1 /*-------------------------------------------------------------------------
4 * top level executor interface routines
11 * The old ExecutorMain() has been replaced by ExecutorStart(),
12 * ExecutorRun() and ExecutorEnd()
14 * These three procedures are the external interfaces to the executor.
15 * In each case, the query descriptor is required as an argument.
17 * ExecutorStart() must be called at the beginning of execution of any
18 * query plan and ExecutorEnd() should always be called at the end of
19 * 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.
24 * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
25 * Portions Copyright (c) 1994, Regents of the University of California
29 * $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.254 2005/08/20 00:39:55 tgl Exp $
31 *-------------------------------------------------------------------------
35 #include "access/heapam.h"
36 #include "access/xlog.h"
37 #include "catalog/heap.h"
38 #include "catalog/namespace.h"
39 #include "commands/tablecmds.h"
40 #include "commands/trigger.h"
41 #include "executor/execdebug.h"
42 #include "executor/execdefs.h"
43 #include "executor/instrument.h"
44 #include "miscadmin.h"
45 #include "optimizer/clauses.h"
46 #include "optimizer/var.h"
47 #include "parser/parsetree.h"
48 #include "storage/smgr.h"
49 #include "utils/acl.h"
50 #include "utils/guc.h"
51 #include "utils/lsyscache.h"
52 #include "utils/memutils.h"
55 typedef struct execRowMark
62 typedef struct evalPlanQual
67 struct evalPlanQual *next; /* stack of active PlanQual plans */
68 struct evalPlanQual *free; /* list of free PlanQual plans */
71 /* decls for local routines only used within this module */
72 static void InitPlan(QueryDesc *queryDesc, bool explainOnly);
73 static void initResultRelInfo(ResultRelInfo *resultRelInfo,
74 Index resultRelationIndex,
78 static TupleTableSlot *ExecutePlan(EState *estate, PlanState *planstate,
81 ScanDirection direction,
83 static void ExecSelect(TupleTableSlot *slot,
86 static void ExecInsert(TupleTableSlot *slot, ItemPointer tupleid,
88 static void ExecDelete(TupleTableSlot *slot, ItemPointer tupleid,
90 static void ExecUpdate(TupleTableSlot *slot, ItemPointer tupleid,
92 static TupleTableSlot *EvalPlanQualNext(EState *estate);
93 static void EndEvalPlanQual(EState *estate);
94 static void ExecCheckRTEPerms(RangeTblEntry *rte);
95 static void ExecCheckXactReadOnly(Query *parsetree);
96 static void EvalPlanQualStart(evalPlanQual *epq, EState *estate,
97 evalPlanQual *priorepq);
98 static void EvalPlanQualStop(evalPlanQual *epq);
100 /* end of local decls */
103 /* ----------------------------------------------------------------
106 * This routine must be called at the beginning of any execution of any
109 * Takes a QueryDesc previously created by CreateQueryDesc (it's not real
110 * clear why we bother to separate the two functions, but...). The tupDesc
111 * field of the QueryDesc is filled in to describe the tuples that will be
112 * returned, and the internal fields (estate and planstate) are set up.
114 * If explainOnly is true, we are not actually intending to run the plan,
115 * only to set up for EXPLAIN; so skip unwanted side-effects.
117 * NB: the CurrentMemoryContext when this is called will become the parent
118 * of the per-query context used for this Executor invocation.
119 * ----------------------------------------------------------------
122 ExecutorStart(QueryDesc *queryDesc, bool explainOnly)
125 MemoryContext oldcontext;
127 /* sanity checks: queryDesc must not be started already */
128 Assert(queryDesc != NULL);
129 Assert(queryDesc->estate == NULL);
132 * If the transaction is read-only, we need to check if any writes are
133 * planned to non-temporary tables.
135 if (XactReadOnly && !explainOnly)
136 ExecCheckXactReadOnly(queryDesc->parsetree);
139 * Build EState, switch into per-query memory context for startup.
141 estate = CreateExecutorState();
142 queryDesc->estate = estate;
144 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
147 * Fill in parameters, if any, from queryDesc
149 estate->es_param_list_info = queryDesc->params;
151 if (queryDesc->plantree->nParamExec > 0)
152 estate->es_param_exec_vals = (ParamExecData *)
153 palloc0(queryDesc->plantree->nParamExec * sizeof(ParamExecData));
156 * Copy other important information into the EState
158 estate->es_snapshot = queryDesc->snapshot;
159 estate->es_crosscheck_snapshot = queryDesc->crosscheck_snapshot;
160 estate->es_instrument = queryDesc->doInstrument;
163 * Initialize the plan state tree
165 InitPlan(queryDesc, explainOnly);
167 MemoryContextSwitchTo(oldcontext);
170 /* ----------------------------------------------------------------
173 * This is the main routine of the executor module. It accepts
174 * the query descriptor from the traffic cop and executes the
177 * ExecutorStart must have been called already.
179 * If direction is NoMovementScanDirection then nothing is done
180 * except to start up/shut down the destination. Otherwise,
181 * we retrieve up to 'count' tuples in the specified direction.
183 * Note: count = 0 is interpreted as no portal limit, i.e., run to
186 * ----------------------------------------------------------------
189 ExecutorRun(QueryDesc *queryDesc,
190 ScanDirection direction, long count)
195 TupleTableSlot *result;
196 MemoryContext oldcontext;
199 Assert(queryDesc != NULL);
201 estate = queryDesc->estate;
203 Assert(estate != NULL);
206 * Switch into per-query memory context
208 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
211 * extract information from the query descriptor and the query
214 operation = queryDesc->operation;
215 dest = queryDesc->dest;
218 * startup tuple receiver
220 estate->es_processed = 0;
221 estate->es_lastoid = InvalidOid;
223 (*dest->rStartup) (dest, operation, queryDesc->tupDesc);
228 if (direction == NoMovementScanDirection)
231 result = ExecutePlan(estate,
232 queryDesc->planstate,
241 (*dest->rShutdown) (dest);
243 MemoryContextSwitchTo(oldcontext);
248 /* ----------------------------------------------------------------
251 * This routine must be called at the end of execution of any
253 * ----------------------------------------------------------------
256 ExecutorEnd(QueryDesc *queryDesc)
259 MemoryContext oldcontext;
262 Assert(queryDesc != NULL);
264 estate = queryDesc->estate;
266 Assert(estate != NULL);
269 * Switch into per-query memory context to run ExecEndPlan
271 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
273 ExecEndPlan(queryDesc->planstate, estate);
276 * Must switch out of context before destroying it
278 MemoryContextSwitchTo(oldcontext);
281 * Release EState and per-query memory context. This should release
282 * everything the executor has allocated.
284 FreeExecutorState(estate);
286 /* Reset queryDesc fields that no longer point to anything */
287 queryDesc->tupDesc = NULL;
288 queryDesc->estate = NULL;
289 queryDesc->planstate = NULL;
292 /* ----------------------------------------------------------------
295 * This routine may be called on an open queryDesc to rewind it
297 * ----------------------------------------------------------------
300 ExecutorRewind(QueryDesc *queryDesc)
303 MemoryContext oldcontext;
306 Assert(queryDesc != NULL);
308 estate = queryDesc->estate;
310 Assert(estate != NULL);
312 /* It's probably not sensible to rescan updating queries */
313 Assert(queryDesc->operation == CMD_SELECT);
316 * Switch into per-query memory context
318 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
323 ExecReScan(queryDesc->planstate, NULL);
325 MemoryContextSwitchTo(oldcontext);
331 * Check access permissions for all relations listed in a range table.
334 ExecCheckRTPerms(List *rangeTable)
338 foreach(l, rangeTable)
340 RangeTblEntry *rte = lfirst(l);
342 ExecCheckRTEPerms(rte);
348 * Check access permissions for a single RTE.
351 ExecCheckRTEPerms(RangeTblEntry *rte)
353 AclMode requiredPerms;
358 * Only plain-relation RTEs need to be checked here. Subquery RTEs
359 * are checked by ExecInitSubqueryScan if the subquery is still a
360 * separate subquery --- if it's been pulled up into our query level
361 * then the RTEs are in our rangetable and will be checked here.
362 * Function RTEs are checked by init_fcache when the function is
363 * prepared for execution. Join and special RTEs need no checks.
365 if (rte->rtekind != RTE_RELATION)
369 * No work if requiredPerms is empty.
371 requiredPerms = rte->requiredPerms;
372 if (requiredPerms == 0)
378 * userid to check as: current user unless we have a setuid
381 * Note: GetUserId() is presently fast enough that there's no harm in
382 * calling it separately for each RTE. If that stops being true, we
383 * could call it once in ExecCheckRTPerms and pass the userid down
384 * from there. But for now, no need for the extra clutter.
386 userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
389 * We must have *all* the requiredPerms bits, so use aclmask not
392 if (pg_class_aclmask(relOid, userid, requiredPerms, ACLMASK_ALL)
394 aclcheck_error(ACLCHECK_NO_PRIV, ACL_KIND_CLASS,
395 get_rel_name(relOid));
399 * Check that the query does not imply any writes to non-temp tables.
402 ExecCheckXactReadOnly(Query *parsetree)
407 * CREATE TABLE AS or SELECT INTO?
409 * XXX should we allow this if the destination is temp?
411 if (parsetree->into != NULL)
414 /* Fail if write permissions are requested on any non-temp table */
415 foreach(l, parsetree->rtable)
417 RangeTblEntry *rte = lfirst(l);
419 if (rte->rtekind == RTE_SUBQUERY)
421 ExecCheckXactReadOnly(rte->subquery);
425 if (rte->rtekind != RTE_RELATION)
428 if ((rte->requiredPerms & (~ACL_SELECT)) == 0)
431 if (isTempNamespace(get_rel_namespace(rte->relid)))
441 (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
442 errmsg("transaction is read-only")));
446 /* ----------------------------------------------------------------
449 * Initializes the query plan: open files, allocate storage
450 * and start up the rule manager
451 * ----------------------------------------------------------------
454 InitPlan(QueryDesc *queryDesc, bool explainOnly)
456 CmdType operation = queryDesc->operation;
457 Query *parseTree = queryDesc->parsetree;
458 Plan *plan = queryDesc->plantree;
459 EState *estate = queryDesc->estate;
460 PlanState *planstate;
462 Relation intoRelationDesc;
467 * Do permissions checks. It's sufficient to examine the query's top
468 * rangetable here --- subplan RTEs will be checked during
471 ExecCheckRTPerms(parseTree->rtable);
474 * get information from query descriptor
476 rangeTable = parseTree->rtable;
479 * initialize the node's execution state
481 estate->es_range_table = rangeTable;
484 * if there is a result relation, initialize result relation stuff
486 if (parseTree->resultRelation != 0 && operation != CMD_SELECT)
488 List *resultRelations = parseTree->resultRelations;
489 int numResultRelations;
490 ResultRelInfo *resultRelInfos;
492 if (resultRelations != NIL)
495 * Multiple result relations (due to inheritance)
496 * parseTree->resultRelations identifies them all
498 ResultRelInfo *resultRelInfo;
501 numResultRelations = list_length(resultRelations);
502 resultRelInfos = (ResultRelInfo *)
503 palloc(numResultRelations * sizeof(ResultRelInfo));
504 resultRelInfo = resultRelInfos;
505 foreach(l, resultRelations)
507 initResultRelInfo(resultRelInfo,
511 estate->es_instrument);
518 * Single result relation identified by
519 * parseTree->resultRelation
521 numResultRelations = 1;
522 resultRelInfos = (ResultRelInfo *) palloc(sizeof(ResultRelInfo));
523 initResultRelInfo(resultRelInfos,
524 parseTree->resultRelation,
527 estate->es_instrument);
530 estate->es_result_relations = resultRelInfos;
531 estate->es_num_result_relations = numResultRelations;
532 /* Initialize to first or only result rel */
533 estate->es_result_relation_info = resultRelInfos;
538 * if no result relation, then set state appropriately
540 estate->es_result_relations = NULL;
541 estate->es_num_result_relations = 0;
542 estate->es_result_relation_info = NULL;
546 * Detect whether we're doing SELECT INTO. If so, set the es_into_oids
547 * flag appropriately so that the plan tree will be initialized with
548 * the correct tuple descriptors.
550 do_select_into = false;
552 if (operation == CMD_SELECT && parseTree->into != NULL)
554 do_select_into = true;
555 estate->es_select_into = true;
556 estate->es_into_oids = parseTree->intoHasOids;
560 * Have to lock relations selected FOR UPDATE/FOR SHARE
562 estate->es_rowMarks = NIL;
563 estate->es_forUpdate = parseTree->forUpdate;
564 estate->es_rowNoWait = parseTree->rowNoWait;
565 if (parseTree->rowMarks != NIL)
569 foreach(l, parseTree->rowMarks)
571 Index rti = lfirst_int(l);
572 Oid relid = getrelid(rti, rangeTable);
576 relation = heap_open(relid, RowShareLock);
577 erm = (execRowMark *) palloc(sizeof(execRowMark));
578 erm->relation = relation;
580 snprintf(erm->resname, sizeof(erm->resname), "ctid%u", rti);
581 estate->es_rowMarks = lappend(estate->es_rowMarks, erm);
586 * initialize the executor "tuple" table. We need slots for all the
587 * plan nodes, plus possibly output slots for the junkfilter(s). At
588 * this point we aren't sure if we need junkfilters, so just add slots
589 * for them unconditionally.
592 int nSlots = ExecCountSlotsNode(plan);
594 if (parseTree->resultRelations != NIL)
595 nSlots += list_length(parseTree->resultRelations);
598 estate->es_tupleTable = ExecCreateTupleTable(nSlots);
601 /* mark EvalPlanQual not active */
602 estate->es_topPlan = plan;
603 estate->es_evalPlanQual = NULL;
604 estate->es_evTupleNull = NULL;
605 estate->es_evTuple = NULL;
606 estate->es_useEvalPlan = false;
609 * initialize the private state information for all the nodes in the
610 * query tree. This opens files, allocates storage and leaves us
611 * ready to start processing tuples.
613 planstate = ExecInitNode(plan, estate);
616 * Get the tuple descriptor describing the type of tuples to return.
617 * (this is especially important if we are creating a relation with
620 tupType = ExecGetResultType(planstate);
623 * Initialize the junk filter if needed. SELECT and INSERT queries
624 * need a filter if there are any junk attrs in the tlist. INSERT and
625 * SELECT INTO also need a filter if the plan may return raw disk
626 * tuples (else heap_insert will be scribbling on the source
627 * relation!). UPDATE and DELETE always need a filter, since there's
628 * always a junk 'ctid' attribute present --- no need to look first.
631 bool junk_filter_needed = false;
638 foreach(tlist, plan->targetlist)
640 TargetEntry *tle = (TargetEntry *) lfirst(tlist);
644 junk_filter_needed = true;
648 if (!junk_filter_needed &&
649 (operation == CMD_INSERT || do_select_into) &&
650 ExecMayReturnRawTuples(planstate))
651 junk_filter_needed = true;
655 junk_filter_needed = true;
661 if (junk_filter_needed)
664 * If there are multiple result relations, each one needs its
665 * own junk filter. Note this is only possible for
666 * UPDATE/DELETE, so we can't be fooled by some needing a
667 * filter and some not.
669 if (parseTree->resultRelations != NIL)
671 PlanState **appendplans;
673 ResultRelInfo *resultRelInfo;
676 /* Top plan had better be an Append here. */
677 Assert(IsA(plan, Append));
678 Assert(((Append *) plan)->isTarget);
679 Assert(IsA(planstate, AppendState));
680 appendplans = ((AppendState *) planstate)->appendplans;
681 as_nplans = ((AppendState *) planstate)->as_nplans;
682 Assert(as_nplans == estate->es_num_result_relations);
683 resultRelInfo = estate->es_result_relations;
684 for (i = 0; i < as_nplans; i++)
686 PlanState *subplan = appendplans[i];
689 j = ExecInitJunkFilter(subplan->plan->targetlist,
690 resultRelInfo->ri_RelationDesc->rd_att->tdhasoid,
691 ExecAllocTableSlot(estate->es_tupleTable));
692 resultRelInfo->ri_junkFilter = j;
697 * Set active junkfilter too; at this point ExecInitAppend
698 * has already selected an active result relation...
700 estate->es_junkFilter =
701 estate->es_result_relation_info->ri_junkFilter;
705 /* Normal case with just one JunkFilter */
708 j = ExecInitJunkFilter(planstate->plan->targetlist,
710 ExecAllocTableSlot(estate->es_tupleTable));
711 estate->es_junkFilter = j;
712 if (estate->es_result_relation_info)
713 estate->es_result_relation_info->ri_junkFilter = j;
715 /* For SELECT, want to return the cleaned tuple type */
716 if (operation == CMD_SELECT)
717 tupType = j->jf_cleanTupType;
721 estate->es_junkFilter = NULL;
725 * If doing SELECT INTO, initialize the "into" relation. We must wait
726 * till now so we have the "clean" result tuple type to create the new
729 * If EXPLAIN, skip creating the "into" relation.
731 intoRelationDesc = NULL;
733 if (do_select_into && !explainOnly)
742 * find namespace to create in, check permissions
744 intoName = parseTree->into->relname;
745 namespaceId = RangeVarGetCreationNamespace(parseTree->into);
747 aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
749 if (aclresult != ACLCHECK_OK)
750 aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
751 get_namespace_name(namespaceId));
754 * have to copy tupType to get rid of constraints
756 tupdesc = CreateTupleDescCopy(tupType);
758 intoRelationId = heap_create_with_catalog(intoName,
768 allowSystemTableMods);
770 FreeTupleDesc(tupdesc);
773 * Advance command counter so that the newly-created relation's
774 * catalog tuples will be visible to heap_open.
776 CommandCounterIncrement();
779 * If necessary, create a TOAST table for the into relation. Note
780 * that AlterTableCreateToastTable ends with
781 * CommandCounterIncrement(), so that the TOAST table will be
782 * visible for insertion.
784 AlterTableCreateToastTable(intoRelationId, true);
787 * And open the constructed table for writing.
789 intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
791 /* use_wal off requires rd_targblock be initially invalid */
792 Assert(intoRelationDesc->rd_targblock == InvalidBlockNumber);
795 * We can skip WAL-logging the insertions, unless PITR is in use.
797 * Note that for a non-temp INTO table, this is safe only because
798 * we know that the catalog changes above will have been WAL-logged,
799 * and so RecordTransactionCommit will think it needs to WAL-log the
800 * eventual transaction commit. Else the commit might be lost, even
801 * though all the data is safely fsync'd ...
803 estate->es_into_relation_use_wal = XLogArchivingActive();
806 estate->es_into_relation_descriptor = intoRelationDesc;
808 queryDesc->tupDesc = tupType;
809 queryDesc->planstate = planstate;
813 * Initialize ResultRelInfo data for one result relation
816 initResultRelInfo(ResultRelInfo *resultRelInfo,
817 Index resultRelationIndex,
822 Oid resultRelationOid;
823 Relation resultRelationDesc;
825 resultRelationOid = getrelid(resultRelationIndex, rangeTable);
826 resultRelationDesc = heap_open(resultRelationOid, RowExclusiveLock);
828 switch (resultRelationDesc->rd_rel->relkind)
830 case RELKIND_SEQUENCE:
832 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
833 errmsg("cannot change sequence \"%s\"",
834 RelationGetRelationName(resultRelationDesc))));
836 case RELKIND_TOASTVALUE:
838 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
839 errmsg("cannot change TOAST relation \"%s\"",
840 RelationGetRelationName(resultRelationDesc))));
844 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
845 errmsg("cannot change view \"%s\"",
846 RelationGetRelationName(resultRelationDesc))));
850 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
851 resultRelInfo->type = T_ResultRelInfo;
852 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
853 resultRelInfo->ri_RelationDesc = resultRelationDesc;
854 resultRelInfo->ri_NumIndices = 0;
855 resultRelInfo->ri_IndexRelationDescs = NULL;
856 resultRelInfo->ri_IndexRelationInfo = NULL;
857 /* make a copy so as not to depend on relcache info not changing... */
858 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
859 if (resultRelInfo->ri_TrigDesc)
861 int n = resultRelInfo->ri_TrigDesc->numtriggers;
863 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
864 palloc0(n * sizeof(FmgrInfo));
866 resultRelInfo->ri_TrigInstrument = InstrAlloc(n);
868 resultRelInfo->ri_TrigInstrument = NULL;
872 resultRelInfo->ri_TrigFunctions = NULL;
873 resultRelInfo->ri_TrigInstrument = NULL;
875 resultRelInfo->ri_ConstraintExprs = NULL;
876 resultRelInfo->ri_junkFilter = NULL;
879 * If there are indices on the result relation, open them and save
880 * descriptors in the result relation info, so that we can add new
881 * index entries for the tuples we add/update. We need not do this
882 * for a DELETE, however, since deletion doesn't affect indexes.
884 if (resultRelationDesc->rd_rel->relhasindex &&
885 operation != CMD_DELETE)
886 ExecOpenIndices(resultRelInfo);
890 * ExecContextForcesOids
892 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
893 * we need to ensure that result tuples have space for an OID iff they are
894 * going to be stored into a relation that has OIDs. In other contexts
895 * we are free to choose whether to leave space for OIDs in result tuples
896 * (we generally don't want to, but we do if a physical-tlist optimization
897 * is possible). This routine checks the plan context and returns TRUE if the
898 * choice is forced, FALSE if the choice is not forced. In the TRUE case,
899 * *hasoids is set to the required value.
901 * One reason this is ugly is that all plan nodes in the plan tree will emit
902 * tuples with space for an OID, though we really only need the topmost node
903 * to do so. However, node types like Sort don't project new tuples but just
904 * return their inputs, and in those cases the requirement propagates down
905 * to the input node. Eventually we might make this code smart enough to
906 * recognize how far down the requirement really goes, but for now we just
907 * make all plan nodes do the same thing if the top level forces the choice.
909 * We assume that estate->es_result_relation_info is already set up to
910 * describe the target relation. Note that in an UPDATE that spans an
911 * inheritance tree, some of the target relations may have OIDs and some not.
912 * We have to make the decisions on a per-relation basis as we initialize
913 * each of the child plans of the topmost Append plan.
915 * SELECT INTO is even uglier, because we don't have the INTO relation's
916 * descriptor available when this code runs; we have to look aside at a
917 * flag set by InitPlan().
920 ExecContextForcesOids(PlanState *planstate, bool *hasoids)
922 if (planstate->state->es_select_into)
924 *hasoids = planstate->state->es_into_oids;
929 ResultRelInfo *ri = planstate->state->es_result_relation_info;
933 Relation rel = ri->ri_RelationDesc;
937 *hasoids = rel->rd_rel->relhasoids;
946 /* ----------------------------------------------------------------
949 * Cleans up the query plan -- closes files and frees up storage
951 * NOTE: we are no longer very worried about freeing storage per se
952 * in this code; FreeExecutorState should be guaranteed to release all
953 * memory that needs to be released. What we are worried about doing
954 * is closing relations and dropping buffer pins. Thus, for example,
955 * tuple tables must be cleared or dropped to ensure pins are released.
956 * ----------------------------------------------------------------
959 ExecEndPlan(PlanState *planstate, EState *estate)
961 ResultRelInfo *resultRelInfo;
966 * shut down any PlanQual processing we were doing
968 if (estate->es_evalPlanQual != NULL)
969 EndEvalPlanQual(estate);
972 * shut down the node-type-specific query processing
974 ExecEndNode(planstate);
977 * destroy the executor "tuple" table.
979 ExecDropTupleTable(estate->es_tupleTable, true);
980 estate->es_tupleTable = NULL;
983 * close the result relation(s) if any, but hold locks until xact
986 resultRelInfo = estate->es_result_relations;
987 for (i = estate->es_num_result_relations; i > 0; i--)
989 /* Close indices and then the relation itself */
990 ExecCloseIndices(resultRelInfo);
991 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
996 * close the "into" relation if necessary, again keeping lock
998 if (estate->es_into_relation_descriptor != NULL)
1001 * If we skipped using WAL, and it's not a temp relation,
1002 * we must force the relation down to disk before it's
1003 * safe to commit the transaction. This requires forcing
1004 * out any dirty buffers and then doing a forced fsync.
1006 if (!estate->es_into_relation_use_wal &&
1007 !estate->es_into_relation_descriptor->rd_istemp)
1009 FlushRelationBuffers(estate->es_into_relation_descriptor);
1010 smgrimmedsync(estate->es_into_relation_descriptor->rd_smgr);
1013 heap_close(estate->es_into_relation_descriptor, NoLock);
1017 * close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
1019 foreach(l, estate->es_rowMarks)
1021 execRowMark *erm = lfirst(l);
1023 heap_close(erm->relation, NoLock);
1027 /* ----------------------------------------------------------------
1030 * processes the query plan to retrieve 'numberTuples' tuples in the
1031 * direction specified.
1033 * Retrieves all tuples if numberTuples is 0
1035 * result is either a slot containing the last tuple in the case
1036 * of a SELECT or NULL otherwise.
1038 * Note: the ctid attribute is a 'junk' attribute that is removed before the
1040 * ----------------------------------------------------------------
1042 static TupleTableSlot *
1043 ExecutePlan(EState *estate,
1044 PlanState *planstate,
1047 ScanDirection direction,
1050 JunkFilter *junkfilter;
1051 TupleTableSlot *slot;
1052 ItemPointer tupleid = NULL;
1053 ItemPointerData tuple_ctid;
1054 long current_tuple_count;
1055 TupleTableSlot *result;
1058 * initialize local variables
1061 current_tuple_count = 0;
1065 * Set the direction.
1067 estate->es_direction = direction;
1070 * Process BEFORE EACH STATEMENT triggers
1075 ExecBSUpdateTriggers(estate, estate->es_result_relation_info);
1078 ExecBSDeleteTriggers(estate, estate->es_result_relation_info);
1081 ExecBSInsertTriggers(estate, estate->es_result_relation_info);
1089 * Loop until we've processed the proper number of tuples from the
1095 /* Reset the per-output-tuple exprcontext */
1096 ResetPerTupleExprContext(estate);
1099 * Execute the plan and obtain a tuple
1102 if (estate->es_useEvalPlan)
1104 slot = EvalPlanQualNext(estate);
1105 if (TupIsNull(slot))
1106 slot = ExecProcNode(planstate);
1109 slot = ExecProcNode(planstate);
1112 * if the tuple is null, then we assume there is nothing more to
1113 * process so we just return null...
1115 if (TupIsNull(slot))
1122 * if we have a junk filter, then project a new tuple with the
1125 * Store this new "clean" tuple in the junkfilter's resultSlot.
1126 * (Formerly, we stored it back over the "dirty" tuple, which is
1127 * WRONG because that tuple slot has the wrong descriptor.)
1129 * Also, extract all the junk information we need.
1131 if ((junkfilter = estate->es_junkFilter) != NULL)
1137 * extract the 'ctid' junk attribute.
1139 if (operation == CMD_UPDATE || operation == CMD_DELETE)
1141 if (!ExecGetJunkAttribute(junkfilter,
1146 elog(ERROR, "could not find junk ctid column");
1148 /* shouldn't ever get a null result... */
1150 elog(ERROR, "ctid is NULL");
1152 tupleid = (ItemPointer) DatumGetPointer(datum);
1153 tuple_ctid = *tupleid; /* make sure we don't free the
1155 tupleid = &tuple_ctid;
1158 * Process any FOR UPDATE or FOR SHARE locking requested.
1160 else if (estate->es_rowMarks != NIL)
1165 foreach(l, estate->es_rowMarks)
1167 execRowMark *erm = lfirst(l);
1168 HeapTupleData tuple;
1170 ItemPointerData update_ctid;
1171 TransactionId update_xmax;
1172 TupleTableSlot *newSlot;
1173 LockTupleMode lockmode;
1176 if (!ExecGetJunkAttribute(junkfilter,
1181 elog(ERROR, "could not find junk \"%s\" column",
1184 /* shouldn't ever get a null result... */
1186 elog(ERROR, "\"%s\" is NULL", erm->resname);
1188 tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1190 if (estate->es_forUpdate)
1191 lockmode = LockTupleExclusive;
1193 lockmode = LockTupleShared;
1195 test = heap_lock_tuple(erm->relation, &tuple, &buffer,
1196 &update_ctid, &update_xmax,
1197 estate->es_snapshot->curcid,
1198 lockmode, estate->es_rowNoWait);
1199 ReleaseBuffer(buffer);
1202 case HeapTupleSelfUpdated:
1203 /* treat it as deleted; do not process */
1206 case HeapTupleMayBeUpdated:
1209 case HeapTupleUpdated:
1210 if (IsXactIsoLevelSerializable)
1212 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1213 errmsg("could not serialize access due to concurrent update")));
1214 if (!ItemPointerEquals(&update_ctid,
1217 /* updated, so look at updated version */
1218 newSlot = EvalPlanQual(estate,
1222 if (!TupIsNull(newSlot))
1225 estate->es_useEvalPlan = true;
1231 * if tuple was deleted or PlanQual failed for
1232 * updated tuple - we must not return this
1238 elog(ERROR, "unrecognized heap_lock_tuple status: %u",
1246 * Finally create a new "clean" tuple with all junk attributes
1249 slot = ExecFilterJunk(junkfilter, slot);
1253 * now that we have a tuple, do the appropriate thing with it..
1254 * either return it to the user, add it to a relation someplace,
1255 * delete it from a relation, or modify some of its attributes.
1260 ExecSelect(slot, /* slot containing tuple */
1261 dest, /* destination's tuple-receiver obj */
1267 ExecInsert(slot, tupleid, estate);
1272 ExecDelete(slot, tupleid, estate);
1277 ExecUpdate(slot, tupleid, estate);
1282 elog(ERROR, "unrecognized operation code: %d",
1289 * check our tuple count.. if we've processed the proper number
1290 * then quit, else loop again and process more tuples. Zero
1291 * numberTuples means no limit.
1293 current_tuple_count++;
1294 if (numberTuples && numberTuples == current_tuple_count)
1299 * Process AFTER EACH STATEMENT triggers
1304 ExecASUpdateTriggers(estate, estate->es_result_relation_info);
1307 ExecASDeleteTriggers(estate, estate->es_result_relation_info);
1310 ExecASInsertTriggers(estate, estate->es_result_relation_info);
1318 * here, result is either a slot containing a tuple in the case of a
1319 * SELECT or NULL otherwise.
1324 /* ----------------------------------------------------------------
1327 * SELECTs are easy.. we just pass the tuple to the appropriate
1328 * print function. The only complexity is when we do a
1329 * "SELECT INTO", in which case we insert the tuple into
1330 * the appropriate relation (note: this is a newly created relation
1331 * so we don't need to worry about indices or locks.)
1332 * ----------------------------------------------------------------
1335 ExecSelect(TupleTableSlot *slot,
1340 * insert the tuple into the "into relation"
1342 * XXX this probably ought to be replaced by a separate destination
1344 if (estate->es_into_relation_descriptor != NULL)
1348 tuple = ExecCopySlotTuple(slot);
1349 heap_insert(estate->es_into_relation_descriptor, tuple,
1350 estate->es_snapshot->curcid,
1351 estate->es_into_relation_use_wal,
1352 false); /* never any point in using FSM */
1353 /* we know there are no indexes to update */
1354 heap_freetuple(tuple);
1359 * send the tuple to the destination
1361 (*dest->receiveSlot) (slot, dest);
1363 (estate->es_processed)++;
1366 /* ----------------------------------------------------------------
1369 * INSERTs are trickier.. we have to insert the tuple into
1370 * the base relation and insert appropriate tuples into the
1372 * ----------------------------------------------------------------
1375 ExecInsert(TupleTableSlot *slot,
1376 ItemPointer tupleid,
1380 ResultRelInfo *resultRelInfo;
1381 Relation resultRelationDesc;
1385 * get the heap tuple out of the tuple table slot, making sure
1386 * we have a writable copy
1388 tuple = ExecMaterializeSlot(slot);
1391 * get information on the (current) result relation
1393 resultRelInfo = estate->es_result_relation_info;
1394 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1396 /* BEFORE ROW INSERT Triggers */
1397 if (resultRelInfo->ri_TrigDesc &&
1398 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
1402 newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
1404 if (newtuple == NULL) /* "do nothing" */
1407 if (newtuple != tuple) /* modified by Trigger(s) */
1410 * Insert modified tuple into tuple table slot, replacing the
1411 * original. We assume that it was allocated in per-tuple
1412 * memory context, and therefore will go away by itself. The
1413 * tuple table slot should not try to clear it.
1415 ExecStoreTuple(newtuple, slot, InvalidBuffer, false);
1421 * Check the constraints of the tuple
1423 if (resultRelationDesc->rd_att->constr)
1424 ExecConstraints(resultRelInfo, slot, estate);
1429 * Note: heap_insert returns the tid (location) of the new tuple
1430 * in the t_self field.
1432 newId = heap_insert(resultRelationDesc, tuple,
1433 estate->es_snapshot->curcid,
1437 (estate->es_processed)++;
1438 estate->es_lastoid = newId;
1439 setLastTid(&(tuple->t_self));
1442 * insert index entries for tuple
1444 if (resultRelInfo->ri_NumIndices > 0)
1445 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1447 /* AFTER ROW INSERT Triggers */
1448 ExecARInsertTriggers(estate, resultRelInfo, tuple);
1451 /* ----------------------------------------------------------------
1454 * DELETE is like UPDATE, we delete the tuple and its
1456 * ----------------------------------------------------------------
1459 ExecDelete(TupleTableSlot *slot,
1460 ItemPointer tupleid,
1463 ResultRelInfo *resultRelInfo;
1464 Relation resultRelationDesc;
1466 ItemPointerData update_ctid;
1467 TransactionId update_xmax;
1470 * get information on the (current) result relation
1472 resultRelInfo = estate->es_result_relation_info;
1473 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1475 /* BEFORE ROW DELETE Triggers */
1476 if (resultRelInfo->ri_TrigDesc &&
1477 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
1481 dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
1482 estate->es_snapshot->curcid);
1484 if (!dodelete) /* "do nothing" */
1491 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1492 * the row to be deleted is visible to that snapshot, and throw a can't-
1493 * serialize error if not. This is a special-case behavior needed for
1494 * referential integrity updates in serializable transactions.
1497 result = heap_delete(resultRelationDesc, tupleid,
1498 &update_ctid, &update_xmax,
1499 estate->es_snapshot->curcid,
1500 estate->es_crosscheck_snapshot,
1501 true /* wait for commit */ );
1504 case HeapTupleSelfUpdated:
1505 /* already deleted by self; nothing to do */
1508 case HeapTupleMayBeUpdated:
1511 case HeapTupleUpdated:
1512 if (IsXactIsoLevelSerializable)
1514 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1515 errmsg("could not serialize access due to concurrent update")));
1516 else if (!ItemPointerEquals(tupleid, &update_ctid))
1518 TupleTableSlot *epqslot;
1520 epqslot = EvalPlanQual(estate,
1521 resultRelInfo->ri_RangeTableIndex,
1524 if (!TupIsNull(epqslot))
1526 *tupleid = update_ctid;
1530 /* tuple already deleted; nothing to do */
1534 elog(ERROR, "unrecognized heap_delete status: %u", result);
1539 (estate->es_processed)++;
1542 * Note: Normally one would think that we have to delete index tuples
1543 * associated with the heap tuple now..
1545 * ... but in POSTGRES, we have no need to do this because the vacuum
1546 * daemon automatically opens an index scan and deletes index tuples
1547 * when it finds deleted heap tuples. -cim 9/27/89
1550 /* AFTER ROW DELETE Triggers */
1551 ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
1554 /* ----------------------------------------------------------------
1557 * note: we can't run UPDATE queries with transactions
1558 * off because UPDATEs are actually INSERTs and our
1559 * scan will mistakenly loop forever, updating the tuple
1560 * it just inserted.. This should be fixed but until it
1561 * is, we don't want to get stuck in an infinite loop
1562 * which corrupts your database..
1563 * ----------------------------------------------------------------
1566 ExecUpdate(TupleTableSlot *slot,
1567 ItemPointer tupleid,
1571 ResultRelInfo *resultRelInfo;
1572 Relation resultRelationDesc;
1574 ItemPointerData update_ctid;
1575 TransactionId update_xmax;
1578 * abort the operation if not running transactions
1580 if (IsBootstrapProcessingMode())
1581 elog(ERROR, "cannot UPDATE during bootstrap");
1584 * get the heap tuple out of the tuple table slot, making sure
1585 * we have a writable copy
1587 tuple = ExecMaterializeSlot(slot);
1590 * get information on the (current) result relation
1592 resultRelInfo = estate->es_result_relation_info;
1593 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1595 /* BEFORE ROW UPDATE Triggers */
1596 if (resultRelInfo->ri_TrigDesc &&
1597 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
1601 newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
1603 estate->es_snapshot->curcid);
1605 if (newtuple == NULL) /* "do nothing" */
1608 if (newtuple != tuple) /* modified by Trigger(s) */
1611 * Insert modified tuple into tuple table slot, replacing the
1612 * original. We assume that it was allocated in per-tuple
1613 * memory context, and therefore will go away by itself. The
1614 * tuple table slot should not try to clear it.
1616 ExecStoreTuple(newtuple, slot, InvalidBuffer, false);
1622 * Check the constraints of the tuple
1624 * If we generate a new candidate tuple after EvalPlanQual testing, we
1625 * must loop back here and recheck constraints. (We don't need to
1626 * redo triggers, however. If there are any BEFORE triggers then
1627 * trigger.c will have done heap_lock_tuple to lock the correct tuple,
1628 * so there's no need to do them again.)
1631 if (resultRelationDesc->rd_att->constr)
1632 ExecConstraints(resultRelInfo, slot, estate);
1635 * replace the heap tuple
1637 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1638 * the row to be updated is visible to that snapshot, and throw a can't-
1639 * serialize error if not. This is a special-case behavior needed for
1640 * referential integrity updates in serializable transactions.
1642 result = heap_update(resultRelationDesc, tupleid, tuple,
1643 &update_ctid, &update_xmax,
1644 estate->es_snapshot->curcid,
1645 estate->es_crosscheck_snapshot,
1646 true /* wait for commit */ );
1649 case HeapTupleSelfUpdated:
1650 /* already deleted by self; nothing to do */
1653 case HeapTupleMayBeUpdated:
1656 case HeapTupleUpdated:
1657 if (IsXactIsoLevelSerializable)
1659 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1660 errmsg("could not serialize access due to concurrent update")));
1661 else if (!ItemPointerEquals(tupleid, &update_ctid))
1663 TupleTableSlot *epqslot;
1665 epqslot = EvalPlanQual(estate,
1666 resultRelInfo->ri_RangeTableIndex,
1669 if (!TupIsNull(epqslot))
1671 *tupleid = update_ctid;
1672 slot = ExecFilterJunk(estate->es_junkFilter, epqslot);
1673 tuple = ExecMaterializeSlot(slot);
1677 /* tuple already deleted; nothing to do */
1681 elog(ERROR, "unrecognized heap_update status: %u", result);
1686 (estate->es_processed)++;
1689 * Note: instead of having to update the old index tuples associated
1690 * with the heap tuple, all we do is form and insert new index tuples.
1691 * This is because UPDATEs are actually DELETEs and INSERTs, and index
1692 * tuple deletion is done automagically by the vacuum daemon. All we
1693 * do is insert new index tuples. -cim 9/27/89
1697 * insert index entries for tuple
1699 * Note: heap_update returns the tid (location) of the new tuple
1700 * in the t_self field.
1702 if (resultRelInfo->ri_NumIndices > 0)
1703 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1705 /* AFTER ROW UPDATE Triggers */
1706 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
1710 ExecRelCheck(ResultRelInfo *resultRelInfo,
1711 TupleTableSlot *slot, EState *estate)
1713 Relation rel = resultRelInfo->ri_RelationDesc;
1714 int ncheck = rel->rd_att->constr->num_check;
1715 ConstrCheck *check = rel->rd_att->constr->check;
1716 ExprContext *econtext;
1717 MemoryContext oldContext;
1722 * If first time through for this result relation, build expression
1723 * nodetrees for rel's constraint expressions. Keep them in the
1724 * per-query memory context so they'll survive throughout the query.
1726 if (resultRelInfo->ri_ConstraintExprs == NULL)
1728 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1729 resultRelInfo->ri_ConstraintExprs =
1730 (List **) palloc(ncheck * sizeof(List *));
1731 for (i = 0; i < ncheck; i++)
1733 /* ExecQual wants implicit-AND form */
1734 qual = make_ands_implicit(stringToNode(check[i].ccbin));
1735 resultRelInfo->ri_ConstraintExprs[i] = (List *)
1736 ExecPrepareExpr((Expr *) qual, estate);
1738 MemoryContextSwitchTo(oldContext);
1742 * We will use the EState's per-tuple context for evaluating
1743 * constraint expressions (creating it if it's not already there).
1745 econtext = GetPerTupleExprContext(estate);
1747 /* Arrange for econtext's scan tuple to be the tuple under test */
1748 econtext->ecxt_scantuple = slot;
1750 /* And evaluate the constraints */
1751 for (i = 0; i < ncheck; i++)
1753 qual = resultRelInfo->ri_ConstraintExprs[i];
1756 * NOTE: SQL92 specifies that a NULL result from a constraint
1757 * expression is not to be treated as a failure. Therefore, tell
1758 * ExecQual to return TRUE for NULL.
1760 if (!ExecQual(qual, econtext, true))
1761 return check[i].ccname;
1764 /* NULL result means no error */
1769 ExecConstraints(ResultRelInfo *resultRelInfo,
1770 TupleTableSlot *slot, EState *estate)
1772 Relation rel = resultRelInfo->ri_RelationDesc;
1773 TupleConstr *constr = rel->rd_att->constr;
1777 if (constr->has_not_null)
1779 int natts = rel->rd_att->natts;
1782 for (attrChk = 1; attrChk <= natts; attrChk++)
1784 if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1785 slot_attisnull(slot, attrChk))
1787 (errcode(ERRCODE_NOT_NULL_VIOLATION),
1788 errmsg("null value in column \"%s\" violates not-null constraint",
1789 NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1793 if (constr->num_check > 0)
1797 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1799 (errcode(ERRCODE_CHECK_VIOLATION),
1800 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1801 RelationGetRelationName(rel), failed)));
1806 * Check a modified tuple to see if we want to process its updated version
1807 * under READ COMMITTED rules.
1809 * See backend/executor/README for some info about how this works.
1811 * estate - executor state data
1812 * rti - rangetable index of table containing tuple
1813 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1814 * priorXmax - t_xmax from the outdated tuple
1816 * *tid is also an output parameter: it's modified to hold the TID of the
1817 * latest version of the tuple (note this may be changed even on failure)
1819 * Returns a slot containing the new candidate update/delete tuple, or
1820 * NULL if we determine we shouldn't process the row.
1823 EvalPlanQual(EState *estate, Index rti,
1824 ItemPointer tid, TransactionId priorXmax)
1829 HeapTupleData tuple;
1830 HeapTuple copyTuple = NULL;
1836 * find relation containing target tuple
1838 if (estate->es_result_relation_info != NULL &&
1839 estate->es_result_relation_info->ri_RangeTableIndex == rti)
1840 relation = estate->es_result_relation_info->ri_RelationDesc;
1846 foreach(l, estate->es_rowMarks)
1848 if (((execRowMark *) lfirst(l))->rti == rti)
1850 relation = ((execRowMark *) lfirst(l))->relation;
1854 if (relation == NULL)
1855 elog(ERROR, "could not find RowMark for RT index %u", rti);
1861 * Loop here to deal with updated or busy tuples
1863 tuple.t_self = *tid;
1868 if (heap_fetch(relation, SnapshotDirty, &tuple, &buffer, true, NULL))
1871 * If xmin isn't what we're expecting, the slot must have been
1872 * recycled and reused for an unrelated tuple. This implies
1873 * that the latest version of the row was deleted, so we need
1874 * do nothing. (Should be safe to examine xmin without getting
1875 * buffer's content lock, since xmin never changes in an existing
1878 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1881 ReleaseBuffer(buffer);
1885 /* otherwise xmin should not be dirty... */
1886 if (TransactionIdIsValid(SnapshotDirty->xmin))
1887 elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1890 * If tuple is being updated by other transaction then we have
1891 * to wait for its commit/abort.
1893 if (TransactionIdIsValid(SnapshotDirty->xmax))
1895 ReleaseBuffer(buffer);
1896 XactLockTableWait(SnapshotDirty->xmax);
1897 continue; /* loop back to repeat heap_fetch */
1901 * We got tuple - now copy it for use by recheck query.
1903 copyTuple = heap_copytuple(&tuple);
1904 ReleaseBuffer(buffer);
1909 * If the referenced slot was actually empty, the latest version
1910 * of the row must have been deleted, so we need do nothing.
1912 if (tuple.t_data == NULL)
1914 ReleaseBuffer(buffer);
1919 * As above, if xmin isn't what we're expecting, do nothing.
1921 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1924 ReleaseBuffer(buffer);
1929 * If we get here, the tuple was found but failed SnapshotDirty.
1930 * Assuming the xmin is either a committed xact or our own xact
1931 * (as it certainly should be if we're trying to modify the tuple),
1932 * this must mean that the row was updated or deleted by either
1933 * a committed xact or our own xact. If it was deleted, we can
1934 * ignore it; if it was updated then chain up to the next version
1935 * and repeat the whole test.
1937 * As above, it should be safe to examine xmax and t_ctid without
1938 * the buffer content lock, because they can't be changing.
1940 if (ItemPointerEquals(&tuple.t_self, &tuple.t_data->t_ctid))
1942 /* deleted, so forget about it */
1943 ReleaseBuffer(buffer);
1947 /* updated, so look at the updated row */
1948 tuple.t_self = tuple.t_data->t_ctid;
1949 /* updated row should have xmin matching this xmax */
1950 priorXmax = HeapTupleHeaderGetXmax(tuple.t_data);
1951 ReleaseBuffer(buffer);
1952 /* loop back to fetch next in chain */
1956 * For UPDATE/DELETE we have to return tid of actual row we're
1959 *tid = tuple.t_self;
1962 * Need to run a recheck subquery. Find or create a PQ stack entry.
1964 epq = estate->es_evalPlanQual;
1967 if (epq != NULL && epq->rti == 0)
1969 /* Top PQ stack entry is idle, so re-use it */
1970 Assert(!(estate->es_useEvalPlan) && epq->next == NULL);
1976 * If this is request for another RTE - Ra, - then we have to check
1977 * wasn't PlanQual requested for Ra already and if so then Ra' row was
1978 * updated again and we have to re-start old execution for Ra and
1979 * forget all what we done after Ra was suspended. Cool? -:))
1981 if (epq != NULL && epq->rti != rti &&
1982 epq->estate->es_evTuple[rti - 1] != NULL)
1986 evalPlanQual *oldepq;
1988 /* stop execution */
1989 EvalPlanQualStop(epq);
1990 /* pop previous PlanQual from the stack */
1992 Assert(oldepq && oldepq->rti != 0);
1993 /* push current PQ to freePQ stack */
1996 estate->es_evalPlanQual = epq;
1997 } while (epq->rti != rti);
2001 * If we are requested for another RTE then we have to suspend
2002 * execution of current PlanQual and start execution for new one.
2004 if (epq == NULL || epq->rti != rti)
2006 /* try to reuse plan used previously */
2007 evalPlanQual *newepq = (epq != NULL) ? epq->free : NULL;
2009 if (newepq == NULL) /* first call or freePQ stack is empty */
2011 newepq = (evalPlanQual *) palloc0(sizeof(evalPlanQual));
2012 newepq->free = NULL;
2013 newepq->estate = NULL;
2014 newepq->planstate = NULL;
2018 /* recycle previously used PlanQual */
2019 Assert(newepq->estate == NULL);
2022 /* push current PQ to the stack */
2025 estate->es_evalPlanQual = epq;
2030 Assert(epq->rti == rti);
2033 * Ok - we're requested for the same RTE. Unfortunately we still have
2034 * to end and restart execution of the plan, because ExecReScan
2035 * wouldn't ensure that upper plan nodes would reset themselves. We
2036 * could make that work if insertion of the target tuple were
2037 * integrated with the Param mechanism somehow, so that the upper plan
2038 * nodes know that their children's outputs have changed.
2040 * Note that the stack of free evalPlanQual nodes is quite useless at the
2041 * moment, since it only saves us from pallocing/releasing the
2042 * evalPlanQual nodes themselves. But it will be useful once we
2043 * implement ReScan instead of end/restart for re-using PlanQual
2048 /* stop execution */
2049 EvalPlanQualStop(epq);
2053 * Initialize new recheck query.
2055 * Note: if we were re-using PlanQual plans via ExecReScan, we'd need to
2056 * instead copy down changeable state from the top plan (including
2057 * es_result_relation_info, es_junkFilter) and reset locally
2058 * changeable state in the epq (including es_param_exec_vals,
2061 EvalPlanQualStart(epq, estate, epq->next);
2064 * free old RTE' tuple, if any, and store target tuple where
2065 * relation's scan node will see it
2067 epqstate = epq->estate;
2068 if (epqstate->es_evTuple[rti - 1] != NULL)
2069 heap_freetuple(epqstate->es_evTuple[rti - 1]);
2070 epqstate->es_evTuple[rti - 1] = copyTuple;
2072 return EvalPlanQualNext(estate);
2075 static TupleTableSlot *
2076 EvalPlanQualNext(EState *estate)
2078 evalPlanQual *epq = estate->es_evalPlanQual;
2079 MemoryContext oldcontext;
2080 TupleTableSlot *slot;
2082 Assert(epq->rti != 0);
2085 oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt);
2086 slot = ExecProcNode(epq->planstate);
2087 MemoryContextSwitchTo(oldcontext);
2090 * No more tuples for this PQ. Continue previous one.
2092 if (TupIsNull(slot))
2094 evalPlanQual *oldepq;
2096 /* stop execution */
2097 EvalPlanQualStop(epq);
2098 /* pop old PQ from the stack */
2102 /* this is the first (oldest) PQ - mark as free */
2104 estate->es_useEvalPlan = false;
2105 /* and continue Query execution */
2108 Assert(oldepq->rti != 0);
2109 /* push current PQ to freePQ stack */
2112 estate->es_evalPlanQual = epq;
2120 EndEvalPlanQual(EState *estate)
2122 evalPlanQual *epq = estate->es_evalPlanQual;
2124 if (epq->rti == 0) /* plans already shutdowned */
2126 Assert(epq->next == NULL);
2132 evalPlanQual *oldepq;
2134 /* stop execution */
2135 EvalPlanQualStop(epq);
2136 /* pop old PQ from the stack */
2140 /* this is the first (oldest) PQ - mark as free */
2142 estate->es_useEvalPlan = false;
2145 Assert(oldepq->rti != 0);
2146 /* push current PQ to freePQ stack */
2149 estate->es_evalPlanQual = epq;
2154 * Start execution of one level of PlanQual.
2156 * This is a cut-down version of ExecutorStart(): we copy some state from
2157 * the top-level estate rather than initializing it fresh.
2160 EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
2164 MemoryContext oldcontext;
2166 rtsize = list_length(estate->es_range_table);
2168 epq->estate = epqstate = CreateExecutorState();
2170 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2173 * The epqstates share the top query's copy of unchanging state such
2174 * as the snapshot, rangetable, result-rel info, and external Param
2175 * info. They need their own copies of local state, including a tuple
2176 * table, es_param_exec_vals, etc.
2178 epqstate->es_direction = ForwardScanDirection;
2179 epqstate->es_snapshot = estate->es_snapshot;
2180 epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot;
2181 epqstate->es_range_table = estate->es_range_table;
2182 epqstate->es_result_relations = estate->es_result_relations;
2183 epqstate->es_num_result_relations = estate->es_num_result_relations;
2184 epqstate->es_result_relation_info = estate->es_result_relation_info;
2185 epqstate->es_junkFilter = estate->es_junkFilter;
2186 epqstate->es_into_relation_descriptor = estate->es_into_relation_descriptor;
2187 epqstate->es_into_relation_use_wal = estate->es_into_relation_use_wal;
2188 epqstate->es_param_list_info = estate->es_param_list_info;
2189 if (estate->es_topPlan->nParamExec > 0)
2190 epqstate->es_param_exec_vals = (ParamExecData *)
2191 palloc0(estate->es_topPlan->nParamExec * sizeof(ParamExecData));
2192 epqstate->es_rowMarks = estate->es_rowMarks;
2193 epqstate->es_forUpdate = estate->es_forUpdate;
2194 epqstate->es_rowNoWait = estate->es_rowNoWait;
2195 epqstate->es_instrument = estate->es_instrument;
2196 epqstate->es_select_into = estate->es_select_into;
2197 epqstate->es_into_oids = estate->es_into_oids;
2198 epqstate->es_topPlan = estate->es_topPlan;
2201 * Each epqstate must have its own es_evTupleNull state, but all the
2202 * stack entries share es_evTuple state. This allows sub-rechecks to
2203 * inherit the value being examined by an outer recheck.
2205 epqstate->es_evTupleNull = (bool *) palloc0(rtsize * sizeof(bool));
2206 if (priorepq == NULL)
2207 /* first PQ stack entry */
2208 epqstate->es_evTuple = (HeapTuple *)
2209 palloc0(rtsize * sizeof(HeapTuple));
2211 /* later stack entries share the same storage */
2212 epqstate->es_evTuple = priorepq->estate->es_evTuple;
2214 epqstate->es_tupleTable =
2215 ExecCreateTupleTable(estate->es_tupleTable->size);
2217 epq->planstate = ExecInitNode(estate->es_topPlan, epqstate);
2219 MemoryContextSwitchTo(oldcontext);
2223 * End execution of one level of PlanQual.
2225 * This is a cut-down version of ExecutorEnd(); basically we want to do most
2226 * of the normal cleanup, but *not* close result relations (which we are
2227 * just sharing from the outer query).
2230 EvalPlanQualStop(evalPlanQual *epq)
2232 EState *epqstate = epq->estate;
2233 MemoryContext oldcontext;
2235 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2237 ExecEndNode(epq->planstate);
2239 ExecDropTupleTable(epqstate->es_tupleTable, true);
2240 epqstate->es_tupleTable = NULL;
2242 if (epqstate->es_evTuple[epq->rti - 1] != NULL)
2244 heap_freetuple(epqstate->es_evTuple[epq->rti - 1]);
2245 epqstate->es_evTuple[epq->rti - 1] = NULL;
2248 MemoryContextSwitchTo(oldcontext);
2250 FreeExecutorState(epqstate);
2253 epq->planstate = NULL;