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11 * The old ExecutorMain() has been replaced by ExecutorStart(),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) 1996-2006, PostgreSQL Global Development Group
25 * Portions Copyright (c) 1994, Regents of the University of California
29 * $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.279 2006/08/12 20:05:55 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 TupleTableSlot *ExecutePlan(EState *estate, PlanState *planstate,
76 ScanDirection direction,
78 static void ExecSelect(TupleTableSlot *slot,
79 DestReceiver *dest, EState *estate);
80 static void ExecInsert(TupleTableSlot *slot, ItemPointer tupleid,
81 TupleTableSlot *planSlot,
82 DestReceiver *dest, EState *estate);
83 static void ExecDelete(ItemPointer tupleid,
84 TupleTableSlot *planSlot,
85 DestReceiver *dest, EState *estate);
86 static void ExecUpdate(TupleTableSlot *slot, ItemPointer tupleid,
87 TupleTableSlot *planSlot,
88 DestReceiver *dest, EState *estate);
89 static void ExecProcessReturning(ProjectionInfo *projectReturning,
90 TupleTableSlot *tupleSlot,
91 TupleTableSlot *planSlot,
93 static TupleTableSlot *EvalPlanQualNext(EState *estate);
94 static void EndEvalPlanQual(EState *estate);
95 static void ExecCheckRTEPerms(RangeTblEntry *rte);
96 static void ExecCheckXactReadOnly(Query *parsetree);
97 static void EvalPlanQualStart(evalPlanQual *epq, EState *estate,
98 evalPlanQual *priorepq);
99 static void EvalPlanQualStop(evalPlanQual *epq);
100 static void OpenIntoRel(QueryDesc *queryDesc);
101 static void CloseIntoRel(QueryDesc *queryDesc);
102 static void intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo);
103 static void intorel_receive(TupleTableSlot *slot, DestReceiver *self);
104 static void intorel_shutdown(DestReceiver *self);
105 static void intorel_destroy(DestReceiver *self);
107 /* end of local decls */
110 /* ----------------------------------------------------------------
113 * This routine must be called at the beginning of any execution of any
116 * Takes a QueryDesc previously created by CreateQueryDesc (it's not real
117 * clear why we bother to separate the two functions, but...). The tupDesc
118 * field of the QueryDesc is filled in to describe the tuples that will be
119 * returned, and the internal fields (estate and planstate) are set up.
121 * eflags contains flag bits as described in executor.h.
123 * NB: the CurrentMemoryContext when this is called will become the parent
124 * of the per-query context used for this Executor invocation.
125 * ----------------------------------------------------------------
128 ExecutorStart(QueryDesc *queryDesc, int eflags)
131 MemoryContext oldcontext;
133 /* sanity checks: queryDesc must not be started already */
134 Assert(queryDesc != NULL);
135 Assert(queryDesc->estate == NULL);
138 * If the transaction is read-only, we need to check if any writes are
139 * planned to non-temporary tables. EXPLAIN is considered read-only.
141 if (XactReadOnly && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
142 ExecCheckXactReadOnly(queryDesc->parsetree);
145 * Build EState, switch into per-query memory context for startup.
147 estate = CreateExecutorState();
148 queryDesc->estate = estate;
150 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
153 * Fill in parameters, if any, from queryDesc
155 estate->es_param_list_info = queryDesc->params;
157 if (queryDesc->plantree->nParamExec > 0)
158 estate->es_param_exec_vals = (ParamExecData *)
159 palloc0(queryDesc->plantree->nParamExec * sizeof(ParamExecData));
162 * Copy other important information into the EState
164 estate->es_snapshot = queryDesc->snapshot;
165 estate->es_crosscheck_snapshot = queryDesc->crosscheck_snapshot;
166 estate->es_instrument = queryDesc->doInstrument;
169 * Initialize the plan state tree
171 InitPlan(queryDesc, eflags);
173 MemoryContextSwitchTo(oldcontext);
176 /* ----------------------------------------------------------------
179 * This is the main routine of the executor module. It accepts
180 * the query descriptor from the traffic cop and executes the
183 * ExecutorStart must have been called already.
185 * If direction is NoMovementScanDirection then nothing is done
186 * except to start up/shut down the destination. Otherwise,
187 * we retrieve up to 'count' tuples in the specified direction.
189 * Note: count = 0 is interpreted as no portal limit, i.e., run to
192 * ----------------------------------------------------------------
195 ExecutorRun(QueryDesc *queryDesc,
196 ScanDirection direction, long count)
202 TupleTableSlot *result;
203 MemoryContext oldcontext;
206 Assert(queryDesc != NULL);
208 estate = queryDesc->estate;
210 Assert(estate != NULL);
213 * Switch into per-query memory context
215 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
218 * extract information from the query descriptor and the query feature.
220 operation = queryDesc->operation;
221 dest = queryDesc->dest;
224 * startup tuple receiver, if we will be emitting tuples
226 estate->es_processed = 0;
227 estate->es_lastoid = InvalidOid;
229 sendTuples = (operation == CMD_SELECT ||
230 queryDesc->parsetree->returningList);
233 (*dest->rStartup) (dest, operation, queryDesc->tupDesc);
238 if (ScanDirectionIsNoMovement(direction))
241 result = ExecutePlan(estate,
242 queryDesc->planstate,
249 * shutdown tuple receiver, if we started it
252 (*dest->rShutdown) (dest);
254 MemoryContextSwitchTo(oldcontext);
259 /* ----------------------------------------------------------------
262 * This routine must be called at the end of execution of any
264 * ----------------------------------------------------------------
267 ExecutorEnd(QueryDesc *queryDesc)
270 MemoryContext oldcontext;
273 Assert(queryDesc != NULL);
275 estate = queryDesc->estate;
277 Assert(estate != NULL);
280 * Switch into per-query memory context to run ExecEndPlan
282 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
284 ExecEndPlan(queryDesc->planstate, estate);
287 * Close the SELECT INTO relation if any
289 if (estate->es_select_into)
290 CloseIntoRel(queryDesc);
293 * Must switch out of context before destroying it
295 MemoryContextSwitchTo(oldcontext);
298 * Release EState and per-query memory context. This should release
299 * everything the executor has allocated.
301 FreeExecutorState(estate);
303 /* Reset queryDesc fields that no longer point to anything */
304 queryDesc->tupDesc = NULL;
305 queryDesc->estate = NULL;
306 queryDesc->planstate = NULL;
309 /* ----------------------------------------------------------------
312 * This routine may be called on an open queryDesc to rewind it
314 * ----------------------------------------------------------------
317 ExecutorRewind(QueryDesc *queryDesc)
320 MemoryContext oldcontext;
323 Assert(queryDesc != NULL);
325 estate = queryDesc->estate;
327 Assert(estate != NULL);
329 /* It's probably not sensible to rescan updating queries */
330 Assert(queryDesc->operation == CMD_SELECT);
333 * Switch into per-query memory context
335 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
340 ExecReScan(queryDesc->planstate, NULL);
342 MemoryContextSwitchTo(oldcontext);
348 * Check access permissions for all relations listed in a range table.
351 ExecCheckRTPerms(List *rangeTable)
355 foreach(l, rangeTable)
357 RangeTblEntry *rte = lfirst(l);
359 ExecCheckRTEPerms(rte);
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. Subquery RTEs are
376 * checked by ExecInitSubqueryScan if the subquery is still a separate
377 * subquery --- if it's been pulled up into our query level then the RTEs
378 * are in our rangetable and will be checked here. Function RTEs are
379 * checked by init_fcache when the function is prepared for execution.
380 * Join and special RTEs need no checks.
382 if (rte->rtekind != RTE_RELATION)
386 * No work if requiredPerms is empty.
388 requiredPerms = rte->requiredPerms;
389 if (requiredPerms == 0)
395 * userid to check as: current user unless we have a setuid indication.
397 * Note: GetUserId() is presently fast enough that there's no harm in
398 * calling it separately for each RTE. If that stops being true, we could
399 * call it once in ExecCheckRTPerms and pass the userid down from there.
400 * But for now, no need for the extra clutter.
402 userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
405 * We must have *all* the requiredPerms bits, so use aclmask not aclcheck.
407 if (pg_class_aclmask(relOid, userid, requiredPerms, ACLMASK_ALL)
409 aclcheck_error(ACLCHECK_NO_PRIV, ACL_KIND_CLASS,
410 get_rel_name(relOid));
414 * Check that the query does not imply any writes to non-temp tables.
417 ExecCheckXactReadOnly(Query *parsetree)
422 * CREATE TABLE AS or SELECT INTO?
424 * XXX should we allow this if the destination is temp?
426 if (parsetree->into != NULL)
429 /* Fail if write permissions are requested on any non-temp table */
430 foreach(l, parsetree->rtable)
432 RangeTblEntry *rte = lfirst(l);
434 if (rte->rtekind == RTE_SUBQUERY)
436 ExecCheckXactReadOnly(rte->subquery);
440 if (rte->rtekind != RTE_RELATION)
443 if ((rte->requiredPerms & (~ACL_SELECT)) == 0)
446 if (isTempNamespace(get_rel_namespace(rte->relid)))
456 (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
457 errmsg("transaction is read-only")));
461 /* ----------------------------------------------------------------
464 * Initializes the query plan: open files, allocate storage
465 * and start up the rule manager
466 * ----------------------------------------------------------------
469 InitPlan(QueryDesc *queryDesc, int eflags)
471 CmdType operation = queryDesc->operation;
472 Query *parseTree = queryDesc->parsetree;
473 Plan *plan = queryDesc->plantree;
474 EState *estate = queryDesc->estate;
475 PlanState *planstate;
481 * Do permissions checks. It's sufficient to examine the query's top
482 * rangetable here --- subplan RTEs will be checked during
485 ExecCheckRTPerms(parseTree->rtable);
488 * get information from query descriptor
490 rangeTable = parseTree->rtable;
493 * initialize the node's execution state
495 estate->es_range_table = rangeTable;
498 * if there is a result relation, initialize result relation stuff
500 if (parseTree->resultRelation)
502 List *resultRelations = parseTree->resultRelations;
503 int numResultRelations;
504 ResultRelInfo *resultRelInfos;
506 if (resultRelations != NIL)
509 * Multiple result relations (due to inheritance)
510 * parseTree->resultRelations identifies them all
512 ResultRelInfo *resultRelInfo;
514 numResultRelations = list_length(resultRelations);
515 resultRelInfos = (ResultRelInfo *)
516 palloc(numResultRelations * sizeof(ResultRelInfo));
517 resultRelInfo = resultRelInfos;
518 foreach(l, resultRelations)
520 initResultRelInfo(resultRelInfo,
524 estate->es_instrument);
531 * Single result relation identified by parseTree->resultRelation
533 numResultRelations = 1;
534 resultRelInfos = (ResultRelInfo *) palloc(sizeof(ResultRelInfo));
535 initResultRelInfo(resultRelInfos,
536 parseTree->resultRelation,
539 estate->es_instrument);
542 estate->es_result_relations = resultRelInfos;
543 estate->es_num_result_relations = numResultRelations;
544 /* Initialize to first or only result rel */
545 estate->es_result_relation_info = resultRelInfos;
550 * if no result relation, then set state appropriately
552 estate->es_result_relations = NULL;
553 estate->es_num_result_relations = 0;
554 estate->es_result_relation_info = NULL;
558 * Detect whether we're doing SELECT INTO. If so, set the es_into_oids
559 * flag appropriately so that the plan tree will be initialized with the
560 * correct tuple descriptors. (Other SELECT INTO stuff comes later.)
562 estate->es_select_into = false;
563 if (operation == CMD_SELECT && parseTree->into != NULL)
565 estate->es_select_into = true;
566 estate->es_into_oids = interpretOidsOption(parseTree->intoOptions);
570 * Have to lock relations selected FOR UPDATE/FOR SHARE
572 estate->es_rowMarks = NIL;
573 foreach(l, parseTree->rowMarks)
575 RowMarkClause *rc = (RowMarkClause *) lfirst(l);
576 Oid relid = getrelid(rc->rti, rangeTable);
580 relation = heap_open(relid, RowShareLock);
581 erm = (ExecRowMark *) palloc(sizeof(ExecRowMark));
582 erm->relation = relation;
584 erm->forUpdate = rc->forUpdate;
585 erm->noWait = rc->noWait;
586 snprintf(erm->resname, sizeof(erm->resname), "ctid%u", rc->rti);
587 estate->es_rowMarks = lappend(estate->es_rowMarks, erm);
591 * initialize the executor "tuple" table. We need slots for all the plan
592 * nodes, plus possibly output slots for the junkfilter(s). At this point
593 * we aren't sure if we need junkfilters, so just add slots for them
594 * unconditionally. Also, if it's not a SELECT, set up a slot for use for
595 * trigger output tuples.
598 int nSlots = ExecCountSlotsNode(plan);
600 if (parseTree->resultRelations != NIL)
601 nSlots += list_length(parseTree->resultRelations);
604 if (operation != CMD_SELECT)
605 nSlots++; /* for es_trig_tuple_slot */
606 if (parseTree->returningLists)
607 nSlots++; /* for RETURNING projection */
609 estate->es_tupleTable = ExecCreateTupleTable(nSlots);
611 if (operation != CMD_SELECT)
612 estate->es_trig_tuple_slot =
613 ExecAllocTableSlot(estate->es_tupleTable);
616 /* mark EvalPlanQual not active */
617 estate->es_topPlan = plan;
618 estate->es_evalPlanQual = NULL;
619 estate->es_evTupleNull = NULL;
620 estate->es_evTuple = NULL;
621 estate->es_useEvalPlan = false;
624 * initialize the private state information for all the nodes in the query
625 * tree. This opens files, allocates storage and leaves us ready to start
628 planstate = ExecInitNode(plan, estate, eflags);
631 * Get the tuple descriptor describing the type of tuples to return. (this
632 * is especially important if we are creating a relation with "SELECT
635 tupType = ExecGetResultType(planstate);
638 * Initialize the junk filter if needed. SELECT and INSERT queries need a
639 * filter if there are any junk attrs in the tlist. INSERT and SELECT
640 * INTO also need a filter if the plan may return raw disk tuples (else
641 * heap_insert will be scribbling on the source relation!). UPDATE and
642 * DELETE always need a filter, since there's always a junk 'ctid'
643 * attribute present --- no need to look first.
646 bool junk_filter_needed = false;
653 foreach(tlist, plan->targetlist)
655 TargetEntry *tle = (TargetEntry *) lfirst(tlist);
659 junk_filter_needed = true;
663 if (!junk_filter_needed &&
664 (operation == CMD_INSERT || estate->es_select_into) &&
665 ExecMayReturnRawTuples(planstate))
666 junk_filter_needed = true;
670 junk_filter_needed = true;
676 if (junk_filter_needed)
679 * If there are multiple result relations, each one needs its own
680 * junk filter. Note this is only possible for UPDATE/DELETE, so
681 * we can't be fooled by some needing a filter and some not.
683 if (parseTree->resultRelations != NIL)
685 PlanState **appendplans;
687 ResultRelInfo *resultRelInfo;
690 /* Top plan had better be an Append here. */
691 Assert(IsA(plan, Append));
692 Assert(((Append *) plan)->isTarget);
693 Assert(IsA(planstate, AppendState));
694 appendplans = ((AppendState *) planstate)->appendplans;
695 as_nplans = ((AppendState *) planstate)->as_nplans;
696 Assert(as_nplans == estate->es_num_result_relations);
697 resultRelInfo = estate->es_result_relations;
698 for (i = 0; i < as_nplans; i++)
700 PlanState *subplan = appendplans[i];
703 j = ExecInitJunkFilter(subplan->plan->targetlist,
704 resultRelInfo->ri_RelationDesc->rd_att->tdhasoid,
705 ExecAllocTableSlot(estate->es_tupleTable));
706 resultRelInfo->ri_junkFilter = j;
711 * Set active junkfilter too; at this point ExecInitAppend has
712 * already selected an active result relation...
714 estate->es_junkFilter =
715 estate->es_result_relation_info->ri_junkFilter;
719 /* Normal case with just one JunkFilter */
722 j = ExecInitJunkFilter(planstate->plan->targetlist,
724 ExecAllocTableSlot(estate->es_tupleTable));
725 estate->es_junkFilter = j;
726 if (estate->es_result_relation_info)
727 estate->es_result_relation_info->ri_junkFilter = j;
729 /* For SELECT, want to return the cleaned tuple type */
730 if (operation == CMD_SELECT)
731 tupType = j->jf_cleanTupType;
735 estate->es_junkFilter = NULL;
739 * Initialize RETURNING projections if needed.
741 if (parseTree->returningLists)
743 TupleTableSlot *slot;
744 ExprContext *econtext;
745 ResultRelInfo *resultRelInfo;
748 * We set QueryDesc.tupDesc to be the RETURNING rowtype in this case.
749 * We assume all the sublists will generate the same output tupdesc.
751 tupType = ExecTypeFromTL((List *) linitial(parseTree->returningLists),
754 /* Set up a slot for the output of the RETURNING projection(s) */
755 slot = ExecAllocTableSlot(estate->es_tupleTable);
756 ExecSetSlotDescriptor(slot, tupType);
757 /* Need an econtext too */
758 econtext = CreateExprContext(estate);
761 * Build a projection for each result rel. Note that any SubPlans
762 * in the RETURNING lists get attached to the topmost plan node.
764 Assert(list_length(parseTree->returningLists) == estate->es_num_result_relations);
765 resultRelInfo = estate->es_result_relations;
766 foreach(l, parseTree->returningLists)
768 List *rlist = (List *) lfirst(l);
771 rliststate = (List *) ExecInitExpr((Expr *) rlist, planstate);
772 resultRelInfo->ri_projectReturning =
773 ExecBuildProjectionInfo(rliststate, econtext, slot);
777 * Because we already ran ExecInitNode() for the top plan node,
778 * any subplans we just attached to it won't have been initialized;
779 * so we have to do it here. (Ugly, but the alternatives seem worse.)
781 foreach(l, planstate->subPlan)
783 SubPlanState *sstate = (SubPlanState *) lfirst(l);
785 Assert(IsA(sstate, SubPlanState));
786 if (sstate->planstate == NULL) /* already inited? */
787 ExecInitSubPlan(sstate, estate, eflags);
791 queryDesc->tupDesc = tupType;
792 queryDesc->planstate = planstate;
795 * If doing SELECT INTO, initialize the "into" relation. We must wait
796 * till now so we have the "clean" result tuple type to create the new
799 * If EXPLAIN, skip creating the "into" relation.
801 if (estate->es_select_into && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
802 OpenIntoRel(queryDesc);
806 * Initialize ResultRelInfo data for one result relation
809 initResultRelInfo(ResultRelInfo *resultRelInfo,
810 Index resultRelationIndex,
815 Oid resultRelationOid;
816 Relation resultRelationDesc;
818 resultRelationOid = getrelid(resultRelationIndex, rangeTable);
819 resultRelationDesc = heap_open(resultRelationOid, RowExclusiveLock);
821 switch (resultRelationDesc->rd_rel->relkind)
823 case RELKIND_SEQUENCE:
825 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
826 errmsg("cannot change sequence \"%s\"",
827 RelationGetRelationName(resultRelationDesc))));
829 case RELKIND_TOASTVALUE:
831 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
832 errmsg("cannot change TOAST relation \"%s\"",
833 RelationGetRelationName(resultRelationDesc))));
837 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
838 errmsg("cannot change view \"%s\"",
839 RelationGetRelationName(resultRelationDesc))));
843 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
844 resultRelInfo->type = T_ResultRelInfo;
845 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
846 resultRelInfo->ri_RelationDesc = resultRelationDesc;
847 resultRelInfo->ri_NumIndices = 0;
848 resultRelInfo->ri_IndexRelationDescs = NULL;
849 resultRelInfo->ri_IndexRelationInfo = NULL;
850 /* make a copy so as not to depend on relcache info not changing... */
851 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
852 if (resultRelInfo->ri_TrigDesc)
854 int n = resultRelInfo->ri_TrigDesc->numtriggers;
856 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
857 palloc0(n * sizeof(FmgrInfo));
859 resultRelInfo->ri_TrigInstrument = InstrAlloc(n);
861 resultRelInfo->ri_TrigInstrument = NULL;
865 resultRelInfo->ri_TrigFunctions = NULL;
866 resultRelInfo->ri_TrigInstrument = NULL;
868 resultRelInfo->ri_ConstraintExprs = NULL;
869 resultRelInfo->ri_junkFilter = NULL;
870 resultRelInfo->ri_projectReturning = NULL;
873 * If there are indices on the result relation, open them and save
874 * descriptors in the result relation info, so that we can add new index
875 * entries for the tuples we add/update. We need not do this for a
876 * DELETE, however, since deletion doesn't affect indexes.
878 if (resultRelationDesc->rd_rel->relhasindex &&
879 operation != CMD_DELETE)
880 ExecOpenIndices(resultRelInfo);
884 * ExecContextForcesOids
886 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
887 * we need to ensure that result tuples have space for an OID iff they are
888 * going to be stored into a relation that has OIDs. In other contexts
889 * we are free to choose whether to leave space for OIDs in result tuples
890 * (we generally don't want to, but we do if a physical-tlist optimization
891 * is possible). This routine checks the plan context and returns TRUE if the
892 * choice is forced, FALSE if the choice is not forced. In the TRUE case,
893 * *hasoids is set to the required value.
895 * One reason this is ugly is that all plan nodes in the plan tree will emit
896 * tuples with space for an OID, though we really only need the topmost node
897 * to do so. However, node types like Sort don't project new tuples but just
898 * return their inputs, and in those cases the requirement propagates down
899 * to the input node. Eventually we might make this code smart enough to
900 * recognize how far down the requirement really goes, but for now we just
901 * make all plan nodes do the same thing if the top level forces the choice.
903 * We assume that estate->es_result_relation_info is already set up to
904 * describe the target relation. Note that in an UPDATE that spans an
905 * inheritance tree, some of the target relations may have OIDs and some not.
906 * We have to make the decisions on a per-relation basis as we initialize
907 * each of the child plans of the topmost Append plan.
909 * SELECT INTO is even uglier, because we don't have the INTO relation's
910 * descriptor available when this code runs; we have to look aside at a
911 * flag set by InitPlan().
914 ExecContextForcesOids(PlanState *planstate, bool *hasoids)
916 if (planstate->state->es_select_into)
918 *hasoids = planstate->state->es_into_oids;
923 ResultRelInfo *ri = planstate->state->es_result_relation_info;
927 Relation rel = ri->ri_RelationDesc;
931 *hasoids = rel->rd_rel->relhasoids;
940 /* ----------------------------------------------------------------
943 * Cleans up the query plan -- closes files and frees up storage
945 * NOTE: we are no longer very worried about freeing storage per se
946 * in this code; FreeExecutorState should be guaranteed to release all
947 * memory that needs to be released. What we are worried about doing
948 * is closing relations and dropping buffer pins. Thus, for example,
949 * tuple tables must be cleared or dropped to ensure pins are released.
950 * ----------------------------------------------------------------
953 ExecEndPlan(PlanState *planstate, EState *estate)
955 ResultRelInfo *resultRelInfo;
960 * shut down any PlanQual processing we were doing
962 if (estate->es_evalPlanQual != NULL)
963 EndEvalPlanQual(estate);
966 * shut down the node-type-specific query processing
968 ExecEndNode(planstate);
971 * destroy the executor "tuple" table.
973 ExecDropTupleTable(estate->es_tupleTable, true);
974 estate->es_tupleTable = NULL;
977 * close the result relation(s) if any, but hold locks until xact commit.
979 resultRelInfo = estate->es_result_relations;
980 for (i = estate->es_num_result_relations; i > 0; i--)
982 /* Close indices and then the relation itself */
983 ExecCloseIndices(resultRelInfo);
984 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
989 * close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
991 foreach(l, estate->es_rowMarks)
993 ExecRowMark *erm = lfirst(l);
995 heap_close(erm->relation, NoLock);
999 /* ----------------------------------------------------------------
1002 * processes the query plan to retrieve 'numberTuples' tuples in the
1003 * direction specified.
1005 * Retrieves all tuples if numberTuples is 0
1007 * result is either a slot containing the last tuple in the case
1008 * of a SELECT or NULL otherwise.
1010 * Note: the ctid attribute is a 'junk' attribute that is removed before the
1012 * ----------------------------------------------------------------
1014 static TupleTableSlot *
1015 ExecutePlan(EState *estate,
1016 PlanState *planstate,
1019 ScanDirection direction,
1022 JunkFilter *junkfilter;
1023 TupleTableSlot *planSlot;
1024 TupleTableSlot *slot;
1025 ItemPointer tupleid = NULL;
1026 ItemPointerData tuple_ctid;
1027 long current_tuple_count;
1028 TupleTableSlot *result;
1031 * initialize local variables
1033 current_tuple_count = 0;
1037 * Set the direction.
1039 estate->es_direction = direction;
1042 * Process BEFORE EACH STATEMENT triggers
1047 ExecBSUpdateTriggers(estate, estate->es_result_relation_info);
1050 ExecBSDeleteTriggers(estate, estate->es_result_relation_info);
1053 ExecBSInsertTriggers(estate, estate->es_result_relation_info);
1061 * Loop until we've processed the proper number of tuples from the plan.
1066 /* Reset the per-output-tuple exprcontext */
1067 ResetPerTupleExprContext(estate);
1070 * Execute the plan and obtain a tuple
1073 if (estate->es_useEvalPlan)
1075 planSlot = EvalPlanQualNext(estate);
1076 if (TupIsNull(planSlot))
1077 planSlot = ExecProcNode(planstate);
1080 planSlot = ExecProcNode(planstate);
1083 * if the tuple is null, then we assume there is nothing more to
1084 * process so we just return null...
1086 if (TupIsNull(planSlot))
1094 * if we have a junk filter, then project a new tuple with the junk
1097 * Store this new "clean" tuple in the junkfilter's resultSlot.
1098 * (Formerly, we stored it back over the "dirty" tuple, which is WRONG
1099 * because that tuple slot has the wrong descriptor.)
1101 * Also, extract all the junk information we need.
1103 if ((junkfilter = estate->es_junkFilter) != NULL)
1109 * extract the 'ctid' junk attribute.
1111 if (operation == CMD_UPDATE || operation == CMD_DELETE)
1113 if (!ExecGetJunkAttribute(junkfilter,
1118 elog(ERROR, "could not find junk ctid column");
1120 /* shouldn't ever get a null result... */
1122 elog(ERROR, "ctid is NULL");
1124 tupleid = (ItemPointer) DatumGetPointer(datum);
1125 tuple_ctid = *tupleid; /* make sure we don't free the ctid!! */
1126 tupleid = &tuple_ctid;
1130 * Process any FOR UPDATE or FOR SHARE locking requested.
1132 else if (estate->es_rowMarks != NIL)
1137 foreach(l, estate->es_rowMarks)
1139 ExecRowMark *erm = lfirst(l);
1140 HeapTupleData tuple;
1142 ItemPointerData update_ctid;
1143 TransactionId update_xmax;
1144 TupleTableSlot *newSlot;
1145 LockTupleMode lockmode;
1148 if (!ExecGetJunkAttribute(junkfilter,
1153 elog(ERROR, "could not find junk \"%s\" column",
1156 /* shouldn't ever get a null result... */
1158 elog(ERROR, "\"%s\" is NULL", erm->resname);
1160 tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1163 lockmode = LockTupleExclusive;
1165 lockmode = LockTupleShared;
1167 test = heap_lock_tuple(erm->relation, &tuple, &buffer,
1168 &update_ctid, &update_xmax,
1169 estate->es_snapshot->curcid,
1170 lockmode, erm->noWait);
1171 ReleaseBuffer(buffer);
1174 case HeapTupleSelfUpdated:
1175 /* treat it as deleted; do not process */
1178 case HeapTupleMayBeUpdated:
1181 case HeapTupleUpdated:
1182 if (IsXactIsoLevelSerializable)
1184 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1185 errmsg("could not serialize access due to concurrent update")));
1186 if (!ItemPointerEquals(&update_ctid,
1189 /* updated, so look at updated version */
1190 newSlot = EvalPlanQual(estate,
1194 estate->es_snapshot->curcid);
1195 if (!TupIsNull(newSlot))
1197 slot = planSlot = newSlot;
1198 estate->es_useEvalPlan = true;
1204 * if tuple was deleted or PlanQual failed for
1205 * updated tuple - we must not return this tuple!
1210 elog(ERROR, "unrecognized heap_lock_tuple status: %u",
1218 * Create a new "clean" tuple with all junk attributes removed.
1219 * We don't need to do this for DELETE, however (there will
1220 * in fact be no non-junk attributes in a DELETE!)
1222 if (operation != CMD_DELETE)
1223 slot = ExecFilterJunk(junkfilter, slot);
1227 * now that we have a tuple, do the appropriate thing with it.. either
1228 * return it to the user, add it to a relation someplace, delete it
1229 * from a relation, or modify some of its attributes.
1234 ExecSelect(slot, dest, estate);
1239 ExecInsert(slot, tupleid, planSlot, dest, estate);
1244 ExecDelete(tupleid, planSlot, dest, estate);
1249 ExecUpdate(slot, tupleid, planSlot, dest, estate);
1254 elog(ERROR, "unrecognized operation code: %d",
1261 * check our tuple count.. if we've processed the proper number then
1262 * quit, else loop again and process more tuples. Zero numberTuples
1265 current_tuple_count++;
1266 if (numberTuples && numberTuples == current_tuple_count)
1271 * Process AFTER EACH STATEMENT triggers
1276 ExecASUpdateTriggers(estate, estate->es_result_relation_info);
1279 ExecASDeleteTriggers(estate, estate->es_result_relation_info);
1282 ExecASInsertTriggers(estate, estate->es_result_relation_info);
1290 * here, result is either a slot containing a tuple in the case of a
1291 * SELECT or NULL otherwise.
1296 /* ----------------------------------------------------------------
1299 * SELECTs are easy.. we just pass the tuple to the appropriate
1301 * ----------------------------------------------------------------
1304 ExecSelect(TupleTableSlot *slot,
1308 (*dest->receiveSlot) (slot, dest);
1310 (estate->es_processed)++;
1313 /* ----------------------------------------------------------------
1316 * INSERTs are trickier.. we have to insert the tuple into
1317 * the base relation and insert appropriate tuples into the
1319 * ----------------------------------------------------------------
1322 ExecInsert(TupleTableSlot *slot,
1323 ItemPointer tupleid,
1324 TupleTableSlot *planSlot,
1329 ResultRelInfo *resultRelInfo;
1330 Relation resultRelationDesc;
1334 * get the heap tuple out of the tuple table slot, making sure we have a
1337 tuple = ExecMaterializeSlot(slot);
1340 * get information on the (current) result relation
1342 resultRelInfo = estate->es_result_relation_info;
1343 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1345 /* BEFORE ROW INSERT Triggers */
1346 if (resultRelInfo->ri_TrigDesc &&
1347 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
1351 newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
1353 if (newtuple == NULL) /* "do nothing" */
1356 if (newtuple != tuple) /* modified by Trigger(s) */
1359 * Put the modified tuple into a slot for convenience of routines
1360 * below. We assume the tuple was allocated in per-tuple memory
1361 * context, and therefore will go away by itself. The tuple table
1362 * slot should not try to clear it.
1364 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1366 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1367 ExecSetSlotDescriptor(newslot, slot->tts_tupleDescriptor);
1368 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1375 * Check the constraints of the tuple
1377 if (resultRelationDesc->rd_att->constr)
1378 ExecConstraints(resultRelInfo, slot, estate);
1383 * Note: heap_insert returns the tid (location) of the new tuple in the
1386 newId = heap_insert(resultRelationDesc, tuple,
1387 estate->es_snapshot->curcid,
1391 (estate->es_processed)++;
1392 estate->es_lastoid = newId;
1393 setLastTid(&(tuple->t_self));
1396 * insert index entries for tuple
1398 if (resultRelInfo->ri_NumIndices > 0)
1399 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1401 /* AFTER ROW INSERT Triggers */
1402 ExecARInsertTriggers(estate, resultRelInfo, tuple);
1404 /* Process RETURNING if present */
1405 if (resultRelInfo->ri_projectReturning)
1406 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1407 slot, planSlot, dest);
1410 /* ----------------------------------------------------------------
1413 * DELETE is like UPDATE, except that we delete the tuple and no
1414 * index modifications are needed
1415 * ----------------------------------------------------------------
1418 ExecDelete(ItemPointer tupleid,
1419 TupleTableSlot *planSlot,
1423 ResultRelInfo *resultRelInfo;
1424 Relation resultRelationDesc;
1426 ItemPointerData update_ctid;
1427 TransactionId update_xmax;
1430 * get information on the (current) result relation
1432 resultRelInfo = estate->es_result_relation_info;
1433 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1435 /* BEFORE ROW DELETE Triggers */
1436 if (resultRelInfo->ri_TrigDesc &&
1437 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
1441 dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
1442 estate->es_snapshot->curcid);
1444 if (!dodelete) /* "do nothing" */
1451 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1452 * the row to be deleted is visible to that snapshot, and throw a can't-
1453 * serialize error if not. This is a special-case behavior needed for
1454 * referential integrity updates in serializable transactions.
1457 result = heap_delete(resultRelationDesc, tupleid,
1458 &update_ctid, &update_xmax,
1459 estate->es_snapshot->curcid,
1460 estate->es_crosscheck_snapshot,
1461 true /* wait for commit */ );
1464 case HeapTupleSelfUpdated:
1465 /* already deleted by self; nothing to do */
1468 case HeapTupleMayBeUpdated:
1471 case HeapTupleUpdated:
1472 if (IsXactIsoLevelSerializable)
1474 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1475 errmsg("could not serialize access due to concurrent update")));
1476 else if (!ItemPointerEquals(tupleid, &update_ctid))
1478 TupleTableSlot *epqslot;
1480 epqslot = EvalPlanQual(estate,
1481 resultRelInfo->ri_RangeTableIndex,
1484 estate->es_snapshot->curcid);
1485 if (!TupIsNull(epqslot))
1487 *tupleid = update_ctid;
1491 /* tuple already deleted; nothing to do */
1495 elog(ERROR, "unrecognized heap_delete status: %u", result);
1500 (estate->es_processed)++;
1503 * Note: Normally one would think that we have to delete index tuples
1504 * associated with the heap tuple now...
1506 * ... but in POSTGRES, we have no need to do this because VACUUM will
1507 * take care of it later. We can't delete index tuples immediately
1508 * anyway, since the tuple is still visible to other transactions.
1511 /* AFTER ROW DELETE Triggers */
1512 ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
1514 /* Process RETURNING if present */
1515 if (resultRelInfo->ri_projectReturning)
1518 * We have to put the target tuple into a slot, which means
1519 * first we gotta fetch it. We can use the trigger tuple slot.
1521 TupleTableSlot *slot = estate->es_trig_tuple_slot;
1522 HeapTupleData deltuple;
1525 deltuple.t_self = *tupleid;
1526 if (!heap_fetch(resultRelationDesc, SnapshotAny,
1527 &deltuple, &delbuffer, false, NULL))
1528 elog(ERROR, "failed to fetch deleted tuple for DELETE RETURNING");
1530 if (slot->tts_tupleDescriptor != RelationGetDescr(resultRelationDesc))
1531 ExecSetSlotDescriptor(slot, RelationGetDescr(resultRelationDesc));
1532 ExecStoreTuple(&deltuple, slot, InvalidBuffer, false);
1534 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1535 slot, planSlot, dest);
1537 ExecClearTuple(slot);
1538 ReleaseBuffer(delbuffer);
1542 /* ----------------------------------------------------------------
1545 * note: we can't run UPDATE queries with transactions
1546 * off because UPDATEs are actually INSERTs and our
1547 * scan will mistakenly loop forever, updating the tuple
1548 * it just inserted.. This should be fixed but until it
1549 * is, we don't want to get stuck in an infinite loop
1550 * which corrupts your database..
1551 * ----------------------------------------------------------------
1554 ExecUpdate(TupleTableSlot *slot,
1555 ItemPointer tupleid,
1556 TupleTableSlot *planSlot,
1561 ResultRelInfo *resultRelInfo;
1562 Relation resultRelationDesc;
1564 ItemPointerData update_ctid;
1565 TransactionId update_xmax;
1568 * abort the operation if not running transactions
1570 if (IsBootstrapProcessingMode())
1571 elog(ERROR, "cannot UPDATE during bootstrap");
1574 * get the heap tuple out of the tuple table slot, making sure we have a
1577 tuple = ExecMaterializeSlot(slot);
1580 * get information on the (current) result relation
1582 resultRelInfo = estate->es_result_relation_info;
1583 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1585 /* BEFORE ROW UPDATE Triggers */
1586 if (resultRelInfo->ri_TrigDesc &&
1587 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
1591 newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
1593 estate->es_snapshot->curcid);
1595 if (newtuple == NULL) /* "do nothing" */
1598 if (newtuple != tuple) /* modified by Trigger(s) */
1601 * Put the modified tuple into a slot for convenience of routines
1602 * below. We assume the tuple was allocated in per-tuple memory
1603 * context, and therefore will go away by itself. The tuple table
1604 * slot should not try to clear it.
1606 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1608 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1609 ExecSetSlotDescriptor(newslot, slot->tts_tupleDescriptor);
1610 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1617 * Check the constraints of the tuple
1619 * If we generate a new candidate tuple after EvalPlanQual testing, we
1620 * must loop back here and recheck constraints. (We don't need to redo
1621 * triggers, however. If there are any BEFORE triggers then trigger.c
1622 * will have done heap_lock_tuple to lock the correct tuple, so there's no
1623 * need to do them again.)
1626 if (resultRelationDesc->rd_att->constr)
1627 ExecConstraints(resultRelInfo, slot, estate);
1630 * replace the heap tuple
1632 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1633 * the row to be updated is visible to that snapshot, and throw a can't-
1634 * serialize error if not. This is a special-case behavior needed for
1635 * referential integrity updates in serializable transactions.
1637 result = heap_update(resultRelationDesc, tupleid, tuple,
1638 &update_ctid, &update_xmax,
1639 estate->es_snapshot->curcid,
1640 estate->es_crosscheck_snapshot,
1641 true /* wait for commit */ );
1644 case HeapTupleSelfUpdated:
1645 /* already deleted by self; nothing to do */
1648 case HeapTupleMayBeUpdated:
1651 case HeapTupleUpdated:
1652 if (IsXactIsoLevelSerializable)
1654 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1655 errmsg("could not serialize access due to concurrent update")));
1656 else if (!ItemPointerEquals(tupleid, &update_ctid))
1658 TupleTableSlot *epqslot;
1660 epqslot = EvalPlanQual(estate,
1661 resultRelInfo->ri_RangeTableIndex,
1664 estate->es_snapshot->curcid);
1665 if (!TupIsNull(epqslot))
1667 *tupleid = update_ctid;
1668 slot = ExecFilterJunk(estate->es_junkFilter, epqslot);
1669 tuple = ExecMaterializeSlot(slot);
1673 /* tuple already deleted; nothing to do */
1677 elog(ERROR, "unrecognized heap_update status: %u", result);
1682 (estate->es_processed)++;
1685 * Note: instead of having to update the old index tuples associated with
1686 * the heap tuple, all we do is form and insert new index tuples. This is
1687 * because UPDATEs are actually DELETEs and INSERTs, and index tuple
1688 * deletion is done later by VACUUM (see notes in ExecDelete). All we do
1689 * here is insert new index tuples. -cim 9/27/89
1693 * insert index entries for tuple
1695 * Note: heap_update returns the tid (location) of the new tuple in the
1698 if (resultRelInfo->ri_NumIndices > 0)
1699 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1701 /* AFTER ROW UPDATE Triggers */
1702 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
1704 /* Process RETURNING if present */
1705 if (resultRelInfo->ri_projectReturning)
1706 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1707 slot, planSlot, dest);
1711 * ExecRelCheck --- check that tuple meets constraints for result relation
1714 ExecRelCheck(ResultRelInfo *resultRelInfo,
1715 TupleTableSlot *slot, EState *estate)
1717 Relation rel = resultRelInfo->ri_RelationDesc;
1718 int ncheck = rel->rd_att->constr->num_check;
1719 ConstrCheck *check = rel->rd_att->constr->check;
1720 ExprContext *econtext;
1721 MemoryContext oldContext;
1726 * If first time through for this result relation, build expression
1727 * nodetrees for rel's constraint expressions. Keep them in the per-query
1728 * memory context so they'll survive throughout the query.
1730 if (resultRelInfo->ri_ConstraintExprs == NULL)
1732 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1733 resultRelInfo->ri_ConstraintExprs =
1734 (List **) palloc(ncheck * sizeof(List *));
1735 for (i = 0; i < ncheck; i++)
1737 /* ExecQual wants implicit-AND form */
1738 qual = make_ands_implicit(stringToNode(check[i].ccbin));
1739 resultRelInfo->ri_ConstraintExprs[i] = (List *)
1740 ExecPrepareExpr((Expr *) qual, estate);
1742 MemoryContextSwitchTo(oldContext);
1746 * We will use the EState's per-tuple context for evaluating constraint
1747 * expressions (creating it if it's not already there).
1749 econtext = GetPerTupleExprContext(estate);
1751 /* Arrange for econtext's scan tuple to be the tuple under test */
1752 econtext->ecxt_scantuple = slot;
1754 /* And evaluate the constraints */
1755 for (i = 0; i < ncheck; i++)
1757 qual = resultRelInfo->ri_ConstraintExprs[i];
1760 * NOTE: SQL92 specifies that a NULL result from a constraint
1761 * expression is not to be treated as a failure. Therefore, tell
1762 * ExecQual to return TRUE for NULL.
1764 if (!ExecQual(qual, econtext, true))
1765 return check[i].ccname;
1768 /* NULL result means no error */
1773 ExecConstraints(ResultRelInfo *resultRelInfo,
1774 TupleTableSlot *slot, EState *estate)
1776 Relation rel = resultRelInfo->ri_RelationDesc;
1777 TupleConstr *constr = rel->rd_att->constr;
1781 if (constr->has_not_null)
1783 int natts = rel->rd_att->natts;
1786 for (attrChk = 1; attrChk <= natts; attrChk++)
1788 if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1789 slot_attisnull(slot, attrChk))
1791 (errcode(ERRCODE_NOT_NULL_VIOLATION),
1792 errmsg("null value in column \"%s\" violates not-null constraint",
1793 NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1797 if (constr->num_check > 0)
1801 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1803 (errcode(ERRCODE_CHECK_VIOLATION),
1804 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1805 RelationGetRelationName(rel), failed)));
1810 * ExecProcessReturning --- evaluate a RETURNING list and send to dest
1812 * projectReturning: RETURNING projection info for current result rel
1813 * tupleSlot: slot holding tuple actually inserted/updated/deleted
1814 * planSlot: slot holding tuple returned by top plan node
1815 * dest: where to send the output
1818 ExecProcessReturning(ProjectionInfo *projectReturning,
1819 TupleTableSlot *tupleSlot,
1820 TupleTableSlot *planSlot,
1823 ExprContext *econtext = projectReturning->pi_exprContext;
1824 TupleTableSlot *retSlot;
1827 * Reset per-tuple memory context to free any expression evaluation
1828 * storage allocated in the previous cycle.
1830 ResetExprContext(econtext);
1832 /* Make tuple and any needed join variables available to ExecProject */
1833 econtext->ecxt_scantuple = tupleSlot;
1834 econtext->ecxt_outertuple = planSlot;
1836 /* Compute the RETURNING expressions */
1837 retSlot = ExecProject(projectReturning, NULL);
1840 (*dest->receiveSlot) (retSlot, dest);
1842 ExecClearTuple(retSlot);
1846 * Check a modified tuple to see if we want to process its updated version
1847 * under READ COMMITTED rules.
1849 * See backend/executor/README for some info about how this works.
1851 * estate - executor state data
1852 * rti - rangetable index of table containing tuple
1853 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1854 * priorXmax - t_xmax from the outdated tuple
1855 * curCid - command ID of current command of my transaction
1857 * *tid is also an output parameter: it's modified to hold the TID of the
1858 * latest version of the tuple (note this may be changed even on failure)
1860 * Returns a slot containing the new candidate update/delete tuple, or
1861 * NULL if we determine we shouldn't process the row.
1864 EvalPlanQual(EState *estate, Index rti,
1865 ItemPointer tid, TransactionId priorXmax, CommandId curCid)
1870 HeapTupleData tuple;
1871 HeapTuple copyTuple = NULL;
1877 * find relation containing target tuple
1879 if (estate->es_result_relation_info != NULL &&
1880 estate->es_result_relation_info->ri_RangeTableIndex == rti)
1881 relation = estate->es_result_relation_info->ri_RelationDesc;
1887 foreach(l, estate->es_rowMarks)
1889 if (((ExecRowMark *) lfirst(l))->rti == rti)
1891 relation = ((ExecRowMark *) lfirst(l))->relation;
1895 if (relation == NULL)
1896 elog(ERROR, "could not find RowMark for RT index %u", rti);
1902 * Loop here to deal with updated or busy tuples
1904 tuple.t_self = *tid;
1909 if (heap_fetch(relation, SnapshotDirty, &tuple, &buffer, true, NULL))
1912 * If xmin isn't what we're expecting, the slot must have been
1913 * recycled and reused for an unrelated tuple. This implies that
1914 * the latest version of the row was deleted, so we need do
1915 * nothing. (Should be safe to examine xmin without getting
1916 * buffer's content lock, since xmin never changes in an existing
1919 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1922 ReleaseBuffer(buffer);
1926 /* otherwise xmin should not be dirty... */
1927 if (TransactionIdIsValid(SnapshotDirty->xmin))
1928 elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1931 * If tuple is being updated by other transaction then we have to
1932 * wait for its commit/abort.
1934 if (TransactionIdIsValid(SnapshotDirty->xmax))
1936 ReleaseBuffer(buffer);
1937 XactLockTableWait(SnapshotDirty->xmax);
1938 continue; /* loop back to repeat heap_fetch */
1942 * If tuple was inserted by our own transaction, we have to check
1943 * cmin against curCid: cmin >= curCid means our command cannot
1944 * see the tuple, so we should ignore it. Without this we are
1945 * open to the "Halloween problem" of indefinitely re-updating
1946 * the same tuple. (We need not check cmax because
1947 * HeapTupleSatisfiesDirty will consider a tuple deleted by
1948 * our transaction dead, regardless of cmax.) We just checked
1949 * that priorXmax == xmin, so we can test that variable instead
1950 * of doing HeapTupleHeaderGetXmin again.
1952 if (TransactionIdIsCurrentTransactionId(priorXmax) &&
1953 HeapTupleHeaderGetCmin(tuple.t_data) >= curCid)
1955 ReleaseBuffer(buffer);
1960 * We got tuple - now copy it for use by recheck query.
1962 copyTuple = heap_copytuple(&tuple);
1963 ReleaseBuffer(buffer);
1968 * If the referenced slot was actually empty, the latest version of
1969 * the row must have been deleted, so we need do nothing.
1971 if (tuple.t_data == NULL)
1973 ReleaseBuffer(buffer);
1978 * As above, if xmin isn't what we're expecting, do nothing.
1980 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1983 ReleaseBuffer(buffer);
1988 * If we get here, the tuple was found but failed SnapshotDirty.
1989 * Assuming the xmin is either a committed xact or our own xact (as it
1990 * certainly should be if we're trying to modify the tuple), this must
1991 * mean that the row was updated or deleted by either a committed xact
1992 * or our own xact. If it was deleted, we can ignore it; if it was
1993 * updated then chain up to the next version and repeat the whole
1996 * As above, it should be safe to examine xmax and t_ctid without the
1997 * buffer content lock, because they can't be changing.
1999 if (ItemPointerEquals(&tuple.t_self, &tuple.t_data->t_ctid))
2001 /* deleted, so forget about it */
2002 ReleaseBuffer(buffer);
2006 /* updated, so look at the updated row */
2007 tuple.t_self = tuple.t_data->t_ctid;
2008 /* updated row should have xmin matching this xmax */
2009 priorXmax = HeapTupleHeaderGetXmax(tuple.t_data);
2010 ReleaseBuffer(buffer);
2011 /* loop back to fetch next in chain */
2015 * For UPDATE/DELETE we have to return tid of actual row we're executing
2018 *tid = tuple.t_self;
2021 * Need to run a recheck subquery. Find or create a PQ stack entry.
2023 epq = estate->es_evalPlanQual;
2026 if (epq != NULL && epq->rti == 0)
2028 /* Top PQ stack entry is idle, so re-use it */
2029 Assert(!(estate->es_useEvalPlan) && epq->next == NULL);
2035 * If this is request for another RTE - Ra, - then we have to check wasn't
2036 * PlanQual requested for Ra already and if so then Ra' row was updated
2037 * again and we have to re-start old execution for Ra and forget all what
2038 * we done after Ra was suspended. Cool? -:))
2040 if (epq != NULL && epq->rti != rti &&
2041 epq->estate->es_evTuple[rti - 1] != NULL)
2045 evalPlanQual *oldepq;
2047 /* stop execution */
2048 EvalPlanQualStop(epq);
2049 /* pop previous PlanQual from the stack */
2051 Assert(oldepq && oldepq->rti != 0);
2052 /* push current PQ to freePQ stack */
2055 estate->es_evalPlanQual = epq;
2056 } while (epq->rti != rti);
2060 * If we are requested for another RTE then we have to suspend execution
2061 * of current PlanQual and start execution for new one.
2063 if (epq == NULL || epq->rti != rti)
2065 /* try to reuse plan used previously */
2066 evalPlanQual *newepq = (epq != NULL) ? epq->free : NULL;
2068 if (newepq == NULL) /* first call or freePQ stack is empty */
2070 newepq = (evalPlanQual *) palloc0(sizeof(evalPlanQual));
2071 newepq->free = NULL;
2072 newepq->estate = NULL;
2073 newepq->planstate = NULL;
2077 /* recycle previously used PlanQual */
2078 Assert(newepq->estate == NULL);
2081 /* push current PQ to the stack */
2084 estate->es_evalPlanQual = epq;
2089 Assert(epq->rti == rti);
2092 * Ok - we're requested for the same RTE. Unfortunately we still have to
2093 * end and restart execution of the plan, because ExecReScan wouldn't
2094 * ensure that upper plan nodes would reset themselves. We could make
2095 * that work if insertion of the target tuple were integrated with the
2096 * Param mechanism somehow, so that the upper plan nodes know that their
2097 * children's outputs have changed.
2099 * Note that the stack of free evalPlanQual nodes is quite useless at the
2100 * moment, since it only saves us from pallocing/releasing the
2101 * evalPlanQual nodes themselves. But it will be useful once we implement
2102 * ReScan instead of end/restart for re-using PlanQual nodes.
2106 /* stop execution */
2107 EvalPlanQualStop(epq);
2111 * Initialize new recheck query.
2113 * Note: if we were re-using PlanQual plans via ExecReScan, we'd need to
2114 * instead copy down changeable state from the top plan (including
2115 * es_result_relation_info, es_junkFilter) and reset locally changeable
2116 * state in the epq (including es_param_exec_vals, es_evTupleNull).
2118 EvalPlanQualStart(epq, estate, epq->next);
2121 * free old RTE' tuple, if any, and store target tuple where relation's
2122 * scan node will see it
2124 epqstate = epq->estate;
2125 if (epqstate->es_evTuple[rti - 1] != NULL)
2126 heap_freetuple(epqstate->es_evTuple[rti - 1]);
2127 epqstate->es_evTuple[rti - 1] = copyTuple;
2129 return EvalPlanQualNext(estate);
2132 static TupleTableSlot *
2133 EvalPlanQualNext(EState *estate)
2135 evalPlanQual *epq = estate->es_evalPlanQual;
2136 MemoryContext oldcontext;
2137 TupleTableSlot *slot;
2139 Assert(epq->rti != 0);
2142 oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt);
2143 slot = ExecProcNode(epq->planstate);
2144 MemoryContextSwitchTo(oldcontext);
2147 * No more tuples for this PQ. Continue previous one.
2149 if (TupIsNull(slot))
2151 evalPlanQual *oldepq;
2153 /* stop execution */
2154 EvalPlanQualStop(epq);
2155 /* pop old PQ from the stack */
2159 /* this is the first (oldest) PQ - mark as free */
2161 estate->es_useEvalPlan = false;
2162 /* and continue Query execution */
2165 Assert(oldepq->rti != 0);
2166 /* push current PQ to freePQ stack */
2169 estate->es_evalPlanQual = epq;
2177 EndEvalPlanQual(EState *estate)
2179 evalPlanQual *epq = estate->es_evalPlanQual;
2181 if (epq->rti == 0) /* plans already shutdowned */
2183 Assert(epq->next == NULL);
2189 evalPlanQual *oldepq;
2191 /* stop execution */
2192 EvalPlanQualStop(epq);
2193 /* pop old PQ from the stack */
2197 /* this is the first (oldest) PQ - mark as free */
2199 estate->es_useEvalPlan = false;
2202 Assert(oldepq->rti != 0);
2203 /* push current PQ to freePQ stack */
2206 estate->es_evalPlanQual = epq;
2211 * Start execution of one level of PlanQual.
2213 * This is a cut-down version of ExecutorStart(): we copy some state from
2214 * the top-level estate rather than initializing it fresh.
2217 EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
2221 MemoryContext oldcontext;
2223 rtsize = list_length(estate->es_range_table);
2225 epq->estate = epqstate = CreateExecutorState();
2227 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2230 * The epqstates share the top query's copy of unchanging state such as
2231 * the snapshot, rangetable, result-rel info, and external Param info.
2232 * They need their own copies of local state, including a tuple table,
2233 * es_param_exec_vals, etc.
2235 epqstate->es_direction = ForwardScanDirection;
2236 epqstate->es_snapshot = estate->es_snapshot;
2237 epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot;
2238 epqstate->es_range_table = estate->es_range_table;
2239 epqstate->es_result_relations = estate->es_result_relations;
2240 epqstate->es_num_result_relations = estate->es_num_result_relations;
2241 epqstate->es_result_relation_info = estate->es_result_relation_info;
2242 epqstate->es_junkFilter = estate->es_junkFilter;
2243 epqstate->es_into_relation_descriptor = estate->es_into_relation_descriptor;
2244 epqstate->es_into_relation_use_wal = estate->es_into_relation_use_wal;
2245 epqstate->es_param_list_info = estate->es_param_list_info;
2246 if (estate->es_topPlan->nParamExec > 0)
2247 epqstate->es_param_exec_vals = (ParamExecData *)
2248 palloc0(estate->es_topPlan->nParamExec * sizeof(ParamExecData));
2249 epqstate->es_rowMarks = estate->es_rowMarks;
2250 epqstate->es_instrument = estate->es_instrument;
2251 epqstate->es_select_into = estate->es_select_into;
2252 epqstate->es_into_oids = estate->es_into_oids;
2253 epqstate->es_topPlan = estate->es_topPlan;
2256 * Each epqstate must have its own es_evTupleNull state, but all the stack
2257 * entries share es_evTuple state. This allows sub-rechecks to inherit
2258 * the value being examined by an outer recheck.
2260 epqstate->es_evTupleNull = (bool *) palloc0(rtsize * sizeof(bool));
2261 if (priorepq == NULL)
2262 /* first PQ stack entry */
2263 epqstate->es_evTuple = (HeapTuple *)
2264 palloc0(rtsize * sizeof(HeapTuple));
2266 /* later stack entries share the same storage */
2267 epqstate->es_evTuple = priorepq->estate->es_evTuple;
2269 epqstate->es_tupleTable =
2270 ExecCreateTupleTable(estate->es_tupleTable->size);
2272 epq->planstate = ExecInitNode(estate->es_topPlan, epqstate, 0);
2274 MemoryContextSwitchTo(oldcontext);
2278 * End execution of one level of PlanQual.
2280 * This is a cut-down version of ExecutorEnd(); basically we want to do most
2281 * of the normal cleanup, but *not* close result relations (which we are
2282 * just sharing from the outer query).
2285 EvalPlanQualStop(evalPlanQual *epq)
2287 EState *epqstate = epq->estate;
2288 MemoryContext oldcontext;
2290 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2292 ExecEndNode(epq->planstate);
2294 ExecDropTupleTable(epqstate->es_tupleTable, true);
2295 epqstate->es_tupleTable = NULL;
2297 if (epqstate->es_evTuple[epq->rti - 1] != NULL)
2299 heap_freetuple(epqstate->es_evTuple[epq->rti - 1]);
2300 epqstate->es_evTuple[epq->rti - 1] = NULL;
2303 MemoryContextSwitchTo(oldcontext);
2305 FreeExecutorState(epqstate);
2308 epq->planstate = NULL;
2313 * Support for SELECT INTO (a/k/a CREATE TABLE AS)
2315 * We implement SELECT INTO by diverting SELECT's normal output with
2316 * a specialized DestReceiver type.
2318 * TODO: remove some of the INTO-specific cruft from EState, and keep
2319 * it in the DestReceiver instead.
2324 DestReceiver pub; /* publicly-known function pointers */
2325 EState *estate; /* EState we are working with */
2329 * OpenIntoRel --- actually create the SELECT INTO target relation
2331 * This also replaces QueryDesc->dest with the special DestReceiver for
2332 * SELECT INTO. We assume that the correct result tuple type has already
2333 * been placed in queryDesc->tupDesc.
2336 OpenIntoRel(QueryDesc *queryDesc)
2338 Query *parseTree = queryDesc->parsetree;
2339 EState *estate = queryDesc->estate;
2340 Relation intoRelationDesc;
2345 AclResult aclresult;
2348 DR_intorel *myState;
2351 * Check consistency of arguments
2353 if (parseTree->intoOnCommit != ONCOMMIT_NOOP && !parseTree->into->istemp)
2355 (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
2356 errmsg("ON COMMIT can only be used on temporary tables")));
2359 * Find namespace to create in, check its permissions
2361 intoName = parseTree->into->relname;
2362 namespaceId = RangeVarGetCreationNamespace(parseTree->into);
2364 aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
2366 if (aclresult != ACLCHECK_OK)
2367 aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
2368 get_namespace_name(namespaceId));
2371 * Select tablespace to use. If not specified, use default_tablespace
2372 * (which may in turn default to database's default).
2374 if (parseTree->intoTableSpaceName)
2376 tablespaceId = get_tablespace_oid(parseTree->intoTableSpaceName);
2377 if (!OidIsValid(tablespaceId))
2379 (errcode(ERRCODE_UNDEFINED_OBJECT),
2380 errmsg("tablespace \"%s\" does not exist",
2381 parseTree->intoTableSpaceName)));
2384 tablespaceId = GetDefaultTablespace();
2385 /* note InvalidOid is OK in this case */
2388 /* Check permissions except when using the database's default space */
2389 if (OidIsValid(tablespaceId))
2391 AclResult aclresult;
2393 aclresult = pg_tablespace_aclcheck(tablespaceId, GetUserId(),
2396 if (aclresult != ACLCHECK_OK)
2397 aclcheck_error(aclresult, ACL_KIND_TABLESPACE,
2398 get_tablespace_name(tablespaceId));
2401 /* Parse and validate any reloptions */
2402 reloptions = transformRelOptions((Datum) 0,
2403 parseTree->intoOptions,
2406 (void) heap_reloptions(RELKIND_RELATION, reloptions, true);
2408 /* have to copy the actual tupdesc to get rid of any constraints */
2409 tupdesc = CreateTupleDescCopy(queryDesc->tupDesc);
2411 /* Now we can actually create the new relation */
2412 intoRelationId = heap_create_with_catalog(intoName,
2422 parseTree->intoOnCommit,
2424 allowSystemTableMods);
2426 FreeTupleDesc(tupdesc);
2429 * Advance command counter so that the newly-created relation's
2430 * catalog tuples will be visible to heap_open.
2432 CommandCounterIncrement();
2435 * If necessary, create a TOAST table for the INTO relation. Note that
2436 * AlterTableCreateToastTable ends with CommandCounterIncrement(), so
2437 * that the TOAST table will be visible for insertion.
2439 AlterTableCreateToastTable(intoRelationId);
2442 * And open the constructed table for writing.
2444 intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
2446 /* use_wal off requires rd_targblock be initially invalid */
2447 Assert(intoRelationDesc->rd_targblock == InvalidBlockNumber);
2450 * We can skip WAL-logging the insertions, unless PITR is in use.
2452 * Note that for a non-temp INTO table, this is safe only because we
2453 * know that the catalog changes above will have been WAL-logged, and
2454 * so RecordTransactionCommit will think it needs to WAL-log the
2455 * eventual transaction commit. Else the commit might be lost, even
2456 * though all the data is safely fsync'd ...
2458 estate->es_into_relation_use_wal = XLogArchivingActive();
2459 estate->es_into_relation_descriptor = intoRelationDesc;
2462 * Now replace the query's DestReceiver with one for SELECT INTO
2464 queryDesc->dest = CreateDestReceiver(DestIntoRel, NULL);
2465 myState = (DR_intorel *) queryDesc->dest;
2466 Assert(myState->pub.mydest == DestIntoRel);
2467 myState->estate = estate;
2471 * CloseIntoRel --- clean up SELECT INTO at ExecutorEnd time
2474 CloseIntoRel(QueryDesc *queryDesc)
2476 EState *estate = queryDesc->estate;
2478 /* OpenIntoRel might never have gotten called */
2479 if (estate->es_into_relation_descriptor)
2482 * If we skipped using WAL, and it's not a temp relation, we must
2483 * force the relation down to disk before it's safe to commit the
2484 * transaction. This requires forcing out any dirty buffers and then
2485 * doing a forced fsync.
2487 if (!estate->es_into_relation_use_wal &&
2488 !estate->es_into_relation_descriptor->rd_istemp)
2490 FlushRelationBuffers(estate->es_into_relation_descriptor);
2491 /* FlushRelationBuffers will have opened rd_smgr */
2492 smgrimmedsync(estate->es_into_relation_descriptor->rd_smgr);
2495 /* close rel, but keep lock until commit */
2496 heap_close(estate->es_into_relation_descriptor, NoLock);
2498 estate->es_into_relation_descriptor = NULL;
2503 * CreateIntoRelDestReceiver -- create a suitable DestReceiver object
2505 * Since CreateDestReceiver doesn't accept the parameters we'd need,
2506 * we just leave the private fields empty here. OpenIntoRel will
2510 CreateIntoRelDestReceiver(void)
2512 DR_intorel *self = (DR_intorel *) palloc(sizeof(DR_intorel));
2514 self->pub.receiveSlot = intorel_receive;
2515 self->pub.rStartup = intorel_startup;
2516 self->pub.rShutdown = intorel_shutdown;
2517 self->pub.rDestroy = intorel_destroy;
2518 self->pub.mydest = DestIntoRel;
2520 self->estate = NULL;
2522 return (DestReceiver *) self;
2526 * intorel_startup --- executor startup
2529 intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo)
2535 * intorel_receive --- receive one tuple
2538 intorel_receive(TupleTableSlot *slot, DestReceiver *self)
2540 DR_intorel *myState = (DR_intorel *) self;
2541 EState *estate = myState->estate;
2544 tuple = ExecCopySlotTuple(slot);
2546 heap_insert(estate->es_into_relation_descriptor,
2548 estate->es_snapshot->curcid,
2549 estate->es_into_relation_use_wal,
2550 false); /* never any point in using FSM */
2552 /* We know this is a newly created relation, so there are no indexes */
2554 heap_freetuple(tuple);
2560 * intorel_shutdown --- executor end
2563 intorel_shutdown(DestReceiver *self)
2569 * intorel_destroy --- release DestReceiver object
2572 intorel_destroy(DestReceiver *self)