execMain.c 56.5 KB
Newer Older
1 2
/*-------------------------------------------------------------------------
 *
3
 * execMain.c
4
 *	  top level executor interface routines
5 6
 *
 * INTERFACE ROUTINES
7 8 9
 *	ExecutorStart()
 *	ExecutorRun()
 *	ExecutorEnd()
10
 *
11 12 13 14
 *	The old ExecutorMain() has been replaced by ExecutorStart(),
 *	ExecutorRun() and ExecutorEnd()
 *
 *	These three procedures are the external interfaces to the executor.
15
 *	In each case, the query descriptor is required as an argument.
16
 *
17
 *	ExecutorStart() must be called at the beginning of execution of any
18 19 20
 *	query plan and ExecutorEnd() should always be called at the end of
 *	execution of a plan.
 *
21
 *	ExecutorRun accepts direction and count arguments that specify whether
22
 *	the plan is to be executed forwards, backwards, and for how many tuples.
23
 *
B
Bruce Momjian 已提交
24
 * Portions Copyright (c) 1996-2003, PostgreSQL Global Development Group
B
Add:  
Bruce Momjian 已提交
25
 * Portions Copyright (c) 1994, Regents of the University of California
26 27 28
 *
 *
 * IDENTIFICATION
29
 *	  $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.229 2004/03/02 18:56:15 tgl Exp $
30 31 32
 *
 *-------------------------------------------------------------------------
 */
33 34
#include "postgres.h"

35 36
#include "access/heapam.h"
#include "catalog/heap.h"
37
#include "catalog/namespace.h"
38
#include "commands/tablecmds.h"
39
#include "commands/trigger.h"
B
Bruce Momjian 已提交
40 41 42
#include "executor/execdebug.h"
#include "executor/execdefs.h"
#include "miscadmin.h"
43
#include "optimizer/clauses.h"
B
Bruce Momjian 已提交
44 45 46
#include "optimizer/var.h"
#include "parser/parsetree.h"
#include "utils/acl.h"
47
#include "utils/guc.h"
48
#include "utils/lsyscache.h"
49

50

51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
typedef struct execRowMark
{
	Relation	relation;
	Index		rti;
	char		resname[32];
} execRowMark;

typedef struct evalPlanQual
{
	Index		rti;
	EState	   *estate;
	PlanState  *planstate;
	struct evalPlanQual *next;	/* stack of active PlanQual plans */
	struct evalPlanQual *free;	/* list of free PlanQual plans */
} evalPlanQual;

67
/* decls for local routines only used within this module */
68
static void InitPlan(QueryDesc *queryDesc, bool explainOnly);
69
static void initResultRelInfo(ResultRelInfo *resultRelInfo,
B
Bruce Momjian 已提交
70 71 72
				  Index resultRelationIndex,
				  List *rangeTable,
				  CmdType operation);
73
static TupleTableSlot *ExecutePlan(EState *estate, PlanState *planstate,
B
Bruce Momjian 已提交
74 75 76
			CmdType operation,
			long numberTuples,
			ScanDirection direction,
77
			DestReceiver *dest);
78
static void ExecSelect(TupleTableSlot *slot,
79
		   DestReceiver *dest,
B
Bruce Momjian 已提交
80
		   EState *estate);
81
static void ExecInsert(TupleTableSlot *slot, ItemPointer tupleid,
B
Bruce Momjian 已提交
82
		   EState *estate);
83
static void ExecDelete(TupleTableSlot *slot, ItemPointer tupleid,
B
Bruce Momjian 已提交
84
		   EState *estate);
85
static void ExecUpdate(TupleTableSlot *slot, ItemPointer tupleid,
B
Bruce Momjian 已提交
86
		   EState *estate);
87
static TupleTableSlot *EvalPlanQualNext(EState *estate);
88
static void EndEvalPlanQual(EState *estate);
89 90
static void ExecCheckRTEPerms(RangeTblEntry *rte);
static void ExecCheckXactReadOnly(Query *parsetree);
91
static void EvalPlanQualStart(evalPlanQual *epq, EState *estate,
B
Bruce Momjian 已提交
92
				  evalPlanQual *priorepq);
93
static void EvalPlanQualStop(evalPlanQual *epq);
94

95 96
/* end of local decls */

97

98
/* ----------------------------------------------------------------
99 100 101 102 103
 *		ExecutorStart
 *
 *		This routine must be called at the beginning of any execution of any
 *		query plan
 *
104
 * Takes a QueryDesc previously created by CreateQueryDesc (it's not real
B
Bruce Momjian 已提交
105
 * clear why we bother to separate the two functions, but...).	The tupDesc
106 107
 * field of the QueryDesc is filled in to describe the tuples that will be
 * returned, and the internal fields (estate and planstate) are set up.
108
 *
109 110 111 112 113 114 115 116
 * If useCurrentSnapshot is true, run the query with the latest available
 * snapshot, instead of the normal QuerySnapshot.  Also, if it's an update
 * or delete query, check that the rows to be updated or deleted would be
 * visible to the normal QuerySnapshot.  (This is a special-case behavior
 * needed for referential integrity updates in serializable transactions.
 * We must check all currently-committed rows, but we want to throw a
 * can't-serialize error if any rows that would need updates would not be
 * visible under the normal serializable snapshot.)
117
 *
118 119 120
 * If explainOnly is true, we are not actually intending to run the plan,
 * only to set up for EXPLAIN; so skip unwanted side-effects.
 *
121 122
 * NB: the CurrentMemoryContext when this is called will become the parent
 * of the per-query context used for this Executor invocation.
123 124
 * ----------------------------------------------------------------
 */
125
void
126
ExecutorStart(QueryDesc *queryDesc, bool useCurrentSnapshot, bool explainOnly)
127
{
128
	EState	   *estate;
129
	MemoryContext oldcontext;
130

131
	/* sanity checks: queryDesc must not be started already */
132
	Assert(queryDesc != NULL);
133 134
	Assert(queryDesc->estate == NULL);

135
	/*
B
Bruce Momjian 已提交
136 137
	 * If the transaction is read-only, we need to check if any writes are
	 * planned to non-temporary tables.
138
	 */
139 140
	if (XactReadOnly && !explainOnly)
		ExecCheckXactReadOnly(queryDesc->parsetree);
141

142
	/*
143
	 * Build EState, switch into per-query memory context for startup.
144 145 146 147
	 */
	estate = CreateExecutorState();
	queryDesc->estate = estate;

148 149 150 151 152
	oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);

	/*
	 * Fill in parameters, if any, from queryDesc
	 */
153
	estate->es_param_list_info = queryDesc->params;
154

V
Vadim B. Mikheev 已提交
155
	if (queryDesc->plantree->nParamExec > 0)
156
		estate->es_param_exec_vals = (ParamExecData *)
157
			palloc0(queryDesc->plantree->nParamExec * sizeof(ParamExecData));
158

159 160
	estate->es_instrument = queryDesc->doInstrument;

161
	/*
162 163
	 * Make our own private copy of the current query snapshot data.
	 *
B
Bruce Momjian 已提交
164 165 166
	 * This "freezes" our idea of which tuples are good and which are not for
	 * the life of this query, even if it outlives the current command and
	 * current snapshot.
167
	 */
168
	if (useCurrentSnapshot)
169
	{
170 171 172 173
		/* RI update/delete query --- must use an up-to-date snapshot */
		estate->es_snapshot = CopyCurrentSnapshot();
		/* crosscheck updates/deletes against transaction snapshot */
		estate->es_crosscheck_snapshot = CopyQuerySnapshot();
174 175 176
	}
	else
	{
177
		/* normal query --- use query snapshot, no crosscheck */
178
		estate->es_snapshot = CopyQuerySnapshot();
179
		estate->es_crosscheck_snapshot = SnapshotAny;
180
	}
181

182
	/*
183
	 * Initialize the plan state tree
184
	 */
185
	InitPlan(queryDesc, explainOnly);
186 187

	MemoryContextSwitchTo(oldcontext);
188 189 190
}

/* ----------------------------------------------------------------
191 192 193 194 195 196 197
 *		ExecutorRun
 *
 *		This is the main routine of the executor module. It accepts
 *		the query descriptor from the traffic cop and executes the
 *		query plan.
 *
 *		ExecutorStart must have been called already.
198
 *
199 200 201
 *		If direction is NoMovementScanDirection then nothing is done
 *		except to start up/shut down the destination.  Otherwise,
 *		we retrieve up to 'count' tuples in the specified direction.
202
 *
203
 *		Note: count = 0 is interpreted as no portal limit, i.e., run to
204
 *		completion.
205
 *
206 207
 * ----------------------------------------------------------------
 */
208
TupleTableSlot *
209
ExecutorRun(QueryDesc *queryDesc,
210
			ScanDirection direction, long count)
211
{
212
	EState	   *estate;
213
	CmdType		operation;
214
	DestReceiver *dest;
215
	TupleTableSlot *result;
216 217 218 219 220 221 222 223
	MemoryContext oldcontext;

	/* sanity checks */
	Assert(queryDesc != NULL);

	estate = queryDesc->estate;

	Assert(estate != NULL);
224

B
Bruce Momjian 已提交
225
	/*
226
	 * Switch into per-query memory context
227
	 */
228
	oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
229

B
Bruce Momjian 已提交
230
	/*
B
Bruce Momjian 已提交
231 232
	 * extract information from the query descriptor and the query
	 * feature.
233
	 */
234 235 236
	operation = queryDesc->operation;
	dest = queryDesc->dest;

B
Bruce Momjian 已提交
237
	/*
238
	 * startup tuple receiver
239
	 */
240 241
	estate->es_processed = 0;
	estate->es_lastoid = InvalidOid;
242

243
	(*dest->rStartup) (dest, operation, queryDesc->tupDesc);
244

245 246 247 248 249 250 251
	/*
	 * run plan
	 */
	if (direction == NoMovementScanDirection)
		result = NULL;
	else
		result = ExecutePlan(estate,
252
							 queryDesc->planstate,
253 254 255
							 operation,
							 count,
							 direction,
256
							 dest);
257

258 259 260
	/*
	 * shutdown receiver
	 */
261
	(*dest->rShutdown) (dest);
262

263 264
	MemoryContextSwitchTo(oldcontext);

265
	return result;
266 267 268
}

/* ----------------------------------------------------------------
269 270
 *		ExecutorEnd
 *
271
 *		This routine must be called at the end of execution of any
272
 *		query plan
273 274 275
 * ----------------------------------------------------------------
 */
void
276
ExecutorEnd(QueryDesc *queryDesc)
277
{
278
	EState	   *estate;
279
	MemoryContext oldcontext;
280

281 282
	/* sanity checks */
	Assert(queryDesc != NULL);
283

284 285
	estate = queryDesc->estate;

286
	Assert(estate != NULL);
287

288
	/*
289
	 * Switch into per-query memory context to run ExecEndPlan
290
	 */
291 292 293
	oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);

	ExecEndPlan(queryDesc->planstate, estate);
294

295
	/*
296
	 * Must switch out of context before destroying it
297
	 */
298
	MemoryContextSwitchTo(oldcontext);
299

300
	/*
301 302
	 * Release EState and per-query memory context.  This should release
	 * everything the executor has allocated.
303
	 */
304 305 306 307 308 309
	FreeExecutorState(estate);

	/* Reset queryDesc fields that no longer point to anything */
	queryDesc->tupDesc = NULL;
	queryDesc->estate = NULL;
	queryDesc->planstate = NULL;
310
}
311

312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
/* ----------------------------------------------------------------
 *		ExecutorRewind
 *
 *		This routine may be called on an open queryDesc to rewind it
 *		to the start.
 * ----------------------------------------------------------------
 */
void
ExecutorRewind(QueryDesc *queryDesc)
{
	EState	   *estate;
	MemoryContext oldcontext;

	/* sanity checks */
	Assert(queryDesc != NULL);

	estate = queryDesc->estate;

	Assert(estate != NULL);

	/* It's probably not sensible to rescan updating queries */
	Assert(queryDesc->operation == CMD_SELECT);

	/*
	 * Switch into per-query memory context
	 */
	oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);

	/*
	 * rescan plan
	 */
	ExecReScan(queryDesc->planstate, NULL);

	MemoryContextSwitchTo(oldcontext);
}

348

349 350 351 352
/*
 * ExecCheckRTPerms
 *		Check access permissions for all relations listed in a range table.
 */
353
void
354
ExecCheckRTPerms(List *rangeTable)
355 356 357 358
{
	List	   *lp;

	foreach(lp, rangeTable)
359
	{
360 361
		RangeTblEntry *rte = lfirst(lp);

362
		ExecCheckRTEPerms(rte);
363 364 365 366 367 368 369 370
	}
}

/*
 * ExecCheckRTEPerms
 *		Check access permissions for a single RTE.
 */
static void
371
ExecCheckRTEPerms(RangeTblEntry *rte)
372
{
373
	AclMode		requiredPerms;
374
	Oid			relOid;
375
	AclId		userid;
376

B
Bruce Momjian 已提交
377
	/*
378 379 380 381
	 * If it's a subquery, recursively examine its rangetable.
	 */
	if (rte->rtekind == RTE_SUBQUERY)
	{
382
		ExecCheckRTPerms(rte->subquery->rtable);
383 384 385 386 387
		return;
	}

	/*
	 * Otherwise, only plain-relation RTEs need to be checked here.
B
Bruce Momjian 已提交
388 389
	 * Function RTEs are checked by init_fcache when the function is
	 * prepared for execution. Join and special RTEs need no checks.
B
Bruce Momjian 已提交
390
	 */
391
	if (rte->rtekind != RTE_RELATION)
392 393
		return;

394 395 396 397 398 399 400
	/*
	 * No work if requiredPerms is empty.
	 */
	requiredPerms = rte->requiredPerms;
	if (requiredPerms == 0)
		return;

401
	relOid = rte->relid;
402 403

	/*
B
Bruce Momjian 已提交
404 405
	 * userid to check as: current user unless we have a setuid
	 * indication.
406
	 *
B
Bruce Momjian 已提交
407 408
	 * Note: GetUserId() is presently fast enough that there's no harm in
	 * calling it separately for each RTE.	If that stops being true, we
409
	 * could call it once in ExecCheckRTPerms and pass the userid down
B
Bruce Momjian 已提交
410
	 * from there.	But for now, no need for the extra clutter.
411
	 */
412
	userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
413

414 415 416 417 418 419 420 421 422
	/*
	 * For each bit in requiredPerms, apply the required check.  (We can't
	 * do this in one aclcheck call because aclcheck treats multiple bits
	 * as OR semantics, when we want AND.)
	 *
	 * We use a well-known cute trick for isolating the rightmost one-bit
	 * in a nonzero word.  See nodes/bitmapset.c for commentary.
	 */
#define RIGHTMOST_ONE(x) ((int32) (x) & -((int32) (x)))
423

424
	while (requiredPerms != 0)
425
	{
426 427
		AclMode		thisPerm;
		AclResult	aclcheck_result;
428

429 430 431 432
		thisPerm = RIGHTMOST_ONE(requiredPerms);
		requiredPerms &= ~thisPerm;

		aclcheck_result = pg_class_aclcheck(relOid, userid, thisPerm);
433
		if (aclcheck_result != ACLCHECK_OK)
434 435
			aclcheck_error(aclcheck_result, ACL_KIND_CLASS,
						   get_rel_name(relOid));
436
	}
437 438
}

439 440 441
/*
 * Check that the query does not imply any writes to non-temp tables.
 */
442
static void
443
ExecCheckXactReadOnly(Query *parsetree)
444
{
445
	List	   *lp;
446

447 448 449 450 451 452
	/*
	 * CREATE TABLE AS or SELECT INTO?
	 *
	 * XXX should we allow this if the destination is temp?
	 */
	if (parsetree->into != NULL)
453 454
		goto fail;

455 456
	/* Fail if write permissions are requested on any non-temp table */
	foreach(lp, parsetree->rtable)
457
	{
458
		RangeTblEntry *rte = lfirst(lp);
459

460
		if (rte->rtekind == RTE_SUBQUERY)
461
		{
462 463 464
			ExecCheckXactReadOnly(rte->subquery);
			continue;
		}
465

466 467
		if (rte->rtekind != RTE_RELATION)
			continue;
468

469 470
		if ((rte->requiredPerms & (~ACL_SELECT)) == 0)
			continue;
471

472 473
		if (isTempNamespace(get_rel_namespace(rte->relid)))
			continue;
474

475
		goto fail;
476 477 478 479 480
	}

	return;

fail:
481 482 483
	ereport(ERROR,
			(errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
			 errmsg("transaction is read-only")));
484 485 486
}


487
/* ----------------------------------------------------------------
488 489 490 491
 *		InitPlan
 *
 *		Initializes the query plan: open files, allocate storage
 *		and start up the rule manager
492 493
 * ----------------------------------------------------------------
 */
494
static void
495
InitPlan(QueryDesc *queryDesc, bool explainOnly)
496
{
497
	CmdType		operation = queryDesc->operation;
B
Bruce Momjian 已提交
498 499 500
	Query	   *parseTree = queryDesc->parsetree;
	Plan	   *plan = queryDesc->plantree;
	EState	   *estate = queryDesc->estate;
501
	PlanState  *planstate;
B
Bruce Momjian 已提交
502 503
	List	   *rangeTable;
	Relation	intoRelationDesc;
504
	bool		do_select_into;
B
Bruce Momjian 已提交
505
	TupleDesc	tupType;
506

507
	/*
B
Bruce Momjian 已提交
508 509
	 * Do permissions checks.  It's sufficient to examine the query's top
	 * rangetable here --- subplan RTEs will be checked during
510
	 * ExecInitSubPlan().
511
	 */
512
	ExecCheckRTPerms(parseTree->rtable);
513

B
Bruce Momjian 已提交
514
	/*
B
Bruce Momjian 已提交
515
	 * get information from query descriptor
516
	 */
517
	rangeTable = parseTree->rtable;
518

B
Bruce Momjian 已提交
519
	/*
B
Bruce Momjian 已提交
520
	 * initialize the node's execution state
521
	 */
522 523
	estate->es_range_table = rangeTable;

B
Bruce Momjian 已提交
524
	/*
525
	 * if there is a result relation, initialize result relation stuff
526
	 */
527
	if (parseTree->resultRelation != 0 && operation != CMD_SELECT)
528
	{
529 530 531
		List	   *resultRelations = parseTree->resultRelations;
		int			numResultRelations;
		ResultRelInfo *resultRelInfos;
B
Bruce Momjian 已提交
532

533 534 535 536 537 538 539
		if (resultRelations != NIL)
		{
			/*
			 * Multiple result relations (due to inheritance)
			 * parseTree->resultRelations identifies them all
			 */
			ResultRelInfo *resultRelInfo;
540

541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557
			numResultRelations = length(resultRelations);
			resultRelInfos = (ResultRelInfo *)
				palloc(numResultRelations * sizeof(ResultRelInfo));
			resultRelInfo = resultRelInfos;
			while (resultRelations != NIL)
			{
				initResultRelInfo(resultRelInfo,
								  lfirsti(resultRelations),
								  rangeTable,
								  operation);
				resultRelInfo++;
				resultRelations = lnext(resultRelations);
			}
		}
		else
		{
			/*
B
Bruce Momjian 已提交
558 559
			 * Single result relation identified by
			 * parseTree->resultRelation
560 561 562 563 564 565 566 567
			 */
			numResultRelations = 1;
			resultRelInfos = (ResultRelInfo *) palloc(sizeof(ResultRelInfo));
			initResultRelInfo(resultRelInfos,
							  parseTree->resultRelation,
							  rangeTable,
							  operation);
		}
568

569 570 571 572
		estate->es_result_relations = resultRelInfos;
		estate->es_num_result_relations = numResultRelations;
		/* Initialize to first or only result rel */
		estate->es_result_relation_info = resultRelInfos;
573
	}
574 575
	else
	{
B
Bruce Momjian 已提交
576
		/*
B
Bruce Momjian 已提交
577
		 * if no result relation, then set state appropriately
578
		 */
579 580
		estate->es_result_relations = NULL;
		estate->es_num_result_relations = 0;
581 582 583
		estate->es_result_relation_info = NULL;
	}

584 585 586 587 588 589 590
	/*
	 * Detect whether we're doing SELECT INTO.  If so, set the force_oids
	 * flag appropriately so that the plan tree will be initialized with
	 * the correct tuple descriptors.
	 */
	do_select_into = false;

591
	if (operation == CMD_SELECT && parseTree->into != NULL)
592 593
	{
		do_select_into = true;
594 595
		estate->es_select_into = true;
		estate->es_into_oids = parseTree->intoHasOids;
596 597
	}

598 599 600
	/*
	 * Have to lock relations selected for update
	 */
601 602
	estate->es_rowMark = NIL;
	if (parseTree->rowMarks != NIL)
603
	{
B
Bruce Momjian 已提交
604
		List	   *l;
605

606
		foreach(l, parseTree->rowMarks)
607
		{
608 609
			Index		rti = lfirsti(l);
			Oid			relid = getrelid(rti, rangeTable);
610 611 612 613
			Relation	relation;
			execRowMark *erm;

			relation = heap_open(relid, RowShareLock);
B
Bruce Momjian 已提交
614
			erm = (execRowMark *) palloc(sizeof(execRowMark));
615
			erm->relation = relation;
616
			erm->rti = rti;
617
			snprintf(erm->resname, sizeof(erm->resname), "ctid%u", rti);
618 619 620
			estate->es_rowMark = lappend(estate->es_rowMark, erm);
		}
	}
621

B
Bruce Momjian 已提交
622
	/*
623
	 * initialize the executor "tuple" table.  We need slots for all the
624 625 626
	 * plan nodes, plus possibly output slots for the junkfilter(s). At
	 * this point we aren't sure if we need junkfilters, so just add slots
	 * for them unconditionally.
627 628
	 */
	{
629
		int			nSlots = ExecCountSlotsNode(plan);
630

631 632 633 634 635
		if (parseTree->resultRelations != NIL)
			nSlots += length(parseTree->resultRelations);
		else
			nSlots += 1;
		estate->es_tupleTable = ExecCreateTupleTable(nSlots);
636
	}
637

638
	/* mark EvalPlanQual not active */
639
	estate->es_topPlan = plan;
640 641
	estate->es_evalPlanQual = NULL;
	estate->es_evTupleNull = NULL;
642
	estate->es_evTuple = NULL;
643 644
	estate->es_useEvalPlan = false;

B
Bruce Momjian 已提交
645
	/*
B
Bruce Momjian 已提交
646 647
	 * initialize the private state information for all the nodes in the
	 * query tree.	This opens files, allocates storage and leaves us
648
	 * ready to start processing tuples.
649
	 */
650
	planstate = ExecInitNode(plan, estate);
651

B
Bruce Momjian 已提交
652
	/*
653
	 * Get the tuple descriptor describing the type of tuples to return.
B
Bruce Momjian 已提交
654
	 * (this is especially important if we are creating a relation with
655
	 * "SELECT INTO")
656
	 */
657
	tupType = ExecGetResultType(planstate);
658

B
Bruce Momjian 已提交
659
	/*
B
Bruce Momjian 已提交
660 661
	 * Initialize the junk filter if needed.  SELECT and INSERT queries
	 * need a filter if there are any junk attrs in the tlist.	INSERT and
662 663 664 665
	 * SELECT INTO also need a filter if the plan may return raw disk tuples
	 * (else heap_insert will be scribbling on the source relation!).
	 * UPDATE and DELETE always need a filter, since there's always a junk
	 * 'ctid' attribute present --- no need to look first.
666 667
	 */
	{
668 669 670
		bool		junk_filter_needed = false;
		List	   *tlist;

671
		switch (operation)
672
		{
673 674
			case CMD_SELECT:
			case CMD_INSERT:
675
				foreach(tlist, plan->targetlist)
676
				{
677 678 679 680 681 682 683
					TargetEntry *tle = (TargetEntry *) lfirst(tlist);

					if (tle->resdom->resjunk)
					{
						junk_filter_needed = true;
						break;
					}
684
				}
685
				if (!junk_filter_needed &&
686 687 688
					(operation == CMD_INSERT || do_select_into) &&
					ExecMayReturnRawTuples(planstate))
					junk_filter_needed = true;
689 690 691 692 693 694 695
				break;
			case CMD_UPDATE:
			case CMD_DELETE:
				junk_filter_needed = true;
				break;
			default:
				break;
696 697
		}

698
		if (junk_filter_needed)
699
		{
700
			/*
B
Bruce Momjian 已提交
701 702 703 704
			 * If there are multiple result relations, each one needs its
			 * own junk filter.  Note this is only possible for
			 * UPDATE/DELETE, so we can't be fooled by some needing a
			 * filter and some not.
705 706 707
			 */
			if (parseTree->resultRelations != NIL)
			{
708 709
				PlanState **appendplans;
				int			as_nplans;
710
				ResultRelInfo *resultRelInfo;
711
				int			i;
712 713 714 715

				/* Top plan had better be an Append here. */
				Assert(IsA(plan, Append));
				Assert(((Append *) plan)->isTarget);
716 717 718 719
				Assert(IsA(planstate, AppendState));
				appendplans = ((AppendState *) planstate)->appendplans;
				as_nplans = ((AppendState *) planstate)->as_nplans;
				Assert(as_nplans == estate->es_num_result_relations);
720
				resultRelInfo = estate->es_result_relations;
721
				for (i = 0; i < as_nplans; i++)
722
				{
723
					PlanState  *subplan = appendplans[i];
724 725
					JunkFilter *j;

726
					j = ExecInitJunkFilter(subplan->plan->targetlist,
727
										   ExecGetResultType(subplan),
728
							  ExecAllocTableSlot(estate->es_tupleTable));
729 730 731
					resultRelInfo->ri_junkFilter = j;
					resultRelInfo++;
				}
B
Bruce Momjian 已提交
732

733 734 735 736 737 738 739 740 741 742
				/*
				 * Set active junkfilter too; at this point ExecInitAppend
				 * has already selected an active result relation...
				 */
				estate->es_junkFilter =
					estate->es_result_relation_info->ri_junkFilter;
			}
			else
			{
				/* Normal case with just one JunkFilter */
743
				JunkFilter *j;
744

745
				j = ExecInitJunkFilter(planstate->plan->targetlist,
746
									   tupType,
747
							  ExecAllocTableSlot(estate->es_tupleTable));
748 749 750
				estate->es_junkFilter = j;
				if (estate->es_result_relation_info)
					estate->es_result_relation_info->ri_junkFilter = j;
751

752 753 754 755
				/* For SELECT, want to return the cleaned tuple type */
				if (operation == CMD_SELECT)
					tupType = j->jf_cleanTupType;
			}
756 757 758 759
		}
		else
			estate->es_junkFilter = NULL;
	}
760

B
Bruce Momjian 已提交
761
	/*
762
	 * If doing SELECT INTO, initialize the "into" relation.  We must wait
B
Bruce Momjian 已提交
763 764
	 * till now so we have the "clean" result tuple type to create the new
	 * table from.
765 766
	 *
	 * If EXPLAIN, skip creating the "into" relation.
767
	 */
768
	intoRelationDesc = NULL;
769

770
	if (do_select_into && !explainOnly)
771
	{
772 773 774 775 776
		char	   *intoName;
		Oid			namespaceId;
		AclResult	aclresult;
		Oid			intoRelationId;
		TupleDesc	tupdesc;
777

778 779 780 781 782
		/*
		 * find namespace to create in, check permissions
		 */
		intoName = parseTree->into->relname;
		namespaceId = RangeVarGetCreationNamespace(parseTree->into);
783

784 785 786
		aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
										  ACL_CREATE);
		if (aclresult != ACLCHECK_OK)
787 788
			aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
						   get_namespace_name(namespaceId));
789

790 791 792 793
		/*
		 * have to copy tupType to get rid of constraints
		 */
		tupdesc = CreateTupleDescCopy(tupType);
794

795 796 797 798 799 800 801
		intoRelationId = heap_create_with_catalog(intoName,
												  namespaceId,
												  tupdesc,
												  RELKIND_RELATION,
												  false,
												  ONCOMMIT_NOOP,
												  allowSystemTableMods);
802

803
		FreeTupleDesc(tupdesc);
804

805
		/*
B
Bruce Momjian 已提交
806 807
		 * Advance command counter so that the newly-created relation's
		 * catalog tuples will be visible to heap_open.
808 809
		 */
		CommandCounterIncrement();
810

811
		/*
B
Bruce Momjian 已提交
812 813 814 815
		 * If necessary, create a TOAST table for the into relation. Note
		 * that AlterTableCreateToastTable ends with
		 * CommandCounterIncrement(), so that the TOAST table will be
		 * visible for insertion.
816 817
		 */
		AlterTableCreateToastTable(intoRelationId, true);
818

819 820 821 822
		/*
		 * And open the constructed table for writing.
		 */
		intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
823 824 825 826
	}

	estate->es_into_relation_descriptor = intoRelationDesc;

827 828
	queryDesc->tupDesc = tupType;
	queryDesc->planstate = planstate;
829 830
}

831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848
/*
 * Initialize ResultRelInfo data for one result relation
 */
static void
initResultRelInfo(ResultRelInfo *resultRelInfo,
				  Index resultRelationIndex,
				  List *rangeTable,
				  CmdType operation)
{
	Oid			resultRelationOid;
	Relation	resultRelationDesc;

	resultRelationOid = getrelid(resultRelationIndex, rangeTable);
	resultRelationDesc = heap_open(resultRelationOid, RowExclusiveLock);

	switch (resultRelationDesc->rd_rel->relkind)
	{
		case RELKIND_SEQUENCE:
849 850
			ereport(ERROR,
					(errcode(ERRCODE_WRONG_OBJECT_TYPE),
851
					 errmsg("cannot change sequence \"%s\"",
B
Bruce Momjian 已提交
852
						  RelationGetRelationName(resultRelationDesc))));
853 854
			break;
		case RELKIND_TOASTVALUE:
855 856
			ereport(ERROR,
					(errcode(ERRCODE_WRONG_OBJECT_TYPE),
857
					 errmsg("cannot change TOAST relation \"%s\"",
B
Bruce Momjian 已提交
858
						  RelationGetRelationName(resultRelationDesc))));
859 860
			break;
		case RELKIND_VIEW:
861 862
			ereport(ERROR,
					(errcode(ERRCODE_WRONG_OBJECT_TYPE),
863
					 errmsg("cannot change view \"%s\"",
B
Bruce Momjian 已提交
864
						  RelationGetRelationName(resultRelationDesc))));
865 866 867 868 869 870 871 872 873 874
			break;
	}

	MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
	resultRelInfo->type = T_ResultRelInfo;
	resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
	resultRelInfo->ri_RelationDesc = resultRelationDesc;
	resultRelInfo->ri_NumIndices = 0;
	resultRelInfo->ri_IndexRelationDescs = NULL;
	resultRelInfo->ri_IndexRelationInfo = NULL;
875 876
	/* make a copy so as not to depend on relcache info not changing... */
	resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
877
	resultRelInfo->ri_TrigFunctions = NULL;
878 879 880 881 882 883
	resultRelInfo->ri_ConstraintExprs = NULL;
	resultRelInfo->ri_junkFilter = NULL;

	/*
	 * If there are indices on the result relation, open them and save
	 * descriptors in the result relation info, so that we can add new
B
Bruce Momjian 已提交
884 885
	 * index entries for the tuples we add/update.	We need not do this
	 * for a DELETE, however, since deletion doesn't affect indexes.
886 887 888 889 890 891
	 */
	if (resultRelationDesc->rd_rel->relhasindex &&
		operation != CMD_DELETE)
		ExecOpenIndices(resultRelInfo);
}

892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948
/*
 *		ExecContextForcesOids
 *
 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
 * we need to ensure that result tuples have space for an OID iff they are
 * going to be stored into a relation that has OIDs.  In other contexts
 * we are free to choose whether to leave space for OIDs in result tuples
 * (we generally don't want to, but we do if a physical-tlist optimization
 * is possible).  This routine checks the plan context and returns TRUE if the
 * choice is forced, FALSE if the choice is not forced.  In the TRUE case,
 * *hasoids is set to the required value.
 *
 * One reason this is ugly is that all plan nodes in the plan tree will emit
 * tuples with space for an OID, though we really only need the topmost node
 * to do so.  However, node types like Sort don't project new tuples but just
 * return their inputs, and in those cases the requirement propagates down
 * to the input node.  Eventually we might make this code smart enough to
 * recognize how far down the requirement really goes, but for now we just
 * make all plan nodes do the same thing if the top level forces the choice.
 *
 * We assume that estate->es_result_relation_info is already set up to
 * describe the target relation.  Note that in an UPDATE that spans an
 * inheritance tree, some of the target relations may have OIDs and some not.
 * We have to make the decisions on a per-relation basis as we initialize
 * each of the child plans of the topmost Append plan.
 *
 * SELECT INTO is even uglier, because we don't have the INTO relation's
 * descriptor available when this code runs; we have to look aside at a
 * flag set by InitPlan().
 */
bool
ExecContextForcesOids(PlanState *planstate, bool *hasoids)
{
	if (planstate->state->es_select_into)
	{
		*hasoids = planstate->state->es_into_oids;
		return true;
	}
	else
	{
		ResultRelInfo *ri = planstate->state->es_result_relation_info;

		if (ri != NULL)
		{
			Relation	rel = ri->ri_RelationDesc;

			if (rel != NULL)
			{
				*hasoids = rel->rd_rel->relhasoids;
				return true;
			}
		}
	}

	return false;
}

949
/* ----------------------------------------------------------------
950
 *		ExecEndPlan
951
 *
952
 *		Cleans up the query plan -- closes files and frees up storage
953 954 955 956 957 958
 *
 * NOTE: we are no longer very worried about freeing storage per se
 * in this code; FreeExecutorState should be guaranteed to release all
 * memory that needs to be released.  What we are worried about doing
 * is closing relations and dropping buffer pins.  Thus, for example,
 * tuple tables must be cleared or dropped to ensure pins are released.
959 960
 * ----------------------------------------------------------------
 */
961
void
962
ExecEndPlan(PlanState *planstate, EState *estate)
963
{
964 965
	ResultRelInfo *resultRelInfo;
	int			i;
966
	List	   *l;
967

968 969 970 971 972 973
	/*
	 * shut down any PlanQual processing we were doing
	 */
	if (estate->es_evalPlanQual != NULL)
		EndEvalPlanQual(estate);

B
Bruce Momjian 已提交
974
	/*
975
	 * shut down the node-type-specific query processing
976
	 */
977
	ExecEndNode(planstate);
978

B
Bruce Momjian 已提交
979
	/*
B
Bruce Momjian 已提交
980
	 * destroy the executor "tuple" table.
981
	 */
982 983
	ExecDropTupleTable(estate->es_tupleTable, true);
	estate->es_tupleTable = NULL;
984

B
Bruce Momjian 已提交
985
	/*
B
Bruce Momjian 已提交
986
	 * close the result relation(s) if any, but hold locks until xact
987
	 * commit.
988
	 */
989 990
	resultRelInfo = estate->es_result_relations;
	for (i = estate->es_num_result_relations; i > 0; i--)
991
	{
992 993 994 995
		/* Close indices and then the relation itself */
		ExecCloseIndices(resultRelInfo);
		heap_close(resultRelInfo->ri_RelationDesc, NoLock);
		resultRelInfo++;
996 997
	}

B
Bruce Momjian 已提交
998
	/*
999
	 * close the "into" relation if necessary, again keeping lock
1000
	 */
1001 1002
	if (estate->es_into_relation_descriptor != NULL)
		heap_close(estate->es_into_relation_descriptor, NoLock);
1003 1004 1005 1006 1007 1008 1009 1010 1011 1012

	/*
	 * close any relations selected FOR UPDATE, again keeping locks
	 */
	foreach(l, estate->es_rowMark)
	{
		execRowMark *erm = lfirst(l);

		heap_close(erm->relation, NoLock);
	}
1013 1014 1015
}

/* ----------------------------------------------------------------
1016 1017
 *		ExecutePlan
 *
1018
 *		processes the query plan to retrieve 'numberTuples' tuples in the
1019
 *		direction specified.
1020
 *
1021
 *		Retrieves all tuples if numberTuples is 0
1022
 *
1023
 *		result is either a slot containing the last tuple in the case
1024
 *		of a SELECT or NULL otherwise.
1025
 *
1026 1027
 * Note: the ctid attribute is a 'junk' attribute that is removed before the
 * user can see it
1028 1029 1030
 * ----------------------------------------------------------------
 */
static TupleTableSlot *
1031
ExecutePlan(EState *estate,
1032
			PlanState *planstate,
1033
			CmdType operation,
1034
			long numberTuples,
1035
			ScanDirection direction,
1036
			DestReceiver *dest)
1037
{
B
Bruce Momjian 已提交
1038 1039 1040 1041 1042 1043
	JunkFilter *junkfilter;
	TupleTableSlot *slot;
	ItemPointer tupleid = NULL;
	ItemPointerData tuple_ctid;
	long		current_tuple_count;
	TupleTableSlot *result;
1044

B
Bruce Momjian 已提交
1045
	/*
B
Bruce Momjian 已提交
1046
	 * initialize local variables
1047
	 */
1048 1049 1050 1051
	slot = NULL;
	current_tuple_count = 0;
	result = NULL;

B
Bruce Momjian 已提交
1052 1053
	/*
	 * Set the direction.
1054
	 */
1055 1056
	estate->es_direction = direction;

1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
	/*
	 * Process BEFORE EACH STATEMENT triggers
	 */
	switch (operation)
	{
		case CMD_UPDATE:
			ExecBSUpdateTriggers(estate, estate->es_result_relation_info);
			break;
		case CMD_DELETE:
			ExecBSDeleteTriggers(estate, estate->es_result_relation_info);
			break;
		case CMD_INSERT:
			ExecBSInsertTriggers(estate, estate->es_result_relation_info);
			break;
		default:
			/* do nothing */
1073
			break;
1074 1075
	}

B
Bruce Momjian 已提交
1076
	/*
B
Bruce Momjian 已提交
1077
	 * Loop until we've processed the proper number of tuples from the
1078
	 * plan.
1079 1080 1081 1082
	 */

	for (;;)
	{
1083 1084
		/* Reset the per-output-tuple exprcontext */
		ResetPerTupleExprContext(estate);
B
Bruce Momjian 已提交
1085

B
Bruce Momjian 已提交
1086
		/*
B
Bruce Momjian 已提交
1087
		 * Execute the plan and obtain a tuple
1088
		 */
B
Bruce Momjian 已提交
1089
lnext:	;
1090 1091 1092 1093
		if (estate->es_useEvalPlan)
		{
			slot = EvalPlanQualNext(estate);
			if (TupIsNull(slot))
1094
				slot = ExecProcNode(planstate);
1095 1096
		}
		else
1097
			slot = ExecProcNode(planstate);
1098

B
Bruce Momjian 已提交
1099
		/*
B
Bruce Momjian 已提交
1100 1101
		 * if the tuple is null, then we assume there is nothing more to
		 * process so we just return null...
1102 1103 1104 1105 1106
		 */
		if (TupIsNull(slot))
		{
			result = NULL;
			break;
1107 1108
		}

B
Bruce Momjian 已提交
1109
		/*
B
Bruce Momjian 已提交
1110 1111
		 * if we have a junk filter, then project a new tuple with the
		 * junk removed.
1112
		 *
1113 1114 1115
		 * Store this new "clean" tuple in the junkfilter's resultSlot.
		 * (Formerly, we stored it back over the "dirty" tuple, which is
		 * WRONG because that tuple slot has the wrong descriptor.)
1116
		 *
B
Bruce Momjian 已提交
1117
		 * Also, extract all the junk information we need.
1118
		 */
1119
		if ((junkfilter = estate->es_junkFilter) != NULL)
1120
		{
1121 1122 1123
			Datum		datum;
			HeapTuple	newTuple;
			bool		isNull;
1124

B
Bruce Momjian 已提交
1125
			/*
1126 1127 1128 1129 1130 1131 1132 1133 1134
			 * extract the 'ctid' junk attribute.
			 */
			if (operation == CMD_UPDATE || operation == CMD_DELETE)
			{
				if (!ExecGetJunkAttribute(junkfilter,
										  slot,
										  "ctid",
										  &datum,
										  &isNull))
1135
					elog(ERROR, "could not find junk ctid column");
1136

1137
				/* shouldn't ever get a null result... */
1138
				if (isNull)
1139
					elog(ERROR, "ctid is NULL");
1140 1141 1142 1143 1144 1145

				tupleid = (ItemPointer) DatumGetPointer(datum);
				tuple_ctid = *tupleid;	/* make sure we don't free the
										 * ctid!! */
				tupleid = &tuple_ctid;
			}
1146
			else if (estate->es_rowMark != NIL)
1147
			{
B
Bruce Momjian 已提交
1148
				List	   *l;
1149

B
Bruce Momjian 已提交
1150 1151
		lmark:	;
				foreach(l, estate->es_rowMark)
1152
				{
1153 1154 1155 1156 1157 1158
					execRowMark *erm = lfirst(l);
					Buffer		buffer;
					HeapTupleData tuple;
					TupleTableSlot *newSlot;
					int			test;

1159 1160 1161 1162 1163
					if (!ExecGetJunkAttribute(junkfilter,
											  slot,
											  erm->resname,
											  &datum,
											  &isNull))
1164
						elog(ERROR, "could not find junk \"%s\" column",
1165
							 erm->resname);
1166

1167
					/* shouldn't ever get a null result... */
1168
					if (isNull)
1169
						elog(ERROR, "\"%s\" is NULL", erm->resname);
1170 1171

					tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1172
					test = heap_mark4update(erm->relation, &tuple, &buffer,
1173
											estate->es_snapshot->curcid);
1174 1175 1176 1177
					ReleaseBuffer(buffer);
					switch (test)
					{
						case HeapTupleSelfUpdated:
1178 1179 1180
							/* treat it as deleted; do not process */
							goto lnext;

1181 1182 1183 1184
						case HeapTupleMayBeUpdated:
							break;

						case HeapTupleUpdated:
1185
							if (IsXactIsoLevelSerializable)
1186 1187
								ereport(ERROR,
										(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1188
										 errmsg("could not serialize access due to concurrent update")));
1189
							if (!(ItemPointerEquals(&(tuple.t_self),
B
Bruce Momjian 已提交
1190
								  (ItemPointer) DatumGetPointer(datum))))
1191
							{
B
Bruce Momjian 已提交
1192
								newSlot = EvalPlanQual(estate, erm->rti, &(tuple.t_self));
1193 1194 1195 1196 1197 1198 1199
								if (!(TupIsNull(newSlot)))
								{
									slot = newSlot;
									estate->es_useEvalPlan = true;
									goto lmark;
								}
							}
B
Bruce Momjian 已提交
1200 1201 1202

							/*
							 * if tuple was deleted or PlanQual failed for
1203
							 * updated tuple - we must not return this
B
Bruce Momjian 已提交
1204
							 * tuple!
1205 1206
							 */
							goto lnext;
1207 1208

						default:
1209 1210
							elog(ERROR, "unrecognized heap_mark4update status: %u",
								 test);
B
Bruce Momjian 已提交
1211
							return (NULL);
1212 1213 1214
					}
				}
			}
1215

B
Bruce Momjian 已提交
1216
			/*
1217 1218 1219 1220 1221 1222
			 * Finally create a new "clean" tuple with all junk attributes
			 * removed
			 */
			newTuple = ExecRemoveJunk(junkfilter, slot);

			slot = ExecStoreTuple(newTuple,		/* tuple to store */
1223
								  junkfilter->jf_resultSlot,	/* dest slot */
B
Bruce Momjian 已提交
1224 1225
								  InvalidBuffer,		/* this tuple has no
														 * buffer */
1226
								  true);		/* tuple should be pfreed */
1227
		}
1228

B
Bruce Momjian 已提交
1229
		/*
B
Bruce Momjian 已提交
1230 1231
		 * now that we have a tuple, do the appropriate thing with it..
		 * either return it to the user, add it to a relation someplace,
B
Bruce Momjian 已提交
1232
		 * delete it from a relation, or modify some of its attributes.
1233 1234 1235
		 */
		switch (operation)
		{
1236
			case CMD_SELECT:
B
Bruce Momjian 已提交
1237
				ExecSelect(slot,	/* slot containing tuple */
1238
						   dest,	/* destination's tuple-receiver obj */
1239
						   estate);
1240 1241
				result = slot;
				break;
1242

1243
			case CMD_INSERT:
1244
				ExecInsert(slot, tupleid, estate);
1245 1246
				result = NULL;
				break;
1247

1248 1249 1250 1251
			case CMD_DELETE:
				ExecDelete(slot, tupleid, estate);
				result = NULL;
				break;
1252

1253
			case CMD_UPDATE:
1254
				ExecUpdate(slot, tupleid, estate);
1255 1256
				result = NULL;
				break;
1257

1258
			default:
1259 1260
				elog(ERROR, "unrecognized operation code: %d",
					 (int) operation);
1261
				result = NULL;
1262
				break;
1263
		}
B
Bruce Momjian 已提交
1264

B
Bruce Momjian 已提交
1265
		/*
1266
		 * check our tuple count.. if we've processed the proper number
B
Bruce Momjian 已提交
1267
		 * then quit, else loop again and process more tuples.	Zero
1268
		 * numberTuples means no limit.
1269
		 */
1270
		current_tuple_count++;
1271
		if (numberTuples && numberTuples == current_tuple_count)
1272
			break;
1273
	}
1274

1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290
	/*
	 * Process AFTER EACH STATEMENT triggers
	 */
	switch (operation)
	{
		case CMD_UPDATE:
			ExecASUpdateTriggers(estate, estate->es_result_relation_info);
			break;
		case CMD_DELETE:
			ExecASDeleteTriggers(estate, estate->es_result_relation_info);
			break;
		case CMD_INSERT:
			ExecASInsertTriggers(estate, estate->es_result_relation_info);
			break;
		default:
			/* do nothing */
1291
			break;
1292 1293
	}

B
Bruce Momjian 已提交
1294
	/*
B
Bruce Momjian 已提交
1295
	 * here, result is either a slot containing a tuple in the case of a
1296
	 * SELECT or NULL otherwise.
1297
	 */
1298
	return result;
1299 1300 1301
}

/* ----------------------------------------------------------------
1302
 *		ExecSelect
1303
 *
1304
 *		SELECTs are easy.. we just pass the tuple to the appropriate
1305
 *		print function.  The only complexity is when we do a
1306
 *		"SELECT INTO", in which case we insert the tuple into
1307 1308
 *		the appropriate relation (note: this is a newly created relation
 *		so we don't need to worry about indices or locks.)
1309 1310 1311
 * ----------------------------------------------------------------
 */
static void
1312
ExecSelect(TupleTableSlot *slot,
1313
		   DestReceiver *dest,
1314
		   EState *estate)
1315
{
1316 1317
	HeapTuple	tuple;
	TupleDesc	attrtype;
1318

B
Bruce Momjian 已提交
1319
	/*
B
Bruce Momjian 已提交
1320
	 * get the heap tuple out of the tuple table slot
1321 1322 1323 1324
	 */
	tuple = slot->val;
	attrtype = slot->ttc_tupleDescriptor;

B
Bruce Momjian 已提交
1325
	/*
B
Bruce Momjian 已提交
1326
	 * insert the tuple into the "into relation"
1327 1328
	 *
	 * XXX this probably ought to be replaced by a separate destination
1329 1330 1331
	 */
	if (estate->es_into_relation_descriptor != NULL)
	{
1332
		heap_insert(estate->es_into_relation_descriptor, tuple,
1333
					estate->es_snapshot->curcid);
1334 1335 1336
		IncrAppended();
	}

B
Bruce Momjian 已提交
1337
	/*
1338
	 * send the tuple to the destination
1339
	 */
1340
	(*dest->receiveTuple) (tuple, attrtype, dest);
1341 1342
	IncrRetrieved();
	(estate->es_processed)++;
1343 1344 1345
}

/* ----------------------------------------------------------------
1346
 *		ExecInsert
1347
 *
1348
 *		INSERTs are trickier.. we have to insert the tuple into
1349 1350
 *		the base relation and insert appropriate tuples into the
 *		index relations.
1351 1352 1353
 * ----------------------------------------------------------------
 */
static void
1354
ExecInsert(TupleTableSlot *slot,
1355
		   ItemPointer tupleid,
1356
		   EState *estate)
1357
{
1358
	HeapTuple	tuple;
1359
	ResultRelInfo *resultRelInfo;
1360 1361 1362
	Relation	resultRelationDesc;
	int			numIndices;
	Oid			newId;
1363

B
Bruce Momjian 已提交
1364
	/*
B
Bruce Momjian 已提交
1365
	 * get the heap tuple out of the tuple table slot
1366 1367 1368
	 */
	tuple = slot->val;

B
Bruce Momjian 已提交
1369
	/*
1370
	 * get information on the (current) result relation
1371
	 */
1372 1373
	resultRelInfo = estate->es_result_relation_info;
	resultRelationDesc = resultRelInfo->ri_RelationDesc;
1374 1375

	/* BEFORE ROW INSERT Triggers */
1376
	if (resultRelInfo->ri_TrigDesc &&
B
Bruce Momjian 已提交
1377
	  resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
1378
	{
1379
		HeapTuple	newtuple;
1380

1381
		newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
1382 1383 1384 1385 1386 1387

		if (newtuple == NULL)	/* "do nothing" */
			return;

		if (newtuple != tuple)	/* modified by Trigger(s) */
		{
1388 1389 1390
			/*
			 * Insert modified tuple into tuple table slot, replacing the
			 * original.  We assume that it was allocated in per-tuple
B
Bruce Momjian 已提交
1391 1392
			 * memory context, and therefore will go away by itself. The
			 * tuple table slot should not try to clear it.
1393 1394 1395
			 */
			ExecStoreTuple(newtuple, slot, InvalidBuffer, false);
			tuple = newtuple;
1396 1397 1398
		}
	}

B
Bruce Momjian 已提交
1399
	/*
1400
	 * Check the constraints of the tuple
1401 1402
	 */
	if (resultRelationDesc->rd_att->constr)
1403
		ExecConstraints(resultRelInfo, slot, estate);
1404

B
Bruce Momjian 已提交
1405
	/*
B
Bruce Momjian 已提交
1406
	 * insert the tuple
1407
	 */
1408
	newId = heap_insert(resultRelationDesc, tuple,
1409
						estate->es_snapshot->curcid);
1410

1411
	IncrAppended();
1412 1413
	(estate->es_processed)++;
	estate->es_lastoid = newId;
T
Tom Lane 已提交
1414
	setLastTid(&(tuple->t_self));
1415

B
Bruce Momjian 已提交
1416
	/*
B
Bruce Momjian 已提交
1417
	 * process indices
1418
	 *
B
Bruce Momjian 已提交
1419 1420 1421
	 * Note: heap_insert adds a new tuple to a relation.  As a side effect,
	 * the tupleid of the new tuple is placed in the new tuple's t_ctid
	 * field.
1422
	 */
1423
	numIndices = resultRelInfo->ri_NumIndices;
1424
	if (numIndices > 0)
1425
		ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1426 1427

	/* AFTER ROW INSERT Triggers */
1428
	ExecARInsertTriggers(estate, resultRelInfo, tuple);
1429 1430 1431
}

/* ----------------------------------------------------------------
1432
 *		ExecDelete
1433
 *
1434
 *		DELETE is like UPDATE, we delete the tuple and its
1435
 *		index tuples.
1436 1437 1438
 * ----------------------------------------------------------------
 */
static void
1439
ExecDelete(TupleTableSlot *slot,
1440
		   ItemPointer tupleid,
1441
		   EState *estate)
1442
{
1443
	ResultRelInfo *resultRelInfo;
B
Bruce Momjian 已提交
1444 1445 1446
	Relation	resultRelationDesc;
	ItemPointerData ctid;
	int			result;
1447

B
Bruce Momjian 已提交
1448
	/*
1449
	 * get information on the (current) result relation
1450
	 */
1451 1452
	resultRelInfo = estate->es_result_relation_info;
	resultRelationDesc = resultRelInfo->ri_RelationDesc;
1453 1454

	/* BEFORE ROW DELETE Triggers */
1455
	if (resultRelInfo->ri_TrigDesc &&
1456
	  resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
1457
	{
1458
		bool		dodelete;
1459

1460
		dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
1461
										estate->es_snapshot->curcid);
1462 1463 1464 1465 1466

		if (!dodelete)			/* "do nothing" */
			return;
	}

V
Vadim B. Mikheev 已提交
1467
	/*
B
Bruce Momjian 已提交
1468
	 * delete the tuple
1469
	 */
1470
ldelete:;
1471 1472
	result = heap_delete(resultRelationDesc, tupleid,
						 &ctid,
1473 1474
						 estate->es_snapshot->curcid,
						 estate->es_crosscheck_snapshot,
1475
						 true /* wait for commit */);
V
Vadim B. Mikheev 已提交
1476 1477 1478
	switch (result)
	{
		case HeapTupleSelfUpdated:
1479
			/* already deleted by self; nothing to do */
V
Vadim B. Mikheev 已提交
1480 1481 1482 1483 1484 1485
			return;

		case HeapTupleMayBeUpdated:
			break;

		case HeapTupleUpdated:
1486
			if (IsXactIsoLevelSerializable)
1487 1488
				ereport(ERROR,
						(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1489
						 errmsg("could not serialize access due to concurrent update")));
1490 1491
			else if (!(ItemPointerEquals(tupleid, &ctid)))
			{
B
Bruce Momjian 已提交
1492
				TupleTableSlot *epqslot = EvalPlanQual(estate,
B
Bruce Momjian 已提交
1493
							   resultRelInfo->ri_RangeTableIndex, &ctid);
1494

V
Vadim B. Mikheev 已提交
1495
				if (!TupIsNull(epqslot))
1496 1497 1498 1499 1500
				{
					*tupleid = ctid;
					goto ldelete;
				}
			}
1501
			/* tuple already deleted; nothing to do */
V
Vadim B. Mikheev 已提交
1502 1503 1504
			return;

		default:
1505
			elog(ERROR, "unrecognized heap_delete status: %u", result);
V
Vadim B. Mikheev 已提交
1506 1507
			return;
	}
1508 1509 1510 1511

	IncrDeleted();
	(estate->es_processed)++;

B
Bruce Momjian 已提交
1512
	/*
B
Bruce Momjian 已提交
1513 1514
	 * Note: Normally one would think that we have to delete index tuples
	 * associated with the heap tuple now..
1515
	 *
B
Bruce Momjian 已提交
1516 1517 1518
	 * ... but in POSTGRES, we have no need to do this because the vacuum
	 * daemon automatically opens an index scan and deletes index tuples
	 * when it finds deleted heap tuples. -cim 9/27/89
1519 1520 1521
	 */

	/* AFTER ROW DELETE Triggers */
1522
	ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
1523 1524 1525
}

/* ----------------------------------------------------------------
1526
 *		ExecUpdate
1527
 *
1528 1529 1530 1531
 *		note: we can't run UPDATE queries with transactions
 *		off because UPDATEs are actually INSERTs and our
 *		scan will mistakenly loop forever, updating the tuple
 *		it just inserted..	This should be fixed but until it
1532 1533
 *		is, we don't want to get stuck in an infinite loop
 *		which corrupts your database..
1534 1535 1536
 * ----------------------------------------------------------------
 */
static void
1537
ExecUpdate(TupleTableSlot *slot,
B
Bruce Momjian 已提交
1538 1539
		   ItemPointer tupleid,
		   EState *estate)
1540
{
B
Bruce Momjian 已提交
1541
	HeapTuple	tuple;
1542
	ResultRelInfo *resultRelInfo;
B
Bruce Momjian 已提交
1543 1544 1545 1546
	Relation	resultRelationDesc;
	ItemPointerData ctid;
	int			result;
	int			numIndices;
1547

B
Bruce Momjian 已提交
1548
	/*
B
Bruce Momjian 已提交
1549
	 * abort the operation if not running transactions
1550 1551
	 */
	if (IsBootstrapProcessingMode())
1552
		elog(ERROR, "cannot UPDATE during bootstrap");
1553

B
Bruce Momjian 已提交
1554
	/*
B
Bruce Momjian 已提交
1555
	 * get the heap tuple out of the tuple table slot
1556 1557 1558
	 */
	tuple = slot->val;

B
Bruce Momjian 已提交
1559
	/*
1560
	 * get information on the (current) result relation
1561
	 */
1562 1563
	resultRelInfo = estate->es_result_relation_info;
	resultRelationDesc = resultRelInfo->ri_RelationDesc;
1564 1565

	/* BEFORE ROW UPDATE Triggers */
1566
	if (resultRelInfo->ri_TrigDesc &&
1567
	  resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
1568
	{
1569
		HeapTuple	newtuple;
1570

1571
		newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
1572
										tupleid, tuple,
1573
										estate->es_snapshot->curcid);
1574 1575 1576 1577 1578 1579

		if (newtuple == NULL)	/* "do nothing" */
			return;

		if (newtuple != tuple)	/* modified by Trigger(s) */
		{
1580 1581 1582
			/*
			 * Insert modified tuple into tuple table slot, replacing the
			 * original.  We assume that it was allocated in per-tuple
B
Bruce Momjian 已提交
1583 1584
			 * memory context, and therefore will go away by itself. The
			 * tuple table slot should not try to clear it.
1585 1586 1587
			 */
			ExecStoreTuple(newtuple, slot, InvalidBuffer, false);
			tuple = newtuple;
1588 1589 1590
		}
	}

B
Bruce Momjian 已提交
1591
	/*
1592
	 * Check the constraints of the tuple
1593
	 *
1594 1595 1596 1597 1598
	 * If we generate a new candidate tuple after EvalPlanQual testing, we
	 * must loop back here and recheck constraints.  (We don't need to
	 * redo triggers, however.	If there are any BEFORE triggers then
	 * trigger.c will have done mark4update to lock the correct tuple, so
	 * there's no need to do them again.)
1599
	 */
1600
lreplace:;
1601
	if (resultRelationDesc->rd_att->constr)
1602
		ExecConstraints(resultRelInfo, slot, estate);
1603

V
Vadim B. Mikheev 已提交
1604
	/*
B
Bruce Momjian 已提交
1605
	 * replace the heap tuple
1606
	 */
1607 1608
	result = heap_update(resultRelationDesc, tupleid, tuple,
						 &ctid,
1609 1610
						 estate->es_snapshot->curcid,
						 estate->es_crosscheck_snapshot,
1611
						 true /* wait for commit */);
V
Vadim B. Mikheev 已提交
1612 1613 1614
	switch (result)
	{
		case HeapTupleSelfUpdated:
1615
			/* already deleted by self; nothing to do */
V
Vadim B. Mikheev 已提交
1616 1617 1618 1619 1620 1621
			return;

		case HeapTupleMayBeUpdated:
			break;

		case HeapTupleUpdated:
1622
			if (IsXactIsoLevelSerializable)
1623 1624
				ereport(ERROR,
						(errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1625
						 errmsg("could not serialize access due to concurrent update")));
1626 1627
			else if (!(ItemPointerEquals(tupleid, &ctid)))
			{
B
Bruce Momjian 已提交
1628
				TupleTableSlot *epqslot = EvalPlanQual(estate,
B
Bruce Momjian 已提交
1629
							   resultRelInfo->ri_RangeTableIndex, &ctid);
1630

V
Vadim B. Mikheev 已提交
1631
				if (!TupIsNull(epqslot))
1632 1633
				{
					*tupleid = ctid;
V
Vadim B. Mikheev 已提交
1634
					tuple = ExecRemoveJunk(estate->es_junkFilter, epqslot);
1635
					slot = ExecStoreTuple(tuple,
1636
									estate->es_junkFilter->jf_resultSlot,
1637
										  InvalidBuffer, true);
1638 1639 1640
					goto lreplace;
				}
			}
1641
			/* tuple already deleted; nothing to do */
V
Vadim B. Mikheev 已提交
1642 1643 1644
			return;

		default:
1645
			elog(ERROR, "unrecognized heap_update status: %u", result);
V
Vadim B. Mikheev 已提交
1646
			return;
1647 1648 1649 1650 1651
	}

	IncrReplaced();
	(estate->es_processed)++;

B
Bruce Momjian 已提交
1652
	/*
B
Bruce Momjian 已提交
1653
	 * Note: instead of having to update the old index tuples associated
B
Bruce Momjian 已提交
1654
	 * with the heap tuple, all we do is form and insert new index tuples.
1655
	 * This is because UPDATEs are actually DELETEs and INSERTs and index
B
Bruce Momjian 已提交
1656 1657
	 * tuple deletion is done automagically by the vacuum daemon. All we
	 * do is insert new index tuples.  -cim 9/27/89
1658 1659
	 */

B
Bruce Momjian 已提交
1660
	/*
B
Bruce Momjian 已提交
1661
	 * process indices
1662
	 *
1663
	 * heap_update updates a tuple in the base relation by invalidating it
1664
	 * and then inserting a new tuple to the relation.	As a side effect,
B
Bruce Momjian 已提交
1665 1666 1667
	 * the tupleid of the new tuple is placed in the new tuple's t_ctid
	 * field.  So we now insert index tuples using the new tupleid stored
	 * there.
1668 1669
	 */

1670
	numIndices = resultRelInfo->ri_NumIndices;
1671
	if (numIndices > 0)
1672
		ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1673 1674

	/* AFTER ROW UPDATE Triggers */
1675
	ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
1676
}
V
Vadim B. Mikheev 已提交
1677

1678
static const char *
1679 1680
ExecRelCheck(ResultRelInfo *resultRelInfo,
			 TupleTableSlot *slot, EState *estate)
V
Vadim B. Mikheev 已提交
1681
{
1682
	Relation	rel = resultRelInfo->ri_RelationDesc;
1683 1684
	int			ncheck = rel->rd_att->constr->num_check;
	ConstrCheck *check = rel->rd_att->constr->check;
1685
	ExprContext *econtext;
1686
	MemoryContext oldContext;
1687 1688
	List	   *qual;
	int			i;
1689

1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701
	/*
	 * If first time through for this result relation, build expression
	 * nodetrees for rel's constraint expressions.  Keep them in the
	 * per-query memory context so they'll survive throughout the query.
	 */
	if (resultRelInfo->ri_ConstraintExprs == NULL)
	{
		oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
		resultRelInfo->ri_ConstraintExprs =
			(List **) palloc(ncheck * sizeof(List *));
		for (i = 0; i < ncheck; i++)
		{
1702 1703
			/* ExecQual wants implicit-AND form */
			qual = make_ands_implicit(stringToNode(check[i].ccbin));
1704
			resultRelInfo->ri_ConstraintExprs[i] = (List *)
1705
				ExecPrepareExpr((Expr *) qual, estate);
1706 1707 1708 1709
		}
		MemoryContextSwitchTo(oldContext);
	}

1710
	/*
B
Bruce Momjian 已提交
1711 1712
	 * We will use the EState's per-tuple context for evaluating
	 * constraint expressions (creating it if it's not already there).
1713
	 */
1714
	econtext = GetPerTupleExprContext(estate);
1715

1716 1717 1718 1719
	/* Arrange for econtext's scan tuple to be the tuple under test */
	econtext->ecxt_scantuple = slot;

	/* And evaluate the constraints */
1720 1721
	for (i = 0; i < ncheck; i++)
	{
1722
		qual = resultRelInfo->ri_ConstraintExprs[i];
1723

1724 1725
		/*
		 * NOTE: SQL92 specifies that a NULL result from a constraint
1726 1727
		 * expression is not to be treated as a failure.  Therefore, tell
		 * ExecQual to return TRUE for NULL.
1728
		 */
1729
		if (!ExecQual(qual, econtext, true))
1730
			return check[i].ccname;
1731 1732
	}

1733
	/* NULL result means no error */
1734
	return NULL;
V
Vadim B. Mikheev 已提交
1735 1736
}

1737
void
1738
ExecConstraints(ResultRelInfo *resultRelInfo,
1739
				TupleTableSlot *slot, EState *estate)
V
Vadim B. Mikheev 已提交
1740
{
1741
	Relation	rel = resultRelInfo->ri_RelationDesc;
1742 1743 1744 1745
	HeapTuple	tuple = slot->val;
	TupleConstr *constr = rel->rd_att->constr;

	Assert(constr);
1746

1747
	if (constr->has_not_null)
V
Vadim B. Mikheev 已提交
1748
	{
1749
		int			natts = rel->rd_att->natts;
1750
		int			attrChk;
1751

1752
		for (attrChk = 1; attrChk <= natts; attrChk++)
1753
		{
B
Bruce Momjian 已提交
1754
			if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1755
				heap_attisnull(tuple, attrChk))
1756 1757
				ereport(ERROR,
						(errcode(ERRCODE_NOT_NULL_VIOLATION),
1758
						 errmsg("null value in column \"%s\" violates not-null constraint",
B
Bruce Momjian 已提交
1759
					NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1760 1761 1762
		}
	}

1763
	if (constr->num_check > 0)
1764
	{
B
Bruce Momjian 已提交
1765
		const char *failed;
1766

1767
		if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1768 1769
			ereport(ERROR,
					(errcode(ERRCODE_CHECK_VIOLATION),
1770
					 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1771
							RelationGetRelationName(rel), failed)));
1772
	}
V
Vadim B. Mikheev 已提交
1773
}
1774

1775 1776 1777 1778 1779 1780
/*
 * Check a modified tuple to see if we want to process its updated version
 * under READ COMMITTED rules.
 *
 * See backend/executor/README for some info about how this works.
 */
B
Bruce Momjian 已提交
1781
TupleTableSlot *
1782 1783
EvalPlanQual(EState *estate, Index rti, ItemPointer tid)
{
1784 1785
	evalPlanQual *epq;
	EState	   *epqstate;
B
Bruce Momjian 已提交
1786 1787
	Relation	relation;
	HeapTupleData tuple;
1788 1789
	HeapTuple	copyTuple = NULL;
	bool		endNode;
1790 1791 1792

	Assert(rti != 0);

1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812
	/*
	 * find relation containing target tuple
	 */
	if (estate->es_result_relation_info != NULL &&
		estate->es_result_relation_info->ri_RangeTableIndex == rti)
		relation = estate->es_result_relation_info->ri_RelationDesc;
	else
	{
		List	   *l;

		relation = NULL;
		foreach(l, estate->es_rowMark)
		{
			if (((execRowMark *) lfirst(l))->rti == rti)
			{
				relation = ((execRowMark *) lfirst(l))->relation;
				break;
			}
		}
		if (relation == NULL)
1813
			elog(ERROR, "could not find RowMark for RT index %u", rti);
1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825
	}

	/*
	 * fetch tid tuple
	 *
	 * Loop here to deal with updated or busy tuples
	 */
	tuple.t_self = *tid;
	for (;;)
	{
		Buffer		buffer;

1826
		if (heap_fetch(relation, SnapshotDirty, &tuple, &buffer, false, NULL))
1827 1828 1829
		{
			TransactionId xwait = SnapshotDirty->xmax;

1830
			/* xmin should not be dirty... */
1831
			if (TransactionIdIsValid(SnapshotDirty->xmin))
1832
				elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878

			/*
			 * If tuple is being updated by other transaction then we have
			 * to wait for its commit/abort.
			 */
			if (TransactionIdIsValid(xwait))
			{
				ReleaseBuffer(buffer);
				XactLockTableWait(xwait);
				continue;
			}

			/*
			 * We got tuple - now copy it for use by recheck query.
			 */
			copyTuple = heap_copytuple(&tuple);
			ReleaseBuffer(buffer);
			break;
		}

		/*
		 * Oops! Invalid tuple. Have to check is it updated or deleted.
		 * Note that it's possible to get invalid SnapshotDirty->tid if
		 * tuple updated by this transaction. Have we to check this ?
		 */
		if (ItemPointerIsValid(&(SnapshotDirty->tid)) &&
			!(ItemPointerEquals(&(tuple.t_self), &(SnapshotDirty->tid))))
		{
			/* updated, so look at the updated copy */
			tuple.t_self = SnapshotDirty->tid;
			continue;
		}

		/*
		 * Deleted or updated by this transaction; forget it.
		 */
		return NULL;
	}

	/*
	 * For UPDATE/DELETE we have to return tid of actual row we're
	 * executing PQ for.
	 */
	*tid = tuple.t_self;

	/*
1879
	 * Need to run a recheck subquery.	Find or create a PQ stack entry.
1880
	 */
1881
	epq = estate->es_evalPlanQual;
1882 1883
	endNode = true;

1884 1885
	if (epq != NULL && epq->rti == 0)
	{
1886
		/* Top PQ stack entry is idle, so re-use it */
1887
		Assert(!(estate->es_useEvalPlan) && epq->next == NULL);
1888 1889 1890 1891 1892 1893
		epq->rti = rti;
		endNode = false;
	}

	/*
	 * If this is request for another RTE - Ra, - then we have to check
B
Bruce Momjian 已提交
1894 1895 1896
	 * wasn't PlanQual requested for Ra already and if so then Ra' row was
	 * updated again and we have to re-start old execution for Ra and
	 * forget all what we done after Ra was suspended. Cool? -:))
1897
	 */
B
Bruce Momjian 已提交
1898
	if (epq != NULL && epq->rti != rti &&
1899
		epq->estate->es_evTuple[rti - 1] != NULL)
1900 1901 1902
	{
		do
		{
1903 1904
			evalPlanQual *oldepq;

1905
			/* stop execution */
1906 1907 1908 1909
			EvalPlanQualStop(epq);
			/* pop previous PlanQual from the stack */
			oldepq = epq->next;
			Assert(oldepq && oldepq->rti != 0);
1910 1911 1912
			/* push current PQ to freePQ stack */
			oldepq->free = epq;
			epq = oldepq;
1913
			estate->es_evalPlanQual = epq;
1914 1915 1916
		} while (epq->rti != rti);
	}

B
Bruce Momjian 已提交
1917
	/*
1918 1919 1920 1921 1922 1923
	 * If we are requested for another RTE then we have to suspend
	 * execution of current PlanQual and start execution for new one.
	 */
	if (epq == NULL || epq->rti != rti)
	{
		/* try to reuse plan used previously */
B
Bruce Momjian 已提交
1924
		evalPlanQual *newepq = (epq != NULL) ? epq->free : NULL;
1925

1926
		if (newepq == NULL)		/* first call or freePQ stack is empty */
1927
		{
1928
			newepq = (evalPlanQual *) palloc0(sizeof(evalPlanQual));
1929
			newepq->free = NULL;
1930 1931
			newepq->estate = NULL;
			newepq->planstate = NULL;
1932 1933
		}
		else
1934
		{
1935 1936 1937
			/* recycle previously used PlanQual */
			Assert(newepq->estate == NULL);
			epq->free = NULL;
1938
		}
1939
		/* push current PQ to the stack */
1940
		newepq->next = epq;
1941
		epq = newepq;
1942
		estate->es_evalPlanQual = epq;
1943 1944 1945 1946
		epq->rti = rti;
		endNode = false;
	}

1947
	Assert(epq->rti == rti);
1948 1949

	/*
1950 1951
	 * Ok - we're requested for the same RTE.  Unfortunately we still have
	 * to end and restart execution of the plan, because ExecReScan
1952
	 * wouldn't ensure that upper plan nodes would reset themselves.  We
1953 1954 1955
	 * could make that work if insertion of the target tuple were
	 * integrated with the Param mechanism somehow, so that the upper plan
	 * nodes know that their children's outputs have changed.
1956
	 *
B
Bruce Momjian 已提交
1957 1958
	 * Note that the stack of free evalPlanQual nodes is quite useless at the
	 * moment, since it only saves us from pallocing/releasing the
1959
	 * evalPlanQual nodes themselves.  But it will be useful once we
B
Bruce Momjian 已提交
1960 1961
	 * implement ReScan instead of end/restart for re-using PlanQual
	 * nodes.
1962 1963
	 */
	if (endNode)
1964
	{
1965
		/* stop execution */
1966
		EvalPlanQualStop(epq);
1967
	}
1968

1969 1970 1971
	/*
	 * Initialize new recheck query.
	 *
B
Bruce Momjian 已提交
1972 1973 1974 1975 1976
	 * Note: if we were re-using PlanQual plans via ExecReScan, we'd need to
	 * instead copy down changeable state from the top plan (including
	 * es_result_relation_info, es_junkFilter) and reset locally
	 * changeable state in the epq (including es_param_exec_vals,
	 * es_evTupleNull).
1977 1978 1979
	 */
	EvalPlanQualStart(epq, estate, epq->next);

1980
	/*
1981 1982
	 * free old RTE' tuple, if any, and store target tuple where
	 * relation's scan node will see it
1983
	 */
1984
	epqstate = epq->estate;
1985 1986 1987
	if (epqstate->es_evTuple[rti - 1] != NULL)
		heap_freetuple(epqstate->es_evTuple[rti - 1]);
	epqstate->es_evTuple[rti - 1] = copyTuple;
1988

1989
	return EvalPlanQualNext(estate);
1990 1991
}

B
Bruce Momjian 已提交
1992
static TupleTableSlot *
1993 1994
EvalPlanQualNext(EState *estate)
{
1995 1996
	evalPlanQual *epq = estate->es_evalPlanQual;
	MemoryContext oldcontext;
B
Bruce Momjian 已提交
1997
	TupleTableSlot *slot;
1998 1999 2000 2001

	Assert(epq->rti != 0);

lpqnext:;
2002
	oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt);
2003
	slot = ExecProcNode(epq->planstate);
2004
	MemoryContextSwitchTo(oldcontext);
2005 2006 2007 2008 2009 2010

	/*
	 * No more tuples for this PQ. Continue previous one.
	 */
	if (TupIsNull(slot))
	{
2011 2012
		evalPlanQual *oldepq;

2013
		/* stop execution */
2014
		EvalPlanQualStop(epq);
2015
		/* pop old PQ from the stack */
2016 2017
		oldepq = epq->next;
		if (oldepq == NULL)
2018
		{
2019 2020 2021 2022 2023
			/* this is the first (oldest) PQ - mark as free */
			epq->rti = 0;
			estate->es_useEvalPlan = false;
			/* and continue Query execution */
			return (NULL);
2024 2025 2026 2027 2028
		}
		Assert(oldepq->rti != 0);
		/* push current PQ to freePQ stack */
		oldepq->free = epq;
		epq = oldepq;
2029
		estate->es_evalPlanQual = epq;
2030 2031 2032 2033 2034
		goto lpqnext;
	}

	return (slot);
}
2035 2036 2037 2038

static void
EndEvalPlanQual(EState *estate)
{
2039
	evalPlanQual *epq = estate->es_evalPlanQual;
2040

2041 2042
	if (epq->rti == 0)			/* plans already shutdowned */
	{
2043
		Assert(epq->next == NULL);
2044
		return;
2045
	}
2046 2047 2048

	for (;;)
	{
2049 2050
		evalPlanQual *oldepq;

2051
		/* stop execution */
2052
		EvalPlanQualStop(epq);
2053
		/* pop old PQ from the stack */
2054 2055
		oldepq = epq->next;
		if (oldepq == NULL)
2056
		{
2057 2058 2059
			/* this is the first (oldest) PQ - mark as free */
			epq->rti = 0;
			estate->es_useEvalPlan = false;
2060 2061 2062 2063 2064 2065
			break;
		}
		Assert(oldepq->rti != 0);
		/* push current PQ to freePQ stack */
		oldepq->free = epq;
		epq = oldepq;
2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090
		estate->es_evalPlanQual = epq;
	}
}

/*
 * Start execution of one level of PlanQual.
 *
 * This is a cut-down version of ExecutorStart(): we copy some state from
 * the top-level estate rather than initializing it fresh.
 */
static void
EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
{
	EState	   *epqstate;
	int			rtsize;
	MemoryContext oldcontext;

	rtsize = length(estate->es_range_table);

	epq->estate = epqstate = CreateExecutorState();

	oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);

	/*
	 * The epqstates share the top query's copy of unchanging state such
B
Bruce Momjian 已提交
2091 2092 2093
	 * as the snapshot, rangetable, result-rel info, and external Param
	 * info. They need their own copies of local state, including a tuple
	 * table, es_param_exec_vals, etc.
2094 2095 2096
	 */
	epqstate->es_direction = ForwardScanDirection;
	epqstate->es_snapshot = estate->es_snapshot;
2097
	epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot;
2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109
	epqstate->es_range_table = estate->es_range_table;
	epqstate->es_result_relations = estate->es_result_relations;
	epqstate->es_num_result_relations = estate->es_num_result_relations;
	epqstate->es_result_relation_info = estate->es_result_relation_info;
	epqstate->es_junkFilter = estate->es_junkFilter;
	epqstate->es_into_relation_descriptor = estate->es_into_relation_descriptor;
	epqstate->es_param_list_info = estate->es_param_list_info;
	if (estate->es_topPlan->nParamExec > 0)
		epqstate->es_param_exec_vals = (ParamExecData *)
			palloc0(estate->es_topPlan->nParamExec * sizeof(ParamExecData));
	epqstate->es_rowMark = estate->es_rowMark;
	epqstate->es_instrument = estate->es_instrument;
2110 2111
	epqstate->es_select_into = estate->es_select_into;
	epqstate->es_into_oids = estate->es_into_oids;
2112
	epqstate->es_topPlan = estate->es_topPlan;
B
Bruce Momjian 已提交
2113

2114
	/*
B
Bruce Momjian 已提交
2115 2116 2117
	 * Each epqstate must have its own es_evTupleNull state, but all the
	 * stack entries share es_evTuple state.  This allows sub-rechecks to
	 * inherit the value being examined by an outer recheck.
2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159
	 */
	epqstate->es_evTupleNull = (bool *) palloc0(rtsize * sizeof(bool));
	if (priorepq == NULL)
		/* first PQ stack entry */
		epqstate->es_evTuple = (HeapTuple *)
			palloc0(rtsize * sizeof(HeapTuple));
	else
		/* later stack entries share the same storage */
		epqstate->es_evTuple = priorepq->estate->es_evTuple;

	epqstate->es_tupleTable =
		ExecCreateTupleTable(estate->es_tupleTable->size);

	epq->planstate = ExecInitNode(estate->es_topPlan, epqstate);

	MemoryContextSwitchTo(oldcontext);
}

/*
 * End execution of one level of PlanQual.
 *
 * This is a cut-down version of ExecutorEnd(); basically we want to do most
 * of the normal cleanup, but *not* close result relations (which we are
 * just sharing from the outer query).
 */
static void
EvalPlanQualStop(evalPlanQual *epq)
{
	EState	   *epqstate = epq->estate;
	MemoryContext oldcontext;

	oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);

	ExecEndNode(epq->planstate);

	ExecDropTupleTable(epqstate->es_tupleTable, true);
	epqstate->es_tupleTable = NULL;

	if (epqstate->es_evTuple[epq->rti - 1] != NULL)
	{
		heap_freetuple(epqstate->es_evTuple[epq->rti - 1]);
		epqstate->es_evTuple[epq->rti - 1] = NULL;
2160
	}
2161 2162 2163 2164 2165 2166 2167

	MemoryContextSwitchTo(oldcontext);

	FreeExecutorState(epqstate);

	epq->estate = NULL;
	epq->planstate = NULL;
2168
}