Debian Codegraph @

src/backend/executor/execMain.c

raw
1 /*-------------------------------------------------------------------------
2 *
3 * execMain.c
4 * top level executor interface routines
5 *
6 * INTERFACE ROUTINES
7 * ExecutorStart()
8 * ExecutorRun()
9 * ExecutorFinish()
10 * ExecutorEnd()
11 *
12 * These four procedures are the external interface to the executor.
13 * In each case, the query descriptor is required as an argument.
14 *
15 * ExecutorStart must be called at the beginning of execution of any
16 * query plan and ExecutorEnd must always be called at the end of
17 * execution of a plan (unless it is aborted due to error).
18 *
19 * ExecutorRun accepts direction and count arguments that specify whether
20 * the plan is to be executed forwards, backwards, and for how many tuples.
21 * In some cases ExecutorRun may be called multiple times to process all
22 * the tuples for a plan. It is also acceptable to stop short of executing
23 * the whole plan (but only if it is a SELECT).
24 *
25 * ExecutorFinish must be called after the final ExecutorRun call and
26 * before ExecutorEnd. This can be omitted only in case of EXPLAIN,
27 * which should also omit ExecutorRun.
28 *
29 * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
30 * Portions Copyright (c) 1994, Regents of the University of California
31 *
32 *
33 * IDENTIFICATION
34 * src/backend/executor/execMain.c
35 *
36 *-------------------------------------------------------------------------
37 */
38 #include "postgres.h"
39
40 #include "access/sysattr.h"
41 #include "access/table.h"
42 #include "access/tableam.h"
43 #include "access/xact.h"
44 #include "catalog/namespace.h"
45 #include "catalog/partition.h"
46 #include "commands/matview.h"
47 #include "commands/trigger.h"
48 #include "executor/executor.h"
49 #include "executor/execPartition.h"
50 #include "executor/nodeSubplan.h"
51 #include "foreign/fdwapi.h"
52 #include "mb/pg_wchar.h"
53 #include "miscadmin.h"
54 #include "nodes/queryjumble.h"
55 #include "parser/parse_relation.h"
56 #include "pgstat.h"
57 #include "rewrite/rewriteHandler.h"
58 #include "tcop/utility.h"
59 #include "utils/acl.h"
60 #include "utils/backend_status.h"
61 #include "utils/lsyscache.h"
62 #include "utils/partcache.h"
63 #include "utils/rls.h"
64 #include "utils/snapmgr.h"
65
66
67 /* Hooks for plugins to get control in ExecutorStart/Run/Finish/End */
68 ExecutorStart_hook_type ExecutorStart_hook = NULL;
69 ExecutorRun_hook_type ExecutorRun_hook = NULL;
70 ExecutorFinish_hook_type ExecutorFinish_hook = NULL;
71 ExecutorEnd_hook_type ExecutorEnd_hook = NULL;
72
73 /* Hook for plugin to get control in ExecCheckPermissions() */
74 ExecutorCheckPerms_hook_type ExecutorCheckPerms_hook = NULL;
75
76 /* decls for local routines only used within this module */
77 static void InitPlan(QueryDesc *queryDesc, int eflags);
78 static void CheckValidRowMarkRel(Relation rel, RowMarkType markType);
79 static void ExecPostprocessPlan(EState *estate);
80 static void ExecEndPlan(PlanState *planstate, EState *estate);
81 static void ExecutePlan(QueryDesc *queryDesc,
82 CmdType operation,
83 bool sendTuples,
84 uint64 numberTuples,
85 ScanDirection direction,
86 DestReceiver *dest);
87 static bool ExecCheckPermissionsModified(Oid relOid, Oid userid,
88 Bitmapset *modifiedCols,
89 AclMode requiredPerms);
90 static void ExecCheckXactReadOnly(PlannedStmt *plannedstmt);
91 static void EvalPlanQualStart(EPQState *epqstate, Plan *planTree);
92 static void ReportNotNullViolationError(ResultRelInfo *resultRelInfo,
93 TupleTableSlot *slot,
94 EState *estate, int attnum);
95
96 /* end of local decls */
97
98
99 /* ----------------------------------------------------------------
100 * ExecutorStart
101 *
102 * This routine must be called at the beginning of any execution of any
103 * query plan
104 *
105 * Takes a QueryDesc previously created by CreateQueryDesc (which is separate
106 * only because some places use QueryDescs for utility commands). The tupDesc
107 * field of the QueryDesc is filled in to describe the tuples that will be
108 * returned, and the internal fields (estate and planstate) are set up.
109 *
110 * eflags contains flag bits as described in executor.h.
111 *
112 * NB: the CurrentMemoryContext when this is called will become the parent
113 * of the per-query context used for this Executor invocation.
114 *
115 * We provide a function hook variable that lets loadable plugins
116 * get control when ExecutorStart is called. Such a plugin would
117 * normally call standard_ExecutorStart().
118 *
119 * ----------------------------------------------------------------
120 */
121 void
122 ExecutorStart(QueryDesc *queryDesc, int eflags)
123 {
124 /*
125 * In some cases (e.g. an EXECUTE statement or an execute message with the
126 * extended query protocol) the query_id won't be reported, so do it now.
127 *
128 * Note that it's harmless to report the query_id multiple times, as the
129 * call will be ignored if the top level query_id has already been
130 * reported.
131 */
132 pgstat_report_query_id(queryDesc->plannedstmt->queryId, false);
133
134 if (ExecutorStart_hook)
135 (*ExecutorStart_hook) (queryDesc, eflags);
136 else
137 standard_ExecutorStart(queryDesc, eflags);
138 }
139
140 void
141 standard_ExecutorStart(QueryDesc *queryDesc, int eflags)
142 {
143 EState *estate;
144 MemoryContext oldcontext;
145
146 /* sanity checks: queryDesc must not be started already */
147 Assert(queryDesc != NULL);
148 Assert(queryDesc->estate == NULL);
149
150 /* caller must ensure the query's snapshot is active */
151 Assert(GetActiveSnapshot() == queryDesc->snapshot);
152
153 /*
154 * If the transaction is read-only, we need to check if any writes are
155 * planned to non-temporary tables. EXPLAIN is considered read-only.
156 *
157 * Don't allow writes in parallel mode. Supporting UPDATE and DELETE
158 * would require (a) storing the combo CID hash in shared memory, rather
159 * than synchronizing it just once at the start of parallelism, and (b) an
160 * alternative to heap_update()'s reliance on xmax for mutual exclusion.
161 * INSERT may have no such troubles, but we forbid it to simplify the
162 * checks.
163 *
164 * We have lower-level defenses in CommandCounterIncrement and elsewhere
165 * against performing unsafe operations in parallel mode, but this gives a
166 * more user-friendly error message.
167 */
168 if ((XactReadOnly || IsInParallelMode()) &&
169 !(eflags & EXEC_FLAG_EXPLAIN_ONLY))
170 ExecCheckXactReadOnly(queryDesc->plannedstmt);
171
172 /*
173 * Build EState, switch into per-query memory context for startup.
174 */
175 estate = CreateExecutorState();
176 queryDesc->estate = estate;
177
178 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
179
180 /*
181 * Fill in external parameters, if any, from queryDesc; and allocate
182 * workspace for internal parameters
183 */
184 estate->es_param_list_info = queryDesc->params;
185
186 if (queryDesc->plannedstmt->paramExecTypes != NIL)
187 {
188 int nParamExec;
189
190 nParamExec = list_length(queryDesc->plannedstmt->paramExecTypes);
191 estate->es_param_exec_vals = (ParamExecData *)
192 palloc0(nParamExec * sizeof(ParamExecData));
193 }
194
195 /* We now require all callers to provide sourceText */
196 Assert(queryDesc->sourceText != NULL);
197 estate->es_sourceText = queryDesc->sourceText;
198
199 /*
200 * Fill in the query environment, if any, from queryDesc.
201 */
202 estate->es_queryEnv = queryDesc->queryEnv;
203
204 /*
205 * If non-read-only query, set the command ID to mark output tuples with
206 */
207 switch (queryDesc->operation)
208 {
209 case CMD_SELECT:
210
211 /*
212 * SELECT FOR [KEY] UPDATE/SHARE and modifying CTEs need to mark
213 * tuples
214 */
215 if (queryDesc->plannedstmt->rowMarks != NIL ||
216 queryDesc->plannedstmt->hasModifyingCTE)
217 estate->es_output_cid = GetCurrentCommandId(true);
218
219 /*
220 * A SELECT without modifying CTEs can't possibly queue triggers,
221 * so force skip-triggers mode. This is just a marginal efficiency
222 * hack, since AfterTriggerBeginQuery/AfterTriggerEndQuery aren't
223 * all that expensive, but we might as well do it.
224 */
225 if (!queryDesc->plannedstmt->hasModifyingCTE)
226 eflags |= EXEC_FLAG_SKIP_TRIGGERS;
227 break;
228
229 case CMD_INSERT:
230 case CMD_DELETE:
231 case CMD_UPDATE:
232 case CMD_MERGE:
233 estate->es_output_cid = GetCurrentCommandId(true);
234 break;
235
236 default:
237 elog(ERROR, "unrecognized operation code: %d",
238 (int) queryDesc->operation);
239 break;
240 }
241
242 /*
243 * Copy other important information into the EState
244 */
245 estate->es_snapshot = RegisterSnapshot(queryDesc->snapshot);
246 estate->es_crosscheck_snapshot = RegisterSnapshot(queryDesc->crosscheck_snapshot);
247 estate->es_top_eflags = eflags;
248 estate->es_instrument = queryDesc->instrument_options;
249 estate->es_jit_flags = queryDesc->plannedstmt->jitFlags;
250
251 /*
252 * Set up an AFTER-trigger statement context, unless told not to, or
253 * unless it's EXPLAIN-only mode (when ExecutorFinish won't be called).
254 */
255 if (!(eflags & (EXEC_FLAG_SKIP_TRIGGERS | EXEC_FLAG_EXPLAIN_ONLY)))
256 AfterTriggerBeginQuery();
257
258 /*
259 * Initialize the plan state tree
260 */
261 InitPlan(queryDesc, eflags);
262
263 MemoryContextSwitchTo(oldcontext);
264 }
265
266 /* ----------------------------------------------------------------
267 * ExecutorRun
268 *
269 * This is the main routine of the executor module. It accepts
270 * the query descriptor from the traffic cop and executes the
271 * query plan.
272 *
273 * ExecutorStart must have been called already.
274 *
275 * If direction is NoMovementScanDirection then nothing is done
276 * except to start up/shut down the destination. Otherwise,
277 * we retrieve up to 'count' tuples in the specified direction.
278 *
279 * Note: count = 0 is interpreted as no portal limit, i.e., run to
280 * completion. Also note that the count limit is only applied to
281 * retrieved tuples, not for instance to those inserted/updated/deleted
282 * by a ModifyTable plan node.
283 *
284 * There is no return value, but output tuples (if any) are sent to
285 * the destination receiver specified in the QueryDesc; and the number
286 * of tuples processed at the top level can be found in
287 * estate->es_processed. The total number of tuples processed in all
288 * the ExecutorRun calls can be found in estate->es_total_processed.
289 *
290 * We provide a function hook variable that lets loadable plugins
291 * get control when ExecutorRun is called. Such a plugin would
292 * normally call standard_ExecutorRun().
293 *
294 * ----------------------------------------------------------------
295 */
296 void
297 ExecutorRun(QueryDesc *queryDesc,
298 ScanDirection direction, uint64 count)
299 {
300 if (ExecutorRun_hook)
301 (*ExecutorRun_hook) (queryDesc, direction, count);
302 else
303 standard_ExecutorRun(queryDesc, direction, count);
304 }
305
306 void
307 standard_ExecutorRun(QueryDesc *queryDesc,
308 ScanDirection direction, uint64 count)
309 {
310 EState *estate;
311 CmdType operation;
312 DestReceiver *dest;
313 bool sendTuples;
314 MemoryContext oldcontext;
315
316 /* sanity checks */
317 Assert(queryDesc != NULL);
318
319 estate = queryDesc->estate;
320
321 Assert(estate != NULL);
322 Assert(!(estate->es_top_eflags & EXEC_FLAG_EXPLAIN_ONLY));
323
324 /* caller must ensure the query's snapshot is active */
325 Assert(GetActiveSnapshot() == estate->es_snapshot);
326
327 /*
328 * Switch into per-query memory context
329 */
330 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
331
332 /* Allow instrumentation of Executor overall runtime */
333 if (queryDesc->totaltime)
334 InstrStartNode(queryDesc->totaltime);
335
336 /*
337 * extract information from the query descriptor and the query feature.
338 */
339 operation = queryDesc->operation;
340 dest = queryDesc->dest;
341
342 /*
343 * startup tuple receiver, if we will be emitting tuples
344 */
345 estate->es_processed = 0;
346
347 sendTuples = (operation == CMD_SELECT ||
348 queryDesc->plannedstmt->hasReturning);
349
350 if (sendTuples)
351 dest->rStartup(dest, operation, queryDesc->tupDesc);
352
353 /*
354 * Run plan, unless direction is NoMovement.
355 *
356 * Note: pquery.c selects NoMovement if a prior call already reached
357 * end-of-data in the user-specified fetch direction. This is important
358 * because various parts of the executor can misbehave if called again
359 * after reporting EOF. For example, heapam.c would actually restart a
360 * heapscan and return all its data afresh. There is also some doubt
361 * about whether a parallel plan would operate properly if an additional,
362 * necessarily non-parallel execution request occurs after completing a
363 * parallel execution. (That case should work, but it's untested.)
364 */
365 if (!ScanDirectionIsNoMovement(direction))
366 ExecutePlan(queryDesc,
367 operation,
368 sendTuples,
369 count,
370 direction,
371 dest);
372
373 /*
374 * Update es_total_processed to keep track of the number of tuples
375 * processed across multiple ExecutorRun() calls.
376 */
377 estate->es_total_processed += estate->es_processed;
378
379 /*
380 * shutdown tuple receiver, if we started it
381 */
382 if (sendTuples)
383 dest->rShutdown(dest);
384
385 if (queryDesc->totaltime)
386 InstrStopNode(queryDesc->totaltime, estate->es_processed);
387
388 MemoryContextSwitchTo(oldcontext);
389 }
390
391 /* ----------------------------------------------------------------
392 * ExecutorFinish
393 *
394 * This routine must be called after the last ExecutorRun call.
395 * It performs cleanup such as firing AFTER triggers. It is
396 * separate from ExecutorEnd because EXPLAIN ANALYZE needs to
397 * include these actions in the total runtime.
398 *
399 * We provide a function hook variable that lets loadable plugins
400 * get control when ExecutorFinish is called. Such a plugin would
401 * normally call standard_ExecutorFinish().
402 *
403 * ----------------------------------------------------------------
404 */
405 void
406 ExecutorFinish(QueryDesc *queryDesc)
407 {
408 if (ExecutorFinish_hook)
409 (*ExecutorFinish_hook) (queryDesc);
410 else
411 standard_ExecutorFinish(queryDesc);
412 }
413
414 void
415 standard_ExecutorFinish(QueryDesc *queryDesc)
416 {
417 EState *estate;
418 MemoryContext oldcontext;
419
420 /* sanity checks */
421 Assert(queryDesc != NULL);
422
423 estate = queryDesc->estate;
424
425 Assert(estate != NULL);
426 Assert(!(estate->es_top_eflags & EXEC_FLAG_EXPLAIN_ONLY));
427
428 /* This should be run once and only once per Executor instance */
429 Assert(!estate->es_finished);
430
431 /* Switch into per-query memory context */
432 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
433
434 /* Allow instrumentation of Executor overall runtime */
435 if (queryDesc->totaltime)
436 InstrStartNode(queryDesc->totaltime);
437
438 /* Run ModifyTable nodes to completion */
439 ExecPostprocessPlan(estate);
440
441 /* Execute queued AFTER triggers, unless told not to */
442 if (!(estate->es_top_eflags & EXEC_FLAG_SKIP_TRIGGERS))
443 AfterTriggerEndQuery(estate);
444
445 if (queryDesc->totaltime)
446 InstrStopNode(queryDesc->totaltime, 0);
447
448 MemoryContextSwitchTo(oldcontext);
449
450 estate->es_finished = true;
451 }
452
453 /* ----------------------------------------------------------------
454 * ExecutorEnd
455 *
456 * This routine must be called at the end of execution of any
457 * query plan
458 *
459 * We provide a function hook variable that lets loadable plugins
460 * get control when ExecutorEnd is called. Such a plugin would
461 * normally call standard_ExecutorEnd().
462 *
463 * ----------------------------------------------------------------
464 */
465 void
466 ExecutorEnd(QueryDesc *queryDesc)
467 {
468 if (ExecutorEnd_hook)
469 (*ExecutorEnd_hook) (queryDesc);
470 else
471 standard_ExecutorEnd(queryDesc);
472 }
473
474 void
475 standard_ExecutorEnd(QueryDesc *queryDesc)
476 {
477 EState *estate;
478 MemoryContext oldcontext;
479
480 /* sanity checks */
481 Assert(queryDesc != NULL);
482
483 estate = queryDesc->estate;
484
485 Assert(estate != NULL);
486
487 if (estate->es_parallel_workers_to_launch > 0)
488 pgstat_update_parallel_workers_stats((PgStat_Counter) estate->es_parallel_workers_to_launch,
489 (PgStat_Counter) estate->es_parallel_workers_launched);
490
491 /*
492 * Check that ExecutorFinish was called, unless in EXPLAIN-only mode. This
493 * Assert is needed because ExecutorFinish is new as of 9.1, and callers
494 * might forget to call it.
495 */
496 Assert(estate->es_finished ||
497 (estate->es_top_eflags & EXEC_FLAG_EXPLAIN_ONLY));
498
499 /*
500 * Switch into per-query memory context to run ExecEndPlan
501 */
502 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
503
504 ExecEndPlan(queryDesc->planstate, estate);
505
506 /* do away with our snapshots */
507 UnregisterSnapshot(estate->es_snapshot);
508 UnregisterSnapshot(estate->es_crosscheck_snapshot);
509
510 /*
511 * Must switch out of context before destroying it
512 */
513 MemoryContextSwitchTo(oldcontext);
514
515 /*
516 * Release EState and per-query memory context. This should release
517 * everything the executor has allocated.
518 */
519 FreeExecutorState(estate);
520
521 /* Reset queryDesc fields that no longer point to anything */
522 queryDesc->tupDesc = NULL;
523 queryDesc->estate = NULL;
524 queryDesc->planstate = NULL;
525 queryDesc->totaltime = NULL;
526 }
527
528 /* ----------------------------------------------------------------
529 * ExecutorRewind
530 *
531 * This routine may be called on an open queryDesc to rewind it
532 * to the start.
533 * ----------------------------------------------------------------
534 */
535 void
536 ExecutorRewind(QueryDesc *queryDesc)
537 {
538 EState *estate;
539 MemoryContext oldcontext;
540
541 /* sanity checks */
542 Assert(queryDesc != NULL);
543
544 estate = queryDesc->estate;
545
546 Assert(estate != NULL);
547
548 /* It's probably not sensible to rescan updating queries */
549 Assert(queryDesc->operation == CMD_SELECT);
550
551 /*
552 * Switch into per-query memory context
553 */
554 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
555
556 /*
557 * rescan plan
558 */
559 ExecReScan(queryDesc->planstate);
560
561 MemoryContextSwitchTo(oldcontext);
562 }
563
564
565 /*
566 * ExecCheckPermissions
567 * Check access permissions of relations mentioned in a query
568 *
569 * Returns true if permissions are adequate. Otherwise, throws an appropriate
570 * error if ereport_on_violation is true, or simply returns false otherwise.
571 *
572 * Note that this does NOT address row-level security policies (aka: RLS). If
573 * rows will be returned to the user as a result of this permission check
574 * passing, then RLS also needs to be consulted (and check_enable_rls()).
575 *
576 * See rewrite/rowsecurity.c.
577 *
578 * NB: rangeTable is no longer used by us, but kept around for the hooks that
579 * might still want to look at the RTEs.
580 */
581 bool
582 ExecCheckPermissions(List *rangeTable, List *rteperminfos,
583 bool ereport_on_violation)
584 {
585 ListCell *l;
586 bool result = true;
587
588 #ifdef USE_ASSERT_CHECKING
589 Bitmapset *indexset = NULL;
590
591 /* Check that rteperminfos is consistent with rangeTable */
592 foreach(l, rangeTable)
593 {
594 RangeTblEntry *rte = lfirst_node(RangeTblEntry, l);
595
596 if (rte->perminfoindex != 0)
597 {
598 /* Sanity checks */
599
600 /*
601 * Only relation RTEs and subquery RTEs that were once relation
602 * RTEs (views) have their perminfoindex set.
603 */
604 Assert(rte->rtekind == RTE_RELATION ||
605 (rte->rtekind == RTE_SUBQUERY &&
606 rte->relkind == RELKIND_VIEW));
607
608 (void) getRTEPermissionInfo(rteperminfos, rte);
609 /* Many-to-one mapping not allowed */
610 Assert(!bms_is_member(rte->perminfoindex, indexset));
611 indexset = bms_add_member(indexset, rte->perminfoindex);
612 }
613 }
614
615 /* All rteperminfos are referenced */
616 Assert(bms_num_members(indexset) == list_length(rteperminfos));
617 #endif
618
619 foreach(l, rteperminfos)
620 {
621 RTEPermissionInfo *perminfo = lfirst_node(RTEPermissionInfo, l);
622
623 Assert(OidIsValid(perminfo->relid));
624 result = ExecCheckOneRelPerms(perminfo);
625 if (!result)
626 {
627 if (ereport_on_violation)
628 aclcheck_error(ACLCHECK_NO_PRIV,
629 get_relkind_objtype(get_rel_relkind(perminfo->relid)),
630 get_rel_name(perminfo->relid));
631 return false;
632 }
633 }
634
635 if (ExecutorCheckPerms_hook)
636 result = (*ExecutorCheckPerms_hook) (rangeTable, rteperminfos,
637 ereport_on_violation);
638 return result;
639 }
640
641 /*
642 * ExecCheckOneRelPerms
643 * Check access permissions for a single relation.
644 */
645 bool
646 ExecCheckOneRelPerms(RTEPermissionInfo *perminfo)
647 {
648 AclMode requiredPerms;
649 AclMode relPerms;
650 AclMode remainingPerms;
651 Oid userid;
652 Oid relOid = perminfo->relid;
653
654 requiredPerms = perminfo->requiredPerms;
655 Assert(requiredPerms != 0);
656
657 /*
658 * userid to check as: current user unless we have a setuid indication.
659 *
660 * Note: GetUserId() is presently fast enough that there's no harm in
661 * calling it separately for each relation. If that stops being true, we
662 * could call it once in ExecCheckPermissions and pass the userid down
663 * from there. But for now, no need for the extra clutter.
664 */
665 userid = OidIsValid(perminfo->checkAsUser) ?
666 perminfo->checkAsUser : GetUserId();
667
668 /*
669 * We must have *all* the requiredPerms bits, but some of the bits can be
670 * satisfied from column-level rather than relation-level permissions.
671 * First, remove any bits that are satisfied by relation permissions.
672 */
673 relPerms = pg_class_aclmask(relOid, userid, requiredPerms, ACLMASK_ALL);
674 remainingPerms = requiredPerms & ~relPerms;
675 if (remainingPerms != 0)
676 {
677 int col = -1;
678
679 /*
680 * If we lack any permissions that exist only as relation permissions,
681 * we can fail straight away.
682 */
683 if (remainingPerms & ~(ACL_SELECT | ACL_INSERT | ACL_UPDATE))
684 return false;
685
686 /*
687 * Check to see if we have the needed privileges at column level.
688 *
689 * Note: failures just report a table-level error; it would be nicer
690 * to report a column-level error if we have some but not all of the
691 * column privileges.
692 */
693 if (remainingPerms & ACL_SELECT)
694 {
695 /*
696 * When the query doesn't explicitly reference any columns (for
697 * example, SELECT COUNT(*) FROM table), allow the query if we
698 * have SELECT on any column of the rel, as per SQL spec.
699 */
700 if (bms_is_empty(perminfo->selectedCols))
701 {
702 if (pg_attribute_aclcheck_all(relOid, userid, ACL_SELECT,
703 ACLMASK_ANY) != ACLCHECK_OK)
704 return false;
705 }
706
707 while ((col = bms_next_member(perminfo->selectedCols, col)) >= 0)
708 {
709 /* bit #s are offset by FirstLowInvalidHeapAttributeNumber */
710 AttrNumber attno = col + FirstLowInvalidHeapAttributeNumber;
711
712 if (attno == InvalidAttrNumber)
713 {
714 /* Whole-row reference, must have priv on all cols */
715 if (pg_attribute_aclcheck_all(relOid, userid, ACL_SELECT,
716 ACLMASK_ALL) != ACLCHECK_OK)
717 return false;
718 }
719 else
720 {
721 if (pg_attribute_aclcheck(relOid, attno, userid,
722 ACL_SELECT) != ACLCHECK_OK)
723 return false;
724 }
725 }
726 }
727
728 /*
729 * Basically the same for the mod columns, for both INSERT and UPDATE
730 * privilege as specified by remainingPerms.
731 */
732 if (remainingPerms & ACL_INSERT &&
733 !ExecCheckPermissionsModified(relOid,
734 userid,
735 perminfo->insertedCols,
736 ACL_INSERT))
737 return false;
738
739 if (remainingPerms & ACL_UPDATE &&
740 !ExecCheckPermissionsModified(relOid,
741 userid,
742 perminfo->updatedCols,
743 ACL_UPDATE))
744 return false;
745 }
746 return true;
747 }
748
749 /*
750 * ExecCheckPermissionsModified
751 * Check INSERT or UPDATE access permissions for a single relation (these
752 * are processed uniformly).
753 */
754 static bool
755 ExecCheckPermissionsModified(Oid relOid, Oid userid, Bitmapset *modifiedCols,
756 AclMode requiredPerms)
757 {
758 int col = -1;
759
760 /*
761 * When the query doesn't explicitly update any columns, allow the query
762 * if we have permission on any column of the rel. This is to handle
763 * SELECT FOR UPDATE as well as possible corner cases in UPDATE.
764 */
765 if (bms_is_empty(modifiedCols))
766 {
767 if (pg_attribute_aclcheck_all(relOid, userid, requiredPerms,
768 ACLMASK_ANY) != ACLCHECK_OK)
769 return false;
770 }
771
772 while ((col = bms_next_member(modifiedCols, col)) >= 0)
773 {
774 /* bit #s are offset by FirstLowInvalidHeapAttributeNumber */
775 AttrNumber attno = col + FirstLowInvalidHeapAttributeNumber;
776
777 if (attno == InvalidAttrNumber)
778 {
779 /* whole-row reference can't happen here */
780 elog(ERROR, "whole-row update is not implemented");
781 }
782 else
783 {
784 if (pg_attribute_aclcheck(relOid, attno, userid,
785 requiredPerms) != ACLCHECK_OK)
786 return false;
787 }
788 }
789 return true;
790 }
791
792 /*
793 * Check that the query does not imply any writes to non-temp tables;
794 * unless we're in parallel mode, in which case don't even allow writes
795 * to temp tables.
796 *
797 * Note: in a Hot Standby this would need to reject writes to temp
798 * tables just as we do in parallel mode; but an HS standby can't have created
799 * any temp tables in the first place, so no need to check that.
800 */
801 static void
802 ExecCheckXactReadOnly(PlannedStmt *plannedstmt)
803 {
804 ListCell *l;
805
806 /*
807 * Fail if write permissions are requested in parallel mode for table
808 * (temp or non-temp), otherwise fail for any non-temp table.
809 */
810 foreach(l, plannedstmt->permInfos)
811 {
812 RTEPermissionInfo *perminfo = lfirst_node(RTEPermissionInfo, l);
813
814 if ((perminfo->requiredPerms & (~ACL_SELECT)) == 0)
815 continue;
816
817 if (isTempNamespace(get_rel_namespace(perminfo->relid)))
818 continue;
819
820 PreventCommandIfReadOnly(CreateCommandName((Node *) plannedstmt));
821 }
822
823 if (plannedstmt->commandType != CMD_SELECT || plannedstmt->hasModifyingCTE)
824 PreventCommandIfParallelMode(CreateCommandName((Node *) plannedstmt));
825 }
826
827
828 /* ----------------------------------------------------------------
829 * InitPlan
830 *
831 * Initializes the query plan: open files, allocate storage
832 * and start up the rule manager
833 * ----------------------------------------------------------------
834 */
835 static void
836 InitPlan(QueryDesc *queryDesc, int eflags)
837 {
838 CmdType operation = queryDesc->operation;
839 PlannedStmt *plannedstmt = queryDesc->plannedstmt;
840 Plan *plan = plannedstmt->planTree;
841 List *rangeTable = plannedstmt->rtable;
842 EState *estate = queryDesc->estate;
843 PlanState *planstate;
844 TupleDesc tupType;
845 ListCell *l;
846 int i;
847
848 /*
849 * Do permissions checks
850 */
851 ExecCheckPermissions(rangeTable, plannedstmt->permInfos, true);
852
853 /*
854 * initialize the node's execution state
855 */
856 ExecInitRangeTable(estate, rangeTable, plannedstmt->permInfos,
857 bms_copy(plannedstmt->unprunableRelids));
858
859 estate->es_plannedstmt = plannedstmt;
860 estate->es_part_prune_infos = plannedstmt->partPruneInfos;
861
862 /*
863 * Perform runtime "initial" pruning to identify which child subplans,
864 * corresponding to the children of plan nodes that contain
865 * PartitionPruneInfo such as Append, will not be executed. The results,
866 * which are bitmapsets of indexes of the child subplans that will be
867 * executed, are saved in es_part_prune_results. These results correspond
868 * to each PartitionPruneInfo entry, and the es_part_prune_results list is
869 * parallel to es_part_prune_infos.
870 */
871 ExecDoInitialPruning(estate);
872
873 /*
874 * Next, build the ExecRowMark array from the PlanRowMark(s), if any.
875 */
876 if (plannedstmt->rowMarks)
877 {
878 estate->es_rowmarks = (ExecRowMark **)
879 palloc0(estate->es_range_table_size * sizeof(ExecRowMark *));
880 foreach(l, plannedstmt->rowMarks)
881 {
882 PlanRowMark *rc = (PlanRowMark *) lfirst(l);
883 RangeTblEntry *rte = exec_rt_fetch(rc->rti, estate);
884 Oid relid;
885 Relation relation;
886 ExecRowMark *erm;
887
888 /* ignore "parent" rowmarks; they are irrelevant at runtime */
889 if (rc->isParent)
890 continue;
891
892 /*
893 * Also ignore rowmarks belonging to child tables that have been
894 * pruned in ExecDoInitialPruning().
895 */
896 if (rte->rtekind == RTE_RELATION &&
897 !bms_is_member(rc->rti, estate->es_unpruned_relids))
898 continue;
899
900 /* get relation's OID (will produce InvalidOid if subquery) */
901 relid = rte->relid;
902
903 /* open relation, if we need to access it for this mark type */
904 switch (rc->markType)
905 {
906 case ROW_MARK_EXCLUSIVE:
907 case ROW_MARK_NOKEYEXCLUSIVE:
908 case ROW_MARK_SHARE:
909 case ROW_MARK_KEYSHARE:
910 case ROW_MARK_REFERENCE:
911 relation = ExecGetRangeTableRelation(estate, rc->rti, false);
912 break;
913 case ROW_MARK_COPY:
914 /* no physical table access is required */
915 relation = NULL;
916 break;
917 default:
918 elog(ERROR, "unrecognized markType: %d", rc->markType);
919 relation = NULL; /* keep compiler quiet */
920 break;
921 }
922
923 /* Check that relation is a legal target for marking */
924 if (relation)
925 CheckValidRowMarkRel(relation, rc->markType);
926
927 erm = (ExecRowMark *) palloc(sizeof(ExecRowMark));
928 erm->relation = relation;
929 erm->relid = relid;
930 erm->rti = rc->rti;
931 erm->prti = rc->prti;
932 erm->rowmarkId = rc->rowmarkId;
933 erm->markType = rc->markType;
934 erm->strength = rc->strength;
935 erm->waitPolicy = rc->waitPolicy;
936 erm->ermActive = false;
937 ItemPointerSetInvalid(&(erm->curCtid));
938 erm->ermExtra = NULL;
939
940 Assert(erm->rti > 0 && erm->rti <= estate->es_range_table_size &&
941 estate->es_rowmarks[erm->rti - 1] == NULL);
942
943 estate->es_rowmarks[erm->rti - 1] = erm;
944 }
945 }
946
947 /*
948 * Initialize the executor's tuple table to empty.
949 */
950 estate->es_tupleTable = NIL;
951
952 /* signal that this EState is not used for EPQ */
953 estate->es_epq_active = NULL;
954
955 /*
956 * Initialize private state information for each SubPlan. We must do this
957 * before running ExecInitNode on the main query tree, since
958 * ExecInitSubPlan expects to be able to find these entries.
959 */
960 Assert(estate->es_subplanstates == NIL);
961 i = 1; /* subplan indices count from 1 */
962 foreach(l, plannedstmt->subplans)
963 {
964 Plan *subplan = (Plan *) lfirst(l);
965 PlanState *subplanstate;
966 int sp_eflags;
967
968 /*
969 * A subplan will never need to do BACKWARD scan nor MARK/RESTORE. If
970 * it is a parameterless subplan (not initplan), we suggest that it be
971 * prepared to handle REWIND efficiently; otherwise there is no need.
972 */
973 sp_eflags = eflags
974 & ~(EXEC_FLAG_REWIND | EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK);
975 if (bms_is_member(i, plannedstmt->rewindPlanIDs))
976 sp_eflags |= EXEC_FLAG_REWIND;
977
978 subplanstate = ExecInitNode(subplan, estate, sp_eflags);
979
980 estate->es_subplanstates = lappend(estate->es_subplanstates,
981 subplanstate);
982
983 i++;
984 }
985
986 /*
987 * Initialize the private state information for all the nodes in the query
988 * tree. This opens files, allocates storage and leaves us ready to start
989 * processing tuples.
990 */
991 planstate = ExecInitNode(plan, estate, eflags);
992
993 /*
994 * Get the tuple descriptor describing the type of tuples to return.
995 */
996 tupType = ExecGetResultType(planstate);
997
998 /*
999 * Initialize the junk filter if needed. SELECT queries need a filter if
1000 * there are any junk attrs in the top-level tlist.
1001 */
1002 if (operation == CMD_SELECT)
1003 {
1004 bool junk_filter_needed = false;
1005 ListCell *tlist;
1006
1007 foreach(tlist, plan->targetlist)
1008 {
1009 TargetEntry *tle = (TargetEntry *) lfirst(tlist);
1010
1011 if (tle->resjunk)
1012 {
1013 junk_filter_needed = true;
1014 break;
1015 }
1016 }
1017
1018 if (junk_filter_needed)
1019 {
1020 JunkFilter *j;
1021 TupleTableSlot *slot;
1022
1023 slot = ExecInitExtraTupleSlot(estate, NULL, &TTSOpsVirtual);
1024 j = ExecInitJunkFilter(planstate->plan->targetlist,
1025 slot);
1026 estate->es_junkFilter = j;
1027
1028 /* Want to return the cleaned tuple type */
1029 tupType = j->jf_cleanTupType;
1030 }
1031 }
1032
1033 queryDesc->tupDesc = tupType;
1034 queryDesc->planstate = planstate;
1035 }
1036
1037 /*
1038 * Check that a proposed result relation is a legal target for the operation
1039 *
1040 * Generally the parser and/or planner should have noticed any such mistake
1041 * already, but let's make sure.
1042 *
1043 * For INSERT ON CONFLICT, the result relation is required to support the
1044 * onConflictAction, regardless of whether a conflict actually occurs.
1045 *
1046 * For MERGE, mergeActions is the list of actions that may be performed. The
1047 * result relation is required to support every action, regardless of whether
1048 * or not they are all executed.
1049 *
1050 * Note: when changing this function, you probably also need to look at
1051 * CheckValidRowMarkRel.
1052 */
1053 void
1054 CheckValidResultRel(ResultRelInfo *resultRelInfo, CmdType operation,
1055 OnConflictAction onConflictAction, List *mergeActions)
1056 {
1057 Relation resultRel = resultRelInfo->ri_RelationDesc;
1058 FdwRoutine *fdwroutine;
1059
1060 /* Expect a fully-formed ResultRelInfo from InitResultRelInfo(). */
1061 Assert(resultRelInfo->ri_needLockTagTuple ==
1062 IsInplaceUpdateRelation(resultRel));
1063
1064 switch (resultRel->rd_rel->relkind)
1065 {
1066 case RELKIND_RELATION:
1067 case RELKIND_PARTITIONED_TABLE:
1068
1069 /*
1070 * For MERGE, check that the target relation supports each action.
1071 * For other operations, just check the operation itself.
1072 */
1073 if (operation == CMD_MERGE)
1074 foreach_node(MergeAction, action, mergeActions)
1075 CheckCmdReplicaIdentity(resultRel, action->commandType);
1076 else
1077 CheckCmdReplicaIdentity(resultRel, operation);
1078
1079 /*
1080 * For INSERT ON CONFLICT DO UPDATE, additionally check that the
1081 * target relation supports UPDATE.
1082 */
1083 if (onConflictAction == ONCONFLICT_UPDATE)
1084 CheckCmdReplicaIdentity(resultRel, CMD_UPDATE);
1085 break;
1086 case RELKIND_SEQUENCE:
1087 ereport(ERROR,
1088 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1089 errmsg("cannot change sequence \"%s\"",
1090 RelationGetRelationName(resultRel))));
1091 break;
1092 case RELKIND_TOASTVALUE:
1093 ereport(ERROR,
1094 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1095 errmsg("cannot change TOAST relation \"%s\"",
1096 RelationGetRelationName(resultRel))));
1097 break;
1098 case RELKIND_VIEW:
1099
1100 /*
1101 * Okay only if there's a suitable INSTEAD OF trigger. Otherwise,
1102 * complain, but omit errdetail because we haven't got the
1103 * information handy (and given that it really shouldn't happen,
1104 * it's not worth great exertion to get).
1105 */
1106 if (!view_has_instead_trigger(resultRel, operation, mergeActions))
1107 error_view_not_updatable(resultRel, operation, mergeActions,
1108 NULL);
1109 break;
1110 case RELKIND_MATVIEW:
1111 if (!MatViewIncrementalMaintenanceIsEnabled())
1112 ereport(ERROR,
1113 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1114 errmsg("cannot change materialized view \"%s\"",
1115 RelationGetRelationName(resultRel))));
1116 break;
1117 case RELKIND_FOREIGN_TABLE:
1118 /* Okay only if the FDW supports it */
1119 fdwroutine = resultRelInfo->ri_FdwRoutine;
1120 switch (operation)
1121 {
1122 case CMD_INSERT:
1123 if (fdwroutine->ExecForeignInsert == NULL)
1124 ereport(ERROR,
1125 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1126 errmsg("cannot insert into foreign table \"%s\"",
1127 RelationGetRelationName(resultRel))));
1128 if (fdwroutine->IsForeignRelUpdatable != NULL &&
1129 (fdwroutine->IsForeignRelUpdatable(resultRel) & (1 << CMD_INSERT)) == 0)
1130 ereport(ERROR,
1131 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1132 errmsg("foreign table \"%s\" does not allow inserts",
1133 RelationGetRelationName(resultRel))));
1134 break;
1135 case CMD_UPDATE:
1136 if (fdwroutine->ExecForeignUpdate == NULL)
1137 ereport(ERROR,
1138 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1139 errmsg("cannot update foreign table \"%s\"",
1140 RelationGetRelationName(resultRel))));
1141 if (fdwroutine->IsForeignRelUpdatable != NULL &&
1142 (fdwroutine->IsForeignRelUpdatable(resultRel) & (1 << CMD_UPDATE)) == 0)
1143 ereport(ERROR,
1144 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1145 errmsg("foreign table \"%s\" does not allow updates",
1146 RelationGetRelationName(resultRel))));
1147 break;
1148 case CMD_DELETE:
1149 if (fdwroutine->ExecForeignDelete == NULL)
1150 ereport(ERROR,
1151 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1152 errmsg("cannot delete from foreign table \"%s\"",
1153 RelationGetRelationName(resultRel))));
1154 if (fdwroutine->IsForeignRelUpdatable != NULL &&
1155 (fdwroutine->IsForeignRelUpdatable(resultRel) & (1 << CMD_DELETE)) == 0)
1156 ereport(ERROR,
1157 (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1158 errmsg("foreign table \"%s\" does not allow deletes",
1159 RelationGetRelationName(resultRel))));
1160 break;
1161 default:
1162 elog(ERROR, "unrecognized CmdType: %d", (int) operation);
1163 break;
1164 }
1165 break;
1166 default:
1167 ereport(ERROR,
1168 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1169 errmsg("cannot change relation \"%s\"",
1170 RelationGetRelationName(resultRel))));
1171 break;
1172 }
1173 }
1174
1175 /*
1176 * Check that a proposed rowmark target relation is a legal target
1177 *
1178 * In most cases parser and/or planner should have noticed this already, but
1179 * they don't cover all cases.
1180 */
1181 static void
1182 CheckValidRowMarkRel(Relation rel, RowMarkType markType)
1183 {
1184 FdwRoutine *fdwroutine;
1185
1186 switch (rel->rd_rel->relkind)
1187 {
1188 case RELKIND_RELATION:
1189 case RELKIND_PARTITIONED_TABLE:
1190 /* OK */
1191 break;
1192 case RELKIND_SEQUENCE:
1193 /* Must disallow this because we don't vacuum sequences */
1194 ereport(ERROR,
1195 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1196 errmsg("cannot lock rows in sequence \"%s\"",
1197 RelationGetRelationName(rel))));
1198 break;
1199 case RELKIND_TOASTVALUE:
1200 /* We could allow this, but there seems no good reason to */
1201 ereport(ERROR,
1202 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1203 errmsg("cannot lock rows in TOAST relation \"%s\"",
1204 RelationGetRelationName(rel))));
1205 break;
1206 case RELKIND_VIEW:
1207 /* Should not get here; planner should have expanded the view */
1208 ereport(ERROR,
1209 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1210 errmsg("cannot lock rows in view \"%s\"",
1211 RelationGetRelationName(rel))));
1212 break;
1213 case RELKIND_MATVIEW:
1214 /* Allow referencing a matview, but not actual locking clauses */
1215 if (markType != ROW_MARK_REFERENCE)
1216 ereport(ERROR,
1217 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1218 errmsg("cannot lock rows in materialized view \"%s\"",
1219 RelationGetRelationName(rel))));
1220 break;
1221 case RELKIND_FOREIGN_TABLE:
1222 /* Okay only if the FDW supports it */
1223 fdwroutine = GetFdwRoutineForRelation(rel, false);
1224 if (fdwroutine->RefetchForeignRow == NULL)
1225 ereport(ERROR,
1226 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1227 errmsg("cannot lock rows in foreign table \"%s\"",
1228 RelationGetRelationName(rel))));
1229 break;
1230 default:
1231 ereport(ERROR,
1232 (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1233 errmsg("cannot lock rows in relation \"%s\"",
1234 RelationGetRelationName(rel))));
1235 break;
1236 }
1237 }
1238
1239 /*
1240 * Initialize ResultRelInfo data for one result relation
1241 *
1242 * Caution: before Postgres 9.1, this function included the relkind checking
1243 * that's now in CheckValidResultRel, and it also did ExecOpenIndices if
1244 * appropriate. Be sure callers cover those needs.
1245 */
1246 void
1247 InitResultRelInfo(ResultRelInfo *resultRelInfo,
1248 Relation resultRelationDesc,
1249 Index resultRelationIndex,
1250 ResultRelInfo *partition_root_rri,
1251 int instrument_options)
1252 {
1253 MemSet(resultRelInfo, 0, sizeof(ResultRelInfo));
1254 resultRelInfo->type = T_ResultRelInfo;
1255 resultRelInfo->ri_RangeTableIndex = resultRelationIndex;
1256 resultRelInfo->ri_RelationDesc = resultRelationDesc;
1257 resultRelInfo->ri_NumIndices = 0;
1258 resultRelInfo->ri_IndexRelationDescs = NULL;
1259 resultRelInfo->ri_IndexRelationInfo = NULL;
1260 resultRelInfo->ri_needLockTagTuple =
1261 IsInplaceUpdateRelation(resultRelationDesc);
1262 /* make a copy so as not to depend on relcache info not changing... */
1263 resultRelInfo->ri_TrigDesc = CopyTriggerDesc(resultRelationDesc->trigdesc);
1264 if (resultRelInfo->ri_TrigDesc)
1265 {
1266 int n = resultRelInfo->ri_TrigDesc->numtriggers;
1267
1268 resultRelInfo->ri_TrigFunctions = (FmgrInfo *)
1269 palloc0(n * sizeof(FmgrInfo));
1270 resultRelInfo->ri_TrigWhenExprs = (ExprState **)
1271 palloc0(n * sizeof(ExprState *));
1272 if (instrument_options)
1273 resultRelInfo->ri_TrigInstrument = InstrAlloc(n, instrument_options, false);
1274 }
1275 else
1276 {
1277 resultRelInfo->ri_TrigFunctions = NULL;
1278 resultRelInfo->ri_TrigWhenExprs = NULL;
1279 resultRelInfo->ri_TrigInstrument = NULL;
1280 }
1281 if (resultRelationDesc->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
1282 resultRelInfo->ri_FdwRoutine = GetFdwRoutineForRelation(resultRelationDesc, true);
1283 else
1284 resultRelInfo->ri_FdwRoutine = NULL;
1285
1286 /* The following fields are set later if needed */
1287 resultRelInfo->ri_RowIdAttNo = 0;
1288 resultRelInfo->ri_extraUpdatedCols = NULL;
1289 resultRelInfo->ri_projectNew = NULL;
1290 resultRelInfo->ri_newTupleSlot = NULL;
1291 resultRelInfo->ri_oldTupleSlot = NULL;
1292 resultRelInfo->ri_projectNewInfoValid = false;
1293 resultRelInfo->ri_FdwState = NULL;
1294 resultRelInfo->ri_usesFdwDirectModify = false;
1295 resultRelInfo->ri_CheckConstraintExprs = NULL;
1296 resultRelInfo->ri_GenVirtualNotNullConstraintExprs = NULL;
1297 resultRelInfo->ri_GeneratedExprsI = NULL;
1298 resultRelInfo->ri_GeneratedExprsU = NULL;
1299 resultRelInfo->ri_projectReturning = NULL;
1300 resultRelInfo->ri_onConflictArbiterIndexes = NIL;
1301 resultRelInfo->ri_onConflict = NULL;
1302 resultRelInfo->ri_ReturningSlot = NULL;
1303 resultRelInfo->ri_TrigOldSlot = NULL;
1304 resultRelInfo->ri_TrigNewSlot = NULL;
1305 resultRelInfo->ri_AllNullSlot = NULL;
1306 resultRelInfo->ri_MergeActions[MERGE_WHEN_MATCHED] = NIL;
1307 resultRelInfo->ri_MergeActions[MERGE_WHEN_NOT_MATCHED_BY_SOURCE] = NIL;
1308 resultRelInfo->ri_MergeActions[MERGE_WHEN_NOT_MATCHED_BY_TARGET] = NIL;
1309 resultRelInfo->ri_MergeJoinCondition = NULL;
1310
1311 /*
1312 * Only ExecInitPartitionInfo() and ExecInitPartitionDispatchInfo() pass
1313 * non-NULL partition_root_rri. For child relations that are part of the
1314 * initial query rather than being dynamically added by tuple routing,
1315 * this field is filled in ExecInitModifyTable().
1316 */
1317 resultRelInfo->ri_RootResultRelInfo = partition_root_rri;
1318 /* Set by ExecGetRootToChildMap */
1319 resultRelInfo->ri_RootToChildMap = NULL;
1320 resultRelInfo->ri_RootToChildMapValid = false;
1321 /* Set by ExecInitRoutingInfo */
1322 resultRelInfo->ri_PartitionTupleSlot = NULL;
1323 resultRelInfo->ri_ChildToRootMap = NULL;
1324 resultRelInfo->ri_ChildToRootMapValid = false;
1325 resultRelInfo->ri_CopyMultiInsertBuffer = NULL;
1326 }
1327
1328 /*
1329 * ExecGetTriggerResultRel
1330 * Get a ResultRelInfo for a trigger target relation.
1331 *
1332 * Most of the time, triggers are fired on one of the result relations of the
1333 * query, and so we can just return a suitable one we already made and stored
1334 * in the es_opened_result_relations or es_tuple_routing_result_relations
1335 * Lists.
1336 *
1337 * However, it is sometimes necessary to fire triggers on other relations;
1338 * this happens mainly when an RI update trigger queues additional triggers
1339 * on other relations, which will be processed in the context of the outer
1340 * query. For efficiency's sake, we want to have a ResultRelInfo for those
1341 * triggers too; that can avoid repeated re-opening of the relation. (It
1342 * also provides a way for EXPLAIN ANALYZE to report the runtimes of such
1343 * triggers.) So we make additional ResultRelInfo's as needed, and save them
1344 * in es_trig_target_relations.
1345 */
1346 ResultRelInfo *
1347 ExecGetTriggerResultRel(EState *estate, Oid relid,
1348 ResultRelInfo *rootRelInfo)
1349 {
1350 ResultRelInfo *rInfo;
1351 ListCell *l;
1352 Relation rel;
1353 MemoryContext oldcontext;
1354
1355 /*
1356 * Before creating a new ResultRelInfo, check if we've already made and
1357 * cached one for this relation. We must ensure that the given
1358 * 'rootRelInfo' matches the one stored in the cached ResultRelInfo as
1359 * trigger handling for partitions can result in mixed requirements for
1360 * what ri_RootResultRelInfo is set to.
1361 */
1362
1363 /* Search through the query result relations */
1364 foreach(l, estate->es_opened_result_relations)
1365 {
1366 rInfo = lfirst(l);
1367 if (RelationGetRelid(rInfo->ri_RelationDesc) == relid &&
1368 rInfo->ri_RootResultRelInfo == rootRelInfo)
1369 return rInfo;
1370 }
1371
1372 /*
1373 * Search through the result relations that were created during tuple
1374 * routing, if any.
1375 */
1376 foreach(l, estate->es_tuple_routing_result_relations)
1377 {
1378 rInfo = (ResultRelInfo *) lfirst(l);
1379 if (RelationGetRelid(rInfo->ri_RelationDesc) == relid &&
1380 rInfo->ri_RootResultRelInfo == rootRelInfo)
1381 return rInfo;
1382 }
1383
1384 /* Nope, but maybe we already made an extra ResultRelInfo for it */
1385 foreach(l, estate->es_trig_target_relations)
1386 {
1387 rInfo = (ResultRelInfo *) lfirst(l);
1388 if (RelationGetRelid(rInfo->ri_RelationDesc) == relid &&
1389 rInfo->ri_RootResultRelInfo == rootRelInfo)
1390 return rInfo;
1391 }
1392 /* Nope, so we need a new one */
1393
1394 /*
1395 * Open the target relation's relcache entry. We assume that an
1396 * appropriate lock is still held by the backend from whenever the trigger
1397 * event got queued, so we need take no new lock here. Also, we need not
1398 * recheck the relkind, so no need for CheckValidResultRel.
1399 */
1400 rel = table_open(relid, NoLock);
1401
1402 /*
1403 * Make the new entry in the right context.
1404 */
1405 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
1406 rInfo = makeNode(ResultRelInfo);
1407 InitResultRelInfo(rInfo,
1408 rel,
1409 0, /* dummy rangetable index */
1410 rootRelInfo,
1411 estate->es_instrument);
1412 estate->es_trig_target_relations =
1413 lappend(estate->es_trig_target_relations, rInfo);
1414 MemoryContextSwitchTo(oldcontext);
1415
1416 /*
1417 * Currently, we don't need any index information in ResultRelInfos used
1418 * only for triggers, so no need to call ExecOpenIndices.
1419 */
1420
1421 return rInfo;
1422 }
1423
1424 /*
1425 * Return the ancestor relations of a given leaf partition result relation
1426 * up to and including the query's root target relation.
1427 *
1428 * These work much like the ones opened by ExecGetTriggerResultRel, except
1429 * that we need to keep them in a separate list.
1430 *
1431 * These are closed by ExecCloseResultRelations.
1432 */
1433 List *
1434 ExecGetAncestorResultRels(EState *estate, ResultRelInfo *resultRelInfo)
1435 {
1436 ResultRelInfo *rootRelInfo = resultRelInfo->ri_RootResultRelInfo;
1437 Relation partRel = resultRelInfo->ri_RelationDesc;
1438 Oid rootRelOid;
1439
1440 if (!partRel->rd_rel->relispartition)
1441 elog(ERROR, "cannot find ancestors of a non-partition result relation");
1442 Assert(rootRelInfo != NULL);
1443 rootRelOid = RelationGetRelid(rootRelInfo->ri_RelationDesc);
1444 if (resultRelInfo->ri_ancestorResultRels == NIL)
1445 {
1446 ListCell *lc;
1447 List *oids = get_partition_ancestors(RelationGetRelid(partRel));
1448 List *ancResultRels = NIL;
1449
1450 foreach(lc, oids)
1451 {
1452 Oid ancOid = lfirst_oid(lc);
1453 Relation ancRel;
1454 ResultRelInfo *rInfo;
1455
1456 /*
1457 * Ignore the root ancestor here, and use ri_RootResultRelInfo
1458 * (below) for it instead. Also, we stop climbing up the
1459 * hierarchy when we find the table that was mentioned in the
1460 * query.
1461 */
1462 if (ancOid == rootRelOid)
1463 break;
1464
1465 /*
1466 * All ancestors up to the root target relation must have been
1467 * locked by the planner or AcquireExecutorLocks().
1468 */
1469 ancRel = table_open(ancOid, NoLock);
1470 rInfo = makeNode(ResultRelInfo);
1471
1472 /* dummy rangetable index */
1473 InitResultRelInfo(rInfo, ancRel, 0, NULL,
1474 estate->es_instrument);
1475 ancResultRels = lappend(ancResultRels, rInfo);
1476 }
1477 ancResultRels = lappend(ancResultRels, rootRelInfo);
1478 resultRelInfo->ri_ancestorResultRels = ancResultRels;
1479 }
1480
1481 /* We must have found some ancestor */
1482 Assert(resultRelInfo->ri_ancestorResultRels != NIL);
1483
1484 return resultRelInfo->ri_ancestorResultRels;
1485 }
1486
1487 /* ----------------------------------------------------------------
1488 * ExecPostprocessPlan
1489 *
1490 * Give plan nodes a final chance to execute before shutdown
1491 * ----------------------------------------------------------------
1492 */
1493 static void
1494 ExecPostprocessPlan(EState *estate)
1495 {
1496 ListCell *lc;
1497
1498 /*
1499 * Make sure nodes run forward.
1500 */
1501 estate->es_direction = ForwardScanDirection;
1502
1503 /*
1504 * Run any secondary ModifyTable nodes to completion, in case the main
1505 * query did not fetch all rows from them. (We do this to ensure that
1506 * such nodes have predictable results.)
1507 */
1508 foreach(lc, estate->es_auxmodifytables)
1509 {
1510 PlanState *ps = (PlanState *) lfirst(lc);
1511
1512 for (;;)
1513 {
1514 TupleTableSlot *slot;
1515
1516 /* Reset the per-output-tuple exprcontext each time */
1517 ResetPerTupleExprContext(estate);
1518
1519 slot = ExecProcNode(ps);
1520
1521 if (TupIsNull(slot))
1522 break;
1523 }
1524 }
1525 }
1526
1527 /* ----------------------------------------------------------------
1528 * ExecEndPlan
1529 *
1530 * Cleans up the query plan -- closes files and frees up storage
1531 *
1532 * NOTE: we are no longer very worried about freeing storage per se
1533 * in this code; FreeExecutorState should be guaranteed to release all
1534 * memory that needs to be released. What we are worried about doing
1535 * is closing relations and dropping buffer pins. Thus, for example,
1536 * tuple tables must be cleared or dropped to ensure pins are released.
1537 * ----------------------------------------------------------------
1538 */
1539 static void
1540 ExecEndPlan(PlanState *planstate, EState *estate)
1541 {
1542 ListCell *l;
1543
1544 /*
1545 * shut down the node-type-specific query processing
1546 */
1547 ExecEndNode(planstate);
1548
1549 /*
1550 * for subplans too
1551 */
1552 foreach(l, estate->es_subplanstates)
1553 {
1554 PlanState *subplanstate = (PlanState *) lfirst(l);
1555
1556 ExecEndNode(subplanstate);
1557 }
1558
1559 /*
1560 * destroy the executor's tuple table. Actually we only care about
1561 * releasing buffer pins and tupdesc refcounts; there's no need to pfree
1562 * the TupleTableSlots, since the containing memory context is about to go
1563 * away anyway.
1564 */
1565 ExecResetTupleTable(estate->es_tupleTable, false);
1566
1567 /*
1568 * Close any Relations that have been opened for range table entries or
1569 * result relations.
1570 */
1571 ExecCloseResultRelations(estate);
1572 ExecCloseRangeTableRelations(estate);
1573 }
1574
1575 /*
1576 * Close any relations that have been opened for ResultRelInfos.
1577 */
1578 void
1579 ExecCloseResultRelations(EState *estate)
1580 {
1581 ListCell *l;
1582
1583 /*
1584 * close indexes of result relation(s) if any. (Rels themselves are
1585 * closed in ExecCloseRangeTableRelations())
1586 *
1587 * In addition, close the stub RTs that may be in each resultrel's
1588 * ri_ancestorResultRels.
1589 */
1590 foreach(l, estate->es_opened_result_relations)
1591 {
1592 ResultRelInfo *resultRelInfo = lfirst(l);
1593 ListCell *lc;
1594
1595 ExecCloseIndices(resultRelInfo);
1596 foreach(lc, resultRelInfo->ri_ancestorResultRels)
1597 {
1598 ResultRelInfo *rInfo = lfirst(lc);
1599
1600 /*
1601 * Ancestors with RTI > 0 (should only be the root ancestor) are
1602 * closed by ExecCloseRangeTableRelations.
1603 */
1604 if (rInfo->ri_RangeTableIndex > 0)
1605 continue;
1606
1607 table_close(rInfo->ri_RelationDesc, NoLock);
1608 }
1609 }
1610
1611 /* Close any relations that have been opened by ExecGetTriggerResultRel(). */
1612 foreach(l, estate->es_trig_target_relations)
1613 {
1614 ResultRelInfo *resultRelInfo = (ResultRelInfo *) lfirst(l);
1615
1616 /*
1617 * Assert this is a "dummy" ResultRelInfo, see above. Otherwise we
1618 * might be issuing a duplicate close against a Relation opened by
1619 * ExecGetRangeTableRelation.
1620 */
1621 Assert(resultRelInfo->ri_RangeTableIndex == 0);
1622
1623 /*
1624 * Since ExecGetTriggerResultRel doesn't call ExecOpenIndices for
1625 * these rels, we needn't call ExecCloseIndices either.
1626 */
1627 Assert(resultRelInfo->ri_NumIndices == 0);
1628
1629 table_close(resultRelInfo->ri_RelationDesc, NoLock);
1630 }
1631 }
1632
1633 /*
1634 * Close all relations opened by ExecGetRangeTableRelation().
1635 *
1636 * We do not release any locks we might hold on those rels.
1637 */
1638 void
1639 ExecCloseRangeTableRelations(EState *estate)
1640 {
1641 int i;
1642
1643 for (i = 0; i < estate->es_range_table_size; i++)
1644 {
1645 if (estate->es_relations[i])
1646 table_close(estate->es_relations[i], NoLock);
1647 }
1648 }
1649
1650 /* ----------------------------------------------------------------
1651 * ExecutePlan
1652 *
1653 * Processes the query plan until we have retrieved 'numberTuples' tuples,
1654 * moving in the specified direction.
1655 *
1656 * Runs to completion if numberTuples is 0
1657 * ----------------------------------------------------------------
1658 */
1659 static void
1660 ExecutePlan(QueryDesc *queryDesc,
1661 CmdType operation,
1662 bool sendTuples,
1663 uint64 numberTuples,
1664 ScanDirection direction,
1665 DestReceiver *dest)
1666 {
1667 EState *estate = queryDesc->estate;
1668 PlanState *planstate = queryDesc->planstate;
1669 bool use_parallel_mode;
1670 TupleTableSlot *slot;
1671 uint64 current_tuple_count;
1672
1673 /*
1674 * initialize local variables
1675 */
1676 current_tuple_count = 0;
1677
1678 /*
1679 * Set the direction.
1680 */
1681 estate->es_direction = direction;
1682
1683 /*
1684 * Set up parallel mode if appropriate.
1685 *
1686 * Parallel mode only supports complete execution of a plan. If we've
1687 * already partially executed it, or if the caller asks us to exit early,
1688 * we must force the plan to run without parallelism.
1689 */
1690 if (queryDesc->already_executed || numberTuples != 0)
1691 use_parallel_mode = false;
1692 else
1693 use_parallel_mode = queryDesc->plannedstmt->parallelModeNeeded;
1694 queryDesc->already_executed = true;
1695
1696 estate->es_use_parallel_mode = use_parallel_mode;
1697 if (use_parallel_mode)
1698 EnterParallelMode();
1699
1700 /*
1701 * Loop until we've processed the proper number of tuples from the plan.
1702 */
1703 for (;;)
1704 {
1705 /* Reset the per-output-tuple exprcontext */
1706 ResetPerTupleExprContext(estate);
1707
1708 /*
1709 * Execute the plan and obtain a tuple
1710 */
1711 slot = ExecProcNode(planstate);
1712
1713 /*
1714 * if the tuple is null, then we assume there is nothing more to
1715 * process so we just end the loop...
1716 */
1717 if (TupIsNull(slot))
1718 break;
1719
1720 /*
1721 * If we have a junk filter, then project a new tuple with the junk
1722 * removed.
1723 *
1724 * Store this new "clean" tuple in the junkfilter's resultSlot.
1725 * (Formerly, we stored it back over the "dirty" tuple, which is WRONG
1726 * because that tuple slot has the wrong descriptor.)
1727 */
1728 if (estate->es_junkFilter != NULL)
1729 slot = ExecFilterJunk(estate->es_junkFilter, slot);
1730
1731 /*
1732 * If we are supposed to send the tuple somewhere, do so. (In
1733 * practice, this is probably always the case at this point.)
1734 */
1735 if (sendTuples)
1736 {
1737 /*
1738 * If we are not able to send the tuple, we assume the destination
1739 * has closed and no more tuples can be sent. If that's the case,
1740 * end the loop.
1741 */
1742 if (!dest->receiveSlot(slot, dest))
1743 break;
1744 }
1745
1746 /*
1747 * Count tuples processed, if this is a SELECT. (For other operation
1748 * types, the ModifyTable plan node must count the appropriate
1749 * events.)
1750 */
1751 if (operation == CMD_SELECT)
1752 (estate->es_processed)++;
1753
1754 /*
1755 * check our tuple count.. if we've processed the proper number then
1756 * quit, else loop again and process more tuples. Zero numberTuples
1757 * means no limit.
1758 */
1759 current_tuple_count++;
1760 if (numberTuples && numberTuples == current_tuple_count)
1761 break;
1762 }
1763
1764 /*
1765 * If we know we won't need to back up, we can release resources at this
1766 * point.
1767 */
1768 if (!(estate->es_top_eflags & EXEC_FLAG_BACKWARD))
1769 ExecShutdownNode(planstate);
1770
1771 if (use_parallel_mode)
1772 ExitParallelMode();
1773 }
1774
1775
1776 /*
1777 * ExecRelCheck --- check that tuple meets check constraints for result relation
1778 *
1779 * Returns NULL if OK, else name of failed check constraint
1780 */
1781 static const char *
1782 ExecRelCheck(ResultRelInfo *resultRelInfo,
1783 TupleTableSlot *slot, EState *estate)
1784 {
1785 Relation rel = resultRelInfo->ri_RelationDesc;
1786 int ncheck = rel->rd_att->constr->num_check;
1787 ConstrCheck *check = rel->rd_att->constr->check;
1788 ExprContext *econtext;
1789 MemoryContext oldContext;
1790
1791 /*
1792 * CheckNNConstraintFetch let this pass with only a warning, but now we
1793 * should fail rather than possibly failing to enforce an important
1794 * constraint.
1795 */
1796 if (ncheck != rel->rd_rel->relchecks)
1797 elog(ERROR, "%d pg_constraint record(s) missing for relation \"%s\"",
1798 rel->rd_rel->relchecks - ncheck, RelationGetRelationName(rel));
1799
1800 /*
1801 * If first time through for this result relation, build expression
1802 * nodetrees for rel's constraint expressions. Keep them in the per-query
1803 * memory context so they'll survive throughout the query.
1804 */
1805 if (resultRelInfo->ri_CheckConstraintExprs == NULL)
1806 {
1807 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
1808 resultRelInfo->ri_CheckConstraintExprs = palloc0_array(ExprState *, ncheck);
1809 for (int i = 0; i < ncheck; i++)
1810 {
1811 Expr *checkconstr;
1812
1813 /* Skip not enforced constraint */
1814 if (!check[i].ccenforced)
1815 continue;
1816
1817 checkconstr = stringToNode(check[i].ccbin);
1818 checkconstr = (Expr *) expand_generated_columns_in_expr((Node *) checkconstr, rel, 1);
1819 resultRelInfo->ri_CheckConstraintExprs[i] =
1820 ExecPrepareExpr(checkconstr, estate);
1821 }
1822 MemoryContextSwitchTo(oldContext);
1823 }
1824
1825 /*
1826 * We will use the EState's per-tuple context for evaluating constraint
1827 * expressions (creating it if it's not already there).
1828 */
1829 econtext = GetPerTupleExprContext(estate);
1830
1831 /* Arrange for econtext's scan tuple to be the tuple under test */
1832 econtext->ecxt_scantuple = slot;
1833
1834 /* And evaluate the constraints */
1835 for (int i = 0; i < ncheck; i++)
1836 {
1837 ExprState *checkconstr = resultRelInfo->ri_CheckConstraintExprs[i];
1838
1839 /*
1840 * NOTE: SQL specifies that a NULL result from a constraint expression
1841 * is not to be treated as a failure. Therefore, use ExecCheck not
1842 * ExecQual.
1843 */
1844 if (checkconstr && !ExecCheck(checkconstr, econtext))
1845 return check[i].ccname;
1846 }
1847
1848 /* NULL result means no error */
1849 return NULL;
1850 }
1851
1852 /*
1853 * ExecPartitionCheck --- check that tuple meets the partition constraint.
1854 *
1855 * Returns true if it meets the partition constraint. If the constraint
1856 * fails and we're asked to emit an error, do so and don't return; otherwise
1857 * return false.
1858 */
1859 bool
1860 ExecPartitionCheck(ResultRelInfo *resultRelInfo, TupleTableSlot *slot,
1861 EState *estate, bool emitError)
1862 {
1863 ExprContext *econtext;
1864 bool success;
1865
1866 /*
1867 * If first time through, build expression state tree for the partition
1868 * check expression. (In the corner case where the partition check
1869 * expression is empty, ie there's a default partition and nothing else,
1870 * we'll be fooled into executing this code each time through. But it's
1871 * pretty darn cheap in that case, so we don't worry about it.)
1872 */
1873 if (resultRelInfo->ri_PartitionCheckExpr == NULL)
1874 {
1875 /*
1876 * Ensure that the qual tree and prepared expression are in the
1877 * query-lifespan context.
1878 */
1879 MemoryContext oldcxt = MemoryContextSwitchTo(estate->es_query_cxt);
1880 List *qual = RelationGetPartitionQual(resultRelInfo->ri_RelationDesc);
1881
1882 resultRelInfo->ri_PartitionCheckExpr = ExecPrepareCheck(qual, estate);
1883 MemoryContextSwitchTo(oldcxt);
1884 }
1885
1886 /*
1887 * We will use the EState's per-tuple context for evaluating constraint
1888 * expressions (creating it if it's not already there).
1889 */
1890 econtext = GetPerTupleExprContext(estate);
1891
1892 /* Arrange for econtext's scan tuple to be the tuple under test */
1893 econtext->ecxt_scantuple = slot;
1894
1895 /*
1896 * As in case of the cataloged constraints, we treat a NULL result as
1897 * success here, not a failure.
1898 */
1899 success = ExecCheck(resultRelInfo->ri_PartitionCheckExpr, econtext);
1900
1901 /* if asked to emit error, don't actually return on failure */
1902 if (!success && emitError)
1903 ExecPartitionCheckEmitError(resultRelInfo, slot, estate);
1904
1905 return success;
1906 }
1907
1908 /*
1909 * ExecPartitionCheckEmitError - Form and emit an error message after a failed
1910 * partition constraint check.
1911 */
1912 void
1913 ExecPartitionCheckEmitError(ResultRelInfo *resultRelInfo,
1914 TupleTableSlot *slot,
1915 EState *estate)
1916 {
1917 Oid root_relid;
1918 TupleDesc tupdesc;
1919 char *val_desc;
1920 Bitmapset *modifiedCols;
1921
1922 /*
1923 * If the tuple has been routed, it's been converted to the partition's
1924 * rowtype, which might differ from the root table's. We must convert it
1925 * back to the root table's rowtype so that val_desc in the error message
1926 * matches the input tuple.
1927 */
1928 if (resultRelInfo->ri_RootResultRelInfo)
1929 {
1930 ResultRelInfo *rootrel = resultRelInfo->ri_RootResultRelInfo;
1931 TupleDesc old_tupdesc;
1932 AttrMap *map;
1933
1934 root_relid = RelationGetRelid(rootrel->ri_RelationDesc);
1935 tupdesc = RelationGetDescr(rootrel->ri_RelationDesc);
1936
1937 old_tupdesc = RelationGetDescr(resultRelInfo->ri_RelationDesc);
1938 /* a reverse map */
1939 map = build_attrmap_by_name_if_req(old_tupdesc, tupdesc, false);
1940
1941 /*
1942 * Partition-specific slot's tupdesc can't be changed, so allocate a
1943 * new one.
1944 */
1945 if (map != NULL)
1946 slot = execute_attr_map_slot(map, slot,
1947 MakeTupleTableSlot(tupdesc, &TTSOpsVirtual));
1948 modifiedCols = bms_union(ExecGetInsertedCols(rootrel, estate),
1949 ExecGetUpdatedCols(rootrel, estate));
1950 }
1951 else
1952 {
1953 root_relid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
1954 tupdesc = RelationGetDescr(resultRelInfo->ri_RelationDesc);
1955 modifiedCols = bms_union(ExecGetInsertedCols(resultRelInfo, estate),
1956 ExecGetUpdatedCols(resultRelInfo, estate));
1957 }
1958
1959 val_desc = ExecBuildSlotValueDescription(root_relid,
1960 slot,
1961 tupdesc,
1962 modifiedCols,
1963 64);
1964 ereport(ERROR,
1965 (errcode(ERRCODE_CHECK_VIOLATION),
1966 errmsg("new row for relation \"%s\" violates partition constraint",
1967 RelationGetRelationName(resultRelInfo->ri_RelationDesc)),
1968 val_desc ? errdetail("Failing row contains %s.", val_desc) : 0,
1969 errtable(resultRelInfo->ri_RelationDesc)));
1970 }
1971
1972 /*
1973 * ExecConstraints - check constraints of the tuple in 'slot'
1974 *
1975 * This checks the traditional NOT NULL and check constraints.
1976 *
1977 * The partition constraint is *NOT* checked.
1978 *
1979 * Note: 'slot' contains the tuple to check the constraints of, which may
1980 * have been converted from the original input tuple after tuple routing.
1981 * 'resultRelInfo' is the final result relation, after tuple routing.
1982 */
1983 void
1984 ExecConstraints(ResultRelInfo *resultRelInfo,
1985 TupleTableSlot *slot, EState *estate)
1986 {
1987 Relation rel = resultRelInfo->ri_RelationDesc;
1988 TupleDesc tupdesc = RelationGetDescr(rel);
1989 TupleConstr *constr = tupdesc->constr;
1990 Bitmapset *modifiedCols;
1991 List *notnull_virtual_attrs = NIL;
1992
1993 Assert(constr); /* we should not be called otherwise */
1994
1995 /*
1996 * Verify not-null constraints.
1997 *
1998 * Not-null constraints on virtual generated columns are collected and
1999 * checked separately below.
2000 */
2001 if (constr->has_not_null)
2002 {
2003 for (AttrNumber attnum = 1; attnum <= tupdesc->natts; attnum++)
2004 {
2005 Form_pg_attribute att = TupleDescAttr(tupdesc, attnum - 1);
2006
2007 if (att->attnotnull && att->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
2008 notnull_virtual_attrs = lappend_int(notnull_virtual_attrs, attnum);
2009 else if (att->attnotnull && slot_attisnull(slot, attnum))
2010 ReportNotNullViolationError(resultRelInfo, slot, estate, attnum);
2011 }
2012 }
2013
2014 /*
2015 * Verify not-null constraints on virtual generated column, if any.
2016 */
2017 if (notnull_virtual_attrs)
2018 {
2019 AttrNumber attnum;
2020
2021 attnum = ExecRelGenVirtualNotNull(resultRelInfo, slot, estate,
2022 notnull_virtual_attrs);
2023 if (attnum != InvalidAttrNumber)
2024 ReportNotNullViolationError(resultRelInfo, slot, estate, attnum);
2025 }
2026
2027 /*
2028 * Verify check constraints.
2029 */
2030 if (rel->rd_rel->relchecks > 0)
2031 {
2032 const char *failed;
2033
2034 if ((failed = ExecRelCheck(resultRelInfo, slot, estate)) != NULL)
2035 {
2036 char *val_desc;
2037 Relation orig_rel = rel;
2038
2039 /*
2040 * If the tuple has been routed, it's been converted to the
2041 * partition's rowtype, which might differ from the root table's.
2042 * We must convert it back to the root table's rowtype so that
2043 * val_desc shown error message matches the input tuple.
2044 */
2045 if (resultRelInfo->ri_RootResultRelInfo)
2046 {
2047 ResultRelInfo *rootrel = resultRelInfo->ri_RootResultRelInfo;
2048 TupleDesc old_tupdesc = RelationGetDescr(rel);
2049 AttrMap *map;
2050
2051 tupdesc = RelationGetDescr(rootrel->ri_RelationDesc);
2052 /* a reverse map */
2053 map = build_attrmap_by_name_if_req(old_tupdesc,
2054 tupdesc,
2055 false);
2056
2057 /*
2058 * Partition-specific slot's tupdesc can't be changed, so
2059 * allocate a new one.
2060 */
2061 if (map != NULL)
2062 slot = execute_attr_map_slot(map, slot,
2063 MakeTupleTableSlot(tupdesc, &TTSOpsVirtual));
2064 modifiedCols = bms_union(ExecGetInsertedCols(rootrel, estate),
2065 ExecGetUpdatedCols(rootrel, estate));
2066 rel = rootrel->ri_RelationDesc;
2067 }
2068 else
2069 modifiedCols = bms_union(ExecGetInsertedCols(resultRelInfo, estate),
2070 ExecGetUpdatedCols(resultRelInfo, estate));
2071 val_desc = ExecBuildSlotValueDescription(RelationGetRelid(rel),
2072 slot,
2073 tupdesc,
2074 modifiedCols,
2075 64);
2076 ereport(ERROR,
2077 (errcode(ERRCODE_CHECK_VIOLATION),
2078 errmsg("new row for relation \"%s\" violates check constraint \"%s\"",
2079 RelationGetRelationName(orig_rel), failed),
2080 val_desc ? errdetail("Failing row contains %s.", val_desc) : 0,
2081 errtableconstraint(orig_rel, failed)));
2082 }
2083 }
2084 }
2085
2086 /*
2087 * Verify not-null constraints on virtual generated columns of the given
2088 * tuple slot.
2089 *
2090 * Return value of InvalidAttrNumber means all not-null constraints on virtual
2091 * generated columns are satisfied. A return value > 0 means a not-null
2092 * violation happened for that attribute.
2093 *
2094 * notnull_virtual_attrs is the list of the attnums of virtual generated column with
2095 * not-null constraints.
2096 */
2097 AttrNumber
2098 ExecRelGenVirtualNotNull(ResultRelInfo *resultRelInfo, TupleTableSlot *slot,
2099 EState *estate, List *notnull_virtual_attrs)
2100 {
2101 Relation rel = resultRelInfo->ri_RelationDesc;
2102 ExprContext *econtext;
2103 MemoryContext oldContext;
2104
2105 /*
2106 * We implement this by building a NullTest node for each virtual
2107 * generated column, which we cache in resultRelInfo, and running those
2108 * through ExecCheck().
2109 */
2110 if (resultRelInfo->ri_GenVirtualNotNullConstraintExprs == NULL)
2111 {
2112 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
2113 resultRelInfo->ri_GenVirtualNotNullConstraintExprs =
2114 palloc0_array(ExprState *, list_length(notnull_virtual_attrs));
2115
2116 foreach_int(attnum, notnull_virtual_attrs)
2117 {
2118 int i = foreach_current_index(attnum);
2119 NullTest *nnulltest;
2120
2121 /* "generated_expression IS NOT NULL" check. */
2122 nnulltest = makeNode(NullTest);
2123 nnulltest->arg = (Expr *) build_generation_expression(rel, attnum);
2124 nnulltest->nulltesttype = IS_NOT_NULL;
2125 nnulltest->argisrow = false;
2126 nnulltest->location = -1;
2127
2128 resultRelInfo->ri_GenVirtualNotNullConstraintExprs[i] =
2129 ExecPrepareExpr((Expr *) nnulltest, estate);
2130 }
2131 MemoryContextSwitchTo(oldContext);
2132 }
2133
2134 /*
2135 * We will use the EState's per-tuple context for evaluating virtual
2136 * generated column not null constraint expressions (creating it if it's
2137 * not already there).
2138 */
2139 econtext = GetPerTupleExprContext(estate);
2140
2141 /* Arrange for econtext's scan tuple to be the tuple under test */
2142 econtext->ecxt_scantuple = slot;
2143
2144 /* And evaluate the check constraints for virtual generated column */
2145 foreach_int(attnum, notnull_virtual_attrs)
2146 {
2147 int i = foreach_current_index(attnum);
2148 ExprState *exprstate = resultRelInfo->ri_GenVirtualNotNullConstraintExprs[i];
2149
2150 Assert(exprstate != NULL);
2151 if (!ExecCheck(exprstate, econtext))
2152 return attnum;
2153 }
2154
2155 /* InvalidAttrNumber result means no error */
2156 return InvalidAttrNumber;
2157 }
2158
2159 /*
2160 * Report a violation of a not-null constraint that was already detected.
2161 */
2162 static void
2163 ReportNotNullViolationError(ResultRelInfo *resultRelInfo, TupleTableSlot *slot,
2164 EState *estate, int attnum)
2165 {
2166 Bitmapset *modifiedCols;
2167 char *val_desc;
2168 Relation rel = resultRelInfo->ri_RelationDesc;
2169 Relation orig_rel = rel;
2170 TupleDesc tupdesc = RelationGetDescr(rel);
2171 TupleDesc orig_tupdesc = RelationGetDescr(rel);
2172 Form_pg_attribute att = TupleDescAttr(tupdesc, attnum - 1);
2173
2174 Assert(attnum > 0);
2175
2176 /*
2177 * If the tuple has been routed, it's been converted to the partition's
2178 * rowtype, which might differ from the root table's. We must convert it
2179 * back to the root table's rowtype so that val_desc shown error message
2180 * matches the input tuple.
2181 */
2182 if (resultRelInfo->ri_RootResultRelInfo)
2183 {
2184 ResultRelInfo *rootrel = resultRelInfo->ri_RootResultRelInfo;
2185 AttrMap *map;
2186
2187 tupdesc = RelationGetDescr(rootrel->ri_RelationDesc);
2188 /* a reverse map */
2189 map = build_attrmap_by_name_if_req(orig_tupdesc,
2190 tupdesc,
2191 false);
2192
2193 /*
2194 * Partition-specific slot's tupdesc can't be changed, so allocate a
2195 * new one.
2196 */
2197 if (map != NULL)
2198 slot = execute_attr_map_slot(map, slot,
2199 MakeTupleTableSlot(tupdesc, &TTSOpsVirtual));
2200 modifiedCols = bms_union(ExecGetInsertedCols(rootrel, estate),
2201 ExecGetUpdatedCols(rootrel, estate));
2202 rel = rootrel->ri_RelationDesc;
2203 }
2204 else
2205 modifiedCols = bms_union(ExecGetInsertedCols(resultRelInfo, estate),
2206 ExecGetUpdatedCols(resultRelInfo, estate));
2207
2208 val_desc = ExecBuildSlotValueDescription(RelationGetRelid(rel),
2209 slot,
2210 tupdesc,
2211 modifiedCols,
2212 64);
2213 ereport(ERROR,
2214 errcode(ERRCODE_NOT_NULL_VIOLATION),
2215 errmsg("null value in column \"%s\" of relation \"%s\" violates not-null constraint",
2216 NameStr(att->attname),
2217 RelationGetRelationName(orig_rel)),
2218 val_desc ? errdetail("Failing row contains %s.", val_desc) : 0,
2219 errtablecol(orig_rel, attnum));
2220 }
2221
2222 /*
2223 * ExecWithCheckOptions -- check that tuple satisfies any WITH CHECK OPTIONs
2224 * of the specified kind.
2225 *
2226 * Note that this needs to be called multiple times to ensure that all kinds of
2227 * WITH CHECK OPTIONs are handled (both those from views which have the WITH
2228 * CHECK OPTION set and from row-level security policies). See ExecInsert()
2229 * and ExecUpdate().
2230 */
2231 void
2232 ExecWithCheckOptions(WCOKind kind, ResultRelInfo *resultRelInfo,
2233 TupleTableSlot *slot, EState *estate)
2234 {
2235 Relation rel = resultRelInfo->ri_RelationDesc;
2236 TupleDesc tupdesc = RelationGetDescr(rel);
2237 ExprContext *econtext;
2238 ListCell *l1,
2239 *l2;
2240
2241 /*
2242 * We will use the EState's per-tuple context for evaluating constraint
2243 * expressions (creating it if it's not already there).
2244 */
2245 econtext = GetPerTupleExprContext(estate);
2246
2247 /* Arrange for econtext's scan tuple to be the tuple under test */
2248 econtext->ecxt_scantuple = slot;
2249
2250 /* Check each of the constraints */
2251 forboth(l1, resultRelInfo->ri_WithCheckOptions,
2252 l2, resultRelInfo->ri_WithCheckOptionExprs)
2253 {
2254 WithCheckOption *wco = (WithCheckOption *) lfirst(l1);
2255 ExprState *wcoExpr = (ExprState *) lfirst(l2);
2256
2257 /*
2258 * Skip any WCOs which are not the kind we are looking for at this
2259 * time.
2260 */
2261 if (wco->kind != kind)
2262 continue;
2263
2264 /*
2265 * WITH CHECK OPTION checks are intended to ensure that the new tuple
2266 * is visible (in the case of a view) or that it passes the
2267 * 'with-check' policy (in the case of row security). If the qual
2268 * evaluates to NULL or FALSE, then the new tuple won't be included in
2269 * the view or doesn't pass the 'with-check' policy for the table.
2270 */
2271 if (!ExecQual(wcoExpr, econtext))
2272 {
2273 char *val_desc;
2274 Bitmapset *modifiedCols;
2275
2276 switch (wco->kind)
2277 {
2278 /*
2279 * For WITH CHECK OPTIONs coming from views, we might be
2280 * able to provide the details on the row, depending on
2281 * the permissions on the relation (that is, if the user
2282 * could view it directly anyway). For RLS violations, we
2283 * don't include the data since we don't know if the user
2284 * should be able to view the tuple as that depends on the
2285 * USING policy.
2286 */
2287 case WCO_VIEW_CHECK:
2288 /* See the comment in ExecConstraints(). */
2289 if (resultRelInfo->ri_RootResultRelInfo)
2290 {
2291 ResultRelInfo *rootrel = resultRelInfo->ri_RootResultRelInfo;
2292 TupleDesc old_tupdesc = RelationGetDescr(rel);
2293 AttrMap *map;
2294
2295 tupdesc = RelationGetDescr(rootrel->ri_RelationDesc);
2296 /* a reverse map */
2297 map = build_attrmap_by_name_if_req(old_tupdesc,
2298 tupdesc,
2299 false);
2300
2301 /*
2302 * Partition-specific slot's tupdesc can't be changed,
2303 * so allocate a new one.
2304 */
2305 if (map != NULL)
2306 slot = execute_attr_map_slot(map, slot,
2307 MakeTupleTableSlot(tupdesc, &TTSOpsVirtual));
2308
2309 modifiedCols = bms_union(ExecGetInsertedCols(rootrel, estate),
2310 ExecGetUpdatedCols(rootrel, estate));
2311 rel = rootrel->ri_RelationDesc;
2312 }
2313 else
2314 modifiedCols = bms_union(ExecGetInsertedCols(resultRelInfo, estate),
2315 ExecGetUpdatedCols(resultRelInfo, estate));
2316 val_desc = ExecBuildSlotValueDescription(RelationGetRelid(rel),
2317 slot,
2318 tupdesc,
2319 modifiedCols,
2320 64);
2321
2322 ereport(ERROR,
2323 (errcode(ERRCODE_WITH_CHECK_OPTION_VIOLATION),
2324 errmsg("new row violates check option for view \"%s\"",
2325 wco->relname),
2326 val_desc ? errdetail("Failing row contains %s.",
2327 val_desc) : 0));
2328 break;
2329 case WCO_RLS_INSERT_CHECK:
2330 case WCO_RLS_UPDATE_CHECK:
2331 if (wco->polname != NULL)
2332 ereport(ERROR,
2333 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2334 errmsg("new row violates row-level security policy \"%s\" for table \"%s\"",
2335 wco->polname, wco->relname)));
2336 else
2337 ereport(ERROR,
2338 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2339 errmsg("new row violates row-level security policy for table \"%s\"",
2340 wco->relname)));
2341 break;
2342 case WCO_RLS_MERGE_UPDATE_CHECK:
2343 case WCO_RLS_MERGE_DELETE_CHECK:
2344 if (wco->polname != NULL)
2345 ereport(ERROR,
2346 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2347 errmsg("target row violates row-level security policy \"%s\" (USING expression) for table \"%s\"",
2348 wco->polname, wco->relname)));
2349 else
2350 ereport(ERROR,
2351 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2352 errmsg("target row violates row-level security policy (USING expression) for table \"%s\"",
2353 wco->relname)));
2354 break;
2355 case WCO_RLS_CONFLICT_CHECK:
2356 if (wco->polname != NULL)
2357 ereport(ERROR,
2358 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2359 errmsg("new row violates row-level security policy \"%s\" (USING expression) for table \"%s\"",
2360 wco->polname, wco->relname)));
2361 else
2362 ereport(ERROR,
2363 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
2364 errmsg("new row violates row-level security policy (USING expression) for table \"%s\"",
2365 wco->relname)));
2366 break;
2367 default:
2368 elog(ERROR, "unrecognized WCO kind: %u", wco->kind);
2369 break;
2370 }
2371 }
2372 }
2373 }
2374
2375 /*
2376 * ExecBuildSlotValueDescription -- construct a string representing a tuple
2377 *
2378 * This is intentionally very similar to BuildIndexValueDescription, but
2379 * unlike that function, we truncate long field values (to at most maxfieldlen
2380 * bytes). That seems necessary here since heap field values could be very
2381 * long, whereas index entries typically aren't so wide.
2382 *
2383 * Also, unlike the case with index entries, we need to be prepared to ignore
2384 * dropped columns. We used to use the slot's tuple descriptor to decode the
2385 * data, but the slot's descriptor doesn't identify dropped columns, so we
2386 * now need to be passed the relation's descriptor.
2387 *
2388 * Note that, like BuildIndexValueDescription, if the user does not have
2389 * permission to view any of the columns involved, a NULL is returned. Unlike
2390 * BuildIndexValueDescription, if the user has access to view a subset of the
2391 * column involved, that subset will be returned with a key identifying which
2392 * columns they are.
2393 */
2394 char *
2395 ExecBuildSlotValueDescription(Oid reloid,
2396 TupleTableSlot *slot,
2397 TupleDesc tupdesc,
2398 Bitmapset *modifiedCols,
2399 int maxfieldlen)
2400 {
2401 StringInfoData buf;
2402 StringInfoData collist;
2403 bool write_comma = false;
2404 bool write_comma_collist = false;
2405 int i;
2406 AclResult aclresult;
2407 bool table_perm = false;
2408 bool any_perm = false;
2409
2410 /*
2411 * Check if RLS is enabled and should be active for the relation; if so,
2412 * then don't return anything. Otherwise, go through normal permission
2413 * checks.
2414 */
2415 if (check_enable_rls(reloid, InvalidOid, true) == RLS_ENABLED)
2416 return NULL;
2417
2418 initStringInfo(&buf);
2419
2420 appendStringInfoChar(&buf, '(');
2421
2422 /*
2423 * Check if the user has permissions to see the row. Table-level SELECT
2424 * allows access to all columns. If the user does not have table-level
2425 * SELECT then we check each column and include those the user has SELECT
2426 * rights on. Additionally, we always include columns the user provided
2427 * data for.
2428 */
2429 aclresult = pg_class_aclcheck(reloid, GetUserId(), ACL_SELECT);
2430 if (aclresult != ACLCHECK_OK)
2431 {
2432 /* Set up the buffer for the column list */
2433 initStringInfo(&collist);
2434 appendStringInfoChar(&collist, '(');
2435 }
2436 else
2437 table_perm = any_perm = true;
2438
2439 /* Make sure the tuple is fully deconstructed */
2440 slot_getallattrs(slot);
2441
2442 for (i = 0; i < tupdesc->natts; i++)
2443 {
2444 bool column_perm = false;
2445 char *val;
2446 int vallen;
2447 Form_pg_attribute att = TupleDescAttr(tupdesc, i);
2448
2449 /* ignore dropped columns */
2450 if (att->attisdropped)
2451 continue;
2452
2453 if (!table_perm)
2454 {
2455 /*
2456 * No table-level SELECT, so need to make sure they either have
2457 * SELECT rights on the column or that they have provided the data
2458 * for the column. If not, omit this column from the error
2459 * message.
2460 */
2461 aclresult = pg_attribute_aclcheck(reloid, att->attnum,
2462 GetUserId(), ACL_SELECT);
2463 if (bms_is_member(att->attnum - FirstLowInvalidHeapAttributeNumber,
2464 modifiedCols) || aclresult == ACLCHECK_OK)
2465 {
2466 column_perm = any_perm = true;
2467
2468 if (write_comma_collist)
2469 appendStringInfoString(&collist, ", ");
2470 else
2471 write_comma_collist = true;
2472
2473 appendStringInfoString(&collist, NameStr(att->attname));
2474 }
2475 }
2476
2477 if (table_perm || column_perm)
2478 {
2479 if (att->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
2480 val = "virtual";
2481 else if (slot->tts_isnull[i])
2482 val = "null";
2483 else
2484 {
2485 Oid foutoid;
2486 bool typisvarlena;
2487
2488 getTypeOutputInfo(att->atttypid,
2489 &foutoid, &typisvarlena);
2490 val = OidOutputFunctionCall(foutoid, slot->tts_values[i]);
2491 }
2492
2493 if (write_comma)
2494 appendStringInfoString(&buf, ", ");
2495 else
2496 write_comma = true;
2497
2498 /* truncate if needed */
2499 vallen = strlen(val);
2500 if (vallen <= maxfieldlen)
2501 appendBinaryStringInfo(&buf, val, vallen);
2502 else
2503 {
2504 vallen = pg_mbcliplen(val, vallen, maxfieldlen);
2505 appendBinaryStringInfo(&buf, val, vallen);
2506 appendStringInfoString(&buf, "...");
2507 }
2508 }
2509 }
2510
2511 /* If we end up with zero columns being returned, then return NULL. */
2512 if (!any_perm)
2513 return NULL;
2514
2515 appendStringInfoChar(&buf, ')');
2516
2517 if (!table_perm)
2518 {
2519 appendStringInfoString(&collist, ") = ");
2520 appendBinaryStringInfo(&collist, buf.data, buf.len);
2521
2522 return collist.data;
2523 }
2524
2525 return buf.data;
2526 }
2527
2528
2529 /*
2530 * ExecUpdateLockMode -- find the appropriate UPDATE tuple lock mode for a
2531 * given ResultRelInfo
2532 */
2533 LockTupleMode
2534 ExecUpdateLockMode(EState *estate, ResultRelInfo *relinfo)
2535 {
2536 Bitmapset *keyCols;
2537 Bitmapset *updatedCols;
2538
2539 /*
2540 * Compute lock mode to use. If columns that are part of the key have not
2541 * been modified, then we can use a weaker lock, allowing for better
2542 * concurrency.
2543 */
2544 updatedCols = ExecGetAllUpdatedCols(relinfo, estate);
2545 keyCols = RelationGetIndexAttrBitmap(relinfo->ri_RelationDesc,
2546 INDEX_ATTR_BITMAP_KEY);
2547
2548 if (bms_overlap(keyCols, updatedCols))
2549 return LockTupleExclusive;
2550
2551 return LockTupleNoKeyExclusive;
2552 }
2553
2554 /*
2555 * ExecFindRowMark -- find the ExecRowMark struct for given rangetable index
2556 *
2557 * If no such struct, either return NULL or throw error depending on missing_ok
2558 */
2559 ExecRowMark *
2560 ExecFindRowMark(EState *estate, Index rti, bool missing_ok)
2561 {
2562 if (rti > 0 && rti <= estate->es_range_table_size &&
2563 estate->es_rowmarks != NULL)
2564 {
2565 ExecRowMark *erm = estate->es_rowmarks[rti - 1];
2566
2567 if (erm)
2568 return erm;
2569 }
2570 if (!missing_ok)
2571 elog(ERROR, "failed to find ExecRowMark for rangetable index %u", rti);
2572 return NULL;
2573 }
2574
2575 /*
2576 * ExecBuildAuxRowMark -- create an ExecAuxRowMark struct
2577 *
2578 * Inputs are the underlying ExecRowMark struct and the targetlist of the
2579 * input plan node (not planstate node!). We need the latter to find out
2580 * the column numbers of the resjunk columns.
2581 */
2582 ExecAuxRowMark *
2583 ExecBuildAuxRowMark(ExecRowMark *erm, List *targetlist)
2584 {
2585 ExecAuxRowMark *aerm = (ExecAuxRowMark *) palloc0(sizeof(ExecAuxRowMark));
2586 char resname[32];
2587
2588 aerm->rowmark = erm;
2589
2590 /* Look up the resjunk columns associated with this rowmark */
2591 if (erm->markType != ROW_MARK_COPY)
2592 {
2593 /* need ctid for all methods other than COPY */
2594 snprintf(resname, sizeof(resname), "ctid%u", erm->rowmarkId);
2595 aerm->ctidAttNo = ExecFindJunkAttributeInTlist(targetlist,
2596 resname);
2597 if (!AttributeNumberIsValid(aerm->ctidAttNo))
2598 elog(ERROR, "could not find junk %s column", resname);
2599 }
2600 else
2601 {
2602 /* need wholerow if COPY */
2603 snprintf(resname, sizeof(resname), "wholerow%u", erm->rowmarkId);
2604 aerm->wholeAttNo = ExecFindJunkAttributeInTlist(targetlist,
2605 resname);
2606 if (!AttributeNumberIsValid(aerm->wholeAttNo))
2607 elog(ERROR, "could not find junk %s column", resname);
2608 }
2609
2610 /* if child rel, need tableoid */
2611 if (erm->rti != erm->prti)
2612 {
2613 snprintf(resname, sizeof(resname), "tableoid%u", erm->rowmarkId);
2614 aerm->toidAttNo = ExecFindJunkAttributeInTlist(targetlist,
2615 resname);
2616 if (!AttributeNumberIsValid(aerm->toidAttNo))
2617 elog(ERROR, "could not find junk %s column", resname);
2618 }
2619
2620 return aerm;
2621 }
2622
2623
2624 /*
2625 * EvalPlanQual logic --- recheck modified tuple(s) to see if we want to
2626 * process the updated version under READ COMMITTED rules.
2627 *
2628 * See backend/executor/README for some info about how this works.
2629 */
2630
2631
2632 /*
2633 * Check the updated version of a tuple to see if we want to process it under
2634 * READ COMMITTED rules.
2635 *
2636 * epqstate - state for EvalPlanQual rechecking
2637 * relation - table containing tuple
2638 * rti - rangetable index of table containing tuple
2639 * inputslot - tuple for processing - this can be the slot from
2640 * EvalPlanQualSlot() for this rel, for increased efficiency.
2641 *
2642 * This tests whether the tuple in inputslot still matches the relevant
2643 * quals. For that result to be useful, typically the input tuple has to be
2644 * last row version (otherwise the result isn't particularly useful) and
2645 * locked (otherwise the result might be out of date). That's typically
2646 * achieved by using table_tuple_lock() with the
2647 * TUPLE_LOCK_FLAG_FIND_LAST_VERSION flag.
2648 *
2649 * Returns a slot containing the new candidate update/delete tuple, or
2650 * NULL if we determine we shouldn't process the row.
2651 */
2652 TupleTableSlot *
2653 EvalPlanQual(EPQState *epqstate, Relation relation,
2654 Index rti, TupleTableSlot *inputslot)
2655 {
2656 TupleTableSlot *slot;
2657 TupleTableSlot *testslot;
2658
2659 Assert(rti > 0);
2660
2661 /*
2662 * Need to run a recheck subquery. Initialize or reinitialize EPQ state.
2663 */
2664 EvalPlanQualBegin(epqstate);
2665
2666 /*
2667 * Callers will often use the EvalPlanQualSlot to store the tuple to avoid
2668 * an unnecessary copy.
2669 */
2670 testslot = EvalPlanQualSlot(epqstate, relation, rti);
2671 if (testslot != inputslot)
2672 ExecCopySlot(testslot, inputslot);
2673
2674 /*
2675 * Mark that an EPQ tuple is available for this relation. (If there is
2676 * more than one result relation, the others remain marked as having no
2677 * tuple available.)
2678 */
2679 epqstate->relsubs_done[rti - 1] = false;
2680 epqstate->relsubs_blocked[rti - 1] = false;
2681
2682 /*
2683 * Run the EPQ query. We assume it will return at most one tuple.
2684 */
2685 slot = EvalPlanQualNext(epqstate);
2686
2687 /*
2688 * If we got a tuple, force the slot to materialize the tuple so that it
2689 * is not dependent on any local state in the EPQ query (in particular,
2690 * it's highly likely that the slot contains references to any pass-by-ref
2691 * datums that may be present in copyTuple). As with the next step, this
2692 * is to guard against early re-use of the EPQ query.
2693 */
2694 if (!TupIsNull(slot))
2695 ExecMaterializeSlot(slot);
2696
2697 /*
2698 * Clear out the test tuple, and mark that no tuple is available here.
2699 * This is needed in case the EPQ state is re-used to test a tuple for a
2700 * different target relation.
2701 */
2702 ExecClearTuple(testslot);
2703 epqstate->relsubs_blocked[rti - 1] = true;
2704
2705 return slot;
2706 }
2707
2708 /*
2709 * EvalPlanQualInit -- initialize during creation of a plan state node
2710 * that might need to invoke EPQ processing.
2711 *
2712 * If the caller intends to use EvalPlanQual(), resultRelations should be
2713 * a list of RT indexes of potential target relations for EvalPlanQual(),
2714 * and we will arrange that the other listed relations don't return any
2715 * tuple during an EvalPlanQual() call. Otherwise resultRelations
2716 * should be NIL.
2717 *
2718 * Note: subplan/auxrowmarks can be NULL/NIL if they will be set later
2719 * with EvalPlanQualSetPlan.
2720 */
2721 void
2722 EvalPlanQualInit(EPQState *epqstate, EState *parentestate,
2723 Plan *subplan, List *auxrowmarks,
2724 int epqParam, List *resultRelations)
2725 {
2726 Index rtsize = parentestate->es_range_table_size;
2727
2728 /* initialize data not changing over EPQState's lifetime */
2729 epqstate->parentestate = parentestate;
2730 epqstate->epqParam = epqParam;
2731 epqstate->resultRelations = resultRelations;
2732
2733 /*
2734 * Allocate space to reference a slot for each potential rti - do so now
2735 * rather than in EvalPlanQualBegin(), as done for other dynamically
2736 * allocated resources, so EvalPlanQualSlot() can be used to hold tuples
2737 * that *may* need EPQ later, without forcing the overhead of
2738 * EvalPlanQualBegin().
2739 */
2740 epqstate->tuple_table = NIL;
2741 epqstate->relsubs_slot = (TupleTableSlot **)
2742 palloc0(rtsize * sizeof(TupleTableSlot *));
2743
2744 /* ... and remember data that EvalPlanQualBegin will need */
2745 epqstate->plan = subplan;
2746 epqstate->arowMarks = auxrowmarks;
2747
2748 /* ... and mark the EPQ state inactive */
2749 epqstate->origslot = NULL;
2750 epqstate->recheckestate = NULL;
2751 epqstate->recheckplanstate = NULL;
2752 epqstate->relsubs_rowmark = NULL;
2753 epqstate->relsubs_done = NULL;
2754 epqstate->relsubs_blocked = NULL;
2755 }
2756
2757 /*
2758 * EvalPlanQualSetPlan -- set or change subplan of an EPQState.
2759 *
2760 * We used to need this so that ModifyTable could deal with multiple subplans.
2761 * It could now be refactored out of existence.
2762 */
2763 void
2764 EvalPlanQualSetPlan(EPQState *epqstate, Plan *subplan, List *auxrowmarks)
2765 {
2766 /* If we have a live EPQ query, shut it down */
2767 EvalPlanQualEnd(epqstate);
2768 /* And set/change the plan pointer */
2769 epqstate->plan = subplan;
2770 /* The rowmarks depend on the plan, too */
2771 epqstate->arowMarks = auxrowmarks;
2772 }
2773
2774 /*
2775 * Return, and create if necessary, a slot for an EPQ test tuple.
2776 *
2777 * Note this only requires EvalPlanQualInit() to have been called,
2778 * EvalPlanQualBegin() is not necessary.
2779 */
2780 TupleTableSlot *
2781 EvalPlanQualSlot(EPQState *epqstate,
2782 Relation relation, Index rti)
2783 {
2784 TupleTableSlot **slot;
2785
2786 Assert(relation);
2787 Assert(rti > 0 && rti <= epqstate->parentestate->es_range_table_size);
2788 slot = &epqstate->relsubs_slot[rti - 1];
2789
2790 if (*slot == NULL)
2791 {
2792 MemoryContext oldcontext;
2793
2794 oldcontext = MemoryContextSwitchTo(epqstate->parentestate->es_query_cxt);
2795 *slot = table_slot_create(relation, &epqstate->tuple_table);
2796 MemoryContextSwitchTo(oldcontext);
2797 }
2798
2799 return *slot;
2800 }
2801
2802 /*
2803 * Fetch the current row value for a non-locked relation, identified by rti,
2804 * that needs to be scanned by an EvalPlanQual operation. origslot must have
2805 * been set to contain the current result row (top-level row) that we need to
2806 * recheck. Returns true if a substitution tuple was found, false if not.
2807 */
2808 bool
2809 EvalPlanQualFetchRowMark(EPQState *epqstate, Index rti, TupleTableSlot *slot)
2810 {
2811 ExecAuxRowMark *earm = epqstate->relsubs_rowmark[rti - 1];
2812 ExecRowMark *erm;
2813 Datum datum;
2814 bool isNull;
2815
2816 Assert(earm != NULL);
2817 Assert(epqstate->origslot != NULL);
2818
2819 erm = earm->rowmark;
2820
2821 if (RowMarkRequiresRowShareLock(erm->markType))
2822 elog(ERROR, "EvalPlanQual doesn't support locking rowmarks");
2823
2824 /* if child rel, must check whether it produced this row */
2825 if (erm->rti != erm->prti)
2826 {
2827 Oid tableoid;
2828
2829 datum = ExecGetJunkAttribute(epqstate->origslot,
2830 earm->toidAttNo,
2831 &isNull);
2832 /* non-locked rels could be on the inside of outer joins */
2833 if (isNull)
2834 return false;
2835
2836 tableoid = DatumGetObjectId(datum);
2837
2838 Assert(OidIsValid(erm->relid));
2839 if (tableoid != erm->relid)
2840 {
2841 /* this child is inactive right now */
2842 return false;
2843 }
2844 }
2845
2846 if (erm->markType == ROW_MARK_REFERENCE)
2847 {
2848 Assert(erm->relation != NULL);
2849
2850 /* fetch the tuple's ctid */
2851 datum = ExecGetJunkAttribute(epqstate->origslot,
2852 earm->ctidAttNo,
2853 &isNull);
2854 /* non-locked rels could be on the inside of outer joins */
2855 if (isNull)
2856 return false;
2857
2858 /* fetch requests on foreign tables must be passed to their FDW */
2859 if (erm->relation->rd_rel->relkind == RELKIND_FOREIGN_TABLE)
2860 {
2861 FdwRoutine *fdwroutine;
2862 bool updated = false;
2863
2864 fdwroutine = GetFdwRoutineForRelation(erm->relation, false);
2865 /* this should have been checked already, but let's be safe */
2866 if (fdwroutine->RefetchForeignRow == NULL)
2867 ereport(ERROR,
2868 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2869 errmsg("cannot lock rows in foreign table \"%s\"",
2870 RelationGetRelationName(erm->relation))));
2871
2872 fdwroutine->RefetchForeignRow(epqstate->recheckestate,
2873 erm,
2874 datum,
2875 slot,
2876 &updated);
2877 if (TupIsNull(slot))
2878 elog(ERROR, "failed to fetch tuple for EvalPlanQual recheck");
2879
2880 /*
2881 * Ideally we'd insist on updated == false here, but that assumes
2882 * that FDWs can track that exactly, which they might not be able
2883 * to. So just ignore the flag.
2884 */
2885 return true;
2886 }
2887 else
2888 {
2889 /* ordinary table, fetch the tuple */
2890 if (!table_tuple_fetch_row_version(erm->relation,
2891 (ItemPointer) DatumGetPointer(datum),
2892 SnapshotAny, slot))
2893 elog(ERROR, "failed to fetch tuple for EvalPlanQual recheck");
2894 return true;
2895 }
2896 }
2897 else
2898 {
2899 Assert(erm->markType == ROW_MARK_COPY);
2900
2901 /* fetch the whole-row Var for the relation */
2902 datum = ExecGetJunkAttribute(epqstate->origslot,
2903 earm->wholeAttNo,
2904 &isNull);
2905 /* non-locked rels could be on the inside of outer joins */
2906 if (isNull)
2907 return false;
2908
2909 ExecStoreHeapTupleDatum(datum, slot);
2910 return true;
2911 }
2912 }
2913
2914 /*
2915 * Fetch the next row (if any) from EvalPlanQual testing
2916 *
2917 * (In practice, there should never be more than one row...)
2918 */
2919 TupleTableSlot *
2920 EvalPlanQualNext(EPQState *epqstate)
2921 {
2922 MemoryContext oldcontext;
2923 TupleTableSlot *slot;
2924
2925 oldcontext = MemoryContextSwitchTo(epqstate->recheckestate->es_query_cxt);
2926 slot = ExecProcNode(epqstate->recheckplanstate);
2927 MemoryContextSwitchTo(oldcontext);
2928
2929 return slot;
2930 }
2931
2932 /*
2933 * Initialize or reset an EvalPlanQual state tree
2934 */
2935 void
2936 EvalPlanQualBegin(EPQState *epqstate)
2937 {
2938 EState *parentestate = epqstate->parentestate;
2939 EState *recheckestate = epqstate->recheckestate;
2940
2941 if (recheckestate == NULL)
2942 {
2943 /* First time through, so create a child EState */
2944 EvalPlanQualStart(epqstate, epqstate->plan);
2945 }
2946 else
2947 {
2948 /*
2949 * We already have a suitable child EPQ tree, so just reset it.
2950 */
2951 Index rtsize = parentestate->es_range_table_size;
2952 PlanState *rcplanstate = epqstate->recheckplanstate;
2953
2954 /*
2955 * Reset the relsubs_done[] flags to equal relsubs_blocked[], so that
2956 * the EPQ run will never attempt to fetch tuples from blocked target
2957 * relations.
2958 */
2959 memcpy(epqstate->relsubs_done, epqstate->relsubs_blocked,
2960 rtsize * sizeof(bool));
2961
2962 /* Recopy current values of parent parameters */
2963 if (parentestate->es_plannedstmt->paramExecTypes != NIL)
2964 {
2965 int i;
2966
2967 /*
2968 * Force evaluation of any InitPlan outputs that could be needed
2969 * by the subplan, just in case they got reset since
2970 * EvalPlanQualStart (see comments therein).
2971 */
2972 ExecSetParamPlanMulti(rcplanstate->plan->extParam,
2973 GetPerTupleExprContext(parentestate));
2974
2975 i = list_length(parentestate->es_plannedstmt->paramExecTypes);
2976
2977 while (--i >= 0)
2978 {
2979 /* copy value if any, but not execPlan link */
2980 recheckestate->es_param_exec_vals[i].value =
2981 parentestate->es_param_exec_vals[i].value;
2982 recheckestate->es_param_exec_vals[i].isnull =
2983 parentestate->es_param_exec_vals[i].isnull;
2984 }
2985 }
2986
2987 /*
2988 * Mark child plan tree as needing rescan at all scan nodes. The
2989 * first ExecProcNode will take care of actually doing the rescan.
2990 */
2991 rcplanstate->chgParam = bms_add_member(rcplanstate->chgParam,
2992 epqstate->epqParam);
2993 }
2994 }
2995
2996 /*
2997 * Start execution of an EvalPlanQual plan tree.
2998 *
2999 * This is a cut-down version of ExecutorStart(): we copy some state from
3000 * the top-level estate rather than initializing it fresh.
3001 */
3002 static void
3003 EvalPlanQualStart(EPQState *epqstate, Plan *planTree)
3004 {
3005 EState *parentestate = epqstate->parentestate;
3006 Index rtsize = parentestate->es_range_table_size;
3007 EState *rcestate;
3008 MemoryContext oldcontext;
3009 ListCell *l;
3010
3011 epqstate->recheckestate = rcestate = CreateExecutorState();
3012
3013 oldcontext = MemoryContextSwitchTo(rcestate->es_query_cxt);
3014
3015 /* signal that this is an EState for executing EPQ */
3016 rcestate->es_epq_active = epqstate;
3017
3018 /*
3019 * Child EPQ EStates share the parent's copy of unchanging state such as
3020 * the snapshot, rangetable, and external Param info. They need their own
3021 * copies of local state, including a tuple table, es_param_exec_vals,
3022 * result-rel info, etc.
3023 */
3024 rcestate->es_direction = ForwardScanDirection;
3025 rcestate->es_snapshot = parentestate->es_snapshot;
3026 rcestate->es_crosscheck_snapshot = parentestate->es_crosscheck_snapshot;
3027 rcestate->es_range_table = parentestate->es_range_table;
3028 rcestate->es_range_table_size = parentestate->es_range_table_size;
3029 rcestate->es_relations = parentestate->es_relations;
3030 rcestate->es_rowmarks = parentestate->es_rowmarks;
3031 rcestate->es_rteperminfos = parentestate->es_rteperminfos;
3032 rcestate->es_plannedstmt = parentestate->es_plannedstmt;
3033 rcestate->es_junkFilter = parentestate->es_junkFilter;
3034 rcestate->es_output_cid = parentestate->es_output_cid;
3035 rcestate->es_queryEnv = parentestate->es_queryEnv;
3036
3037 /*
3038 * ResultRelInfos needed by subplans are initialized from scratch when the
3039 * subplans themselves are initialized.
3040 */
3041 rcestate->es_result_relations = NULL;
3042 /* es_trig_target_relations must NOT be copied */
3043 rcestate->es_top_eflags = parentestate->es_top_eflags;
3044 rcestate->es_instrument = parentestate->es_instrument;
3045 /* es_auxmodifytables must NOT be copied */
3046
3047 /*
3048 * The external param list is simply shared from parent. The internal
3049 * param workspace has to be local state, but we copy the initial values
3050 * from the parent, so as to have access to any param values that were
3051 * already set from other parts of the parent's plan tree.
3052 */
3053 rcestate->es_param_list_info = parentestate->es_param_list_info;
3054 if (parentestate->es_plannedstmt->paramExecTypes != NIL)
3055 {
3056 int i;
3057
3058 /*
3059 * Force evaluation of any InitPlan outputs that could be needed by
3060 * the subplan. (With more complexity, maybe we could postpone this
3061 * till the subplan actually demands them, but it doesn't seem worth
3062 * the trouble; this is a corner case already, since usually the
3063 * InitPlans would have been evaluated before reaching EvalPlanQual.)
3064 *
3065 * This will not touch output params of InitPlans that occur somewhere
3066 * within the subplan tree, only those that are attached to the
3067 * ModifyTable node or above it and are referenced within the subplan.
3068 * That's OK though, because the planner would only attach such
3069 * InitPlans to a lower-level SubqueryScan node, and EPQ execution
3070 * will not descend into a SubqueryScan.
3071 *
3072 * The EState's per-output-tuple econtext is sufficiently short-lived
3073 * for this, since it should get reset before there is any chance of
3074 * doing EvalPlanQual again.
3075 */
3076 ExecSetParamPlanMulti(planTree->extParam,
3077 GetPerTupleExprContext(parentestate));
3078
3079 /* now make the internal param workspace ... */
3080 i = list_length(parentestate->es_plannedstmt->paramExecTypes);
3081 rcestate->es_param_exec_vals = (ParamExecData *)
3082 palloc0(i * sizeof(ParamExecData));
3083 /* ... and copy down all values, whether really needed or not */
3084 while (--i >= 0)
3085 {
3086 /* copy value if any, but not execPlan link */
3087 rcestate->es_param_exec_vals[i].value =
3088 parentestate->es_param_exec_vals[i].value;
3089 rcestate->es_param_exec_vals[i].isnull =
3090 parentestate->es_param_exec_vals[i].isnull;
3091 }
3092 }
3093
3094 /*
3095 * Copy es_unpruned_relids so that pruned relations are ignored by
3096 * ExecInitLockRows() and ExecInitModifyTable() when initializing the plan
3097 * trees below.
3098 */
3099 rcestate->es_unpruned_relids = parentestate->es_unpruned_relids;
3100
3101 /*
3102 * Also make the PartitionPruneInfo and the results of pruning available.
3103 * These need to match exactly so that we initialize all the same Append
3104 * and MergeAppend subplans as the parent did.
3105 */
3106 rcestate->es_part_prune_infos = parentestate->es_part_prune_infos;
3107 rcestate->es_part_prune_states = parentestate->es_part_prune_states;
3108 rcestate->es_part_prune_results = parentestate->es_part_prune_results;
3109
3110 /* We'll also borrow the es_partition_directory from the parent state */
3111 rcestate->es_partition_directory = parentestate->es_partition_directory;
3112
3113 /*
3114 * Initialize private state information for each SubPlan. We must do this
3115 * before running ExecInitNode on the main query tree, since
3116 * ExecInitSubPlan expects to be able to find these entries. Some of the
3117 * SubPlans might not be used in the part of the plan tree we intend to
3118 * run, but since it's not easy to tell which, we just initialize them
3119 * all.
3120 */
3121 Assert(rcestate->es_subplanstates == NIL);
3122 foreach(l, parentestate->es_plannedstmt->subplans)
3123 {
3124 Plan *subplan = (Plan *) lfirst(l);
3125 PlanState *subplanstate;
3126
3127 subplanstate = ExecInitNode(subplan, rcestate, 0);
3128 rcestate->es_subplanstates = lappend(rcestate->es_subplanstates,
3129 subplanstate);
3130 }
3131
3132 /*
3133 * Build an RTI indexed array of rowmarks, so that
3134 * EvalPlanQualFetchRowMark() can efficiently access the to be fetched
3135 * rowmark.
3136 */
3137 epqstate->relsubs_rowmark = (ExecAuxRowMark **)
3138 palloc0(rtsize * sizeof(ExecAuxRowMark *));
3139 foreach(l, epqstate->arowMarks)
3140 {
3141 ExecAuxRowMark *earm = (ExecAuxRowMark *) lfirst(l);
3142
3143 epqstate->relsubs_rowmark[earm->rowmark->rti - 1] = earm;
3144 }
3145
3146 /*
3147 * Initialize per-relation EPQ tuple states. Result relations, if any,
3148 * get marked as blocked; others as not-fetched.
3149 */
3150 epqstate->relsubs_done = palloc_array(bool, rtsize);
3151 epqstate->relsubs_blocked = palloc0_array(bool, rtsize);
3152
3153 foreach(l, epqstate->resultRelations)
3154 {
3155 int rtindex = lfirst_int(l);
3156
3157 Assert(rtindex > 0 && rtindex <= rtsize);
3158 epqstate->relsubs_blocked[rtindex - 1] = true;
3159 }
3160
3161 memcpy(epqstate->relsubs_done, epqstate->relsubs_blocked,
3162 rtsize * sizeof(bool));
3163
3164 /*
3165 * Initialize the private state information for all the nodes in the part
3166 * of the plan tree we need to run. This opens files, allocates storage
3167 * and leaves us ready to start processing tuples.
3168 */
3169 epqstate->recheckplanstate = ExecInitNode(planTree, rcestate, 0);
3170
3171 MemoryContextSwitchTo(oldcontext);
3172 }
3173
3174 /*
3175 * EvalPlanQualEnd -- shut down at termination of parent plan state node,
3176 * or if we are done with the current EPQ child.
3177 *
3178 * This is a cut-down version of ExecutorEnd(); basically we want to do most
3179 * of the normal cleanup, but *not* close result relations (which we are
3180 * just sharing from the outer query). We do, however, have to close any
3181 * result and trigger target relations that got opened, since those are not
3182 * shared. (There probably shouldn't be any of the latter, but just in
3183 * case...)
3184 */
3185 void
3186 EvalPlanQualEnd(EPQState *epqstate)
3187 {
3188 EState *estate = epqstate->recheckestate;
3189 Index rtsize;
3190 MemoryContext oldcontext;
3191 ListCell *l;
3192
3193 rtsize = epqstate->parentestate->es_range_table_size;
3194
3195 /*
3196 * We may have a tuple table, even if EPQ wasn't started, because we allow
3197 * use of EvalPlanQualSlot() without calling EvalPlanQualBegin().
3198 */
3199 if (epqstate->tuple_table != NIL)
3200 {
3201 memset(epqstate->relsubs_slot, 0,
3202 rtsize * sizeof(TupleTableSlot *));
3203 ExecResetTupleTable(epqstate->tuple_table, true);
3204 epqstate->tuple_table = NIL;
3205 }
3206
3207 /* EPQ wasn't started, nothing further to do */
3208 if (estate == NULL)
3209 return;
3210
3211 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
3212
3213 ExecEndNode(epqstate->recheckplanstate);
3214
3215 foreach(l, estate->es_subplanstates)
3216 {
3217 PlanState *subplanstate = (PlanState *) lfirst(l);
3218
3219 ExecEndNode(subplanstate);
3220 }
3221
3222 /* throw away the per-estate tuple table, some node may have used it */
3223 ExecResetTupleTable(estate->es_tupleTable, false);
3224
3225 /* Close any result and trigger target relations attached to this EState */
3226 ExecCloseResultRelations(estate);
3227
3228 MemoryContextSwitchTo(oldcontext);
3229
3230 /*
3231 * NULLify the partition directory before freeing the executor state.
3232 * Since EvalPlanQualStart() just borrowed the parent EState's directory,
3233 * we'd better leave it up to the parent to delete it.
3234 */
3235 estate->es_partition_directory = NULL;
3236
3237 FreeExecutorState(estate);
3238
3239 /* Mark EPQState idle */
3240 epqstate->origslot = NULL;
3241 epqstate->recheckestate = NULL;
3242 epqstate->recheckplanstate = NULL;
3243 epqstate->relsubs_rowmark = NULL;
3244 epqstate->relsubs_done = NULL;
3245 epqstate->relsubs_blocked = NULL;
3246 }