proc.c 44.1 KB
Newer Older
1 2
/*-------------------------------------------------------------------------
 *
3
 * proc.c
4
 *	  routines to manage per-process shared memory data structure
5
 *
6
 * Portions Copyright (c) 1996-2008, PostgreSQL Global Development Group
B
Add:  
Bruce Momjian 已提交
7
 * Portions Copyright (c) 1994, Regents of the University of California
8 9 10
 *
 *
 * IDENTIFICATION
11
 *	  $PostgreSQL: pgsql/src/backend/storage/lmgr/proc.c,v 1.198 2008/01/01 19:45:52 momjian Exp $
12 13 14 15 16
 *
 *-------------------------------------------------------------------------
 */
/*
 * Interface (a):
17
 *		ProcSleep(), ProcWakeup(),
18 19
 *		ProcQueueAlloc() -- create a shm queue for sleeping processes
 *		ProcQueueInit() -- create a queue without allocing memory
20
 *
21 22
 * Waiting for a lock causes the backend to be put to sleep.  Whoever releases
 * the lock wakes the process up again (and gives it an error code so it knows
23 24 25 26
 * whether it was awoken on an error condition).
 *
 * Interface (b):
 *
27 28
 * ProcReleaseLocks -- frees the locks associated with current transaction
 *
29
 * ProcKill -- destroys the shared memory state (and locks)
30
 * associated with the process.
31
 */
32 33
#include "postgres.h"

34
#include <signal.h>
35 36
#include <unistd.h>
#include <sys/time.h>
M
Marc G. Fournier 已提交
37

38
#include "access/transam.h"
39
#include "access/xact.h"
40
#include "miscadmin.h"
41
#include "postmaster/autovacuum.h"
42
#include "storage/ipc.h"
43
#include "storage/lmgr.h"
44
#include "storage/proc.h"
45
#include "storage/procarray.h"
46
#include "storage/spin.h"
47

48

49
/* GUC variables */
B
Bruce Momjian 已提交
50
int			DeadlockTimeout = 1000;
51
int			StatementTimeout = 0;
52
bool		log_lock_waits = false;
M
 
Marc G. Fournier 已提交
53

54
/* Pointer to this process's PGPROC struct, if any */
J
Jan Wieck 已提交
55
PGPROC	   *MyProc = NULL;
56 57

/*
J
Jan Wieck 已提交
58
 * This spinlock protects the freelist of recycled PGPROC structures.
59
 * We cannot use an LWLock because the LWLock manager depends on already
J
Jan Wieck 已提交
60
 * having a PGPROC and a wait semaphore!  But these structures are touched
61 62
 * relatively infrequently (only at backend startup or shutdown) and not for
 * very long, so a spinlock is okay.
63
 */
64
NON_EXEC_STATIC slock_t *ProcStructLock = NULL;
65

66
/* Pointers to shared-memory structures */
67
NON_EXEC_STATIC PROC_HDR *ProcGlobal = NULL;
68
NON_EXEC_STATIC PGPROC *AuxiliaryProcs = NULL;
69

70 71
/* If we are waiting for a lock, this points to the associated LOCALLOCK */
static LOCALLOCK *lockAwaited = NULL;
72

73 74 75
/* Mark these volatile because they can be changed by signal handler */
static volatile bool statement_timeout_active = false;
static volatile bool deadlock_timeout_active = false;
76
static volatile DeadLockState deadlock_state = DS_NOT_YET_CHECKED;
77
volatile bool cancel_from_timeout = false;
B
Bruce Momjian 已提交
78

79 80 81
/* timeout_start_time is set when log_lock_waits is true */
static TimestampTz timeout_start_time;

82
/* statement_fin_time is valid only if statement_timeout_active is true */
83
static TimestampTz statement_fin_time;
84 85


86
static void RemoveProcFromArray(int code, Datum arg);
87
static void ProcKill(int code, Datum arg);
88
static void AuxiliaryProcKill(int code, Datum arg);
89
static bool CheckStatementTimeout(void);
90

V
Vadim B. Mikheev 已提交
91

92 93 94
/*
 * Report shared-memory space needed by InitProcGlobal.
 */
95
Size
96
ProcGlobalShmemSize(void)
97
{
98 99 100 101
	Size		size = 0;

	/* ProcGlobal */
	size = add_size(size, sizeof(PROC_HDR));
102 103
	/* AuxiliaryProcs */
	size = add_size(size, mul_size(NUM_AUXILIARY_PROCS, sizeof(PGPROC)));
104
	/* MyProcs, including autovacuum */
105 106 107
	size = add_size(size, mul_size(MaxBackends, sizeof(PGPROC)));
	/* ProcStructLock */
	size = add_size(size, sizeof(slock_t));
108 109 110 111

	return size;
}

112 113 114 115
/*
 * Report number of semaphores needed by InitProcGlobal.
 */
int
116
ProcGlobalSemas(void)
117
{
118 119 120 121
	/*
	 * We need a sema per backend (including autovacuum), plus one for each
	 * auxiliary process.
	 */
122
	return MaxBackends + NUM_AUXILIARY_PROCS;
123 124
}

125 126
/*
 * InitProcGlobal -
127 128
 *	  Initialize the global process table during postmaster or standalone
 *	  backend startup.
129
 *
130
 *	  We also create all the per-process semaphores we will need to support
131 132 133 134 135 136 137
 *	  the requested number of backends.  We used to allocate semaphores
 *	  only when backends were actually started up, but that is bad because
 *	  it lets Postgres fail under load --- a lot of Unix systems are
 *	  (mis)configured with small limits on the number of semaphores, and
 *	  running out when trying to start another backend is a common failure.
 *	  So, now we grab enough semaphores to support the desired max number
 *	  of backends immediately at initialization --- if the sysadmin has set
138 139
 *	  MaxConnections or autovacuum_max_workers higher than his kernel will
 *	  support, he'll find out sooner rather than later.
140 141 142 143
 *
 *	  Another reason for creating semaphores here is that the semaphore
 *	  implementation typically requires us to create semaphores in the
 *	  postmaster, not in backends.
144 145
 *
 * Note: this is NOT called by individual backends under a postmaster,
146
 * not even in the EXEC_BACKEND case.  The ProcGlobal and AuxiliaryProcs
147
 * pointers must be propagated specially for EXEC_BACKEND operation.
148 149
 */
void
150
InitProcGlobal(void)
151
{
152 153 154
	PGPROC	   *procs;
	int			i;
	bool		found;
155

156
	/* Create the ProcGlobal shared structure */
157
	ProcGlobal = (PROC_HDR *)
158 159
		ShmemInitStruct("Proc Header", sizeof(PROC_HDR), &found);
	Assert(!found);
160

161
	/*
162 163
	 * Create the PGPROC structures for auxiliary (bgwriter) processes, too.
	 * These do not get linked into the freeProcs list.
164
	 */
165 166
	AuxiliaryProcs = (PGPROC *)
		ShmemInitStruct("AuxiliaryProcs", NUM_AUXILIARY_PROCS * sizeof(PGPROC),
167 168
						&found);
	Assert(!found);
169

170 171 172 173
	/*
	 * Initialize the data structures.
	 */
	ProcGlobal->freeProcs = INVALID_OFFSET;
174
	ProcGlobal->autovacFreeProcs = INVALID_OFFSET;
175

176
	ProcGlobal->spins_per_delay = DEFAULT_SPINS_PER_DELAY;
177

178 179 180
	/*
	 * Pre-create the PGPROC structures and create a semaphore for each.
	 */
181
	procs = (PGPROC *) ShmemAlloc((MaxConnections) * sizeof(PGPROC));
182 183 184 185
	if (!procs)
		ereport(FATAL,
				(errcode(ERRCODE_OUT_OF_MEMORY),
				 errmsg("out of shared memory")));
186 187
	MemSet(procs, 0, MaxConnections * sizeof(PGPROC));
	for (i = 0; i < MaxConnections; i++)
188 189 190 191 192
	{
		PGSemaphoreCreate(&(procs[i].sem));
		procs[i].links.next = ProcGlobal->freeProcs;
		ProcGlobal->freeProcs = MAKE_OFFSET(&procs[i]);
	}
193

194 195 196 197 198 199 200 201 202 203 204 205 206
	procs = (PGPROC *) ShmemAlloc((autovacuum_max_workers) * sizeof(PGPROC));
	if (!procs)
		ereport(FATAL,
				(errcode(ERRCODE_OUT_OF_MEMORY),
				 errmsg("out of shared memory")));
	MemSet(procs, 0, autovacuum_max_workers * sizeof(PGPROC));
	for (i = 0; i < autovacuum_max_workers; i++)
	{
		PGSemaphoreCreate(&(procs[i].sem));
		procs[i].links.next = ProcGlobal->autovacFreeProcs;
		ProcGlobal->autovacFreeProcs = MAKE_OFFSET(&procs[i]);
	}

207 208
	MemSet(AuxiliaryProcs, 0, NUM_AUXILIARY_PROCS * sizeof(PGPROC));
	for (i = 0; i < NUM_AUXILIARY_PROCS; i++)
209
	{
B
Bruce Momjian 已提交
210
		AuxiliaryProcs[i].pid = 0;		/* marks auxiliary proc as not in use */
211
		PGSemaphoreCreate(&(AuxiliaryProcs[i].sem));
212
	}
213 214 215 216

	/* Create ProcStructLock spinlock, too */
	ProcStructLock = (slock_t *) ShmemAlloc(sizeof(slock_t));
	SpinLockInit(ProcStructLock);
217 218
}

219
/*
220
 * InitProcess -- initialize a per-process data structure for this backend
221 222
 */
void
223
InitProcess(void)
224
{
225 226
	/* use volatile pointer to prevent code rearrangement */
	volatile PROC_HDR *procglobal = ProcGlobal;
227 228
	SHMEM_OFFSET myOffset;
	int			i;
229 230

	/*
231 232
	 * ProcGlobal should be set up already (if we are a backend, we inherit
	 * this by fork() or EXEC_BACKEND mechanism from the postmaster).
233
	 */
234
	if (procglobal == NULL)
235
		elog(PANIC, "proc header uninitialized");
236 237

	if (MyProc != NULL)
238
		elog(ERROR, "you already exist");
239

240
	/*
B
Bruce Momjian 已提交
241 242
	 * Try to get a proc struct from the free list.  If this fails, we must be
	 * out of PGPROC structures (not to mention semaphores).
243
	 *
B
Bruce Momjian 已提交
244 245
	 * While we are holding the ProcStructLock, also copy the current shared
	 * estimate of spins_per_delay to local storage.
246
	 */
247
	SpinLockAcquire(ProcStructLock);
248

249 250
	set_spins_per_delay(procglobal->spins_per_delay);

251 252 253 254
	if (IsAutoVacuumWorkerProcess())
		myOffset = procglobal->autovacFreeProcs;
	else
		myOffset = procglobal->freeProcs;
255 256

	if (myOffset != INVALID_OFFSET)
257
	{
J
Jan Wieck 已提交
258
		MyProc = (PGPROC *) MAKE_PTR(myOffset);
259 260 261 262
		if (IsAutoVacuumWorkerProcess())
			procglobal->autovacFreeProcs = MyProc->links.next;
		else
			procglobal->freeProcs = MyProc->links.next;
263
		SpinLockRelease(ProcStructLock);
264 265 266 267
	}
	else
	{
		/*
B
Bruce Momjian 已提交
268 269
		 * If we reach here, all the PGPROCs are in use.  This is one of the
		 * possible places to detect "too many backends", so give the standard
270 271
		 * error message.  XXX do we need to give a different failure message
		 * in the autovacuum case?
272
		 */
273
		SpinLockRelease(ProcStructLock);
274 275 276
		ereport(FATAL,
				(errcode(ERRCODE_TOO_MANY_CONNECTIONS),
				 errmsg("sorry, too many clients already")));
277
	}
278

279
	/*
B
Bruce Momjian 已提交
280 281
	 * Initialize all fields of MyProc, except for the semaphore which was
	 * prepared for us by InitProcGlobal.
282
	 */
283
	SHMQueueElemInit(&(MyProc->links));
284
	MyProc->waitStatus = STATUS_OK;
285
	MyProc->lxid = InvalidLocalTransactionId;
286
	MyProc->xid = InvalidTransactionId;
287
	MyProc->xmin = InvalidTransactionId;
288
	MyProc->pid = MyProcPid;
289 290
	/* backendId, databaseId and roleId will be filled in later */
	MyProc->backendId = InvalidBackendId;
291
	MyProc->databaseId = InvalidOid;
292
	MyProc->roleId = InvalidOid;
293
	MyProc->inCommit = false;
294 295 296
	MyProc->vacuumFlags = 0;
	if (IsAutoVacuumWorkerProcess())
		MyProc->vacuumFlags |= PROC_IS_AUTOVACUUM;
297 298 299
	MyProc->lwWaiting = false;
	MyProc->lwExclusive = false;
	MyProc->lwWaitLink = NULL;
300
	MyProc->waitLock = NULL;
301
	MyProc->waitProcLock = NULL;
302 303
	for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
		SHMQueueInit(&(MyProc->myProcLocks[i]));
304

305
	/*
306
	 * We might be reusing a semaphore that belonged to a failed process. So
B
Bruce Momjian 已提交
307
	 * be careful and reinitialize its value here.	(This is not strictly
308
	 * necessary anymore, but seems like a good idea for cleanliness.)
309
	 */
310
	PGSemaphoreReset(&MyProc->sem);
311

312
	/*
313
	 * Arrange to clean up at backend exit.
314
	 */
315
	on_shmem_exit(ProcKill, 0);
316 317

	/*
B
Bruce Momjian 已提交
318 319
	 * Now that we have a PGPROC, we could try to acquire locks, so initialize
	 * the deadlock checker.
320 321
	 */
	InitDeadLockChecking();
322 323
}

324 325 326 327 328 329 330 331 332 333 334 335 336
/*
 * InitProcessPhase2 -- make MyProc visible in the shared ProcArray.
 *
 * This is separate from InitProcess because we can't acquire LWLocks until
 * we've created a PGPROC, but in the EXEC_BACKEND case there is a good deal
 * of stuff to be done before this step that will require LWLock access.
 */
void
InitProcessPhase2(void)
{
	Assert(MyProc != NULL);

	/*
B
Bruce Momjian 已提交
337 338 339
	 * We should now know what database we're in, so advertise that.  (We need
	 * not do any locking here, since no other backend can yet see our
	 * PGPROC.)
340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
	 */
	Assert(OidIsValid(MyDatabaseId));
	MyProc->databaseId = MyDatabaseId;

	/*
	 * Add our PGPROC to the PGPROC array in shared memory.
	 */
	ProcArrayAdd(MyProc);

	/*
	 * Arrange to clean that up at backend exit.
	 */
	on_shmem_exit(RemoveProcFromArray, 0);
}

355
/*
356
 * InitAuxiliaryProcess -- create a per-auxiliary-process data structure
357
 *
358 359
 * This is called by bgwriter and similar processes so that they will have a
 * MyProc value that's real enough to let them wait for LWLocks.  The PGPROC
360
 * and sema that are assigned are one of the extra ones created during
361
 * InitProcGlobal.
362
 *
363
 * Auxiliary processes are presently not expected to wait for real (lockmgr)
364
 * locks, so we need not set up the deadlock checker.  They are never added
B
Bruce Momjian 已提交
365
 * to the ProcArray or the sinval messaging mechanism, either.	They also
366 367
 * don't get a VXID assigned, since this is only useful when we actually
 * hold lockmgr locks.
368 369
 */
void
370
InitAuxiliaryProcess(void)
371
{
372
	PGPROC	   *auxproc;
373
	int			proctype;
374
	int			i;
J
Jan Wieck 已提交
375

376
	/*
377 378
	 * ProcGlobal should be set up already (if we are a backend, we inherit
	 * this by fork() or EXEC_BACKEND mechanism from the postmaster).
379
	 */
380
	if (ProcGlobal == NULL || AuxiliaryProcs == NULL)
381
		elog(PANIC, "proc header uninitialized");
382 383

	if (MyProc != NULL)
384
		elog(ERROR, "you already exist");
385

386
	/*
387
	 * We use the ProcStructLock to protect assignment and releasing of
388
	 * AuxiliaryProcs entries.
389
	 *
B
Bruce Momjian 已提交
390 391
	 * While we are holding the ProcStructLock, also copy the current shared
	 * estimate of spins_per_delay to local storage.
392 393 394 395 396
	 */
	SpinLockAcquire(ProcStructLock);

	set_spins_per_delay(ProcGlobal->spins_per_delay);

397
	/*
398
	 * Find a free auxproc ... *big* trouble if there isn't one ...
399
	 */
400
	for (proctype = 0; proctype < NUM_AUXILIARY_PROCS; proctype++)
401
	{
402 403
		auxproc = &AuxiliaryProcs[proctype];
		if (auxproc->pid == 0)
404 405
			break;
	}
406
	if (proctype >= NUM_AUXILIARY_PROCS)
407 408
	{
		SpinLockRelease(ProcStructLock);
409
		elog(FATAL, "all AuxiliaryProcs are in use");
410
	}
411

412
	/* Mark auxiliary proc as in use by me */
413
	/* use volatile pointer to prevent code rearrangement */
414
	((volatile PGPROC *) auxproc)->pid = MyProcPid;
415

416
	MyProc = auxproc;
417 418 419

	SpinLockRelease(ProcStructLock);

420
	/*
421 422
	 * Initialize all fields of MyProc, except for the semaphore which was
	 * prepared for us by InitProcGlobal.
423 424
	 */
	SHMQueueElemInit(&(MyProc->links));
425
	MyProc->waitStatus = STATUS_OK;
426
	MyProc->lxid = InvalidLocalTransactionId;
427 428
	MyProc->xid = InvalidTransactionId;
	MyProc->xmin = InvalidTransactionId;
429
	MyProc->backendId = InvalidBackendId;
430
	MyProc->databaseId = InvalidOid;
431
	MyProc->roleId = InvalidOid;
432
	MyProc->inCommit = false;
433 434
	/* we don't set the "is autovacuum" flag in the launcher */
	MyProc->vacuumFlags = 0;
435 436 437 438
	MyProc->lwWaiting = false;
	MyProc->lwExclusive = false;
	MyProc->lwWaitLink = NULL;
	MyProc->waitLock = NULL;
439
	MyProc->waitProcLock = NULL;
440 441
	for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
		SHMQueueInit(&(MyProc->myProcLocks[i]));
442 443

	/*
B
Bruce Momjian 已提交
444
	 * We might be reusing a semaphore that belonged to a failed process. So
B
Bruce Momjian 已提交
445
	 * be careful and reinitialize its value here.	(This is not strictly
446
	 * necessary anymore, but seems like a good idea for cleanliness.)
447
	 */
448
	PGSemaphoreReset(&MyProc->sem);
449 450 451 452

	/*
	 * Arrange to clean up at process exit.
	 */
453
	on_shmem_exit(AuxiliaryProcKill, Int32GetDatum(proctype));
454 455
}

456 457 458 459 460 461 462 463 464 465
/*
 * Check whether there are at least N free PGPROC objects.
 *
 * Note: this is designed on the assumption that N will generally be small.
 */
bool
HaveNFreeProcs(int n)
{
	SHMEM_OFFSET offset;
	PGPROC	   *proc;
B
Bruce Momjian 已提交
466

467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
	/* use volatile pointer to prevent code rearrangement */
	volatile PROC_HDR *procglobal = ProcGlobal;

	SpinLockAcquire(ProcStructLock);

	offset = procglobal->freeProcs;

	while (n > 0 && offset != INVALID_OFFSET)
	{
		proc = (PGPROC *) MAKE_PTR(offset);
		offset = proc->links.next;
		n--;
	}

	SpinLockRelease(ProcStructLock);

	return (n <= 0);
}

486 487 488
/*
 * Cancel any pending wait for lock, when aborting a transaction.
 *
489 490
 * Returns true if we had been waiting for a lock, else false.
 *
491
 * (Normally, this would only happen if we accept a cancel/die
492
 * interrupt while waiting; but an ereport(ERROR) while waiting is
493 494
 * within the realm of possibility, too.)
 */
495
bool
496 497
LockWaitCancel(void)
{
498 499
	LWLockId	partitionLock;

500
	/* Nothing to do if we weren't waiting for a lock */
501
	if (lockAwaited == NULL)
502 503
		return false;

504
	/* Turn off the deadlock timer, if it's still running (see ProcSleep) */
505
	disable_sig_alarm(false);
506 507

	/* Unlink myself from the wait queue, if on it (might not be anymore!) */
508
	partitionLock = LockHashPartitionLock(lockAwaited->hashcode);
509
	LWLockAcquire(partitionLock, LW_EXCLUSIVE);
510

511
	if (MyProc->links.next != INVALID_OFFSET)
512 513
	{
		/* We could not have been granted the lock yet */
514
		RemoveFromWaitQueue(MyProc, lockAwaited->hashcode);
515 516 517 518 519
	}
	else
	{
		/*
		 * Somebody kicked us off the lock queue already.  Perhaps they
B
Bruce Momjian 已提交
520 521 522
		 * granted us the lock, or perhaps they detected a deadlock. If they
		 * did grant us the lock, we'd better remember it in our local lock
		 * table.
523
		 */
524 525
		if (MyProc->waitStatus == STATUS_OK)
			GrantAwaitedLock();
526 527
	}

528
	lockAwaited = NULL;
529

530
	LWLockRelease(partitionLock);
H
Hiroshi Inoue 已提交
531

532
	/*
533
	 * We used to do PGSemaphoreReset() here to ensure that our proc's wait
B
Bruce Momjian 已提交
534 535 536 537 538 539
	 * semaphore is reset to zero.	This prevented a leftover wakeup signal
	 * from remaining in the semaphore if someone else had granted us the lock
	 * we wanted before we were able to remove ourselves from the wait-list.
	 * However, now that ProcSleep loops until waitStatus changes, a leftover
	 * wakeup signal isn't harmful, and it seems not worth expending cycles to
	 * get rid of a signal that most likely isn't there.
540
	 */
541 542

	/*
B
Bruce Momjian 已提交
543 544
	 * Return true even if we were kicked off the lock before we were able to
	 * remove ourselves.
545 546
	 */
	return true;
H
Hiroshi Inoue 已提交
547
}
548

549

550
/*
551
 * ProcReleaseLocks() -- release locks associated with current transaction
552
 *			at main transaction commit or abort
553 554 555 556 557 558
 *
 * At main transaction commit, we release all locks except session locks.
 * At main transaction abort, we release all locks including session locks;
 * this lets us clean up after a VACUUM FULL failure.
 *
 * At subtransaction commit, we don't release any locks (so this func is not
559
 * needed at all); we will defer the releasing to the parent transaction.
560
 * At subtransaction abort, we release all locks held by the subtransaction;
561 562
 * this is implemented by retail releasing of the locks under control of
 * the ResourceOwner mechanism.
563 564
 *
 * Note that user locks are not released in any case.
565 566
 */
void
567
ProcReleaseLocks(bool isCommit)
568
{
569 570
	if (!MyProc)
		return;
571 572 573
	/* If waiting, get off wait queue (should only be needed after error) */
	LockWaitCancel();
	/* Release locks */
574
	LockReleaseAll(DEFAULT_LOCKMETHOD, !isCommit);
575 576 577
}


578 579 580 581 582 583 584
/*
 * RemoveProcFromArray() -- Remove this process from the shared ProcArray.
 */
static void
RemoveProcFromArray(int code, Datum arg)
{
	Assert(MyProc != NULL);
585
	ProcArrayRemove(MyProc, InvalidTransactionId);
586 587
}

588 589
/*
 * ProcKill() -- Destroy the per-proc data structure for
590
 *		this process. Release any of its held LW locks.
591 592
 */
static void
593
ProcKill(int code, Datum arg)
594
{
595 596 597
	/* use volatile pointer to prevent code rearrangement */
	volatile PROC_HDR *procglobal = ProcGlobal;

598
	Assert(MyProc != NULL);
599

600
	/*
B
Bruce Momjian 已提交
601 602
	 * Release any LW locks I am holding.  There really shouldn't be any, but
	 * it's cheap to check again before we cut the knees off the LWLock
603
	 * facility by releasing our PGPROC ...
604
	 */
605
	LWLockReleaseAll();
606

607
	SpinLockAcquire(ProcStructLock);
608

J
Jan Wieck 已提交
609
	/* Return PGPROC structure (and semaphore) to freelist */
610 611 612 613 614 615 616 617 618 619
	if (IsAutoVacuumWorkerProcess())
	{
		MyProc->links.next = procglobal->autovacFreeProcs;
		procglobal->autovacFreeProcs = MAKE_OFFSET(MyProc);
	}
	else
	{
		MyProc->links.next = procglobal->freeProcs;
		procglobal->freeProcs = MAKE_OFFSET(MyProc);
	}
620

J
Jan Wieck 已提交
621
	/* PGPROC struct isn't mine anymore */
622
	MyProc = NULL;
623

624 625 626
	/* Update shared estimate of spins_per_delay */
	procglobal->spins_per_delay = update_spins_per_delay(procglobal->spins_per_delay);

627
	SpinLockRelease(ProcStructLock);
628 629 630 631

	/* wake autovac launcher if needed -- see comments in FreeWorkerInfo */
	if (AutovacuumLauncherPid != 0)
		kill(AutovacuumLauncherPid, SIGUSR1);
632 633 634
}

/*
635 636 637
 * AuxiliaryProcKill() -- Cut-down version of ProcKill for auxiliary
 *		processes (bgwriter, etc).	The PGPROC and sema are not released, only
 *		marked as not-in-use.
638 639
 */
static void
640
AuxiliaryProcKill(int code, Datum arg)
641
{
B
Bruce Momjian 已提交
642
	int			proctype = DatumGetInt32(arg);
643
	PGPROC	   *auxproc;
J
Jan Wieck 已提交
644

645
	Assert(proctype >= 0 && proctype < NUM_AUXILIARY_PROCS);
J
Jan Wieck 已提交
646

647
	auxproc = &AuxiliaryProcs[proctype];
J
Jan Wieck 已提交
648

649
	Assert(MyProc == auxproc);
650

651
	/* Release any LW locks I am holding (see notes above) */
652 653
	LWLockReleaseAll();

654 655
	SpinLockAcquire(ProcStructLock);

656
	/* Mark auxiliary proc no longer in use */
657 658
	MyProc->pid = 0;

J
Jan Wieck 已提交
659
	/* PGPROC struct isn't mine anymore */
660
	MyProc = NULL;
661 662 663 664 665

	/* Update shared estimate of spins_per_delay */
	ProcGlobal->spins_per_delay = update_spins_per_delay(ProcGlobal->spins_per_delay);

	SpinLockRelease(ProcStructLock);
666 667
}

668

669 670
/*
 * ProcQueue package: routines for putting processes to sleep
671
 *		and  waking them up
672 673 674 675 676 677 678 679
 */

/*
 * ProcQueueAlloc -- alloc/attach to a shared memory process queue
 *
 * Returns: a pointer to the queue or NULL
 * Side Effects: Initializes the queue if we allocated one
 */
680
#ifdef NOT_USED
681
PROC_QUEUE *
682 683
ProcQueueAlloc(char *name)
{
684 685
	bool		found;
	PROC_QUEUE *queue = (PROC_QUEUE *)
B
Bruce Momjian 已提交
686
	ShmemInitStruct(name, sizeof(PROC_QUEUE), &found);
687 688

	if (!queue)
689
		return NULL;
690 691
	if (!found)
		ProcQueueInit(queue);
692
	return queue;
693
}
694
#endif
695 696 697 698 699

/*
 * ProcQueueInit -- initialize a shared memory process queue
 */
void
700
ProcQueueInit(PROC_QUEUE *queue)
701
{
702 703
	SHMQueueInit(&(queue->links));
	queue->size = 0;
704 705 706 707
}


/*
708
 * ProcSleep -- put a process to sleep on the specified lock
709
 *
710 711
 * Caller must have set MyProc->heldLocks to reflect locks already held
 * on the lockable object by this process (under all XIDs).
712
 *
713
 * The lock table's partition lock must be held at entry, and will be held
714
 * at exit.
715
 *
716
 * Result: STATUS_OK if we acquired the lock, STATUS_ERROR if not (deadlock).
717
 *
718
 * ASSUME: that no one will fiddle with the queue until after
719
 *		we release the partition lock.
720 721
 *
 * NOTES: The process queue is now a priority queue for locking.
722 723 724
 *
 * P() on the semaphore should put us to sleep.  The process
 * semaphore is normally zero, so when we try to acquire it, we sleep.
725 726
 */
int
727
ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
728
{
729 730 731
	LOCKMODE	lockmode = locallock->tag.mode;
	LOCK	   *lock = locallock->lock;
	PROCLOCK   *proclock = locallock->proclock;
732 733
	uint32		hashcode = locallock->hashcode;
	LWLockId	partitionLock = LockHashPartitionLock(hashcode);
734
	PROC_QUEUE *waitQueue = &(lock->waitProcs);
735
	LOCKMASK	myHeldLocks = MyProc->heldLocks;
736
	bool		early_deadlock = false;
B
Bruce Momjian 已提交
737
	bool		allow_autovacuum_cancel = true;
738
	int			myWaitStatus;
J
Jan Wieck 已提交
739
	PGPROC	   *proc;
740
	int			i;
741

742
	/*
743 744
	 * Determine where to add myself in the wait queue.
	 *
745 746 747 748
	 * Normally I should go at the end of the queue.  However, if I already
	 * hold locks that conflict with the request of any previous waiter, put
	 * myself in the queue just in front of the first such waiter. This is not
	 * a necessary step, since deadlock detection would move me to before that
B
Bruce Momjian 已提交
749 750
	 * waiter anyway; but it's relatively cheap to detect such a conflict
	 * immediately, and avoid delaying till deadlock timeout.
751
	 *
752 753
	 * Special case: if I find I should go in front of some waiter, check to
	 * see if I conflict with already-held locks or the requests before that
B
Bruce Momjian 已提交
754 755 756 757
	 * waiter.	If not, then just grant myself the requested lock immediately.
	 * This is the same as the test for immediate grant in LockAcquire, except
	 * we are only considering the part of the wait queue before my insertion
	 * point.
758 759
	 */
	if (myHeldLocks != 0)
V
Vadim B. Mikheev 已提交
760
	{
761
		LOCKMASK	aheadRequests = 0;
762

J
Jan Wieck 已提交
763
		proc = (PGPROC *) MAKE_PTR(waitQueue->links.next);
764
		for (i = 0; i < waitQueue->size; i++)
V
Vadim B. Mikheev 已提交
765
		{
766
			/* Must he wait for me? */
B
Bruce Momjian 已提交
767
			if (lockMethodTable->conflictTab[proc->waitLockMode] & myHeldLocks)
V
Vadim B. Mikheev 已提交
768
			{
769
				/* Must I wait for him ? */
B
Bruce Momjian 已提交
770
				if (lockMethodTable->conflictTab[lockmode] & proc->heldLocks)
771
				{
772
					/*
B
Bruce Momjian 已提交
773 774 775 776 777
					 * Yes, so we have a deadlock.	Easiest way to clean up
					 * correctly is to call RemoveFromWaitQueue(), but we
					 * can't do that until we are *on* the wait queue. So, set
					 * a flag to check below, and break out of loop.  Also,
					 * record deadlock info for later message.
778
					 */
779
					RememberSimpleDeadLock(MyProc, lockmode, lock, proc);
780 781
					early_deadlock = true;
					break;
782
				}
783
				/* I must go before this waiter.  Check special case. */
B
Bruce Momjian 已提交
784
				if ((lockMethodTable->conflictTab[lockmode] & aheadRequests) == 0 &&
785 786 787
					LockCheckConflicts(lockMethodTable,
									   lockmode,
									   lock,
788
									   proclock,
789
									   MyProc) == STATUS_OK)
790
				{
791
					/* Skip the wait and just grant myself the lock. */
792
					GrantLock(lock, proclock, lockmode);
793
					GrantAwaitedLock();
794
					return STATUS_OK;
795 796
				}
				/* Break out of loop to put myself before him */
V
Vadim B. Mikheev 已提交
797
				break;
798
			}
799
			/* Nope, so advance to next waiter */
800
			aheadRequests |= LOCKBIT_ON(proc->waitLockMode);
J
Jan Wieck 已提交
801
			proc = (PGPROC *) MAKE_PTR(proc->links.next);
V
Vadim B. Mikheev 已提交
802
		}
B
Bruce Momjian 已提交
803

804
		/*
B
Bruce Momjian 已提交
805 806
		 * If we fall out of loop normally, proc points to waitQueue head, so
		 * we will insert at tail of queue as desired.
807
		 */
808 809 810 811
	}
	else
	{
		/* I hold no locks, so I can't push in front of anyone. */
J
Jan Wieck 已提交
812
		proc = (PGPROC *) &(waitQueue->links);
V
Vadim B. Mikheev 已提交
813
	}
814

815
	/*
B
Bruce Momjian 已提交
816
	 * Insert self into queue, ahead of the given proc (or at tail of queue).
817
	 */
818
	SHMQueueInsertBefore(&(proc->links), &(MyProc->links));
B
Bruce Momjian 已提交
819
	waitQueue->size++;
820

821
	lock->waitMask |= LOCKBIT_ON(lockmode);
822

J
Jan Wieck 已提交
823
	/* Set up wait information in PGPROC object, too */
824
	MyProc->waitLock = lock;
825
	MyProc->waitProcLock = proclock;
826 827
	MyProc->waitLockMode = lockmode;

828
	MyProc->waitStatus = STATUS_WAITING;
829 830

	/*
B
Bruce Momjian 已提交
831 832 833
	 * If we detected deadlock, give up without waiting.  This must agree with
	 * CheckDeadLock's recovery code, except that we shouldn't release the
	 * semaphore since we haven't tried to lock it yet.
834 835 836
	 */
	if (early_deadlock)
	{
837
		RemoveFromWaitQueue(MyProc, hashcode);
838 839
		return STATUS_ERROR;
	}
840

841
	/* mark that we are waiting for a lock */
842
	lockAwaited = locallock;
843

844
	/*
845
	 * Release the lock table's partition lock.
846
	 *
847
	 * NOTE: this may also cause us to exit critical-section state, possibly
B
Bruce Momjian 已提交
848 849
	 * allowing a cancel/die interrupt to be accepted. This is OK because we
	 * have recorded the fact that we are waiting for a lock, and so
850
	 * LockWaitCancel will clean up if cancel/die happens.
851
	 */
852
	LWLockRelease(partitionLock);
853

854 855 856
	/* Reset deadlock_state before enabling the signal handler */
	deadlock_state = DS_NOT_YET_CHECKED;

857
	/*
B
Bruce Momjian 已提交
858 859 860 861
	 * Set timer so we can wake up after awhile and check for a deadlock. If a
	 * deadlock is detected, the handler releases the process's semaphore and
	 * sets MyProc->waitStatus = STATUS_ERROR, allowing us to know that we
	 * must report failure rather than success.
862
	 *
863 864
	 * By delaying the check until we've waited for a bit, we can avoid
	 * running the rather expensive deadlock-check code in most cases.
865
	 */
866
	if (!enable_sig_alarm(DeadlockTimeout, false))
867
		elog(FATAL, "could not set timer for process wakeup");
868

869
	/*
870
	 * If someone wakes us between LWLockRelease and PGSemaphoreLock,
B
Bruce Momjian 已提交
871
	 * PGSemaphoreLock will not block.	The wakeup is "saved" by the semaphore
B
Bruce Momjian 已提交
872 873 874 875 876
	 * implementation.	While this is normally good, there are cases where a
	 * saved wakeup might be leftover from a previous operation (for example,
	 * we aborted ProcWaitForSignal just before someone did ProcSendSignal).
	 * So, loop to wait again if the waitStatus shows we haven't been granted
	 * nor denied the lock yet.
877
	 *
878 879 880 881 882 883 884
	 * We pass interruptOK = true, which eliminates a window in which
	 * cancel/die interrupts would be held off undesirably.  This is a promise
	 * that we don't mind losing control to a cancel/die interrupt here.  We
	 * don't, because we have no shared-state-change work to do after being
	 * granted the lock (the grantor did it all).  We do have to worry about
	 * updating the locallock table, but if we lose control to an error,
	 * LockWaitCancel will fix that up.
885
	 */
B
Bruce Momjian 已提交
886 887
	do
	{
888
		PGSemaphoreLock(&MyProc->sem, true);
889

890 891
		/*
		 * waitStatus could change from STATUS_WAITING to something else
B
Bruce Momjian 已提交
892
		 * asynchronously.	Read it just once per loop to prevent surprising
893 894 895 896
		 * behavior (such as missing log messages).
		 */
		myWaitStatus = MyProc->waitStatus;

897 898
		/*
		 * If we are not deadlocked, but are waiting on an autovacuum-induced
B
Bruce Momjian 已提交
899
		 * task, send a signal to interrupt it.
900 901 902
		 */
		if (deadlock_state == DS_BLOCKED_BY_AUTOVACUUM && allow_autovacuum_cancel)
		{
B
Bruce Momjian 已提交
903
			PGPROC	   *autovac = GetBlockingAutoVacuumPgproc();
904 905 906 907 908 909 910 911 912 913 914

			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);

			/*
			 * Only do it if the worker is not working to protect against Xid
			 * wraparound.
			 */
			if ((autovac != NULL) &&
				(autovac->vacuumFlags & PROC_IS_AUTOVACUUM) &&
				!(autovac->vacuumFlags & PROC_VACUUM_FOR_WRAPAROUND))
			{
B
Bruce Momjian 已提交
915
				int			pid = autovac->pid;
916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938

				elog(DEBUG2, "sending cancel to blocking autovacuum pid = %d",
					 pid);

				/* don't hold the lock across the kill() syscall */
				LWLockRelease(ProcArrayLock);

				/* send the autovacuum worker Back to Old Kent Road */
				if (kill(pid, SIGINT) < 0)
				{
					/* Just a warning to allow multiple callers */
					ereport(WARNING,
							(errmsg("could not send signal to process %d: %m",
									pid)));
				}
			}
			else
				LWLockRelease(ProcArrayLock);

			/* prevent signal from being resent more than once */
			allow_autovacuum_cancel = false;
		}

939 940 941 942
		/*
		 * If awoken after the deadlock check interrupt has run, and
		 * log_lock_waits is on, then report about the wait.
		 */
943
		if (log_lock_waits && deadlock_state != DS_NOT_YET_CHECKED)
944
		{
945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962
			StringInfoData buf;
			const char *modename;
			long		secs;
			int			usecs;
			long		msecs;

			initStringInfo(&buf);
			DescribeLockTag(&buf, &locallock->tag.lock);
			modename = GetLockmodeName(locallock->tag.lock.locktag_lockmethodid,
									   lockmode);
			TimestampDifference(timeout_start_time, GetCurrentTimestamp(),
								&secs, &usecs);
			msecs = secs * 1000 + usecs / 1000;
			usecs = usecs % 1000;

			if (deadlock_state == DS_SOFT_DEADLOCK)
				ereport(LOG,
						(errmsg("process %d avoided deadlock for %s on %s by rearranging queue order after %ld.%03d ms",
B
Bruce Momjian 已提交
963
							  MyProcPid, modename, buf.data, msecs, usecs)));
964
			else if (deadlock_state == DS_HARD_DEADLOCK)
965
			{
966
				/*
B
Bruce Momjian 已提交
967 968 969 970
				 * This message is a bit redundant with the error that will be
				 * reported subsequently, but in some cases the error report
				 * might not make it to the log (eg, if it's caught by an
				 * exception handler), and we want to ensure all long-wait
971 972 973 974
				 * events get logged.
				 */
				ereport(LOG,
						(errmsg("process %d detected deadlock while waiting for %s on %s after %ld.%03d ms",
B
Bruce Momjian 已提交
975
							  MyProcPid, modename, buf.data, msecs, usecs)));
976
			}
977 978 979 980

			if (myWaitStatus == STATUS_WAITING)
				ereport(LOG,
						(errmsg("process %d still waiting for %s on %s after %ld.%03d ms",
B
Bruce Momjian 已提交
981
							  MyProcPid, modename, buf.data, msecs, usecs)));
982 983
			else if (myWaitStatus == STATUS_OK)
				ereport(LOG,
B
Bruce Momjian 已提交
984 985
					(errmsg("process %d acquired %s on %s after %ld.%03d ms",
							MyProcPid, modename, buf.data, msecs, usecs)));
986 987 988
			else
			{
				Assert(myWaitStatus == STATUS_ERROR);
B
Bruce Momjian 已提交
989

990 991
				/*
				 * Currently, the deadlock checker always kicks its own
B
Bruce Momjian 已提交
992 993 994 995 996
				 * process, which means that we'll only see STATUS_ERROR when
				 * deadlock_state == DS_HARD_DEADLOCK, and there's no need to
				 * print redundant messages.  But for completeness and
				 * future-proofing, print a message if it looks like someone
				 * else kicked us off the lock.
997 998 999 1000
				 */
				if (deadlock_state != DS_HARD_DEADLOCK)
					ereport(LOG,
							(errmsg("process %d failed to acquire %s on %s after %ld.%03d ms",
B
Bruce Momjian 已提交
1001
							  MyProcPid, modename, buf.data, msecs, usecs)));
1002 1003 1004
			}

			/*
B
Bruce Momjian 已提交
1005 1006
			 * At this point we might still need to wait for the lock. Reset
			 * state so we don't print the above messages again.
1007 1008 1009 1010
			 */
			deadlock_state = DS_NO_DEADLOCK;

			pfree(buf.data);
1011
		}
1012
	} while (myWaitStatus == STATUS_WAITING);
1013

1014
	/*
1015
	 * Disable the timer, if it's still running
B
Bruce Momjian 已提交
1016
	 */
1017
	if (!disable_sig_alarm(false))
1018
		elog(FATAL, "could not disable timer for process wakeup");
B
Bruce Momjian 已提交
1019

1020
	/*
B
Bruce Momjian 已提交
1021 1022 1023
	 * Re-acquire the lock table's partition lock.  We have to do this to hold
	 * off cancel/die interrupts before we can mess with lockAwaited (else we
	 * might have a missed or duplicated locallock update).
1024
	 */
1025
	LWLockAcquire(partitionLock, LW_EXCLUSIVE);
1026 1027 1028

	/*
	 * We no longer want LockWaitCancel to do anything.
1029
	 */
1030
	lockAwaited = NULL;
1031

1032
	/*
1033
	 * If we got the lock, be sure to remember it in the locallock table.
1034
	 */
1035
	if (MyProc->waitStatus == STATUS_OK)
1036
		GrantAwaitedLock();
1037

1038 1039 1040 1041
	/*
	 * We don't have to do anything else, because the awaker did all the
	 * necessary update of the lock table and MyProc.
	 */
1042
	return MyProc->waitStatus;
1043 1044 1045 1046 1047 1048
}


/*
 * ProcWakeup -- wake up a process by releasing its private semaphore.
 *
1049
 *	 Also remove the process from the wait queue and set its links invalid.
1050
 *	 RETURN: the next process in the wait queue.
1051
 *
1052 1053
 * The appropriate lock partition lock must be held by caller.
 *
1054 1055 1056
 * XXX: presently, this code is only used for the "success" case, and only
 * works correctly for that case.  To clean up in failure case, would need
 * to twiddle the lock's request counts too --- see RemoveFromWaitQueue.
1057
 * Hence, in practice the waitStatus parameter must be STATUS_OK.
1058
 */
J
Jan Wieck 已提交
1059
PGPROC *
1060
ProcWakeup(PGPROC *proc, int waitStatus)
1061
{
J
Jan Wieck 已提交
1062
	PGPROC	   *retProc;
1063

1064
	/* Proc should be sleeping ... */
1065 1066
	if (proc->links.prev == INVALID_OFFSET ||
		proc->links.next == INVALID_OFFSET)
1067
		return NULL;
1068
	Assert(proc->waitStatus == STATUS_WAITING);
1069

1070
	/* Save next process before we zap the list link */
J
Jan Wieck 已提交
1071
	retProc = (PGPROC *) MAKE_PTR(proc->links.next);
1072

1073
	/* Remove process from wait queue */
1074
	SHMQueueDelete(&(proc->links));
1075
	(proc->waitLock->waitProcs.size)--;
1076

1077 1078
	/* Clean up process' state and pass it the ok/fail signal */
	proc->waitLock = NULL;
1079
	proc->waitProcLock = NULL;
1080
	proc->waitStatus = waitStatus;
1081

1082
	/* And awaken it */
1083
	PGSemaphoreUnlock(&proc->sem);
1084 1085

	return retProc;
1086 1087 1088 1089
}

/*
 * ProcLockWakeup -- routine for waking up processes when a lock is
1090 1091
 *		released (or a prior waiter is aborted).  Scan all waiters
 *		for lock, waken any that are no longer blocked.
1092 1093
 *
 * The appropriate lock partition lock must be held by caller.
1094
 */
1095
void
1096
ProcLockWakeup(LockMethod lockMethodTable, LOCK *lock)
1097
{
1098 1099
	PROC_QUEUE *waitQueue = &(lock->waitProcs);
	int			queue_size = waitQueue->size;
J
Jan Wieck 已提交
1100
	PGPROC	   *proc;
1101
	LOCKMASK	aheadRequests = 0;
M
 
Marc G. Fournier 已提交
1102

1103
	Assert(queue_size >= 0);
1104

1105 1106
	if (queue_size == 0)
		return;
1107

J
Jan Wieck 已提交
1108
	proc = (PGPROC *) MAKE_PTR(waitQueue->links.next);
1109

1110 1111
	while (queue_size-- > 0)
	{
B
Bruce Momjian 已提交
1112
		LOCKMODE	lockmode = proc->waitLockMode;
M
 
Marc G. Fournier 已提交
1113 1114

		/*
B
Bruce Momjian 已提交
1115 1116
		 * Waken if (a) doesn't conflict with requests of earlier waiters, and
		 * (b) doesn't conflict with already-held locks.
M
 
Marc G. Fournier 已提交
1117
		 */
B
Bruce Momjian 已提交
1118
		if ((lockMethodTable->conflictTab[lockmode] & aheadRequests) == 0 &&
1119 1120 1121
			LockCheckConflicts(lockMethodTable,
							   lockmode,
							   lock,
1122
							   proc->waitProcLock,
1123
							   proc) == STATUS_OK)
M
 
Marc G. Fournier 已提交
1124
		{
1125
			/* OK to waken */
1126
			GrantLock(lock, proc->waitProcLock, lockmode);
1127
			proc = ProcWakeup(proc, STATUS_OK);
B
Bruce Momjian 已提交
1128

1129
			/*
B
Bruce Momjian 已提交
1130 1131 1132
			 * ProcWakeup removes proc from the lock's waiting process queue
			 * and returns the next proc in chain; don't use proc's next-link,
			 * because it's been cleared.
1133
			 */
M
 
Marc G. Fournier 已提交
1134
		}
1135
		else
1136
		{
B
Bruce Momjian 已提交
1137
			/*
B
Bruce Momjian 已提交
1138
			 * Cannot wake this guy. Remember his request for later checks.
B
Bruce Momjian 已提交
1139
			 */
1140
			aheadRequests |= LOCKBIT_ON(lockmode);
J
Jan Wieck 已提交
1141
			proc = (PGPROC *) MAKE_PTR(proc->links.next);
1142
		}
M
 
Marc G. Fournier 已提交
1143
	}
1144 1145

	Assert(waitQueue->size >= 0);
1146 1147
}

1148 1149 1150
/*
 * CheckDeadLock
 *
1151
 * We only get to this routine if we got SIGALRM after DeadlockTimeout
1152 1153
 * while waiting for a lock to be released by some other process.  Look
 * to see if there's a deadlock; if not, just return and continue waiting.
1154
 * (But signal ProcSleep to log a message, if log_lock_waits is true.)
1155 1156
 * If we have a real deadlock, remove ourselves from the lock's wait queue
 * and signal an error to ProcSleep.
1157 1158 1159
 *
 * NB: this is run inside a signal handler, so be very wary about what is done
 * here or in called routines.
1160
 */
1161
static void
1162
CheckDeadLock(void)
1163
{
1164 1165
	int			i;

1166
	/*
B
Bruce Momjian 已提交
1167 1168
	 * Acquire exclusive lock on the entire shared lock data structures. Must
	 * grab LWLocks in partition-number order to avoid LWLock deadlock.
1169 1170 1171 1172 1173 1174
	 *
	 * Note that the deadlock check interrupt had better not be enabled
	 * anywhere that this process itself holds lock partition locks, else this
	 * will wait forever.  Also note that LWLockAcquire creates a critical
	 * section, so that this routine cannot be interrupted by cancel/die
	 * interrupts.
1175
	 */
1176 1177
	for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
		LWLockAcquire(FirstLockMgrLock + i, LW_EXCLUSIVE);
1178

1179
	/*
1180 1181
	 * Check to see if we've been awoken by anyone in the interim.
	 *
1182 1183 1184
	 * If we have we can return and resume our transaction -- happy day.
	 * Before we are awoken the process releasing the lock grants it to us so
	 * we know that we don't have to wait anymore.
1185
	 *
1186
	 * We check by looking to see if we've been unlinked from the wait queue.
B
Bruce Momjian 已提交
1187
	 * This is quicker than checking our semaphore's state, since no kernel
1188
	 * call is needed, and it is safe because we hold the lock partition lock.
1189
	 */
1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201
	if (MyProc->links.prev == INVALID_OFFSET ||
		MyProc->links.next == INVALID_OFFSET)
		goto check_done;

#ifdef LOCK_DEBUG
	if (Debug_deadlocks)
		DumpAllLocks();
#endif

	/* Run the deadlock check, and set deadlock_state for use by ProcSleep */
	deadlock_state = DeadLockCheck(MyProc);

1202
	if (deadlock_state == DS_HARD_DEADLOCK)
B
Bruce Momjian 已提交
1203
	{
1204 1205 1206
		/*
		 * Oops.  We have a deadlock.
		 *
1207 1208 1209 1210
		 * Get this process out of wait state. (Note: we could do this more
		 * efficiently by relying on lockAwaited, but use this coding to
		 * preserve the flexibility to kill some other transaction than the
		 * one detecting the deadlock.)
1211 1212
		 *
		 * RemoveFromWaitQueue sets MyProc->waitStatus to STATUS_ERROR, so
1213 1214
		 * ProcSleep will report an error after we return from the signal
		 * handler.
1215 1216 1217
		 */
		Assert(MyProc->waitLock != NULL);
		RemoveFromWaitQueue(MyProc, LockTagHashCode(&(MyProc->waitLock->tag)));
1218

1219 1220 1221 1222 1223
		/*
		 * Unlock my semaphore so that the interrupted ProcSleep() call can
		 * finish.
		 */
		PGSemaphoreUnlock(&MyProc->sem);
1224

1225
		/*
1226 1227 1228 1229 1230 1231 1232 1233
		 * We're done here.  Transaction abort caused by the error that
		 * ProcSleep will raise will cause any other locks we hold to be
		 * released, thus allowing other processes to wake up; we don't need
		 * to do that here.  NOTE: an exception is that releasing locks we
		 * hold doesn't consider the possibility of waiters that were blocked
		 * behind us on the lock we just failed to get, and might now be
		 * wakable because we're not in front of them anymore.  However,
		 * RemoveFromWaitQueue took care of waking up any such processes.
1234 1235
		 */
	}
1236
	else if (log_lock_waits || deadlock_state == DS_BLOCKED_BY_AUTOVACUUM)
1237 1238 1239 1240
	{
		/*
		 * Unlock my semaphore so that the interrupted ProcSleep() call can
		 * print the log message (we daren't do it here because we are inside
B
Bruce Momjian 已提交
1241 1242
		 * a signal handler).  It will then sleep again until someone releases
		 * the lock.
1243 1244 1245
		 *
		 * If blocked by autovacuum, this wakeup will enable ProcSleep to send
		 * the cancelling signal to the autovacuum worker.
1246 1247 1248
		 */
		PGSemaphoreUnlock(&MyProc->sem);
	}
1249 1250

	/*
B
Bruce Momjian 已提交
1251 1252 1253 1254 1255
	 * And release locks.  We do this in reverse order for two reasons: (1)
	 * Anyone else who needs more than one of the locks will be trying to lock
	 * them in increasing order; we don't want to release the other process
	 * until it can get all the locks it needs. (2) This avoids O(N^2)
	 * behavior inside LWLockRelease.
1256
	 */
1257
check_done:
B
Bruce Momjian 已提交
1258
	for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
1259
		LWLockRelease(FirstLockMgrLock + i);
1260 1261 1262
}


1263 1264 1265 1266 1267 1268
/*
 * ProcWaitForSignal - wait for a signal from another backend.
 *
 * This can share the semaphore normally used for waiting for locks,
 * since a backend could never be waiting for a lock and a signal at
 * the same time.  As with locks, it's OK if the signal arrives just
B
Bruce Momjian 已提交
1269
 * before we actually reach the waiting state.	Also as with locks,
1270 1271
 * it's necessary that the caller be robust against bogus wakeups:
 * always check that the desired state has occurred, and wait again
B
Bruce Momjian 已提交
1272
 * if not.	This copes with possible "leftover" wakeups.
1273 1274 1275 1276
 */
void
ProcWaitForSignal(void)
{
1277
	PGSemaphoreLock(&MyProc->sem, true);
1278 1279 1280
}

/*
1281
 * ProcSendSignal - send a signal to a backend identified by PID
1282 1283
 */
void
1284
ProcSendSignal(int pid)
1285
{
1286
	PGPROC	   *proc = BackendPidGetProc(pid);
1287 1288

	if (proc != NULL)
1289
		PGSemaphoreUnlock(&proc->sem);
1290 1291 1292
}


1293 1294 1295 1296 1297 1298 1299 1300 1301
/*****************************************************************************
 * SIGALRM interrupt support
 *
 * Maybe these should be in pqsignal.c?
 *****************************************************************************/

/*
 * Enable the SIGALRM interrupt to fire after the specified delay
 *
1302
 * Delay is given in milliseconds.	Caller should be sure a SIGALRM
1303 1304
 * signal handler is installed before this is called.
 *
1305 1306
 * This code properly handles nesting of deadlock timeout alarms within
 * statement timeout alarms.
1307
 *
1308 1309 1310
 * Returns TRUE if okay, FALSE on failure.
 */
bool
1311
enable_sig_alarm(int delayms, bool is_statement_timeout)
1312
{
1313
	TimestampTz fin_time;
1314
	struct itimerval timeval;
1315

1316 1317
	if (is_statement_timeout)
	{
1318 1319 1320 1321 1322 1323 1324 1325 1326 1327
		/*
		 * Begin statement-level timeout
		 *
		 * Note that we compute statement_fin_time with reference to the
		 * statement_timestamp, but apply the specified delay without any
		 * correction; that is, we ignore whatever time has elapsed since
		 * statement_timestamp was set.  In the normal case only a small
		 * interval will have elapsed and so this doesn't matter, but there
		 * are corner cases (involving multi-statement query strings with
		 * embedded COMMIT or ROLLBACK) where we might re-initialize the
B
Bruce Momjian 已提交
1328 1329 1330 1331
		 * statement timeout long after initial receipt of the message. In
		 * such cases the enforcement of the statement timeout will be a bit
		 * inconsistent.  This annoyance is judged not worth the cost of
		 * performing an additional gettimeofday() here.
1332
		 */
1333
		Assert(!deadlock_timeout_active);
1334 1335
		fin_time = GetCurrentStatementStartTimestamp();
		fin_time = TimestampTzPlusMilliseconds(fin_time, delayms);
1336
		statement_fin_time = fin_time;
1337
		cancel_from_timeout = false;
1338
		statement_timeout_active = true;
1339 1340 1341 1342 1343 1344
	}
	else if (statement_timeout_active)
	{
		/*
		 * Begin deadlock timeout with statement-level timeout active
		 *
1345 1346 1347 1348
		 * Here, we want to interrupt at the closer of the two timeout times.
		 * If fin_time >= statement_fin_time then we need not touch the
		 * existing timer setting; else set up to interrupt at the deadlock
		 * timeout time.
1349 1350 1351
		 *
		 * NOTE: in this case it is possible that this routine will be
		 * interrupted by the previously-set timer alarm.  This is okay
B
Bruce Momjian 已提交
1352 1353 1354
		 * because the signal handler will do only what it should do according
		 * to the state variables.	The deadlock checker may get run earlier
		 * than normal, but that does no harm.
1355
		 */
1356 1357
		timeout_start_time = GetCurrentTimestamp();
		fin_time = TimestampTzPlusMilliseconds(timeout_start_time, delayms);
1358
		deadlock_timeout_active = true;
1359
		if (fin_time >= statement_fin_time)
1360 1361 1362 1363 1364 1365
			return true;
	}
	else
	{
		/* Begin deadlock timeout with no statement-level timeout */
		deadlock_timeout_active = true;
1366 1367 1368
		/* GetCurrentTimestamp can be expensive, so only do it if we must */
		if (log_lock_waits)
			timeout_start_time = GetCurrentTimestamp();
1369
	}
1370

1371
	/* If we reach here, okay to set the timer interrupt */
1372
	MemSet(&timeval, 0, sizeof(struct itimerval));
1373 1374
	timeval.it_value.tv_sec = delayms / 1000;
	timeval.it_value.tv_usec = (delayms % 1000) * 1000;
1375
	if (setitimer(ITIMER_REAL, &timeval, NULL))
1376
		return false;
1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397
	return true;
}

/*
 * Cancel the SIGALRM timer, either for a deadlock timeout or a statement
 * timeout.  If a deadlock timeout is canceled, any active statement timeout
 * remains in force.
 *
 * Returns TRUE if okay, FALSE on failure.
 */
bool
disable_sig_alarm(bool is_statement_timeout)
{
	/*
	 * Always disable the interrupt if it is active; this avoids being
	 * interrupted by the signal handler and thereby possibly getting
	 * confused.
	 *
	 * We will re-enable the interrupt if necessary in CheckStatementTimeout.
	 */
	if (statement_timeout_active || deadlock_timeout_active)
1398
	{
1399
		struct itimerval timeval;
1400

1401
		MemSet(&timeval, 0, sizeof(struct itimerval));
1402
		if (setitimer(ITIMER_REAL, &timeval, NULL))
1403
		{
1404 1405 1406
			statement_timeout_active = false;
			cancel_from_timeout = false;
			deadlock_timeout_active = false;
1407 1408
			return false;
		}
1409 1410
	}

1411 1412 1413 1414
	/* Always cancel deadlock timeout, in case this is error cleanup */
	deadlock_timeout_active = false;

	/* Cancel or reschedule statement timeout */
1415
	if (is_statement_timeout)
1416
	{
1417
		statement_timeout_active = false;
1418 1419
		cancel_from_timeout = false;
	}
1420 1421 1422 1423 1424
	else if (statement_timeout_active)
	{
		if (!CheckStatementTimeout())
			return false;
	}
1425 1426 1427
	return true;
}

1428

1429
/*
1430 1431 1432
 * Check for statement timeout.  If the timeout time has come,
 * trigger a query-cancel interrupt; if not, reschedule the SIGALRM
 * interrupt to occur at the right time.
1433
 *
1434
 * Returns true if okay, false if failed to set the interrupt.
1435
 */
1436 1437
static bool
CheckStatementTimeout(void)
1438
{
1439
	TimestampTz now;
B
Bruce Momjian 已提交
1440

1441 1442 1443
	if (!statement_timeout_active)
		return true;			/* do nothing if not active */

1444
	now = GetCurrentTimestamp();
1445

1446
	if (now >= statement_fin_time)
1447
	{
1448 1449
		/* Time to die */
		statement_timeout_active = false;
1450
		cancel_from_timeout = true;
1451 1452 1453 1454
#ifdef HAVE_SETSID
		/* try to signal whole process group */
		kill(-MyProcPid, SIGINT);
#endif
1455
		kill(MyProcPid, SIGINT);
1456 1457 1458 1459
	}
	else
	{
		/* Not time yet, so (re)schedule the interrupt */
1460 1461
		long		secs;
		int			usecs;
1462 1463
		struct itimerval timeval;

1464 1465
		TimestampDifference(now, statement_fin_time,
							&secs, &usecs);
B
Bruce Momjian 已提交
1466

1467 1468 1469 1470 1471 1472
		/*
		 * It's possible that the difference is less than a microsecond;
		 * ensure we don't cancel, rather than set, the interrupt.
		 */
		if (secs == 0 && usecs == 0)
			usecs = 1;
1473
		MemSet(&timeval, 0, sizeof(struct itimerval));
1474 1475
		timeval.it_value.tv_sec = secs;
		timeval.it_value.tv_usec = usecs;
1476
		if (setitimer(ITIMER_REAL, &timeval, NULL))
1477 1478 1479
			return false;
	}

1480 1481
	return true;
}
1482 1483 1484


/*
1485 1486 1487 1488 1489 1490
 * Signal handler for SIGALRM
 *
 * Process deadlock check and/or statement timeout check, as needed.
 * To avoid various edge cases, we must be careful to do nothing
 * when there is nothing to be done.  We also need to be able to
 * reschedule the timer interrupt if called before end of statement.
1491 1492 1493 1494
 */
void
handle_sig_alarm(SIGNAL_ARGS)
{
1495 1496 1497
	int			save_errno = errno;

	if (deadlock_timeout_active)
1498
	{
1499
		deadlock_timeout_active = false;
1500 1501
		CheckDeadLock();
	}
1502 1503 1504 1505 1506

	if (statement_timeout_active)
		(void) CheckStatementTimeout();

	errno = save_errno;
1507
}