 : void
1658 4955392 : GrantLock(LOCK *lock, PROCLOCK *proclock, LOCKMODE lockmode)
1659 : {
1660 4955392 : lock->nGranted++;
1661 4955392 : lock->granted[lockmode]++;
1662 4955392 : lock->grantMask |= LOCKBIT_ON(lockmode);
1663 4955392 : if (lock->granted[lockmode] == lock->requested[lockmode])
1664 4954784 : lock->waitMask &= LOCKBIT_OFF(lockmode);
1665 4955392 : proclock->holdMask |= LOCKBIT_ON(lockmode);
1666 : LOCK_PRINT("GrantLock", lock, lockmode);
1667 : Assert((lock->nGranted > 0) && (lock->granted[lockmode] > 0));
1668 : Assert(lock->nGranted <= lock->nRequested);
1669 4955392 : }
1670 :
1671 : /*
1672 : * UnGrantLock -- opposite of GrantLock.
1673 : *
1674 : * Updates the lock and proclock data structures to show that the lock
1675 : * is no longer held nor requested by the current holder.
1676 : *
1677 : * Returns true if there were any waiters waiting on the lock that
1678 : * should now be woken up with ProcLockWakeup.
1679 : */
1680 : static bool
1681 4955260 : UnGrantLock(LOCK *lock, LOCKMODE lockmode,
1682 : PROCLOCK *proclock, LockMethod lockMethodTable)
1683 : {
1684 4955260 : bool wakeupNeeded = false;
1685 :
1686 : Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
1687 : Assert((lock->nGranted > 0) && (lock->granted[lockmode] > 0));
1688 : Assert(lock->nGranted <= lock->nRequested);
1689 :
1690 : /*
1691 : * fix the general lock stats
1692 : */
1693 4955260 : lock->nRequested--;
1694 4955260 : lock->requested[lockmode]--;
1695 4955260 : lock->nGranted--;
1696 4955260 : lock->granted[lockmode]--;
1697 :
1698 4955260 : if (lock->granted[lockmode] == 0)
1699 : {
1700 : /* change the conflict mask. No more of this lock type. */
1701 4916960 : lock->grantMask &= LOCKBIT_OFF(lockmode);
1702 : }
1703 :
1704 : LOCK_PRINT("UnGrantLock: updated", lock, lockmode);
1705 :
1706 : /*
1707 : * We need only run ProcLockWakeup if the released lock conflicts with at
1708 : * least one of the lock types requested by waiter(s). Otherwise whatever
1709 : * conflict made them wait must still exist. NOTE: before MVCC, we could
1710 : * skip wakeup if lock->granted[lockmode] was still positive. But that's
1711 : * not true anymore, because the remaining granted locks might belong to
1712 : * some waiter, who could now be awakened because he doesn't conflict with
1713 : * his own locks.
1714 : */
1715 4955260 : if (lockMethodTable->conflictTab[lockmode] & lock->waitMask)
1716 2354 : wakeupNeeded = true;
1717 :
1718 : /*
1719 : * Now fix the per-proclock state.
1720 : */
1721 4955260 : proclock->holdMask &= LOCKBIT_OFF(lockmode);
1722 : PROCLOCK_PRINT("UnGrantLock: updated", proclock);
1723 :
1724 4955260 : return wakeupNeeded;
1725 : }
1726 :
1727 : /*
1728 : * CleanUpLock -- clean up after releasing a lock. We garbage-collect the
1729 : * proclock and lock objects if possible, and call ProcLockWakeup if there
1730 : * are remaining requests and the caller says it's OK. (Normally, this
1731 : * should be called after UnGrantLock, and wakeupNeeded is the result from
1732 : * UnGrantLock.)
1733 : *
1734 : * The appropriate partition lock must be held at entry, and will be
1735 : * held at exit.
1736 : */
1737 : static void
1738 4884310 : CleanUpLock(LOCK *lock, PROCLOCK *proclock,
1739 : LockMethod lockMethodTable, uint32 hashcode,
1740 : bool wakeupNeeded)
1741 : {
1742 : /*
1743 : * If this was my last hold on this lock, delete my entry in the proclock
1744 : * table.
1745 : */
1746 4884310 : if (proclock->holdMask == 0)
1747 : {
1748 : uint32 proclock_hashcode;
1749 :
1750 : PROCLOCK_PRINT("CleanUpLock: deleting", proclock);
1751 4520196 : dlist_delete(&proclock->lockLink);
1752 4520196 : dlist_delete(&proclock->procLink);
1753 4520196 : proclock_hashcode = ProcLockHashCode(&proclock->tag, hashcode);
1754 4520196 : if (!hash_search_with_hash_value(LockMethodProcLockHash,
1755 4520196 : &(proclock->tag),
1756 : proclock_hashcode,
1757 : HASH_REMOVE,
1758 : NULL))
1759 0 : elog(PANIC, "proclock table corrupted");
1760 : }
1761 :
1762 4884310 : if (lock->nRequested == 0)
1763 : {
1764 : /*
1765 : * The caller just released the last lock, so garbage-collect the lock
1766 : * object.
1767 : */
1768 : LOCK_PRINT("CleanUpLock: deleting", lock, 0);
1769 : Assert(dlist_is_empty(&lock->procLocks));
1770 4470352 : if (!hash_search_with_hash_value(LockMethodLockHash,
1771 4470352 : &(lock->tag),
1772 : hashcode,
1773 : HASH_REMOVE,
1774 : NULL))
1775 0 : elog(PANIC, "lock table corrupted");
1776 : }
1777 413958 : else if (wakeupNeeded)
1778 : {
1779 : /* There are waiters on this lock, so wake them up. */
1780 2434 : ProcLockWakeup(lockMethodTable, lock);
1781 : }
1782 4884310 : }
1783 :
1784 : /*
1785 : * GrantLockLocal -- update the locallock data structures to show
1786 : * the lock request has been granted.
1787 : *
1788 : * We expect that LockAcquire made sure there is room to add a new
1789 : * ResourceOwner entry.
1790 : */
1791 : static void
1792 43430276 : GrantLockLocal(LOCALLOCK *locallock, ResourceOwner owner)
1793 : {
1794 43430276 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
1795 : int i;
1796 :
1797 : Assert(locallock->numLockOwners < locallock->maxLockOwners);
1798 : /* Count the total */
1799 43430276 : locallock->nLocks++;
1800 : /* Count the per-owner lock */
1801 44998082 : for (i = 0; i < locallock->numLockOwners; i++)
1802 : {
1803 4727036 : if (lockOwners[i].owner == owner)
1804 : {
1805 3159230 : lockOwners[i].nLocks++;
1806 3159230 : return;
1807 : }
1808 : }
1809 40271046 : lockOwners[i].owner = owner;
1810 40271046 : lockOwners[i].nLocks = 1;
1811 40271046 : locallock->numLockOwners++;
1812 40271046 : if (owner != NULL)
1813 39946392 : ResourceOwnerRememberLock(owner, locallock);
1814 :
1815 : /* Indicate that the lock is acquired for certain types of locks. */
1816 40271046 : CheckAndSetLockHeld(locallock, true);
1817 : }
1818 :
1819 : /*
1820 : * BeginStrongLockAcquire - inhibit use of fastpath for a given LOCALLOCK,
1821 : * and arrange for error cleanup if it fails
1822 : */
1823 : static void
1824 375308 : BeginStrongLockAcquire(LOCALLOCK *locallock, uint32 fasthashcode)
1825 : {
1826 : Assert(StrongLockInProgress == NULL);
1827 : Assert(locallock->holdsStrongLockCount == false);
1828 :
1829 : /*
1830 : * Adding to a memory location is not atomic, so we take a spinlock to
1831 : * ensure we don't collide with someone else trying to bump the count at
1832 : * the same time.
1833 : *
1834 : * XXX: It might be worth considering using an atomic fetch-and-add
1835 : * instruction here, on architectures where that is supported.
1836 : */
1837 :
1838 375308 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1839 375308 : FastPathStrongRelationLocks->count[fasthashcode]++;
1840 375308 : locallock->holdsStrongLockCount = true;
1841 375308 : StrongLockInProgress = locallock;
1842 375308 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1843 375308 : }
1844 :
1845 : /*
1846 : * FinishStrongLockAcquire - cancel pending cleanup for a strong lock
1847 : * acquisition once it's no longer needed
1848 : */
1849 : static void
1850 4951866 : FinishStrongLockAcquire(void)
1851 : {
1852 4951866 : StrongLockInProgress = NULL;
1853 4951866 : }
1854 :
1855 : /*
1856 : * AbortStrongLockAcquire - undo strong lock state changes performed by
1857 : * BeginStrongLockAcquire.
1858 : */
1859 : void
1860 1222212 : AbortStrongLockAcquire(void)
1861 : {
1862 : uint32 fasthashcode;
1863 1222212 : LOCALLOCK *locallock = StrongLockInProgress;
1864 :
1865 1222212 : if (locallock == NULL)
1866 1221786 : return;
1867 :
1868 426 : fasthashcode = FastPathStrongLockHashPartition(locallock->hashcode);
1869 : Assert(locallock->holdsStrongLockCount == true);
1870 426 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
1871 : Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
1872 426 : FastPathStrongRelationLocks->count[fasthashcode]--;
1873 426 : locallock->holdsStrongLockCount = false;
1874 426 : StrongLockInProgress = NULL;
1875 426 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
1876 : }
1877 :
1878 : /*
1879 : * GrantAwaitedLock -- call GrantLockLocal for the lock we are doing
1880 : * WaitOnLock on.
1881 : *
1882 : * proc.c needs this for the case where we are booted off the lock by
1883 : * timeout, but discover that someone granted us the lock anyway.
1884 : *
1885 : * We could just export GrantLockLocal, but that would require including
1886 : * resowner.h in lock.h, which creates circularity.
1887 : */
1888 : void
1889 2 : GrantAwaitedLock(void)
1890 : {
1891 2 : GrantLockLocal(awaitedLock, awaitedOwner);
1892 2 : }
1893 :
1894 : /*
1895 : * GetAwaitedLock -- Return the lock we're currently doing WaitOnLock on.
1896 : */
1897 : LOCALLOCK *
1898 1220756 : GetAwaitedLock(void)
1899 : {
1900 1220756 : return awaitedLock;
1901 : }
1902 :
1903 : /*
1904 : * ResetAwaitedLock -- Forget that we are waiting on a lock.
1905 : */
1906 : void
1907 80 : ResetAwaitedLock(void)
1908 : {
1909 80 : awaitedLock = NULL;
1910 80 : }
1911 :
1912 : /*
1913 : * MarkLockClear -- mark an acquired lock as "clear"
1914 : *
1915 : * This means that we know we have absorbed all sinval messages that other
1916 : * sessions generated before we acquired this lock, and so we can confidently
1917 : * assume we know about any catalog changes protected by this lock.
1918 : */
1919 : void
1920 37632230 : MarkLockClear(LOCALLOCK *locallock)
1921 : {
1922 : Assert(locallock->nLocks > 0);
1923 37632230 : locallock->lockCleared = true;
1924 37632230 : }
1925 :
1926 : /*
1927 : * WaitOnLock -- wait to acquire a lock
1928 : *
1929 : * This is a wrapper around ProcSleep, with extra tracing and bookkeeping.
1930 : */
1931 : static ProcWaitStatus
1932 2484 : WaitOnLock(LOCALLOCK *locallock, ResourceOwner owner)
1933 : {
1934 : ProcWaitStatus result;
1935 : ErrorContextCallback waiterrcontext;
1936 :
1937 : TRACE_POSTGRESQL_LOCK_WAIT_START(locallock->tag.lock.locktag_field1,
1938 : locallock->tag.lock.locktag_field2,
1939 : locallock->tag.lock.locktag_field3,
1940 : locallock->tag.lock.locktag_field4,
1941 : locallock->tag.lock.locktag_type,
1942 : locallock->tag.mode);
1943 :
1944 : /* Setup error traceback support for ereport() */
1945 2484 : waiterrcontext.callback = waitonlock_error_callback;
1946 2484 : waiterrcontext.arg = (void *) locallock;
1947 2484 : waiterrcontext.previous = error_context_stack;
1948 2484 : error_context_stack = &waiterrcontext;
1949 :
1950 : /* adjust the process title to indicate that it's waiting */
1951 2484 : set_ps_display_suffix("waiting");
1952 :
1953 : /*
1954 : * Record the fact that we are waiting for a lock, so that
1955 : * LockErrorCleanup will clean up if cancel/die happens.
1956 : */
1957 2484 : awaitedLock = locallock;
1958 2484 : awaitedOwner = owner;
1959 :
1960 : /*
1961 : * NOTE: Think not to put any shared-state cleanup after the call to
1962 : * ProcSleep, in either the normal or failure path. The lock state must
1963 : * be fully set by the lock grantor, or by CheckDeadLock if we give up
1964 : * waiting for the lock. This is necessary because of the possibility
1965 : * that a cancel/die interrupt will interrupt ProcSleep after someone else
1966 : * grants us the lock, but before we've noticed it. Hence, after granting,
1967 : * the locktable state must fully reflect the fact that we own the lock;
1968 : * we can't do additional work on return.
1969 : *
1970 : * We can and do use a PG_TRY block to try to clean up after failure, but
1971 : * this still has a major limitation: elog(FATAL) can occur while waiting
1972 : * (eg, a "die" interrupt), and then control won't come back here. So all
1973 : * cleanup of essential state should happen in LockErrorCleanup, not here.
1974 : * We can use PG_TRY to clear the "waiting" status flags, since doing that
1975 : * is unimportant if the process exits.
1976 : */
1977 2484 : PG_TRY();
1978 : {
1979 2484 : result = ProcSleep(locallock);
1980 : }
1981 72 : PG_CATCH();
1982 : {
1983 : /* In this path, awaitedLock remains set until LockErrorCleanup */
1984 :
1985 : /* reset ps display to remove the suffix */
1986 72 : set_ps_display_remove_suffix();
1987 :
1988 : /* and propagate the error */
1989 72 : PG_RE_THROW();
1990 : }
1991 2404 : PG_END_TRY();
1992 :
1993 : /*
1994 : * We no longer want LockErrorCleanup to do anything.
1995 : */
1996 2404 : awaitedLock = NULL;
1997 :
1998 : /* reset ps display to remove the suffix */
1999 2404 : set_ps_display_remove_suffix();
2000 :
2001 2404 : error_context_stack = waiterrcontext.previous;
2002 :
2003 : TRACE_POSTGRESQL_LOCK_WAIT_DONE(locallock->tag.lock.locktag_field1,
2004 : locallock->tag.lock.locktag_field2,
2005 : locallock->tag.lock.locktag_field3,
2006 : locallock->tag.lock.locktag_field4,
2007 : locallock->tag.lock.locktag_type,
2008 : locallock->tag.mode);
2009 :
2010 2404 : return result;
2011 : }
2012 :
2013 : /*
2014 : * error context callback for failures in WaitOnLock
2015 : *
2016 : * We report which lock was being waited on, in the same style used in
2017 : * deadlock reports. This helps with lock timeout errors in particular.
2018 : */
2019 : static void
2020 1664 : waitonlock_error_callback(void *arg)
2021 : {
2022 1664 : LOCALLOCK *locallock = (LOCALLOCK *) arg;
2023 1664 : const LOCKTAG *tag = &locallock->tag.lock;
2024 1664 : LOCKMODE mode = locallock->tag.mode;
2025 : StringInfoData locktagbuf;
2026 :
2027 1664 : initStringInfo(&locktagbuf);
2028 1664 : DescribeLockTag(&locktagbuf, tag);
2029 :
2030 3328 : errcontext("waiting for %s on %s",
2031 1664 : GetLockmodeName(tag->locktag_lockmethodid, mode),
2032 : locktagbuf.data);
2033 1664 : }
2034 :
2035 : /*
2036 : * Remove a proc from the wait-queue it is on (caller must know it is on one).
2037 : * This is only used when the proc has failed to get the lock, so we set its
2038 : * waitStatus to PROC_WAIT_STATUS_ERROR.
2039 : *
2040 : * Appropriate partition lock must be held by caller. Also, caller is
2041 : * responsible for signaling the proc if needed.
2042 : *
2043 : * NB: this does not clean up any locallock object that may exist for the lock.
2044 : */
2045 : void
2046 88 : RemoveFromWaitQueue(PGPROC *proc, uint32 hashcode)
2047 : {
2048 88 : LOCK *waitLock = proc->waitLock;
2049 88 : PROCLOCK *proclock = proc->waitProcLock;
2050 88 : LOCKMODE lockmode = proc->waitLockMode;
2051 88 : LOCKMETHODID lockmethodid = LOCK_LOCKMETHOD(*waitLock);
2052 :
2053 : /* Make sure proc is waiting */
2054 : Assert(proc->waitStatus == PROC_WAIT_STATUS_WAITING);
2055 : Assert(proc->links.next != NULL);
2056 : Assert(waitLock);
2057 : Assert(!dclist_is_empty(&waitLock->waitProcs));
2058 : Assert(0 < lockmethodid && lockmethodid < lengthof(LockMethods));
2059 :
2060 : /* Remove proc from lock's wait queue */
2061 88 : dclist_delete_from_thoroughly(&waitLock->waitProcs, &proc->links);
2062 :
2063 : /* Undo increments of request counts by waiting process */
2064 : Assert(waitLock->nRequested > 0);
2065 : Assert(waitLock->nRequested > proc->waitLock->nGranted);
2066 88 : waitLock->nRequested--;
2067 : Assert(waitLock->requested[lockmode] > 0);
2068 88 : waitLock->requested[lockmode]--;
2069 : /* don't forget to clear waitMask bit if appropriate */
2070 88 : if (waitLock->granted[lockmode] == waitLock->requested[lockmode])
2071 88 : waitLock->waitMask &= LOCKBIT_OFF(lockmode);
2072 :
2073 : /* Clean up the proc's own state, and pass it the ok/fail signal */
2074 88 : proc->waitLock = NULL;
2075 88 : proc->waitProcLock = NULL;
2076 88 : proc->waitStatus = PROC_WAIT_STATUS_ERROR;
2077 :
2078 : /*
2079 : * Delete the proclock immediately if it represents no already-held locks.
2080 : * (This must happen now because if the owner of the lock decides to
2081 : * release it, and the requested/granted counts then go to zero,
2082 : * LockRelease expects there to be no remaining proclocks.) Then see if
2083 : * any other waiters for the lock can be woken up now.
2084 : */
2085 88 : CleanUpLock(waitLock, proclock,
2086 88 : LockMethods[lockmethodid], hashcode,
2087 : true);
2088 88 : }
2089 :
2090 : /*
2091 : * LockRelease -- look up 'locktag' and release one 'lockmode' lock on it.
2092 : * Release a session lock if 'sessionLock' is true, else release a
2093 : * regular transaction lock.
2094 : *
2095 : * Side Effects: find any waiting processes that are now wakable,
2096 : * grant them their requested locks and awaken them.
2097 : * (We have to grant the lock here to avoid a race between
2098 : * the waking process and any new process to
2099 : * come along and request the lock.)
2100 : */
2101 : bool
2102 38725224 : LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
2103 : {
2104 38725224 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
2105 : LockMethod lockMethodTable;
2106 : LOCALLOCKTAG localtag;
2107 : LOCALLOCK *locallock;
2108 : LOCK *lock;
2109 : PROCLOCK *proclock;
2110 : LWLock *partitionLock;
2111 : bool wakeupNeeded;
2112 :
2113 38725224 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2114 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2115 38725224 : lockMethodTable = LockMethods[lockmethodid];
2116 38725224 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
2117 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
2118 :
2119 : #ifdef LOCK_DEBUG
2120 : if (LOCK_DEBUG_ENABLED(locktag))
2121 : elog(LOG, "LockRelease: lock [%u,%u] %s",
2122 : locktag->locktag_field1, locktag->locktag_field2,
2123 : lockMethodTable->lockModeNames[lockmode]);
2124 : #endif
2125 :
2126 : /*
2127 : * Find the LOCALLOCK entry for this lock and lockmode
2128 : */
2129 38725224 : MemSet(&localtag, 0, sizeof(localtag)); /* must clear padding */
2130 38725224 : localtag.lock = *locktag;
2131 38725224 : localtag.mode = lockmode;
2132 :
2133 38725224 : locallock = (LOCALLOCK *) hash_search(LockMethodLocalHash,
2134 : &localtag,
2135 : HASH_FIND, NULL);
2136 :
2137 : /*
2138 : * let the caller print its own error message, too. Do not ereport(ERROR).
2139 : */
2140 38725224 : if (!locallock || locallock->nLocks <= 0)
2141 : {
2142 26 : elog(WARNING, "you don't own a lock of type %s",
2143 : lockMethodTable->lockModeNames[lockmode]);
2144 26 : return false;
2145 : }
2146 :
2147 : /*
2148 : * Decrease the count for the resource owner.
2149 : */
2150 : {
2151 38725198 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
2152 : ResourceOwner owner;
2153 : int i;
2154 :
2155 : /* Identify owner for lock */
2156 38725198 : if (sessionLock)
2157 324638 : owner = NULL;
2158 : else
2159 38400560 : owner = CurrentResourceOwner;
2160 :
2161 38726972 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
2162 : {
2163 38726948 : if (lockOwners[i].owner == owner)
2164 : {
2165 : Assert(lockOwners[i].nLocks > 0);
2166 38725174 : if (--lockOwners[i].nLocks == 0)
2167 : {
2168 37349730 : if (owner != NULL)
2169 37025152 : ResourceOwnerForgetLock(owner, locallock);
2170 : /* compact out unused slot */
2171 37349730 : locallock->numLockOwners--;
2172 37349730 : if (i < locallock->numLockOwners)
2173 174 : lockOwners[i] = lockOwners[locallock->numLockOwners];
2174 : }
2175 38725174 : break;
2176 : }
2177 : }
2178 38725198 : if (i < 0)
2179 : {
2180 : /* don't release a lock belonging to another owner */
2181 24 : elog(WARNING, "you don't own a lock of type %s",
2182 : lockMethodTable->lockModeNames[lockmode]);
2183 24 : return false;
2184 : }
2185 : }
2186 :
2187 : /*
2188 : * Decrease the total local count. If we're still holding the lock, we're
2189 : * done.
2190 : */
2191 38725174 : locallock->nLocks--;
2192 :
2193 38725174 : if (locallock->nLocks > 0)
2194 2048072 : return true;
2195 :
2196 : /*
2197 : * At this point we can no longer suppose we are clear of invalidation
2198 : * messages related to this lock. Although we'll delete the LOCALLOCK
2199 : * object before any intentional return from this routine, it seems worth
2200 : * the trouble to explicitly reset lockCleared right now, just in case
2201 : * some error prevents us from deleting the LOCALLOCK.
2202 : */
2203 36677102 : locallock->lockCleared = false;
2204 :
2205 : /* Attempt fast release of any lock eligible for the fast path. */
2206 36677102 : if (EligibleForRelationFastPath(locktag, lockmode) &&
2207 33685028 : FastPathLocalUseCounts[FAST_PATH_REL_GROUP(locktag->locktag_field2)] > 0)
2208 : {
2209 : bool released;
2210 :
2211 : /*
2212 : * We might not find the lock here, even if we originally entered it
2213 : * here. Another backend may have moved it to the main table.
2214 : */
2215 33220672 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
2216 33220672 : released = FastPathUnGrantRelationLock(locktag->locktag_field2,
2217 : lockmode);
2218 33220672 : LWLockRelease(&MyProc->fpInfoLock);
2219 33220672 : if (released)
2220 : {
2221 32837080 : RemoveLocalLock(locallock);
2222 32837080 : return true;
2223 : }
2224 : }
2225 :
2226 : /*
2227 : * Otherwise we've got to mess with the shared lock table.
2228 : */
2229 3840022 : partitionLock = LockHashPartitionLock(locallock->hashcode);
2230 :
2231 3840022 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2232 :
2233 : /*
2234 : * Normally, we don't need to re-find the lock or proclock, since we kept
2235 : * their addresses in the locallock table, and they couldn't have been
2236 : * removed while we were holding a lock on them. But it's possible that
2237 : * the lock was taken fast-path and has since been moved to the main hash
2238 : * table by another backend, in which case we will need to look up the
2239 : * objects here. We assume the lock field is NULL if so.
2240 : */
2241 3840022 : lock = locallock->lock;
2242 3840022 : if (!lock)
2243 : {
2244 : PROCLOCKTAG proclocktag;
2245 :
2246 : Assert(EligibleForRelationFastPath(locktag, lockmode));
2247 18 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
2248 : locktag,
2249 : locallock->hashcode,
2250 : HASH_FIND,
2251 : NULL);
2252 18 : if (!lock)
2253 0 : elog(ERROR, "failed to re-find shared lock object");
2254 18 : locallock->lock = lock;
2255 :
2256 18 : proclocktag.myLock = lock;
2257 18 : proclocktag.myProc = MyProc;
2258 18 : locallock->proclock = (PROCLOCK *) hash_search(LockMethodProcLockHash,
2259 : &proclocktag,
2260 : HASH_FIND,
2261 : NULL);
2262 18 : if (!locallock->proclock)
2263 0 : elog(ERROR, "failed to re-find shared proclock object");
2264 : }
2265 : LOCK_PRINT("LockRelease: found", lock, lockmode);
2266 3840022 : proclock = locallock->proclock;
2267 : PROCLOCK_PRINT("LockRelease: found", proclock);
2268 :
2269 : /*
2270 : * Double-check that we are actually holding a lock of the type we want to
2271 : * release.
2272 : */
2273 3840022 : if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
2274 : {
2275 : PROCLOCK_PRINT("LockRelease: WRONGTYPE", proclock);
2276 0 : LWLockRelease(partitionLock);
2277 0 : elog(WARNING, "you don't own a lock of type %s",
2278 : lockMethodTable->lockModeNames[lockmode]);
2279 0 : RemoveLocalLock(locallock);
2280 0 : return false;
2281 : }
2282 :
2283 : /*
2284 : * Do the releasing. CleanUpLock will waken any now-wakable waiters.
2285 : */
2286 3840022 : wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
2287 :
2288 3840022 : CleanUpLock(lock, proclock,
2289 : lockMethodTable, locallock->hashcode,
2290 : wakeupNeeded);
2291 :
2292 3840022 : LWLockRelease(partitionLock);
2293 :
2294 3840022 : RemoveLocalLock(locallock);
2295 3840022 : return true;
2296 : }
2297 :
2298 : /*
2299 : * LockReleaseAll -- Release all locks of the specified lock method that
2300 : * are held by the current process.
2301 : *
2302 : * Well, not necessarily *all* locks. The available behaviors are:
2303 : * allLocks == true: release all locks including session locks.
2304 : * allLocks == false: release all non-session locks.
2305 : */
2306 : void
2307 2356762 : LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
2308 : {
2309 : HASH_SEQ_STATUS status;
2310 : LockMethod lockMethodTable;
2311 : int i,
2312 : numLockModes;
2313 : LOCALLOCK *locallock;
2314 : LOCK *lock;
2315 : int partition;
2316 2356762 : bool have_fast_path_lwlock = false;
2317 :
2318 2356762 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2319 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2320 2356762 : lockMethodTable = LockMethods[lockmethodid];
2321 :
2322 : #ifdef LOCK_DEBUG
2323 : if (*(lockMethodTable->trace_flag))
2324 : elog(LOG, "LockReleaseAll: lockmethod=%d", lockmethodid);
2325 : #endif
2326 :
2327 : /*
2328 : * Get rid of our fast-path VXID lock, if appropriate. Note that this is
2329 : * the only way that the lock we hold on our own VXID can ever get
2330 : * released: it is always and only released when a toplevel transaction
2331 : * ends.
2332 : */
2333 2356762 : if (lockmethodid == DEFAULT_LOCKMETHOD)
2334 1160712 : VirtualXactLockTableCleanup();
2335 :
2336 2356762 : numLockModes = lockMethodTable->numLockModes;
2337 :
2338 : /*
2339 : * First we run through the locallock table and get rid of unwanted
2340 : * entries, then we scan the process's proclocks and get rid of those. We
2341 : * do this separately because we may have multiple locallock entries
2342 : * pointing to the same proclock, and we daren't end up with any dangling
2343 : * pointers. Fast-path locks are cleaned up during the locallock table
2344 : * scan, though.
2345 : */
2346 2356762 : hash_seq_init(&status, LockMethodLocalHash);
2347 :
2348 5353616 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2349 : {
2350 : /*
2351 : * If the LOCALLOCK entry is unused, something must've gone wrong
2352 : * while trying to acquire this lock. Just forget the local entry.
2353 : */
2354 2996854 : if (locallock->nLocks == 0)
2355 : {
2356 88 : RemoveLocalLock(locallock);
2357 88 : continue;
2358 : }
2359 :
2360 : /* Ignore items that are not of the lockmethod to be removed */
2361 2996766 : if (LOCALLOCK_LOCKMETHOD(*locallock) != lockmethodid)
2362 301908 : continue;
2363 :
2364 : /*
2365 : * If we are asked to release all locks, we can just zap the entry.
2366 : * Otherwise, must scan to see if there are session locks. We assume
2367 : * there is at most one lockOwners entry for session locks.
2368 : */
2369 2694858 : if (!allLocks)
2370 : {
2371 2534758 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
2372 :
2373 : /* If session lock is above array position 0, move it down to 0 */
2374 5350278 : for (i = 0; i < locallock->numLockOwners; i++)
2375 : {
2376 2815520 : if (lockOwners[i].owner == NULL)
2377 301372 : lockOwners[0] = lockOwners[i];
2378 : else
2379 2514148 : ResourceOwnerForgetLock(lockOwners[i].owner, locallock);
2380 : }
2381 :
2382 2534758 : if (locallock->numLockOwners > 0 &&
2383 2534758 : lockOwners[0].owner == NULL &&
2384 301372 : lockOwners[0].nLocks > 0)
2385 : {
2386 : /* Fix the locallock to show just the session locks */
2387 301372 : locallock->nLocks = lockOwners[0].nLocks;
2388 301372 : locallock->numLockOwners = 1;
2389 : /* We aren't deleting this locallock, so done */
2390 301372 : continue;
2391 : }
2392 : else
2393 2233386 : locallock->numLockOwners = 0;
2394 : }
2395 :
2396 : #ifdef USE_ASSERT_CHECKING
2397 :
2398 : /*
2399 : * Tuple locks are currently held only for short durations within a
2400 : * transaction. Check that we didn't forget to release one.
2401 : */
2402 : if (LOCALLOCK_LOCKTAG(*locallock) == LOCKTAG_TUPLE && !allLocks)
2403 : elog(WARNING, "tuple lock held at commit");
2404 : #endif
2405 :
2406 : /*
2407 : * If the lock or proclock pointers are NULL, this lock was taken via
2408 : * the relation fast-path (and is not known to have been transferred).
2409 : */
2410 2393486 : if (locallock->proclock == NULL || locallock->lock == NULL)
2411 2274 : {
2412 1282408 : LOCKMODE lockmode = locallock->tag.mode;
2413 : Oid relid;
2414 :
2415 : /* Verify that a fast-path lock is what we've got. */
2416 1282408 : if (!EligibleForRelationFastPath(&locallock->tag.lock, lockmode))
2417 0 : elog(PANIC, "locallock table corrupted");
2418 :
2419 : /*
2420 : * If we don't currently hold the LWLock that protects our
2421 : * fast-path data structures, we must acquire it before attempting
2422 : * to release the lock via the fast-path. We will continue to
2423 : * hold the LWLock until we're done scanning the locallock table,
2424 : * unless we hit a transferred fast-path lock. (XXX is this
2425 : * really such a good idea? There could be a lot of entries ...)
2426 : */
2427 1282408 : if (!have_fast_path_lwlock)
2428 : {
2429 490132 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
2430 490132 : have_fast_path_lwlock = true;
2431 : }
2432 :
2433 : /* Attempt fast-path release. */
2434 1282408 : relid = locallock->tag.lock.locktag_field2;
2435 1282408 : if (FastPathUnGrantRelationLock(relid, lockmode))
2436 : {
2437 1280134 : RemoveLocalLock(locallock);
2438 1280134 : continue;
2439 : }
2440 :
2441 : /*
2442 : * Our lock, originally taken via the fast path, has been
2443 : * transferred to the main lock table. That's going to require
2444 : * some extra work, so release our fast-path lock before starting.
2445 : */
2446 2274 : LWLockRelease(&MyProc->fpInfoLock);
2447 2274 : have_fast_path_lwlock = false;
2448 :
2449 : /*
2450 : * Now dump the lock. We haven't got a pointer to the LOCK or
2451 : * PROCLOCK in this case, so we have to handle this a bit
2452 : * differently than a normal lock release. Unfortunately, this
2453 : * requires an extra LWLock acquire-and-release cycle on the
2454 : * partitionLock, but hopefully it shouldn't happen often.
2455 : */
2456 2274 : LockRefindAndRelease(lockMethodTable, MyProc,
2457 : &locallock->tag.lock, lockmode, false);
2458 2274 : RemoveLocalLock(locallock);
2459 2274 : continue;
2460 : }
2461 :
2462 : /* Mark the proclock to show we need to release this lockmode */
2463 1111078 : if (locallock->nLocks > 0)
2464 1111078 : locallock->proclock->releaseMask |= LOCKBIT_ON(locallock->tag.mode);
2465 :
2466 : /* And remove the locallock hashtable entry */
2467 1111078 : RemoveLocalLock(locallock);
2468 : }
2469 :
2470 : /* Done with the fast-path data structures */
2471 2356762 : if (have_fast_path_lwlock)
2472 487858 : LWLockRelease(&MyProc->fpInfoLock);
2473 :
2474 : /*
2475 : * Now, scan each lock partition separately.
2476 : */
2477 40064954 : for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
2478 : {
2479 : LWLock *partitionLock;
2480 37708192 : dlist_head *procLocks = &MyProc->myProcLocks[partition];
2481 : dlist_mutable_iter proclock_iter;
2482 :
2483 37708192 : partitionLock = LockHashPartitionLockByIndex(partition);
2484 :
2485 : /*
2486 : * If the proclock list for this partition is empty, we can skip
2487 : * acquiring the partition lock. This optimization is trickier than
2488 : * it looks, because another backend could be in process of adding
2489 : * something to our proclock list due to promoting one of our
2490 : * fast-path locks. However, any such lock must be one that we
2491 : * decided not to delete above, so it's okay to skip it again now;
2492 : * we'd just decide not to delete it again. We must, however, be
2493 : * careful to re-fetch the list header once we've acquired the
2494 : * partition lock, to be sure we have a valid, up-to-date pointer.
2495 : * (There is probably no significant risk if pointer fetch/store is
2496 : * atomic, but we don't wish to assume that.)
2497 : *
2498 : * XXX This argument assumes that the locallock table correctly
2499 : * represents all of our fast-path locks. While allLocks mode
2500 : * guarantees to clean up all of our normal locks regardless of the
2501 : * locallock situation, we lose that guarantee for fast-path locks.
2502 : * This is not ideal.
2503 : */
2504 37708192 : if (dlist_is_empty(procLocks))
2505 36297156 : continue; /* needn't examine this partition */
2506 :
2507 1411036 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2508 :
2509 3052944 : dlist_foreach_modify(proclock_iter, procLocks)
2510 : {
2511 1641908 : PROCLOCK *proclock = dlist_container(PROCLOCK, procLink, proclock_iter.cur);
2512 1641908 : bool wakeupNeeded = false;
2513 :
2514 : Assert(proclock->tag.myProc == MyProc);
2515 :
2516 1641908 : lock = proclock->tag.myLock;
2517 :
2518 : /* Ignore items that are not of the lockmethod to be removed */
2519 1641908 : if (LOCK_LOCKMETHOD(*lock) != lockmethodid)
2520 301902 : continue;
2521 :
2522 : /*
2523 : * In allLocks mode, force release of all locks even if locallock
2524 : * table had problems
2525 : */
2526 1340006 : if (allLocks)
2527 87336 : proclock->releaseMask = proclock->holdMask;
2528 : else
2529 : Assert((proclock->releaseMask & ~proclock->holdMask) == 0);
2530 :
2531 : /*
2532 : * Ignore items that have nothing to be released, unless they have
2533 : * holdMask == 0 and are therefore recyclable
2534 : */
2535 1340006 : if (proclock->releaseMask == 0 && proclock->holdMask != 0)
2536 299966 : continue;
2537 :
2538 : PROCLOCK_PRINT("LockReleaseAll", proclock);
2539 : LOCK_PRINT("LockReleaseAll", lock, 0);
2540 : Assert(lock->nRequested >= 0);
2541 : Assert(lock->nGranted >= 0);
2542 : Assert(lock->nGranted <= lock->nRequested);
2543 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
2544 :
2545 : /*
2546 : * Release the previously-marked lock modes
2547 : */
2548 9360360 : for (i = 1; i <= numLockModes; i++)
2549 : {
2550 8320320 : if (proclock->releaseMask & LOCKBIT_ON(i))
2551 1111078 : wakeupNeeded |= UnGrantLock(lock, i, proclock,
2552 : lockMethodTable);
2553 : }
2554 : Assert((lock->nRequested >= 0) && (lock->nGranted >= 0));
2555 : Assert(lock->nGranted <= lock->nRequested);
2556 : LOCK_PRINT("LockReleaseAll: updated", lock, 0);
2557 :
2558 1040040 : proclock->releaseMask = 0;
2559 :
2560 : /* CleanUpLock will wake up waiters if needed. */
2561 1040040 : CleanUpLock(lock, proclock,
2562 : lockMethodTable,
2563 1040040 : LockTagHashCode(&lock->tag),
2564 : wakeupNeeded);
2565 : } /* loop over PROCLOCKs within this partition */
2566 :
2567 1411036 : LWLockRelease(partitionLock);
2568 : } /* loop over partitions */
2569 :
2570 : #ifdef LOCK_DEBUG
2571 : if (*(lockMethodTable->trace_flag))
2572 : elog(LOG, "LockReleaseAll done");
2573 : #endif
2574 2356762 : }
2575 :
2576 : /*
2577 : * LockReleaseSession -- Release all session locks of the specified lock method
2578 : * that are held by the current process.
2579 : */
2580 : void
2581 238 : LockReleaseSession(LOCKMETHODID lockmethodid)
2582 : {
2583 : HASH_SEQ_STATUS status;
2584 : LOCALLOCK *locallock;
2585 :
2586 238 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
2587 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
2588 :
2589 238 : hash_seq_init(&status, LockMethodLocalHash);
2590 :
2591 452 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2592 : {
2593 : /* Ignore items that are not of the specified lock method */
2594 214 : if (LOCALLOCK_LOCKMETHOD(*locallock) != lockmethodid)
2595 20 : continue;
2596 :
2597 194 : ReleaseLockIfHeld(locallock, true);
2598 : }
2599 238 : }
2600 :
2601 : /*
2602 : * LockReleaseCurrentOwner
2603 : * Release all locks belonging to CurrentResourceOwner
2604 : *
2605 : * If the caller knows what those locks are, it can pass them as an array.
2606 : * That speeds up the call significantly, when a lot of locks are held.
2607 : * Otherwise, pass NULL for locallocks, and we'll traverse through our hash
2608 : * table to find them.
2609 : */
2610 : void
2611 10682 : LockReleaseCurrentOwner(LOCALLOCK **locallocks, int nlocks)
2612 : {
2613 10682 : if (locallocks == NULL)
2614 : {
2615 : HASH_SEQ_STATUS status;
2616 : LOCALLOCK *locallock;
2617 :
2618 8 : hash_seq_init(&status, LockMethodLocalHash);
2619 :
2620 544 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2621 536 : ReleaseLockIfHeld(locallock, false);
2622 : }
2623 : else
2624 : {
2625 : int i;
2626 :
2627 15750 : for (i = nlocks - 1; i >= 0; i--)
2628 5076 : ReleaseLockIfHeld(locallocks[i], false);
2629 : }
2630 10682 : }
2631 :
2632 : /*
2633 : * ReleaseLockIfHeld
2634 : * Release any session-level locks on this lockable object if sessionLock
2635 : * is true; else, release any locks held by CurrentResourceOwner.
2636 : *
2637 : * It is tempting to pass this a ResourceOwner pointer (or NULL for session
2638 : * locks), but without refactoring LockRelease() we cannot support releasing
2639 : * locks belonging to resource owners other than CurrentResourceOwner.
2640 : * If we were to refactor, it'd be a good idea to fix it so we don't have to
2641 : * do a hashtable lookup of the locallock, too. However, currently this
2642 : * function isn't used heavily enough to justify refactoring for its
2643 : * convenience.
2644 : */
2645 : static void
2646 5806 : ReleaseLockIfHeld(LOCALLOCK *locallock, bool sessionLock)
2647 : {
2648 : ResourceOwner owner;
2649 : LOCALLOCKOWNER *lockOwners;
2650 : int i;
2651 :
2652 : /* Identify owner for lock (must match LockRelease!) */
2653 5806 : if (sessionLock)
2654 194 : owner = NULL;
2655 : else
2656 5612 : owner = CurrentResourceOwner;
2657 :
2658 : /* Scan to see if there are any locks belonging to the target owner */
2659 5806 : lockOwners = locallock->lockOwners;
2660 6192 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
2661 : {
2662 5806 : if (lockOwners[i].owner == owner)
2663 : {
2664 : Assert(lockOwners[i].nLocks > 0);
2665 5420 : if (lockOwners[i].nLocks < locallock->nLocks)
2666 : {
2667 : /*
2668 : * We will still hold this lock after forgetting this
2669 : * ResourceOwner.
2670 : */
2671 1396 : locallock->nLocks -= lockOwners[i].nLocks;
2672 : /* compact out unused slot */
2673 1396 : locallock->numLockOwners--;
2674 1396 : if (owner != NULL)
2675 1396 : ResourceOwnerForgetLock(owner, locallock);
2676 1396 : if (i < locallock->numLockOwners)
2677 0 : lockOwners[i] = lockOwners[locallock->numLockOwners];
2678 : }
2679 : else
2680 : {
2681 : Assert(lockOwners[i].nLocks == locallock->nLocks);
2682 : /* We want to call LockRelease just once */
2683 4024 : lockOwners[i].nLocks = 1;
2684 4024 : locallock->nLocks = 1;
2685 4024 : if (!LockRelease(&locallock->tag.lock,
2686 : locallock->tag.mode,
2687 : sessionLock))
2688 0 : elog(WARNING, "ReleaseLockIfHeld: failed??");
2689 : }
2690 5420 : break;
2691 : }
2692 : }
2693 5806 : }
2694 :
2695 : /*
2696 : * LockReassignCurrentOwner
2697 : * Reassign all locks belonging to CurrentResourceOwner to belong
2698 : * to its parent resource owner.
2699 : *
2700 : * If the caller knows what those locks are, it can pass them as an array.
2701 : * That speeds up the call significantly, when a lot of locks are held
2702 : * (e.g pg_dump with a large schema). Otherwise, pass NULL for locallocks,
2703 : * and we'll traverse through our hash table to find them.
2704 : */
2705 : void
2706 703978 : LockReassignCurrentOwner(LOCALLOCK **locallocks, int nlocks)
2707 : {
2708 703978 : ResourceOwner parent = ResourceOwnerGetParent(CurrentResourceOwner);
2709 :
2710 : Assert(parent != NULL);
2711 :
2712 703978 : if (locallocks == NULL)
2713 : {
2714 : HASH_SEQ_STATUS status;
2715 : LOCALLOCK *locallock;
2716 :
2717 7072 : hash_seq_init(&status, LockMethodLocalHash);
2718 :
2719 222832 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
2720 215760 : LockReassignOwner(locallock, parent);
2721 : }
2722 : else
2723 : {
2724 : int i;
2725 :
2726 1492572 : for (i = nlocks - 1; i >= 0; i--)
2727 795666 : LockReassignOwner(locallocks[i], parent);
2728 : }
2729 703978 : }
2730 :
2731 : /*
2732 : * Subroutine of LockReassignCurrentOwner. Reassigns a given lock belonging to
2733 : * CurrentResourceOwner to its parent.
2734 : */
2735 : static void
2736 1011426 : LockReassignOwner(LOCALLOCK *locallock, ResourceOwner parent)
2737 : {
2738 : LOCALLOCKOWNER *lockOwners;
2739 : int i;
2740 1011426 : int ic = -1;
2741 1011426 : int ip = -1;
2742 :
2743 : /*
2744 : * Scan to see if there are any locks belonging to current owner or its
2745 : * parent
2746 : */
2747 1011426 : lockOwners = locallock->lockOwners;
2748 2350046 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
2749 : {
2750 1338620 : if (lockOwners[i].owner == CurrentResourceOwner)
2751 971268 : ic = i;
2752 367352 : else if (lockOwners[i].owner == parent)
2753 282902 : ip = i;
2754 : }
2755 :
2756 1011426 : if (ic < 0)
2757 40158 : return; /* no current locks */
2758 :
2759 971268 : if (ip < 0)
2760 : {
2761 : /* Parent has no slot, so just give it the child's slot */
2762 728458 : lockOwners[ic].owner = parent;
2763 728458 : ResourceOwnerRememberLock(parent, locallock);
2764 : }
2765 : else
2766 : {
2767 : /* Merge child's count with parent's */
2768 242810 : lockOwners[ip].nLocks += lockOwners[ic].nLocks;
2769 : /* compact out unused slot */
2770 242810 : locallock->numLockOwners--;
2771 242810 : if (ic < locallock->numLockOwners)
2772 1542 : lockOwners[ic] = lockOwners[locallock->numLockOwners];
2773 : }
2774 971268 : ResourceOwnerForgetLock(CurrentResourceOwner, locallock);
2775 : }
2776 :
2777 : /*
2778 : * FastPathGrantRelationLock
2779 : * Grant lock using per-backend fast-path array, if there is space.
2780 : */
2781 : static bool
2782 34120110 : FastPathGrantRelationLock(Oid relid, LOCKMODE lockmode)
2783 : {
2784 : uint32 i;
2785 34120110 : uint32 unused_slot = FastPathLockSlotsPerBackend();
2786 :
2787 : /* fast-path group the lock belongs to */
2788 34120110 : uint32 group = FAST_PATH_REL_GROUP(relid);
2789 :
2790 : /* Scan for existing entry for this relid, remembering empty slot. */
2791 578804144 : for (i = 0; i < FP_LOCK_SLOTS_PER_GROUP; i++)
2792 : {
2793 : /* index into the whole per-backend array */
2794 545559370 : uint32 f = FAST_PATH_SLOT(group, i);
2795 :
2796 545559370 : if (FAST_PATH_GET_BITS(MyProc, f) == 0)
2797 527786528 : unused_slot = f;
2798 17772842 : else if (MyProc->fpRelId[f] == relid)
2799 : {
2800 : Assert(!FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode));
2801 875336 : FAST_PATH_SET_LOCKMODE(MyProc, f, lockmode);
2802 875336 : return true;
2803 : }
2804 : }
2805 :
2806 : /* If no existing entry, use any empty slot. */
2807 33244774 : if (unused_slot < FastPathLockSlotsPerBackend())
2808 : {
2809 33244774 : MyProc->fpRelId[unused_slot] = relid;
2810 33244774 : FAST_PATH_SET_LOCKMODE(MyProc, unused_slot, lockmode);
2811 33244774 : ++FastPathLocalUseCounts[group];
2812 33244774 : return true;
2813 : }
2814 :
2815 : /* No existing entry, and no empty slot. */
2816 0 : return false;
2817 : }
2818 :
2819 : /*
2820 : * FastPathUnGrantRelationLock
2821 : * Release fast-path lock, if present. Update backend-private local
2822 : * use count, while we're at it.
2823 : */
2824 : static bool
2825 34503080 : FastPathUnGrantRelationLock(Oid relid, LOCKMODE lockmode)
2826 : {
2827 : uint32 i;
2828 34503080 : bool result = false;
2829 :
2830 : /* fast-path group the lock belongs to */
2831 34503080 : uint32 group = FAST_PATH_REL_GROUP(relid);
2832 :
2833 34503080 : FastPathLocalUseCounts[group] = 0;
2834 586552360 : for (i = 0; i < FP_LOCK_SLOTS_PER_GROUP; i++)
2835 : {
2836 : /* index into the whole per-backend array */
2837 552049280 : uint32 f = FAST_PATH_SLOT(group, i);
2838 :
2839 552049280 : if (MyProc->fpRelId[f] == relid
2840 47943814 : && FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode))
2841 : {
2842 : Assert(!result);
2843 34117214 : FAST_PATH_CLEAR_LOCKMODE(MyProc, f, lockmode);
2844 34117214 : result = true;
2845 : /* we continue iterating so as to update FastPathLocalUseCount */
2846 : }
2847 552049280 : if (FAST_PATH_GET_BITS(MyProc, f) != 0)
2848 23335118 : ++FastPathLocalUseCounts[group];
2849 : }
2850 34503080 : return result;
2851 : }
2852 :
2853 : /*
2854 : * FastPathTransferRelationLocks
2855 : * Transfer locks matching the given lock tag from per-backend fast-path
2856 : * arrays to the shared hash table.
2857 : *
2858 : * Returns true if successful, false if ran out of shared memory.
2859 : */
2860 : static bool
2861 375308 : FastPathTransferRelationLocks(LockMethod lockMethodTable, const LOCKTAG *locktag,
2862 : uint32 hashcode)
2863 : {
2864 375308 : LWLock *partitionLock = LockHashPartitionLock(hashcode);
2865 375308 : Oid relid = locktag->locktag_field2;
2866 : uint32 i;
2867 :
2868 : /* fast-path group the lock belongs to */
2869 375308 : uint32 group = FAST_PATH_REL_GROUP(relid);
2870 :
2871 : /*
2872 : * Every PGPROC that can potentially hold a fast-path lock is present in
2873 : * ProcGlobal->allProcs. Prepared transactions are not, but any
2874 : * outstanding fast-path locks held by prepared transactions are
2875 : * transferred to the main lock table.
2876 : */
2877 54784828 : for (i = 0; i < ProcGlobal->allProcCount; i++)
2878 : {
2879 54409520 : PGPROC *proc = &ProcGlobal->allProcs[i];
2880 : uint32 j;
2881 :
2882 54409520 : LWLockAcquire(&proc->fpInfoLock, LW_EXCLUSIVE);
2883 :
2884 : /*
2885 : * If the target backend isn't referencing the same database as the
2886 : * lock, then we needn't examine the individual relation IDs at all;
2887 : * none of them can be relevant.
2888 : *
2889 : * proc->databaseId is set at backend startup time and never changes
2890 : * thereafter, so it might be safe to perform this test before
2891 : * acquiring &proc->fpInfoLock. In particular, it's certainly safe to
2892 : * assume that if the target backend holds any fast-path locks, it
2893 : * must have performed a memory-fencing operation (in particular, an
2894 : * LWLock acquisition) since setting proc->databaseId. However, it's
2895 : * less clear that our backend is certain to have performed a memory
2896 : * fencing operation since the other backend set proc->databaseId. So
2897 : * for now, we test it after acquiring the LWLock just to be safe.
2898 : *
2899 : * Also skip groups without any registered fast-path locks.
2900 : */
2901 54409520 : if (proc->databaseId != locktag->locktag_field1 ||
2902 19037294 : proc->fpLockBits[group] == 0)
2903 : {
2904 54041952 : LWLockRelease(&proc->fpInfoLock);
2905 54041952 : continue;
2906 : }
2907 :
2908 6246170 : for (j = 0; j < FP_LOCK_SLOTS_PER_GROUP; j++)
2909 : {
2910 : uint32 lockmode;
2911 :
2912 : /* index into the whole per-backend array */
2913 5880798 : uint32 f = FAST_PATH_SLOT(group, j);
2914 :
2915 : /* Look for an allocated slot matching the given relid. */
2916 5880798 : if (relid != proc->fpRelId[f] || FAST_PATH_GET_BITS(proc, f) == 0)
2917 5878602 : continue;
2918 :
2919 : /* Find or create lock object. */
2920 2196 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2921 2196 : for (lockmode = FAST_PATH_LOCKNUMBER_OFFSET;
2922 8784 : lockmode < FAST_PATH_LOCKNUMBER_OFFSET + FAST_PATH_BITS_PER_SLOT;
2923 6588 : ++lockmode)
2924 : {
2925 : PROCLOCK *proclock;
2926 :
2927 6588 : if (!FAST_PATH_CHECK_LOCKMODE(proc, f, lockmode))
2928 4280 : continue;
2929 2308 : proclock = SetupLockInTable(lockMethodTable, proc, locktag,
2930 : hashcode, lockmode);
2931 2308 : if (!proclock)
2932 : {
2933 0 : LWLockRelease(partitionLock);
2934 0 : LWLockRelease(&proc->fpInfoLock);
2935 0 : return false;
2936 : }
2937 2308 : GrantLock(proclock->tag.myLock, proclock, lockmode);
2938 2308 : FAST_PATH_CLEAR_LOCKMODE(proc, f, lockmode);
2939 : }
2940 2196 : LWLockRelease(partitionLock);
2941 :
2942 : /* No need to examine remaining slots. */
2943 2196 : break;
2944 : }
2945 367568 : LWLockRelease(&proc->fpInfoLock);
2946 : }
2947 375308 : return true;
2948 : }
2949 :
2950 : /*
2951 : * FastPathGetRelationLockEntry
2952 : * Return the PROCLOCK for a lock originally taken via the fast-path,
2953 : * transferring it to the primary lock table if necessary.
2954 : *
2955 : * Note: caller takes care of updating the locallock object.
2956 : */
2957 : static PROCLOCK *
2958 604 : FastPathGetRelationLockEntry(LOCALLOCK *locallock)
2959 : {
2960 604 : LockMethod lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
2961 604 : LOCKTAG *locktag = &locallock->tag.lock;
2962 604 : PROCLOCK *proclock = NULL;
2963 604 : LWLock *partitionLock = LockHashPartitionLock(locallock->hashcode);
2964 604 : Oid relid = locktag->locktag_field2;
2965 : uint32 i,
2966 : group;
2967 :
2968 : /* fast-path group the lock belongs to */
2969 604 : group = FAST_PATH_REL_GROUP(relid);
2970 :
2971 604 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
2972 :
2973 9666 : for (i = 0; i < FP_LOCK_SLOTS_PER_GROUP; i++)
2974 : {
2975 : uint32 lockmode;
2976 :
2977 : /* index into the whole per-backend array */
2978 9650 : uint32 f = FAST_PATH_SLOT(group, i);
2979 :
2980 : /* Look for an allocated slot matching the given relid. */
2981 9650 : if (relid != MyProc->fpRelId[f] || FAST_PATH_GET_BITS(MyProc, f) == 0)
2982 9062 : continue;
2983 :
2984 : /* If we don't have a lock of the given mode, forget it! */
2985 588 : lockmode = locallock->tag.mode;
2986 588 : if (!FAST_PATH_CHECK_LOCKMODE(MyProc, f, lockmode))
2987 0 : break;
2988 :
2989 : /* Find or create lock object. */
2990 588 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
2991 :
2992 588 : proclock = SetupLockInTable(lockMethodTable, MyProc, locktag,
2993 : locallock->hashcode, lockmode);
2994 588 : if (!proclock)
2995 : {
2996 0 : LWLockRelease(partitionLock);
2997 0 : LWLockRelease(&MyProc->fpInfoLock);
2998 0 : ereport(ERROR,
2999 : (errcode(ERRCODE_OUT_OF_MEMORY),
3000 : errmsg("out of shared memory"),
3001 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
3002 : }
3003 588 : GrantLock(proclock->tag.myLock, proclock, lockmode);
3004 588 : FAST_PATH_CLEAR_LOCKMODE(MyProc, f, lockmode);
3005 :
3006 588 : LWLockRelease(partitionLock);
3007 :
3008 : /* No need to examine remaining slots. */
3009 588 : break;
3010 : }
3011 :
3012 604 : LWLockRelease(&MyProc->fpInfoLock);
3013 :
3014 : /* Lock may have already been transferred by some other backend. */
3015 604 : if (proclock == NULL)
3016 : {
3017 : LOCK *lock;
3018 : PROCLOCKTAG proclocktag;
3019 : uint32 proclock_hashcode;
3020 :
3021 16 : LWLockAcquire(partitionLock, LW_SHARED);
3022 :
3023 16 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
3024 : locktag,
3025 : locallock->hashcode,
3026 : HASH_FIND,
3027 : NULL);
3028 16 : if (!lock)
3029 0 : elog(ERROR, "failed to re-find shared lock object");
3030 :
3031 16 : proclocktag.myLock = lock;
3032 16 : proclocktag.myProc = MyProc;
3033 :
3034 16 : proclock_hashcode = ProcLockHashCode(&proclocktag, locallock->hashcode);
3035 : proclock = (PROCLOCK *)
3036 16 : hash_search_with_hash_value(LockMethodProcLockHash,
3037 : &proclocktag,
3038 : proclock_hashcode,
3039 : HASH_FIND,
3040 : NULL);
3041 16 : if (!proclock)
3042 0 : elog(ERROR, "failed to re-find shared proclock object");
3043 16 : LWLockRelease(partitionLock);
3044 : }
3045 :
3046 604 : return proclock;
3047 : }
3048 :
3049 : /*
3050 : * GetLockConflicts
3051 : * Get an array of VirtualTransactionIds of xacts currently holding locks
3052 : * that would conflict with the specified lock/lockmode.
3053 : * xacts merely awaiting such a lock are NOT reported.
3054 : *
3055 : * The result array is palloc'd and is terminated with an invalid VXID.
3056 : * *countp, if not null, is updated to the number of items set.
3057 : *
3058 : * Of course, the result could be out of date by the time it's returned, so
3059 : * use of this function has to be thought about carefully. Similarly, a
3060 : * PGPROC with no "lxid" will be considered non-conflicting regardless of any
3061 : * lock it holds. Existing callers don't care about a locker after that
3062 : * locker's pg_xact updates complete. CommitTransaction() clears "lxid" after
3063 : * pg_xact updates and before releasing locks.
3064 : *
3065 : * Note we never include the current xact's vxid in the result array,
3066 : * since an xact never blocks itself.
3067 : */
3068 : VirtualTransactionId *
3069 2908 : GetLockConflicts(const LOCKTAG *locktag, LOCKMODE lockmode, int *countp)
3070 : {
3071 : static VirtualTransactionId *vxids;
3072 2908 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
3073 : LockMethod lockMethodTable;
3074 : LOCK *lock;
3075 : LOCKMASK conflictMask;
3076 : dlist_iter proclock_iter;
3077 : PROCLOCK *proclock;
3078 : uint32 hashcode;
3079 : LWLock *partitionLock;
3080 2908 : int count = 0;
3081 2908 : int fast_count = 0;
3082 :
3083 2908 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
3084 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
3085 2908 : lockMethodTable = LockMethods[lockmethodid];
3086 2908 : if (lockmode <= 0 || lockmode > lockMethodTable->numLockModes)
3087 0 : elog(ERROR, "unrecognized lock mode: %d", lockmode);
3088 :
3089 : /*
3090 : * Allocate memory to store results, and fill with InvalidVXID. We only
3091 : * need enough space for MaxBackends + max_prepared_xacts + a terminator.
3092 : * InHotStandby allocate once in TopMemoryContext.
3093 : */
3094 2908 : if (InHotStandby)
3095 : {
3096 8 : if (vxids == NULL)
3097 2 : vxids = (VirtualTransactionId *)
3098 2 : MemoryContextAlloc(TopMemoryContext,
3099 : sizeof(VirtualTransactionId) *
3100 2 : (MaxBackends + max_prepared_xacts + 1));
3101 : }
3102 : else
3103 2900 : vxids = (VirtualTransactionId *)
3104 2900 : palloc0(sizeof(VirtualTransactionId) *
3105 2900 : (MaxBackends + max_prepared_xacts + 1));
3106 :
3107 : /* Compute hash code and partition lock, and look up conflicting modes. */
3108 2908 : hashcode = LockTagHashCode(locktag);
3109 2908 : partitionLock = LockHashPartitionLock(hashcode);
3110 2908 : conflictMask = lockMethodTable->conflictTab[lockmode];
3111 :
3112 : /*
3113 : * Fast path locks might not have been entered in the primary lock table.
3114 : * If the lock we're dealing with could conflict with such a lock, we must
3115 : * examine each backend's fast-path array for conflicts.
3116 : */
3117 2908 : if (ConflictsWithRelationFastPath(locktag, lockmode))
3118 : {
3119 : int i;
3120 2908 : Oid relid = locktag->locktag_field2;
3121 : VirtualTransactionId vxid;
3122 :
3123 : /* fast-path group the lock belongs to */
3124 2908 : uint32 group = FAST_PATH_REL_GROUP(relid);
3125 :
3126 : /*
3127 : * Iterate over relevant PGPROCs. Anything held by a prepared
3128 : * transaction will have been transferred to the primary lock table,
3129 : * so we need not worry about those. This is all a bit fuzzy, because
3130 : * new locks could be taken after we've visited a particular
3131 : * partition, but the callers had better be prepared to deal with that
3132 : * anyway, since the locks could equally well be taken between the
3133 : * time we return the value and the time the caller does something
3134 : * with it.
3135 : */
3136 457092 : for (i = 0; i < ProcGlobal->allProcCount; i++)
3137 : {
3138 454184 : PGPROC *proc = &ProcGlobal->allProcs[i];
3139 : uint32 j;
3140 :
3141 : /* A backend never blocks itself */
3142 454184 : if (proc == MyProc)
3143 2908 : continue;
3144 :
3145 451276 : LWLockAcquire(&proc->fpInfoLock, LW_SHARED);
3146 :
3147 : /*
3148 : * If the target backend isn't referencing the same database as
3149 : * the lock, then we needn't examine the individual relation IDs
3150 : * at all; none of them can be relevant.
3151 : *
3152 : * See FastPathTransferRelationLocks() for discussion of why we do
3153 : * this test after acquiring the lock.
3154 : *
3155 : * Also skip groups without any registered fast-path locks.
3156 : */
3157 451276 : if (proc->databaseId != locktag->locktag_field1 ||
3158 175420 : proc->fpLockBits[group] == 0)
3159 : {
3160 450482 : LWLockRelease(&proc->fpInfoLock);
3161 450482 : continue;
3162 : }
3163 :
3164 13156 : for (j = 0; j < FP_LOCK_SLOTS_PER_GROUP; j++)
3165 : {
3166 : uint32 lockmask;
3167 :
3168 : /* index into the whole per-backend array */
3169 12680 : uint32 f = FAST_PATH_SLOT(group, j);
3170 :
3171 : /* Look for an allocated slot matching the given relid. */
3172 12680 : if (relid != proc->fpRelId[f])
3173 12362 : continue;
3174 318 : lockmask = FAST_PATH_GET_BITS(proc, f);
3175 318 : if (!lockmask)
3176 0 : continue;
3177 318 : lockmask <<= FAST_PATH_LOCKNUMBER_OFFSET;
3178 :
3179 : /*
3180 : * There can only be one entry per relation, so if we found it
3181 : * and it doesn't conflict, we can skip the rest of the slots.
3182 : */
3183 318 : if ((lockmask & conflictMask) == 0)
3184 10 : break;
3185 :
3186 : /* Conflict! */
3187 308 : GET_VXID_FROM_PGPROC(vxid, *proc);
3188 :
3189 308 : if (VirtualTransactionIdIsValid(vxid))
3190 308 : vxids[count++] = vxid;
3191 : /* else, xact already committed or aborted */
3192 :
3193 : /* No need to examine remaining slots. */
3194 308 : break;
3195 : }
3196 :
3197 794 : LWLockRelease(&proc->fpInfoLock);
3198 : }
3199 : }
3200 :
3201 : /* Remember how many fast-path conflicts we found. */
3202 2908 : fast_count = count;
3203 :
3204 : /*
3205 : * Look up the lock object matching the tag.
3206 : */
3207 2908 : LWLockAcquire(partitionLock, LW_SHARED);
3208 :
3209 2908 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
3210 : locktag,
3211 : hashcode,
3212 : HASH_FIND,
3213 : NULL);
3214 2908 : if (!lock)
3215 : {
3216 : /*
3217 : * If the lock object doesn't exist, there is nothing holding a lock
3218 : * on this lockable object.
3219 : */
3220 140 : LWLockRelease(partitionLock);
3221 140 : vxids[count].procNumber = INVALID_PROC_NUMBER;
3222 140 : vxids[count].localTransactionId = InvalidLocalTransactionId;
3223 140 : if (countp)
3224 0 : *countp = count;
3225 140 : return vxids;
3226 : }
3227 :
3228 : /*
3229 : * Examine each existing holder (or awaiter) of the lock.
3230 : */
3231 5554 : dlist_foreach(proclock_iter, &lock->procLocks)
3232 : {
3233 2786 : proclock = dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
3234 :
3235 2786 : if (conflictMask & proclock->holdMask)
3236 : {
3237 2778 : PGPROC *proc = proclock->tag.myProc;
3238 :
3239 : /* A backend never blocks itself */
3240 2778 : if (proc != MyProc)
3241 : {
3242 : VirtualTransactionId vxid;
3243 :
3244 18 : GET_VXID_FROM_PGPROC(vxid, *proc);
3245 :
3246 18 : if (VirtualTransactionIdIsValid(vxid))
3247 : {
3248 : int i;
3249 :
3250 : /* Avoid duplicate entries. */
3251 24 : for (i = 0; i < fast_count; ++i)
3252 6 : if (VirtualTransactionIdEquals(vxids[i], vxid))
3253 0 : break;
3254 18 : if (i >= fast_count)
3255 18 : vxids[count++] = vxid;
3256 : }
3257 : /* else, xact already committed or aborted */
3258 : }
3259 : }
3260 : }
3261 :
3262 2768 : LWLockRelease(partitionLock);
3263 :
3264 2768 : if (count > MaxBackends + max_prepared_xacts) /* should never happen */
3265 0 : elog(PANIC, "too many conflicting locks found");
3266 :
3267 2768 : vxids[count].procNumber = INVALID_PROC_NUMBER;
3268 2768 : vxids[count].localTransactionId = InvalidLocalTransactionId;
3269 2768 : if (countp)
3270 2762 : *countp = count;
3271 2768 : return vxids;
3272 : }
3273 :
3274 : /*
3275 : * Find a lock in the shared lock table and release it. It is the caller's
3276 : * responsibility to verify that this is a sane thing to do. (For example, it
3277 : * would be bad to release a lock here if there might still be a LOCALLOCK
3278 : * object with pointers to it.)
3279 : *
3280 : * We currently use this in two situations: first, to release locks held by
3281 : * prepared transactions on commit (see lock_twophase_postcommit); and second,
3282 : * to release locks taken via the fast-path, transferred to the main hash
3283 : * table, and then released (see LockReleaseAll).
3284 : */
3285 : static void
3286 4160 : LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
3287 : LOCKTAG *locktag, LOCKMODE lockmode,
3288 : bool decrement_strong_lock_count)
3289 : {
3290 : LOCK *lock;
3291 : PROCLOCK *proclock;
3292 : PROCLOCKTAG proclocktag;
3293 : uint32 hashcode;
3294 : uint32 proclock_hashcode;
3295 : LWLock *partitionLock;
3296 : bool wakeupNeeded;
3297 :
3298 4160 : hashcode = LockTagHashCode(locktag);
3299 4160 : partitionLock = LockHashPartitionLock(hashcode);
3300 :
3301 4160 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3302 :
3303 : /*
3304 : * Re-find the lock object (it had better be there).
3305 : */
3306 4160 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
3307 : locktag,
3308 : hashcode,
3309 : HASH_FIND,
3310 : NULL);
3311 4160 : if (!lock)
3312 0 : elog(PANIC, "failed to re-find shared lock object");
3313 :
3314 : /*
3315 : * Re-find the proclock object (ditto).
3316 : */
3317 4160 : proclocktag.myLock = lock;
3318 4160 : proclocktag.myProc = proc;
3319 :
3320 4160 : proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
3321 :
3322 4160 : proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
3323 : &proclocktag,
3324 : proclock_hashcode,
3325 : HASH_FIND,
3326 : NULL);
3327 4160 : if (!proclock)
3328 0 : elog(PANIC, "failed to re-find shared proclock object");
3329 :
3330 : /*
3331 : * Double-check that we are actually holding a lock of the type we want to
3332 : * release.
3333 : */
3334 4160 : if (!(proclock->holdMask & LOCKBIT_ON(lockmode)))
3335 : {
3336 : PROCLOCK_PRINT("lock_twophase_postcommit: WRONGTYPE", proclock);
3337 0 : LWLockRelease(partitionLock);
3338 0 : elog(WARNING, "you don't own a lock of type %s",
3339 : lockMethodTable->lockModeNames[lockmode]);
3340 0 : return;
3341 : }
3342 :
3343 : /*
3344 : * Do the releasing. CleanUpLock will waken any now-wakable waiters.
3345 : */
3346 4160 : wakeupNeeded = UnGrantLock(lock, lockmode, proclock, lockMethodTable);
3347 :
3348 4160 : CleanUpLock(lock, proclock,
3349 : lockMethodTable, hashcode,
3350 : wakeupNeeded);
3351 :
3352 4160 : LWLockRelease(partitionLock);
3353 :
3354 : /*
3355 : * Decrement strong lock count. This logic is needed only for 2PC.
3356 : */
3357 4160 : if (decrement_strong_lock_count
3358 1440 : && ConflictsWithRelationFastPath(locktag, lockmode))
3359 : {
3360 142 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
3361 :
3362 142 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
3363 : Assert(FastPathStrongRelationLocks->count[fasthashcode] > 0);
3364 142 : FastPathStrongRelationLocks->count[fasthashcode]--;
3365 142 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
3366 : }
3367 : }
3368 :
3369 : /*
3370 : * CheckForSessionAndXactLocks
3371 : * Check to see if transaction holds both session-level and xact-level
3372 : * locks on the same object; if so, throw an error.
3373 : *
3374 : * If we have both session- and transaction-level locks on the same object,
3375 : * PREPARE TRANSACTION must fail. This should never happen with regular
3376 : * locks, since we only take those at session level in some special operations
3377 : * like VACUUM. It's possible to hit this with advisory locks, though.
3378 : *
3379 : * It would be nice if we could keep the session hold and give away the
3380 : * transactional hold to the prepared xact. However, that would require two
3381 : * PROCLOCK objects, and we cannot be sure that another PROCLOCK will be
3382 : * available when it comes time for PostPrepare_Locks to do the deed.
3383 : * So for now, we error out while we can still do so safely.
3384 : *
3385 : * Since the LOCALLOCK table stores a separate entry for each lockmode,
3386 : * we can't implement this check by examining LOCALLOCK entries in isolation.
3387 : * We must build a transient hashtable that is indexed by locktag only.
3388 : */
3389 : static void
3390 584 : CheckForSessionAndXactLocks(void)
3391 : {
3392 : typedef struct
3393 : {
3394 : LOCKTAG lock; /* identifies the lockable object */
3395 : bool sessLock; /* is any lockmode held at session level? */
3396 : bool xactLock; /* is any lockmode held at xact level? */
3397 : } PerLockTagEntry;
3398 :
3399 : HASHCTL hash_ctl;
3400 : HTAB *lockhtab;
3401 : HASH_SEQ_STATUS status;
3402 : LOCALLOCK *locallock;
3403 :
3404 : /* Create a local hash table keyed by LOCKTAG only */
3405 584 : hash_ctl.keysize = sizeof(LOCKTAG);
3406 584 : hash_ctl.entrysize = sizeof(PerLockTagEntry);
3407 584 : hash_ctl.hcxt = CurrentMemoryContext;
3408 :
3409 584 : lockhtab = hash_create("CheckForSessionAndXactLocks table",
3410 : 256, /* arbitrary initial size */
3411 : &hash_ctl,
3412 : HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
3413 :
3414 : /* Scan local lock table to find entries for each LOCKTAG */
3415 584 : hash_seq_init(&status, LockMethodLocalHash);
3416 :
3417 1996 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3418 : {
3419 1416 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3420 : PerLockTagEntry *hentry;
3421 : bool found;
3422 : int i;
3423 :
3424 : /*
3425 : * Ignore VXID locks. We don't want those to be held by prepared
3426 : * transactions, since they aren't meaningful after a restart.
3427 : */
3428 1416 : if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3429 0 : continue;
3430 :
3431 : /* Ignore it if we don't actually hold the lock */
3432 1416 : if (locallock->nLocks <= 0)
3433 0 : continue;
3434 :
3435 : /* Otherwise, find or make an entry in lockhtab */
3436 1416 : hentry = (PerLockTagEntry *) hash_search(lockhtab,
3437 1416 : &locallock->tag.lock,
3438 : HASH_ENTER, &found);
3439 1416 : if (!found) /* initialize, if newly created */
3440 1318 : hentry->sessLock = hentry->xactLock = false;
3441 :
3442 : /* Scan to see if we hold lock at session or xact level or both */
3443 2832 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
3444 : {
3445 1416 : if (lockOwners[i].owner == NULL)
3446 18 : hentry->sessLock = true;
3447 : else
3448 1398 : hentry->xactLock = true;
3449 : }
3450 :
3451 : /*
3452 : * We can throw error immediately when we see both types of locks; no
3453 : * need to wait around to see if there are more violations.
3454 : */
3455 1416 : if (hentry->sessLock && hentry->xactLock)
3456 4 : ereport(ERROR,
3457 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3458 : errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3459 : }
3460 :
3461 : /* Success, so clean up */
3462 580 : hash_destroy(lockhtab);
3463 580 : }
3464 :
3465 : /*
3466 : * AtPrepare_Locks
3467 : * Do the preparatory work for a PREPARE: make 2PC state file records
3468 : * for all locks currently held.
3469 : *
3470 : * Session-level locks are ignored, as are VXID locks.
3471 : *
3472 : * For the most part, we don't need to touch shared memory for this ---
3473 : * all the necessary state information is in the locallock table.
3474 : * Fast-path locks are an exception, however: we move any such locks to
3475 : * the main table before allowing PREPARE TRANSACTION to succeed.
3476 : */
3477 : void
3478 584 : AtPrepare_Locks(void)
3479 : {
3480 : HASH_SEQ_STATUS status;
3481 : LOCALLOCK *locallock;
3482 :
3483 : /* First, verify there aren't locks of both xact and session level */
3484 584 : CheckForSessionAndXactLocks();
3485 :
3486 : /* Now do the per-locallock cleanup work */
3487 580 : hash_seq_init(&status, LockMethodLocalHash);
3488 :
3489 1984 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3490 : {
3491 : TwoPhaseLockRecord record;
3492 1404 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3493 : bool haveSessionLock;
3494 : bool haveXactLock;
3495 : int i;
3496 :
3497 : /*
3498 : * Ignore VXID locks. We don't want those to be held by prepared
3499 : * transactions, since they aren't meaningful after a restart.
3500 : */
3501 1404 : if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3502 14 : continue;
3503 :
3504 : /* Ignore it if we don't actually hold the lock */
3505 1404 : if (locallock->nLocks <= 0)
3506 0 : continue;
3507 :
3508 : /* Scan to see whether we hold it at session or transaction level */
3509 1404 : haveSessionLock = haveXactLock = false;
3510 2808 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
3511 : {
3512 1404 : if (lockOwners[i].owner == NULL)
3513 14 : haveSessionLock = true;
3514 : else
3515 1390 : haveXactLock = true;
3516 : }
3517 :
3518 : /* Ignore it if we have only session lock */
3519 1404 : if (!haveXactLock)
3520 14 : continue;
3521 :
3522 : /* This can't happen, because we already checked it */
3523 1390 : if (haveSessionLock)
3524 0 : ereport(ERROR,
3525 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3526 : errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3527 :
3528 : /*
3529 : * If the local lock was taken via the fast-path, we need to move it
3530 : * to the primary lock table, or just get a pointer to the existing
3531 : * primary lock table entry if by chance it's already been
3532 : * transferred.
3533 : */
3534 1390 : if (locallock->proclock == NULL)
3535 : {
3536 604 : locallock->proclock = FastPathGetRelationLockEntry(locallock);
3537 604 : locallock->lock = locallock->proclock->tag.myLock;
3538 : }
3539 :
3540 : /*
3541 : * Arrange to not release any strong lock count held by this lock
3542 : * entry. We must retain the count until the prepared transaction is
3543 : * committed or rolled back.
3544 : */
3545 1390 : locallock->holdsStrongLockCount = false;
3546 :
3547 : /*
3548 : * Create a 2PC record.
3549 : */
3550 1390 : memcpy(&(record.locktag), &(locallock->tag.lock), sizeof(LOCKTAG));
3551 1390 : record.lockmode = locallock->tag.mode;
3552 :
3553 1390 : RegisterTwoPhaseRecord(TWOPHASE_RM_LOCK_ID, 0,
3554 : &record, sizeof(TwoPhaseLockRecord));
3555 : }
3556 580 : }
3557 :
3558 : /*
3559 : * PostPrepare_Locks
3560 : * Clean up after successful PREPARE
3561 : *
3562 : * Here, we want to transfer ownership of our locks to a dummy PGPROC
3563 : * that's now associated with the prepared transaction, and we want to
3564 : * clean out the corresponding entries in the LOCALLOCK table.
3565 : *
3566 : * Note: by removing the LOCALLOCK entries, we are leaving dangling
3567 : * pointers in the transaction's resource owner. This is OK at the
3568 : * moment since resowner.c doesn't try to free locks retail at a toplevel
3569 : * transaction commit or abort. We could alternatively zero out nLocks
3570 : * and leave the LOCALLOCK entries to be garbage-collected by LockReleaseAll,
3571 : * but that probably costs more cycles.
3572 : */
3573 : void
3574 580 : PostPrepare_Locks(FullTransactionId fxid)
3575 : {
3576 580 : PGPROC *newproc = TwoPhaseGetDummyProc(fxid, false);
3577 : HASH_SEQ_STATUS status;
3578 : LOCALLOCK *locallock;
3579 : LOCK *lock;
3580 : PROCLOCK *proclock;
3581 : PROCLOCKTAG proclocktag;
3582 : int partition;
3583 :
3584 : /* Can't prepare a lock group follower. */
3585 : Assert(MyProc->lockGroupLeader == NULL ||
3586 : MyProc->lockGroupLeader == MyProc);
3587 :
3588 : /* This is a critical section: any error means big trouble */
3589 580 : START_CRIT_SECTION();
3590 :
3591 : /*
3592 : * First we run through the locallock table and get rid of unwanted
3593 : * entries, then we scan the process's proclocks and transfer them to the
3594 : * target proc.
3595 : *
3596 : * We do this separately because we may have multiple locallock entries
3597 : * pointing to the same proclock, and we daren't end up with any dangling
3598 : * pointers.
3599 : */
3600 580 : hash_seq_init(&status, LockMethodLocalHash);
3601 :
3602 1984 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3603 : {
3604 1404 : LOCALLOCKOWNER *lockOwners = locallock->lockOwners;
3605 : bool haveSessionLock;
3606 : bool haveXactLock;
3607 : int i;
3608 :
3609 1404 : if (locallock->proclock == NULL || locallock->lock == NULL)
3610 : {
3611 : /*
3612 : * We must've run out of shared memory while trying to set up this
3613 : * lock. Just forget the local entry.
3614 : */
3615 : Assert(locallock->nLocks == 0);
3616 0 : RemoveLocalLock(locallock);
3617 0 : continue;
3618 : }
3619 :
3620 : /* Ignore VXID locks */
3621 1404 : if (locallock->tag.lock.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3622 0 : continue;
3623 :
3624 : /* Scan to see whether we hold it at session or transaction level */
3625 1404 : haveSessionLock = haveXactLock = false;
3626 2808 : for (i = locallock->numLockOwners - 1; i >= 0; i--)
3627 : {
3628 1404 : if (lockOwners[i].owner == NULL)
3629 14 : haveSessionLock = true;
3630 : else
3631 1390 : haveXactLock = true;
3632 : }
3633 :
3634 : /* Ignore it if we have only session lock */
3635 1404 : if (!haveXactLock)
3636 14 : continue;
3637 :
3638 : /* This can't happen, because we already checked it */
3639 1390 : if (haveSessionLock)
3640 0 : ereport(PANIC,
3641 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3642 : errmsg("cannot PREPARE while holding both session-level and transaction-level locks on the same object")));
3643 :
3644 : /* Mark the proclock to show we need to release this lockmode */
3645 1390 : if (locallock->nLocks > 0)
3646 1390 : locallock->proclock->releaseMask |= LOCKBIT_ON(locallock->tag.mode);
3647 :
3648 : /* And remove the locallock hashtable entry */
3649 1390 : RemoveLocalLock(locallock);
3650 : }
3651 :
3652 : /*
3653 : * Now, scan each lock partition separately.
3654 : */
3655 9860 : for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
3656 : {
3657 : LWLock *partitionLock;
3658 9280 : dlist_head *procLocks = &(MyProc->myProcLocks[partition]);
3659 : dlist_mutable_iter proclock_iter;
3660 :
3661 9280 : partitionLock = LockHashPartitionLockByIndex(partition);
3662 :
3663 : /*
3664 : * If the proclock list for this partition is empty, we can skip
3665 : * acquiring the partition lock. This optimization is safer than the
3666 : * situation in LockReleaseAll, because we got rid of any fast-path
3667 : * locks during AtPrepare_Locks, so there cannot be any case where
3668 : * another backend is adding something to our lists now. For safety,
3669 : * though, we code this the same way as in LockReleaseAll.
3670 : */
3671 9280 : if (dlist_is_empty(procLocks))
3672 7994 : continue; /* needn't examine this partition */
3673 :
3674 1286 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
3675 :
3676 2646 : dlist_foreach_modify(proclock_iter, procLocks)
3677 : {
3678 1360 : proclock = dlist_container(PROCLOCK, procLink, proclock_iter.cur);
3679 :
3680 : Assert(proclock->tag.myProc == MyProc);
3681 :
3682 1360 : lock = proclock->tag.myLock;
3683 :
3684 : /* Ignore VXID locks */
3685 1360 : if (lock->tag.locktag_type == LOCKTAG_VIRTUALTRANSACTION)
3686 50 : continue;
3687 :
3688 : PROCLOCK_PRINT("PostPrepare_Locks", proclock);
3689 : LOCK_PRINT("PostPrepare_Locks", lock, 0);
3690 : Assert(lock->nRequested >= 0);
3691 : Assert(lock->nGranted >= 0);
3692 : Assert(lock->nGranted <= lock->nRequested);
3693 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
3694 :
3695 : /* Ignore it if nothing to release (must be a session lock) */
3696 1310 : if (proclock->releaseMask == 0)
3697 14 : continue;
3698 :
3699 : /* Else we should be releasing all locks */
3700 1296 : if (proclock->releaseMask != proclock->holdMask)
3701 0 : elog(PANIC, "we seem to have dropped a bit somewhere");
3702 :
3703 : /*
3704 : * We cannot simply modify proclock->tag.myProc to reassign
3705 : * ownership of the lock, because that's part of the hash key and
3706 : * the proclock would then be in the wrong hash chain. Instead
3707 : * use hash_update_hash_key. (We used to create a new hash entry,
3708 : * but that risks out-of-memory failure if other processes are
3709 : * busy making proclocks too.) We must unlink the proclock from
3710 : * our procLink chain and put it into the new proc's chain, too.
3711 : *
3712 : * Note: the updated proclock hash key will still belong to the
3713 : * same hash partition, cf proclock_hash(). So the partition lock
3714 : * we already hold is sufficient for this.
3715 : */
3716 1296 : dlist_delete(&proclock->procLink);
3717 :
3718 : /*
3719 : * Create the new hash key for the proclock.
3720 : */
3721 1296 : proclocktag.myLock = lock;
3722 1296 : proclocktag.myProc = newproc;
3723 :
3724 : /*
3725 : * Update groupLeader pointer to point to the new proc. (We'd
3726 : * better not be a member of somebody else's lock group!)
3727 : */
3728 : Assert(proclock->groupLeader == proclock->tag.myProc);
3729 1296 : proclock->groupLeader = newproc;
3730 :
3731 : /*
3732 : * Update the proclock. We should not find any existing entry for
3733 : * the same hash key, since there can be only one entry for any
3734 : * given lock with my own proc.
3735 : */
3736 1296 : if (!hash_update_hash_key(LockMethodProcLockHash,
3737 : proclock,
3738 : &proclocktag))
3739 0 : elog(PANIC, "duplicate entry found while reassigning a prepared transaction's locks");
3740 :
3741 : /* Re-link into the new proc's proclock list */
3742 1296 : dlist_push_tail(&newproc->myProcLocks[partition], &proclock->procLink);
3743 :
3744 : PROCLOCK_PRINT("PostPrepare_Locks: updated", proclock);
3745 : } /* loop over PROCLOCKs within this partition */
3746 :
3747 1286 : LWLockRelease(partitionLock);
3748 : } /* loop over partitions */
3749 :
3750 580 : END_CRIT_SECTION();
3751 580 : }
3752 :
3753 :
3754 : /*
3755 : * Estimate shared-memory space used for lock tables
3756 : */
3757 : Size
3758 4088 : LockManagerShmemSize(void)
3759 : {
3760 4088 : Size size = 0;
3761 : long max_table_size;
3762 :
3763 : /* lock hash table */
3764 4088 : max_table_size = NLOCKENTS();
3765 4088 : size = add_size(size, hash_estimate_size(max_table_size, sizeof(LOCK)));
3766 :
3767 : /* proclock hash table */
3768 4088 : max_table_size *= 2;
3769 4088 : size = add_size(size, hash_estimate_size(max_table_size, sizeof(PROCLOCK)));
3770 :
3771 : /*
3772 : * Since NLOCKENTS is only an estimate, add 10% safety margin.
3773 : */
3774 4088 : size = add_size(size, size / 10);
3775 :
3776 4088 : return size;
3777 : }
3778 :
3779 : /*
3780 : * GetLockStatusData - Return a summary of the lock manager's internal
3781 : * status, for use in a user-level reporting function.
3782 : *
3783 : * The return data consists of an array of LockInstanceData objects,
3784 : * which are a lightly abstracted version of the PROCLOCK data structures,
3785 : * i.e. there is one entry for each unique lock and interested PGPROC.
3786 : * It is the caller's responsibility to match up related items (such as
3787 : * references to the same lockable object or PGPROC) if wanted.
3788 : *
3789 : * The design goal is to hold the LWLocks for as short a time as possible;
3790 : * thus, this function simply makes a copy of the necessary data and releases
3791 : * the locks, allowing the caller to contemplate and format the data for as
3792 : * long as it pleases.
3793 : */
3794 : LockData *
3795 574 : GetLockStatusData(void)
3796 : {
3797 : LockData *data;
3798 : PROCLOCK *proclock;
3799 : HASH_SEQ_STATUS seqstat;
3800 : int els;
3801 : int el;
3802 : int i;
3803 :
3804 574 : data = (LockData *) palloc(sizeof(LockData));
3805 :
3806 : /* Guess how much space we'll need. */
3807 574 : els = MaxBackends;
3808 574 : el = 0;
3809 574 : data->locks = (LockInstanceData *) palloc(sizeof(LockInstanceData) * els);
3810 :
3811 : /*
3812 : * First, we iterate through the per-backend fast-path arrays, locking
3813 : * them one at a time. This might produce an inconsistent picture of the
3814 : * system state, but taking all of those LWLocks at the same time seems
3815 : * impractical (in particular, note MAX_SIMUL_LWLOCKS). It shouldn't
3816 : * matter too much, because none of these locks can be involved in lock
3817 : * conflicts anyway - anything that might must be present in the main lock
3818 : * table. (For the same reason, we don't sweat about making leaderPid
3819 : * completely valid. We cannot safely dereference another backend's
3820 : * lockGroupLeader field without holding all lock partition locks, and
3821 : * it's not worth that.)
3822 : */
3823 85476 : for (i = 0; i < ProcGlobal->allProcCount; ++i)
3824 : {
3825 84902 : PGPROC *proc = &ProcGlobal->allProcs[i];
3826 :
3827 : /* Skip backends with pid=0, as they don't hold fast-path locks */
3828 84902 : if (proc->pid == 0)
3829 76562 : continue;
3830 :
3831 8340 : LWLockAcquire(&proc->fpInfoLock, LW_SHARED);
3832 :
3833 41700 : for (uint32 g = 0; g < FastPathLockGroupsPerBackend; g++)
3834 : {
3835 : /* Skip groups without registered fast-path locks */
3836 33360 : if (proc->fpLockBits[g] == 0)
3837 27846 : continue;
3838 :
3839 93738 : for (int j = 0; j < FP_LOCK_SLOTS_PER_GROUP; j++)
3840 : {
3841 : LockInstanceData *instance;
3842 88224 : uint32 f = FAST_PATH_SLOT(g, j);
3843 88224 : uint32 lockbits = FAST_PATH_GET_BITS(proc, f);
3844 :
3845 : /* Skip unallocated slots */
3846 88224 : if (!lockbits)
3847 78886 : continue;
3848 :
3849 9338 : if (el >= els)
3850 : {
3851 30 : els += MaxBackends;
3852 30 : data->locks = (LockInstanceData *)
3853 30 : repalloc(data->locks, sizeof(LockInstanceData) * els);
3854 : }
3855 :
3856 9338 : instance = &data->locks[el];
3857 9338 : SET_LOCKTAG_RELATION(instance->locktag, proc->databaseId,
3858 : proc->fpRelId[f]);
3859 9338 : instance->holdMask = lockbits << FAST_PATH_LOCKNUMBER_OFFSET;
3860 9338 : instance->waitLockMode = NoLock;
3861 9338 : instance->vxid.procNumber = proc->vxid.procNumber;
3862 9338 : instance->vxid.localTransactionId = proc->vxid.lxid;
3863 9338 : instance->pid = proc->pid;
3864 9338 : instance->leaderPid = proc->pid;
3865 9338 : instance->fastpath = true;
3866 :
3867 : /*
3868 : * Successfully taking fast path lock means there were no
3869 : * conflicting locks.
3870 : */
3871 9338 : instance->waitStart = 0;
3872 :
3873 9338 : el++;
3874 : }
3875 : }
3876 :
3877 8340 : if (proc->fpVXIDLock)
3878 : {
3879 : VirtualTransactionId vxid;
3880 : LockInstanceData *instance;
3881 :
3882 3128 : if (el >= els)
3883 : {
3884 24 : els += MaxBackends;
3885 24 : data->locks = (LockInstanceData *)
3886 24 : repalloc(data->locks, sizeof(LockInstanceData) * els);
3887 : }
3888 :
3889 3128 : vxid.procNumber = proc->vxid.procNumber;
3890 3128 : vxid.localTransactionId = proc->fpLocalTransactionId;
3891 :
3892 3128 : instance = &data->locks[el];
3893 3128 : SET_LOCKTAG_VIRTUALTRANSACTION(instance->locktag, vxid);
3894 3128 : instance->holdMask = LOCKBIT_ON(ExclusiveLock);
3895 3128 : instance->waitLockMode = NoLock;
3896 3128 : instance->vxid.procNumber = proc->vxid.procNumber;
3897 3128 : instance->vxid.localTransactionId = proc->vxid.lxid;
3898 3128 : instance->pid = proc->pid;
3899 3128 : instance->leaderPid = proc->pid;
3900 3128 : instance->fastpath = true;
3901 3128 : instance->waitStart = 0;
3902 :
3903 3128 : el++;
3904 : }
3905 :
3906 8340 : LWLockRelease(&proc->fpInfoLock);
3907 : }
3908 :
3909 : /*
3910 : * Next, acquire lock on the entire shared lock data structure. We do
3911 : * this so that, at least for locks in the primary lock table, the state
3912 : * will be self-consistent.
3913 : *
3914 : * Since this is a read-only operation, we take shared instead of
3915 : * exclusive lock. There's not a whole lot of point to this, because all
3916 : * the normal operations require exclusive lock, but it doesn't hurt
3917 : * anything either. It will at least allow two backends to do
3918 : * GetLockStatusData in parallel.
3919 : *
3920 : * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
3921 : */
3922 9758 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
3923 9184 : LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
3924 :
3925 : /* Now we can safely count the number of proclocks */
3926 574 : data->nelements = el + hash_get_num_entries(LockMethodProcLockHash);
3927 574 : if (data->nelements > els)
3928 : {
3929 18 : els = data->nelements;
3930 18 : data->locks = (LockInstanceData *)
3931 18 : repalloc(data->locks, sizeof(LockInstanceData) * els);
3932 : }
3933 :
3934 : /* Now scan the tables to copy the data */
3935 574 : hash_seq_init(&seqstat, LockMethodProcLockHash);
3936 :
3937 6540 : while ((proclock = (PROCLOCK *) hash_seq_search(&seqstat)))
3938 : {
3939 5966 : PGPROC *proc = proclock->tag.myProc;
3940 5966 : LOCK *lock = proclock->tag.myLock;
3941 5966 : LockInstanceData *instance = &data->locks[el];
3942 :
3943 5966 : memcpy(&instance->locktag, &lock->tag, sizeof(LOCKTAG));
3944 5966 : instance->holdMask = proclock->holdMask;
3945 5966 : if (proc->waitLock == proclock->tag.myLock)
3946 18 : instance->waitLockMode = proc->waitLockMode;
3947 : else
3948 5948 : instance->waitLockMode = NoLock;
3949 5966 : instance->vxid.procNumber = proc->vxid.procNumber;
3950 5966 : instance->vxid.localTransactionId = proc->vxid.lxid;
3951 5966 : instance->pid = proc->pid;
3952 5966 : instance->leaderPid = proclock->groupLeader->pid;
3953 5966 : instance->fastpath = false;
3954 5966 : instance->waitStart = (TimestampTz) pg_atomic_read_u64(&proc->waitStart);
3955 :
3956 5966 : el++;
3957 : }
3958 :
3959 : /*
3960 : * And release locks. We do this in reverse order for two reasons: (1)
3961 : * Anyone else who needs more than one of the locks will be trying to lock
3962 : * them in increasing order; we don't want to release the other process
3963 : * until it can get all the locks it needs. (2) This avoids O(N^2)
3964 : * behavior inside LWLockRelease.
3965 : */
3966 9758 : for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
3967 9184 : LWLockRelease(LockHashPartitionLockByIndex(i));
3968 :
3969 : Assert(el == data->nelements);
3970 :
3971 574 : return data;
3972 : }
3973 :
3974 : /*
3975 : * GetBlockerStatusData - Return a summary of the lock manager's state
3976 : * concerning locks that are blocking the specified PID or any member of
3977 : * the PID's lock group, for use in a user-level reporting function.
3978 : *
3979 : * For each PID within the lock group that is awaiting some heavyweight lock,
3980 : * the return data includes an array of LockInstanceData objects, which are
3981 : * the same data structure used by GetLockStatusData; but unlike that function,
3982 : * this one reports only the PROCLOCKs associated with the lock that that PID
3983 : * is blocked on. (Hence, all the locktags should be the same for any one
3984 : * blocked PID.) In addition, we return an array of the PIDs of those backends
3985 : * that are ahead of the blocked PID in the lock's wait queue. These can be
3986 : * compared with the PIDs in the LockInstanceData objects to determine which
3987 : * waiters are ahead of or behind the blocked PID in the queue.
3988 : *
3989 : * If blocked_pid isn't a valid backend PID or nothing in its lock group is
3990 : * waiting on any heavyweight lock, return empty arrays.
3991 : *
3992 : * The design goal is to hold the LWLocks for as short a time as possible;
3993 : * thus, this function simply makes a copy of the necessary data and releases
3994 : * the locks, allowing the caller to contemplate and format the data for as
3995 : * long as it pleases.
3996 : */
3997 : BlockedProcsData *
3998 4080 : GetBlockerStatusData(int blocked_pid)
3999 : {
4000 : BlockedProcsData *data;
4001 : PGPROC *proc;
4002 : int i;
4003 :
4004 4080 : data = (BlockedProcsData *) palloc(sizeof(BlockedProcsData));
4005 :
4006 : /*
4007 : * Guess how much space we'll need, and preallocate. Most of the time
4008 : * this will avoid needing to do repalloc while holding the LWLocks. (We
4009 : * assume, but check with an Assert, that MaxBackends is enough entries
4010 : * for the procs[] array; the other two could need enlargement, though.)
4011 : */
4012 4080 : data->nprocs = data->nlocks = data->npids = 0;
4013 4080 : data->maxprocs = data->maxlocks = data->maxpids = MaxBackends;
4014 4080 : data->procs = (BlockedProcData *) palloc(sizeof(BlockedProcData) * data->maxprocs);
4015 4080 : data->locks = (LockInstanceData *) palloc(sizeof(LockInstanceData) * data->maxlocks);
4016 4080 : data->waiter_pids = (int *) palloc(sizeof(int) * data->maxpids);
4017 :
4018 : /*
4019 : * In order to search the ProcArray for blocked_pid and assume that that
4020 : * entry won't immediately disappear under us, we must hold ProcArrayLock.
4021 : * In addition, to examine the lock grouping fields of any other backend,
4022 : * we must hold all the hash partition locks. (Only one of those locks is
4023 : * actually relevant for any one lock group, but we can't know which one
4024 : * ahead of time.) It's fairly annoying to hold all those locks
4025 : * throughout this, but it's no worse than GetLockStatusData(), and it
4026 : * does have the advantage that we're guaranteed to return a
4027 : * self-consistent instantaneous state.
4028 : */
4029 4080 : LWLockAcquire(ProcArrayLock, LW_SHARED);
4030 :
4031 4080 : proc = BackendPidGetProcWithLock(blocked_pid);
4032 :
4033 : /* Nothing to do if it's gone */
4034 4080 : if (proc != NULL)
4035 : {
4036 : /*
4037 : * Acquire lock on the entire shared lock data structure. See notes
4038 : * in GetLockStatusData().
4039 : */
4040 69360 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
4041 65280 : LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
4042 :
4043 4080 : if (proc->lockGroupLeader == NULL)
4044 : {
4045 : /* Easy case, proc is not a lock group member */
4046 3384 : GetSingleProcBlockerStatusData(proc, data);
4047 : }
4048 : else
4049 : {
4050 : /* Examine all procs in proc's lock group */
4051 : dlist_iter iter;
4052 :
4053 1484 : dlist_foreach(iter, &proc->lockGroupLeader->lockGroupMembers)
4054 : {
4055 : PGPROC *memberProc;
4056 :
4057 788 : memberProc = dlist_container(PGPROC, lockGroupLink, iter.cur);
4058 788 : GetSingleProcBlockerStatusData(memberProc, data);
4059 : }
4060 : }
4061 :
4062 : /*
4063 : * And release locks. See notes in GetLockStatusData().
4064 : */
4065 69360 : for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
4066 65280 : LWLockRelease(LockHashPartitionLockByIndex(i));
4067 :
4068 : Assert(data->nprocs <= data->maxprocs);
4069 : }
4070 :
4071 4080 : LWLockRelease(ProcArrayLock);
4072 :
4073 4080 : return data;
4074 : }
4075 :
4076 : /* Accumulate data about one possibly-blocked proc for GetBlockerStatusData */
4077 : static void
4078 4172 : GetSingleProcBlockerStatusData(PGPROC *blocked_proc, BlockedProcsData *data)
4079 : {
4080 4172 : LOCK *theLock = blocked_proc->waitLock;
4081 : BlockedProcData *bproc;
4082 : dlist_iter proclock_iter;
4083 : dlist_iter proc_iter;
4084 : dclist_head *waitQueue;
4085 : int queue_size;
4086 :
4087 : /* Nothing to do if this proc is not blocked */
4088 4172 : if (theLock == NULL)
4089 1840 : return;
4090 :
4091 : /* Set up a procs[] element */
4092 2332 : bproc = &data->procs[data->nprocs++];
4093 2332 : bproc->pid = blocked_proc->pid;
4094 2332 : bproc->first_lock = data->nlocks;
4095 2332 : bproc->first_waiter = data->npids;
4096 :
4097 : /*
4098 : * We may ignore the proc's fast-path arrays, since nothing in those could
4099 : * be related to a contended lock.
4100 : */
4101 :
4102 : /* Collect all PROCLOCKs associated with theLock */
4103 7090 : dlist_foreach(proclock_iter, &theLock->procLocks)
4104 : {
4105 4758 : PROCLOCK *proclock =
4106 4758 : dlist_container(PROCLOCK, lockLink, proclock_iter.cur);
4107 4758 : PGPROC *proc = proclock->tag.myProc;
4108 4758 : LOCK *lock = proclock->tag.myLock;
4109 : LockInstanceData *instance;
4110 :
4111 4758 : if (data->nlocks >= data->maxlocks)
4112 : {
4113 0 : data->maxlocks += MaxBackends;
4114 0 : data->locks = (LockInstanceData *)
4115 0 : repalloc(data->locks, sizeof(LockInstanceData) * data->maxlocks);
4116 : }
4117 :
4118 4758 : instance = &data->locks[data->nlocks];
4119 4758 : memcpy(&instance->locktag, &lock->tag, sizeof(LOCKTAG));
4120 4758 : instance->holdMask = proclock->holdMask;
4121 4758 : if (proc->waitLock == lock)
4122 2416 : instance->waitLockMode = proc->waitLockMode;
4123 : else
4124 2342 : instance->waitLockMode = NoLock;
4125 4758 : instance->vxid.procNumber = proc->vxid.procNumber;
4126 4758 : instance->vxid.localTransactionId = proc->vxid.lxid;
4127 4758 : instance->pid = proc->pid;
4128 4758 : instance->leaderPid = proclock->groupLeader->pid;
4129 4758 : instance->fastpath = false;
4130 4758 : data->nlocks++;
4131 : }
4132 :
4133 : /* Enlarge waiter_pids[] if it's too small to hold all wait queue PIDs */
4134 2332 : waitQueue = &(theLock->waitProcs);
4135 2332 : queue_size = dclist_count(waitQueue);
4136 :
4137 2332 : if (queue_size > data->maxpids - data->npids)
4138 : {
4139 0 : data->maxpids = Max(data->maxpids + MaxBackends,
4140 : data->npids + queue_size);
4141 0 : data->waiter_pids = (int *) repalloc(data->waiter_pids,
4142 0 : sizeof(int) * data->maxpids);
4143 : }
4144 :
4145 : /* Collect PIDs from the lock's wait queue, stopping at blocked_proc */
4146 2372 : dclist_foreach(proc_iter, waitQueue)
4147 : {
4148 2372 : PGPROC *queued_proc = dlist_container(PGPROC, links, proc_iter.cur);
4149 :
4150 2372 : if (queued_proc == blocked_proc)
4151 2332 : break;
4152 40 : data->waiter_pids[data->npids++] = queued_proc->pid;
4153 40 : queued_proc = (PGPROC *) queued_proc->links.next;
4154 : }
4155 :
4156 2332 : bproc->num_locks = data->nlocks - bproc->first_lock;
4157 2332 : bproc->num_waiters = data->npids - bproc->first_waiter;
4158 : }
4159 :
4160 : /*
4161 : * Returns a list of currently held AccessExclusiveLocks, for use by
4162 : * LogStandbySnapshot(). The result is a palloc'd array,
4163 : * with the number of elements returned into *nlocks.
4164 : *
4165 : * XXX This currently takes a lock on all partitions of the lock table,
4166 : * but it's possible to do better. By reference counting locks and storing
4167 : * the value in the ProcArray entry for each backend we could tell if any
4168 : * locks need recording without having to acquire the partition locks and
4169 : * scan the lock table. Whether that's worth the additional overhead
4170 : * is pretty dubious though.
4171 : */
4172 : xl_standby_lock *
4173 2750 : GetRunningTransactionLocks(int *nlocks)
4174 : {
4175 : xl_standby_lock *accessExclusiveLocks;
4176 : PROCLOCK *proclock;
4177 : HASH_SEQ_STATUS seqstat;
4178 : int i;
4179 : int index;
4180 : int els;
4181 :
4182 : /*
4183 : * Acquire lock on the entire shared lock data structure.
4184 : *
4185 : * Must grab LWLocks in partition-number order to avoid LWLock deadlock.
4186 : */
4187 46750 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
4188 44000 : LWLockAcquire(LockHashPartitionLockByIndex(i), LW_SHARED);
4189 :
4190 : /* Now we can safely count the number of proclocks */
4191 2750 : els = hash_get_num_entries(LockMethodProcLockHash);
4192 :
4193 : /*
4194 : * Allocating enough space for all locks in the lock table is overkill,
4195 : * but it's more convenient and faster than having to enlarge the array.
4196 : */
4197 2750 : accessExclusiveLocks = palloc(els * sizeof(xl_standby_lock));
4198 :
4199 : /* Now scan the tables to copy the data */
4200 2750 : hash_seq_init(&seqstat, LockMethodProcLockHash);
4201 :
4202 : /*
4203 : * If lock is a currently granted AccessExclusiveLock then it will have
4204 : * just one proclock holder, so locks are never accessed twice in this
4205 : * particular case. Don't copy this code for use elsewhere because in the
4206 : * general case this will give you duplicate locks when looking at
4207 : * non-exclusive lock types.
4208 : */
4209 2750 : index = 0;
4210 11976 : while ((proclock = (PROCLOCK *) hash_seq_search(&seqstat)))
4211 : {
4212 : /* make sure this definition matches the one used in LockAcquire */
4213 9226 : if ((proclock->holdMask & LOCKBIT_ON(AccessExclusiveLock)) &&
4214 4682 : proclock->tag.myLock->tag.locktag_type == LOCKTAG_RELATION)
4215 : {
4216 3046 : PGPROC *proc = proclock->tag.myProc;
4217 3046 : LOCK *lock = proclock->tag.myLock;
4218 3046 : TransactionId xid = proc->xid;
4219 :
4220 : /*
4221 : * Don't record locks for transactions if we know they have
4222 : * already issued their WAL record for commit but not yet released
4223 : * lock. It is still possible that we see locks held by already
4224 : * complete transactions, if they haven't yet zeroed their xids.
4225 : */
4226 3046 : if (!TransactionIdIsValid(xid))
4227 8 : continue;
4228 :
4229 3038 : accessExclusiveLocks[index].xid = xid;
4230 3038 : accessExclusiveLocks[index].dbOid = lock->tag.locktag_field1;
4231 3038 : accessExclusiveLocks[index].relOid = lock->tag.locktag_field2;
4232 :
4233 3038 : index++;
4234 : }
4235 : }
4236 :
4237 : Assert(index <= els);
4238 :
4239 : /*
4240 : * And release locks. We do this in reverse order for two reasons: (1)
4241 : * Anyone else who needs more than one of the locks will be trying to lock
4242 : * them in increasing order; we don't want to release the other process
4243 : * until it can get all the locks it needs. (2) This avoids O(N^2)
4244 : * behavior inside LWLockRelease.
4245 : */
4246 46750 : for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
4247 44000 : LWLockRelease(LockHashPartitionLockByIndex(i));
4248 :
4249 2750 : *nlocks = index;
4250 2750 : return accessExclusiveLocks;
4251 : }
4252 :
4253 : /* Provide the textual name of any lock mode */
4254 : const char *
4255 22660 : GetLockmodeName(LOCKMETHODID lockmethodid, LOCKMODE mode)
4256 : {
4257 : Assert(lockmethodid > 0 && lockmethodid < lengthof(LockMethods));
4258 : Assert(mode > 0 && mode <= LockMethods[lockmethodid]->numLockModes);
4259 22660 : return LockMethods[lockmethodid]->lockModeNames[mode];
4260 : }
4261 :
4262 : #ifdef LOCK_DEBUG
4263 : /*
4264 : * Dump all locks in the given proc's myProcLocks lists.
4265 : *
4266 : * Caller is responsible for having acquired appropriate LWLocks.
4267 : */
4268 : void
4269 : DumpLocks(PGPROC *proc)
4270 : {
4271 : int i;
4272 :
4273 : if (proc == NULL)
4274 : return;
4275 :
4276 : if (proc->waitLock)
4277 : LOCK_PRINT("DumpLocks: waiting on", proc->waitLock, 0);
4278 :
4279 : for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
4280 : {
4281 : dlist_head *procLocks = &proc->myProcLocks[i];
4282 : dlist_iter iter;
4283 :
4284 : dlist_foreach(iter, procLocks)
4285 : {
4286 : PROCLOCK *proclock = dlist_container(PROCLOCK, procLink, iter.cur);
4287 : LOCK *lock = proclock->tag.myLock;
4288 :
4289 : Assert(proclock->tag.myProc == proc);
4290 : PROCLOCK_PRINT("DumpLocks", proclock);
4291 : LOCK_PRINT("DumpLocks", lock, 0);
4292 : }
4293 : }
4294 : }
4295 :
4296 : /*
4297 : * Dump all lmgr locks.
4298 : *
4299 : * Caller is responsible for having acquired appropriate LWLocks.
4300 : */
4301 : void
4302 : DumpAllLocks(void)
4303 : {
4304 : PGPROC *proc;
4305 : PROCLOCK *proclock;
4306 : LOCK *lock;
4307 : HASH_SEQ_STATUS status;
4308 :
4309 : proc = MyProc;
4310 :
4311 : if (proc && proc->waitLock)
4312 : LOCK_PRINT("DumpAllLocks: waiting on", proc->waitLock, 0);
4313 :
4314 : hash_seq_init(&status, LockMethodProcLockHash);
4315 :
4316 : while ((proclock = (PROCLOCK *) hash_seq_search(&status)) != NULL)
4317 : {
4318 : PROCLOCK_PRINT("DumpAllLocks", proclock);
4319 :
4320 : lock = proclock->tag.myLock;
4321 : if (lock)
4322 : LOCK_PRINT("DumpAllLocks", lock, 0);
4323 : else
4324 : elog(LOG, "DumpAllLocks: proclock->tag.myLock = NULL");
4325 : }
4326 : }
4327 : #endif /* LOCK_DEBUG */
4328 :
4329 : /*
4330 : * LOCK 2PC resource manager's routines
4331 : */
4332 :
4333 : /*
4334 : * Re-acquire a lock belonging to a transaction that was prepared.
4335 : *
4336 : * Because this function is run at db startup, re-acquiring the locks should
4337 : * never conflict with running transactions because there are none. We
4338 : * assume that the lock state represented by the stored 2PC files is legal.
4339 : *
4340 : * When switching from Hot Standby mode to normal operation, the locks will
4341 : * be already held by the startup process. The locks are acquired for the new
4342 : * procs without checking for conflicts, so we don't get a conflict between the
4343 : * startup process and the dummy procs, even though we will momentarily have
4344 : * a situation where two procs are holding the same AccessExclusiveLock,
4345 : * which isn't normally possible because the conflict. If we're in standby
4346 : * mode, but a recovery snapshot hasn't been established yet, it's possible
4347 : * that some but not all of the locks are already held by the startup process.
4348 : *
4349 : * This approach is simple, but also a bit dangerous, because if there isn't
4350 : * enough shared memory to acquire the locks, an error will be thrown, which
4351 : * is promoted to FATAL and recovery will abort, bringing down postmaster.
4352 : * A safer approach would be to transfer the locks like we do in
4353 : * AtPrepare_Locks, but then again, in hot standby mode it's possible for
4354 : * read-only backends to use up all the shared lock memory anyway, so that
4355 : * replaying the WAL record that needs to acquire a lock will throw an error
4356 : * and PANIC anyway.
4357 : */
4358 : void
4359 180 : lock_twophase_recover(FullTransactionId fxid, uint16 info,
4360 : void *recdata, uint32 len)
4361 : {
4362 180 : TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4363 180 : PGPROC *proc = TwoPhaseGetDummyProc(fxid, false);
4364 : LOCKTAG *locktag;
4365 : LOCKMODE lockmode;
4366 : LOCKMETHODID lockmethodid;
4367 : LOCK *lock;
4368 : PROCLOCK *proclock;
4369 : PROCLOCKTAG proclocktag;
4370 : bool found;
4371 : uint32 hashcode;
4372 : uint32 proclock_hashcode;
4373 : int partition;
4374 : LWLock *partitionLock;
4375 : LockMethod lockMethodTable;
4376 :
4377 : Assert(len == sizeof(TwoPhaseLockRecord));
4378 180 : locktag = &rec->locktag;
4379 180 : lockmode = rec->lockmode;
4380 180 : lockmethodid = locktag->locktag_lockmethodid;
4381 :
4382 180 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4383 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4384 180 : lockMethodTable = LockMethods[lockmethodid];
4385 :
4386 180 : hashcode = LockTagHashCode(locktag);
4387 180 : partition = LockHashPartition(hashcode);
4388 180 : partitionLock = LockHashPartitionLock(hashcode);
4389 :
4390 180 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4391 :
4392 : /*
4393 : * Find or create a lock with this tag.
4394 : */
4395 180 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
4396 : locktag,
4397 : hashcode,
4398 : HASH_ENTER_NULL,
4399 : &found);
4400 180 : if (!lock)
4401 : {
4402 0 : LWLockRelease(partitionLock);
4403 0 : ereport(ERROR,
4404 : (errcode(ERRCODE_OUT_OF_MEMORY),
4405 : errmsg("out of shared memory"),
4406 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
4407 : }
4408 :
4409 : /*
4410 : * if it's a new lock object, initialize it
4411 : */
4412 180 : if (!found)
4413 : {
4414 156 : lock->grantMask = 0;
4415 156 : lock->waitMask = 0;
4416 156 : dlist_init(&lock->procLocks);
4417 156 : dclist_init(&lock->waitProcs);
4418 156 : lock->nRequested = 0;
4419 156 : lock->nGranted = 0;
4420 936 : MemSet(lock->requested, 0, sizeof(int) * MAX_LOCKMODES);
4421 156 : MemSet(lock->granted, 0, sizeof(int) * MAX_LOCKMODES);
4422 : LOCK_PRINT("lock_twophase_recover: new", lock, lockmode);
4423 : }
4424 : else
4425 : {
4426 : LOCK_PRINT("lock_twophase_recover: found", lock, lockmode);
4427 : Assert((lock->nRequested >= 0) && (lock->requested[lockmode] >= 0));
4428 : Assert((lock->nGranted >= 0) && (lock->granted[lockmode] >= 0));
4429 : Assert(lock->nGranted <= lock->nRequested);
4430 : }
4431 :
4432 : /*
4433 : * Create the hash key for the proclock table.
4434 : */
4435 180 : proclocktag.myLock = lock;
4436 180 : proclocktag.myProc = proc;
4437 :
4438 180 : proclock_hashcode = ProcLockHashCode(&proclocktag, hashcode);
4439 :
4440 : /*
4441 : * Find or create a proclock entry with this tag
4442 : */
4443 180 : proclock = (PROCLOCK *) hash_search_with_hash_value(LockMethodProcLockHash,
4444 : &proclocktag,
4445 : proclock_hashcode,
4446 : HASH_ENTER_NULL,
4447 : &found);
4448 180 : if (!proclock)
4449 : {
4450 : /* Oops, not enough shmem for the proclock */
4451 0 : if (lock->nRequested == 0)
4452 : {
4453 : /*
4454 : * There are no other requestors of this lock, so garbage-collect
4455 : * the lock object. We *must* do this to avoid a permanent leak
4456 : * of shared memory, because there won't be anything to cause
4457 : * anyone to release the lock object later.
4458 : */
4459 : Assert(dlist_is_empty(&lock->procLocks));
4460 0 : if (!hash_search_with_hash_value(LockMethodLockHash,
4461 0 : &(lock->tag),
4462 : hashcode,
4463 : HASH_REMOVE,
4464 : NULL))
4465 0 : elog(PANIC, "lock table corrupted");
4466 : }
4467 0 : LWLockRelease(partitionLock);
4468 0 : ereport(ERROR,
4469 : (errcode(ERRCODE_OUT_OF_MEMORY),
4470 : errmsg("out of shared memory"),
4471 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
4472 : }
4473 :
4474 : /*
4475 : * If new, initialize the new entry
4476 : */
4477 180 : if (!found)
4478 : {
4479 : Assert(proc->lockGroupLeader == NULL);
4480 164 : proclock->groupLeader = proc;
4481 164 : proclock->holdMask = 0;
4482 164 : proclock->releaseMask = 0;
4483 : /* Add proclock to appropriate lists */
4484 164 : dlist_push_tail(&lock->procLocks, &proclock->lockLink);
4485 164 : dlist_push_tail(&proc->myProcLocks[partition],
4486 : &proclock->procLink);
4487 : PROCLOCK_PRINT("lock_twophase_recover: new", proclock);
4488 : }
4489 : else
4490 : {
4491 : PROCLOCK_PRINT("lock_twophase_recover: found", proclock);
4492 : Assert((proclock->holdMask & ~lock->grantMask) == 0);
4493 : }
4494 :
4495 : /*
4496 : * lock->nRequested and lock->requested[] count the total number of
4497 : * requests, whether granted or waiting, so increment those immediately.
4498 : */
4499 180 : lock->nRequested++;
4500 180 : lock->requested[lockmode]++;
4501 : Assert((lock->nRequested > 0) && (lock->requested[lockmode] > 0));
4502 :
4503 : /*
4504 : * We shouldn't already hold the desired lock.
4505 : */
4506 180 : if (proclock->holdMask & LOCKBIT_ON(lockmode))
4507 0 : elog(ERROR, "lock %s on object %u/%u/%u is already held",
4508 : lockMethodTable->lockModeNames[lockmode],
4509 : lock->tag.locktag_field1, lock->tag.locktag_field2,
4510 : lock->tag.locktag_field3);
4511 :
4512 : /*
4513 : * We ignore any possible conflicts and just grant ourselves the lock. Not
4514 : * only because we don't bother, but also to avoid deadlocks when
4515 : * switching from standby to normal mode. See function comment.
4516 : */
4517 180 : GrantLock(lock, proclock, lockmode);
4518 :
4519 : /*
4520 : * Bump strong lock count, to make sure any fast-path lock requests won't
4521 : * be granted without consulting the primary lock table.
4522 : */
4523 180 : if (ConflictsWithRelationFastPath(&lock->tag, lockmode))
4524 : {
4525 36 : uint32 fasthashcode = FastPathStrongLockHashPartition(hashcode);
4526 :
4527 36 : SpinLockAcquire(&FastPathStrongRelationLocks->mutex);
4528 36 : FastPathStrongRelationLocks->count[fasthashcode]++;
4529 36 : SpinLockRelease(&FastPathStrongRelationLocks->mutex);
4530 : }
4531 :
4532 180 : LWLockRelease(partitionLock);
4533 180 : }
4534 :
4535 : /*
4536 : * Re-acquire a lock belonging to a transaction that was prepared, when
4537 : * starting up into hot standby mode.
4538 : */
4539 : void
4540 0 : lock_twophase_standby_recover(FullTransactionId fxid, uint16 info,
4541 : void *recdata, uint32 len)
4542 : {
4543 0 : TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4544 : LOCKTAG *locktag;
4545 : LOCKMODE lockmode;
4546 : LOCKMETHODID lockmethodid;
4547 :
4548 : Assert(len == sizeof(TwoPhaseLockRecord));
4549 0 : locktag = &rec->locktag;
4550 0 : lockmode = rec->lockmode;
4551 0 : lockmethodid = locktag->locktag_lockmethodid;
4552 :
4553 0 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4554 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4555 :
4556 0 : if (lockmode == AccessExclusiveLock &&
4557 0 : locktag->locktag_type == LOCKTAG_RELATION)
4558 : {
4559 0 : StandbyAcquireAccessExclusiveLock(XidFromFullTransactionId(fxid),
4560 : locktag->locktag_field1 /* dboid */ ,
4561 : locktag->locktag_field2 /* reloid */ );
4562 : }
4563 0 : }
4564 :
4565 :
4566 : /*
4567 : * 2PC processing routine for COMMIT PREPARED case.
4568 : *
4569 : * Find and release the lock indicated by the 2PC record.
4570 : */
4571 : void
4572 1440 : lock_twophase_postcommit(FullTransactionId fxid, uint16 info,
4573 : void *recdata, uint32 len)
4574 : {
4575 1440 : TwoPhaseLockRecord *rec = (TwoPhaseLockRecord *) recdata;
4576 1440 : PGPROC *proc = TwoPhaseGetDummyProc(fxid, true);
4577 : LOCKTAG *locktag;
4578 : LOCKMETHODID lockmethodid;
4579 : LockMethod lockMethodTable;
4580 :
4581 : Assert(len == sizeof(TwoPhaseLockRecord));
4582 1440 : locktag = &rec->locktag;
4583 1440 : lockmethodid = locktag->locktag_lockmethodid;
4584 :
4585 1440 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4586 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4587 1440 : lockMethodTable = LockMethods[lockmethodid];
4588 :
4589 1440 : LockRefindAndRelease(lockMethodTable, proc, locktag, rec->lockmode, true);
4590 1440 : }
4591 :
4592 : /*
4593 : * 2PC processing routine for ROLLBACK PREPARED case.
4594 : *
4595 : * This is actually just the same as the COMMIT case.
4596 : */
4597 : void
4598 252 : lock_twophase_postabort(FullTransactionId fxid, uint16 info,
4599 : void *recdata, uint32 len)
4600 : {
4601 252 : lock_twophase_postcommit(fxid, info, recdata, len);
4602 252 : }
4603 :
4604 : /*
4605 : * VirtualXactLockTableInsert
4606 : *
4607 : * Take vxid lock via the fast-path. There can't be any pre-existing
4608 : * lockers, as we haven't advertised this vxid via the ProcArray yet.
4609 : *
4610 : * Since MyProc->fpLocalTransactionId will normally contain the same data
4611 : * as MyProc->vxid.lxid, you might wonder if we really need both. The
4612 : * difference is that MyProc->vxid.lxid is set and cleared unlocked, and
4613 : * examined by procarray.c, while fpLocalTransactionId is protected by
4614 : * fpInfoLock and is used only by the locking subsystem. Doing it this
4615 : * way makes it easier to verify that there are no funny race conditions.
4616 : *
4617 : * We don't bother recording this lock in the local lock table, since it's
4618 : * only ever released at the end of a transaction. Instead,
4619 : * LockReleaseAll() calls VirtualXactLockTableCleanup().
4620 : */
4621 : void
4622 1159984 : VirtualXactLockTableInsert(VirtualTransactionId vxid)
4623 : {
4624 : Assert(VirtualTransactionIdIsValid(vxid));
4625 :
4626 1159984 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
4627 :
4628 : Assert(MyProc->vxid.procNumber == vxid.procNumber);
4629 : Assert(MyProc->fpLocalTransactionId == InvalidLocalTransactionId);
4630 : Assert(MyProc->fpVXIDLock == false);
4631 :
4632 1159984 : MyProc->fpVXIDLock = true;
4633 1159984 : MyProc->fpLocalTransactionId = vxid.localTransactionId;
4634 :
4635 1159984 : LWLockRelease(&MyProc->fpInfoLock);
4636 1159984 : }
4637 :
4638 : /*
4639 : * VirtualXactLockTableCleanup
4640 : *
4641 : * Check whether a VXID lock has been materialized; if so, release it,
4642 : * unblocking waiters.
4643 : */
4644 : void
4645 1160924 : VirtualXactLockTableCleanup(void)
4646 : {
4647 : bool fastpath;
4648 : LocalTransactionId lxid;
4649 :
4650 : Assert(MyProc->vxid.procNumber != INVALID_PROC_NUMBER);
4651 :
4652 : /*
4653 : * Clean up shared memory state.
4654 : */
4655 1160924 : LWLockAcquire(&MyProc->fpInfoLock, LW_EXCLUSIVE);
4656 :
4657 1160924 : fastpath = MyProc->fpVXIDLock;
4658 1160924 : lxid = MyProc->fpLocalTransactionId;
4659 1160924 : MyProc->fpVXIDLock = false;
4660 1160924 : MyProc->fpLocalTransactionId = InvalidLocalTransactionId;
4661 :
4662 1160924 : LWLockRelease(&MyProc->fpInfoLock);
4663 :
4664 : /*
4665 : * If fpVXIDLock has been cleared without touching fpLocalTransactionId,
4666 : * that means someone transferred the lock to the main lock table.
4667 : */
4668 1160924 : if (!fastpath && LocalTransactionIdIsValid(lxid))
4669 : {
4670 : VirtualTransactionId vxid;
4671 : LOCKTAG locktag;
4672 :
4673 446 : vxid.procNumber = MyProcNumber;
4674 446 : vxid.localTransactionId = lxid;
4675 446 : SET_LOCKTAG_VIRTUALTRANSACTION(locktag, vxid);
4676 :
4677 446 : LockRefindAndRelease(LockMethods[DEFAULT_LOCKMETHOD], MyProc,
4678 : &locktag, ExclusiveLock, false);
4679 : }
4680 1160924 : }
4681 :
4682 : /*
4683 : * XactLockForVirtualXact
4684 : *
4685 : * If TransactionIdIsValid(xid), this is essentially XactLockTableWait(xid,
4686 : * NULL, NULL, XLTW_None) or ConditionalXactLockTableWait(xid). Unlike those
4687 : * functions, it assumes "xid" is never a subtransaction and that "xid" is
4688 : * prepared, committed, or aborted.
4689 : *
4690 : * If !TransactionIdIsValid(xid), this locks every prepared XID having been
4691 : * known as "vxid" before its PREPARE TRANSACTION.
4692 : */
4693 : static bool
4694 446 : XactLockForVirtualXact(VirtualTransactionId vxid,
4695 : TransactionId xid, bool wait)
4696 : {
4697 446 : bool more = false;
4698 :
4699 : /* There is no point to wait for 2PCs if you have no 2PCs. */
4700 446 : if (max_prepared_xacts == 0)
4701 116 : return true;
4702 :
4703 : do
4704 : {
4705 : LockAcquireResult lar;
4706 : LOCKTAG tag;
4707 :
4708 : /* Clear state from previous iterations. */
4709 330 : if (more)
4710 : {
4711 0 : xid = InvalidTransactionId;
4712 0 : more = false;
4713 : }
4714 :
4715 : /* If we have no xid, try to find one. */
4716 330 : if (!TransactionIdIsValid(xid))
4717 170 : xid = TwoPhaseGetXidByVirtualXID(vxid, &more);
4718 330 : if (!TransactionIdIsValid(xid))
4719 : {
4720 : Assert(!more);
4721 142 : return true;
4722 : }
4723 :
4724 : /* Check or wait for XID completion. */
4725 188 : SET_LOCKTAG_TRANSACTION(tag, xid);
4726 188 : lar = LockAcquire(&tag, ShareLock, false, !wait);
4727 188 : if (lar == LOCKACQUIRE_NOT_AVAIL)
4728 0 : return false;
4729 188 : LockRelease(&tag, ShareLock, false);
4730 188 : } while (more);
4731 :
4732 188 : return true;
4733 : }
4734 :
4735 : /*
4736 : * VirtualXactLock
4737 : *
4738 : * If wait = true, wait as long as the given VXID or any XID acquired by the
4739 : * same transaction is still running. Then, return true.
4740 : *
4741 : * If wait = false, just check whether that VXID or one of those XIDs is still
4742 : * running, and return true or false.
4743 : */
4744 : bool
4745 526 : VirtualXactLock(VirtualTransactionId vxid, bool wait)
4746 : {
4747 : LOCKTAG tag;
4748 : PGPROC *proc;
4749 526 : TransactionId xid = InvalidTransactionId;
4750 :
4751 : Assert(VirtualTransactionIdIsValid(vxid));
4752 :
4753 526 : if (VirtualTransactionIdIsRecoveredPreparedXact(vxid))
4754 : /* no vxid lock; localTransactionId is a normal, locked XID */
4755 2 : return XactLockForVirtualXact(vxid, vxid.localTransactionId, wait);
4756 :
4757 524 : SET_LOCKTAG_VIRTUALTRANSACTION(tag, vxid);
4758 :
4759 : /*
4760 : * If a lock table entry must be made, this is the PGPROC on whose behalf
4761 : * it must be done. Note that the transaction might end or the PGPROC
4762 : * might be reassigned to a new backend before we get around to examining
4763 : * it, but it doesn't matter. If we find upon examination that the
4764 : * relevant lxid is no longer running here, that's enough to prove that
4765 : * it's no longer running anywhere.
4766 : */
4767 524 : proc = ProcNumberGetProc(vxid.procNumber);
4768 524 : if (proc == NULL)
4769 6 : return XactLockForVirtualXact(vxid, InvalidTransactionId, wait);
4770 :
4771 : /*
4772 : * We must acquire this lock before checking the procNumber and lxid
4773 : * against the ones we're waiting for. The target backend will only set
4774 : * or clear lxid while holding this lock.
4775 : */
4776 518 : LWLockAcquire(&proc->fpInfoLock, LW_EXCLUSIVE);
4777 :
4778 518 : if (proc->vxid.procNumber != vxid.procNumber
4779 518 : || proc->fpLocalTransactionId != vxid.localTransactionId)
4780 : {
4781 : /* VXID ended */
4782 28 : LWLockRelease(&proc->fpInfoLock);
4783 28 : return XactLockForVirtualXact(vxid, InvalidTransactionId, wait);
4784 : }
4785 :
4786 : /*
4787 : * If we aren't asked to wait, there's no need to set up a lock table
4788 : * entry. The transaction is still in progress, so just return false.
4789 : */
4790 490 : if (!wait)
4791 : {
4792 30 : LWLockRelease(&proc->fpInfoLock);
4793 30 : return false;
4794 : }
4795 :
4796 : /*
4797 : * OK, we're going to need to sleep on the VXID. But first, we must set
4798 : * up the primary lock table entry, if needed (ie, convert the proc's
4799 : * fast-path lock on its VXID to a regular lock).
4800 : */
4801 460 : if (proc->fpVXIDLock)
4802 : {
4803 : PROCLOCK *proclock;
4804 : uint32 hashcode;
4805 : LWLock *partitionLock;
4806 :
4807 446 : hashcode = LockTagHashCode(&tag);
4808 :
4809 446 : partitionLock = LockHashPartitionLock(hashcode);
4810 446 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4811 :
4812 446 : proclock = SetupLockInTable(LockMethods[DEFAULT_LOCKMETHOD], proc,
4813 : &tag, hashcode, ExclusiveLock);
4814 446 : if (!proclock)
4815 : {
4816 0 : LWLockRelease(partitionLock);
4817 0 : LWLockRelease(&proc->fpInfoLock);
4818 0 : ereport(ERROR,
4819 : (errcode(ERRCODE_OUT_OF_MEMORY),
4820 : errmsg("out of shared memory"),
4821 : errhint("You might need to increase \"%s\".", "max_locks_per_transaction")));
4822 : }
4823 446 : GrantLock(proclock->tag.myLock, proclock, ExclusiveLock);
4824 :
4825 446 : LWLockRelease(partitionLock);
4826 :
4827 446 : proc->fpVXIDLock = false;
4828 : }
4829 :
4830 : /*
4831 : * If the proc has an XID now, we'll avoid a TwoPhaseGetXidByVirtualXID()
4832 : * search. The proc might have assigned this XID but not yet locked it,
4833 : * in which case the proc will lock this XID before releasing the VXID.
4834 : * The fpInfoLock critical section excludes VirtualXactLockTableCleanup(),
4835 : * so we won't save an XID of a different VXID. It doesn't matter whether
4836 : * we save this before or after setting up the primary lock table entry.
4837 : */
4838 460 : xid = proc->xid;
4839 :
4840 : /* Done with proc->fpLockBits */
4841 460 : LWLockRelease(&proc->fpInfoLock);
4842 :
4843 : /* Time to wait. */
4844 460 : (void) LockAcquire(&tag, ShareLock, false, false);
4845 :
4846 410 : LockRelease(&tag, ShareLock, false);
4847 410 : return XactLockForVirtualXact(vxid, xid, wait);
4848 : }
4849 :
4850 : /*
4851 : * LockWaiterCount
4852 : *
4853 : * Find the number of lock requester on this locktag
4854 : */
4855 : int
4856 134398 : LockWaiterCount(const LOCKTAG *locktag)
4857 : {
4858 134398 : LOCKMETHODID lockmethodid = locktag->locktag_lockmethodid;
4859 : LOCK *lock;
4860 : bool found;
4861 : uint32 hashcode;
4862 : LWLock *partitionLock;
4863 134398 : int waiters = 0;
4864 :
4865 134398 : if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
4866 0 : elog(ERROR, "unrecognized lock method: %d", lockmethodid);
4867 :
4868 134398 : hashcode = LockTagHashCode(locktag);
4869 134398 : partitionLock = LockHashPartitionLock(hashcode);
4870 134398 : LWLockAcquire(partitionLock, LW_EXCLUSIVE);
4871 :
4872 134398 : lock = (LOCK *) hash_search_with_hash_value(LockMethodLockHash,
4873 : locktag,
4874 : hashcode,
4875 : HASH_FIND,
4876 : &found);
4877 134398 : if (found)
4878 : {
4879 : Assert(lock != NULL);
4880 40 : waiters = lock->nRequested;
4881 : }
4882 134398 : LWLockRelease(partitionLock);
4883 :
4884 134398 : return waiters;
4885 : }
|
