1 /*-------------------------------------------------------------------------
4 * top level executor interface routines
11 * The old ExecutorMain() has been replaced by ExecutorStart(),
12 * ExecutorRun() and ExecutorEnd()
14 * These three procedures are the external interfaces to the executor.
15 * In each case, the query descriptor is required as an argument.
17 * ExecutorStart() must be called at the beginning of execution of any
18 * query plan and ExecutorEnd() should always be called at the end of
19 * execution of a plan.
21 * ExecutorRun accepts direction and count arguments that specify whether
22 * the plan is to be executed forwards, backwards, and for how many tuples.
24 * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
25 * Portions Copyright (c) 1994, Regents of the University of California
29 * $PostgreSQL: pgsql/src/backend/executor/execMain.c,v 1.265 2006/01/12 21:48:53 tgl Exp $
31 *-------------------------------------------------------------------------
35 #include "access/heapam.h"
36 #include "access/xlog.h"
37 #include "catalog/heap.h"
38 #include "catalog/namespace.h"
39 #include "commands/tablecmds.h"
40 #include "commands/trigger.h"
41 #include "executor/execdebug.h"
42 #include "executor/execdefs.h"
43 #include "executor/instrument.h"
44 #include "miscadmin.h"
45 #include "optimizer/clauses.h"
46 #include "optimizer/var.h"
47 #include "parser/parsetree.h"
48 #include "storage/smgr.h"
49 #include "utils/acl.h"
50 #include "utils/guc.h"
51 #include "utils/lsyscache.h"
52 #include "utils/memutils.h"
55 typedef struct evalPlanQual
60 struct evalPlanQual *next; /* stack of active PlanQual plans */
61 struct evalPlanQual *free; /* list of free PlanQual plans */
64 /* decls for local routines only used within this module */
65 static void InitPlan(QueryDesc *queryDesc, bool explainOnly);
66 static void initResultRelInfo(ResultRelInfo *resultRelInfo,
67 Index resultRelationIndex,
71 static TupleTableSlot *ExecutePlan(EState *estate, PlanState *planstate,
74 ScanDirection direction,
76 static void ExecSelect(TupleTableSlot *slot,
79 static void ExecInsert(TupleTableSlot *slot, ItemPointer tupleid,
81 static void ExecDelete(TupleTableSlot *slot, ItemPointer tupleid,
83 static void ExecUpdate(TupleTableSlot *slot, ItemPointer tupleid,
85 static TupleTableSlot *EvalPlanQualNext(EState *estate);
86 static void EndEvalPlanQual(EState *estate);
87 static void ExecCheckRTEPerms(RangeTblEntry *rte);
88 static void ExecCheckXactReadOnly(Query *parsetree);
89 static void EvalPlanQualStart(evalPlanQual *epq, EState *estate,
90 evalPlanQual *priorepq);
91 static void EvalPlanQualStop(evalPlanQual *epq);
93 /* end of local decls */
96 /* ----------------------------------------------------------------
99 * This routine must be called at the beginning of any execution of any
102 * Takes a QueryDesc previously created by CreateQueryDesc (it's not real
103 * clear why we bother to separate the two functions, but...). The tupDesc
104 * field of the QueryDesc is filled in to describe the tuples that will be
105 * returned, and the internal fields (estate and planstate) are set up.
107 * If explainOnly is true, we are not actually intending to run the plan,
108 * only to set up for EXPLAIN; so skip unwanted side-effects.
110 * NB: the CurrentMemoryContext when this is called will become the parent
111 * of the per-query context used for this Executor invocation.
112 * ----------------------------------------------------------------
115 ExecutorStart(QueryDesc *queryDesc, bool explainOnly)
118 MemoryContext oldcontext;
120 /* sanity checks: queryDesc must not be started already */
121 Assert(queryDesc != NULL);
122 Assert(queryDesc->estate == NULL);
125 * If the transaction is read-only, we need to check if any writes are
126 * planned to non-temporary tables.
128 if (XactReadOnly && !explainOnly)
129 ExecCheckXactReadOnly(queryDesc->parsetree);
132 * Build EState, switch into per-query memory context for startup.
134 estate = CreateExecutorState();
135 queryDesc->estate = estate;
137 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
140 * Fill in parameters, if any, from queryDesc
142 estate->es_param_list_info = queryDesc->params;
144 if (queryDesc->plantree->nParamExec > 0)
145 estate->es_param_exec_vals = (ParamExecData *)
146 palloc0(queryDesc->plantree->nParamExec * sizeof(ParamExecData));
149 * Copy other important information into the EState
151 estate->es_snapshot = queryDesc->snapshot;
152 estate->es_crosscheck_snapshot = queryDesc->crosscheck_snapshot;
153 estate->es_instrument = queryDesc->doInstrument;
156 * Initialize the plan state tree
158 InitPlan(queryDesc, explainOnly);
160 MemoryContextSwitchTo(oldcontext);
163 /* ----------------------------------------------------------------
166 * This is the main routine of the executor module. It accepts
167 * the query descriptor from the traffic cop and executes the
170 * ExecutorStart must have been called already.
172 * If direction is NoMovementScanDirection then nothing is done
173 * except to start up/shut down the destination. Otherwise,
174 * we retrieve up to 'count' tuples in the specified direction.
176 * Note: count = 0 is interpreted as no portal limit, i.e., run to
179 * ----------------------------------------------------------------
182 ExecutorRun(QueryDesc *queryDesc,
183 ScanDirection direction, long count)
188 TupleTableSlot *result;
189 MemoryContext oldcontext;
192 Assert(queryDesc != NULL);
194 estate = queryDesc->estate;
196 Assert(estate != NULL);
199 * Switch into per-query memory context
201 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
204 * extract information from the query descriptor and the query feature.
206 operation = queryDesc->operation;
207 dest = queryDesc->dest;
210 * startup tuple receiver
212 estate->es_processed = 0;
213 estate->es_lastoid = InvalidOid;
215 (*dest->rStartup) (dest, operation, queryDesc->tupDesc);
220 if (direction == NoMovementScanDirection)
223 result = ExecutePlan(estate,
224 queryDesc->planstate,
233 (*dest->rShutdown) (dest);
235 MemoryContextSwitchTo(oldcontext);
240 /* ----------------------------------------------------------------
243 * This routine must be called at the end of execution of any
245 * ----------------------------------------------------------------
248 ExecutorEnd(QueryDesc *queryDesc)
251 MemoryContext oldcontext;
254 Assert(queryDesc != NULL);
256 estate = queryDesc->estate;
258 Assert(estate != NULL);
261 * Switch into per-query memory context to run ExecEndPlan
263 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
265 ExecEndPlan(queryDesc->planstate, estate);
268 * Must switch out of context before destroying it
270 MemoryContextSwitchTo(oldcontext);
273 * Release EState and per-query memory context. This should release
274 * everything the executor has allocated.
276 FreeExecutorState(estate);
278 /* Reset queryDesc fields that no longer point to anything */
279 queryDesc->tupDesc = NULL;
280 queryDesc->estate = NULL;
281 queryDesc->planstate = NULL;
284 /* ----------------------------------------------------------------
287 * This routine may be called on an open queryDesc to rewind it
289 * ----------------------------------------------------------------
292 ExecutorRewind(QueryDesc *queryDesc)
295 MemoryContext oldcontext;
298 Assert(queryDesc != NULL);
300 estate = queryDesc->estate;
302 Assert(estate != NULL);
304 /* It's probably not sensible to rescan updating queries */
305 Assert(queryDesc->operation == CMD_SELECT);
308 * Switch into per-query memory context
310 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
315 ExecReScan(queryDesc->planstate, NULL);
317 MemoryContextSwitchTo(oldcontext);
323 * Check access permissions for all relations listed in a range table.
326 ExecCheckRTPerms(List *rangeTable)
330 foreach(l, rangeTable)
332 RangeTblEntry *rte = lfirst(l);
334 ExecCheckRTEPerms(rte);
340 * Check access permissions for a single RTE.
343 ExecCheckRTEPerms(RangeTblEntry *rte)
345 AclMode requiredPerms;
350 * Only plain-relation RTEs need to be checked here. Subquery RTEs are
351 * checked by ExecInitSubqueryScan if the subquery is still a separate
352 * subquery --- if it's been pulled up into our query level then the RTEs
353 * are in our rangetable and will be checked here. Function RTEs are
354 * checked by init_fcache when the function is prepared for execution.
355 * Join and special RTEs need no checks.
357 if (rte->rtekind != RTE_RELATION)
361 * No work if requiredPerms is empty.
363 requiredPerms = rte->requiredPerms;
364 if (requiredPerms == 0)
370 * userid to check as: current user unless we have a setuid indication.
372 * Note: GetUserId() is presently fast enough that there's no harm in
373 * calling it separately for each RTE. If that stops being true, we could
374 * call it once in ExecCheckRTPerms and pass the userid down from there.
375 * But for now, no need for the extra clutter.
377 userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
380 * We must have *all* the requiredPerms bits, so use aclmask not aclcheck.
382 if (pg_class_aclmask(relOid, userid, requiredPerms, ACLMASK_ALL)
384 aclcheck_error(ACLCHECK_NO_PRIV, ACL_KIND_CLASS,
385 get_rel_name(relOid));
389 * Check that the query does not imply any writes to non-temp tables.
392 ExecCheckXactReadOnly(Query *parsetree)
397 * CREATE TABLE AS or SELECT INTO?
399 * XXX should we allow this if the destination is temp?
401 if (parsetree->into != NULL)
404 /* Fail if write permissions are requested on any non-temp table */
405 foreach(l, parsetree->rtable)
407 RangeTblEntry *rte = lfirst(l);
409 if (rte->rtekind == RTE_SUBQUERY)
411 ExecCheckXactReadOnly(rte->subquery);
415 if (rte->rtekind != RTE_RELATION)
418 if ((rte->requiredPerms & (~ACL_SELECT)) == 0)
421 if (isTempNamespace(get_rel_namespace(rte->relid)))
431 (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
432 errmsg("transaction is read-only")));
436 /* ----------------------------------------------------------------
439 * Initializes the query plan: open files, allocate storage
440 * and start up the rule manager
441 * ----------------------------------------------------------------
444 InitPlan(QueryDesc *queryDesc, bool explainOnly)
446 CmdType operation = queryDesc->operation;
447 Query *parseTree = queryDesc->parsetree;
448 Plan *plan = queryDesc->plantree;
449 EState *estate = queryDesc->estate;
450 PlanState *planstate;
452 Relation intoRelationDesc;
457 * Do permissions checks. It's sufficient to examine the query's top
458 * rangetable here --- subplan RTEs will be checked during
461 ExecCheckRTPerms(parseTree->rtable);
464 * get information from query descriptor
466 rangeTable = parseTree->rtable;
469 * initialize the node's execution state
471 estate->es_range_table = rangeTable;
474 * if there is a result relation, initialize result relation stuff
476 if (parseTree->resultRelation != 0 && operation != CMD_SELECT)
478 List *resultRelations = parseTree->resultRelations;
479 int numResultRelations;
480 ResultRelInfo *resultRelInfos;
482 if (resultRelations != NIL)
485 * Multiple result relations (due to inheritance)
486 * parseTree->resultRelations identifies them all
488 ResultRelInfo *resultRelInfo;
491 numResultRelations = list_length(resultRelations);
492 resultRelInfos = (ResultRelInfo *)
493 palloc(numResultRelations * sizeof(ResultRelInfo));
494 resultRelInfo = resultRelInfos;
495 foreach(l, resultRelations)
497 initResultRelInfo(resultRelInfo,
501 estate->es_instrument);
508 * Single result relation identified by parseTree->resultRelation
510 numResultRelations = 1;
511 resultRelInfos = (ResultRelInfo *) palloc(sizeof(ResultRelInfo));
512 initResultRelInfo(resultRelInfos,
513 parseTree->resultRelation,
516 estate->es_instrument);
519 estate->es_result_relations = resultRelInfos;
520 estate->es_num_result_relations = numResultRelations;
521 /* Initialize to first or only result rel */
522 estate->es_result_relation_info = resultRelInfos;
527 * if no result relation, then set state appropriately
529 estate->es_result_relations = NULL;
530 estate->es_num_result_relations = 0;
531 estate->es_result_relation_info = NULL;
535 * Detect whether we're doing SELECT INTO. If so, set the es_into_oids
536 * flag appropriately so that the plan tree will be initialized with the
537 * correct tuple descriptors.
539 do_select_into = false;
541 if (operation == CMD_SELECT && parseTree->into != NULL)
543 do_select_into = true;
544 estate->es_select_into = true;
545 estate->es_into_oids = parseTree->intoHasOids;
549 * Have to lock relations selected FOR UPDATE/FOR SHARE
551 estate->es_rowMarks = NIL;
552 estate->es_forUpdate = parseTree->forUpdate;
553 estate->es_rowNoWait = parseTree->rowNoWait;
554 if (parseTree->rowMarks != NIL)
558 foreach(l, parseTree->rowMarks)
560 Index rti = lfirst_int(l);
561 Oid relid = getrelid(rti, rangeTable);
565 relation = heap_open(relid, RowShareLock);
566 erm = (ExecRowMark *) palloc(sizeof(ExecRowMark));
567 erm->relation = relation;
569 snprintf(erm->resname, sizeof(erm->resname), "ctid%u", rti);
570 estate->es_rowMarks = lappend(estate->es_rowMarks, erm);
575 * initialize the executor "tuple" table. We need slots for all the plan
576 * nodes, plus possibly output slots for the junkfilter(s). At this point
577 * we aren't sure if we need junkfilters, so just add slots for them
578 * unconditionally. Also, if it's not a SELECT, set up a slot for use for
579 * trigger output tuples.
582 int nSlots = ExecCountSlotsNode(plan);
584 if (parseTree->resultRelations != NIL)
585 nSlots += list_length(parseTree->resultRelations);
588 if (operation != CMD_SELECT)
591 estate->es_tupleTable = ExecCreateTupleTable(nSlots);
593 if (operation != CMD_SELECT)
594 estate->es_trig_tuple_slot =
595 ExecAllocTableSlot(estate->es_tupleTable);
598 /* mark EvalPlanQual not active */
599 estate->es_topPlan = plan;
600 estate->es_evalPlanQual = NULL;
601 estate->es_evTupleNull = NULL;
602 estate->es_evTuple = NULL;
603 estate->es_useEvalPlan = false;
606 * initialize the private state information for all the nodes in the query
607 * tree. This opens files, allocates storage and leaves us ready to start
610 planstate = ExecInitNode(plan, estate);
613 * Get the tuple descriptor describing the type of tuples to return. (this
614 * is especially important if we are creating a relation with "SELECT
617 tupType = ExecGetResultType(planstate);
620 * Initialize the junk filter if needed. SELECT and INSERT queries need a
621 * filter if there are any junk attrs in the tlist. INSERT and SELECT
622 * INTO also need a filter if the plan may return raw disk tuples (else
623 * heap_insert will be scribbling on the source relation!). UPDATE and
624 * DELETE always need a filter, since there's always a junk 'ctid'
625 * attribute present --- no need to look first.
628 bool junk_filter_needed = false;
635 foreach(tlist, plan->targetlist)
637 TargetEntry *tle = (TargetEntry *) lfirst(tlist);
641 junk_filter_needed = true;
645 if (!junk_filter_needed &&
646 (operation == CMD_INSERT || do_select_into) &&
647 ExecMayReturnRawTuples(planstate))
648 junk_filter_needed = true;
652 junk_filter_needed = true;
658 if (junk_filter_needed)
661 * If there are multiple result relations, each one needs its own
662 * junk filter. Note this is only possible for UPDATE/DELETE, so
663 * we can't be fooled by some needing a filter and some not.
665 if (parseTree->resultRelations != NIL)
667 PlanState **appendplans;
669 ResultRelInfo *resultRelInfo;
672 /* Top plan had better be an Append here. */
673 Assert(IsA(plan, Append));
674 Assert(((Append *) plan)->isTarget);
675 Assert(IsA(planstate, AppendState));
676 appendplans = ((AppendState *) planstate)->appendplans;
677 as_nplans = ((AppendState *) planstate)->as_nplans;
678 Assert(as_nplans == estate->es_num_result_relations);
679 resultRelInfo = estate->es_result_relations;
680 for (i = 0; i < as_nplans; i++)
682 PlanState *subplan = appendplans[i];
685 j = ExecInitJunkFilter(subplan->plan->targetlist,
686 resultRelInfo->ri_RelationDesc->rd_att->tdhasoid,
687 ExecAllocTableSlot(estate->es_tupleTable));
688 resultRelInfo->ri_junkFilter = j;
693 * Set active junkfilter too; at this point ExecInitAppend has
694 * already selected an active result relation...
696 estate->es_junkFilter =
697 estate->es_result_relation_info->ri_junkFilter;
701 /* Normal case with just one JunkFilter */
704 j = ExecInitJunkFilter(planstate->plan->targetlist,
706 ExecAllocTableSlot(estate->es_tupleTable));
707 estate->es_junkFilter = j;
708 if (estate->es_result_relation_info)
709 estate->es_result_relation_info->ri_junkFilter = j;
711 /* For SELECT, want to return the cleaned tuple type */
712 if (operation == CMD_SELECT)
713 tupType = j->jf_cleanTupType;
717 estate->es_junkFilter = NULL;
721 * If doing SELECT INTO, initialize the "into" relation. We must wait
722 * till now so we have the "clean" result tuple type to create the new
725 * If EXPLAIN, skip creating the "into" relation.
727 intoRelationDesc = NULL;
729 if (do_select_into && !explainOnly)
738 * find namespace to create in, check permissions
740 intoName = parseTree->into->relname;
741 namespaceId = RangeVarGetCreationNamespace(parseTree->into);
743 aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
745 if (aclresult != ACLCHECK_OK)
746 aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
747 get_namespace_name(namespaceId));
750 * have to copy tupType to get rid of constraints
752 tupdesc = CreateTupleDescCopy(tupType);
754 intoRelationId = heap_create_with_catalog(intoName,
765 allowSystemTableMods);
767 FreeTupleDesc(tupdesc);
770 * Advance command counter so that the newly-created relation's
771 * catalog tuples will be visible to heap_open.
773 CommandCounterIncrement();
776 * If necessary, create a TOAST table for the into relation. Note that
777 * AlterTableCreateToastTable ends with CommandCounterIncrement(), so
778 * that the TOAST table will be visible for insertion.
780 AlterTableCreateToastTable(intoRelationId, true);
783 * And open the constructed table for writing.
785 intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
787 /* use_wal off requires rd_targblock be initially invalid */
788 Assert(intoRelationDesc->rd_targblock == InvalidBlockNumber);
791 * We can skip WAL-logging the insertions, unless PITR is in use.
793 * Note that for a non-temp INTO table, this is safe only because we
794 * know that the catalog changes above will have been WAL-logged, and
795 * so RecordTransactionCommit will think it needs to WAL-log the
796 * eventual transaction commit. Else the commit might be lost, even
797 * though all the data is safely fsync'd ...
799 estate->es_into_relation_use_wal = XLogArchivingActive();
802 estate->es_into_relation_descriptor = intoRelationDesc;
804 queryDesc->tupDesc = tupType;
805 queryDesc->planstate = planstate;
809 * Initialize ResultRelInfo data for one result relation
812 initResultRelInfo(ResultRelInfo *resultRelInfo,
813 Index resultRelationIndex,
818 Oid resultRelationOid;
819 Relation resultRelationDesc;
821 resultRelationOid = getrelid(resultRelationIndex, rangeTable);
822 resultRelationDesc = heap_open(resultRelationOid, RowExclusiveLock);
824 switch (resultRelationDesc->rd_rel->relkind)
826 case RELKIND_SEQUENCE:
828 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
829 errmsg("cannot change sequence \"%s\"",
830 RelationGetRelationName(resultRelationDesc))));
832 case RELKIND_TOASTVALUE:
834 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
835 errmsg("cannot change TOAST relation \"%s\"",
836 RelationGetRelationName(resultRelationDesc))));
840 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
841 errmsg("cannot change view \"%s\"",
842 RelationGetRelationName(resultRelationDesc))));
846 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
847 resultRelInfo->type = T_ResultRelInfo;
848 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
849 resultRelInfo->ri_RelationDesc = resultRelationDesc;
850 resultRelInfo->ri_NumIndices = 0;
851 resultRelInfo->ri_IndexRelationDescs = NULL;
852 resultRelInfo->ri_IndexRelationInfo = NULL;
853 /* make a copy so as not to depend on relcache info not changing... */
854 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
855 if (resultRelInfo->ri_TrigDesc)
857 int n = resultRelInfo->ri_TrigDesc->numtriggers;
859 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
860 palloc0(n * sizeof(FmgrInfo));
862 resultRelInfo->ri_TrigInstrument = InstrAlloc(n);
864 resultRelInfo->ri_TrigInstrument = NULL;
868 resultRelInfo->ri_TrigFunctions = NULL;
869 resultRelInfo->ri_TrigInstrument = NULL;
871 resultRelInfo->ri_ConstraintExprs = NULL;
872 resultRelInfo->ri_junkFilter = NULL;
875 * If there are indices on the result relation, open them and save
876 * descriptors in the result relation info, so that we can add new index
877 * entries for the tuples we add/update. We need not do this for a
878 * DELETE, however, since deletion doesn't affect indexes.
880 if (resultRelationDesc->rd_rel->relhasindex &&
881 operation != CMD_DELETE)
882 ExecOpenIndices(resultRelInfo);
886 * ExecContextForcesOids
888 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
889 * we need to ensure that result tuples have space for an OID iff they are
890 * going to be stored into a relation that has OIDs. In other contexts
891 * we are free to choose whether to leave space for OIDs in result tuples
892 * (we generally don't want to, but we do if a physical-tlist optimization
893 * is possible). This routine checks the plan context and returns TRUE if the
894 * choice is forced, FALSE if the choice is not forced. In the TRUE case,
895 * *hasoids is set to the required value.
897 * One reason this is ugly is that all plan nodes in the plan tree will emit
898 * tuples with space for an OID, though we really only need the topmost node
899 * to do so. However, node types like Sort don't project new tuples but just
900 * return their inputs, and in those cases the requirement propagates down
901 * to the input node. Eventually we might make this code smart enough to
902 * recognize how far down the requirement really goes, but for now we just
903 * make all plan nodes do the same thing if the top level forces the choice.
905 * We assume that estate->es_result_relation_info is already set up to
906 * describe the target relation. Note that in an UPDATE that spans an
907 * inheritance tree, some of the target relations may have OIDs and some not.
908 * We have to make the decisions on a per-relation basis as we initialize
909 * each of the child plans of the topmost Append plan.
911 * SELECT INTO is even uglier, because we don't have the INTO relation's
912 * descriptor available when this code runs; we have to look aside at a
913 * flag set by InitPlan().
916 ExecContextForcesOids(PlanState *planstate, bool *hasoids)
918 if (planstate->state->es_select_into)
920 *hasoids = planstate->state->es_into_oids;
925 ResultRelInfo *ri = planstate->state->es_result_relation_info;
929 Relation rel = ri->ri_RelationDesc;
933 *hasoids = rel->rd_rel->relhasoids;
942 /* ----------------------------------------------------------------
945 * Cleans up the query plan -- closes files and frees up storage
947 * NOTE: we are no longer very worried about freeing storage per se
948 * in this code; FreeExecutorState should be guaranteed to release all
949 * memory that needs to be released. What we are worried about doing
950 * is closing relations and dropping buffer pins. Thus, for example,
951 * tuple tables must be cleared or dropped to ensure pins are released.
952 * ----------------------------------------------------------------
955 ExecEndPlan(PlanState *planstate, EState *estate)
957 ResultRelInfo *resultRelInfo;
962 * shut down any PlanQual processing we were doing
964 if (estate->es_evalPlanQual != NULL)
965 EndEvalPlanQual(estate);
968 * shut down the node-type-specific query processing
970 ExecEndNode(planstate);
973 * destroy the executor "tuple" table.
975 ExecDropTupleTable(estate->es_tupleTable, true);
976 estate->es_tupleTable = NULL;
979 * close the result relation(s) if any, but hold locks until xact commit.
981 resultRelInfo = estate->es_result_relations;
982 for (i = estate->es_num_result_relations; i > 0; i--)
984 /* Close indices and then the relation itself */
985 ExecCloseIndices(resultRelInfo);
986 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
991 * close the "into" relation if necessary, again keeping lock
993 if (estate->es_into_relation_descriptor != NULL)
996 * If we skipped using WAL, and it's not a temp relation, we must
997 * force the relation down to disk before it's safe to commit the
998 * transaction. This requires forcing out any dirty buffers and then
999 * doing a forced fsync.
1001 if (!estate->es_into_relation_use_wal &&
1002 !estate->es_into_relation_descriptor->rd_istemp)
1004 FlushRelationBuffers(estate->es_into_relation_descriptor);
1005 /* FlushRelationBuffers will have opened rd_smgr */
1006 smgrimmedsync(estate->es_into_relation_descriptor->rd_smgr);
1009 heap_close(estate->es_into_relation_descriptor, NoLock);
1013 * close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
1015 foreach(l, estate->es_rowMarks)
1017 ExecRowMark *erm = lfirst(l);
1019 heap_close(erm->relation, NoLock);
1023 /* ----------------------------------------------------------------
1026 * processes the query plan to retrieve 'numberTuples' tuples in the
1027 * direction specified.
1029 * Retrieves all tuples if numberTuples is 0
1031 * result is either a slot containing the last tuple in the case
1032 * of a SELECT or NULL otherwise.
1034 * Note: the ctid attribute is a 'junk' attribute that is removed before the
1036 * ----------------------------------------------------------------
1038 static TupleTableSlot *
1039 ExecutePlan(EState *estate,
1040 PlanState *planstate,
1043 ScanDirection direction,
1046 JunkFilter *junkfilter;
1047 TupleTableSlot *slot;
1048 ItemPointer tupleid = NULL;
1049 ItemPointerData tuple_ctid;
1050 long current_tuple_count;
1051 TupleTableSlot *result;
1054 * initialize local variables
1057 current_tuple_count = 0;
1061 * Set the direction.
1063 estate->es_direction = direction;
1066 * Process BEFORE EACH STATEMENT triggers
1071 ExecBSUpdateTriggers(estate, estate->es_result_relation_info);
1074 ExecBSDeleteTriggers(estate, estate->es_result_relation_info);
1077 ExecBSInsertTriggers(estate, estate->es_result_relation_info);
1085 * Loop until we've processed the proper number of tuples from the plan.
1090 /* Reset the per-output-tuple exprcontext */
1091 ResetPerTupleExprContext(estate);
1094 * Execute the plan and obtain a tuple
1097 if (estate->es_useEvalPlan)
1099 slot = EvalPlanQualNext(estate);
1100 if (TupIsNull(slot))
1101 slot = ExecProcNode(planstate);
1104 slot = ExecProcNode(planstate);
1107 * if the tuple is null, then we assume there is nothing more to
1108 * process so we just return null...
1110 if (TupIsNull(slot))
1117 * if we have a junk filter, then project a new tuple with the junk
1120 * Store this new "clean" tuple in the junkfilter's resultSlot.
1121 * (Formerly, we stored it back over the "dirty" tuple, which is WRONG
1122 * because that tuple slot has the wrong descriptor.)
1124 * Also, extract all the junk information we need.
1126 if ((junkfilter = estate->es_junkFilter) != NULL)
1132 * extract the 'ctid' junk attribute.
1134 if (operation == CMD_UPDATE || operation == CMD_DELETE)
1136 if (!ExecGetJunkAttribute(junkfilter,
1141 elog(ERROR, "could not find junk ctid column");
1143 /* shouldn't ever get a null result... */
1145 elog(ERROR, "ctid is NULL");
1147 tupleid = (ItemPointer) DatumGetPointer(datum);
1148 tuple_ctid = *tupleid; /* make sure we don't free the ctid!! */
1149 tupleid = &tuple_ctid;
1153 * Process any FOR UPDATE or FOR SHARE locking requested.
1155 else if (estate->es_rowMarks != NIL)
1160 foreach(l, estate->es_rowMarks)
1162 ExecRowMark *erm = lfirst(l);
1163 HeapTupleData tuple;
1165 ItemPointerData update_ctid;
1166 TransactionId update_xmax;
1167 TupleTableSlot *newSlot;
1168 LockTupleMode lockmode;
1171 if (!ExecGetJunkAttribute(junkfilter,
1176 elog(ERROR, "could not find junk \"%s\" column",
1179 /* shouldn't ever get a null result... */
1181 elog(ERROR, "\"%s\" is NULL", erm->resname);
1183 tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1185 if (estate->es_forUpdate)
1186 lockmode = LockTupleExclusive;
1188 lockmode = LockTupleShared;
1190 test = heap_lock_tuple(erm->relation, &tuple, &buffer,
1191 &update_ctid, &update_xmax,
1192 estate->es_snapshot->curcid,
1193 lockmode, estate->es_rowNoWait);
1194 ReleaseBuffer(buffer);
1197 case HeapTupleSelfUpdated:
1198 /* treat it as deleted; do not process */
1201 case HeapTupleMayBeUpdated:
1204 case HeapTupleUpdated:
1205 if (IsXactIsoLevelSerializable)
1207 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1208 errmsg("could not serialize access due to concurrent update")));
1209 if (!ItemPointerEquals(&update_ctid,
1212 /* updated, so look at updated version */
1213 newSlot = EvalPlanQual(estate,
1217 estate->es_snapshot->curcid);
1218 if (!TupIsNull(newSlot))
1221 estate->es_useEvalPlan = true;
1227 * if tuple was deleted or PlanQual failed for
1228 * updated tuple - we must not return this tuple!
1233 elog(ERROR, "unrecognized heap_lock_tuple status: %u",
1241 * Finally create a new "clean" tuple with all junk attributes
1244 slot = ExecFilterJunk(junkfilter, slot);
1248 * now that we have a tuple, do the appropriate thing with it.. either
1249 * return it to the user, add it to a relation someplace, delete it
1250 * from a relation, or modify some of its attributes.
1255 ExecSelect(slot, /* slot containing tuple */
1256 dest, /* destination's tuple-receiver obj */
1262 ExecInsert(slot, tupleid, estate);
1267 ExecDelete(slot, tupleid, estate);
1272 ExecUpdate(slot, tupleid, estate);
1277 elog(ERROR, "unrecognized operation code: %d",
1284 * check our tuple count.. if we've processed the proper number then
1285 * quit, else loop again and process more tuples. Zero numberTuples
1288 current_tuple_count++;
1289 if (numberTuples && numberTuples == current_tuple_count)
1294 * Process AFTER EACH STATEMENT triggers
1299 ExecASUpdateTriggers(estate, estate->es_result_relation_info);
1302 ExecASDeleteTriggers(estate, estate->es_result_relation_info);
1305 ExecASInsertTriggers(estate, estate->es_result_relation_info);
1313 * here, result is either a slot containing a tuple in the case of a
1314 * SELECT or NULL otherwise.
1319 /* ----------------------------------------------------------------
1322 * SELECTs are easy.. we just pass the tuple to the appropriate
1323 * print function. The only complexity is when we do a
1324 * "SELECT INTO", in which case we insert the tuple into
1325 * the appropriate relation (note: this is a newly created relation
1326 * so we don't need to worry about indices or locks.)
1327 * ----------------------------------------------------------------
1330 ExecSelect(TupleTableSlot *slot,
1335 * insert the tuple into the "into relation"
1337 * XXX this probably ought to be replaced by a separate destination
1339 if (estate->es_into_relation_descriptor != NULL)
1343 tuple = ExecCopySlotTuple(slot);
1344 heap_insert(estate->es_into_relation_descriptor, tuple,
1345 estate->es_snapshot->curcid,
1346 estate->es_into_relation_use_wal,
1347 false); /* never any point in using FSM */
1348 /* we know there are no indexes to update */
1349 heap_freetuple(tuple);
1354 * send the tuple to the destination
1356 (*dest->receiveSlot) (slot, dest);
1358 (estate->es_processed)++;
1361 /* ----------------------------------------------------------------
1364 * INSERTs are trickier.. we have to insert the tuple into
1365 * the base relation and insert appropriate tuples into the
1367 * ----------------------------------------------------------------
1370 ExecInsert(TupleTableSlot *slot,
1371 ItemPointer tupleid,
1375 ResultRelInfo *resultRelInfo;
1376 Relation resultRelationDesc;
1380 * get the heap tuple out of the tuple table slot, making sure we have a
1383 tuple = ExecMaterializeSlot(slot);
1386 * get information on the (current) result relation
1388 resultRelInfo = estate->es_result_relation_info;
1389 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1391 /* BEFORE ROW INSERT Triggers */
1392 if (resultRelInfo->ri_TrigDesc &&
1393 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_INSERT] > 0)
1397 newtuple = ExecBRInsertTriggers(estate, resultRelInfo, tuple);
1399 if (newtuple == NULL) /* "do nothing" */
1402 if (newtuple != tuple) /* modified by Trigger(s) */
1405 * Put the modified tuple into a slot for convenience of routines
1406 * below. We assume the tuple was allocated in per-tuple memory
1407 * context, and therefore will go away by itself. The tuple table
1408 * slot should not try to clear it.
1410 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1412 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1413 ExecSetSlotDescriptor(newslot,
1414 slot->tts_tupleDescriptor,
1416 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1423 * Check the constraints of the tuple
1425 if (resultRelationDesc->rd_att->constr)
1426 ExecConstraints(resultRelInfo, slot, estate);
1431 * Note: heap_insert returns the tid (location) of the new tuple in the
1434 newId = heap_insert(resultRelationDesc, tuple,
1435 estate->es_snapshot->curcid,
1439 (estate->es_processed)++;
1440 estate->es_lastoid = newId;
1441 setLastTid(&(tuple->t_self));
1444 * insert index entries for tuple
1446 if (resultRelInfo->ri_NumIndices > 0)
1447 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1449 /* AFTER ROW INSERT Triggers */
1450 ExecARInsertTriggers(estate, resultRelInfo, tuple);
1453 /* ----------------------------------------------------------------
1456 * DELETE is like UPDATE, except that we delete the tuple and no
1457 * index modifications are needed
1458 * ----------------------------------------------------------------
1461 ExecDelete(TupleTableSlot *slot,
1462 ItemPointer tupleid,
1465 ResultRelInfo *resultRelInfo;
1466 Relation resultRelationDesc;
1468 ItemPointerData update_ctid;
1469 TransactionId update_xmax;
1472 * get information on the (current) result relation
1474 resultRelInfo = estate->es_result_relation_info;
1475 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1477 /* BEFORE ROW DELETE Triggers */
1478 if (resultRelInfo->ri_TrigDesc &&
1479 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_DELETE] > 0)
1483 dodelete = ExecBRDeleteTriggers(estate, resultRelInfo, tupleid,
1484 estate->es_snapshot->curcid);
1486 if (!dodelete) /* "do nothing" */
1493 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1494 * the row to be deleted is visible to that snapshot, and throw a can't-
1495 * serialize error if not. This is a special-case behavior needed for
1496 * referential integrity updates in serializable transactions.
1499 result = heap_delete(resultRelationDesc, tupleid,
1500 &update_ctid, &update_xmax,
1501 estate->es_snapshot->curcid,
1502 estate->es_crosscheck_snapshot,
1503 true /* wait for commit */ );
1506 case HeapTupleSelfUpdated:
1507 /* already deleted by self; nothing to do */
1510 case HeapTupleMayBeUpdated:
1513 case HeapTupleUpdated:
1514 if (IsXactIsoLevelSerializable)
1516 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1517 errmsg("could not serialize access due to concurrent update")));
1518 else if (!ItemPointerEquals(tupleid, &update_ctid))
1520 TupleTableSlot *epqslot;
1522 epqslot = EvalPlanQual(estate,
1523 resultRelInfo->ri_RangeTableIndex,
1526 estate->es_snapshot->curcid);
1527 if (!TupIsNull(epqslot))
1529 *tupleid = update_ctid;
1533 /* tuple already deleted; nothing to do */
1537 elog(ERROR, "unrecognized heap_delete status: %u", result);
1542 (estate->es_processed)++;
1545 * Note: Normally one would think that we have to delete index tuples
1546 * associated with the heap tuple now...
1548 * ... but in POSTGRES, we have no need to do this because VACUUM will
1549 * take care of it later. We can't delete index tuples immediately
1550 * anyway, since the tuple is still visible to other transactions.
1553 /* AFTER ROW DELETE Triggers */
1554 ExecARDeleteTriggers(estate, resultRelInfo, tupleid);
1557 /* ----------------------------------------------------------------
1560 * note: we can't run UPDATE queries with transactions
1561 * off because UPDATEs are actually INSERTs and our
1562 * scan will mistakenly loop forever, updating the tuple
1563 * it just inserted.. This should be fixed but until it
1564 * is, we don't want to get stuck in an infinite loop
1565 * which corrupts your database..
1566 * ----------------------------------------------------------------
1569 ExecUpdate(TupleTableSlot *slot,
1570 ItemPointer tupleid,
1574 ResultRelInfo *resultRelInfo;
1575 Relation resultRelationDesc;
1577 ItemPointerData update_ctid;
1578 TransactionId update_xmax;
1581 * abort the operation if not running transactions
1583 if (IsBootstrapProcessingMode())
1584 elog(ERROR, "cannot UPDATE during bootstrap");
1587 * get the heap tuple out of the tuple table slot, making sure we have a
1590 tuple = ExecMaterializeSlot(slot);
1593 * get information on the (current) result relation
1595 resultRelInfo = estate->es_result_relation_info;
1596 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1598 /* BEFORE ROW UPDATE Triggers */
1599 if (resultRelInfo->ri_TrigDesc &&
1600 resultRelInfo->ri_TrigDesc->n_before_row[TRIGGER_EVENT_UPDATE] > 0)
1604 newtuple = ExecBRUpdateTriggers(estate, resultRelInfo,
1606 estate->es_snapshot->curcid);
1608 if (newtuple == NULL) /* "do nothing" */
1611 if (newtuple != tuple) /* modified by Trigger(s) */
1614 * Put the modified tuple into a slot for convenience of routines
1615 * below. We assume the tuple was allocated in per-tuple memory
1616 * context, and therefore will go away by itself. The tuple table
1617 * slot should not try to clear it.
1619 TupleTableSlot *newslot = estate->es_trig_tuple_slot;
1621 if (newslot->tts_tupleDescriptor != slot->tts_tupleDescriptor)
1622 ExecSetSlotDescriptor(newslot,
1623 slot->tts_tupleDescriptor,
1625 ExecStoreTuple(newtuple, newslot, InvalidBuffer, false);
1632 * Check the constraints of the tuple
1634 * If we generate a new candidate tuple after EvalPlanQual testing, we
1635 * must loop back here and recheck constraints. (We don't need to redo
1636 * triggers, however. If there are any BEFORE triggers then trigger.c
1637 * will have done heap_lock_tuple to lock the correct tuple, so there's no
1638 * need to do them again.)
1641 if (resultRelationDesc->rd_att->constr)
1642 ExecConstraints(resultRelInfo, slot, estate);
1645 * replace the heap tuple
1647 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check that
1648 * the row to be updated is visible to that snapshot, and throw a can't-
1649 * serialize error if not. This is a special-case behavior needed for
1650 * referential integrity updates in serializable transactions.
1652 result = heap_update(resultRelationDesc, tupleid, tuple,
1653 &update_ctid, &update_xmax,
1654 estate->es_snapshot->curcid,
1655 estate->es_crosscheck_snapshot,
1656 true /* wait for commit */ );
1659 case HeapTupleSelfUpdated:
1660 /* already deleted by self; nothing to do */
1663 case HeapTupleMayBeUpdated:
1666 case HeapTupleUpdated:
1667 if (IsXactIsoLevelSerializable)
1669 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1670 errmsg("could not serialize access due to concurrent update")));
1671 else if (!ItemPointerEquals(tupleid, &update_ctid))
1673 TupleTableSlot *epqslot;
1675 epqslot = EvalPlanQual(estate,
1676 resultRelInfo->ri_RangeTableIndex,
1679 estate->es_snapshot->curcid);
1680 if (!TupIsNull(epqslot))
1682 *tupleid = update_ctid;
1683 slot = ExecFilterJunk(estate->es_junkFilter, epqslot);
1684 tuple = ExecMaterializeSlot(slot);
1688 /* tuple already deleted; nothing to do */
1692 elog(ERROR, "unrecognized heap_update status: %u", result);
1697 (estate->es_processed)++;
1700 * Note: instead of having to update the old index tuples associated with
1701 * the heap tuple, all we do is form and insert new index tuples. This is
1702 * because UPDATEs are actually DELETEs and INSERTs, and index tuple
1703 * deletion is done later by VACUUM (see notes in ExecDelete). All we do
1704 * here is insert new index tuples. -cim 9/27/89
1708 * insert index entries for tuple
1710 * Note: heap_update returns the tid (location) of the new tuple in the
1713 if (resultRelInfo->ri_NumIndices > 0)
1714 ExecInsertIndexTuples(slot, &(tuple->t_self), estate, false);
1716 /* AFTER ROW UPDATE Triggers */
1717 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, tuple);
1721 ExecRelCheck(ResultRelInfo *resultRelInfo,
1722 TupleTableSlot *slot, EState *estate)
1724 Relation rel = resultRelInfo->ri_RelationDesc;
1725 int ncheck = rel->rd_att->constr->num_check;
1726 ConstrCheck *check = rel->rd_att->constr->check;
1727 ExprContext *econtext;
1728 MemoryContext oldContext;
1733 * If first time through for this result relation, build expression
1734 * nodetrees for rel's constraint expressions. Keep them in the per-query
1735 * memory context so they'll survive throughout the query.
1737 if (resultRelInfo->ri_ConstraintExprs == NULL)
1739 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1740 resultRelInfo->ri_ConstraintExprs =
1741 (List **) palloc(ncheck * sizeof(List *));
1742 for (i = 0; i < ncheck; i++)
1744 /* ExecQual wants implicit-AND form */
1745 qual = make_ands_implicit(stringToNode(check[i].ccbin));
1746 resultRelInfo->ri_ConstraintExprs[i] = (List *)
1747 ExecPrepareExpr((Expr *) qual, estate);
1749 MemoryContextSwitchTo(oldContext);
1753 * We will use the EState's per-tuple context for evaluating constraint
1754 * expressions (creating it if it's not already there).
1756 econtext = GetPerTupleExprContext(estate);
1758 /* Arrange for econtext's scan tuple to be the tuple under test */
1759 econtext->ecxt_scantuple = slot;
1761 /* And evaluate the constraints */
1762 for (i = 0; i < ncheck; i++)
1764 qual = resultRelInfo->ri_ConstraintExprs[i];
1767 * NOTE: SQL92 specifies that a NULL result from a constraint
1768 * expression is not to be treated as a failure. Therefore, tell
1769 * ExecQual to return TRUE for NULL.
1771 if (!ExecQual(qual, econtext, true))
1772 return check[i].ccname;
1775 /* NULL result means no error */
1780 ExecConstraints(ResultRelInfo *resultRelInfo,
1781 TupleTableSlot *slot, EState *estate)
1783 Relation rel = resultRelInfo->ri_RelationDesc;
1784 TupleConstr *constr = rel->rd_att->constr;
1788 if (constr->has_not_null)
1790 int natts = rel->rd_att->natts;
1793 for (attrChk = 1; attrChk <= natts; attrChk++)
1795 if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1796 slot_attisnull(slot, attrChk))
1798 (errcode(ERRCODE_NOT_NULL_VIOLATION),
1799 errmsg("null value in column \"%s\" violates not-null constraint",
1800 NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1804 if (constr->num_check > 0)
1808 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1810 (errcode(ERRCODE_CHECK_VIOLATION),
1811 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1812 RelationGetRelationName(rel), failed)));
1817 * Check a modified tuple to see if we want to process its updated version
1818 * under READ COMMITTED rules.
1820 * See backend/executor/README for some info about how this works.
1822 * estate - executor state data
1823 * rti - rangetable index of table containing tuple
1824 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1825 * priorXmax - t_xmax from the outdated tuple
1826 * curCid - command ID of current command of my transaction
1828 * *tid is also an output parameter: it's modified to hold the TID of the
1829 * latest version of the tuple (note this may be changed even on failure)
1831 * Returns a slot containing the new candidate update/delete tuple, or
1832 * NULL if we determine we shouldn't process the row.
1835 EvalPlanQual(EState *estate, Index rti,
1836 ItemPointer tid, TransactionId priorXmax, CommandId curCid)
1841 HeapTupleData tuple;
1842 HeapTuple copyTuple = NULL;
1848 * find relation containing target tuple
1850 if (estate->es_result_relation_info != NULL &&
1851 estate->es_result_relation_info->ri_RangeTableIndex == rti)
1852 relation = estate->es_result_relation_info->ri_RelationDesc;
1858 foreach(l, estate->es_rowMarks)
1860 if (((ExecRowMark *) lfirst(l))->rti == rti)
1862 relation = ((ExecRowMark *) lfirst(l))->relation;
1866 if (relation == NULL)
1867 elog(ERROR, "could not find RowMark for RT index %u", rti);
1873 * Loop here to deal with updated or busy tuples
1875 tuple.t_self = *tid;
1880 if (heap_fetch(relation, SnapshotDirty, &tuple, &buffer, true, NULL))
1883 * If xmin isn't what we're expecting, the slot must have been
1884 * recycled and reused for an unrelated tuple. This implies that
1885 * the latest version of the row was deleted, so we need do
1886 * nothing. (Should be safe to examine xmin without getting
1887 * buffer's content lock, since xmin never changes in an existing
1890 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1893 ReleaseBuffer(buffer);
1897 /* otherwise xmin should not be dirty... */
1898 if (TransactionIdIsValid(SnapshotDirty->xmin))
1899 elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1902 * If tuple is being updated by other transaction then we have to
1903 * wait for its commit/abort.
1905 if (TransactionIdIsValid(SnapshotDirty->xmax))
1907 ReleaseBuffer(buffer);
1908 XactLockTableWait(SnapshotDirty->xmax);
1909 continue; /* loop back to repeat heap_fetch */
1913 * If tuple was inserted by our own transaction, we have to check
1914 * cmin against curCid: cmin >= curCid means our command cannot
1915 * see the tuple, so we should ignore it. Without this we are
1916 * open to the "Halloween problem" of indefinitely re-updating
1917 * the same tuple. (We need not check cmax because
1918 * HeapTupleSatisfiesDirty will consider a tuple deleted by
1919 * our transaction dead, regardless of cmax.) We just checked
1920 * that priorXmax == xmin, so we can test that variable instead
1921 * of doing HeapTupleHeaderGetXmin again.
1923 if (TransactionIdIsCurrentTransactionId(priorXmax) &&
1924 HeapTupleHeaderGetCmin(tuple.t_data) >= curCid)
1926 ReleaseBuffer(buffer);
1931 * We got tuple - now copy it for use by recheck query.
1933 copyTuple = heap_copytuple(&tuple);
1934 ReleaseBuffer(buffer);
1939 * If the referenced slot was actually empty, the latest version of
1940 * the row must have been deleted, so we need do nothing.
1942 if (tuple.t_data == NULL)
1944 ReleaseBuffer(buffer);
1949 * As above, if xmin isn't what we're expecting, do nothing.
1951 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1954 ReleaseBuffer(buffer);
1959 * If we get here, the tuple was found but failed SnapshotDirty.
1960 * Assuming the xmin is either a committed xact or our own xact (as it
1961 * certainly should be if we're trying to modify the tuple), this must
1962 * mean that the row was updated or deleted by either a committed xact
1963 * or our own xact. If it was deleted, we can ignore it; if it was
1964 * updated then chain up to the next version and repeat the whole
1967 * As above, it should be safe to examine xmax and t_ctid without the
1968 * buffer content lock, because they can't be changing.
1970 if (ItemPointerEquals(&tuple.t_self, &tuple.t_data->t_ctid))
1972 /* deleted, so forget about it */
1973 ReleaseBuffer(buffer);
1977 /* updated, so look at the updated row */
1978 tuple.t_self = tuple.t_data->t_ctid;
1979 /* updated row should have xmin matching this xmax */
1980 priorXmax = HeapTupleHeaderGetXmax(tuple.t_data);
1981 ReleaseBuffer(buffer);
1982 /* loop back to fetch next in chain */
1986 * For UPDATE/DELETE we have to return tid of actual row we're executing
1989 *tid = tuple.t_self;
1992 * Need to run a recheck subquery. Find or create a PQ stack entry.
1994 epq = estate->es_evalPlanQual;
1997 if (epq != NULL && epq->rti == 0)
1999 /* Top PQ stack entry is idle, so re-use it */
2000 Assert(!(estate->es_useEvalPlan) && epq->next == NULL);
2006 * If this is request for another RTE - Ra, - then we have to check wasn't
2007 * PlanQual requested for Ra already and if so then Ra' row was updated
2008 * again and we have to re-start old execution for Ra and forget all what
2009 * we done after Ra was suspended. Cool? -:))
2011 if (epq != NULL && epq->rti != rti &&
2012 epq->estate->es_evTuple[rti - 1] != NULL)
2016 evalPlanQual *oldepq;
2018 /* stop execution */
2019 EvalPlanQualStop(epq);
2020 /* pop previous PlanQual from the stack */
2022 Assert(oldepq && oldepq->rti != 0);
2023 /* push current PQ to freePQ stack */
2026 estate->es_evalPlanQual = epq;
2027 } while (epq->rti != rti);
2031 * If we are requested for another RTE then we have to suspend execution
2032 * of current PlanQual and start execution for new one.
2034 if (epq == NULL || epq->rti != rti)
2036 /* try to reuse plan used previously */
2037 evalPlanQual *newepq = (epq != NULL) ? epq->free : NULL;
2039 if (newepq == NULL) /* first call or freePQ stack is empty */
2041 newepq = (evalPlanQual *) palloc0(sizeof(evalPlanQual));
2042 newepq->free = NULL;
2043 newepq->estate = NULL;
2044 newepq->planstate = NULL;
2048 /* recycle previously used PlanQual */
2049 Assert(newepq->estate == NULL);
2052 /* push current PQ to the stack */
2055 estate->es_evalPlanQual = epq;
2060 Assert(epq->rti == rti);
2063 * Ok - we're requested for the same RTE. Unfortunately we still have to
2064 * end and restart execution of the plan, because ExecReScan wouldn't
2065 * ensure that upper plan nodes would reset themselves. We could make
2066 * that work if insertion of the target tuple were integrated with the
2067 * Param mechanism somehow, so that the upper plan nodes know that their
2068 * children's outputs have changed.
2070 * Note that the stack of free evalPlanQual nodes is quite useless at the
2071 * moment, since it only saves us from pallocing/releasing the
2072 * evalPlanQual nodes themselves. But it will be useful once we implement
2073 * ReScan instead of end/restart for re-using PlanQual nodes.
2077 /* stop execution */
2078 EvalPlanQualStop(epq);
2082 * Initialize new recheck query.
2084 * Note: if we were re-using PlanQual plans via ExecReScan, we'd need to
2085 * instead copy down changeable state from the top plan (including
2086 * es_result_relation_info, es_junkFilter) and reset locally changeable
2087 * state in the epq (including es_param_exec_vals, es_evTupleNull).
2089 EvalPlanQualStart(epq, estate, epq->next);
2092 * free old RTE' tuple, if any, and store target tuple where relation's
2093 * scan node will see it
2095 epqstate = epq->estate;
2096 if (epqstate->es_evTuple[rti - 1] != NULL)
2097 heap_freetuple(epqstate->es_evTuple[rti - 1]);
2098 epqstate->es_evTuple[rti - 1] = copyTuple;
2100 return EvalPlanQualNext(estate);
2103 static TupleTableSlot *
2104 EvalPlanQualNext(EState *estate)
2106 evalPlanQual *epq = estate->es_evalPlanQual;
2107 MemoryContext oldcontext;
2108 TupleTableSlot *slot;
2110 Assert(epq->rti != 0);
2113 oldcontext = MemoryContextSwitchTo(epq->estate->es_query_cxt);
2114 slot = ExecProcNode(epq->planstate);
2115 MemoryContextSwitchTo(oldcontext);
2118 * No more tuples for this PQ. Continue previous one.
2120 if (TupIsNull(slot))
2122 evalPlanQual *oldepq;
2124 /* stop execution */
2125 EvalPlanQualStop(epq);
2126 /* pop old PQ from the stack */
2130 /* this is the first (oldest) PQ - mark as free */
2132 estate->es_useEvalPlan = false;
2133 /* and continue Query execution */
2136 Assert(oldepq->rti != 0);
2137 /* push current PQ to freePQ stack */
2140 estate->es_evalPlanQual = epq;
2148 EndEvalPlanQual(EState *estate)
2150 evalPlanQual *epq = estate->es_evalPlanQual;
2152 if (epq->rti == 0) /* plans already shutdowned */
2154 Assert(epq->next == NULL);
2160 evalPlanQual *oldepq;
2162 /* stop execution */
2163 EvalPlanQualStop(epq);
2164 /* pop old PQ from the stack */
2168 /* this is the first (oldest) PQ - mark as free */
2170 estate->es_useEvalPlan = false;
2173 Assert(oldepq->rti != 0);
2174 /* push current PQ to freePQ stack */
2177 estate->es_evalPlanQual = epq;
2182 * Start execution of one level of PlanQual.
2184 * This is a cut-down version of ExecutorStart(): we copy some state from
2185 * the top-level estate rather than initializing it fresh.
2188 EvalPlanQualStart(evalPlanQual *epq, EState *estate, evalPlanQual *priorepq)
2192 MemoryContext oldcontext;
2194 rtsize = list_length(estate->es_range_table);
2196 epq->estate = epqstate = CreateExecutorState();
2198 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2201 * The epqstates share the top query's copy of unchanging state such as
2202 * the snapshot, rangetable, result-rel info, and external Param info.
2203 * They need their own copies of local state, including a tuple table,
2204 * es_param_exec_vals, etc.
2206 epqstate->es_direction = ForwardScanDirection;
2207 epqstate->es_snapshot = estate->es_snapshot;
2208 epqstate->es_crosscheck_snapshot = estate->es_crosscheck_snapshot;
2209 epqstate->es_range_table = estate->es_range_table;
2210 epqstate->es_result_relations = estate->es_result_relations;
2211 epqstate->es_num_result_relations = estate->es_num_result_relations;
2212 epqstate->es_result_relation_info = estate->es_result_relation_info;
2213 epqstate->es_junkFilter = estate->es_junkFilter;
2214 epqstate->es_into_relation_descriptor = estate->es_into_relation_descriptor;
2215 epqstate->es_into_relation_use_wal = estate->es_into_relation_use_wal;
2216 epqstate->es_param_list_info = estate->es_param_list_info;
2217 if (estate->es_topPlan->nParamExec > 0)
2218 epqstate->es_param_exec_vals = (ParamExecData *)
2219 palloc0(estate->es_topPlan->nParamExec * sizeof(ParamExecData));
2220 epqstate->es_rowMarks = estate->es_rowMarks;
2221 epqstate->es_forUpdate = estate->es_forUpdate;
2222 epqstate->es_rowNoWait = estate->es_rowNoWait;
2223 epqstate->es_instrument = estate->es_instrument;
2224 epqstate->es_select_into = estate->es_select_into;
2225 epqstate->es_into_oids = estate->es_into_oids;
2226 epqstate->es_topPlan = estate->es_topPlan;
2229 * Each epqstate must have its own es_evTupleNull state, but all the stack
2230 * entries share es_evTuple state. This allows sub-rechecks to inherit
2231 * the value being examined by an outer recheck.
2233 epqstate->es_evTupleNull = (bool *) palloc0(rtsize * sizeof(bool));
2234 if (priorepq == NULL)
2235 /* first PQ stack entry */
2236 epqstate->es_evTuple = (HeapTuple *)
2237 palloc0(rtsize * sizeof(HeapTuple));
2239 /* later stack entries share the same storage */
2240 epqstate->es_evTuple = priorepq->estate->es_evTuple;
2242 epqstate->es_tupleTable =
2243 ExecCreateTupleTable(estate->es_tupleTable->size);
2245 epq->planstate = ExecInitNode(estate->es_topPlan, epqstate);
2247 MemoryContextSwitchTo(oldcontext);
2251 * End execution of one level of PlanQual.
2253 * This is a cut-down version of ExecutorEnd(); basically we want to do most
2254 * of the normal cleanup, but *not* close result relations (which we are
2255 * just sharing from the outer query).
2258 EvalPlanQualStop(evalPlanQual *epq)
2260 EState *epqstate = epq->estate;
2261 MemoryContext oldcontext;
2263 oldcontext = MemoryContextSwitchTo(epqstate->es_query_cxt);
2265 ExecEndNode(epq->planstate);
2267 ExecDropTupleTable(epqstate->es_tupleTable, true);
2268 epqstate->es_tupleTable = NULL;
2270 if (epqstate->es_evTuple[epq->rti - 1] != NULL)
2272 heap_freetuple(epqstate->es_evTuple[epq->rti - 1]);
2273 epqstate->es_evTuple[epq->rti - 1] = NULL;
2276 MemoryContextSwitchTo(oldcontext);
2278 FreeExecutorState(epqstate);
2281 epq->planstate = NULL;