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11 * The old ExecutorMain() has been replaced by ExecutorStart(),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) 1996-2011, PostgreSQL Global Development Group
25 * Portions Copyright (c) 1994, Regents of the University of California
29 * src/backend/executor/execMain.c
31 *-------------------------------------------------------------------------
35 #include "access/reloptions.h"
36 #include "access/sysattr.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 "miscadmin.h"
47 #include "optimizer/clauses.h"
48 #include "parser/parse_clause.h"
49 #include "parser/parsetree.h"
50 #include "storage/bufmgr.h"
51 #include "storage/lmgr.h"
52 #include "storage/smgr.h"
53 #include "tcop/utility.h"
54 #include "utils/acl.h"
55 #include "utils/lsyscache.h"
56 #include "utils/memutils.h"
57 #include "utils/snapmgr.h"
58 #include "utils/tqual.h"
61 /* Hooks for plugins to get control in ExecutorStart/Run/End() */
62 ExecutorStart_hook_type ExecutorStart_hook = NULL;
63 ExecutorRun_hook_type ExecutorRun_hook = NULL;
64 ExecutorEnd_hook_type ExecutorEnd_hook = NULL;
66 /* Hook for plugin to get control in ExecCheckRTPerms() */
67 ExecutorCheckPerms_hook_type ExecutorCheckPerms_hook = NULL;
69 /* decls for local routines only used within this module */
70 static void InitPlan(QueryDesc *queryDesc, int eflags);
71 static void ExecEndPlan(PlanState *planstate, EState *estate);
72 static void ExecutePlan(EState *estate, PlanState *planstate,
76 ScanDirection direction,
78 static bool ExecCheckRTEPerms(RangeTblEntry *rte);
79 static void ExecCheckXactReadOnly(PlannedStmt *plannedstmt);
80 static void EvalPlanQualStart(EPQState *epqstate, EState *parentestate,
82 static void OpenIntoRel(QueryDesc *queryDesc);
83 static void CloseIntoRel(QueryDesc *queryDesc);
84 static void intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo);
85 static void intorel_receive(TupleTableSlot *slot, DestReceiver *self);
86 static void intorel_shutdown(DestReceiver *self);
87 static void intorel_destroy(DestReceiver *self);
89 /* end of local decls */
92 /* ----------------------------------------------------------------
95 * This routine must be called at the beginning of any execution of any
98 * Takes a QueryDesc previously created by CreateQueryDesc (it's not real
99 * clear why we bother to separate the two functions, but...). The tupDesc
100 * field of the QueryDesc is filled in to describe the tuples that will be
101 * returned, and the internal fields (estate and planstate) are set up.
103 * eflags contains flag bits as described in executor.h.
105 * NB: the CurrentMemoryContext when this is called will become the parent
106 * of the per-query context used for this Executor invocation.
108 * We provide a function hook variable that lets loadable plugins
109 * get control when ExecutorStart is called. Such a plugin would
110 * normally call standard_ExecutorStart().
112 * ----------------------------------------------------------------
115 ExecutorStart(QueryDesc *queryDesc, int eflags)
117 if (ExecutorStart_hook)
118 (*ExecutorStart_hook) (queryDesc, eflags);
120 standard_ExecutorStart(queryDesc, eflags);
124 standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
127 MemoryContext oldcontext;
129 /* sanity checks: queryDesc must not be started already */
130 Assert(queryDesc != NULL);
131 Assert(queryDesc->estate == NULL);
134 * If the transaction is read-only, we need to check if any writes are
135 * planned to non-temporary tables. EXPLAIN is considered read-only.
137 if (XactReadOnly && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
138 ExecCheckXactReadOnly(queryDesc->plannedstmt);
141 * Build EState, switch into per-query memory context for startup.
143 estate = CreateExecutorState();
144 queryDesc->estate = estate;
146 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
149 * Fill in external parameters, if any, from queryDesc; and allocate
150 * workspace for internal parameters
152 estate->es_param_list_info = queryDesc->params;
154 if (queryDesc->plannedstmt->nParamExec > 0)
155 estate->es_param_exec_vals = (ParamExecData *)
156 palloc0(queryDesc->plannedstmt->nParamExec * sizeof(ParamExecData));
159 * If non-read-only query, set the command ID to mark output tuples with
161 switch (queryDesc->operation)
164 /* SELECT INTO and SELECT FOR UPDATE/SHARE need to mark tuples */
165 if (queryDesc->plannedstmt->intoClause != NULL ||
166 queryDesc->plannedstmt->rowMarks != NIL)
167 estate->es_output_cid = GetCurrentCommandId(true);
173 estate->es_output_cid = GetCurrentCommandId(true);
177 elog(ERROR, "unrecognized operation code: %d",
178 (int) queryDesc->operation);
183 * Copy other important information into the EState
185 estate->es_snapshot = RegisterSnapshot(queryDesc->snapshot);
186 estate->es_crosscheck_snapshot = RegisterSnapshot(queryDesc->crosscheck_snapshot);
187 estate->es_instrument = queryDesc->instrument_options;
190 * Initialize the plan state tree
192 InitPlan(queryDesc, eflags);
194 MemoryContextSwitchTo(oldcontext);
197 /* ----------------------------------------------------------------
200 * This is the main routine of the executor module. It accepts
201 * the query descriptor from the traffic cop and executes the
204 * ExecutorStart must have been called already.
206 * If direction is NoMovementScanDirection then nothing is done
207 * except to start up/shut down the destination. Otherwise,
208 * we retrieve up to 'count' tuples in the specified direction.
210 * Note: count = 0 is interpreted as no portal limit, i.e., run to
213 * There is no return value, but output tuples (if any) are sent to
214 * the destination receiver specified in the QueryDesc; and the number
215 * of tuples processed at the top level can be found in
216 * estate->es_processed.
218 * We provide a function hook variable that lets loadable plugins
219 * get control when ExecutorRun is called. Such a plugin would
220 * normally call standard_ExecutorRun().
222 * ----------------------------------------------------------------
225 ExecutorRun(QueryDesc *queryDesc,
226 ScanDirection direction, long count)
228 if (ExecutorRun_hook)
229 (*ExecutorRun_hook) (queryDesc, direction, count);
231 standard_ExecutorRun(queryDesc, direction, count);
235 standard_ExecutorRun(QueryDesc *queryDesc,
236 ScanDirection direction, long count)
242 MemoryContext oldcontext;
245 Assert(queryDesc != NULL);
247 estate = queryDesc->estate;
249 Assert(estate != NULL);
252 * Switch into per-query memory context
254 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
256 /* Allow instrumentation of ExecutorRun overall runtime */
257 if (queryDesc->totaltime)
258 InstrStartNode(queryDesc->totaltime);
261 * extract information from the query descriptor and the query feature.
263 operation = queryDesc->operation;
264 dest = queryDesc->dest;
267 * startup tuple receiver, if we will be emitting tuples
269 estate->es_processed = 0;
270 estate->es_lastoid = InvalidOid;
272 sendTuples = (operation == CMD_SELECT ||
273 queryDesc->plannedstmt->hasReturning);
276 (*dest->rStartup) (dest, operation, queryDesc->tupDesc);
281 if (!ScanDirectionIsNoMovement(direction))
283 queryDesc->planstate,
291 * shutdown tuple receiver, if we started it
294 (*dest->rShutdown) (dest);
296 if (queryDesc->totaltime)
297 InstrStopNode(queryDesc->totaltime, estate->es_processed);
299 MemoryContextSwitchTo(oldcontext);
302 /* ----------------------------------------------------------------
305 * This routine must be called at the end of execution of any
308 * We provide a function hook variable that lets loadable plugins
309 * get control when ExecutorEnd is called. Such a plugin would
310 * normally call standard_ExecutorEnd().
312 * ----------------------------------------------------------------
315 ExecutorEnd(QueryDesc *queryDesc)
317 if (ExecutorEnd_hook)
318 (*ExecutorEnd_hook) (queryDesc);
320 standard_ExecutorEnd(queryDesc);
324 standard_ExecutorEnd(QueryDesc *queryDesc)
327 MemoryContext oldcontext;
330 Assert(queryDesc != NULL);
332 estate = queryDesc->estate;
334 Assert(estate != NULL);
337 * Switch into per-query memory context to run ExecEndPlan
339 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
341 ExecEndPlan(queryDesc->planstate, estate);
344 * Close the SELECT INTO relation if any
346 if (estate->es_select_into)
347 CloseIntoRel(queryDesc);
349 /* do away with our snapshots */
350 UnregisterSnapshot(estate->es_snapshot);
351 UnregisterSnapshot(estate->es_crosscheck_snapshot);
354 * Must switch out of context before destroying it
356 MemoryContextSwitchTo(oldcontext);
359 * Release EState and per-query memory context. This should release
360 * everything the executor has allocated.
362 FreeExecutorState(estate);
364 /* Reset queryDesc fields that no longer point to anything */
365 queryDesc->tupDesc = NULL;
366 queryDesc->estate = NULL;
367 queryDesc->planstate = NULL;
368 queryDesc->totaltime = NULL;
371 /* ----------------------------------------------------------------
374 * This routine may be called on an open queryDesc to rewind it
376 * ----------------------------------------------------------------
379 ExecutorRewind(QueryDesc *queryDesc)
382 MemoryContext oldcontext;
385 Assert(queryDesc != NULL);
387 estate = queryDesc->estate;
389 Assert(estate != NULL);
391 /* It's probably not sensible to rescan updating queries */
392 Assert(queryDesc->operation == CMD_SELECT);
395 * Switch into per-query memory context
397 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
402 ExecReScan(queryDesc->planstate);
404 MemoryContextSwitchTo(oldcontext);
410 * Check access permissions for all relations listed in a range table.
412 * Returns true if permissions are adequate. Otherwise, throws an appropriate
413 * error if ereport_on_violation is true, or simply returns false otherwise.
416 ExecCheckRTPerms(List *rangeTable, bool ereport_on_violation)
421 foreach(l, rangeTable)
423 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
425 result = ExecCheckRTEPerms(rte);
428 Assert(rte->rtekind == RTE_RELATION);
429 if (ereport_on_violation)
430 aclcheck_error(ACLCHECK_NO_PRIV, ACL_KIND_CLASS,
431 get_rel_name(rte->relid));
436 if (ExecutorCheckPerms_hook)
437 result = (*ExecutorCheckPerms_hook)(rangeTable,
438 ereport_on_violation);
444 * Check access permissions for a single RTE.
447 ExecCheckRTEPerms(RangeTblEntry *rte)
449 AclMode requiredPerms;
451 AclMode remainingPerms;
458 * Only plain-relation RTEs need to be checked here. Function RTEs are
459 * checked by init_fcache when the function is prepared for execution.
460 * Join, subquery, and special RTEs need no checks.
462 if (rte->rtekind != RTE_RELATION)
466 * No work if requiredPerms is empty.
468 requiredPerms = rte->requiredPerms;
469 if (requiredPerms == 0)
475 * userid to check as: current user unless we have a setuid indication.
477 * Note: GetUserId() is presently fast enough that there's no harm in
478 * calling it separately for each RTE. If that stops being true, we could
479 * call it once in ExecCheckRTPerms and pass the userid down from there.
480 * But for now, no need for the extra clutter.
482 userid = rte->checkAsUser ? rte->checkAsUser : GetUserId();
485 * We must have *all* the requiredPerms bits, but some of the bits can be
486 * satisfied from column-level rather than relation-level permissions.
487 * First, remove any bits that are satisfied by relation permissions.
489 relPerms = pg_class_aclmask(relOid, userid, requiredPerms, ACLMASK_ALL);
490 remainingPerms = requiredPerms & ~relPerms;
491 if (remainingPerms != 0)
494 * If we lack any permissions that exist only as relation permissions,
495 * we can fail straight away.
497 if (remainingPerms & ~(ACL_SELECT | ACL_INSERT | ACL_UPDATE))
501 * Check to see if we have the needed privileges at column level.
503 * Note: failures just report a table-level error; it would be nicer
504 * to report a column-level error if we have some but not all of the
507 if (remainingPerms & ACL_SELECT)
510 * When the query doesn't explicitly reference any columns (for
511 * example, SELECT COUNT(*) FROM table), allow the query if we
512 * have SELECT on any column of the rel, as per SQL spec.
514 if (bms_is_empty(rte->selectedCols))
516 if (pg_attribute_aclcheck_all(relOid, userid, ACL_SELECT,
517 ACLMASK_ANY) != ACLCHECK_OK)
521 tmpset = bms_copy(rte->selectedCols);
522 while ((col = bms_first_member(tmpset)) >= 0)
524 /* remove the column number offset */
525 col += FirstLowInvalidHeapAttributeNumber;
526 if (col == InvalidAttrNumber)
528 /* Whole-row reference, must have priv on all cols */
529 if (pg_attribute_aclcheck_all(relOid, userid, ACL_SELECT,
530 ACLMASK_ALL) != ACLCHECK_OK)
535 if (pg_attribute_aclcheck(relOid, col, userid,
536 ACL_SELECT) != ACLCHECK_OK)
544 * Basically the same for the mod columns, with either INSERT or
545 * UPDATE privilege as specified by remainingPerms.
547 remainingPerms &= ~ACL_SELECT;
548 if (remainingPerms != 0)
551 * When the query doesn't explicitly change any columns, allow the
552 * query if we have permission on any column of the rel. This is
553 * to handle SELECT FOR UPDATE as well as possible corner cases in
556 if (bms_is_empty(rte->modifiedCols))
558 if (pg_attribute_aclcheck_all(relOid, userid, remainingPerms,
559 ACLMASK_ANY) != ACLCHECK_OK)
563 tmpset = bms_copy(rte->modifiedCols);
564 while ((col = bms_first_member(tmpset)) >= 0)
566 /* remove the column number offset */
567 col += FirstLowInvalidHeapAttributeNumber;
568 if (col == InvalidAttrNumber)
570 /* whole-row reference can't happen here */
571 elog(ERROR, "whole-row update is not implemented");
575 if (pg_attribute_aclcheck(relOid, col, userid,
576 remainingPerms) != ACLCHECK_OK)
587 * Check that the query does not imply any writes to non-temp tables.
589 * Note: in a Hot Standby slave this would need to reject writes to temp
590 * tables as well; but an HS slave can't have created any temp tables
591 * in the first place, so no need to check that.
594 ExecCheckXactReadOnly(PlannedStmt *plannedstmt)
599 * CREATE TABLE AS or SELECT INTO?
601 * XXX should we allow this if the destination is temp? Considering that
602 * it would still require catalog changes, probably not.
604 if (plannedstmt->intoClause != NULL)
605 PreventCommandIfReadOnly(CreateCommandTag((Node *) plannedstmt));
607 /* Fail if write permissions are requested on any non-temp table */
608 foreach(l, plannedstmt->rtable)
610 RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
612 if (rte->rtekind != RTE_RELATION)
615 if ((rte->requiredPerms & (~ACL_SELECT)) == 0)
618 if (isTempNamespace(get_rel_namespace(rte->relid)))
621 PreventCommandIfReadOnly(CreateCommandTag((Node *) plannedstmt));
626 /* ----------------------------------------------------------------
629 * Initializes the query plan: open files, allocate storage
630 * and start up the rule manager
631 * ----------------------------------------------------------------
634 InitPlan(QueryDesc *queryDesc, int eflags)
636 CmdType operation = queryDesc->operation;
637 PlannedStmt *plannedstmt = queryDesc->plannedstmt;
638 Plan *plan = plannedstmt->planTree;
639 List *rangeTable = plannedstmt->rtable;
640 EState *estate = queryDesc->estate;
641 PlanState *planstate;
647 * Do permissions checks
649 ExecCheckRTPerms(rangeTable, true);
652 * initialize the node's execution state
654 estate->es_range_table = rangeTable;
655 estate->es_plannedstmt = plannedstmt;
658 * initialize result relation stuff, and open/lock the result rels.
660 * We must do this before initializing the plan tree, else we might try to
661 * do a lock upgrade if a result rel is also a source rel.
663 if (plannedstmt->resultRelations)
665 List *resultRelations = plannedstmt->resultRelations;
666 int numResultRelations = list_length(resultRelations);
667 ResultRelInfo *resultRelInfos;
668 ResultRelInfo *resultRelInfo;
670 resultRelInfos = (ResultRelInfo *)
671 palloc(numResultRelations * sizeof(ResultRelInfo));
672 resultRelInfo = resultRelInfos;
673 foreach(l, resultRelations)
675 Index resultRelationIndex = lfirst_int(l);
676 Oid resultRelationOid;
677 Relation resultRelation;
679 resultRelationOid = getrelid(resultRelationIndex, rangeTable);
680 resultRelation = heap_open(resultRelationOid, RowExclusiveLock);
681 InitResultRelInfo(resultRelInfo,
685 estate->es_instrument);
688 estate->es_result_relations = resultRelInfos;
689 estate->es_num_result_relations = numResultRelations;
690 /* es_result_relation_info is NULL except when within ModifyTable */
691 estate->es_result_relation_info = NULL;
696 * if no result relation, then set state appropriately
698 estate->es_result_relations = NULL;
699 estate->es_num_result_relations = 0;
700 estate->es_result_relation_info = NULL;
704 * Similarly, we have to lock relations selected FOR UPDATE/FOR SHARE
705 * before we initialize the plan tree, else we'd be risking lock upgrades.
706 * While we are at it, build the ExecRowMark list.
708 estate->es_rowMarks = NIL;
709 foreach(l, plannedstmt->rowMarks)
711 PlanRowMark *rc = (PlanRowMark *) lfirst(l);
716 /* ignore "parent" rowmarks; they are irrelevant at runtime */
720 switch (rc->markType)
722 case ROW_MARK_EXCLUSIVE:
724 relid = getrelid(rc->rti, rangeTable);
725 relation = heap_open(relid, RowShareLock);
727 case ROW_MARK_REFERENCE:
728 relid = getrelid(rc->rti, rangeTable);
729 relation = heap_open(relid, AccessShareLock);
732 /* there's no real table here ... */
736 elog(ERROR, "unrecognized markType: %d", rc->markType);
737 relation = NULL; /* keep compiler quiet */
741 erm = (ExecRowMark *) palloc(sizeof(ExecRowMark));
742 erm->relation = relation;
744 erm->prti = rc->prti;
745 erm->rowmarkId = rc->rowmarkId;
746 erm->markType = rc->markType;
747 erm->noWait = rc->noWait;
748 ItemPointerSetInvalid(&(erm->curCtid));
749 estate->es_rowMarks = lappend(estate->es_rowMarks, erm);
753 * Detect whether we're doing SELECT INTO. If so, set the es_into_oids
754 * flag appropriately so that the plan tree will be initialized with the
755 * correct tuple descriptors. (Other SELECT INTO stuff comes later.)
757 estate->es_select_into = false;
758 if (operation == CMD_SELECT && plannedstmt->intoClause != NULL)
760 estate->es_select_into = true;
761 estate->es_into_oids = interpretOidsOption(plannedstmt->intoClause->options);
765 * Initialize the executor's tuple table to empty.
767 estate->es_tupleTable = NIL;
768 estate->es_trig_tuple_slot = NULL;
769 estate->es_trig_oldtup_slot = NULL;
771 /* mark EvalPlanQual not active */
772 estate->es_epqTuple = NULL;
773 estate->es_epqTupleSet = NULL;
774 estate->es_epqScanDone = NULL;
777 * Initialize private state information for each SubPlan. We must do this
778 * before running ExecInitNode on the main query tree, since
779 * ExecInitSubPlan expects to be able to find these entries.
781 Assert(estate->es_subplanstates == NIL);
782 i = 1; /* subplan indices count from 1 */
783 foreach(l, plannedstmt->subplans)
785 Plan *subplan = (Plan *) lfirst(l);
786 PlanState *subplanstate;
790 * A subplan will never need to do BACKWARD scan nor MARK/RESTORE. If
791 * it is a parameterless subplan (not initplan), we suggest that it be
792 * prepared to handle REWIND efficiently; otherwise there is no need.
794 sp_eflags = eflags & EXEC_FLAG_EXPLAIN_ONLY;
795 if (bms_is_member(i, plannedstmt->rewindPlanIDs))
796 sp_eflags |= EXEC_FLAG_REWIND;
798 subplanstate = ExecInitNode(subplan, estate, sp_eflags);
800 estate->es_subplanstates = lappend(estate->es_subplanstates,
807 * Initialize the private state information for all the nodes in the query
808 * tree. This opens files, allocates storage and leaves us ready to start
811 planstate = ExecInitNode(plan, estate, eflags);
814 * Get the tuple descriptor describing the type of tuples to return. (this
815 * is especially important if we are creating a relation with "SELECT
818 tupType = ExecGetResultType(planstate);
821 * Initialize the junk filter if needed. SELECT queries need a filter if
822 * there are any junk attrs in the top-level tlist.
824 if (operation == CMD_SELECT)
826 bool junk_filter_needed = false;
829 foreach(tlist, plan->targetlist)
831 TargetEntry *tle = (TargetEntry *) lfirst(tlist);
835 junk_filter_needed = true;
840 if (junk_filter_needed)
844 j = ExecInitJunkFilter(planstate->plan->targetlist,
846 ExecInitExtraTupleSlot(estate));
847 estate->es_junkFilter = j;
849 /* Want to return the cleaned tuple type */
850 tupType = j->jf_cleanTupType;
854 queryDesc->tupDesc = tupType;
855 queryDesc->planstate = planstate;
858 * If doing SELECT INTO, initialize the "into" relation. We must wait
859 * till now so we have the "clean" result tuple type to create the new
862 * If EXPLAIN, skip creating the "into" relation.
864 if (estate->es_select_into && !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
865 OpenIntoRel(queryDesc);
869 * Initialize ResultRelInfo data for one result relation
872 InitResultRelInfo(ResultRelInfo *resultRelInfo,
873 Relation resultRelationDesc,
874 Index resultRelationIndex,
876 int instrument_options)
878 TriggerDesc *trigDesc = resultRelationDesc->trigdesc;
881 * Check valid relkind ... in most cases parser and/or planner should have
882 * noticed this already, but let's make sure. In the view case we do need
883 * a test here, because if the view wasn't rewritten by a rule, it had
884 * better have an INSTEAD trigger.
886 switch (resultRelationDesc->rd_rel->relkind)
888 case RELKIND_RELATION:
891 case RELKIND_SEQUENCE:
893 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
894 errmsg("cannot change sequence \"%s\"",
895 RelationGetRelationName(resultRelationDesc))));
897 case RELKIND_TOASTVALUE:
899 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
900 errmsg("cannot change TOAST relation \"%s\"",
901 RelationGetRelationName(resultRelationDesc))));
907 if (!trigDesc || !trigDesc->trig_insert_instead_row)
909 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
910 errmsg("cannot insert into view \"%s\"",
911 RelationGetRelationName(resultRelationDesc)),
912 errhint("You need an unconditional ON INSERT DO INSTEAD rule or an INSTEAD OF INSERT trigger.")));
915 if (!trigDesc || !trigDesc->trig_update_instead_row)
917 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
918 errmsg("cannot update view \"%s\"",
919 RelationGetRelationName(resultRelationDesc)),
920 errhint("You need an unconditional ON UPDATE DO INSTEAD rule or an INSTEAD OF UPDATE trigger.")));
923 if (!trigDesc || !trigDesc->trig_delete_instead_row)
925 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
926 errmsg("cannot delete from view \"%s\"",
927 RelationGetRelationName(resultRelationDesc)),
928 errhint("You need an unconditional ON DELETE DO INSTEAD rule or an INSTEAD OF DELETE trigger.")));
931 elog(ERROR, "unrecognized CmdType: %d", (int) operation);
935 case RELKIND_FOREIGN_TABLE:
937 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
938 errmsg("cannot change foreign table \"%s\"",
939 RelationGetRelationName(resultRelationDesc))));
943 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
944 errmsg("cannot change relation \"%s\"",
945 RelationGetRelationName(resultRelationDesc))));
949 /* OK, fill in the node */
950 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
951 resultRelInfo->type = T_ResultRelInfo;
952 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
953 resultRelInfo->ri_RelationDesc = resultRelationDesc;
954 resultRelInfo->ri_NumIndices = 0;
955 resultRelInfo->ri_IndexRelationDescs = NULL;
956 resultRelInfo->ri_IndexRelationInfo = NULL;
957 /* make a copy so as not to depend on relcache info not changing... */
958 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(trigDesc);
959 if (resultRelInfo->ri_TrigDesc)
961 int n = resultRelInfo->ri_TrigDesc->numtriggers;
963 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
964 palloc0(n * sizeof(FmgrInfo));
965 resultRelInfo->ri_TrigWhenExprs = (List **)
966 palloc0(n * sizeof(List *));
967 if (instrument_options)
968 resultRelInfo->ri_TrigInstrument = InstrAlloc(n, instrument_options);
972 resultRelInfo->ri_TrigFunctions = NULL;
973 resultRelInfo->ri_TrigWhenExprs = NULL;
974 resultRelInfo->ri_TrigInstrument = NULL;
976 resultRelInfo->ri_ConstraintExprs = NULL;
977 resultRelInfo->ri_junkFilter = NULL;
978 resultRelInfo->ri_projectReturning = NULL;
981 * If there are indices on the result relation, open them and save
982 * descriptors in the result relation info, so that we can add new index
983 * entries for the tuples we add/update. We need not do this for a
984 * DELETE, however, since deletion doesn't affect indexes.
986 if (resultRelationDesc->rd_rel->relhasindex &&
987 operation != CMD_DELETE)
988 ExecOpenIndices(resultRelInfo);
992 * ExecGetTriggerResultRel
994 * Get a ResultRelInfo for a trigger target relation. Most of the time,
995 * triggers are fired on one of the result relations of the query, and so
996 * we can just return a member of the es_result_relations array. (Note: in
997 * self-join situations there might be multiple members with the same OID;
998 * if so it doesn't matter which one we pick.) However, it is sometimes
999 * necessary to fire triggers on other relations; this happens mainly when an
1000 * RI update trigger queues additional triggers on other relations, which will
1001 * be processed in the context of the outer query. For efficiency's sake,
1002 * we want to have a ResultRelInfo for those triggers too; that can avoid
1003 * repeated re-opening of the relation. (It also provides a way for EXPLAIN
1004 * ANALYZE to report the runtimes of such triggers.) So we make additional
1005 * ResultRelInfo's as needed, and save them in es_trig_target_relations.
1008 ExecGetTriggerResultRel(EState *estate, Oid relid)
1010 ResultRelInfo *rInfo;
1014 MemoryContext oldcontext;
1016 /* First, search through the query result relations */
1017 rInfo = estate->es_result_relations;
1018 nr = estate->es_num_result_relations;
1021 if (RelationGetRelid(rInfo->ri_RelationDesc) == relid)
1026 /* Nope, but maybe we already made an extra ResultRelInfo for it */
1027 foreach(l, estate->es_trig_target_relations)
1029 rInfo = (ResultRelInfo *) lfirst(l);
1030 if (RelationGetRelid(rInfo->ri_RelationDesc) == relid)
1033 /* Nope, so we need a new one */
1036 * Open the target relation's relcache entry. We assume that an
1037 * appropriate lock is still held by the backend from whenever the trigger
1038 * event got queued, so we need take no new lock here.
1040 rel = heap_open(relid, NoLock);
1043 * Make the new entry in the right context. Currently, we don't need any
1044 * index information in ResultRelInfos used only for triggers, so tell
1045 * InitResultRelInfo it's a DELETE.
1047 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
1048 rInfo = makeNode(ResultRelInfo);
1049 InitResultRelInfo(rInfo,
1051 0, /* dummy rangetable index */
1053 estate->es_instrument);
1054 estate->es_trig_target_relations =
1055 lappend(estate->es_trig_target_relations, rInfo);
1056 MemoryContextSwitchTo(oldcontext);
1062 * ExecContextForcesOids
1064 * This is pretty grotty: when doing INSERT, UPDATE, or SELECT INTO,
1065 * we need to ensure that result tuples have space for an OID iff they are
1066 * going to be stored into a relation that has OIDs. In other contexts
1067 * we are free to choose whether to leave space for OIDs in result tuples
1068 * (we generally don't want to, but we do if a physical-tlist optimization
1069 * is possible). This routine checks the plan context and returns TRUE if the
1070 * choice is forced, FALSE if the choice is not forced. In the TRUE case,
1071 * *hasoids is set to the required value.
1073 * One reason this is ugly is that all plan nodes in the plan tree will emit
1074 * tuples with space for an OID, though we really only need the topmost node
1075 * to do so. However, node types like Sort don't project new tuples but just
1076 * return their inputs, and in those cases the requirement propagates down
1077 * to the input node. Eventually we might make this code smart enough to
1078 * recognize how far down the requirement really goes, but for now we just
1079 * make all plan nodes do the same thing if the top level forces the choice.
1081 * We assume that if we are generating tuples for INSERT or UPDATE,
1082 * estate->es_result_relation_info is already set up to describe the target
1083 * relation. Note that in an UPDATE that spans an inheritance tree, some of
1084 * the target relations may have OIDs and some not. We have to make the
1085 * decisions on a per-relation basis as we initialize each of the subplans of
1086 * the ModifyTable node, so ModifyTable has to set es_result_relation_info
1087 * while initializing each subplan.
1089 * SELECT INTO is even uglier, because we don't have the INTO relation's
1090 * descriptor available when this code runs; we have to look aside at a
1091 * flag set by InitPlan().
1094 ExecContextForcesOids(PlanState *planstate, bool *hasoids)
1096 ResultRelInfo *ri = planstate->state->es_result_relation_info;
1100 Relation rel = ri->ri_RelationDesc;
1104 *hasoids = rel->rd_rel->relhasoids;
1109 if (planstate->state->es_select_into)
1111 *hasoids = planstate->state->es_into_oids;
1118 /* ----------------------------------------------------------------
1121 * Cleans up the query plan -- closes files and frees up storage
1123 * NOTE: we are no longer very worried about freeing storage per se
1124 * in this code; FreeExecutorState should be guaranteed to release all
1125 * memory that needs to be released. What we are worried about doing
1126 * is closing relations and dropping buffer pins. Thus, for example,
1127 * tuple tables must be cleared or dropped to ensure pins are released.
1128 * ----------------------------------------------------------------
1131 ExecEndPlan(PlanState *planstate, EState *estate)
1133 ResultRelInfo *resultRelInfo;
1138 * shut down the node-type-specific query processing
1140 ExecEndNode(planstate);
1145 foreach(l, estate->es_subplanstates)
1147 PlanState *subplanstate = (PlanState *) lfirst(l);
1149 ExecEndNode(subplanstate);
1153 * destroy the executor's tuple table. Actually we only care about
1154 * releasing buffer pins and tupdesc refcounts; there's no need to pfree
1155 * the TupleTableSlots, since the containing memory context is about to go
1158 ExecResetTupleTable(estate->es_tupleTable, false);
1161 * close the result relation(s) if any, but hold locks until xact commit.
1163 resultRelInfo = estate->es_result_relations;
1164 for (i = estate->es_num_result_relations; i > 0; i--)
1166 /* Close indices and then the relation itself */
1167 ExecCloseIndices(resultRelInfo);
1168 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
1173 * likewise close any trigger target relations
1175 foreach(l, estate->es_trig_target_relations)
1177 resultRelInfo = (ResultRelInfo *) lfirst(l);
1178 /* Close indices and then the relation itself */
1179 ExecCloseIndices(resultRelInfo);
1180 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
1184 * close any relations selected FOR UPDATE/FOR SHARE, again keeping locks
1186 foreach(l, estate->es_rowMarks)
1188 ExecRowMark *erm = (ExecRowMark *) lfirst(l);
1191 heap_close(erm->relation, NoLock);
1195 /* ----------------------------------------------------------------
1198 * Processes the query plan until we have processed 'numberTuples' tuples,
1199 * moving in the specified direction.
1201 * Runs to completion if numberTuples is 0
1203 * Note: the ctid attribute is a 'junk' attribute that is removed before the
1205 * ----------------------------------------------------------------
1208 ExecutePlan(EState *estate,
1209 PlanState *planstate,
1213 ScanDirection direction,
1216 TupleTableSlot *slot;
1217 long current_tuple_count;
1220 * initialize local variables
1222 current_tuple_count = 0;
1225 * Set the direction.
1227 estate->es_direction = direction;
1230 * Loop until we've processed the proper number of tuples from the plan.
1234 /* Reset the per-output-tuple exprcontext */
1235 ResetPerTupleExprContext(estate);
1238 * Execute the plan and obtain a tuple
1240 slot = ExecProcNode(planstate);
1243 * if the tuple is null, then we assume there is nothing more to
1244 * process so we just end the loop...
1246 if (TupIsNull(slot))
1250 * If we have a junk filter, then project a new tuple with the junk
1253 * Store this new "clean" tuple in the junkfilter's resultSlot.
1254 * (Formerly, we stored it back over the "dirty" tuple, which is WRONG
1255 * because that tuple slot has the wrong descriptor.)
1257 if (estate->es_junkFilter != NULL)
1258 slot = ExecFilterJunk(estate->es_junkFilter, slot);
1261 * If we are supposed to send the tuple somewhere, do so. (In
1262 * practice, this is probably always the case at this point.)
1265 (*dest->receiveSlot) (slot, dest);
1268 * Count tuples processed, if this is a SELECT. (For other operation
1269 * types, the ModifyTable plan node must count the appropriate
1272 if (operation == CMD_SELECT)
1273 (estate->es_processed)++;
1276 * check our tuple count.. if we've processed the proper number then
1277 * quit, else loop again and process more tuples. Zero numberTuples
1280 current_tuple_count++;
1281 if (numberTuples && numberTuples == current_tuple_count)
1288 * ExecRelCheck --- check that tuple meets constraints for result relation
1291 ExecRelCheck(ResultRelInfo *resultRelInfo,
1292 TupleTableSlot *slot, EState *estate)
1294 Relation rel = resultRelInfo->ri_RelationDesc;
1295 int ncheck = rel->rd_att->constr->num_check;
1296 ConstrCheck *check = rel->rd_att->constr->check;
1297 ExprContext *econtext;
1298 MemoryContext oldContext;
1303 * If first time through for this result relation, build expression
1304 * nodetrees for rel's constraint expressions. Keep them in the per-query
1305 * memory context so they'll survive throughout the query.
1307 if (resultRelInfo->ri_ConstraintExprs == NULL)
1309 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1310 resultRelInfo->ri_ConstraintExprs =
1311 (List **) palloc(ncheck * sizeof(List *));
1312 for (i = 0; i < ncheck; i++)
1314 /* ExecQual wants implicit-AND form */
1315 qual = make_ands_implicit(stringToNode(check[i].ccbin));
1316 resultRelInfo->ri_ConstraintExprs[i] = (List *)
1317 ExecPrepareExpr((Expr *) qual, estate);
1319 MemoryContextSwitchTo(oldContext);
1323 * We will use the EState's per-tuple context for evaluating constraint
1324 * expressions (creating it if it's not already there).
1326 econtext = GetPerTupleExprContext(estate);
1328 /* Arrange for econtext's scan tuple to be the tuple under test */
1329 econtext->ecxt_scantuple = slot;
1331 /* And evaluate the constraints */
1332 for (i = 0; i < ncheck; i++)
1334 qual = resultRelInfo->ri_ConstraintExprs[i];
1337 * NOTE: SQL92 specifies that a NULL result from a constraint
1338 * expression is not to be treated as a failure. Therefore, tell
1339 * ExecQual to return TRUE for NULL.
1341 if (!ExecQual(qual, econtext, true))
1342 return check[i].ccname;
1345 /* NULL result means no error */
1350 ExecConstraints(ResultRelInfo *resultRelInfo,
1351 TupleTableSlot *slot, EState *estate)
1353 Relation rel = resultRelInfo->ri_RelationDesc;
1354 TupleConstr *constr = rel->rd_att->constr;
1358 if (constr->has_not_null)
1360 int natts = rel->rd_att->natts;
1363 for (attrChk = 1; attrChk <= natts; attrChk++)
1365 if (rel->rd_att->attrs[attrChk - 1]->attnotnull &&
1366 slot_attisnull(slot, attrChk))
1368 (errcode(ERRCODE_NOT_NULL_VIOLATION),
1369 errmsg("null value in column \"%s\" violates not-null constraint",
1370 NameStr(rel->rd_att->attrs[attrChk - 1]->attname))));
1374 if (constr->num_check > 0)
1378 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
1380 (errcode(ERRCODE_CHECK_VIOLATION),
1381 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
1382 RelationGetRelationName(rel), failed)));
1388 * ExecFindRowMark -- find the ExecRowMark struct for given rangetable index
1391 ExecFindRowMark(EState *estate, Index rti)
1395 foreach(lc, estate->es_rowMarks)
1397 ExecRowMark *erm = (ExecRowMark *) lfirst(lc);
1399 if (erm->rti == rti)
1402 elog(ERROR, "failed to find ExecRowMark for rangetable index %u", rti);
1403 return NULL; /* keep compiler quiet */
1407 * ExecBuildAuxRowMark -- create an ExecAuxRowMark struct
1409 * Inputs are the underlying ExecRowMark struct and the targetlist of the
1410 * input plan node (not planstate node!). We need the latter to find out
1411 * the column numbers of the resjunk columns.
1414 ExecBuildAuxRowMark(ExecRowMark *erm, List *targetlist)
1416 ExecAuxRowMark *aerm = (ExecAuxRowMark *) palloc0(sizeof(ExecAuxRowMark));
1419 aerm->rowmark = erm;
1421 /* Look up the resjunk columns associated with this rowmark */
1424 Assert(erm->markType != ROW_MARK_COPY);
1426 /* if child rel, need tableoid */
1427 if (erm->rti != erm->prti)
1429 snprintf(resname, sizeof(resname), "tableoid%u", erm->rowmarkId);
1430 aerm->toidAttNo = ExecFindJunkAttributeInTlist(targetlist,
1432 if (!AttributeNumberIsValid(aerm->toidAttNo))
1433 elog(ERROR, "could not find junk %s column", resname);
1436 /* always need ctid for real relations */
1437 snprintf(resname, sizeof(resname), "ctid%u", erm->rowmarkId);
1438 aerm->ctidAttNo = ExecFindJunkAttributeInTlist(targetlist,
1440 if (!AttributeNumberIsValid(aerm->ctidAttNo))
1441 elog(ERROR, "could not find junk %s column", resname);
1445 Assert(erm->markType == ROW_MARK_COPY);
1447 snprintf(resname, sizeof(resname), "wholerow%u", erm->rowmarkId);
1448 aerm->wholeAttNo = ExecFindJunkAttributeInTlist(targetlist,
1450 if (!AttributeNumberIsValid(aerm->wholeAttNo))
1451 elog(ERROR, "could not find junk %s column", resname);
1459 * EvalPlanQual logic --- recheck modified tuple(s) to see if we want to
1460 * process the updated version under READ COMMITTED rules.
1462 * See backend/executor/README for some info about how this works.
1467 * Check a modified tuple to see if we want to process its updated version
1468 * under READ COMMITTED rules.
1470 * estate - outer executor state data
1471 * epqstate - state for EvalPlanQual rechecking
1472 * relation - table containing tuple
1473 * rti - rangetable index of table containing tuple
1474 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1475 * priorXmax - t_xmax from the outdated tuple
1477 * *tid is also an output parameter: it's modified to hold the TID of the
1478 * latest version of the tuple (note this may be changed even on failure)
1480 * Returns a slot containing the new candidate update/delete tuple, or
1481 * NULL if we determine we shouldn't process the row.
1484 EvalPlanQual(EState *estate, EPQState *epqstate,
1485 Relation relation, Index rti,
1486 ItemPointer tid, TransactionId priorXmax)
1488 TupleTableSlot *slot;
1489 HeapTuple copyTuple;
1494 * Get and lock the updated version of the row; if fail, return NULL.
1496 copyTuple = EvalPlanQualFetch(estate, relation, LockTupleExclusive,
1499 if (copyTuple == NULL)
1503 * For UPDATE/DELETE we have to return tid of actual row we're executing
1506 *tid = copyTuple->t_self;
1509 * Need to run a recheck subquery. Initialize or reinitialize EPQ state.
1511 EvalPlanQualBegin(epqstate, estate);
1514 * Free old test tuple, if any, and store new tuple where relation's scan
1517 EvalPlanQualSetTuple(epqstate, rti, copyTuple);
1520 * Fetch any non-locked source rows
1522 EvalPlanQualFetchRowMarks(epqstate);
1525 * Run the EPQ query. We assume it will return at most one tuple.
1527 slot = EvalPlanQualNext(epqstate);
1530 * If we got a tuple, force the slot to materialize the tuple so that it
1531 * is not dependent on any local state in the EPQ query (in particular,
1532 * it's highly likely that the slot contains references to any pass-by-ref
1533 * datums that may be present in copyTuple). As with the next step, this
1534 * is to guard against early re-use of the EPQ query.
1536 if (!TupIsNull(slot))
1537 (void) ExecMaterializeSlot(slot);
1540 * Clear out the test tuple. This is needed in case the EPQ query is
1541 * re-used to test a tuple for a different relation. (Not clear that can
1542 * really happen, but let's be safe.)
1544 EvalPlanQualSetTuple(epqstate, rti, NULL);
1550 * Fetch a copy of the newest version of an outdated tuple
1552 * estate - executor state data
1553 * relation - table containing tuple
1554 * lockmode - requested tuple lock mode
1555 * *tid - t_ctid from the outdated tuple (ie, next updated version)
1556 * priorXmax - t_xmax from the outdated tuple
1558 * Returns a palloc'd copy of the newest tuple version, or NULL if we find
1559 * that there is no newest version (ie, the row was deleted not updated).
1560 * If successful, we have locked the newest tuple version, so caller does not
1561 * need to worry about it changing anymore.
1563 * Note: properly, lockmode should be declared as enum LockTupleMode,
1564 * but we use "int" to avoid having to include heapam.h in executor.h.
1567 EvalPlanQualFetch(EState *estate, Relation relation, int lockmode,
1568 ItemPointer tid, TransactionId priorXmax)
1570 HeapTuple copyTuple = NULL;
1571 HeapTupleData tuple;
1572 SnapshotData SnapshotDirty;
1575 * fetch target tuple
1577 * Loop here to deal with updated or busy tuples
1579 InitDirtySnapshot(SnapshotDirty);
1580 tuple.t_self = *tid;
1585 if (heap_fetch(relation, &SnapshotDirty, &tuple, &buffer, true, NULL))
1588 ItemPointerData update_ctid;
1589 TransactionId update_xmax;
1592 * If xmin isn't what we're expecting, the slot must have been
1593 * recycled and reused for an unrelated tuple. This implies that
1594 * the latest version of the row was deleted, so we need do
1595 * nothing. (Should be safe to examine xmin without getting
1596 * buffer's content lock, since xmin never changes in an existing
1599 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1602 ReleaseBuffer(buffer);
1606 /* otherwise xmin should not be dirty... */
1607 if (TransactionIdIsValid(SnapshotDirty.xmin))
1608 elog(ERROR, "t_xmin is uncommitted in tuple to be updated");
1611 * If tuple is being updated by other transaction then we have to
1612 * wait for its commit/abort.
1614 if (TransactionIdIsValid(SnapshotDirty.xmax))
1616 ReleaseBuffer(buffer);
1617 XactLockTableWait(SnapshotDirty.xmax);
1618 continue; /* loop back to repeat heap_fetch */
1622 * If tuple was inserted by our own transaction, we have to check
1623 * cmin against es_output_cid: cmin >= current CID means our
1624 * command cannot see the tuple, so we should ignore it. Without
1625 * this we are open to the "Halloween problem" of indefinitely
1626 * re-updating the same tuple. (We need not check cmax because
1627 * HeapTupleSatisfiesDirty will consider a tuple deleted by our
1628 * transaction dead, regardless of cmax.) We just checked that
1629 * priorXmax == xmin, so we can test that variable instead of
1630 * doing HeapTupleHeaderGetXmin again.
1632 if (TransactionIdIsCurrentTransactionId(priorXmax) &&
1633 HeapTupleHeaderGetCmin(tuple.t_data) >= estate->es_output_cid)
1635 ReleaseBuffer(buffer);
1640 * This is a live tuple, so now try to lock it.
1642 test = heap_lock_tuple(relation, &tuple, &buffer,
1643 &update_ctid, &update_xmax,
1644 estate->es_output_cid,
1646 /* We now have two pins on the buffer, get rid of one */
1647 ReleaseBuffer(buffer);
1651 case HeapTupleSelfUpdated:
1652 /* treat it as deleted; do not process */
1653 ReleaseBuffer(buffer);
1656 case HeapTupleMayBeUpdated:
1657 /* successfully locked */
1660 case HeapTupleUpdated:
1661 ReleaseBuffer(buffer);
1662 if (IsolationUsesXactSnapshot())
1664 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1665 errmsg("could not serialize access due to concurrent update")));
1666 if (!ItemPointerEquals(&update_ctid, &tuple.t_self))
1668 /* it was updated, so look at the updated version */
1669 tuple.t_self = update_ctid;
1670 /* updated row should have xmin matching this xmax */
1671 priorXmax = update_xmax;
1674 /* tuple was deleted, so give up */
1678 ReleaseBuffer(buffer);
1679 elog(ERROR, "unrecognized heap_lock_tuple status: %u",
1681 return NULL; /* keep compiler quiet */
1685 * We got tuple - now copy it for use by recheck query.
1687 copyTuple = heap_copytuple(&tuple);
1688 ReleaseBuffer(buffer);
1693 * If the referenced slot was actually empty, the latest version of
1694 * the row must have been deleted, so we need do nothing.
1696 if (tuple.t_data == NULL)
1698 ReleaseBuffer(buffer);
1703 * As above, if xmin isn't what we're expecting, do nothing.
1705 if (!TransactionIdEquals(HeapTupleHeaderGetXmin(tuple.t_data),
1708 ReleaseBuffer(buffer);
1713 * If we get here, the tuple was found but failed SnapshotDirty.
1714 * Assuming the xmin is either a committed xact or our own xact (as it
1715 * certainly should be if we're trying to modify the tuple), this must
1716 * mean that the row was updated or deleted by either a committed xact
1717 * or our own xact. If it was deleted, we can ignore it; if it was
1718 * updated then chain up to the next version and repeat the whole
1721 * As above, it should be safe to examine xmax and t_ctid without the
1722 * buffer content lock, because they can't be changing.
1724 if (ItemPointerEquals(&tuple.t_self, &tuple.t_data->t_ctid))
1726 /* deleted, so forget about it */
1727 ReleaseBuffer(buffer);
1731 /* updated, so look at the updated row */
1732 tuple.t_self = tuple.t_data->t_ctid;
1733 /* updated row should have xmin matching this xmax */
1734 priorXmax = HeapTupleHeaderGetXmax(tuple.t_data);
1735 ReleaseBuffer(buffer);
1736 /* loop back to fetch next in chain */
1740 * Return the copied tuple
1746 * EvalPlanQualInit -- initialize during creation of a plan state node
1747 * that might need to invoke EPQ processing.
1749 * Note: subplan/auxrowmarks can be NULL/NIL if they will be set later
1750 * with EvalPlanQualSetPlan.
1753 EvalPlanQualInit(EPQState *epqstate, EState *estate,
1754 Plan *subplan, List *auxrowmarks, int epqParam)
1756 /* Mark the EPQ state inactive */
1757 epqstate->estate = NULL;
1758 epqstate->planstate = NULL;
1759 epqstate->origslot = NULL;
1760 /* ... and remember data that EvalPlanQualBegin will need */
1761 epqstate->plan = subplan;
1762 epqstate->arowMarks = auxrowmarks;
1763 epqstate->epqParam = epqParam;
1767 * EvalPlanQualSetPlan -- set or change subplan of an EPQState.
1769 * We need this so that ModifyTuple can deal with multiple subplans.
1772 EvalPlanQualSetPlan(EPQState *epqstate, Plan *subplan, List *auxrowmarks)
1774 /* If we have a live EPQ query, shut it down */
1775 EvalPlanQualEnd(epqstate);
1776 /* And set/change the plan pointer */
1777 epqstate->plan = subplan;
1778 /* The rowmarks depend on the plan, too */
1779 epqstate->arowMarks = auxrowmarks;
1783 * Install one test tuple into EPQ state, or clear test tuple if tuple == NULL
1785 * NB: passed tuple must be palloc'd; it may get freed later
1788 EvalPlanQualSetTuple(EPQState *epqstate, Index rti, HeapTuple tuple)
1790 EState *estate = epqstate->estate;
1795 * free old test tuple, if any, and store new tuple where relation's scan
1798 if (estate->es_epqTuple[rti - 1] != NULL)
1799 heap_freetuple(estate->es_epqTuple[rti - 1]);
1800 estate->es_epqTuple[rti - 1] = tuple;
1801 estate->es_epqTupleSet[rti - 1] = true;
1805 * Fetch back the current test tuple (if any) for the specified RTI
1808 EvalPlanQualGetTuple(EPQState *epqstate, Index rti)
1810 EState *estate = epqstate->estate;
1814 return estate->es_epqTuple[rti - 1];
1818 * Fetch the current row values for any non-locked relations that need
1819 * to be scanned by an EvalPlanQual operation. origslot must have been set
1820 * to contain the current result row (top-level row) that we need to recheck.
1823 EvalPlanQualFetchRowMarks(EPQState *epqstate)
1827 Assert(epqstate->origslot != NULL);
1829 foreach(l, epqstate->arowMarks)
1831 ExecAuxRowMark *aerm = (ExecAuxRowMark *) lfirst(l);
1832 ExecRowMark *erm = aerm->rowmark;
1835 HeapTupleData tuple;
1837 if (RowMarkRequiresRowShareLock(erm->markType))
1838 elog(ERROR, "EvalPlanQual doesn't support locking rowmarks");
1840 /* clear any leftover test tuple for this rel */
1841 EvalPlanQualSetTuple(epqstate, erm->rti, NULL);
1847 Assert(erm->markType == ROW_MARK_REFERENCE);
1849 /* if child rel, must check whether it produced this row */
1850 if (erm->rti != erm->prti)
1854 datum = ExecGetJunkAttribute(epqstate->origslot,
1857 /* non-locked rels could be on the inside of outer joins */
1860 tableoid = DatumGetObjectId(datum);
1862 if (tableoid != RelationGetRelid(erm->relation))
1864 /* this child is inactive right now */
1869 /* fetch the tuple's ctid */
1870 datum = ExecGetJunkAttribute(epqstate->origslot,
1873 /* non-locked rels could be on the inside of outer joins */
1876 tuple.t_self = *((ItemPointer) DatumGetPointer(datum));
1878 /* okay, fetch the tuple */
1879 if (!heap_fetch(erm->relation, SnapshotAny, &tuple, &buffer,
1881 elog(ERROR, "failed to fetch tuple for EvalPlanQual recheck");
1883 /* successful, copy and store tuple */
1884 EvalPlanQualSetTuple(epqstate, erm->rti,
1885 heap_copytuple(&tuple));
1886 ReleaseBuffer(buffer);
1892 Assert(erm->markType == ROW_MARK_COPY);
1894 /* fetch the whole-row Var for the relation */
1895 datum = ExecGetJunkAttribute(epqstate->origslot,
1898 /* non-locked rels could be on the inside of outer joins */
1901 td = DatumGetHeapTupleHeader(datum);
1903 /* build a temporary HeapTuple control structure */
1904 tuple.t_len = HeapTupleHeaderGetDatumLength(td);
1905 ItemPointerSetInvalid(&(tuple.t_self));
1906 tuple.t_tableOid = InvalidOid;
1909 /* copy and store tuple */
1910 EvalPlanQualSetTuple(epqstate, erm->rti,
1911 heap_copytuple(&tuple));
1917 * Fetch the next row (if any) from EvalPlanQual testing
1919 * (In practice, there should never be more than one row...)
1922 EvalPlanQualNext(EPQState *epqstate)
1924 MemoryContext oldcontext;
1925 TupleTableSlot *slot;
1927 oldcontext = MemoryContextSwitchTo(epqstate->estate->es_query_cxt);
1928 slot = ExecProcNode(epqstate->planstate);
1929 MemoryContextSwitchTo(oldcontext);
1935 * Initialize or reset an EvalPlanQual state tree
1938 EvalPlanQualBegin(EPQState *epqstate, EState *parentestate)
1940 EState *estate = epqstate->estate;
1944 /* First time through, so create a child EState */
1945 EvalPlanQualStart(epqstate, parentestate, epqstate->plan);
1950 * We already have a suitable child EPQ tree, so just reset it.
1952 int rtsize = list_length(parentestate->es_range_table);
1953 PlanState *planstate = epqstate->planstate;
1955 MemSet(estate->es_epqScanDone, 0, rtsize * sizeof(bool));
1957 /* Recopy current values of parent parameters */
1958 if (parentestate->es_plannedstmt->nParamExec > 0)
1960 int i = parentestate->es_plannedstmt->nParamExec;
1964 /* copy value if any, but not execPlan link */
1965 estate->es_param_exec_vals[i].value =
1966 parentestate->es_param_exec_vals[i].value;
1967 estate->es_param_exec_vals[i].isnull =
1968 parentestate->es_param_exec_vals[i].isnull;
1973 * Mark child plan tree as needing rescan at all scan nodes. The
1974 * first ExecProcNode will take care of actually doing the rescan.
1976 planstate->chgParam = bms_add_member(planstate->chgParam,
1977 epqstate->epqParam);
1982 * Start execution of an EvalPlanQual plan tree.
1984 * This is a cut-down version of ExecutorStart(): we copy some state from
1985 * the top-level estate rather than initializing it fresh.
1988 EvalPlanQualStart(EPQState *epqstate, EState *parentestate, Plan *planTree)
1992 MemoryContext oldcontext;
1995 rtsize = list_length(parentestate->es_range_table);
1997 epqstate->estate = estate = CreateExecutorState();
1999 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
2002 * Child EPQ EStates share the parent's copy of unchanging state such as
2003 * the snapshot, rangetable, result-rel info, and external Param info.
2004 * They need their own copies of local state, including a tuple table,
2005 * es_param_exec_vals, etc.
2007 estate->es_direction = ForwardScanDirection;
2008 estate->es_snapshot = parentestate->es_snapshot;
2009 estate->es_crosscheck_snapshot = parentestate->es_crosscheck_snapshot;
2010 estate->es_range_table = parentestate->es_range_table;
2011 estate->es_plannedstmt = parentestate->es_plannedstmt;
2012 estate->es_junkFilter = parentestate->es_junkFilter;
2013 estate->es_output_cid = parentestate->es_output_cid;
2014 estate->es_result_relations = parentestate->es_result_relations;
2015 estate->es_num_result_relations = parentestate->es_num_result_relations;
2016 estate->es_result_relation_info = parentestate->es_result_relation_info;
2017 /* es_trig_target_relations must NOT be copied */
2018 estate->es_rowMarks = parentestate->es_rowMarks;
2019 estate->es_instrument = parentestate->es_instrument;
2020 estate->es_select_into = parentestate->es_select_into;
2021 estate->es_into_oids = parentestate->es_into_oids;
2024 * The external param list is simply shared from parent. The internal
2025 * param workspace has to be local state, but we copy the initial values
2026 * from the parent, so as to have access to any param values that were
2027 * already set from other parts of the parent's plan tree.
2029 estate->es_param_list_info = parentestate->es_param_list_info;
2030 if (parentestate->es_plannedstmt->nParamExec > 0)
2032 int i = parentestate->es_plannedstmt->nParamExec;
2034 estate->es_param_exec_vals = (ParamExecData *)
2035 palloc0(i * sizeof(ParamExecData));
2038 /* copy value if any, but not execPlan link */
2039 estate->es_param_exec_vals[i].value =
2040 parentestate->es_param_exec_vals[i].value;
2041 estate->es_param_exec_vals[i].isnull =
2042 parentestate->es_param_exec_vals[i].isnull;
2047 * Each EState must have its own es_epqScanDone state, but if we have
2048 * nested EPQ checks they should share es_epqTuple arrays. This allows
2049 * sub-rechecks to inherit the values being examined by an outer recheck.
2051 estate->es_epqScanDone = (bool *) palloc0(rtsize * sizeof(bool));
2052 if (parentestate->es_epqTuple != NULL)
2054 estate->es_epqTuple = parentestate->es_epqTuple;
2055 estate->es_epqTupleSet = parentestate->es_epqTupleSet;
2059 estate->es_epqTuple = (HeapTuple *)
2060 palloc0(rtsize * sizeof(HeapTuple));
2061 estate->es_epqTupleSet = (bool *)
2062 palloc0(rtsize * sizeof(bool));
2066 * Each estate also has its own tuple table.
2068 estate->es_tupleTable = NIL;
2071 * Initialize private state information for each SubPlan. We must do this
2072 * before running ExecInitNode on the main query tree, since
2073 * ExecInitSubPlan expects to be able to find these entries. Some of the
2074 * SubPlans might not be used in the part of the plan tree we intend to
2075 * run, but since it's not easy to tell which, we just initialize them
2078 Assert(estate->es_subplanstates == NIL);
2079 foreach(l, parentestate->es_plannedstmt->subplans)
2081 Plan *subplan = (Plan *) lfirst(l);
2082 PlanState *subplanstate;
2084 subplanstate = ExecInitNode(subplan, estate, 0);
2086 estate->es_subplanstates = lappend(estate->es_subplanstates,
2091 * Initialize the private state information for all the nodes in the part
2092 * of the plan tree we need to run. This opens files, allocates storage
2093 * and leaves us ready to start processing tuples.
2095 epqstate->planstate = ExecInitNode(planTree, estate, 0);
2097 MemoryContextSwitchTo(oldcontext);
2101 * EvalPlanQualEnd -- shut down at termination of parent plan state node,
2102 * or if we are done with the current EPQ child.
2104 * This is a cut-down version of ExecutorEnd(); basically we want to do most
2105 * of the normal cleanup, but *not* close result relations (which we are
2106 * just sharing from the outer query). We do, however, have to close any
2107 * trigger target relations that got opened, since those are not shared.
2108 * (There probably shouldn't be any of the latter, but just in case...)
2111 EvalPlanQualEnd(EPQState *epqstate)
2113 EState *estate = epqstate->estate;
2114 MemoryContext oldcontext;
2118 return; /* idle, so nothing to do */
2120 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
2122 ExecEndNode(epqstate->planstate);
2124 foreach(l, estate->es_subplanstates)
2126 PlanState *subplanstate = (PlanState *) lfirst(l);
2128 ExecEndNode(subplanstate);
2131 /* throw away the per-estate tuple table */
2132 ExecResetTupleTable(estate->es_tupleTable, false);
2134 /* close any trigger target relations attached to this EState */
2135 foreach(l, estate->es_trig_target_relations)
2137 ResultRelInfo *resultRelInfo = (ResultRelInfo *) lfirst(l);
2139 /* Close indices and then the relation itself */
2140 ExecCloseIndices(resultRelInfo);
2141 heap_close(resultRelInfo->ri_RelationDesc, NoLock);
2144 MemoryContextSwitchTo(oldcontext);
2146 FreeExecutorState(estate);
2148 /* Mark EPQState idle */
2149 epqstate->estate = NULL;
2150 epqstate->planstate = NULL;
2151 epqstate->origslot = NULL;
2156 * Support for SELECT INTO (a/k/a CREATE TABLE AS)
2158 * We implement SELECT INTO by diverting SELECT's normal output with
2159 * a specialized DestReceiver type.
2164 DestReceiver pub; /* publicly-known function pointers */
2165 EState *estate; /* EState we are working with */
2166 Relation rel; /* Relation to write to */
2167 int hi_options; /* heap_insert performance options */
2168 BulkInsertState bistate; /* bulk insert state */
2172 * OpenIntoRel --- actually create the SELECT INTO target relation
2174 * This also replaces QueryDesc->dest with the special DestReceiver for
2175 * SELECT INTO. We assume that the correct result tuple type has already
2176 * been placed in queryDesc->tupDesc.
2179 OpenIntoRel(QueryDesc *queryDesc)
2181 IntoClause *into = queryDesc->plannedstmt->intoClause;
2182 EState *estate = queryDesc->estate;
2183 Relation intoRelationDesc;
2188 AclResult aclresult;
2191 DR_intorel *myState;
2192 static char *validnsps[] = HEAP_RELOPT_NAMESPACES;
2197 * XXX This code needs to be kept in sync with DefineRelation(). Maybe we
2198 * should try to use that function instead.
2202 * Check consistency of arguments
2204 if (into->onCommit != ONCOMMIT_NOOP
2205 && into->rel->relpersistence != RELPERSISTENCE_TEMP)
2207 (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
2208 errmsg("ON COMMIT can only be used on temporary tables")));
2211 * Security check: disallow creating temp tables from security-restricted
2212 * code. This is needed because calling code might not expect untrusted
2213 * tables to appear in pg_temp at the front of its search path.
2215 if (into->rel->relpersistence == RELPERSISTENCE_TEMP
2216 && InSecurityRestrictedOperation())
2218 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2219 errmsg("cannot create temporary table within security-restricted operation")));
2222 * Find namespace to create in, check its permissions
2224 intoName = into->rel->relname;
2225 namespaceId = RangeVarGetCreationNamespace(into->rel);
2227 aclresult = pg_namespace_aclcheck(namespaceId, GetUserId(),
2229 if (aclresult != ACLCHECK_OK)
2230 aclcheck_error(aclresult, ACL_KIND_NAMESPACE,
2231 get_namespace_name(namespaceId));
2234 * Select tablespace to use. If not specified, use default tablespace
2235 * (which may in turn default to database's default).
2237 if (into->tableSpaceName)
2239 tablespaceId = get_tablespace_oid(into->tableSpaceName, false);
2243 tablespaceId = GetDefaultTablespace(into->rel->relpersistence);
2244 /* note InvalidOid is OK in this case */
2247 /* Check permissions except when using the database's default space */
2248 if (OidIsValid(tablespaceId) && tablespaceId != MyDatabaseTableSpace)
2250 AclResult aclresult;
2252 aclresult = pg_tablespace_aclcheck(tablespaceId, GetUserId(),
2255 if (aclresult != ACLCHECK_OK)
2256 aclcheck_error(aclresult, ACL_KIND_TABLESPACE,
2257 get_tablespace_name(tablespaceId));
2260 /* Parse and validate any reloptions */
2261 reloptions = transformRelOptions((Datum) 0,
2267 (void) heap_reloptions(RELKIND_RELATION, reloptions, true);
2269 /* Copy the tupdesc because heap_create_with_catalog modifies it */
2270 tupdesc = CreateTupleDescCopy(queryDesc->tupDesc);
2272 /* Now we can actually create the new relation */
2273 intoRelationId = heap_create_with_catalog(intoName,
2283 into->rel->relpersistence,
2291 allowSystemTableMods,
2293 Assert(intoRelationId != InvalidOid);
2295 FreeTupleDesc(tupdesc);
2298 * Advance command counter so that the newly-created relation's catalog
2299 * tuples will be visible to heap_open.
2301 CommandCounterIncrement();
2304 * If necessary, create a TOAST table for the INTO relation. Note that
2305 * AlterTableCreateToastTable ends with CommandCounterIncrement(), so that
2306 * the TOAST table will be visible for insertion.
2308 reloptions = transformRelOptions((Datum) 0,
2315 (void) heap_reloptions(RELKIND_TOASTVALUE, reloptions, true);
2317 AlterTableCreateToastTable(intoRelationId, reloptions);
2320 * And open the constructed table for writing.
2322 intoRelationDesc = heap_open(intoRelationId, AccessExclusiveLock);
2325 * Now replace the query's DestReceiver with one for SELECT INTO
2327 queryDesc->dest = CreateDestReceiver(DestIntoRel);
2328 myState = (DR_intorel *) queryDesc->dest;
2329 Assert(myState->pub.mydest == DestIntoRel);
2330 myState->estate = estate;
2331 myState->rel = intoRelationDesc;
2334 * We can skip WAL-logging the insertions, unless PITR or streaming
2335 * replication is in use. We can skip the FSM in any case.
2337 myState->hi_options = HEAP_INSERT_SKIP_FSM |
2338 (XLogIsNeeded() ? 0 : HEAP_INSERT_SKIP_WAL);
2339 myState->bistate = GetBulkInsertState();
2341 /* Not using WAL requires smgr_targblock be initially invalid */
2342 Assert(RelationGetTargetBlock(intoRelationDesc) == InvalidBlockNumber);
2346 * CloseIntoRel --- clean up SELECT INTO at ExecutorEnd time
2349 CloseIntoRel(QueryDesc *queryDesc)
2351 DR_intorel *myState = (DR_intorel *) queryDesc->dest;
2353 /* OpenIntoRel might never have gotten called */
2354 if (myState && myState->pub.mydest == DestIntoRel && myState->rel)
2356 FreeBulkInsertState(myState->bistate);
2358 /* If we skipped using WAL, must heap_sync before commit */
2359 if (myState->hi_options & HEAP_INSERT_SKIP_WAL)
2360 heap_sync(myState->rel);
2362 /* close rel, but keep lock until commit */
2363 heap_close(myState->rel, NoLock);
2365 myState->rel = NULL;
2370 * CreateIntoRelDestReceiver -- create a suitable DestReceiver object
2373 CreateIntoRelDestReceiver(void)
2375 DR_intorel *self = (DR_intorel *) palloc0(sizeof(DR_intorel));
2377 self->pub.receiveSlot = intorel_receive;
2378 self->pub.rStartup = intorel_startup;
2379 self->pub.rShutdown = intorel_shutdown;
2380 self->pub.rDestroy = intorel_destroy;
2381 self->pub.mydest = DestIntoRel;
2383 /* private fields will be set by OpenIntoRel */
2385 return (DestReceiver *) self;
2389 * intorel_startup --- executor startup
2392 intorel_startup(DestReceiver *self, int operation, TupleDesc typeinfo)
2398 * intorel_receive --- receive one tuple
2401 intorel_receive(TupleTableSlot *slot, DestReceiver *self)
2403 DR_intorel *myState = (DR_intorel *) self;
2407 * get the heap tuple out of the tuple table slot, making sure we have a
2410 tuple = ExecMaterializeSlot(slot);
2413 * force assignment of new OID (see comments in ExecInsert)
2415 if (myState->rel->rd_rel->relhasoids)
2416 HeapTupleSetOid(tuple, InvalidOid);
2418 heap_insert(myState->rel,
2420 myState->estate->es_output_cid,
2421 myState->hi_options,
2424 /* We know this is a newly created relation, so there are no indexes */
2428 * intorel_shutdown --- executor end
2431 intorel_shutdown(DestReceiver *self)
2437 * intorel_destroy --- release DestReceiver object
2440 intorel_destroy(DestReceiver *self)