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11 * The old ExecutorMain() has been replaced by ExecutorStart(),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 1.280 2006/10/04 00:29:52 momjian 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 in
762 * 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);
778 * Because we already ran ExecInitNode() for the top plan node, any
779 * subplans we just attached to it won't have been initialized; so we
780 * have to do it here. (Ugly, but the alternatives seem worse.)
782 foreach(l, planstate->subPlan)
784 SubPlanState *sstate = (SubPlanState *) lfirst(l);
786 Assert(IsA(sstate, SubPlanState));
787 if (sstate->planstate == NULL) /* already inited? */
788 ExecInitSubPlan(sstate, estate, eflags);
792 queryDesc->tupDesc = tupType;
793 queryDesc->planstate = planstate;
796 * If doing SELECT INTO, initialize the "into" relation. We must wait
797 * till now so we have the "clean" result tuple type to create the new
800 * If EXPLAIN, skip creating the "into" relation.
802 if (estate->es_select_into && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
803 OpenIntoRel(queryDesc);
807 * Initialize ResultRelInfo data for one result relation
810 initResultRelInfo(ResultRelInfo *resultRelInfo,
811 Index resultRelationIndex,
816 Oid resultRelationOid;
817 Relation resultRelationDesc;
819 resultRelationOid = getrelid(resultRelationIndex, rangeTable);
820 resultRelationDesc = heap_open(resultRelationOid, RowExclusiveLock);
822 switch (resultRelationDesc->rd_rel->relkind)
824 case RELKIND_SEQUENCE:
826 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
827 errmsg("cannot change sequence \"%s\"",
828 RelationGetRelationName(resultRelationDesc))));
830 case RELKIND_TOASTVALUE:
832 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
833 errmsg("cannot change TOAST relation \"%s\"",
834 RelationGetRelationName(resultRelationDesc))));
838 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
839 errmsg("cannot change view \"%s\"",
840 RelationGetRelationName(resultRelationDesc))));
844 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
845 resultRelInfo->type = T_ResultRelInfo;
846 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
847 resultRelInfo->ri_RelationDesc = resultRelationDesc;
848 resultRelInfo->ri_NumIndices = 0;
849 resultRelInfo->ri_IndexRelationDescs = NULL;
850 resultRelInfo->ri_IndexRelationInfo = NULL;
851 /* make a copy so as not to depend on relcache info not changing... */
852 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
853 if (resultRelInfo->ri_TrigDesc)
855 int n = resultRelInfo->ri_TrigDesc->numtriggers;
857 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
858 palloc0(n * sizeof(FmgrInfo));
860 resultRelInfo->ri_TrigInstrument = InstrAlloc(n);
862 resultRelInfo->ri_TrigInstrument = NULL;
866 resultRelInfo->ri_TrigFunctions = NULL;
867 resultRelInfo->ri_TrigInstrument = NULL;
869 resultRelInfo->ri_ConstraintExprs = NULL;
870 resultRelInfo->ri_junkFilter = NULL;
871 resultRelInfo->ri_projectReturning = NULL;
874 * If there are indices on the result relation, open them and save
875 * descriptors in the result relation info, so that we can add new index
876 * entries for the tuples we add/update. We need not do this for a
877 * DELETE, however, since deletion doesn't affect indexes.
879 if (resultRelationDesc->rd_rel->relhasindex &&
880 operation != CMD_DELETE)
881 ExecOpenIndices(resultRelInfo);
885 * ExecContextForcesOids
887 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
888 * we need to ensure that result tuples have space for an OID iff they are
889 * going to be stored into a relation that has OIDs. In other contexts
890 * we are free to choose whether to leave space for OIDs in result tuples
891 * (we generally don't want to, but we do if a physical-tlist optimization
892 * is possible). This routine checks the plan context and returns TRUE if the
893 * choice is forced, FALSE if the choice is not forced. In the TRUE case,
894 * *hasoids is set to the required value.
896 * One reason this is ugly is that all plan nodes in the plan tree will emit
897 * tuples with space for an OID, though we really only need the topmost node
898 * to do so. However, node types like Sort don't project new tuples but just
899 * return their inputs, and in those cases the requirement propagates down
900 * to the input node. Eventually we might make this code smart enough to
901 * recognize how far down the requirement really goes, but for now we just
902 * make all plan nodes do the same thing if the top level forces the choice.
904 * We assume that estate->es_result_relation_info is already set up to
905 * describe the target relation. Note that in an UPDATE that spans an
906 * inheritance tree, some of the target relations may have OIDs and some not.
907 * We have to make the decisions on a per-relation basis as we initialize
908 * each of the child plans of the topmost Append plan.
910 * SELECT INTO is even uglier, because we don't have the INTO relation's
911 * descriptor available when this code runs; we have to look aside at a
912 * flag set by InitPlan().
915 ExecContextForcesOids(PlanState *planstate, bool *hasoids)
917 if (planstate->state->es_select_into)
919 *hasoids = planstate->state->es_into_oids;
924 ResultRelInfo *ri = planstate->state->es_result_relation_info;
928 Relation rel = ri->ri_RelationDesc;
932 *hasoids = rel->rd_rel->relhasoids;
941 /* ----------------------------------------------------------------
944 * Cleans up the query plan -- closes files and frees up storage
946 * NOTE: we are no longer very worried about freeing storage per se
947 * in this code; FreeExecutorState should be guaranteed to release all
948 * memory that needs to be released. What we are worried about doing
949 * is closing relations and dropping buffer pins. Thus, for example,
950 * tuple tables must be cleared or dropped to ensure pins are released.
951 * ----------------------------------------------------------------
954 ExecEndPlan(PlanState *planstate, EState *estate)
956 ResultRelInfo *resultRelInfo;
961 * shut down any PlanQual processing we were doing
963 if (estate->es_evalPlanQual != NULL)
964 EndEvalPlanQual(estate);
967 * shut down the node-type-specific query processing
969 ExecEndNode(planstate);
972 * destroy the executor "tuple" table.
974 ExecDropTupleTable(estate->es_tupleTable, true);
975 estate->es_tupleTable = NULL;
978 * close the result relation(s) if any, but hold locks until xact commit.
980 resultRelInfo = estate->es_result_relations;
981 for (i = estate->es_num_result_relations; i > 0; i--)
983 /* Close indices and then the relation itself */
984 ExecCloseIndices(resultRelInfo);
985 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
990 * close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
992 foreach(l, estate->es_rowMarks)
994 ExecRowMark *erm = lfirst(l);
996 heap_close(erm->relation, NoLock);
1000 /* ----------------------------------------------------------------
1003 * processes the query plan to retrieve 'numberTuples' tuples in the
1004 * direction specified.
1006 * Retrieves all tuples if numberTuples is 0
1008 * result is either a slot containing the last tuple in the case
1009 * of a SELECT or NULL otherwise.
1011 * Note: the ctid attribute is a 'junk' attribute that is removed before the
1013 * ----------------------------------------------------------------
1015 static TupleTableSlot *
1016 ExecutePlan(EState *estate,
1017 PlanState *planstate,
1020 ScanDirection direction,
1023 JunkFilter *junkfilter;
1024 TupleTableSlot *planSlot;
1025 TupleTableSlot *slot;
1026 ItemPointer tupleid = NULL;
1027 ItemPointerData tuple_ctid;
1028 long current_tuple_count;
1029 TupleTableSlot *result;
1032 * initialize local variables
1034 current_tuple_count = 0;
1038 * Set the direction.
1040 estate->es_direction = direction;
1043 * Process BEFORE EACH STATEMENT triggers
1048 ExecBSUpdateTriggers(estate, estate->es_result_relation_info);
1051 ExecBSDeleteTriggers(estate, estate->es_result_relation_info);
1054 ExecBSInsertTriggers(estate, estate->es_result_relation_info);
1062 * Loop until we've processed the proper number of tuples from the plan.
1067 /* Reset the per-output-tuple exprcontext */
1068 ResetPerTupleExprContext(estate);
1071 * Execute the plan and obtain a tuple
1074 if (estate->es_useEvalPlan)
1076 planSlot = EvalPlanQualNext(estate);
1077 if (TupIsNull(planSlot))
1078 planSlot = ExecProcNode(planstate);
1081 planSlot = ExecProcNode(planstate);
1084 * if the tuple is null, then we assume there is nothing more to
1085 * process so we just return null...
1087 if (TupIsNull(planSlot))
1095 * if we have a junk filter, then project a new tuple with the junk
1098 * Store this new "clean" tuple in the junkfilter's resultSlot.
1099 * (Formerly, we stored it back over the "dirty" tuple, which is WRONG
1100 * because that tuple slot has the wrong descriptor.)
1102 * Also, extract all the junk information we need.
1104 if ((junkfilter = estate->es_junkFilter) != NULL)
1110 * extract the 'ctid' junk attribute.
1112 if (operation == CMD_UPDATE || operation == CMD_DELETE)
1114 if (!ExecGetJunkAttribute(junkfilter,
1119 elog(ERROR, "could not find junk ctid column");
1121 /* shouldn't ever get a null result... */
1123 elog(ERROR, "ctid is NULL");
1125 tupleid = (ItemPointer) DatumGetPointer(datum);
1126 tuple_ctid = *tupleid; /* make sure we don't free the ctid!! */
1127 tupleid = &tuple_ctid;
1131 * Process any FOR UPDATE or FOR SHARE locking requested.
1133 else if (estate->es_rowMarks != NIL)
1138 foreach(l, estate->es_rowMarks)
1140 ExecRowMark *erm = lfirst(l);
1141 HeapTupleData tuple;
1143 ItemPointerData update_ctid;
1144 TransactionId update_xmax;
1145 TupleTableSlot *newSlot;
1146 LockTupleMode lockmode;
1149 if (!ExecGetJunkAttribute(junkfilter,
1154 elog(ERROR, "could not find junk \"%s\" column",
1157 /* shouldn't ever get a null result... */
1159 elog(ERROR, "\"%s\" is NULL", erm->resname);
1161 tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1164 lockmode = LockTupleExclusive;
1166 lockmode = LockTupleShared;
1168 test = heap_lock_tuple(erm->relation, &tuple, &buffer,
1169 &update_ctid, &update_xmax,
1170 estate->es_snapshot->curcid,
1171 lockmode, erm->noWait);
1172 ReleaseBuffer(buffer);
1175 case HeapTupleSelfUpdated:
1176 /* treat it as deleted; do not process */
1179 case HeapTupleMayBeUpdated:
1182 case HeapTupleUpdated:
1183 if (IsXactIsoLevelSerializable)
1185 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1186 errmsg("could not serialize access due to concurrent update")));
1187 if (!ItemPointerEquals(&update_ctid,
1190 /* updated, so look at updated version */
1191 newSlot = EvalPlanQual(estate,
1195 estate->es_snapshot->curcid);
1196 if (!TupIsNull(newSlot))
1198 slot = planSlot = newSlot;
1199 estate->es_useEvalPlan = true;
1205 * if tuple was deleted or PlanQual failed for
1206 * updated tuple - we must not return this tuple!
1211 elog(ERROR, "unrecognized heap_lock_tuple status: %u",
1219 * Create a new "clean" tuple with all junk attributes removed. We
1220 * don't need to do this for DELETE, however (there will in fact
1221 * be no non-junk attributes in a DELETE!)
1223 if (operation != CMD_DELETE)
1224 slot = ExecFilterJunk(junkfilter, slot);
1228 * now that we have a tuple, do the appropriate thing with it.. either
1229 * return it to the user, add it to a relation someplace, delete it
1230 * from a relation, or modify some of its attributes.
1235 ExecSelect(slot, dest, estate);
1240 ExecInsert(slot, tupleid, planSlot, dest, estate);
1245 ExecDelete(tupleid, planSlot, dest, estate);
1250 ExecUpdate(slot, tupleid, planSlot, dest, estate);
1255 elog(ERROR, "unrecognized operation code: %d",
1262 * check our tuple count.. if we've processed the proper number then
1263 * quit, else loop again and process more tuples. Zero numberTuples
1266 current_tuple_count++;
1267 if (numberTuples && numberTuples == current_tuple_count)
1272 * Process AFTER EACH STATEMENT triggers
1277 ExecASUpdateTriggers(estate, estate->es_result_relation_info);
1280 ExecASDeleteTriggers(estate, estate->es_result_relation_info);
1283 ExecASInsertTriggers(estate, estate->es_result_relation_info);
1291 * here, result is either a slot containing a tuple in the case of a
1292 * SELECT or NULL otherwise.
1297 /* ----------------------------------------------------------------
1300 * SELECTs are easy.. we just pass the tuple to the appropriate
1302 * ----------------------------------------------------------------
1305 ExecSelect(TupleTableSlot *slot,
1309 (*dest->receiveSlot) (slot, dest);
1311 (estate->es_processed)++;
1314 /* ----------------------------------------------------------------
1317 * INSERTs are trickier.. we have to insert the tuple into
1318 * the base relation and insert appropriate tuples into the
1320 * ----------------------------------------------------------------
1323 ExecInsert(TupleTableSlot *slot,
1324 ItemPointer tupleid,
1325 TupleTableSlot *planSlot,
1330 ResultRelInfo *resultRelInfo;
1331 Relation resultRelationDesc;
1335 * get the heap tuple out of the tuple table slot, making sure we have a
1338 tuple = ExecMaterializeSlot(slot);
1341 * get information on the (current) result relation
1343 resultRelInfo = estate->es_result_relation_info;
1344 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1346 /* BEFORE ROW INSERT Triggers */
1347 if (resultRelInfo->ri_TrigDesc &&
1348 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
1352 newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
1354 if (newtuple == NULL) /* "do nothing" */
1357 if (newtuple != tuple) /* modified by Trigger(s) */
1360 * Put the modified tuple into a slot for convenience of routines
1361 * below. We assume the tuple was allocated in per-tuple memory
1362 * context, and therefore will go away by itself. The tuple table
1363 * slot should not try to clear it.
1365 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1367 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1368 ExecSetSlotDescriptor(newslot, slot->tts_tupleDescriptor);
1369 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1376 * Check the constraints of the tuple
1378 if (resultRelationDesc->rd_att->constr)
1379 ExecConstraints(resultRelInfo, slot, estate);
1384 * Note: heap_insert returns the tid (location) of the new tuple in the
1387 newId = heap_insert(resultRelationDesc, tuple,
1388 estate->es_snapshot->curcid,
1392 (estate->es_processed)++;
1393 estate->es_lastoid = newId;
1394 setLastTid(&(tuple->t_self));
1397 * insert index entries for tuple
1399 if (resultRelInfo->ri_NumIndices > 0)
1400 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1402 /* AFTER ROW INSERT Triggers */
1403 ExecARInsertTriggers(estate, resultRelInfo, tuple);
1405 /* Process RETURNING if present */
1406 if (resultRelInfo->ri_projectReturning)
1407 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1408 slot, planSlot, dest);
1411 /* ----------------------------------------------------------------
1414 * DELETE is like UPDATE, except that we delete the tuple and no
1415 * index modifications are needed
1416 * ----------------------------------------------------------------
1419 ExecDelete(ItemPointer tupleid,
1420 TupleTableSlot *planSlot,
1424 ResultRelInfo *resultRelInfo;
1425 Relation resultRelationDesc;
1427 ItemPointerData update_ctid;
1428 TransactionId update_xmax;
1431 * get information on the (current) result relation
1433 resultRelInfo = estate->es_result_relation_info;
1434 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1436 /* BEFORE ROW DELETE Triggers */
1437 if (resultRelInfo->ri_TrigDesc &&
1438 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
1442 dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
1443 estate->es_snapshot->curcid);
1445 if (!dodelete) /* "do nothing" */
1452 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1453 * the row to be deleted is visible to that snapshot, and throw a can't-
1454 * serialize error if not. This is a special-case behavior needed for
1455 * referential integrity updates in serializable transactions.
1458 result = heap_delete(resultRelationDesc, tupleid,
1459 &update_ctid, &update_xmax,
1460 estate->es_snapshot->curcid,
1461 estate->es_crosscheck_snapshot,
1462 true /* wait for commit */ );
1465 case HeapTupleSelfUpdated:
1466 /* already deleted by self; nothing to do */
1469 case HeapTupleMayBeUpdated:
1472 case HeapTupleUpdated:
1473 if (IsXactIsoLevelSerializable)
1475 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1476 errmsg("could not serialize access due to concurrent update")));
1477 else if (!ItemPointerEquals(tupleid, &update_ctid))
1479 TupleTableSlot *epqslot;
1481 epqslot = EvalPlanQual(estate,
1482 resultRelInfo->ri_RangeTableIndex,
1485 estate->es_snapshot->curcid);
1486 if (!TupIsNull(epqslot))
1488 *tupleid = update_ctid;
1492 /* tuple already deleted; nothing to do */
1496 elog(ERROR, "unrecognized heap_delete status: %u", result);
1501 (estate->es_processed)++;
1504 * Note: Normally one would think that we have to delete index tuples
1505 * associated with the heap tuple now...
1507 * ... but in POSTGRES, we have no need to do this because VACUUM will
1508 * take care of it later. We can't delete index tuples immediately
1509 * anyway, since the tuple is still visible to other transactions.
1512 /* AFTER ROW DELETE Triggers */
1513 ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
1515 /* Process RETURNING if present */
1516 if (resultRelInfo->ri_projectReturning)
1519 * We have to put the target tuple into a slot, which means first we
1520 * gotta fetch it. We can use the trigger tuple slot.
1522 TupleTableSlot *slot = estate->es_trig_tuple_slot;
1523 HeapTupleData deltuple;
1526 deltuple.t_self = *tupleid;
1527 if (!heap_fetch(resultRelationDesc, SnapshotAny,
1528 &deltuple, &delbuffer, false, NULL))
1529 elog(ERROR, "failed to fetch deleted tuple for DELETE RETURNING");
1531 if (slot->tts_tupleDescriptor != RelationGetDescr(resultRelationDesc))
1532 ExecSetSlotDescriptor(slot, RelationGetDescr(resultRelationDesc));
1533 ExecStoreTuple(&deltuple, slot, InvalidBuffer, false);
1535 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1536 slot, planSlot, dest);
1538 ExecClearTuple(slot);
1539 ReleaseBuffer(delbuffer);
1543 /* ----------------------------------------------------------------
1546 * note: we can't run UPDATE queries with transactions
1547 * off because UPDATEs are actually INSERTs and our
1548 * scan will mistakenly loop forever, updating the tuple
1549 * it just inserted.. This should be fixed but until it
1550 * is, we don't want to get stuck in an infinite loop
1551 * which corrupts your database..
1552 * ----------------------------------------------------------------
1555 ExecUpdate(TupleTableSlot *slot,
1556 ItemPointer tupleid,
1557 TupleTableSlot *planSlot,
1562 ResultRelInfo *resultRelInfo;
1563 Relation resultRelationDesc;
1565 ItemPointerData update_ctid;
1566 TransactionId update_xmax;
1569 * abort the operation if not running transactions
1571 if (IsBootstrapProcessingMode())
1572 elog(ERROR, "cannot UPDATE during bootstrap");
1575 * get the heap tuple out of the tuple table slot, making sure we have a
1578 tuple = ExecMaterializeSlot(slot);
1581 * get information on the (current) result relation
1583 resultRelInfo = estate->es_result_relation_info;
1584 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1586 /* BEFORE ROW UPDATE Triggers */
1587 if (resultRelInfo->ri_TrigDesc &&
1588 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
1592 newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
1594 estate->es_snapshot->curcid);
1596 if (newtuple == NULL) /* "do nothing" */
1599 if (newtuple != tuple) /* modified by Trigger(s) */
1602 * Put the modified tuple into a slot for convenience of routines
1603 * below. We assume the tuple was allocated in per-tuple memory
1604 * context, and therefore will go away by itself. The tuple table
1605 * slot should not try to clear it.
1607 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1609 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1610 ExecSetSlotDescriptor(newslot, slot->tts_tupleDescriptor);
1611 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1618 * Check the constraints of the tuple
1620 * If we generate a new candidate tuple after EvalPlanQual testing, we
1621 * must loop back here and recheck constraints. (We don't need to redo
1622 * triggers, however. If there are any BEFORE triggers then trigger.c
1623 * will have done heap_lock_tuple to lock the correct tuple, so there's no
1624 * need to do them again.)
1627 if (resultRelationDesc->rd_att->constr)
1628 ExecConstraints(resultRelInfo, slot, estate);
1631 * replace the heap tuple
1633 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1634 * the row to be updated is visible to that snapshot, and throw a can't-
1635 * serialize error if not. This is a special-case behavior needed for
1636 * referential integrity updates in serializable transactions.
1638 result = heap_update(resultRelationDesc, tupleid, tuple,
1639 &update_ctid, &update_xmax,
1640 estate->es_snapshot->curcid,
1641 estate->es_crosscheck_snapshot,
1642 true /* wait for commit */ );
1645 case HeapTupleSelfUpdated:
1646 /* already deleted by self; nothing to do */
1649 case HeapTupleMayBeUpdated:
1652 case HeapTupleUpdated:
1653 if (IsXactIsoLevelSerializable)
1655 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1656 errmsg("could not serialize access due to concurrent update")));
1657 else if (!ItemPointerEquals(tupleid, &update_ctid))
1659 TupleTableSlot *epqslot;
1661 epqslot = EvalPlanQual(estate,
1662 resultRelInfo->ri_RangeTableIndex,
1665 estate->es_snapshot->curcid);
1666 if (!TupIsNull(epqslot))
1668 *tupleid = update_ctid;
1669 slot = ExecFilterJunk(estate->es_junkFilter, epqslot);
1670 tuple = ExecMaterializeSlot(slot);
1674 /* tuple already deleted; nothing to do */
1678 elog(ERROR, "unrecognized heap_update status: %u", result);
1683 (estate->es_processed)++;
1686 * Note: instead of having to update the old index tuples associated with
1687 * the heap tuple, all we do is form and insert new index tuples. This is
1688 * because UPDATEs are actually DELETEs and INSERTs, and index tuple
1689 * deletion is done later by VACUUM (see notes in ExecDelete). All we do
1690 * here is insert new index tuples. -cim 9/27/89
1694 * insert index entries for tuple
1696 * Note: heap_update returns the tid (location) of the new tuple in the
1699 if (resultRelInfo->ri_NumIndices > 0)
1700 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1702 /* AFTER ROW UPDATE Triggers */
1703 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
1705 /* Process RETURNING if present */
1706 if (resultRelInfo->ri_projectReturning)
1707 ExecProcessReturning(resultRelInfo->ri_projectReturning,
1708 slot, planSlot, dest);
1712 * ExecRelCheck --- check that tuple meets constraints for result relation
1715 ExecRelCheck(ResultRelInfo *resultRelInfo,
1716 TupleTableSlot *slot, EState *estate)
1718 Relation rel = resultRelInfo->ri_RelationDesc;
1719 int ncheck = rel->rd_att->constr->num_check;
1720 ConstrCheck *check = rel->rd_att->constr->check;
1721 ExprContext *econtext;
1722 MemoryContext oldContext;
1727 * If first time through for this result relation, build expression
1728 * nodetrees for rel's constraint expressions. Keep them in the per-query
1729 * memory context so they'll survive throughout the query.
1731 if (resultRelInfo->ri_ConstraintExprs == NULL)
1733 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1734 resultRelInfo->ri_ConstraintExprs =
1735 (List **) palloc(ncheck * sizeof(List *));
1736 for (i = 0; i < ncheck; i++)
1738 /* ExecQual wants implicit-AND form */
1739 qual = make_ands_implicit(stringToNode(check[i].ccbin));
1740 resultRelInfo->ri_ConstraintExprs[i] = (List *)
1741 ExecPrepareExpr((Expr *) qual, estate);
1743 MemoryContextSwitchTo(oldContext);
1747 * We will use the EState's per-tuple context for evaluating constraint
1748 * expressions (creating it if it's not already there).
1750 econtext = GetPerTupleExprContext(estate);
1752 /* Arrange for econtext's scan tuple to be the tuple under test */
1753 econtext->ecxt_scantuple = slot;
1755 /* And evaluate the constraints */
1756 for (i = 0; i < ncheck; i++)
1758 qual = resultRelInfo->ri_ConstraintExprs[i];
1761 * NOTE: SQL92 specifies that a NULL result from a constraint
1762 * expression is not to be treated as a failure. Therefore, tell
1763 * ExecQual to return TRUE for NULL.
1765 if (!ExecQual(qual, econtext, true))
1766 return check[i].ccname;
1769 /* NULL result means no error */
1774 ExecConstraints(ResultRelInfo *resultRelInfo,
1775 TupleTableSlot *slot, EState *estate)
1777 Relation rel = resultRelInfo->ri_RelationDesc;
1778 TupleConstr *constr = rel->rd_att->constr;
1782 if (constr->has_not_null)
1784 int natts = rel->rd_att->natts;
1787 for (attrChk = 1; attrChk <= natts; attrChk++)
1789 if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1790 slot_attisnull(slot, attrChk))
1792 (errcode(ERRCODE_NOT_NULL_VIOLATION),
1793 errmsg("null value in column \"%s\" violates not-null constraint",
1794 NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1798 if (constr->num_check > 0)
1802 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1804 (errcode(ERRCODE_CHECK_VIOLATION),
1805 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1806 RelationGetRelationName(rel), failed)));
1811 * ExecProcessReturning --- evaluate a RETURNING list and send to dest
1813 * projectReturning: RETURNING projection info for current result rel
1814 * tupleSlot: slot holding tuple actually inserted/updated/deleted
1815 * planSlot: slot holding tuple returned by top plan node
1816 * dest: where to send the output
1819 ExecProcessReturning(ProjectionInfo *projectReturning,
1820 TupleTableSlot *tupleSlot,
1821 TupleTableSlot *planSlot,
1824 ExprContext *econtext = projectReturning->pi_exprContext;
1825 TupleTableSlot *retSlot;
1828 * Reset per-tuple memory context to free any expression evaluation
1829 * storage allocated in the previous cycle.
1831 ResetExprContext(econtext);
1833 /* Make tuple and any needed join variables available to ExecProject */
1834 econtext->ecxt_scantuple = tupleSlot;
1835 econtext->ecxt_outertuple = planSlot;
1837 /* Compute the RETURNING expressions */
1838 retSlot = ExecProject(projectReturning, NULL);
1841 (*dest->receiveSlot) (retSlot, dest);
1843 ExecClearTuple(retSlot);
1847 * Check a modified tuple to see if we want to process its updated version
1848 * under READ COMMITTED rules.
1850 * See backend/executor/README for some info about how this works.
1852 * estate - executor state data
1853 * rti - rangetable index of table containing tuple
1854 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1855 * priorXmax - t_xmax from the outdated tuple
1856 * curCid - command ID of current command of my transaction
1858 * *tid is also an output parameter: it's modified to hold the TID of the
1859 * latest version of the tuple (note this may be changed even on failure)
1861 * Returns a slot containing the new candidate update/delete tuple, or
1862 * NULL if we determine we shouldn't process the row.
1865 EvalPlanQual(EState *estate, Index rti,
1866 ItemPointer tid, TransactionId priorXmax, CommandId curCid)
1871 HeapTupleData tuple;
1872 HeapTuple copyTuple = NULL;
1878 * find relation containing target tuple
1880 if (estate->es_result_relation_info != NULL &&
1881 estate->es_result_relation_info->ri_RangeTableIndex == rti)
1882 relation = estate->es_result_relation_info->ri_RelationDesc;
1888 foreach(l, estate->es_rowMarks)
1890 if (((ExecRowMark *) lfirst(l))->rti == rti)
1892 relation = ((ExecRowMark *) lfirst(l))->relation;
1896 if (relation == NULL)
1897 elog(ERROR, "could not find RowMark for RT index %u", rti);
1903 * Loop here to deal with updated or busy tuples
1905 tuple.t_self = *tid;
1910 if (heap_fetch(relation, SnapshotDirty, &tuple, &buffer, true, NULL))
1913 * If xmin isn't what we're expecting, the slot must have been
1914 * recycled and reused for an unrelated tuple. This implies that
1915 * the latest version of the row was deleted, so we need do
1916 * nothing. (Should be safe to examine xmin without getting
1917 * buffer's content lock, since xmin never changes in an existing
1920 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1923 ReleaseBuffer(buffer);
1927 /* otherwise xmin should not be dirty... */
1928 if (TransactionIdIsValid(SnapshotDirty->xmin))
1929 elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1932 * If tuple is being updated by other transaction then we have to
1933 * wait for its commit/abort.
1935 if (TransactionIdIsValid(SnapshotDirty->xmax))
1937 ReleaseBuffer(buffer);
1938 XactLockTableWait(SnapshotDirty->xmax);
1939 continue; /* loop back to repeat heap_fetch */
1943 * If tuple was inserted by our own transaction, we have to check
1944 * cmin against curCid: cmin >= curCid means our command cannot
1945 * see the tuple, so we should ignore it. Without this we are
1946 * open to the "Halloween problem" of indefinitely re-updating the
1947 * same tuple. (We need not check cmax because
1948 * HeapTupleSatisfiesDirty will consider a tuple deleted by our
1949 * transaction dead, regardless of cmax.) We just checked that
1950 * priorXmax == xmin, so we can test that variable instead of
1951 * doing HeapTupleHeaderGetXmin again.
1953 if (TransactionIdIsCurrentTransactionId(priorXmax) &&
1954 HeapTupleHeaderGetCmin(tuple.t_data) >= curCid)
1956 ReleaseBuffer(buffer);
1961 * We got tuple - now copy it for use by recheck query.
1963 copyTuple = heap_copytuple(&tuple);
1964 ReleaseBuffer(buffer);
1969 * If the referenced slot was actually empty, the latest version of
1970 * the row must have been deleted, so we need do nothing.
1972 if (tuple.t_data == NULL)
1974 ReleaseBuffer(buffer);
1979 * As above, if xmin isn't what we're expecting, do nothing.
1981 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1984 ReleaseBuffer(buffer);
1989 * If we get here, the tuple was found but failed SnapshotDirty.
1990 * Assuming the xmin is either a committed xact or our own xact (as it
1991 * certainly should be if we're trying to modify the tuple), this must
1992 * mean that the row was updated or deleted by either a committed xact
1993 * or our own xact. If it was deleted, we can ignore it; if it was
1994 * updated then chain up to the next version and repeat the whole
1997 * As above, it should be safe to examine xmax and t_ctid without the
1998 * buffer content lock, because they can't be changing.
2000 if (ItemPointerEquals(&tuple.t_self, &tuple.t_data->t_ctid))
2002 /* deleted, so forget about it */
2003 ReleaseBuffer(buffer);
2007 /* updated, so look at the updated row */
2008 tuple.t_self = tuple.t_data->t_ctid;
2009 /* updated row should have xmin matching this xmax */
2010 priorXmax = HeapTupleHeaderGetXmax(tuple.t_data);
2011 ReleaseBuffer(buffer);
2012 /* loop back to fetch next in chain */
2016 * For UPDATE/DELETE we have to return tid of actual row we're executing
2019 *tid = tuple.t_self;
2022 * Need to run a recheck subquery. Find or create a PQ stack entry.
2024 epq = estate->es_evalPlanQual;
2027 if (epq != NULL && epq->rti == 0)
2029 /* Top PQ stack entry is idle, so re-use it */
2030 Assert(!(estate->es_useEvalPlan) && epq->next == NULL);
2036 * If this is request for another RTE - Ra, - then we have to check wasn't
2037 * PlanQual requested for Ra already and if so then Ra' row was updated
2038 * again and we have to re-start old execution for Ra and forget all what
2039 * we done after Ra was suspended. Cool? -:))
2041 if (epq != NULL && epq->rti != rti &&
2042 epq->estate->es_evTuple[rti - 1] != NULL)
2046 evalPlanQual *oldepq;
2048 /* stop execution */
2049 EvalPlanQualStop(epq);
2050 /* pop previous PlanQual from the stack */
2052 Assert(oldepq && oldepq->rti != 0);
2053 /* push current PQ to freePQ stack */
2056 estate->es_evalPlanQual = epq;
2057 } while (epq->rti != rti);
2061 * If we are requested for another RTE then we have to suspend execution
2062 * of current PlanQual and start execution for new one.
2064 if (epq == NULL || epq->rti != rti)
2066 /* try to reuse plan used previously */
2067 evalPlanQual *newepq = (epq != NULL) ? epq->free : NULL;
2069 if (newepq == NULL) /* first call or freePQ stack is empty */
2071 newepq = (evalPlanQual *) palloc0(sizeof(evalPlanQual));
2072 newepq->free = NULL;
2073 newepq->estate = NULL;
2074 newepq->planstate = NULL;
2078 /* recycle previously used PlanQual */
2079 Assert(newepq->estate == NULL);
2082 /* push current PQ to the stack */
2085 estate->es_evalPlanQual = epq;
2090 Assert(epq->rti == rti);
2093 * Ok - we're requested for the same RTE. Unfortunately we still have to
2094 * end and restart execution of the plan, because ExecReScan wouldn't
2095 * ensure that upper plan nodes would reset themselves. We could make
2096 * that work if insertion of the target tuple were integrated with the
2097 * Param mechanism somehow, so that the upper plan nodes know that their
2098 * children's outputs have changed.
2100 * Note that the stack of free evalPlanQual nodes is quite useless at the
2101 * moment, since it only saves us from pallocing/releasing the
2102 * evalPlanQual nodes themselves. But it will be useful once we implement
2103 * ReScan instead of end/restart for re-using PlanQual nodes.
2107 /* stop execution */
2108 EvalPlanQualStop(epq);
2112 * Initialize new recheck query.
2114 * Note: if we were re-using PlanQual plans via ExecReScan, we'd need to
2115 * instead copy down changeable state from the top plan (including
2116 * es_result_relation_info, es_junkFilter) and reset locally changeable
2117 * state in the epq (including es_param_exec_vals, es_evTupleNull).
2119 EvalPlanQualStart(epq, estate, epq->next);
2122 * free old RTE' tuple, if any, and store target tuple where relation's
2123 * scan node will see it
2125 epqstate = epq->estate;
2126 if (epqstate->es_evTuple[rti - 1] != NULL)
2127 heap_freetuple(epqstate->es_evTuple[rti - 1]);
2128 epqstate->es_evTuple[rti - 1] = copyTuple;
2130 return EvalPlanQualNext(estate);
2133 static TupleTableSlot *
2134 EvalPlanQualNext(EState *estate)
2136 evalPlanQual *epq = estate->es_evalPlanQual;
2137 MemoryContext oldcontext;
2138 TupleTableSlot *slot;
2140 Assert(epq->rti != 0);
2143 oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt);
2144 slot = ExecProcNode(epq->planstate);
2145 MemoryContextSwitchTo(oldcontext);
2148 * No more tuples for this PQ. Continue previous one.
2150 if (TupIsNull(slot))
2152 evalPlanQual *oldepq;
2154 /* stop execution */
2155 EvalPlanQualStop(epq);
2156 /* pop old PQ from the stack */
2160 /* this is the first (oldest) PQ - mark as free */
2162 estate->es_useEvalPlan = false;
2163 /* and continue Query execution */
2166 Assert(oldepq->rti != 0);
2167 /* push current PQ to freePQ stack */
2170 estate->es_evalPlanQual = epq;
2178 EndEvalPlanQual(EState *estate)
2180 evalPlanQual *epq = estate->es_evalPlanQual;
2182 if (epq->rti == 0) /* plans already shutdowned */
2184 Assert(epq->next == NULL);
2190 evalPlanQual *oldepq;
2192 /* stop execution */
2193 EvalPlanQualStop(epq);
2194 /* pop old PQ from the stack */
2198 /* this is the first (oldest) PQ - mark as free */
2200 estate->es_useEvalPlan = false;
2203 Assert(oldepq->rti != 0);
2204 /* push current PQ to freePQ stack */
2207 estate->es_evalPlanQual = epq;
2212 * Start execution of one level of PlanQual.
2214 * This is a cut-down version of ExecutorStart(): we copy some state from
2215 * the top-level estate rather than initializing it fresh.
2218 EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
2222 MemoryContext oldcontext;
2224 rtsize = list_length(estate->es_range_table);
2226 epq->estate = epqstate = CreateExecutorState();
2228 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2231 * The epqstates share the top query's copy of unchanging state such as
2232 * the snapshot, rangetable, result-rel info, and external Param info.
2233 * They need their own copies of local state, including a tuple table,
2234 * es_param_exec_vals, etc.
2236 epqstate->es_direction = ForwardScanDirection;
2237 epqstate->es_snapshot = estate->es_snapshot;
2238 epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot;
2239 epqstate->es_range_table = estate->es_range_table;
2240 epqstate->es_result_relations = estate->es_result_relations;
2241 epqstate->es_num_result_relations = estate->es_num_result_relations;
2242 epqstate->es_result_relation_info = estate->es_result_relation_info;
2243 epqstate->es_junkFilter = estate->es_junkFilter;
2244 epqstate->es_into_relation_descriptor = estate->es_into_relation_descriptor;
2245 epqstate->es_into_relation_use_wal = estate->es_into_relation_use_wal;
2246 epqstate->es_param_list_info = estate->es_param_list_info;
2247 if (estate->es_topPlan->nParamExec > 0)
2248 epqstate->es_param_exec_vals = (ParamExecData *)
2249 palloc0(estate->es_topPlan->nParamExec * sizeof(ParamExecData));
2250 epqstate->es_rowMarks = estate->es_rowMarks;
2251 epqstate->es_instrument = estate->es_instrument;
2252 epqstate->es_select_into = estate->es_select_into;
2253 epqstate->es_into_oids = estate->es_into_oids;
2254 epqstate->es_topPlan = estate->es_topPlan;
2257 * Each epqstate must have its own es_evTupleNull state, but all the stack
2258 * entries share es_evTuple state. This allows sub-rechecks to inherit
2259 * the value being examined by an outer recheck.
2261 epqstate->es_evTupleNull = (bool *) palloc0(rtsize * sizeof(bool));
2262 if (priorepq == NULL)
2263 /* first PQ stack entry */
2264 epqstate->es_evTuple = (HeapTuple *)
2265 palloc0(rtsize * sizeof(HeapTuple));
2267 /* later stack entries share the same storage */
2268 epqstate->es_evTuple = priorepq->estate->es_evTuple;
2270 epqstate->es_tupleTable =
2271 ExecCreateTupleTable(estate->es_tupleTable->size);
2273 epq->planstate = ExecInitNode(estate->es_topPlan, epqstate, 0);
2275 MemoryContextSwitchTo(oldcontext);
2279 * End execution of one level of PlanQual.
2281 * This is a cut-down version of ExecutorEnd(); basically we want to do most
2282 * of the normal cleanup, but *not* close result relations (which we are
2283 * just sharing from the outer query).
2286 EvalPlanQualStop(evalPlanQual *epq)
2288 EState *epqstate = epq->estate;
2289 MemoryContext oldcontext;
2291 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2293 ExecEndNode(epq->planstate);
2295 ExecDropTupleTable(epqstate->es_tupleTable, true);
2296 epqstate->es_tupleTable = NULL;
2298 if (epqstate->es_evTuple[epq->rti - 1] != NULL)
2300 heap_freetuple(epqstate->es_evTuple[epq->rti - 1]);
2301 epqstate->es_evTuple[epq->rti - 1] = NULL;
2304 MemoryContextSwitchTo(oldcontext);
2306 FreeExecutorState(epqstate);
2309 epq->planstate = NULL;
2314 * Support for SELECT INTO (a/k/a CREATE TABLE AS)
2316 * We implement SELECT INTO by diverting SELECT's normal output with
2317 * a specialized DestReceiver type.
2319 * TODO: remove some of the INTO-specific cruft from EState, and keep
2320 * it in the DestReceiver instead.
2325 DestReceiver pub; /* publicly-known function pointers */
2326 EState *estate; /* EState we are working with */
2330 * OpenIntoRel --- actually create the SELECT INTO target relation
2332 * This also replaces QueryDesc->dest with the special DestReceiver for
2333 * SELECT INTO. We assume that the correct result tuple type has already
2334 * been placed in queryDesc->tupDesc.
2337 OpenIntoRel(QueryDesc *queryDesc)
2339 Query *parseTree = queryDesc->parsetree;
2340 EState *estate = queryDesc->estate;
2341 Relation intoRelationDesc;
2346 AclResult aclresult;
2349 DR_intorel *myState;
2352 * Check consistency of arguments
2354 if (parseTree->intoOnCommit != ONCOMMIT_NOOP && !parseTree->into->istemp)
2356 (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
2357 errmsg("ON COMMIT can only be used on temporary tables")));
2360 * Find namespace to create in, check its permissions
2362 intoName = parseTree->into->relname;
2363 namespaceId = RangeVarGetCreationNamespace(parseTree->into);
2365 aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
2367 if (aclresult != ACLCHECK_OK)
2368 aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
2369 get_namespace_name(namespaceId));
2372 * Select tablespace to use. If not specified, use default_tablespace
2373 * (which may in turn default to database's default).
2375 if (parseTree->intoTableSpaceName)
2377 tablespaceId = get_tablespace_oid(parseTree->intoTableSpaceName);
2378 if (!OidIsValid(tablespaceId))
2380 (errcode(ERRCODE_UNDEFINED_OBJECT),
2381 errmsg("tablespace \"%s\" does not exist",
2382 parseTree->intoTableSpaceName)));
2386 tablespaceId = GetDefaultTablespace();
2387 /* note InvalidOid is OK in this case */
2390 /* Check permissions except when using the database's default space */
2391 if (OidIsValid(tablespaceId))
2393 AclResult aclresult;
2395 aclresult = pg_tablespace_aclcheck(tablespaceId, GetUserId(),
2398 if (aclresult != ACLCHECK_OK)
2399 aclcheck_error(aclresult, ACL_KIND_TABLESPACE,
2400 get_tablespace_name(tablespaceId));
2403 /* Parse and validate any reloptions */
2404 reloptions = transformRelOptions((Datum) 0,
2405 parseTree->intoOptions,
2408 (void) heap_reloptions(RELKIND_RELATION, reloptions, true);
2410 /* have to copy the actual tupdesc to get rid of any constraints */
2411 tupdesc = CreateTupleDescCopy(queryDesc->tupDesc);
2413 /* Now we can actually create the new relation */
2414 intoRelationId = heap_create_with_catalog(intoName,
2424 parseTree->intoOnCommit,
2426 allowSystemTableMods);
2428 FreeTupleDesc(tupdesc);
2431 * Advance command counter so that the newly-created relation's catalog
2432 * tuples will be visible to heap_open.
2434 CommandCounterIncrement();
2437 * If necessary, create a TOAST table for the INTO relation. Note that
2438 * AlterTableCreateToastTable ends with CommandCounterIncrement(), so that
2439 * the TOAST table will be visible for insertion.
2441 AlterTableCreateToastTable(intoRelationId);
2444 * And open the constructed table for writing.
2446 intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
2448 /* use_wal off requires rd_targblock be initially invalid */
2449 Assert(intoRelationDesc->rd_targblock == InvalidBlockNumber);
2452 * We can skip WAL-logging the insertions, unless PITR is in use.
2454 * Note that for a non-temp INTO table, this is safe only because we know
2455 * that the catalog changes above will have been WAL-logged, and so
2456 * RecordTransactionCommit will think it needs to WAL-log the eventual
2457 * transaction commit. Else the commit might be lost, even though all the
2458 * data is safely fsync'd ...
2460 estate->es_into_relation_use_wal = XLogArchivingActive();
2461 estate->es_into_relation_descriptor = intoRelationDesc;
2464 * Now replace the query's DestReceiver with one for SELECT INTO
2466 queryDesc->dest = CreateDestReceiver(DestIntoRel, NULL);
2467 myState = (DR_intorel *) queryDesc->dest;
2468 Assert(myState->pub.mydest == DestIntoRel);
2469 myState->estate = estate;
2473 * CloseIntoRel --- clean up SELECT INTO at ExecutorEnd time
2476 CloseIntoRel(QueryDesc *queryDesc)
2478 EState *estate = queryDesc->estate;
2480 /* OpenIntoRel might never have gotten called */
2481 if (estate->es_into_relation_descriptor)
2484 * If we skipped using WAL, and it's not a temp relation, we must
2485 * force the relation down to disk before it's safe to commit the
2486 * transaction. This requires forcing out any dirty buffers and then
2487 * doing a forced fsync.
2489 if (!estate->es_into_relation_use_wal &&
2490 !estate->es_into_relation_descriptor->rd_istemp)
2492 FlushRelationBuffers(estate->es_into_relation_descriptor);
2493 /* FlushRelationBuffers will have opened rd_smgr */
2494 smgrimmedsync(estate->es_into_relation_descriptor->rd_smgr);
2497 /* close rel, but keep lock until commit */
2498 heap_close(estate->es_into_relation_descriptor, NoLock);
2500 estate->es_into_relation_descriptor = NULL;
2505 * CreateIntoRelDestReceiver -- create a suitable DestReceiver object
2507 * Since CreateDestReceiver doesn't accept the parameters we'd need,
2508 * we just leave the private fields empty here. OpenIntoRel will
2512 CreateIntoRelDestReceiver(void)
2514 DR_intorel *self = (DR_intorel *) palloc(sizeof(DR_intorel));
2516 self->pub.receiveSlot = intorel_receive;
2517 self->pub.rStartup = intorel_startup;
2518 self->pub.rShutdown = intorel_shutdown;
2519 self->pub.rDestroy = intorel_destroy;
2520 self->pub.mydest = DestIntoRel;
2522 self->estate = NULL;
2524 return (DestReceiver *) self;
2528 * intorel_startup --- executor startup
2531 intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo)
2537 * intorel_receive --- receive one tuple
2540 intorel_receive(TupleTableSlot *slot, DestReceiver *self)
2542 DR_intorel *myState = (DR_intorel *) self;
2543 EState *estate = myState->estate;
2546 tuple = ExecCopySlotTuple(slot);
2548 heap_insert(estate->es_into_relation_descriptor,
2550 estate->es_snapshot->curcid,
2551 estate->es_into_relation_use_wal,
2552 false); /* never any point in using FSM */
2554 /* We know this is a newly created relation, so there are no indexes */
2556 heap_freetuple(tuple);
2562 * intorel_shutdown --- executor end
2565 intorel_shutdown(DestReceiver *self)
2571 * intorel_destroy --- release DestReceiver object
2574 intorel_destroy(DestReceiver *self)