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11 * The old ExecutorMain() has been replaced by ExecutorStart(),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) 1996-2007, PostgreSQL Global Development Group
25 * Portions Copyright (c) 1994, Regents of the University of California
29 * $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.292 2007/03/29 00:15:38 tgl Exp $
31 *-------------------------------------------------------------------------
35 #include "access/heapam.h"
36 #include "access/reloptions.h"
37 #include "access/transam.h"
38 #include "access/xact.h"
39 #include "catalog/heap.h"
40 #include "catalog/namespace.h"
41 #include "catalog/toasting.h"
42 #include "commands/tablespace.h"
43 #include "commands/trigger.h"
44 #include "executor/execdebug.h"
45 #include "executor/instrument.h"
46 #include "executor/nodeSubplan.h"
47 #include "miscadmin.h"
48 #include "optimizer/clauses.h"
49 #include "parser/parse_clause.h"
50 #include "parser/parsetree.h"
51 #include "storage/smgr.h"
52 #include "utils/acl.h"
53 #include "utils/lsyscache.h"
54 #include "utils/memutils.h"
57 typedef struct evalPlanQual
62 struct evalPlanQual *next; /* stack of active PlanQual plans */
63 struct evalPlanQual *free; /* list of free PlanQual plans */
66 /* decls for local routines only used within this module */
67 static void InitPlan(QueryDesc *queryDesc, int eflags);
68 static void initResultRelInfo(ResultRelInfo *resultRelInfo,
69 Index resultRelationIndex,
73 static void ExecEndPlan(PlanState *planstate, EState *estate);
74 static TupleTableSlot *ExecutePlan(EState *estate, PlanState *planstate,
77 ScanDirection direction,
79 static void ExecSelect(TupleTableSlot *slot,
80 DestReceiver *dest, EState *estate);
81 static void ExecInsert(TupleTableSlot *slot, ItemPointer tupleid,
82 TupleTableSlot *planSlot,
83 DestReceiver *dest, EState *estate);
84 static void ExecDelete(ItemPointer tupleid,
85 TupleTableSlot *planSlot,
86 DestReceiver *dest, EState *estate);
87 static void ExecUpdate(TupleTableSlot *slot, ItemPointer tupleid,
88 TupleTableSlot *planSlot,
89 DestReceiver *dest, EState *estate);
90 static void ExecProcessReturning(ProjectionInfo *projectReturning,
91 TupleTableSlot *tupleSlot,
92 TupleTableSlot *planSlot,
94 static TupleTableSlot *EvalPlanQualNext(EState *estate);
95 static void EndEvalPlanQual(EState *estate);
96 static void ExecCheckRTPerms(List *rangeTable);
97 static void ExecCheckRTEPerms(RangeTblEntry *rte);
98 static void ExecCheckXactReadOnly(PlannedStmt *plannedstmt);
99 static void EvalPlanQualStart(evalPlanQual *epq, EState *estate,
100 evalPlanQual *priorepq);
101 static void EvalPlanQualStop(evalPlanQual *epq);
102 static void OpenIntoRel(QueryDesc *queryDesc);
103 static void CloseIntoRel(QueryDesc *queryDesc);
104 static void intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo);
105 static void intorel_receive(TupleTableSlot *slot, DestReceiver *self);
106 static void intorel_shutdown(DestReceiver *self);
107 static void intorel_destroy(DestReceiver *self);
109 /* end of local decls */
112 /* ----------------------------------------------------------------
115 * This routine must be called at the beginning of any execution of any
118 * Takes a QueryDesc previously created by CreateQueryDesc (it's not real
119 * clear why we bother to separate the two functions, but...). The tupDesc
120 * field of the QueryDesc is filled in to describe the tuples that will be
121 * returned, and the internal fields (estate and planstate) are set up.
123 * eflags contains flag bits as described in executor.h.
125 * NB: the CurrentMemoryContext when this is called will become the parent
126 * of the per-query context used for this Executor invocation.
127 * ----------------------------------------------------------------
130 ExecutorStart(QueryDesc *queryDesc, int eflags)
133 MemoryContext oldcontext;
135 /* sanity checks: queryDesc must not be started already */
136 Assert(queryDesc != NULL);
137 Assert(queryDesc->estate == NULL);
140 * If the transaction is read-only, we need to check if any writes are
141 * planned to non-temporary tables. EXPLAIN is considered read-only.
143 if (XactReadOnly && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
144 ExecCheckXactReadOnly(queryDesc->plannedstmt);
147 * Build EState, switch into per-query memory context for startup.
149 estate = CreateExecutorState();
150 queryDesc->estate = estate;
152 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
155 * Fill in parameters, if any, from queryDesc
157 estate->es_param_list_info = queryDesc->params;
159 if (queryDesc->plannedstmt->nParamExec > 0)
160 estate->es_param_exec_vals = (ParamExecData *)
161 palloc0(queryDesc->plannedstmt->nParamExec * sizeof(ParamExecData));
164 * Copy other important information into the EState
166 estate->es_snapshot = queryDesc->snapshot;
167 estate->es_crosscheck_snapshot = queryDesc->crosscheck_snapshot;
168 estate->es_instrument = queryDesc->doInstrument;
171 * Initialize the plan state tree
173 InitPlan(queryDesc, eflags);
175 MemoryContextSwitchTo(oldcontext);
178 /* ----------------------------------------------------------------
181 * This is the main routine of the executor module. It accepts
182 * the query descriptor from the traffic cop and executes the
185 * ExecutorStart must have been called already.
187 * If direction is NoMovementScanDirection then nothing is done
188 * except to start up/shut down the destination. Otherwise,
189 * we retrieve up to 'count' tuples in the specified direction.
191 * Note: count = 0 is interpreted as no portal limit, i.e., run to
194 * ----------------------------------------------------------------
197 ExecutorRun(QueryDesc *queryDesc,
198 ScanDirection direction, long count)
204 TupleTableSlot *result;
205 MemoryContext oldcontext;
208 Assert(queryDesc != NULL);
210 estate = queryDesc->estate;
212 Assert(estate != NULL);
215 * Switch into per-query memory context
217 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
220 * extract information from the query descriptor and the query feature.
222 operation = queryDesc->operation;
223 dest = queryDesc->dest;
226 * startup tuple receiver, if we will be emitting tuples
228 estate->es_processed = 0;
229 estate->es_lastoid = InvalidOid;
231 sendTuples = (operation == CMD_SELECT ||
232 queryDesc->plannedstmt->returningLists);
235 (*dest->rStartup) (dest, operation, queryDesc->tupDesc);
240 if (ScanDirectionIsNoMovement(direction))
243 result = ExecutePlan(estate,
244 queryDesc->planstate,
251 * shutdown tuple receiver, if we started it
254 (*dest->rShutdown) (dest);
256 MemoryContextSwitchTo(oldcontext);
261 /* ----------------------------------------------------------------
264 * This routine must be called at the end of execution of any
266 * ----------------------------------------------------------------
269 ExecutorEnd(QueryDesc *queryDesc)
272 MemoryContext oldcontext;
275 Assert(queryDesc != NULL);
277 estate = queryDesc->estate;
279 Assert(estate != NULL);
282 * Switch into per-query memory context to run ExecEndPlan
284 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
286 ExecEndPlan(queryDesc->planstate, estate);
289 * Close the SELECT INTO relation if any
291 if (estate->es_select_into)
292 CloseIntoRel(queryDesc);
295 * Must switch out of context before destroying it
297 MemoryContextSwitchTo(oldcontext);
300 * Release EState and per-query memory context. This should release
301 * everything the executor has allocated.
303 FreeExecutorState(estate);
305 /* Reset queryDesc fields that no longer point to anything */
306 queryDesc->tupDesc = NULL;
307 queryDesc->estate = NULL;
308 queryDesc->planstate = NULL;
311 /* ----------------------------------------------------------------
314 * This routine may be called on an open queryDesc to rewind it
316 * ----------------------------------------------------------------
319 ExecutorRewind(QueryDesc *queryDesc)
322 MemoryContext oldcontext;
325 Assert(queryDesc != NULL);
327 estate = queryDesc->estate;
329 Assert(estate != NULL);
331 /* It's probably not sensible to rescan updating queries */
332 Assert(queryDesc->operation == CMD_SELECT);
335 * Switch into per-query memory context
337 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
342 ExecReScan(queryDesc->planstate, NULL);
344 MemoryContextSwitchTo(oldcontext);
350 * Check access permissions for all relations listed in a range table.
353 ExecCheckRTPerms(List *rangeTable)
357 foreach(l, rangeTable)
359 ExecCheckRTEPerms((RangeTblEntry *) lfirst(l));
365 * Check access permissions for a single RTE.
368 ExecCheckRTEPerms(RangeTblEntry *rte)
370 AclMode requiredPerms;
375 * Only plain-relation RTEs need to be checked here. Function RTEs are
376 * checked by init_fcache when the function is prepared for execution.
377 * Join, subquery, and special RTEs need no checks.
379 if (rte->rtekind != RTE_RELATION)
383 * No work if requiredPerms is empty.
385 requiredPerms = rte->requiredPerms;
386 if (requiredPerms == 0)
392 * userid to check as: current user unless we have a setuid indication.
394 * Note: GetUserId() is presently fast enough that there's no harm in
395 * calling it separately for each RTE. If that stops being true, we could
396 * call it once in ExecCheckRTPerms and pass the userid down from there.
397 * But for now, no need for the extra clutter.
399 userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
402 * We must have *all* the requiredPerms bits, so use aclmask not aclcheck.
404 if (pg_class_aclmask(relOid, userid, requiredPerms, ACLMASK_ALL)
406 aclcheck_error(ACLCHECK_NO_PRIV, ACL_KIND_CLASS,
407 get_rel_name(relOid));
411 * Check that the query does not imply any writes to non-temp tables.
414 ExecCheckXactReadOnly(PlannedStmt *plannedstmt)
419 * CREATE TABLE AS or SELECT INTO?
421 * XXX should we allow this if the destination is temp?
423 if (plannedstmt->into != NULL)
426 /* Fail if write permissions are requested on any non-temp table */
427 foreach(l, plannedstmt->rtable)
429 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
431 if (rte->rtekind != RTE_RELATION)
434 if ((rte->requiredPerms & (~ACL_SELECT)) == 0)
437 if (isTempNamespace(get_rel_namespace(rte->relid)))
447 (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
448 errmsg("transaction is read-only")));
452 /* ----------------------------------------------------------------
455 * Initializes the query plan: open files, allocate storage
456 * and start up the rule manager
457 * ----------------------------------------------------------------
460 InitPlan(QueryDesc *queryDesc, int eflags)
462 CmdType operation = queryDesc->operation;
463 PlannedStmt *plannedstmt = queryDesc->plannedstmt;
464 Plan *plan = plannedstmt->planTree;
465 List *rangeTable = plannedstmt->rtable;
466 EState *estate = queryDesc->estate;
467 PlanState *planstate;
473 * Do permissions checks
475 ExecCheckRTPerms(rangeTable);
478 * initialize the node's execution state
480 estate->es_range_table = rangeTable;
483 * initialize result relation stuff
485 if (plannedstmt->resultRelations)
487 List *resultRelations = plannedstmt->resultRelations;
488 int numResultRelations = list_length(resultRelations);
489 ResultRelInfo *resultRelInfos;
490 ResultRelInfo *resultRelInfo;
492 resultRelInfos = (ResultRelInfo *)
493 palloc(numResultRelations * sizeof(ResultRelInfo));
494 resultRelInfo = resultRelInfos;
495 foreach(l, resultRelations)
497 initResultRelInfo(resultRelInfo,
501 estate->es_instrument);
504 estate->es_result_relations = resultRelInfos;
505 estate->es_num_result_relations = numResultRelations;
506 /* Initialize to first or only result rel */
507 estate->es_result_relation_info = resultRelInfos;
512 * if no result relation, then set state appropriately
514 estate->es_result_relations = NULL;
515 estate->es_num_result_relations = 0;
516 estate->es_result_relation_info = NULL;
520 * Detect whether we're doing SELECT INTO. If so, set the es_into_oids
521 * flag appropriately so that the plan tree will be initialized with the
522 * correct tuple descriptors. (Other SELECT INTO stuff comes later.)
524 estate->es_select_into = false;
525 if (operation == CMD_SELECT && plannedstmt->into != NULL)
527 estate->es_select_into = true;
528 estate->es_into_oids = interpretOidsOption(plannedstmt->into->options);
532 * Have to lock relations selected FOR UPDATE/FOR SHARE before we
533 * initialize the plan tree, else we'd be doing a lock upgrade.
534 * While we are at it, build the ExecRowMark list.
536 estate->es_rowMarks = NIL;
537 foreach(l, plannedstmt->rowMarks)
539 RowMarkClause *rc = (RowMarkClause *) lfirst(l);
540 Oid relid = getrelid(rc->rti, rangeTable);
544 relation = heap_open(relid, RowShareLock);
545 erm = (ExecRowMark *) palloc(sizeof(ExecRowMark));
546 erm->relation = relation;
548 erm->forUpdate = rc->forUpdate;
549 erm->noWait = rc->noWait;
550 /* We'll set up ctidAttno below */
551 erm->ctidAttNo = InvalidAttrNumber;
552 estate->es_rowMarks = lappend(estate->es_rowMarks, erm);
556 * Initialize the executor "tuple" table. We need slots for all the plan
557 * nodes, plus possibly output slots for the junkfilter(s). At this point
558 * we aren't sure if we need junkfilters, so just add slots for them
559 * unconditionally. Also, if it's not a SELECT, set up a slot for use for
560 * trigger output tuples. Also, one for RETURNING-list evaluation.
565 /* Slots for the main plan tree */
566 nSlots = ExecCountSlotsNode(plan);
567 /* Add slots for subplans and initplans */
568 foreach(l, plannedstmt->subplans)
570 Plan *subplan = (Plan *) lfirst(l);
572 nSlots += ExecCountSlotsNode(subplan);
574 /* Add slots for junkfilter(s) */
575 if (plannedstmt->resultRelations != NIL)
576 nSlots += list_length(plannedstmt->resultRelations);
579 if (operation != CMD_SELECT)
580 nSlots++; /* for es_trig_tuple_slot */
581 if (plannedstmt->returningLists)
582 nSlots++; /* for RETURNING projection */
584 estate->es_tupleTable = ExecCreateTupleTable(nSlots);
586 if (operation != CMD_SELECT)
587 estate->es_trig_tuple_slot =
588 ExecAllocTableSlot(estate->es_tupleTable);
591 /* mark EvalPlanQual not active */
592 estate->es_plannedstmt = plannedstmt;
593 estate->es_evalPlanQual = NULL;
594 estate->es_evTupleNull = NULL;
595 estate->es_evTuple = NULL;
596 estate->es_useEvalPlan = false;
599 * Initialize private state information for each SubPlan. We must do
600 * this before running ExecInitNode on the main query tree, since
601 * ExecInitSubPlan expects to be able to find these entries.
603 Assert(estate->es_subplanstates == NIL);
604 i = 1; /* subplan indices count from 1 */
605 foreach(l, plannedstmt->subplans)
607 Plan *subplan = (Plan *) lfirst(l);
608 PlanState *subplanstate;
612 * A subplan will never need to do BACKWARD scan nor MARK/RESTORE.
613 * If it is a parameterless subplan (not initplan), we suggest that it
614 * be prepared to handle REWIND efficiently; otherwise there is no
617 sp_eflags = eflags & EXEC_FLAG_EXPLAIN_ONLY;
618 if (bms_is_member(i, plannedstmt->rewindPlanIDs))
619 sp_eflags |= EXEC_FLAG_REWIND;
621 subplanstate = ExecInitNode(subplan, estate, sp_eflags);
623 estate->es_subplanstates = lappend(estate->es_subplanstates,
630 * Initialize the private state information for all the nodes in the query
631 * tree. This opens files, allocates storage and leaves us ready to start
634 planstate = ExecInitNode(plan, estate, eflags);
637 * Get the tuple descriptor describing the type of tuples to return. (this
638 * is especially important if we are creating a relation with "SELECT
641 tupType = ExecGetResultType(planstate);
644 * Initialize the junk filter if needed. SELECT and INSERT queries need a
645 * filter if there are any junk attrs in the tlist. INSERT and SELECT
646 * INTO also need a filter if the plan may return raw disk tuples (else
647 * heap_insert will be scribbling on the source relation!). UPDATE and
648 * DELETE always need a filter, since there's always a junk 'ctid'
649 * attribute present --- no need to look first.
652 bool junk_filter_needed = false;
659 foreach(tlist, plan->targetlist)
661 TargetEntry *tle = (TargetEntry *) lfirst(tlist);
665 junk_filter_needed = true;
669 if (!junk_filter_needed &&
670 (operation == CMD_INSERT || estate->es_select_into) &&
671 ExecMayReturnRawTuples(planstate))
672 junk_filter_needed = true;
676 junk_filter_needed = true;
682 if (junk_filter_needed)
685 * If there are multiple result relations, each one needs its own
686 * junk filter. Note this is only possible for UPDATE/DELETE, so
687 * we can't be fooled by some needing a filter and some not.
689 if (list_length(plannedstmt->resultRelations) > 1)
691 PlanState **appendplans;
693 ResultRelInfo *resultRelInfo;
695 /* Top plan had better be an Append here. */
696 Assert(IsA(plan, Append));
697 Assert(((Append *) plan)->isTarget);
698 Assert(IsA(planstate, AppendState));
699 appendplans = ((AppendState *) planstate)->appendplans;
700 as_nplans = ((AppendState *) planstate)->as_nplans;
701 Assert(as_nplans == estate->es_num_result_relations);
702 resultRelInfo = estate->es_result_relations;
703 for (i = 0; i < as_nplans; i++)
705 PlanState *subplan = appendplans[i];
708 j = ExecInitJunkFilter(subplan->plan->targetlist,
709 resultRelInfo->ri_RelationDesc->rd_att->tdhasoid,
710 ExecAllocTableSlot(estate->es_tupleTable));
712 * Since it must be UPDATE/DELETE, there had better be
713 * a "ctid" junk attribute in the tlist ... but ctid could
714 * be at a different resno for each result relation.
715 * We look up the ctid resnos now and save them in the
718 j->jf_junkAttNo = ExecFindJunkAttribute(j, "ctid");
719 if (!AttributeNumberIsValid(j->jf_junkAttNo))
720 elog(ERROR, "could not find junk ctid column");
721 resultRelInfo->ri_junkFilter = j;
726 * Set active junkfilter too; at this point ExecInitAppend has
727 * already selected an active result relation...
729 estate->es_junkFilter =
730 estate->es_result_relation_info->ri_junkFilter;
734 /* Normal case with just one JunkFilter */
737 j = ExecInitJunkFilter(planstate->plan->targetlist,
739 ExecAllocTableSlot(estate->es_tupleTable));
740 estate->es_junkFilter = j;
741 if (estate->es_result_relation_info)
742 estate->es_result_relation_info->ri_junkFilter = j;
744 if (operation == CMD_SELECT)
746 /* For SELECT, want to return the cleaned tuple type */
747 tupType = j->jf_cleanTupType;
748 /* For SELECT FOR UPDATE/SHARE, find the ctid attrs now */
749 foreach(l, estate->es_rowMarks)
751 ExecRowMark *erm = (ExecRowMark *) lfirst(l);
754 snprintf(resname, sizeof(resname), "ctid%u", erm->rti);
755 erm->ctidAttNo = ExecFindJunkAttribute(j, resname);
756 if (!AttributeNumberIsValid(erm->ctidAttNo))
757 elog(ERROR, "could not find junk \"%s\" column",
761 else if (operation == CMD_UPDATE || operation == CMD_DELETE)
763 /* For UPDATE/DELETE, find the ctid junk attr now */
764 j->jf_junkAttNo = ExecFindJunkAttribute(j, "ctid");
765 if (!AttributeNumberIsValid(j->jf_junkAttNo))
766 elog(ERROR, "could not find junk ctid column");
771 estate->es_junkFilter = NULL;
775 * Initialize RETURNING projections if needed.
777 if (plannedstmt->returningLists)
779 TupleTableSlot *slot;
780 ExprContext *econtext;
781 ResultRelInfo *resultRelInfo;
784 * We set QueryDesc.tupDesc to be the RETURNING rowtype in this case.
785 * We assume all the sublists will generate the same output tupdesc.
787 tupType = ExecTypeFromTL((List *) linitial(plannedstmt->returningLists),
790 /* Set up a slot for the output of the RETURNING projection(s) */
791 slot = ExecAllocTableSlot(estate->es_tupleTable);
792 ExecSetSlotDescriptor(slot, tupType);
793 /* Need an econtext too */
794 econtext = CreateExprContext(estate);
797 * Build a projection for each result rel. Note that any SubPlans in
798 * the RETURNING lists get attached to the topmost plan node.
800 Assert(list_length(plannedstmt->returningLists) == estate->es_num_result_relations);
801 resultRelInfo = estate->es_result_relations;
802 foreach(l, plannedstmt->returningLists)
804 List *rlist = (List *) lfirst(l);
807 rliststate = (List *) ExecInitExpr((Expr *) rlist, planstate);
808 resultRelInfo->ri_projectReturning =
809 ExecBuildProjectionInfo(rliststate, econtext, slot,
810 resultRelInfo->ri_RelationDesc->rd_att);
815 queryDesc->tupDesc = tupType;
816 queryDesc->planstate = planstate;
819 * If doing SELECT INTO, initialize the "into" relation. We must wait
820 * till now so we have the "clean" result tuple type to create the new
823 * If EXPLAIN, skip creating the "into" relation.
825 if (estate->es_select_into && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
826 OpenIntoRel(queryDesc);
830 * Initialize ResultRelInfo data for one result relation
833 initResultRelInfo(ResultRelInfo *resultRelInfo,
834 Index resultRelationIndex,
839 Oid resultRelationOid;
840 Relation resultRelationDesc;
842 resultRelationOid = getrelid(resultRelationIndex, rangeTable);
843 resultRelationDesc = heap_open(resultRelationOid, RowExclusiveLock);
845 switch (resultRelationDesc->rd_rel->relkind)
847 case RELKIND_SEQUENCE:
849 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
850 errmsg("cannot change sequence \"%s\"",
851 RelationGetRelationName(resultRelationDesc))));
853 case RELKIND_TOASTVALUE:
855 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
856 errmsg("cannot change TOAST relation \"%s\"",
857 RelationGetRelationName(resultRelationDesc))));
861 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
862 errmsg("cannot change view \"%s\"",
863 RelationGetRelationName(resultRelationDesc))));
867 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
868 resultRelInfo->type = T_ResultRelInfo;
869 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
870 resultRelInfo->ri_RelationDesc = resultRelationDesc;
871 resultRelInfo->ri_NumIndices = 0;
872 resultRelInfo->ri_IndexRelationDescs = NULL;
873 resultRelInfo->ri_IndexRelationInfo = NULL;
874 /* make a copy so as not to depend on relcache info not changing... */
875 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
876 if (resultRelInfo->ri_TrigDesc)
878 int n = resultRelInfo->ri_TrigDesc->numtriggers;
880 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
881 palloc0(n * sizeof(FmgrInfo));
883 resultRelInfo->ri_TrigInstrument = InstrAlloc(n);
885 resultRelInfo->ri_TrigInstrument = NULL;
889 resultRelInfo->ri_TrigFunctions = NULL;
890 resultRelInfo->ri_TrigInstrument = NULL;
892 resultRelInfo->ri_ConstraintExprs = NULL;
893 resultRelInfo->ri_junkFilter = NULL;
894 resultRelInfo->ri_projectReturning = NULL;
897 * If there are indices on the result relation, open them and save
898 * descriptors in the result relation info, so that we can add new index
899 * entries for the tuples we add/update. We need not do this for a
900 * DELETE, however, since deletion doesn't affect indexes.
902 if (resultRelationDesc->rd_rel->relhasindex &&
903 operation != CMD_DELETE)
904 ExecOpenIndices(resultRelInfo);
908 * ExecContextForcesOids
910 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
911 * we need to ensure that result tuples have space for an OID iff they are
912 * going to be stored into a relation that has OIDs. In other contexts
913 * we are free to choose whether to leave space for OIDs in result tuples
914 * (we generally don't want to, but we do if a physical-tlist optimization
915 * is possible). This routine checks the plan context and returns TRUE if the
916 * choice is forced, FALSE if the choice is not forced. In the TRUE case,
917 * *hasoids is set to the required value.
919 * One reason this is ugly is that all plan nodes in the plan tree will emit
920 * tuples with space for an OID, though we really only need the topmost node
921 * to do so. However, node types like Sort don't project new tuples but just
922 * return their inputs, and in those cases the requirement propagates down
923 * to the input node. Eventually we might make this code smart enough to
924 * recognize how far down the requirement really goes, but for now we just
925 * make all plan nodes do the same thing if the top level forces the choice.
927 * We assume that estate->es_result_relation_info is already set up to
928 * describe the target relation. Note that in an UPDATE that spans an
929 * inheritance tree, some of the target relations may have OIDs and some not.
930 * We have to make the decisions on a per-relation basis as we initialize
931 * each of the child plans of the topmost Append plan.
933 * SELECT INTO is even uglier, because we don't have the INTO relation's
934 * descriptor available when this code runs; we have to look aside at a
935 * flag set by InitPlan().
938 ExecContextForcesOids(PlanState *planstate, bool *hasoids)
940 if (planstate->state->es_select_into)
942 *hasoids = planstate->state->es_into_oids;
947 ResultRelInfo *ri = planstate->state->es_result_relation_info;
951 Relation rel = ri->ri_RelationDesc;
955 *hasoids = rel->rd_rel->relhasoids;
964 /* ----------------------------------------------------------------
967 * Cleans up the query plan -- closes files and frees up storage
969 * NOTE: we are no longer very worried about freeing storage per se
970 * in this code; FreeExecutorState should be guaranteed to release all
971 * memory that needs to be released. What we are worried about doing
972 * is closing relations and dropping buffer pins. Thus, for example,
973 * tuple tables must be cleared or dropped to ensure pins are released.
974 * ----------------------------------------------------------------
977 ExecEndPlan(PlanState *planstate, EState *estate)
979 ResultRelInfo *resultRelInfo;
984 * shut down any PlanQual processing we were doing
986 if (estate->es_evalPlanQual != NULL)
987 EndEvalPlanQual(estate);
990 * shut down the node-type-specific query processing
992 ExecEndNode(planstate);
997 foreach(l, estate->es_subplanstates)
999 PlanState *subplanstate = (PlanState *) lfirst(l);
1001 ExecEndNode(subplanstate);
1005 * destroy the executor "tuple" table.
1007 ExecDropTupleTable(estate->es_tupleTable, true);
1008 estate->es_tupleTable = NULL;
1011 * close the result relation(s) if any, but hold locks until xact commit.
1013 resultRelInfo = estate->es_result_relations;
1014 for (i = estate->es_num_result_relations; i > 0; i--)
1016 /* Close indices and then the relation itself */
1017 ExecCloseIndices(resultRelInfo);
1018 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
1023 * close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
1025 foreach(l, estate->es_rowMarks)
1027 ExecRowMark *erm = lfirst(l);
1029 heap_close(erm->relation, NoLock);
1033 /* ----------------------------------------------------------------
1036 * processes the query plan to retrieve 'numberTuples' tuples in the
1037 * direction specified.
1039 * Retrieves all tuples if numberTuples is 0
1041 * result is either a slot containing the last tuple in the case
1042 * of a SELECT or NULL otherwise.
1044 * Note: the ctid attribute is a 'junk' attribute that is removed before the
1046 * ----------------------------------------------------------------
1048 static TupleTableSlot *
1049 ExecutePlan(EState *estate,
1050 PlanState *planstate,
1053 ScanDirection direction,
1056 JunkFilter *junkfilter;
1057 TupleTableSlot *planSlot;
1058 TupleTableSlot *slot;
1059 ItemPointer tupleid = NULL;
1060 ItemPointerData tuple_ctid;
1061 long current_tuple_count;
1062 TupleTableSlot *result;
1065 * initialize local variables
1067 current_tuple_count = 0;
1071 * Set the direction.
1073 estate->es_direction = direction;
1076 * Process BEFORE EACH STATEMENT triggers
1081 ExecBSUpdateTriggers(estate, estate->es_result_relation_info);
1084 ExecBSDeleteTriggers(estate, estate->es_result_relation_info);
1087 ExecBSInsertTriggers(estate, estate->es_result_relation_info);
1095 * Loop until we've processed the proper number of tuples from the plan.
1100 /* Reset the per-output-tuple exprcontext */
1101 ResetPerTupleExprContext(estate);
1104 * Execute the plan and obtain a tuple
1107 if (estate->es_useEvalPlan)
1109 planSlot = EvalPlanQualNext(estate);
1110 if (TupIsNull(planSlot))
1111 planSlot = ExecProcNode(planstate);
1114 planSlot = ExecProcNode(planstate);
1117 * if the tuple is null, then we assume there is nothing more to
1118 * process so we just return null...
1120 if (TupIsNull(planSlot))
1128 * if we have a junk filter, then project a new tuple with the junk
1131 * Store this new "clean" tuple in the junkfilter's resultSlot.
1132 * (Formerly, we stored it back over the "dirty" tuple, which is WRONG
1133 * because that tuple slot has the wrong descriptor.)
1135 * Also, extract all the junk information we need.
1137 if ((junkfilter = estate->es_junkFilter) != NULL)
1143 * extract the 'ctid' junk attribute.
1145 if (operation == CMD_UPDATE || operation == CMD_DELETE)
1147 datum = ExecGetJunkAttribute(slot, junkfilter->jf_junkAttNo,
1149 /* shouldn't ever get a null result... */
1151 elog(ERROR, "ctid is NULL");
1153 tupleid = (ItemPointer) DatumGetPointer(datum);
1154 tuple_ctid = *tupleid; /* make sure we don't free the ctid!! */
1155 tupleid = &tuple_ctid;
1159 * Process any FOR UPDATE or FOR SHARE locking requested.
1161 else if (estate->es_rowMarks != NIL)
1166 foreach(l, estate->es_rowMarks)
1168 ExecRowMark *erm = lfirst(l);
1169 HeapTupleData tuple;
1171 ItemPointerData update_ctid;
1172 TransactionId update_xmax;
1173 TupleTableSlot *newSlot;
1174 LockTupleMode lockmode;
1177 datum = ExecGetJunkAttribute(slot,
1180 /* shouldn't ever get a null result... */
1182 elog(ERROR, "ctid is NULL");
1184 tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1187 lockmode = LockTupleExclusive;
1189 lockmode = LockTupleShared;
1191 test = heap_lock_tuple(erm->relation, &tuple, &buffer,
1192 &update_ctid, &update_xmax,
1193 estate->es_snapshot->curcid,
1194 lockmode, erm->noWait);
1195 ReleaseBuffer(buffer);
1198 case HeapTupleSelfUpdated:
1199 /* treat it as deleted; do not process */
1202 case HeapTupleMayBeUpdated:
1205 case HeapTupleUpdated:
1206 if (IsXactIsoLevelSerializable)
1208 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1209 errmsg("could not serialize access due to concurrent update")));
1210 if (!ItemPointerEquals(&update_ctid,
1213 /* updated, so look at updated version */
1214 newSlot = EvalPlanQual(estate,
1218 estate->es_snapshot->curcid);
1219 if (!TupIsNull(newSlot))
1221 slot = planSlot = newSlot;
1222 estate->es_useEvalPlan = true;
1228 * if tuple was deleted or PlanQual failed for
1229 * updated tuple - we must not return this tuple!
1234 elog(ERROR, "unrecognized heap_lock_tuple status: %u",
1242 * Create a new "clean" tuple with all junk attributes removed. We
1243 * don't need to do this for DELETE, however (there will in fact
1244 * be no non-junk attributes in a DELETE!)
1246 if (operation != CMD_DELETE)
1247 slot = ExecFilterJunk(junkfilter, slot);
1251 * now that we have a tuple, do the appropriate thing with it.. either
1252 * return it to the user, add it to a relation someplace, delete it
1253 * from a relation, or modify some of its attributes.
1258 ExecSelect(slot, dest, estate);
1263 ExecInsert(slot, tupleid, planSlot, dest, estate);
1268 ExecDelete(tupleid, planSlot, dest, estate);
1273 ExecUpdate(slot, tupleid, planSlot, dest, estate);
1278 elog(ERROR, "unrecognized operation code: %d",
1285 * check our tuple count.. if we've processed the proper number then
1286 * quit, else loop again and process more tuples. Zero numberTuples
1289 current_tuple_count++;
1290 if (numberTuples && numberTuples == current_tuple_count)
1295 * Process AFTER EACH STATEMENT triggers
1300 ExecASUpdateTriggers(estate, estate->es_result_relation_info);
1303 ExecASDeleteTriggers(estate, estate->es_result_relation_info);
1306 ExecASInsertTriggers(estate, estate->es_result_relation_info);
1314 * here, result is either a slot containing a tuple in the case of a
1315 * SELECT or NULL otherwise.
1320 /* ----------------------------------------------------------------
1323 * SELECTs are easy.. we just pass the tuple to the appropriate
1325 * ----------------------------------------------------------------
1328 ExecSelect(TupleTableSlot *slot,
1332 (*dest->receiveSlot) (slot, dest);
1334 (estate->es_processed)++;
1337 /* ----------------------------------------------------------------
1340 * INSERTs are trickier.. we have to insert the tuple into
1341 * the base relation and insert appropriate tuples into the
1343 * ----------------------------------------------------------------
1346 ExecInsert(TupleTableSlot *slot,
1347 ItemPointer tupleid,
1348 TupleTableSlot *planSlot,
1353 ResultRelInfo *resultRelInfo;
1354 Relation resultRelationDesc;
1358 * get the heap tuple out of the tuple table slot, making sure we have a
1361 tuple = ExecMaterializeSlot(slot);
1364 * get information on the (current) result relation
1366 resultRelInfo = estate->es_result_relation_info;
1367 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1369 /* BEFORE ROW INSERT Triggers */
1370 if (resultRelInfo->ri_TrigDesc &&
1371 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
1375 newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
1377 if (newtuple == NULL) /* "do nothing" */
1380 if (newtuple != tuple) /* modified by Trigger(s) */
1383 * Put the modified tuple into a slot for convenience of routines
1384 * below. We assume the tuple was allocated in per-tuple memory
1385 * context, and therefore will go away by itself. The tuple table
1386 * slot should not try to clear it.
1388 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1390 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1391 ExecSetSlotDescriptor(newslot, slot->tts_tupleDescriptor);
1392 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1399 * Check the constraints of the tuple
1401 if (resultRelationDesc->rd_att->constr)
1402 ExecConstraints(resultRelInfo, slot, estate);
1407 * Note: heap_insert returns the tid (location) of the new tuple in the
1410 newId = heap_insert(resultRelationDesc, tuple,
1411 estate->es_snapshot->curcid,
1415 (estate->es_processed)++;
1416 estate->es_lastoid = newId;
1417 setLastTid(&(tuple->t_self));
1420 * insert index entries for tuple
1422 if (resultRelInfo->ri_NumIndices > 0)
1423 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1425 /* AFTER ROW INSERT Triggers */
1426 ExecARInsertTriggers(estate, resultRelInfo, tuple);
1428 /* Process RETURNING if present */
1429 if (resultRelInfo->ri_projectReturning)
1430 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1431 slot, planSlot, dest);
1434 /* ----------------------------------------------------------------
1437 * DELETE is like UPDATE, except that we delete the tuple and no
1438 * index modifications are needed
1439 * ----------------------------------------------------------------
1442 ExecDelete(ItemPointer tupleid,
1443 TupleTableSlot *planSlot,
1447 ResultRelInfo *resultRelInfo;
1448 Relation resultRelationDesc;
1450 ItemPointerData update_ctid;
1451 TransactionId update_xmax;
1454 * get information on the (current) result relation
1456 resultRelInfo = estate->es_result_relation_info;
1457 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1459 /* BEFORE ROW DELETE Triggers */
1460 if (resultRelInfo->ri_TrigDesc &&
1461 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
1465 dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
1466 estate->es_snapshot->curcid);
1468 if (!dodelete) /* "do nothing" */
1475 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1476 * the row to be deleted is visible to that snapshot, and throw a can't-
1477 * serialize error if not. This is a special-case behavior needed for
1478 * referential integrity updates in serializable transactions.
1481 result = heap_delete(resultRelationDesc, tupleid,
1482 &update_ctid, &update_xmax,
1483 estate->es_snapshot->curcid,
1484 estate->es_crosscheck_snapshot,
1485 true /* wait for commit */ );
1488 case HeapTupleSelfUpdated:
1489 /* already deleted by self; nothing to do */
1492 case HeapTupleMayBeUpdated:
1495 case HeapTupleUpdated:
1496 if (IsXactIsoLevelSerializable)
1498 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1499 errmsg("could not serialize access due to concurrent update")));
1500 else if (!ItemPointerEquals(tupleid, &update_ctid))
1502 TupleTableSlot *epqslot;
1504 epqslot = EvalPlanQual(estate,
1505 resultRelInfo->ri_RangeTableIndex,
1508 estate->es_snapshot->curcid);
1509 if (!TupIsNull(epqslot))
1511 *tupleid = update_ctid;
1515 /* tuple already deleted; nothing to do */
1519 elog(ERROR, "unrecognized heap_delete status: %u", result);
1524 (estate->es_processed)++;
1527 * Note: Normally one would think that we have to delete index tuples
1528 * associated with the heap tuple now...
1530 * ... but in POSTGRES, we have no need to do this because VACUUM will
1531 * take care of it later. We can't delete index tuples immediately
1532 * anyway, since the tuple is still visible to other transactions.
1535 /* AFTER ROW DELETE Triggers */
1536 ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
1538 /* Process RETURNING if present */
1539 if (resultRelInfo->ri_projectReturning)
1542 * We have to put the target tuple into a slot, which means first we
1543 * gotta fetch it. We can use the trigger tuple slot.
1545 TupleTableSlot *slot = estate->es_trig_tuple_slot;
1546 HeapTupleData deltuple;
1549 deltuple.t_self = *tupleid;
1550 if (!heap_fetch(resultRelationDesc, SnapshotAny,
1551 &deltuple, &delbuffer, false, NULL))
1552 elog(ERROR, "failed to fetch deleted tuple for DELETE RETURNING");
1554 if (slot->tts_tupleDescriptor != RelationGetDescr(resultRelationDesc))
1555 ExecSetSlotDescriptor(slot, RelationGetDescr(resultRelationDesc));
1556 ExecStoreTuple(&deltuple, slot, InvalidBuffer, false);
1558 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1559 slot, planSlot, dest);
1561 ExecClearTuple(slot);
1562 ReleaseBuffer(delbuffer);
1566 /* ----------------------------------------------------------------
1569 * note: we can't run UPDATE queries with transactions
1570 * off because UPDATEs are actually INSERTs and our
1571 * scan will mistakenly loop forever, updating the tuple
1572 * it just inserted.. This should be fixed but until it
1573 * is, we don't want to get stuck in an infinite loop
1574 * which corrupts your database..
1575 * ----------------------------------------------------------------
1578 ExecUpdate(TupleTableSlot *slot,
1579 ItemPointer tupleid,
1580 TupleTableSlot *planSlot,
1585 ResultRelInfo *resultRelInfo;
1586 Relation resultRelationDesc;
1588 ItemPointerData update_ctid;
1589 TransactionId update_xmax;
1592 * abort the operation if not running transactions
1594 if (IsBootstrapProcessingMode())
1595 elog(ERROR, "cannot UPDATE during bootstrap");
1598 * get the heap tuple out of the tuple table slot, making sure we have a
1601 tuple = ExecMaterializeSlot(slot);
1604 * get information on the (current) result relation
1606 resultRelInfo = estate->es_result_relation_info;
1607 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1609 /* BEFORE ROW UPDATE Triggers */
1610 if (resultRelInfo->ri_TrigDesc &&
1611 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
1615 newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
1617 estate->es_snapshot->curcid);
1619 if (newtuple == NULL) /* "do nothing" */
1622 if (newtuple != tuple) /* modified by Trigger(s) */
1625 * Put the modified tuple into a slot for convenience of routines
1626 * below. We assume the tuple was allocated in per-tuple memory
1627 * context, and therefore will go away by itself. The tuple table
1628 * slot should not try to clear it.
1630 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1632 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1633 ExecSetSlotDescriptor(newslot, slot->tts_tupleDescriptor);
1634 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1641 * Check the constraints of the tuple
1643 * If we generate a new candidate tuple after EvalPlanQual testing, we
1644 * must loop back here and recheck constraints. (We don't need to redo
1645 * triggers, however. If there are any BEFORE triggers then trigger.c
1646 * will have done heap_lock_tuple to lock the correct tuple, so there's no
1647 * need to do them again.)
1650 if (resultRelationDesc->rd_att->constr)
1651 ExecConstraints(resultRelInfo, slot, estate);
1654 * replace the heap tuple
1656 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1657 * the row to be updated is visible to that snapshot, and throw a can't-
1658 * serialize error if not. This is a special-case behavior needed for
1659 * referential integrity updates in serializable transactions.
1661 result = heap_update(resultRelationDesc, tupleid, tuple,
1662 &update_ctid, &update_xmax,
1663 estate->es_snapshot->curcid,
1664 estate->es_crosscheck_snapshot,
1665 true /* wait for commit */ );
1668 case HeapTupleSelfUpdated:
1669 /* already deleted by self; nothing to do */
1672 case HeapTupleMayBeUpdated:
1675 case HeapTupleUpdated:
1676 if (IsXactIsoLevelSerializable)
1678 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1679 errmsg("could not serialize access due to concurrent update")));
1680 else if (!ItemPointerEquals(tupleid, &update_ctid))
1682 TupleTableSlot *epqslot;
1684 epqslot = EvalPlanQual(estate,
1685 resultRelInfo->ri_RangeTableIndex,
1688 estate->es_snapshot->curcid);
1689 if (!TupIsNull(epqslot))
1691 *tupleid = update_ctid;
1692 slot = ExecFilterJunk(estate->es_junkFilter, epqslot);
1693 tuple = ExecMaterializeSlot(slot);
1697 /* tuple already deleted; nothing to do */
1701 elog(ERROR, "unrecognized heap_update status: %u", result);
1706 (estate->es_processed)++;
1709 * Note: instead of having to update the old index tuples associated with
1710 * the heap tuple, all we do is form and insert new index tuples. This is
1711 * because UPDATEs are actually DELETEs and INSERTs, and index tuple
1712 * deletion is done later by VACUUM (see notes in ExecDelete). All we do
1713 * here is insert new index tuples. -cim 9/27/89
1717 * insert index entries for tuple
1719 * Note: heap_update returns the tid (location) of the new tuple in the
1722 if (resultRelInfo->ri_NumIndices > 0)
1723 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1725 /* AFTER ROW UPDATE Triggers */
1726 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
1728 /* Process RETURNING if present */
1729 if (resultRelInfo->ri_projectReturning)
1730 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1731 slot, planSlot, dest);
1735 * ExecRelCheck --- check that tuple meets constraints for result relation
1738 ExecRelCheck(ResultRelInfo *resultRelInfo,
1739 TupleTableSlot *slot, EState *estate)
1741 Relation rel = resultRelInfo->ri_RelationDesc;
1742 int ncheck = rel->rd_att->constr->num_check;
1743 ConstrCheck *check = rel->rd_att->constr->check;
1744 ExprContext *econtext;
1745 MemoryContext oldContext;
1750 * If first time through for this result relation, build expression
1751 * nodetrees for rel's constraint expressions. Keep them in the per-query
1752 * memory context so they'll survive throughout the query.
1754 if (resultRelInfo->ri_ConstraintExprs == NULL)
1756 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1757 resultRelInfo->ri_ConstraintExprs =
1758 (List **) palloc(ncheck * sizeof(List *));
1759 for (i = 0; i < ncheck; i++)
1761 /* ExecQual wants implicit-AND form */
1762 qual = make_ands_implicit(stringToNode(check[i].ccbin));
1763 resultRelInfo->ri_ConstraintExprs[i] = (List *)
1764 ExecPrepareExpr((Expr *) qual, estate);
1766 MemoryContextSwitchTo(oldContext);
1770 * We will use the EState's per-tuple context for evaluating constraint
1771 * expressions (creating it if it's not already there).
1773 econtext = GetPerTupleExprContext(estate);
1775 /* Arrange for econtext's scan tuple to be the tuple under test */
1776 econtext->ecxt_scantuple = slot;
1778 /* And evaluate the constraints */
1779 for (i = 0; i < ncheck; i++)
1781 qual = resultRelInfo->ri_ConstraintExprs[i];
1784 * NOTE: SQL92 specifies that a NULL result from a constraint
1785 * expression is not to be treated as a failure. Therefore, tell
1786 * ExecQual to return TRUE for NULL.
1788 if (!ExecQual(qual, econtext, true))
1789 return check[i].ccname;
1792 /* NULL result means no error */
1797 ExecConstraints(ResultRelInfo *resultRelInfo,
1798 TupleTableSlot *slot, EState *estate)
1800 Relation rel = resultRelInfo->ri_RelationDesc;
1801 TupleConstr *constr = rel->rd_att->constr;
1805 if (constr->has_not_null)
1807 int natts = rel->rd_att->natts;
1810 for (attrChk = 1; attrChk <= natts; attrChk++)
1812 if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1813 slot_attisnull(slot, attrChk))
1815 (errcode(ERRCODE_NOT_NULL_VIOLATION),
1816 errmsg("null value in column \"%s\" violates not-null constraint",
1817 NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1821 if (constr->num_check > 0)
1825 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1827 (errcode(ERRCODE_CHECK_VIOLATION),
1828 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1829 RelationGetRelationName(rel), failed)));
1834 * ExecProcessReturning --- evaluate a RETURNING list and send to dest
1836 * projectReturning: RETURNING projection info for current result rel
1837 * tupleSlot: slot holding tuple actually inserted/updated/deleted
1838 * planSlot: slot holding tuple returned by top plan node
1839 * dest: where to send the output
1842 ExecProcessReturning(ProjectionInfo *projectReturning,
1843 TupleTableSlot *tupleSlot,
1844 TupleTableSlot *planSlot,
1847 ExprContext *econtext = projectReturning->pi_exprContext;
1848 TupleTableSlot *retSlot;
1851 * Reset per-tuple memory context to free any expression evaluation
1852 * storage allocated in the previous cycle.
1854 ResetExprContext(econtext);
1856 /* Make tuple and any needed join variables available to ExecProject */
1857 econtext->ecxt_scantuple = tupleSlot;
1858 econtext->ecxt_outertuple = planSlot;
1860 /* Compute the RETURNING expressions */
1861 retSlot = ExecProject(projectReturning, NULL);
1864 (*dest->receiveSlot) (retSlot, dest);
1866 ExecClearTuple(retSlot);
1870 * Check a modified tuple to see if we want to process its updated version
1871 * under READ COMMITTED rules.
1873 * See backend/executor/README for some info about how this works.
1875 * estate - executor state data
1876 * rti - rangetable index of table containing tuple
1877 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1878 * priorXmax - t_xmax from the outdated tuple
1879 * curCid - command ID of current command of my transaction
1881 * *tid is also an output parameter: it's modified to hold the TID of the
1882 * latest version of the tuple (note this may be changed even on failure)
1884 * Returns a slot containing the new candidate update/delete tuple, or
1885 * NULL if we determine we shouldn't process the row.
1888 EvalPlanQual(EState *estate, Index rti,
1889 ItemPointer tid, TransactionId priorXmax, CommandId curCid)
1894 HeapTupleData tuple;
1895 HeapTuple copyTuple = NULL;
1896 SnapshotData SnapshotDirty;
1902 * find relation containing target tuple
1904 if (estate->es_result_relation_info != NULL &&
1905 estate->es_result_relation_info->ri_RangeTableIndex == rti)
1906 relation = estate->es_result_relation_info->ri_RelationDesc;
1912 foreach(l, estate->es_rowMarks)
1914 if (((ExecRowMark *) lfirst(l))->rti == rti)
1916 relation = ((ExecRowMark *) lfirst(l))->relation;
1920 if (relation == NULL)
1921 elog(ERROR, "could not find RowMark for RT index %u", rti);
1927 * Loop here to deal with updated or busy tuples
1929 InitDirtySnapshot(SnapshotDirty);
1930 tuple.t_self = *tid;
1935 if (heap_fetch(relation, &SnapshotDirty, &tuple, &buffer, true, NULL))
1938 * If xmin isn't what we're expecting, the slot must have been
1939 * recycled and reused for an unrelated tuple. This implies that
1940 * the latest version of the row was deleted, so we need do
1941 * nothing. (Should be safe to examine xmin without getting
1942 * buffer's content lock, since xmin never changes in an existing
1945 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1948 ReleaseBuffer(buffer);
1952 /* otherwise xmin should not be dirty... */
1953 if (TransactionIdIsValid(SnapshotDirty.xmin))
1954 elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1957 * If tuple is being updated by other transaction then we have to
1958 * wait for its commit/abort.
1960 if (TransactionIdIsValid(SnapshotDirty.xmax))
1962 ReleaseBuffer(buffer);
1963 XactLockTableWait(SnapshotDirty.xmax);
1964 continue; /* loop back to repeat heap_fetch */
1968 * If tuple was inserted by our own transaction, we have to check
1969 * cmin against curCid: cmin >= curCid means our command cannot
1970 * see the tuple, so we should ignore it. Without this we are
1971 * open to the "Halloween problem" of indefinitely re-updating the
1972 * same tuple. (We need not check cmax because
1973 * HeapTupleSatisfiesDirty will consider a tuple deleted by our
1974 * transaction dead, regardless of cmax.) We just checked that
1975 * priorXmax == xmin, so we can test that variable instead of
1976 * doing HeapTupleHeaderGetXmin again.
1978 if (TransactionIdIsCurrentTransactionId(priorXmax) &&
1979 HeapTupleHeaderGetCmin(tuple.t_data) >= curCid)
1981 ReleaseBuffer(buffer);
1986 * We got tuple - now copy it for use by recheck query.
1988 copyTuple = heap_copytuple(&tuple);
1989 ReleaseBuffer(buffer);
1994 * If the referenced slot was actually empty, the latest version of
1995 * the row must have been deleted, so we need do nothing.
1997 if (tuple.t_data == NULL)
1999 ReleaseBuffer(buffer);
2004 * As above, if xmin isn't what we're expecting, do nothing.
2006 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
2009 ReleaseBuffer(buffer);
2014 * If we get here, the tuple was found but failed SnapshotDirty.
2015 * Assuming the xmin is either a committed xact or our own xact (as it
2016 * certainly should be if we're trying to modify the tuple), this must
2017 * mean that the row was updated or deleted by either a committed xact
2018 * or our own xact. If it was deleted, we can ignore it; if it was
2019 * updated then chain up to the next version and repeat the whole
2022 * As above, it should be safe to examine xmax and t_ctid without the
2023 * buffer content lock, because they can't be changing.
2025 if (ItemPointerEquals(&tuple.t_self, &tuple.t_data->t_ctid))
2027 /* deleted, so forget about it */
2028 ReleaseBuffer(buffer);
2032 /* updated, so look at the updated row */
2033 tuple.t_self = tuple.t_data->t_ctid;
2034 /* updated row should have xmin matching this xmax */
2035 priorXmax = HeapTupleHeaderGetXmax(tuple.t_data);
2036 ReleaseBuffer(buffer);
2037 /* loop back to fetch next in chain */
2041 * For UPDATE/DELETE we have to return tid of actual row we're executing
2044 *tid = tuple.t_self;
2047 * Need to run a recheck subquery. Find or create a PQ stack entry.
2049 epq = estate->es_evalPlanQual;
2052 if (epq != NULL && epq->rti == 0)
2054 /* Top PQ stack entry is idle, so re-use it */
2055 Assert(!(estate->es_useEvalPlan) && epq->next == NULL);
2061 * If this is request for another RTE - Ra, - then we have to check wasn't
2062 * PlanQual requested for Ra already and if so then Ra' row was updated
2063 * again and we have to re-start old execution for Ra and forget all what
2064 * we done after Ra was suspended. Cool? -:))
2066 if (epq != NULL && epq->rti != rti &&
2067 epq->estate->es_evTuple[rti - 1] != NULL)
2071 evalPlanQual *oldepq;
2073 /* stop execution */
2074 EvalPlanQualStop(epq);
2075 /* pop previous PlanQual from the stack */
2077 Assert(oldepq && oldepq->rti != 0);
2078 /* push current PQ to freePQ stack */
2081 estate->es_evalPlanQual = epq;
2082 } while (epq->rti != rti);
2086 * If we are requested for another RTE then we have to suspend execution
2087 * of current PlanQual and start execution for new one.
2089 if (epq == NULL || epq->rti != rti)
2091 /* try to reuse plan used previously */
2092 evalPlanQual *newepq = (epq != NULL) ? epq->free : NULL;
2094 if (newepq == NULL) /* first call or freePQ stack is empty */
2096 newepq = (evalPlanQual *) palloc0(sizeof(evalPlanQual));
2097 newepq->free = NULL;
2098 newepq->estate = NULL;
2099 newepq->planstate = NULL;
2103 /* recycle previously used PlanQual */
2104 Assert(newepq->estate == NULL);
2107 /* push current PQ to the stack */
2110 estate->es_evalPlanQual = epq;
2115 Assert(epq->rti == rti);
2118 * Ok - we're requested for the same RTE. Unfortunately we still have to
2119 * end and restart execution of the plan, because ExecReScan wouldn't
2120 * ensure that upper plan nodes would reset themselves. We could make
2121 * that work if insertion of the target tuple were integrated with the
2122 * Param mechanism somehow, so that the upper plan nodes know that their
2123 * children's outputs have changed.
2125 * Note that the stack of free evalPlanQual nodes is quite useless at the
2126 * moment, since it only saves us from pallocing/releasing the
2127 * evalPlanQual nodes themselves. But it will be useful once we implement
2128 * ReScan instead of end/restart for re-using PlanQual nodes.
2132 /* stop execution */
2133 EvalPlanQualStop(epq);
2137 * Initialize new recheck query.
2139 * Note: if we were re-using PlanQual plans via ExecReScan, we'd need to
2140 * instead copy down changeable state from the top plan (including
2141 * es_result_relation_info, es_junkFilter) and reset locally changeable
2142 * state in the epq (including es_param_exec_vals, es_evTupleNull).
2144 EvalPlanQualStart(epq, estate, epq->next);
2147 * free old RTE' tuple, if any, and store target tuple where relation's
2148 * scan node will see it
2150 epqstate = epq->estate;
2151 if (epqstate->es_evTuple[rti - 1] != NULL)
2152 heap_freetuple(epqstate->es_evTuple[rti - 1]);
2153 epqstate->es_evTuple[rti - 1] = copyTuple;
2155 return EvalPlanQualNext(estate);
2158 static TupleTableSlot *
2159 EvalPlanQualNext(EState *estate)
2161 evalPlanQual *epq = estate->es_evalPlanQual;
2162 MemoryContext oldcontext;
2163 TupleTableSlot *slot;
2165 Assert(epq->rti != 0);
2168 oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt);
2169 slot = ExecProcNode(epq->planstate);
2170 MemoryContextSwitchTo(oldcontext);
2173 * No more tuples for this PQ. Continue previous one.
2175 if (TupIsNull(slot))
2177 evalPlanQual *oldepq;
2179 /* stop execution */
2180 EvalPlanQualStop(epq);
2181 /* pop old PQ from the stack */
2185 /* this is the first (oldest) PQ - mark as free */
2187 estate->es_useEvalPlan = false;
2188 /* and continue Query execution */
2191 Assert(oldepq->rti != 0);
2192 /* push current PQ to freePQ stack */
2195 estate->es_evalPlanQual = epq;
2203 EndEvalPlanQual(EState *estate)
2205 evalPlanQual *epq = estate->es_evalPlanQual;
2207 if (epq->rti == 0) /* plans already shutdowned */
2209 Assert(epq->next == NULL);
2215 evalPlanQual *oldepq;
2217 /* stop execution */
2218 EvalPlanQualStop(epq);
2219 /* pop old PQ from the stack */
2223 /* this is the first (oldest) PQ - mark as free */
2225 estate->es_useEvalPlan = false;
2228 Assert(oldepq->rti != 0);
2229 /* push current PQ to freePQ stack */
2232 estate->es_evalPlanQual = epq;
2237 * Start execution of one level of PlanQual.
2239 * This is a cut-down version of ExecutorStart(): we copy some state from
2240 * the top-level estate rather than initializing it fresh.
2243 EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
2247 MemoryContext oldcontext;
2250 rtsize = list_length(estate->es_range_table);
2252 epq->estate = epqstate = CreateExecutorState();
2254 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2257 * The epqstates share the top query's copy of unchanging state such as
2258 * the snapshot, rangetable, result-rel info, and external Param info.
2259 * They need their own copies of local state, including a tuple table,
2260 * es_param_exec_vals, etc.
2262 epqstate->es_direction = ForwardScanDirection;
2263 epqstate->es_snapshot = estate->es_snapshot;
2264 epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot;
2265 epqstate->es_range_table = estate->es_range_table;
2266 epqstate->es_result_relations = estate->es_result_relations;
2267 epqstate->es_num_result_relations = estate->es_num_result_relations;
2268 epqstate->es_result_relation_info = estate->es_result_relation_info;
2269 epqstate->es_junkFilter = estate->es_junkFilter;
2270 epqstate->es_into_relation_descriptor = estate->es_into_relation_descriptor;
2271 epqstate->es_into_relation_use_wal = estate->es_into_relation_use_wal;
2272 epqstate->es_param_list_info = estate->es_param_list_info;
2273 if (estate->es_plannedstmt->nParamExec > 0)
2274 epqstate->es_param_exec_vals = (ParamExecData *)
2275 palloc0(estate->es_plannedstmt->nParamExec * sizeof(ParamExecData));
2276 epqstate->es_rowMarks = estate->es_rowMarks;
2277 epqstate->es_instrument = estate->es_instrument;
2278 epqstate->es_select_into = estate->es_select_into;
2279 epqstate->es_into_oids = estate->es_into_oids;
2280 epqstate->es_plannedstmt = estate->es_plannedstmt;
2283 * Each epqstate must have its own es_evTupleNull state, but all the stack
2284 * entries share es_evTuple state. This allows sub-rechecks to inherit
2285 * the value being examined by an outer recheck.
2287 epqstate->es_evTupleNull = (bool *) palloc0(rtsize * sizeof(bool));
2288 if (priorepq == NULL)
2289 /* first PQ stack entry */
2290 epqstate->es_evTuple = (HeapTuple *)
2291 palloc0(rtsize * sizeof(HeapTuple));
2293 /* later stack entries share the same storage */
2294 epqstate->es_evTuple = priorepq->estate->es_evTuple;
2297 * Create sub-tuple-table; we needn't redo the CountSlots work though.
2299 epqstate->es_tupleTable =
2300 ExecCreateTupleTable(estate->es_tupleTable->size);
2303 * Initialize private state information for each SubPlan. We must do
2304 * this before running ExecInitNode on the main query tree, since
2305 * ExecInitSubPlan expects to be able to find these entries.
2307 Assert(epqstate->es_subplanstates == NIL);
2308 foreach(l, estate->es_plannedstmt->subplans)
2310 Plan *subplan = (Plan *) lfirst(l);
2311 PlanState *subplanstate;
2313 subplanstate = ExecInitNode(subplan, epqstate, 0);
2315 epqstate->es_subplanstates = lappend(epqstate->es_subplanstates,
2320 * Initialize the private state information for all the nodes in the query
2321 * tree. This opens files, allocates storage and leaves us ready to start
2322 * processing tuples.
2324 epq->planstate = ExecInitNode(estate->es_plannedstmt->planTree, epqstate, 0);
2326 MemoryContextSwitchTo(oldcontext);
2330 * End execution of one level of PlanQual.
2332 * This is a cut-down version of ExecutorEnd(); basically we want to do most
2333 * of the normal cleanup, but *not* close result relations (which we are
2334 * just sharing from the outer query).
2337 EvalPlanQualStop(evalPlanQual *epq)
2339 EState *epqstate = epq->estate;
2340 MemoryContext oldcontext;
2343 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2345 ExecEndNode(epq->planstate);
2347 foreach(l, epqstate->es_subplanstates)
2349 PlanState *subplanstate = (PlanState *) lfirst(l);
2351 ExecEndNode(subplanstate);
2354 ExecDropTupleTable(epqstate->es_tupleTable, true);
2355 epqstate->es_tupleTable = NULL;
2357 if (epqstate->es_evTuple[epq->rti - 1] != NULL)
2359 heap_freetuple(epqstate->es_evTuple[epq->rti - 1]);
2360 epqstate->es_evTuple[epq->rti - 1] = NULL;
2363 MemoryContextSwitchTo(oldcontext);
2365 FreeExecutorState(epqstate);
2368 epq->planstate = NULL;
2373 * Support for SELECT INTO (a/k/a CREATE TABLE AS)
2375 * We implement SELECT INTO by diverting SELECT's normal output with
2376 * a specialized DestReceiver type.
2378 * TODO: remove some of the INTO-specific cruft from EState, and keep
2379 * it in the DestReceiver instead.
2384 DestReceiver pub; /* publicly-known function pointers */
2385 EState *estate; /* EState we are working with */
2389 * OpenIntoRel --- actually create the SELECT INTO target relation
2391 * This also replaces QueryDesc->dest with the special DestReceiver for
2392 * SELECT INTO. We assume that the correct result tuple type has already
2393 * been placed in queryDesc->tupDesc.
2396 OpenIntoRel(QueryDesc *queryDesc)
2398 IntoClause *into = queryDesc->plannedstmt->into;
2399 EState *estate = queryDesc->estate;
2400 Relation intoRelationDesc;
2405 AclResult aclresult;
2408 DR_intorel *myState;
2413 * Check consistency of arguments
2415 if (into->onCommit != ONCOMMIT_NOOP && !into->rel->istemp)
2417 (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
2418 errmsg("ON COMMIT can only be used on temporary tables")));
2421 * Find namespace to create in, check its permissions
2423 intoName = into->rel->relname;
2424 namespaceId = RangeVarGetCreationNamespace(into->rel);
2426 aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
2428 if (aclresult != ACLCHECK_OK)
2429 aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
2430 get_namespace_name(namespaceId));
2433 * Select tablespace to use. If not specified, use default_tablespace
2434 * (which may in turn default to database's default).
2436 if (into->tableSpaceName)
2438 tablespaceId = get_tablespace_oid(into->tableSpaceName);
2439 if (!OidIsValid(tablespaceId))
2441 (errcode(ERRCODE_UNDEFINED_OBJECT),
2442 errmsg("tablespace \"%s\" does not exist",
2443 into->tableSpaceName)));
2447 tablespaceId = GetDefaultTablespace();
2448 /* note InvalidOid is OK in this case */
2451 /* Check permissions except when using the database's default space */
2452 if (OidIsValid(tablespaceId))
2454 AclResult aclresult;
2456 aclresult = pg_tablespace_aclcheck(tablespaceId, GetUserId(),
2459 if (aclresult != ACLCHECK_OK)
2460 aclcheck_error(aclresult, ACL_KIND_TABLESPACE,
2461 get_tablespace_name(tablespaceId));
2464 /* Parse and validate any reloptions */
2465 reloptions = transformRelOptions((Datum) 0,
2469 (void) heap_reloptions(RELKIND_RELATION, reloptions, true);
2471 /* have to copy the actual tupdesc to get rid of any constraints */
2472 tupdesc = CreateTupleDescCopy(queryDesc->tupDesc);
2474 /* Now we can actually create the new relation */
2475 intoRelationId = heap_create_with_catalog(intoName,
2487 allowSystemTableMods);
2489 FreeTupleDesc(tupdesc);
2492 * Advance command counter so that the newly-created relation's catalog
2493 * tuples will be visible to heap_open.
2495 CommandCounterIncrement();
2498 * If necessary, create a TOAST table for the INTO relation. Note that
2499 * AlterTableCreateToastTable ends with CommandCounterIncrement(), so that
2500 * the TOAST table will be visible for insertion.
2502 AlterTableCreateToastTable(intoRelationId);
2505 * And open the constructed table for writing.
2507 intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
2509 /* use_wal off requires rd_targblock be initially invalid */
2510 Assert(intoRelationDesc->rd_targblock == InvalidBlockNumber);
2513 * We can skip WAL-logging the insertions, unless PITR is in use.
2515 * Note that for a non-temp INTO table, this is safe only because we know
2516 * that the catalog changes above will have been WAL-logged, and so
2517 * RecordTransactionCommit will think it needs to WAL-log the eventual
2518 * transaction commit. Else the commit might be lost, even though all the
2519 * data is safely fsync'd ...
2521 estate->es_into_relation_use_wal = XLogArchivingActive();
2522 estate->es_into_relation_descriptor = intoRelationDesc;
2525 * Now replace the query's DestReceiver with one for SELECT INTO
2527 queryDesc->dest = CreateDestReceiver(DestIntoRel, NULL);
2528 myState = (DR_intorel *) queryDesc->dest;
2529 Assert(myState->pub.mydest == DestIntoRel);
2530 myState->estate = estate;
2534 * CloseIntoRel --- clean up SELECT INTO at ExecutorEnd time
2537 CloseIntoRel(QueryDesc *queryDesc)
2539 EState *estate = queryDesc->estate;
2541 /* OpenIntoRel might never have gotten called */
2542 if (estate->es_into_relation_descriptor)
2544 /* If we skipped using WAL, must heap_sync before commit */
2545 if (!estate->es_into_relation_use_wal)
2546 heap_sync(estate->es_into_relation_descriptor);
2548 /* close rel, but keep lock until commit */
2549 heap_close(estate->es_into_relation_descriptor, NoLock);
2551 estate->es_into_relation_descriptor = NULL;
2556 * CreateIntoRelDestReceiver -- create a suitable DestReceiver object
2558 * Since CreateDestReceiver doesn't accept the parameters we'd need,
2559 * we just leave the private fields empty here. OpenIntoRel will
2563 CreateIntoRelDestReceiver(void)
2565 DR_intorel *self = (DR_intorel *) palloc(sizeof(DR_intorel));
2567 self->pub.receiveSlot = intorel_receive;
2568 self->pub.rStartup = intorel_startup;
2569 self->pub.rShutdown = intorel_shutdown;
2570 self->pub.rDestroy = intorel_destroy;
2571 self->pub.mydest = DestIntoRel;
2573 self->estate = NULL;
2575 return (DestReceiver *) self;
2579 * intorel_startup --- executor startup
2582 intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo)
2588 * intorel_receive --- receive one tuple
2591 intorel_receive(TupleTableSlot *slot, DestReceiver *self)
2593 DR_intorel *myState = (DR_intorel *) self;
2594 EState *estate = myState->estate;
2597 tuple = ExecCopySlotTuple(slot);
2599 heap_insert(estate->es_into_relation_descriptor,
2601 estate->es_snapshot->curcid,
2602 estate->es_into_relation_use_wal,
2603 false); /* never any point in using FSM */
2605 /* We know this is a newly created relation, so there are no indexes */
2607 heap_freetuple(tuple);
2613 * intorel_shutdown --- executor end
2616 intorel_shutdown(DestReceiver *self)
2622 * intorel_destroy --- release DestReceiver object
2625 intorel_destroy(DestReceiver *self)