1/*-------------------------------------------------------------------------
2 *
3 * nodeModifyTable.c
4 * routines to handle ModifyTable nodes.
5 *
6 * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
8 *
9 *
10 * IDENTIFICATION
11 * src/backend/executor/nodeModifyTable.c
12 *
13 *-------------------------------------------------------------------------
14 */
15/* INTERFACE ROUTINES
16 * ExecInitModifyTable - initialize the ModifyTable node
17 * ExecModifyTable - retrieve the next tuple from the node
18 * ExecEndModifyTable - shut down the ModifyTable node
19 * ExecReScanModifyTable - rescan the ModifyTable node
20 *
21 * NOTES
22 * Each ModifyTable node contains a list of one or more subplans,
23 * much like an Append node. There is one subplan per result relation.
24 * The key reason for this is that in an inherited UPDATE command, each
25 * result relation could have a different schema (more or different
26 * columns) requiring a different plan tree to produce it. In an
27 * inherited DELETE, all the subplans should produce the same output
28 * rowtype, but we might still find that different plans are appropriate
29 * for different child relations.
30 *
31 * If the query specifies RETURNING, then the ModifyTable returns a
32 * RETURNING tuple after completing each row insert, update, or delete.
33 * It must be called again to continue the operation. Without RETURNING,
34 * we just loop within the node until all the work is done, then
35 * return NULL. This avoids useless call/return overhead.
36 */
37
38#include "postgres.h"
39
40#include "access/heapam.h"
41#include "access/htup_details.h"
42#include "access/tableam.h"
43#include "access/xact.h"
44#include "catalog/catalog.h"
45#include "commands/trigger.h"
46#include "executor/execPartition.h"
47#include "executor/executor.h"
48#include "executor/nodeModifyTable.h"
49#include "foreign/fdwapi.h"
50#include "miscadmin.h"
51#include "nodes/nodeFuncs.h"
52#include "rewrite/rewriteHandler.h"
53#include "storage/bufmgr.h"
54#include "storage/lmgr.h"
55#include "utils/builtins.h"
56#include "utils/datum.h"
57#include "utils/memutils.h"
58#include "utils/rel.h"
59
60
61static bool ExecOnConflictUpdate(ModifyTableState *mtstate,
62 ResultRelInfo *resultRelInfo,
63 ItemPointer conflictTid,
64 TupleTableSlot *planSlot,
65 TupleTableSlot *excludedSlot,
66 EState *estate,
67 bool canSetTag,
68 TupleTableSlot **returning);
69static TupleTableSlot *ExecPrepareTupleRouting(ModifyTableState *mtstate,
70 EState *estate,
71 PartitionTupleRouting *proute,
72 ResultRelInfo *targetRelInfo,
73 TupleTableSlot *slot);
74static ResultRelInfo *getTargetResultRelInfo(ModifyTableState *node);
75static void ExecSetupChildParentMapForSubplan(ModifyTableState *mtstate);
76static TupleConversionMap *tupconv_map_for_subplan(ModifyTableState *node,
77 int whichplan);
78
79/*
80 * Verify that the tuples to be produced by INSERT or UPDATE match the
81 * target relation's rowtype
82 *
83 * We do this to guard against stale plans. If plan invalidation is
84 * functioning properly then we should never get a failure here, but better
85 * safe than sorry. Note that this is called after we have obtained lock
86 * on the target rel, so the rowtype can't change underneath us.
87 *
88 * The plan output is represented by its targetlist, because that makes
89 * handling the dropped-column case easier.
90 */
91static void
92ExecCheckPlanOutput(Relation resultRel, List *targetList)
93{
94 TupleDesc resultDesc = RelationGetDescr(resultRel);
95 int attno = 0;
96 ListCell *lc;
97
98 foreach(lc, targetList)
99 {
100 TargetEntry *tle = (TargetEntry *) lfirst(lc);
101 Form_pg_attribute attr;
102
103 if (tle->resjunk)
104 continue; /* ignore junk tlist items */
105
106 if (attno >= resultDesc->natts)
107 ereport(ERROR,
108 (errcode(ERRCODE_DATATYPE_MISMATCH),
109 errmsg("table row type and query-specified row type do not match"),
110 errdetail("Query has too many columns.")));
111 attr = TupleDescAttr(resultDesc, attno);
112 attno++;
113
114 if (!attr->attisdropped)
115 {
116 /* Normal case: demand type match */
117 if (exprType((Node *) tle->expr) != attr->atttypid)
118 ereport(ERROR,
119 (errcode(ERRCODE_DATATYPE_MISMATCH),
120 errmsg("table row type and query-specified row type do not match"),
121 errdetail("Table has type %s at ordinal position %d, but query expects %s.",
122 format_type_be(attr->atttypid),
123 attno,
124 format_type_be(exprType((Node *) tle->expr)))));
125 }
126 else
127 {
128 /*
129 * For a dropped column, we can't check atttypid (it's likely 0).
130 * In any case the planner has most likely inserted an INT4 null.
131 * What we insist on is just *some* NULL constant.
132 */
133 if (!IsA(tle->expr, Const) ||
134 !((Const *) tle->expr)->constisnull)
135 ereport(ERROR,
136 (errcode(ERRCODE_DATATYPE_MISMATCH),
137 errmsg("table row type and query-specified row type do not match"),
138 errdetail("Query provides a value for a dropped column at ordinal position %d.",
139 attno)));
140 }
141 }
142 if (attno != resultDesc->natts)
143 ereport(ERROR,
144 (errcode(ERRCODE_DATATYPE_MISMATCH),
145 errmsg("table row type and query-specified row type do not match"),
146 errdetail("Query has too few columns.")));
147}
148
149/*
150 * ExecProcessReturning --- evaluate a RETURNING list
151 *
152 * resultRelInfo: current result rel
153 * tupleSlot: slot holding tuple actually inserted/updated/deleted
154 * planSlot: slot holding tuple returned by top subplan node
155 *
156 * Note: If tupleSlot is NULL, the FDW should have already provided econtext's
157 * scan tuple.
158 *
159 * Returns a slot holding the result tuple
160 */
161static TupleTableSlot *
162ExecProcessReturning(ResultRelInfo *resultRelInfo,
163 TupleTableSlot *tupleSlot,
164 TupleTableSlot *planSlot)
165{
166 ProjectionInfo *projectReturning = resultRelInfo->ri_projectReturning;
167 ExprContext *econtext = projectReturning->pi_exprContext;
168
169 /* Make tuple and any needed join variables available to ExecProject */
170 if (tupleSlot)
171 econtext->ecxt_scantuple = tupleSlot;
172 econtext->ecxt_outertuple = planSlot;
173
174 /*
175 * RETURNING expressions might reference the tableoid column, so
176 * reinitialize tts_tableOid before evaluating them.
177 */
178 econtext->ecxt_scantuple->tts_tableOid =
179 RelationGetRelid(resultRelInfo->ri_RelationDesc);
180
181 /* Compute the RETURNING expressions */
182 return ExecProject(projectReturning);
183}
184
185/*
186 * ExecCheckTupleVisible -- verify tuple is visible
187 *
188 * It would not be consistent with guarantees of the higher isolation levels to
189 * proceed with avoiding insertion (taking speculative insertion's alternative
190 * path) on the basis of another tuple that is not visible to MVCC snapshot.
191 * Check for the need to raise a serialization failure, and do so as necessary.
192 */
193static void
194ExecCheckTupleVisible(EState *estate,
195 Relation rel,
196 TupleTableSlot *slot)
197{
198 if (!IsolationUsesXactSnapshot())
199 return;
200
201 if (!table_tuple_satisfies_snapshot(rel, slot, estate->es_snapshot))
202 {
203 Datum xminDatum;
204 TransactionId xmin;
205 bool isnull;
206
207 xminDatum = slot_getsysattr(slot, MinTransactionIdAttributeNumber, &isnull);
208 Assert(!isnull);
209 xmin = DatumGetTransactionId(xminDatum);
210
211 /*
212 * We should not raise a serialization failure if the conflict is
213 * against a tuple inserted by our own transaction, even if it's not
214 * visible to our snapshot. (This would happen, for example, if
215 * conflicting keys are proposed for insertion in a single command.)
216 */
217 if (!TransactionIdIsCurrentTransactionId(xmin))
218 ereport(ERROR,
219 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
220 errmsg("could not serialize access due to concurrent update")));
221 }
222}
223
224/*
225 * ExecCheckTIDVisible -- convenience variant of ExecCheckTupleVisible()
226 */
227static void
228ExecCheckTIDVisible(EState *estate,
229 ResultRelInfo *relinfo,
230 ItemPointer tid,
231 TupleTableSlot *tempSlot)
232{
233 Relation rel = relinfo->ri_RelationDesc;
234
235 /* Redundantly check isolation level */
236 if (!IsolationUsesXactSnapshot())
237 return;
238
239 if (!table_tuple_fetch_row_version(rel, tid, SnapshotAny, tempSlot))
240 elog(ERROR, "failed to fetch conflicting tuple for ON CONFLICT");
241 ExecCheckTupleVisible(estate, rel, tempSlot);
242 ExecClearTuple(tempSlot);
243}
244
245/*
246 * Compute stored generated columns for a tuple
247 */
248void
249ExecComputeStoredGenerated(EState *estate, TupleTableSlot *slot)
250{
251 ResultRelInfo *resultRelInfo = estate->es_result_relation_info;
252 Relation rel = resultRelInfo->ri_RelationDesc;
253 TupleDesc tupdesc = RelationGetDescr(rel);
254 int natts = tupdesc->natts;
255 MemoryContext oldContext;
256 Datum *values;
257 bool *nulls;
258
259 Assert(tupdesc->constr && tupdesc->constr->has_generated_stored);
260
261 /*
262 * If first time through for this result relation, build expression
263 * nodetrees for rel's stored generation expressions. Keep them in the
264 * per-query memory context so they'll survive throughout the query.
265 */
266 if (resultRelInfo->ri_GeneratedExprs == NULL)
267 {
268 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
269
270 resultRelInfo->ri_GeneratedExprs =
271 (ExprState **) palloc(natts * sizeof(ExprState *));
272
273 for (int i = 0; i < natts; i++)
274 {
275 if (TupleDescAttr(tupdesc, i)->attgenerated == ATTRIBUTE_GENERATED_STORED)
276 {
277 Expr *expr;
278
279 expr = (Expr *) build_column_default(rel, i + 1);
280 if (expr == NULL)
281 elog(ERROR, "no generation expression found for column number %d of table \"%s\"",
282 i + 1, RelationGetRelationName(rel));
283
284 resultRelInfo->ri_GeneratedExprs[i] = ExecPrepareExpr(expr, estate);
285 }
286 }
287
288 MemoryContextSwitchTo(oldContext);
289 }
290
291 oldContext = MemoryContextSwitchTo(GetPerTupleMemoryContext(estate));
292
293 values = palloc(sizeof(*values) * natts);
294 nulls = palloc(sizeof(*nulls) * natts);
295
296 slot_getallattrs(slot);
297 memcpy(nulls, slot->tts_isnull, sizeof(*nulls) * natts);
298
299 for (int i = 0; i < natts; i++)
300 {
301 Form_pg_attribute attr = TupleDescAttr(tupdesc, i);
302
303 if (attr->attgenerated == ATTRIBUTE_GENERATED_STORED)
304 {
305 ExprContext *econtext;
306 Datum val;
307 bool isnull;
308
309 econtext = GetPerTupleExprContext(estate);
310 econtext->ecxt_scantuple = slot;
311
312 val = ExecEvalExpr(resultRelInfo->ri_GeneratedExprs[i], econtext, &isnull);
313
314 values[i] = val;
315 nulls[i] = isnull;
316 }
317 else
318 {
319 if (!nulls[i])
320 values[i] = datumCopy(slot->tts_values[i], attr->attbyval, attr->attlen);
321 }
322 }
323
324 ExecClearTuple(slot);
325 memcpy(slot->tts_values, values, sizeof(*values) * natts);
326 memcpy(slot->tts_isnull, nulls, sizeof(*nulls) * natts);
327 ExecStoreVirtualTuple(slot);
328 ExecMaterializeSlot(slot);
329
330 MemoryContextSwitchTo(oldContext);
331}
332
333/* ----------------------------------------------------------------
334 * ExecInsert
335 *
336 * For INSERT, we have to insert the tuple into the target relation
337 * and insert appropriate tuples into the index relations.
338 *
339 * Returns RETURNING result if any, otherwise NULL.
340 * ----------------------------------------------------------------
341 */
342static TupleTableSlot *
343ExecInsert(ModifyTableState *mtstate,
344 TupleTableSlot *slot,
345 TupleTableSlot *planSlot,
346 EState *estate,
347 bool canSetTag)
348{
349 ResultRelInfo *resultRelInfo;
350 Relation resultRelationDesc;
351 List *recheckIndexes = NIL;
352 TupleTableSlot *result = NULL;
353 TransitionCaptureState *ar_insert_trig_tcs;
354 ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
355 OnConflictAction onconflict = node->onConflictAction;
356
357 ExecMaterializeSlot(slot);
358
359 /*
360 * get information on the (current) result relation
361 */
362 resultRelInfo = estate->es_result_relation_info;
363 resultRelationDesc = resultRelInfo->ri_RelationDesc;
364
365 /*
366 * BEFORE ROW INSERT Triggers.
367 *
368 * Note: We fire BEFORE ROW TRIGGERS for every attempted insertion in an
369 * INSERT ... ON CONFLICT statement. We cannot check for constraint
370 * violations before firing these triggers, because they can change the
371 * values to insert. Also, they can run arbitrary user-defined code with
372 * side-effects that we can't cancel by just not inserting the tuple.
373 */
374 if (resultRelInfo->ri_TrigDesc &&
375 resultRelInfo->ri_TrigDesc->trig_insert_before_row)
376 {
377 if (!ExecBRInsertTriggers(estate, resultRelInfo, slot))
378 return NULL; /* "do nothing" */
379 }
380
381 /* INSTEAD OF ROW INSERT Triggers */
382 if (resultRelInfo->ri_TrigDesc &&
383 resultRelInfo->ri_TrigDesc->trig_insert_instead_row)
384 {
385 if (!ExecIRInsertTriggers(estate, resultRelInfo, slot))
386 return NULL; /* "do nothing" */
387 }
388 else if (resultRelInfo->ri_FdwRoutine)
389 {
390 /*
391 * Compute stored generated columns
392 */
393 if (resultRelationDesc->rd_att->constr &&
394 resultRelationDesc->rd_att->constr->has_generated_stored)
395 ExecComputeStoredGenerated(estate, slot);
396
397 /*
398 * insert into foreign table: let the FDW do it
399 */
400 slot = resultRelInfo->ri_FdwRoutine->ExecForeignInsert(estate,
401 resultRelInfo,
402 slot,
403 planSlot);
404
405 if (slot == NULL) /* "do nothing" */
406 return NULL;
407
408 /*
409 * AFTER ROW Triggers or RETURNING expressions might reference the
410 * tableoid column, so (re-)initialize tts_tableOid before evaluating
411 * them.
412 */
413 slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
414
415 }
416 else
417 {
418 WCOKind wco_kind;
419
420 /*
421 * Constraints might reference the tableoid column, so (re-)initialize
422 * tts_tableOid before evaluating them.
423 */
424 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
425
426 /*
427 * Compute stored generated columns
428 */
429 if (resultRelationDesc->rd_att->constr &&
430 resultRelationDesc->rd_att->constr->has_generated_stored)
431 ExecComputeStoredGenerated(estate, slot);
432
433 /*
434 * Check any RLS WITH CHECK policies.
435 *
436 * Normally we should check INSERT policies. But if the insert is the
437 * result of a partition key update that moved the tuple to a new
438 * partition, we should instead check UPDATE policies, because we are
439 * executing policies defined on the target table, and not those
440 * defined on the child partitions.
441 */
442 wco_kind = (mtstate->operation == CMD_UPDATE) ?
443 WCO_RLS_UPDATE_CHECK : WCO_RLS_INSERT_CHECK;
444
445 /*
446 * ExecWithCheckOptions() will skip any WCOs which are not of the kind
447 * we are looking for at this point.
448 */
449 if (resultRelInfo->ri_WithCheckOptions != NIL)
450 ExecWithCheckOptions(wco_kind, resultRelInfo, slot, estate);
451
452 /*
453 * Check the constraints of the tuple.
454 */
455 if (resultRelationDesc->rd_att->constr)
456 ExecConstraints(resultRelInfo, slot, estate);
457
458 /*
459 * Also check the tuple against the partition constraint, if there is
460 * one; except that if we got here via tuple-routing, we don't need to
461 * if there's no BR trigger defined on the partition.
462 */
463 if (resultRelInfo->ri_PartitionCheck &&
464 (resultRelInfo->ri_PartitionRoot == NULL ||
465 (resultRelInfo->ri_TrigDesc &&
466 resultRelInfo->ri_TrigDesc->trig_insert_before_row)))
467 ExecPartitionCheck(resultRelInfo, slot, estate, true);
468
469 if (onconflict != ONCONFLICT_NONE && resultRelInfo->ri_NumIndices > 0)
470 {
471 /* Perform a speculative insertion. */
472 uint32 specToken;
473 ItemPointerData conflictTid;
474 bool specConflict;
475 List *arbiterIndexes;
476
477 arbiterIndexes = resultRelInfo->ri_onConflictArbiterIndexes;
478
479 /*
480 * Do a non-conclusive check for conflicts first.
481 *
482 * We're not holding any locks yet, so this doesn't guarantee that
483 * the later insert won't conflict. But it avoids leaving behind
484 * a lot of canceled speculative insertions, if you run a lot of
485 * INSERT ON CONFLICT statements that do conflict.
486 *
487 * We loop back here if we find a conflict below, either during
488 * the pre-check, or when we re-check after inserting the tuple
489 * speculatively.
490 */
491 vlock:
492 specConflict = false;
493 if (!ExecCheckIndexConstraints(slot, estate, &conflictTid,
494 arbiterIndexes))
495 {
496 /* committed conflict tuple found */
497 if (onconflict == ONCONFLICT_UPDATE)
498 {
499 /*
500 * In case of ON CONFLICT DO UPDATE, execute the UPDATE
501 * part. Be prepared to retry if the UPDATE fails because
502 * of another concurrent UPDATE/DELETE to the conflict
503 * tuple.
504 */
505 TupleTableSlot *returning = NULL;
506
507 if (ExecOnConflictUpdate(mtstate, resultRelInfo,
508 &conflictTid, planSlot, slot,
509 estate, canSetTag, &returning))
510 {
511 InstrCountTuples2(&mtstate->ps, 1);
512 return returning;
513 }
514 else
515 goto vlock;
516 }
517 else
518 {
519 /*
520 * In case of ON CONFLICT DO NOTHING, do nothing. However,
521 * verify that the tuple is visible to the executor's MVCC
522 * snapshot at higher isolation levels.
523 *
524 * Using ExecGetReturningSlot() to store the tuple for the
525 * recheck isn't that pretty, but we can't trivially use
526 * the input slot, because it might not be of a compatible
527 * type. As there's no conflicting usage of
528 * ExecGetReturningSlot() in the DO NOTHING case...
529 */
530 Assert(onconflict == ONCONFLICT_NOTHING);
531 ExecCheckTIDVisible(estate, resultRelInfo, &conflictTid,
532 ExecGetReturningSlot(estate, resultRelInfo));
533 InstrCountTuples2(&mtstate->ps, 1);
534 return NULL;
535 }
536 }
537
538 /*
539 * Before we start insertion proper, acquire our "speculative
540 * insertion lock". Others can use that to wait for us to decide
541 * if we're going to go ahead with the insertion, instead of
542 * waiting for the whole transaction to complete.
543 */
544 specToken = SpeculativeInsertionLockAcquire(GetCurrentTransactionId());
545
546 /* insert the tuple, with the speculative token */
547 table_tuple_insert_speculative(resultRelationDesc, slot,
548 estate->es_output_cid,
549 0,
550 NULL,
551 specToken);
552
553 /* insert index entries for tuple */
554 recheckIndexes = ExecInsertIndexTuples(slot, estate, true,
555 &specConflict,
556 arbiterIndexes);
557
558 /* adjust the tuple's state accordingly */
559 table_tuple_complete_speculative(resultRelationDesc, slot,
560 specToken, !specConflict);
561
562 /*
563 * Wake up anyone waiting for our decision. They will re-check
564 * the tuple, see that it's no longer speculative, and wait on our
565 * XID as if this was a regularly inserted tuple all along. Or if
566 * we killed the tuple, they will see it's dead, and proceed as if
567 * the tuple never existed.
568 */
569 SpeculativeInsertionLockRelease(GetCurrentTransactionId());
570
571 /*
572 * If there was a conflict, start from the beginning. We'll do
573 * the pre-check again, which will now find the conflicting tuple
574 * (unless it aborts before we get there).
575 */
576 if (specConflict)
577 {
578 list_free(recheckIndexes);
579 goto vlock;
580 }
581
582 /* Since there was no insertion conflict, we're done */
583 }
584 else
585 {
586 /* insert the tuple normally */
587 table_tuple_insert(resultRelationDesc, slot,
588 estate->es_output_cid,
589 0, NULL);
590
591 /* insert index entries for tuple */
592 if (resultRelInfo->ri_NumIndices > 0)
593 recheckIndexes = ExecInsertIndexTuples(slot, estate, false, NULL,
594 NIL);
595 }
596 }
597
598 if (canSetTag)
599 {
600 (estate->es_processed)++;
601 setLastTid(&slot->tts_tid);
602 }
603
604 /*
605 * If this insert is the result of a partition key update that moved the
606 * tuple to a new partition, put this row into the transition NEW TABLE,
607 * if there is one. We need to do this separately for DELETE and INSERT
608 * because they happen on different tables.
609 */
610 ar_insert_trig_tcs = mtstate->mt_transition_capture;
611 if (mtstate->operation == CMD_UPDATE && mtstate->mt_transition_capture
612 && mtstate->mt_transition_capture->tcs_update_new_table)
613 {
614 ExecARUpdateTriggers(estate, resultRelInfo, NULL,
615 NULL,
616 slot,
617 NULL,
618 mtstate->mt_transition_capture);
619
620 /*
621 * We've already captured the NEW TABLE row, so make sure any AR
622 * INSERT trigger fired below doesn't capture it again.
623 */
624 ar_insert_trig_tcs = NULL;
625 }
626
627 /* AFTER ROW INSERT Triggers */
628 ExecARInsertTriggers(estate, resultRelInfo, slot, recheckIndexes,
629 ar_insert_trig_tcs);
630
631 list_free(recheckIndexes);
632
633 /*
634 * Check any WITH CHECK OPTION constraints from parent views. We are
635 * required to do this after testing all constraints and uniqueness
636 * violations per the SQL spec, so we do it after actually inserting the
637 * record into the heap and all indexes.
638 *
639 * ExecWithCheckOptions will elog(ERROR) if a violation is found, so the
640 * tuple will never be seen, if it violates the WITH CHECK OPTION.
641 *
642 * ExecWithCheckOptions() will skip any WCOs which are not of the kind we
643 * are looking for at this point.
644 */
645 if (resultRelInfo->ri_WithCheckOptions != NIL)
646 ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
647
648 /* Process RETURNING if present */
649 if (resultRelInfo->ri_projectReturning)
650 result = ExecProcessReturning(resultRelInfo, slot, planSlot);
651
652 return result;
653}
654
655/* ----------------------------------------------------------------
656 * ExecDelete
657 *
658 * DELETE is like UPDATE, except that we delete the tuple and no
659 * index modifications are needed.
660 *
661 * When deleting from a table, tupleid identifies the tuple to
662 * delete and oldtuple is NULL. When deleting from a view,
663 * oldtuple is passed to the INSTEAD OF triggers and identifies
664 * what to delete, and tupleid is invalid. When deleting from a
665 * foreign table, tupleid is invalid; the FDW has to figure out
666 * which row to delete using data from the planSlot. oldtuple is
667 * passed to foreign table triggers; it is NULL when the foreign
668 * table has no relevant triggers. We use tupleDeleted to indicate
669 * whether the tuple is actually deleted, callers can use it to
670 * decide whether to continue the operation. When this DELETE is a
671 * part of an UPDATE of partition-key, then the slot returned by
672 * EvalPlanQual() is passed back using output parameter epqslot.
673 *
674 * Returns RETURNING result if any, otherwise NULL.
675 * ----------------------------------------------------------------
676 */
677static TupleTableSlot *
678ExecDelete(ModifyTableState *mtstate,
679 ItemPointer tupleid,
680 HeapTuple oldtuple,
681 TupleTableSlot *planSlot,
682 EPQState *epqstate,
683 EState *estate,
684 bool processReturning,
685 bool canSetTag,
686 bool changingPart,
687 bool *tupleDeleted,
688 TupleTableSlot **epqreturnslot)
689{
690 ResultRelInfo *resultRelInfo;
691 Relation resultRelationDesc;
692 TM_Result result;
693 TM_FailureData tmfd;
694 TupleTableSlot *slot = NULL;
695 TransitionCaptureState *ar_delete_trig_tcs;
696
697 if (tupleDeleted)
698 *tupleDeleted = false;
699
700 /*
701 * get information on the (current) result relation
702 */
703 resultRelInfo = estate->es_result_relation_info;
704 resultRelationDesc = resultRelInfo->ri_RelationDesc;
705
706 /* BEFORE ROW DELETE Triggers */
707 if (resultRelInfo->ri_TrigDesc &&
708 resultRelInfo->ri_TrigDesc->trig_delete_before_row)
709 {
710 bool dodelete;
711
712 dodelete = ExecBRDeleteTriggers(estate, epqstate, resultRelInfo,
713 tupleid, oldtuple, epqreturnslot);
714
715 if (!dodelete) /* "do nothing" */
716 return NULL;
717 }
718
719 /* INSTEAD OF ROW DELETE Triggers */
720 if (resultRelInfo->ri_TrigDesc &&
721 resultRelInfo->ri_TrigDesc->trig_delete_instead_row)
722 {
723 bool dodelete;
724
725 Assert(oldtuple != NULL);
726 dodelete = ExecIRDeleteTriggers(estate, resultRelInfo, oldtuple);
727
728 if (!dodelete) /* "do nothing" */
729 return NULL;
730 }
731 else if (resultRelInfo->ri_FdwRoutine)
732 {
733 /*
734 * delete from foreign table: let the FDW do it
735 *
736 * We offer the returning slot as a place to store RETURNING data,
737 * although the FDW can return some other slot if it wants.
738 */
739 slot = ExecGetReturningSlot(estate, resultRelInfo);
740 slot = resultRelInfo->ri_FdwRoutine->ExecForeignDelete(estate,
741 resultRelInfo,
742 slot,
743 planSlot);
744
745 if (slot == NULL) /* "do nothing" */
746 return NULL;
747
748 /*
749 * RETURNING expressions might reference the tableoid column, so
750 * (re)initialize tts_tableOid before evaluating them.
751 */
752 if (TTS_EMPTY(slot))
753 ExecStoreAllNullTuple(slot);
754
755 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
756 }
757 else
758 {
759 /*
760 * delete the tuple
761 *
762 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check
763 * that the row to be deleted is visible to that snapshot, and throw a
764 * can't-serialize error if not. This is a special-case behavior
765 * needed for referential integrity updates in transaction-snapshot
766 * mode transactions.
767 */
768ldelete:;
769 result = table_tuple_delete(resultRelationDesc, tupleid,
770 estate->es_output_cid,
771 estate->es_snapshot,
772 estate->es_crosscheck_snapshot,
773 true /* wait for commit */ ,
774 &tmfd,
775 changingPart);
776
777 switch (result)
778 {
779 case TM_SelfModified:
780
781 /*
782 * The target tuple was already updated or deleted by the
783 * current command, or by a later command in the current
784 * transaction. The former case is possible in a join DELETE
785 * where multiple tuples join to the same target tuple. This
786 * is somewhat questionable, but Postgres has always allowed
787 * it: we just ignore additional deletion attempts.
788 *
789 * The latter case arises if the tuple is modified by a
790 * command in a BEFORE trigger, or perhaps by a command in a
791 * volatile function used in the query. In such situations we
792 * should not ignore the deletion, but it is equally unsafe to
793 * proceed. We don't want to discard the original DELETE
794 * while keeping the triggered actions based on its deletion;
795 * and it would be no better to allow the original DELETE
796 * while discarding updates that it triggered. The row update
797 * carries some information that might be important according
798 * to business rules; so throwing an error is the only safe
799 * course.
800 *
801 * If a trigger actually intends this type of interaction, it
802 * can re-execute the DELETE and then return NULL to cancel
803 * the outer delete.
804 */
805 if (tmfd.cmax != estate->es_output_cid)
806 ereport(ERROR,
807 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
808 errmsg("tuple to be deleted was already modified by an operation triggered by the current command"),
809 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
810
811 /* Else, already deleted by self; nothing to do */
812 return NULL;
813
814 case TM_Ok:
815 break;
816
817 case TM_Updated:
818 {
819 TupleTableSlot *inputslot;
820 TupleTableSlot *epqslot;
821
822 if (IsolationUsesXactSnapshot())
823 ereport(ERROR,
824 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
825 errmsg("could not serialize access due to concurrent update")));
826
827 /*
828 * Already know that we're going to need to do EPQ, so
829 * fetch tuple directly into the right slot.
830 */
831 EvalPlanQualBegin(epqstate);
832 inputslot = EvalPlanQualSlot(epqstate, resultRelationDesc,
833 resultRelInfo->ri_RangeTableIndex);
834
835 result = table_tuple_lock(resultRelationDesc, tupleid,
836 estate->es_snapshot,
837 inputslot, estate->es_output_cid,
838 LockTupleExclusive, LockWaitBlock,
839 TUPLE_LOCK_FLAG_FIND_LAST_VERSION,
840 &tmfd);
841
842 switch (result)
843 {
844 case TM_Ok:
845 Assert(tmfd.traversed);
846 epqslot = EvalPlanQual(epqstate,
847 resultRelationDesc,
848 resultRelInfo->ri_RangeTableIndex,
849 inputslot);
850 if (TupIsNull(epqslot))
851 /* Tuple not passing quals anymore, exiting... */
852 return NULL;
853
854 /*
855 * If requested, skip delete and pass back the
856 * updated row.
857 */
858 if (epqreturnslot)
859 {
860 *epqreturnslot = epqslot;
861 return NULL;
862 }
863 else
864 goto ldelete;
865
866 case TM_SelfModified:
867
868 /*
869 * This can be reached when following an update
870 * chain from a tuple updated by another session,
871 * reaching a tuple that was already updated in
872 * this transaction. If previously updated by this
873 * command, ignore the delete, otherwise error
874 * out.
875 *
876 * See also TM_SelfModified response to
877 * table_tuple_delete() above.
878 */
879 if (tmfd.cmax != estate->es_output_cid)
880 ereport(ERROR,
881 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
882 errmsg("tuple to be deleted was already modified by an operation triggered by the current command"),
883 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
884 return NULL;
885
886 case TM_Deleted:
887 /* tuple already deleted; nothing to do */
888 return NULL;
889
890 default:
891
892 /*
893 * TM_Invisible should be impossible because we're
894 * waiting for updated row versions, and would
895 * already have errored out if the first version
896 * is invisible.
897 *
898 * TM_Updated should be impossible, because we're
899 * locking the latest version via
900 * TUPLE_LOCK_FLAG_FIND_LAST_VERSION.
901 */
902 elog(ERROR, "unexpected table_tuple_lock status: %u",
903 result);
904 return NULL;
905 }
906
907 Assert(false);
908 break;
909 }
910
911 case TM_Deleted:
912 if (IsolationUsesXactSnapshot())
913 ereport(ERROR,
914 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
915 errmsg("could not serialize access due to concurrent delete")));
916 /* tuple already deleted; nothing to do */
917 return NULL;
918
919 default:
920 elog(ERROR, "unrecognized table_tuple_delete status: %u",
921 result);
922 return NULL;
923 }
924
925 /*
926 * Note: Normally one would think that we have to delete index tuples
927 * associated with the heap tuple now...
928 *
929 * ... but in POSTGRES, we have no need to do this because VACUUM will
930 * take care of it later. We can't delete index tuples immediately
931 * anyway, since the tuple is still visible to other transactions.
932 */
933 }
934
935 if (canSetTag)
936 (estate->es_processed)++;
937
938 /* Tell caller that the delete actually happened. */
939 if (tupleDeleted)
940 *tupleDeleted = true;
941
942 /*
943 * If this delete is the result of a partition key update that moved the
944 * tuple to a new partition, put this row into the transition OLD TABLE,
945 * if there is one. We need to do this separately for DELETE and INSERT
946 * because they happen on different tables.
947 */
948 ar_delete_trig_tcs = mtstate->mt_transition_capture;
949 if (mtstate->operation == CMD_UPDATE && mtstate->mt_transition_capture
950 && mtstate->mt_transition_capture->tcs_update_old_table)
951 {
952 ExecARUpdateTriggers(estate, resultRelInfo,
953 tupleid,
954 oldtuple,
955 NULL,
956 NULL,
957 mtstate->mt_transition_capture);
958
959 /*
960 * We've already captured the NEW TABLE row, so make sure any AR
961 * DELETE trigger fired below doesn't capture it again.
962 */
963 ar_delete_trig_tcs = NULL;
964 }
965
966 /* AFTER ROW DELETE Triggers */
967 ExecARDeleteTriggers(estate, resultRelInfo, tupleid, oldtuple,
968 ar_delete_trig_tcs);
969
970 /* Process RETURNING if present and if requested */
971 if (processReturning && resultRelInfo->ri_projectReturning)
972 {
973 /*
974 * We have to put the target tuple into a slot, which means first we
975 * gotta fetch it. We can use the trigger tuple slot.
976 */
977 TupleTableSlot *rslot;
978
979 if (resultRelInfo->ri_FdwRoutine)
980 {
981 /* FDW must have provided a slot containing the deleted row */
982 Assert(!TupIsNull(slot));
983 }
984 else
985 {
986 slot = ExecGetReturningSlot(estate, resultRelInfo);
987 if (oldtuple != NULL)
988 {
989 ExecForceStoreHeapTuple(oldtuple, slot, false);
990 }
991 else
992 {
993 if (!table_tuple_fetch_row_version(resultRelationDesc, tupleid,
994 SnapshotAny, slot))
995 elog(ERROR, "failed to fetch deleted tuple for DELETE RETURNING");
996 }
997 }
998
999 rslot = ExecProcessReturning(resultRelInfo, slot, planSlot);
1000
1001 /*
1002 * Before releasing the target tuple again, make sure rslot has a
1003 * local copy of any pass-by-reference values.
1004 */
1005 ExecMaterializeSlot(rslot);
1006
1007 ExecClearTuple(slot);
1008
1009 return rslot;
1010 }
1011
1012 return NULL;
1013}
1014
1015/* ----------------------------------------------------------------
1016 * ExecUpdate
1017 *
1018 * note: we can't run UPDATE queries with transactions
1019 * off because UPDATEs are actually INSERTs and our
1020 * scan will mistakenly loop forever, updating the tuple
1021 * it just inserted.. This should be fixed but until it
1022 * is, we don't want to get stuck in an infinite loop
1023 * which corrupts your database..
1024 *
1025 * When updating a table, tupleid identifies the tuple to
1026 * update and oldtuple is NULL. When updating a view, oldtuple
1027 * is passed to the INSTEAD OF triggers and identifies what to
1028 * update, and tupleid is invalid. When updating a foreign table,
1029 * tupleid is invalid; the FDW has to figure out which row to
1030 * update using data from the planSlot. oldtuple is passed to
1031 * foreign table triggers; it is NULL when the foreign table has
1032 * no relevant triggers.
1033 *
1034 * Returns RETURNING result if any, otherwise NULL.
1035 * ----------------------------------------------------------------
1036 */
1037static TupleTableSlot *
1038ExecUpdate(ModifyTableState *mtstate,
1039 ItemPointer tupleid,
1040 HeapTuple oldtuple,
1041 TupleTableSlot *slot,
1042 TupleTableSlot *planSlot,
1043 EPQState *epqstate,
1044 EState *estate,
1045 bool canSetTag)
1046{
1047 ResultRelInfo *resultRelInfo;
1048 Relation resultRelationDesc;
1049 TM_Result result;
1050 TM_FailureData tmfd;
1051 List *recheckIndexes = NIL;
1052 TupleConversionMap *saved_tcs_map = NULL;
1053
1054 /*
1055 * abort the operation if not running transactions
1056 */
1057 if (IsBootstrapProcessingMode())
1058 elog(ERROR, "cannot UPDATE during bootstrap");
1059
1060 ExecMaterializeSlot(slot);
1061
1062 /*
1063 * get information on the (current) result relation
1064 */
1065 resultRelInfo = estate->es_result_relation_info;
1066 resultRelationDesc = resultRelInfo->ri_RelationDesc;
1067
1068 /* BEFORE ROW UPDATE Triggers */
1069 if (resultRelInfo->ri_TrigDesc &&
1070 resultRelInfo->ri_TrigDesc->trig_update_before_row)
1071 {
1072 if (!ExecBRUpdateTriggers(estate, epqstate, resultRelInfo,
1073 tupleid, oldtuple, slot))
1074 return NULL; /* "do nothing" */
1075 }
1076
1077 /* INSTEAD OF ROW UPDATE Triggers */
1078 if (resultRelInfo->ri_TrigDesc &&
1079 resultRelInfo->ri_TrigDesc->trig_update_instead_row)
1080 {
1081 if (!ExecIRUpdateTriggers(estate, resultRelInfo,
1082 oldtuple, slot))
1083 return NULL; /* "do nothing" */
1084 }
1085 else if (resultRelInfo->ri_FdwRoutine)
1086 {
1087 /*
1088 * Compute stored generated columns
1089 */
1090 if (resultRelationDesc->rd_att->constr &&
1091 resultRelationDesc->rd_att->constr->has_generated_stored)
1092 ExecComputeStoredGenerated(estate, slot);
1093
1094 /*
1095 * update in foreign table: let the FDW do it
1096 */
1097 slot = resultRelInfo->ri_FdwRoutine->ExecForeignUpdate(estate,
1098 resultRelInfo,
1099 slot,
1100 planSlot);
1101
1102 if (slot == NULL) /* "do nothing" */
1103 return NULL;
1104
1105 /*
1106 * AFTER ROW Triggers or RETURNING expressions might reference the
1107 * tableoid column, so (re-)initialize tts_tableOid before evaluating
1108 * them.
1109 */
1110 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
1111 }
1112 else
1113 {
1114 LockTupleMode lockmode;
1115 bool partition_constraint_failed;
1116 bool update_indexes;
1117
1118 /*
1119 * Constraints might reference the tableoid column, so (re-)initialize
1120 * tts_tableOid before evaluating them.
1121 */
1122 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
1123
1124 /*
1125 * Compute stored generated columns
1126 */
1127 if (resultRelationDesc->rd_att->constr &&
1128 resultRelationDesc->rd_att->constr->has_generated_stored)
1129 ExecComputeStoredGenerated(estate, slot);
1130
1131 /*
1132 * Check any RLS UPDATE WITH CHECK policies
1133 *
1134 * If we generate a new candidate tuple after EvalPlanQual testing, we
1135 * must loop back here and recheck any RLS policies and constraints.
1136 * (We don't need to redo triggers, however. If there are any BEFORE
1137 * triggers then trigger.c will have done table_tuple_lock to lock the
1138 * correct tuple, so there's no need to do them again.)
1139 */
1140lreplace:;
1141
1142 /* ensure slot is independent, consider e.g. EPQ */
1143 ExecMaterializeSlot(slot);
1144
1145 /*
1146 * If partition constraint fails, this row might get moved to another
1147 * partition, in which case we should check the RLS CHECK policy just
1148 * before inserting into the new partition, rather than doing it here.
1149 * This is because a trigger on that partition might again change the
1150 * row. So skip the WCO checks if the partition constraint fails.
1151 */
1152 partition_constraint_failed =
1153 resultRelInfo->ri_PartitionCheck &&
1154 !ExecPartitionCheck(resultRelInfo, slot, estate, false);
1155
1156 if (!partition_constraint_failed &&
1157 resultRelInfo->ri_WithCheckOptions != NIL)
1158 {
1159 /*
1160 * ExecWithCheckOptions() will skip any WCOs which are not of the
1161 * kind we are looking for at this point.
1162 */
1163 ExecWithCheckOptions(WCO_RLS_UPDATE_CHECK,
1164 resultRelInfo, slot, estate);
1165 }
1166
1167 /*
1168 * If a partition check failed, try to move the row into the right
1169 * partition.
1170 */
1171 if (partition_constraint_failed)
1172 {
1173 bool tuple_deleted;
1174 TupleTableSlot *ret_slot;
1175 TupleTableSlot *epqslot = NULL;
1176 PartitionTupleRouting *proute = mtstate->mt_partition_tuple_routing;
1177 int map_index;
1178 TupleConversionMap *tupconv_map;
1179
1180 /*
1181 * Disallow an INSERT ON CONFLICT DO UPDATE that causes the
1182 * original row to migrate to a different partition. Maybe this
1183 * can be implemented some day, but it seems a fringe feature with
1184 * little redeeming value.
1185 */
1186 if (((ModifyTable *) mtstate->ps.plan)->onConflictAction == ONCONFLICT_UPDATE)
1187 ereport(ERROR,
1188 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1189 errmsg("invalid ON UPDATE specification"),
1190 errdetail("The result tuple would appear in a different partition than the original tuple.")));
1191
1192 /*
1193 * When an UPDATE is run on a leaf partition, we will not have
1194 * partition tuple routing set up. In that case, fail with
1195 * partition constraint violation error.
1196 */
1197 if (proute == NULL)
1198 ExecPartitionCheckEmitError(resultRelInfo, slot, estate);
1199
1200 /*
1201 * Row movement, part 1. Delete the tuple, but skip RETURNING
1202 * processing. We want to return rows from INSERT.
1203 */
1204 ExecDelete(mtstate, tupleid, oldtuple, planSlot, epqstate,
1205 estate, false, false /* canSetTag */ ,
1206 true /* changingPart */ , &tuple_deleted, &epqslot);
1207
1208 /*
1209 * For some reason if DELETE didn't happen (e.g. trigger prevented
1210 * it, or it was already deleted by self, or it was concurrently
1211 * deleted by another transaction), then we should skip the insert
1212 * as well; otherwise, an UPDATE could cause an increase in the
1213 * total number of rows across all partitions, which is clearly
1214 * wrong.
1215 *
1216 * For a normal UPDATE, the case where the tuple has been the
1217 * subject of a concurrent UPDATE or DELETE would be handled by
1218 * the EvalPlanQual machinery, but for an UPDATE that we've
1219 * translated into a DELETE from this partition and an INSERT into
1220 * some other partition, that's not available, because CTID chains
1221 * can't span relation boundaries. We mimic the semantics to a
1222 * limited extent by skipping the INSERT if the DELETE fails to
1223 * find a tuple. This ensures that two concurrent attempts to
1224 * UPDATE the same tuple at the same time can't turn one tuple
1225 * into two, and that an UPDATE of a just-deleted tuple can't
1226 * resurrect it.
1227 */
1228 if (!tuple_deleted)
1229 {
1230 /*
1231 * epqslot will be typically NULL. But when ExecDelete()
1232 * finds that another transaction has concurrently updated the
1233 * same row, it re-fetches the row, skips the delete, and
1234 * epqslot is set to the re-fetched tuple slot. In that case,
1235 * we need to do all the checks again.
1236 */
1237 if (TupIsNull(epqslot))
1238 return NULL;
1239 else
1240 {
1241 slot = ExecFilterJunk(resultRelInfo->ri_junkFilter, epqslot);
1242 goto lreplace;
1243 }
1244 }
1245
1246 /*
1247 * Updates set the transition capture map only when a new subplan
1248 * is chosen. But for inserts, it is set for each row. So after
1249 * INSERT, we need to revert back to the map created for UPDATE;
1250 * otherwise the next UPDATE will incorrectly use the one created
1251 * for INSERT. So first save the one created for UPDATE.
1252 */
1253 if (mtstate->mt_transition_capture)
1254 saved_tcs_map = mtstate->mt_transition_capture->tcs_map;
1255
1256 /*
1257 * resultRelInfo is one of the per-subplan resultRelInfos. So we
1258 * should convert the tuple into root's tuple descriptor, since
1259 * ExecInsert() starts the search from root. The tuple conversion
1260 * map list is in the order of mtstate->resultRelInfo[], so to
1261 * retrieve the one for this resultRel, we need to know the
1262 * position of the resultRel in mtstate->resultRelInfo[].
1263 */
1264 map_index = resultRelInfo - mtstate->resultRelInfo;
1265 Assert(map_index >= 0 && map_index < mtstate->mt_nplans);
1266 tupconv_map = tupconv_map_for_subplan(mtstate, map_index);
1267 if (tupconv_map != NULL)
1268 slot = execute_attr_map_slot(tupconv_map->attrMap,
1269 slot,
1270 mtstate->mt_root_tuple_slot);
1271
1272 /*
1273 * Prepare for tuple routing, making it look like we're inserting
1274 * into the root.
1275 */
1276 Assert(mtstate->rootResultRelInfo != NULL);
1277 slot = ExecPrepareTupleRouting(mtstate, estate, proute,
1278 mtstate->rootResultRelInfo, slot);
1279
1280 ret_slot = ExecInsert(mtstate, slot, planSlot,
1281 estate, canSetTag);
1282
1283 /* Revert ExecPrepareTupleRouting's node change. */
1284 estate->es_result_relation_info = resultRelInfo;
1285 if (mtstate->mt_transition_capture)
1286 {
1287 mtstate->mt_transition_capture->tcs_original_insert_tuple = NULL;
1288 mtstate->mt_transition_capture->tcs_map = saved_tcs_map;
1289 }
1290
1291 return ret_slot;
1292 }
1293
1294 /*
1295 * Check the constraints of the tuple. We've already checked the
1296 * partition constraint above; however, we must still ensure the tuple
1297 * passes all other constraints, so we will call ExecConstraints() and
1298 * have it validate all remaining checks.
1299 */
1300 if (resultRelationDesc->rd_att->constr)
1301 ExecConstraints(resultRelInfo, slot, estate);
1302
1303 /*
1304 * replace the heap tuple
1305 *
1306 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check
1307 * that the row to be updated is visible to that snapshot, and throw a
1308 * can't-serialize error if not. This is a special-case behavior
1309 * needed for referential integrity updates in transaction-snapshot
1310 * mode transactions.
1311 */
1312 result = table_tuple_update(resultRelationDesc, tupleid, slot,
1313 estate->es_output_cid,
1314 estate->es_snapshot,
1315 estate->es_crosscheck_snapshot,
1316 true /* wait for commit */ ,
1317 &tmfd, &lockmode, &update_indexes);
1318
1319 switch (result)
1320 {
1321 case TM_SelfModified:
1322
1323 /*
1324 * The target tuple was already updated or deleted by the
1325 * current command, or by a later command in the current
1326 * transaction. The former case is possible in a join UPDATE
1327 * where multiple tuples join to the same target tuple. This
1328 * is pretty questionable, but Postgres has always allowed it:
1329 * we just execute the first update action and ignore
1330 * additional update attempts.
1331 *
1332 * The latter case arises if the tuple is modified by a
1333 * command in a BEFORE trigger, or perhaps by a command in a
1334 * volatile function used in the query. In such situations we
1335 * should not ignore the update, but it is equally unsafe to
1336 * proceed. We don't want to discard the original UPDATE
1337 * while keeping the triggered actions based on it; and we
1338 * have no principled way to merge this update with the
1339 * previous ones. So throwing an error is the only safe
1340 * course.
1341 *
1342 * If a trigger actually intends this type of interaction, it
1343 * can re-execute the UPDATE (assuming it can figure out how)
1344 * and then return NULL to cancel the outer update.
1345 */
1346 if (tmfd.cmax != estate->es_output_cid)
1347 ereport(ERROR,
1348 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
1349 errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
1350 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
1351
1352 /* Else, already updated by self; nothing to do */
1353 return NULL;
1354
1355 case TM_Ok:
1356 break;
1357
1358 case TM_Updated:
1359 {
1360 TupleTableSlot *inputslot;
1361 TupleTableSlot *epqslot;
1362
1363 if (IsolationUsesXactSnapshot())
1364 ereport(ERROR,
1365 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1366 errmsg("could not serialize access due to concurrent update")));
1367
1368 /*
1369 * Already know that we're going to need to do EPQ, so
1370 * fetch tuple directly into the right slot.
1371 */
1372 inputslot = EvalPlanQualSlot(epqstate, resultRelationDesc,
1373 resultRelInfo->ri_RangeTableIndex);
1374
1375 result = table_tuple_lock(resultRelationDesc, tupleid,
1376 estate->es_snapshot,
1377 inputslot, estate->es_output_cid,
1378 lockmode, LockWaitBlock,
1379 TUPLE_LOCK_FLAG_FIND_LAST_VERSION,
1380 &tmfd);
1381
1382 switch (result)
1383 {
1384 case TM_Ok:
1385 Assert(tmfd.traversed);
1386
1387 epqslot = EvalPlanQual(epqstate,
1388 resultRelationDesc,
1389 resultRelInfo->ri_RangeTableIndex,
1390 inputslot);
1391 if (TupIsNull(epqslot))
1392 /* Tuple not passing quals anymore, exiting... */
1393 return NULL;
1394
1395 slot = ExecFilterJunk(resultRelInfo->ri_junkFilter, epqslot);
1396 goto lreplace;
1397
1398 case TM_Deleted:
1399 /* tuple already deleted; nothing to do */
1400 return NULL;
1401
1402 case TM_SelfModified:
1403
1404 /*
1405 * This can be reached when following an update
1406 * chain from a tuple updated by another session,
1407 * reaching a tuple that was already updated in
1408 * this transaction. If previously modified by
1409 * this command, ignore the redundant update,
1410 * otherwise error out.
1411 *
1412 * See also TM_SelfModified response to
1413 * table_tuple_update() above.
1414 */
1415 if (tmfd.cmax != estate->es_output_cid)
1416 ereport(ERROR,
1417 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
1418 errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
1419 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
1420 return NULL;
1421
1422 default:
1423 /* see table_tuple_lock call in ExecDelete() */
1424 elog(ERROR, "unexpected table_tuple_lock status: %u",
1425 result);
1426 return NULL;
1427 }
1428 }
1429
1430 break;
1431
1432 case TM_Deleted:
1433 if (IsolationUsesXactSnapshot())
1434 ereport(ERROR,
1435 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1436 errmsg("could not serialize access due to concurrent delete")));
1437 /* tuple already deleted; nothing to do */
1438 return NULL;
1439
1440 default:
1441 elog(ERROR, "unrecognized table_tuple_update status: %u",
1442 result);
1443 return NULL;
1444 }
1445
1446 /* insert index entries for tuple if necessary */
1447 if (resultRelInfo->ri_NumIndices > 0 && update_indexes)
1448 recheckIndexes = ExecInsertIndexTuples(slot, estate, false, NULL, NIL);
1449 }
1450
1451 if (canSetTag)
1452 (estate->es_processed)++;
1453
1454 /* AFTER ROW UPDATE Triggers */
1455 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, oldtuple, slot,
1456 recheckIndexes,
1457 mtstate->operation == CMD_INSERT ?
1458 mtstate->mt_oc_transition_capture :
1459 mtstate->mt_transition_capture);
1460
1461 list_free(recheckIndexes);
1462
1463 /*
1464 * Check any WITH CHECK OPTION constraints from parent views. We are
1465 * required to do this after testing all constraints and uniqueness
1466 * violations per the SQL spec, so we do it after actually updating the
1467 * record in the heap and all indexes.
1468 *
1469 * ExecWithCheckOptions() will skip any WCOs which are not of the kind we
1470 * are looking for at this point.
1471 */
1472 if (resultRelInfo->ri_WithCheckOptions != NIL)
1473 ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
1474
1475 /* Process RETURNING if present */
1476 if (resultRelInfo->ri_projectReturning)
1477 return ExecProcessReturning(resultRelInfo, slot, planSlot);
1478
1479 return NULL;
1480}
1481
1482/*
1483 * ExecOnConflictUpdate --- execute UPDATE of INSERT ON CONFLICT DO UPDATE
1484 *
1485 * Try to lock tuple for update as part of speculative insertion. If
1486 * a qual originating from ON CONFLICT DO UPDATE is satisfied, update
1487 * (but still lock row, even though it may not satisfy estate's
1488 * snapshot).
1489 *
1490 * Returns true if we're done (with or without an update), or false if
1491 * the caller must retry the INSERT from scratch.
1492 */
1493static bool
1494ExecOnConflictUpdate(ModifyTableState *mtstate,
1495 ResultRelInfo *resultRelInfo,
1496 ItemPointer conflictTid,
1497 TupleTableSlot *planSlot,
1498 TupleTableSlot *excludedSlot,
1499 EState *estate,
1500 bool canSetTag,
1501 TupleTableSlot **returning)
1502{
1503 ExprContext *econtext = mtstate->ps.ps_ExprContext;
1504 Relation relation = resultRelInfo->ri_RelationDesc;
1505 ExprState *onConflictSetWhere = resultRelInfo->ri_onConflict->oc_WhereClause;
1506 TupleTableSlot *existing = resultRelInfo->ri_onConflict->oc_Existing;
1507 TM_FailureData tmfd;
1508 LockTupleMode lockmode;
1509 TM_Result test;
1510 Datum xminDatum;
1511 TransactionId xmin;
1512 bool isnull;
1513
1514 /* Determine lock mode to use */
1515 lockmode = ExecUpdateLockMode(estate, resultRelInfo);
1516
1517 /*
1518 * Lock tuple for update. Don't follow updates when tuple cannot be
1519 * locked without doing so. A row locking conflict here means our
1520 * previous conclusion that the tuple is conclusively committed is not
1521 * true anymore.
1522 */
1523 test = table_tuple_lock(relation, conflictTid,
1524 estate->es_snapshot,
1525 existing, estate->es_output_cid,
1526 lockmode, LockWaitBlock, 0,
1527 &tmfd);
1528 switch (test)
1529 {
1530 case TM_Ok:
1531 /* success! */
1532 break;
1533
1534 case TM_Invisible:
1535
1536 /*
1537 * This can occur when a just inserted tuple is updated again in
1538 * the same command. E.g. because multiple rows with the same
1539 * conflicting key values are inserted.
1540 *
1541 * This is somewhat similar to the ExecUpdate() TM_SelfModified
1542 * case. We do not want to proceed because it would lead to the
1543 * same row being updated a second time in some unspecified order,
1544 * and in contrast to plain UPDATEs there's no historical behavior
1545 * to break.
1546 *
1547 * It is the user's responsibility to prevent this situation from
1548 * occurring. These problems are why SQL-2003 similarly specifies
1549 * that for SQL MERGE, an exception must be raised in the event of
1550 * an attempt to update the same row twice.
1551 */
1552 xminDatum = slot_getsysattr(existing,
1553 MinTransactionIdAttributeNumber,
1554 &isnull);
1555 Assert(!isnull);
1556 xmin = DatumGetTransactionId(xminDatum);
1557
1558 if (TransactionIdIsCurrentTransactionId(xmin))
1559 ereport(ERROR,
1560 (errcode(ERRCODE_CARDINALITY_VIOLATION),
1561 errmsg("ON CONFLICT DO UPDATE command cannot affect row a second time"),
1562 errhint("Ensure that no rows proposed for insertion within the same command have duplicate constrained values.")));
1563
1564 /* This shouldn't happen */
1565 elog(ERROR, "attempted to lock invisible tuple");
1566 break;
1567
1568 case TM_SelfModified:
1569
1570 /*
1571 * This state should never be reached. As a dirty snapshot is used
1572 * to find conflicting tuples, speculative insertion wouldn't have
1573 * seen this row to conflict with.
1574 */
1575 elog(ERROR, "unexpected self-updated tuple");
1576 break;
1577
1578 case TM_Updated:
1579 if (IsolationUsesXactSnapshot())
1580 ereport(ERROR,
1581 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1582 errmsg("could not serialize access due to concurrent update")));
1583
1584 /*
1585 * As long as we don't support an UPDATE of INSERT ON CONFLICT for
1586 * a partitioned table we shouldn't reach to a case where tuple to
1587 * be lock is moved to another partition due to concurrent update
1588 * of the partition key.
1589 */
1590 Assert(!ItemPointerIndicatesMovedPartitions(&tmfd.ctid));
1591
1592 /*
1593 * Tell caller to try again from the very start.
1594 *
1595 * It does not make sense to use the usual EvalPlanQual() style
1596 * loop here, as the new version of the row might not conflict
1597 * anymore, or the conflicting tuple has actually been deleted.
1598 */
1599 ExecClearTuple(existing);
1600 return false;
1601
1602 case TM_Deleted:
1603 if (IsolationUsesXactSnapshot())
1604 ereport(ERROR,
1605 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1606 errmsg("could not serialize access due to concurrent delete")));
1607
1608 /* see TM_Updated case */
1609 Assert(!ItemPointerIndicatesMovedPartitions(&tmfd.ctid));
1610 ExecClearTuple(existing);
1611 return false;
1612
1613 default:
1614 elog(ERROR, "unrecognized table_tuple_lock status: %u", test);
1615 }
1616
1617 /* Success, the tuple is locked. */
1618
1619 /*
1620 * Verify that the tuple is visible to our MVCC snapshot if the current
1621 * isolation level mandates that.
1622 *
1623 * It's not sufficient to rely on the check within ExecUpdate() as e.g.
1624 * CONFLICT ... WHERE clause may prevent us from reaching that.
1625 *
1626 * This means we only ever continue when a new command in the current
1627 * transaction could see the row, even though in READ COMMITTED mode the
1628 * tuple will not be visible according to the current statement's
1629 * snapshot. This is in line with the way UPDATE deals with newer tuple
1630 * versions.
1631 */
1632 ExecCheckTupleVisible(estate, relation, existing);
1633
1634 /*
1635 * Make tuple and any needed join variables available to ExecQual and
1636 * ExecProject. The EXCLUDED tuple is installed in ecxt_innertuple, while
1637 * the target's existing tuple is installed in the scantuple. EXCLUDED
1638 * has been made to reference INNER_VAR in setrefs.c, but there is no
1639 * other redirection.
1640 */
1641 econtext->ecxt_scantuple = existing;
1642 econtext->ecxt_innertuple = excludedSlot;
1643 econtext->ecxt_outertuple = NULL;
1644
1645 if (!ExecQual(onConflictSetWhere, econtext))
1646 {
1647 ExecClearTuple(existing); /* see return below */
1648 InstrCountFiltered1(&mtstate->ps, 1);
1649 return true; /* done with the tuple */
1650 }
1651
1652 if (resultRelInfo->ri_WithCheckOptions != NIL)
1653 {
1654 /*
1655 * Check target's existing tuple against UPDATE-applicable USING
1656 * security barrier quals (if any), enforced here as RLS checks/WCOs.
1657 *
1658 * The rewriter creates UPDATE RLS checks/WCOs for UPDATE security
1659 * quals, and stores them as WCOs of "kind" WCO_RLS_CONFLICT_CHECK,
1660 * but that's almost the extent of its special handling for ON
1661 * CONFLICT DO UPDATE.
1662 *
1663 * The rewriter will also have associated UPDATE applicable straight
1664 * RLS checks/WCOs for the benefit of the ExecUpdate() call that
1665 * follows. INSERTs and UPDATEs naturally have mutually exclusive WCO
1666 * kinds, so there is no danger of spurious over-enforcement in the
1667 * INSERT or UPDATE path.
1668 */
1669 ExecWithCheckOptions(WCO_RLS_CONFLICT_CHECK, resultRelInfo,
1670 existing,
1671 mtstate->ps.state);
1672 }
1673
1674 /* Project the new tuple version */
1675 ExecProject(resultRelInfo->ri_onConflict->oc_ProjInfo);
1676
1677 /*
1678 * Note that it is possible that the target tuple has been modified in
1679 * this session, after the above table_tuple_lock. We choose to not error
1680 * out in that case, in line with ExecUpdate's treatment of similar cases.
1681 * This can happen if an UPDATE is triggered from within ExecQual(),
1682 * ExecWithCheckOptions() or ExecProject() above, e.g. by selecting from a
1683 * wCTE in the ON CONFLICT's SET.
1684 */
1685
1686 /* Execute UPDATE with projection */
1687 *returning = ExecUpdate(mtstate, conflictTid, NULL,
1688 resultRelInfo->ri_onConflict->oc_ProjSlot,
1689 planSlot,
1690 &mtstate->mt_epqstate, mtstate->ps.state,
1691 canSetTag);
1692
1693 /*
1694 * Clear out existing tuple, as there might not be another conflict among
1695 * the next input rows. Don't want to hold resources till the end of the
1696 * query.
1697 */
1698 ExecClearTuple(existing);
1699 return true;
1700}
1701
1702
1703/*
1704 * Process BEFORE EACH STATEMENT triggers
1705 */
1706static void
1707fireBSTriggers(ModifyTableState *node)
1708{
1709 ModifyTable *plan = (ModifyTable *) node->ps.plan;
1710 ResultRelInfo *resultRelInfo = node->resultRelInfo;
1711
1712 /*
1713 * If the node modifies a partitioned table, we must fire its triggers.
1714 * Note that in that case, node->resultRelInfo points to the first leaf
1715 * partition, not the root table.
1716 */
1717 if (node->rootResultRelInfo != NULL)
1718 resultRelInfo = node->rootResultRelInfo;
1719
1720 switch (node->operation)
1721 {
1722 case CMD_INSERT:
1723 ExecBSInsertTriggers(node->ps.state, resultRelInfo);
1724 if (plan->onConflictAction == ONCONFLICT_UPDATE)
1725 ExecBSUpdateTriggers(node->ps.state,
1726 resultRelInfo);
1727 break;
1728 case CMD_UPDATE:
1729 ExecBSUpdateTriggers(node->ps.state, resultRelInfo);
1730 break;
1731 case CMD_DELETE:
1732 ExecBSDeleteTriggers(node->ps.state, resultRelInfo);
1733 break;
1734 default:
1735 elog(ERROR, "unknown operation");
1736 break;
1737 }
1738}
1739
1740/*
1741 * Return the target rel ResultRelInfo.
1742 *
1743 * This relation is the same as :
1744 * - the relation for which we will fire AFTER STATEMENT triggers.
1745 * - the relation into whose tuple format all captured transition tuples must
1746 * be converted.
1747 * - the root partitioned table.
1748 */
1749static ResultRelInfo *
1750getTargetResultRelInfo(ModifyTableState *node)
1751{
1752 /*
1753 * Note that if the node modifies a partitioned table, node->resultRelInfo
1754 * points to the first leaf partition, not the root table.
1755 */
1756 if (node->rootResultRelInfo != NULL)
1757 return node->rootResultRelInfo;
1758 else
1759 return node->resultRelInfo;
1760}
1761
1762/*
1763 * Process AFTER EACH STATEMENT triggers
1764 */
1765static void
1766fireASTriggers(ModifyTableState *node)
1767{
1768 ModifyTable *plan = (ModifyTable *) node->ps.plan;
1769 ResultRelInfo *resultRelInfo = getTargetResultRelInfo(node);
1770
1771 switch (node->operation)
1772 {
1773 case CMD_INSERT:
1774 if (plan->onConflictAction == ONCONFLICT_UPDATE)
1775 ExecASUpdateTriggers(node->ps.state,
1776 resultRelInfo,
1777 node->mt_oc_transition_capture);
1778 ExecASInsertTriggers(node->ps.state, resultRelInfo,
1779 node->mt_transition_capture);
1780 break;
1781 case CMD_UPDATE:
1782 ExecASUpdateTriggers(node->ps.state, resultRelInfo,
1783 node->mt_transition_capture);
1784 break;
1785 case CMD_DELETE:
1786 ExecASDeleteTriggers(node->ps.state, resultRelInfo,
1787 node->mt_transition_capture);
1788 break;
1789 default:
1790 elog(ERROR, "unknown operation");
1791 break;
1792 }
1793}
1794
1795/*
1796 * Set up the state needed for collecting transition tuples for AFTER
1797 * triggers.
1798 */
1799static void
1800ExecSetupTransitionCaptureState(ModifyTableState *mtstate, EState *estate)
1801{
1802 ModifyTable *plan = (ModifyTable *) mtstate->ps.plan;
1803 ResultRelInfo *targetRelInfo = getTargetResultRelInfo(mtstate);
1804
1805 /* Check for transition tables on the directly targeted relation. */
1806 mtstate->mt_transition_capture =
1807 MakeTransitionCaptureState(targetRelInfo->ri_TrigDesc,
1808 RelationGetRelid(targetRelInfo->ri_RelationDesc),
1809 mtstate->operation);
1810 if (plan->operation == CMD_INSERT &&
1811 plan->onConflictAction == ONCONFLICT_UPDATE)
1812 mtstate->mt_oc_transition_capture =
1813 MakeTransitionCaptureState(targetRelInfo->ri_TrigDesc,
1814 RelationGetRelid(targetRelInfo->ri_RelationDesc),
1815 CMD_UPDATE);
1816
1817 /*
1818 * If we found that we need to collect transition tuples then we may also
1819 * need tuple conversion maps for any children that have TupleDescs that
1820 * aren't compatible with the tuplestores. (We can share these maps
1821 * between the regular and ON CONFLICT cases.)
1822 */
1823 if (mtstate->mt_transition_capture != NULL ||
1824 mtstate->mt_oc_transition_capture != NULL)
1825 {
1826 ExecSetupChildParentMapForSubplan(mtstate);
1827
1828 /*
1829 * Install the conversion map for the first plan for UPDATE and DELETE
1830 * operations. It will be advanced each time we switch to the next
1831 * plan. (INSERT operations set it every time, so we need not update
1832 * mtstate->mt_oc_transition_capture here.)
1833 */
1834 if (mtstate->mt_transition_capture && mtstate->operation != CMD_INSERT)
1835 mtstate->mt_transition_capture->tcs_map =
1836 tupconv_map_for_subplan(mtstate, 0);
1837 }
1838}
1839
1840/*
1841 * ExecPrepareTupleRouting --- prepare for routing one tuple
1842 *
1843 * Determine the partition in which the tuple in slot is to be inserted,
1844 * and modify mtstate and estate to prepare for it.
1845 *
1846 * Caller must revert the estate changes after executing the insertion!
1847 * In mtstate, transition capture changes may also need to be reverted.
1848 *
1849 * Returns a slot holding the tuple of the partition rowtype.
1850 */
1851static TupleTableSlot *
1852ExecPrepareTupleRouting(ModifyTableState *mtstate,
1853 EState *estate,
1854 PartitionTupleRouting *proute,
1855 ResultRelInfo *targetRelInfo,
1856 TupleTableSlot *slot)
1857{
1858 ResultRelInfo *partrel;
1859 PartitionRoutingInfo *partrouteinfo;
1860 TupleConversionMap *map;
1861
1862 /*
1863 * Lookup the target partition's ResultRelInfo. If ExecFindPartition does
1864 * not find a valid partition for the tuple in 'slot' then an error is
1865 * raised. An error may also be raised if the found partition is not a
1866 * valid target for INSERTs. This is required since a partitioned table
1867 * UPDATE to another partition becomes a DELETE+INSERT.
1868 */
1869 partrel = ExecFindPartition(mtstate, targetRelInfo, proute, slot, estate);
1870 partrouteinfo = partrel->ri_PartitionInfo;
1871 Assert(partrouteinfo != NULL);
1872
1873 /*
1874 * Make it look like we are inserting into the partition.
1875 */
1876 estate->es_result_relation_info = partrel;
1877
1878 /*
1879 * If we're capturing transition tuples, we might need to convert from the
1880 * partition rowtype to root partitioned table's rowtype.
1881 */
1882 if (mtstate->mt_transition_capture != NULL)
1883 {
1884 if (partrel->ri_TrigDesc &&
1885 partrel->ri_TrigDesc->trig_insert_before_row)
1886 {
1887 /*
1888 * If there are any BEFORE triggers on the partition, we'll have
1889 * to be ready to convert their result back to tuplestore format.
1890 */
1891 mtstate->mt_transition_capture->tcs_original_insert_tuple = NULL;
1892 mtstate->mt_transition_capture->tcs_map =
1893 partrouteinfo->pi_PartitionToRootMap;
1894 }
1895 else
1896 {
1897 /*
1898 * Otherwise, just remember the original unconverted tuple, to
1899 * avoid a needless round trip conversion.
1900 */
1901 mtstate->mt_transition_capture->tcs_original_insert_tuple = slot;
1902 mtstate->mt_transition_capture->tcs_map = NULL;
1903 }
1904 }
1905 if (mtstate->mt_oc_transition_capture != NULL)
1906 {
1907 mtstate->mt_oc_transition_capture->tcs_map =
1908 partrouteinfo->pi_PartitionToRootMap;
1909 }
1910
1911 /*
1912 * Convert the tuple, if necessary.
1913 */
1914 map = partrouteinfo->pi_RootToPartitionMap;
1915 if (map != NULL)
1916 {
1917 TupleTableSlot *new_slot = partrouteinfo->pi_PartitionTupleSlot;
1918
1919 slot = execute_attr_map_slot(map->attrMap, slot, new_slot);
1920 }
1921
1922 return slot;
1923}
1924
1925/*
1926 * Initialize the child-to-root tuple conversion map array for UPDATE subplans.
1927 *
1928 * This map array is required to convert the tuple from the subplan result rel
1929 * to the target table descriptor. This requirement arises for two independent
1930 * scenarios:
1931 * 1. For update-tuple-routing.
1932 * 2. For capturing tuples in transition tables.
1933 */
1934static void
1935ExecSetupChildParentMapForSubplan(ModifyTableState *mtstate)
1936{
1937 ResultRelInfo *targetRelInfo = getTargetResultRelInfo(mtstate);
1938 ResultRelInfo *resultRelInfos = mtstate->resultRelInfo;
1939 TupleDesc outdesc;
1940 int numResultRelInfos = mtstate->mt_nplans;
1941 int i;
1942
1943 /*
1944 * Build array of conversion maps from each child's TupleDesc to the one
1945 * used in the target relation. The map pointers may be NULL when no
1946 * conversion is necessary, which is hopefully a common case.
1947 */
1948
1949 /* Get tuple descriptor of the target rel. */
1950 outdesc = RelationGetDescr(targetRelInfo->ri_RelationDesc);
1951
1952 mtstate->mt_per_subplan_tupconv_maps = (TupleConversionMap **)
1953 palloc(sizeof(TupleConversionMap *) * numResultRelInfos);
1954
1955 for (i = 0; i < numResultRelInfos; ++i)
1956 {
1957 mtstate->mt_per_subplan_tupconv_maps[i] =
1958 convert_tuples_by_name(RelationGetDescr(resultRelInfos[i].ri_RelationDesc),
1959 outdesc,
1960 gettext_noop("could not convert row type"));
1961 }
1962}
1963
1964/*
1965 * For a given subplan index, get the tuple conversion map.
1966 */
1967static TupleConversionMap *
1968tupconv_map_for_subplan(ModifyTableState *mtstate, int whichplan)
1969{
1970 /* If nobody else set the per-subplan array of maps, do so ourselves. */
1971 if (mtstate->mt_per_subplan_tupconv_maps == NULL)
1972 ExecSetupChildParentMapForSubplan(mtstate);
1973
1974 Assert(whichplan >= 0 && whichplan < mtstate->mt_nplans);
1975 return mtstate->mt_per_subplan_tupconv_maps[whichplan];
1976}
1977
1978/* ----------------------------------------------------------------
1979 * ExecModifyTable
1980 *
1981 * Perform table modifications as required, and return RETURNING results
1982 * if needed.
1983 * ----------------------------------------------------------------
1984 */
1985static TupleTableSlot *
1986ExecModifyTable(PlanState *pstate)
1987{
1988 ModifyTableState *node = castNode(ModifyTableState, pstate);
1989 PartitionTupleRouting *proute = node->mt_partition_tuple_routing;
1990 EState *estate = node->ps.state;
1991 CmdType operation = node->operation;
1992 ResultRelInfo *saved_resultRelInfo;
1993 ResultRelInfo *resultRelInfo;
1994 PlanState *subplanstate;
1995 JunkFilter *junkfilter;
1996 TupleTableSlot *slot;
1997 TupleTableSlot *planSlot;
1998 ItemPointer tupleid;
1999 ItemPointerData tuple_ctid;
2000 HeapTupleData oldtupdata;
2001 HeapTuple oldtuple;
2002
2003 CHECK_FOR_INTERRUPTS();
2004
2005 /*
2006 * This should NOT get called during EvalPlanQual; we should have passed a
2007 * subplan tree to EvalPlanQual, instead. Use a runtime test not just
2008 * Assert because this condition is easy to miss in testing. (Note:
2009 * although ModifyTable should not get executed within an EvalPlanQual
2010 * operation, we do have to allow it to be initialized and shut down in
2011 * case it is within a CTE subplan. Hence this test must be here, not in
2012 * ExecInitModifyTable.)
2013 */
2014 if (estate->es_epq_active != NULL)
2015 elog(ERROR, "ModifyTable should not be called during EvalPlanQual");
2016
2017 /*
2018 * If we've already completed processing, don't try to do more. We need
2019 * this test because ExecPostprocessPlan might call us an extra time, and
2020 * our subplan's nodes aren't necessarily robust against being called
2021 * extra times.
2022 */
2023 if (node->mt_done)
2024 return NULL;
2025
2026 /*
2027 * On first call, fire BEFORE STATEMENT triggers before proceeding.
2028 */
2029 if (node->fireBSTriggers)
2030 {
2031 fireBSTriggers(node);
2032 node->fireBSTriggers = false;
2033 }
2034
2035 /* Preload local variables */
2036 resultRelInfo = node->resultRelInfo + node->mt_whichplan;
2037 subplanstate = node->mt_plans[node->mt_whichplan];
2038 junkfilter = resultRelInfo->ri_junkFilter;
2039
2040 /*
2041 * es_result_relation_info must point to the currently active result
2042 * relation while we are within this ModifyTable node. Even though
2043 * ModifyTable nodes can't be nested statically, they can be nested
2044 * dynamically (since our subplan could include a reference to a modifying
2045 * CTE). So we have to save and restore the caller's value.
2046 */
2047 saved_resultRelInfo = estate->es_result_relation_info;
2048
2049 estate->es_result_relation_info = resultRelInfo;
2050
2051 /*
2052 * Fetch rows from subplan(s), and execute the required table modification
2053 * for each row.
2054 */
2055 for (;;)
2056 {
2057 /*
2058 * Reset the per-output-tuple exprcontext. This is needed because
2059 * triggers expect to use that context as workspace. It's a bit ugly
2060 * to do this below the top level of the plan, however. We might need
2061 * to rethink this later.
2062 */
2063 ResetPerTupleExprContext(estate);
2064
2065 /*
2066 * Reset per-tuple memory context used for processing on conflict and
2067 * returning clauses, to free any expression evaluation storage
2068 * allocated in the previous cycle.
2069 */
2070 if (pstate->ps_ExprContext)
2071 ResetExprContext(pstate->ps_ExprContext);
2072
2073 planSlot = ExecProcNode(subplanstate);
2074
2075 if (TupIsNull(planSlot))
2076 {
2077 /* advance to next subplan if any */
2078 node->mt_whichplan++;
2079 if (node->mt_whichplan < node->mt_nplans)
2080 {
2081 resultRelInfo++;
2082 subplanstate = node->mt_plans[node->mt_whichplan];
2083 junkfilter = resultRelInfo->ri_junkFilter;
2084 estate->es_result_relation_info = resultRelInfo;
2085 EvalPlanQualSetPlan(&node->mt_epqstate, subplanstate->plan,
2086 node->mt_arowmarks[node->mt_whichplan]);
2087 /* Prepare to convert transition tuples from this child. */
2088 if (node->mt_transition_capture != NULL)
2089 {
2090 node->mt_transition_capture->tcs_map =
2091 tupconv_map_for_subplan(node, node->mt_whichplan);
2092 }
2093 if (node->mt_oc_transition_capture != NULL)
2094 {
2095 node->mt_oc_transition_capture->tcs_map =
2096 tupconv_map_for_subplan(node, node->mt_whichplan);
2097 }
2098 continue;
2099 }
2100 else
2101 break;
2102 }
2103
2104 /*
2105 * Ensure input tuple is the right format for the target relation.
2106 */
2107 if (node->mt_scans[node->mt_whichplan]->tts_ops != planSlot->tts_ops)
2108 {
2109 ExecCopySlot(node->mt_scans[node->mt_whichplan], planSlot);
2110 planSlot = node->mt_scans[node->mt_whichplan];
2111 }
2112
2113 /*
2114 * If resultRelInfo->ri_usesFdwDirectModify is true, all we need to do
2115 * here is compute the RETURNING expressions.
2116 */
2117 if (resultRelInfo->ri_usesFdwDirectModify)
2118 {
2119 Assert(resultRelInfo->ri_projectReturning);
2120
2121 /*
2122 * A scan slot containing the data that was actually inserted,
2123 * updated or deleted has already been made available to
2124 * ExecProcessReturning by IterateDirectModify, so no need to
2125 * provide it here.
2126 */
2127 slot = ExecProcessReturning(resultRelInfo, NULL, planSlot);
2128
2129 estate->es_result_relation_info = saved_resultRelInfo;
2130 return slot;
2131 }
2132
2133 EvalPlanQualSetSlot(&node->mt_epqstate, planSlot);
2134 slot = planSlot;
2135
2136 tupleid = NULL;
2137 oldtuple = NULL;
2138 if (junkfilter != NULL)
2139 {
2140 /*
2141 * extract the 'ctid' or 'wholerow' junk attribute.
2142 */
2143 if (operation == CMD_UPDATE || operation == CMD_DELETE)
2144 {
2145 char relkind;
2146 Datum datum;
2147 bool isNull;
2148
2149 relkind = resultRelInfo->ri_RelationDesc->rd_rel->relkind;
2150 if (relkind == RELKIND_RELATION || relkind == RELKIND_MATVIEW)
2151 {
2152 datum = ExecGetJunkAttribute(slot,
2153 junkfilter->jf_junkAttNo,
2154 &isNull);
2155 /* shouldn't ever get a null result... */
2156 if (isNull)
2157 elog(ERROR, "ctid is NULL");
2158
2159 tupleid = (ItemPointer) DatumGetPointer(datum);
2160 tuple_ctid = *tupleid; /* be sure we don't free ctid!! */
2161 tupleid = &tuple_ctid;
2162 }
2163
2164 /*
2165 * Use the wholerow attribute, when available, to reconstruct
2166 * the old relation tuple.
2167 *
2168 * Foreign table updates have a wholerow attribute when the
2169 * relation has a row-level trigger. Note that the wholerow
2170 * attribute does not carry system columns. Foreign table
2171 * triggers miss seeing those, except that we know enough here
2172 * to set t_tableOid. Quite separately from this, the FDW may
2173 * fetch its own junk attrs to identify the row.
2174 *
2175 * Other relevant relkinds, currently limited to views, always
2176 * have a wholerow attribute.
2177 */
2178 else if (AttributeNumberIsValid(junkfilter->jf_junkAttNo))
2179 {
2180 datum = ExecGetJunkAttribute(slot,
2181 junkfilter->jf_junkAttNo,
2182 &isNull);
2183 /* shouldn't ever get a null result... */
2184 if (isNull)
2185 elog(ERROR, "wholerow is NULL");
2186
2187 oldtupdata.t_data = DatumGetHeapTupleHeader(datum);
2188 oldtupdata.t_len =
2189 HeapTupleHeaderGetDatumLength(oldtupdata.t_data);
2190 ItemPointerSetInvalid(&(oldtupdata.t_self));
2191 /* Historically, view triggers see invalid t_tableOid. */
2192 oldtupdata.t_tableOid =
2193 (relkind == RELKIND_VIEW) ? InvalidOid :
2194 RelationGetRelid(resultRelInfo->ri_RelationDesc);
2195
2196 oldtuple = &oldtupdata;
2197 }
2198 else
2199 Assert(relkind == RELKIND_FOREIGN_TABLE);
2200 }
2201
2202 /*
2203 * apply the junkfilter if needed.
2204 */
2205 if (operation != CMD_DELETE)
2206 slot = ExecFilterJunk(junkfilter, slot);
2207 }
2208
2209 switch (operation)
2210 {
2211 case CMD_INSERT:
2212 /* Prepare for tuple routing if needed. */
2213 if (proute)
2214 slot = ExecPrepareTupleRouting(node, estate, proute,
2215 resultRelInfo, slot);
2216 slot = ExecInsert(node, slot, planSlot,
2217 estate, node->canSetTag);
2218 /* Revert ExecPrepareTupleRouting's state change. */
2219 if (proute)
2220 estate->es_result_relation_info = resultRelInfo;
2221 break;
2222 case CMD_UPDATE:
2223 slot = ExecUpdate(node, tupleid, oldtuple, slot, planSlot,
2224 &node->mt_epqstate, estate, node->canSetTag);
2225 break;
2226 case CMD_DELETE:
2227 slot = ExecDelete(node, tupleid, oldtuple, planSlot,
2228 &node->mt_epqstate, estate,
2229 true, node->canSetTag,
2230 false /* changingPart */ , NULL, NULL);
2231 break;
2232 default:
2233 elog(ERROR, "unknown operation");
2234 break;
2235 }
2236
2237 /*
2238 * If we got a RETURNING result, return it to caller. We'll continue
2239 * the work on next call.
2240 */
2241 if (slot)
2242 {
2243 estate->es_result_relation_info = saved_resultRelInfo;
2244 return slot;
2245 }
2246 }
2247
2248 /* Restore es_result_relation_info before exiting */
2249 estate->es_result_relation_info = saved_resultRelInfo;
2250
2251 /*
2252 * We're done, but fire AFTER STATEMENT triggers before exiting.
2253 */
2254 fireASTriggers(node);
2255
2256 node->mt_done = true;
2257
2258 return NULL;
2259}
2260
2261/* ----------------------------------------------------------------
2262 * ExecInitModifyTable
2263 * ----------------------------------------------------------------
2264 */
2265ModifyTableState *
2266ExecInitModifyTable(ModifyTable *node, EState *estate, int eflags)
2267{
2268 ModifyTableState *mtstate;
2269 CmdType operation = node->operation;
2270 int nplans = list_length(node->plans);
2271 ResultRelInfo *saved_resultRelInfo;
2272 ResultRelInfo *resultRelInfo;
2273 Plan *subplan;
2274 ListCell *l;
2275 int i;
2276 Relation rel;
2277 bool update_tuple_routing_needed = node->partColsUpdated;
2278
2279 /* check for unsupported flags */
2280 Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
2281
2282 /*
2283 * create state structure
2284 */
2285 mtstate = makeNode(ModifyTableState);
2286 mtstate->ps.plan = (Plan *) node;
2287 mtstate->ps.state = estate;
2288 mtstate->ps.ExecProcNode = ExecModifyTable;
2289
2290 mtstate->operation = operation;
2291 mtstate->canSetTag = node->canSetTag;
2292 mtstate->mt_done = false;
2293
2294 mtstate->mt_plans = (PlanState **) palloc0(sizeof(PlanState *) * nplans);
2295 mtstate->resultRelInfo = estate->es_result_relations + node->resultRelIndex;
2296 mtstate->mt_scans = (TupleTableSlot **) palloc0(sizeof(TupleTableSlot *) * nplans);
2297
2298 /* If modifying a partitioned table, initialize the root table info */
2299 if (node->rootResultRelIndex >= 0)
2300 mtstate->rootResultRelInfo = estate->es_root_result_relations +
2301 node->rootResultRelIndex;
2302
2303 mtstate->mt_arowmarks = (List **) palloc0(sizeof(List *) * nplans);
2304 mtstate->mt_nplans = nplans;
2305
2306 /* set up epqstate with dummy subplan data for the moment */
2307 EvalPlanQualInit(&mtstate->mt_epqstate, estate, NULL, NIL, node->epqParam);
2308 mtstate->fireBSTriggers = true;
2309
2310 /*
2311 * call ExecInitNode on each of the plans to be executed and save the
2312 * results into the array "mt_plans". This is also a convenient place to
2313 * verify that the proposed target relations are valid and open their
2314 * indexes for insertion of new index entries. Note we *must* set
2315 * estate->es_result_relation_info correctly while we initialize each
2316 * sub-plan; external modules such as FDWs may depend on that (see
2317 * contrib/postgres_fdw/postgres_fdw.c: postgresBeginDirectModify() as one
2318 * example).
2319 */
2320 saved_resultRelInfo = estate->es_result_relation_info;
2321
2322 resultRelInfo = mtstate->resultRelInfo;
2323 i = 0;
2324 foreach(l, node->plans)
2325 {
2326 subplan = (Plan *) lfirst(l);
2327
2328 /* Initialize the usesFdwDirectModify flag */
2329 resultRelInfo->ri_usesFdwDirectModify = bms_is_member(i,
2330 node->fdwDirectModifyPlans);
2331
2332 /*
2333 * Verify result relation is a valid target for the current operation
2334 */
2335 CheckValidResultRel(resultRelInfo, operation);
2336
2337 /*
2338 * If there are indices on the result relation, open them and save
2339 * descriptors in the result relation info, so that we can add new
2340 * index entries for the tuples we add/update. We need not do this
2341 * for a DELETE, however, since deletion doesn't affect indexes. Also,
2342 * inside an EvalPlanQual operation, the indexes might be open
2343 * already, since we share the resultrel state with the original
2344 * query.
2345 */
2346 if (resultRelInfo->ri_RelationDesc->rd_rel->relhasindex &&
2347 operation != CMD_DELETE &&
2348 resultRelInfo->ri_IndexRelationDescs == NULL)
2349 ExecOpenIndices(resultRelInfo,
2350 node->onConflictAction != ONCONFLICT_NONE);
2351
2352 /*
2353 * If this is an UPDATE and a BEFORE UPDATE trigger is present, the
2354 * trigger itself might modify the partition-key values. So arrange
2355 * for tuple routing.
2356 */
2357 if (resultRelInfo->ri_TrigDesc &&
2358 resultRelInfo->ri_TrigDesc->trig_update_before_row &&
2359 operation == CMD_UPDATE)
2360 update_tuple_routing_needed = true;
2361
2362 /* Now init the plan for this result rel */
2363 estate->es_result_relation_info = resultRelInfo;
2364 mtstate->mt_plans[i] = ExecInitNode(subplan, estate, eflags);
2365 mtstate->mt_scans[i] =
2366 ExecInitExtraTupleSlot(mtstate->ps.state, ExecGetResultType(mtstate->mt_plans[i]),
2367 table_slot_callbacks(resultRelInfo->ri_RelationDesc));
2368
2369 /* Also let FDWs init themselves for foreign-table result rels */
2370 if (!resultRelInfo->ri_usesFdwDirectModify &&
2371 resultRelInfo->ri_FdwRoutine != NULL &&
2372 resultRelInfo->ri_FdwRoutine->BeginForeignModify != NULL)
2373 {
2374 List *fdw_private = (List *) list_nth(node->fdwPrivLists, i);
2375
2376 resultRelInfo->ri_FdwRoutine->BeginForeignModify(mtstate,
2377 resultRelInfo,
2378 fdw_private,
2379 i,
2380 eflags);
2381 }
2382
2383 resultRelInfo++;
2384 i++;
2385 }
2386
2387 estate->es_result_relation_info = saved_resultRelInfo;
2388
2389 /* Get the target relation */
2390 rel = (getTargetResultRelInfo(mtstate))->ri_RelationDesc;
2391
2392 /*
2393 * If it's not a partitioned table after all, UPDATE tuple routing should
2394 * not be attempted.
2395 */
2396 if (rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
2397 update_tuple_routing_needed = false;
2398
2399 /*
2400 * Build state for tuple routing if it's an INSERT or if it's an UPDATE of
2401 * partition key.
2402 */
2403 if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE &&
2404 (operation == CMD_INSERT || update_tuple_routing_needed))
2405 mtstate->mt_partition_tuple_routing =
2406 ExecSetupPartitionTupleRouting(estate, mtstate, rel);
2407
2408 /*
2409 * Build state for collecting transition tuples. This requires having a
2410 * valid trigger query context, so skip it in explain-only mode.
2411 */
2412 if (!(eflags & EXEC_FLAG_EXPLAIN_ONLY))
2413 ExecSetupTransitionCaptureState(mtstate, estate);
2414
2415 /*
2416 * Construct mapping from each of the per-subplan partition attnos to the
2417 * root attno. This is required when during update row movement the tuple
2418 * descriptor of a source partition does not match the root partitioned
2419 * table descriptor. In such a case we need to convert tuples to the root
2420 * tuple descriptor, because the search for destination partition starts
2421 * from the root. We'll also need a slot to store these converted tuples.
2422 * We can skip this setup if it's not a partition key update.
2423 */
2424 if (update_tuple_routing_needed)
2425 {
2426 ExecSetupChildParentMapForSubplan(mtstate);
2427 mtstate->mt_root_tuple_slot = table_slot_create(rel, NULL);
2428 }
2429
2430 /*
2431 * Initialize any WITH CHECK OPTION constraints if needed.
2432 */
2433 resultRelInfo = mtstate->resultRelInfo;
2434 i = 0;
2435 foreach(l, node->withCheckOptionLists)
2436 {
2437 List *wcoList = (List *) lfirst(l);
2438 List *wcoExprs = NIL;
2439 ListCell *ll;
2440
2441 foreach(ll, wcoList)
2442 {
2443 WithCheckOption *wco = (WithCheckOption *) lfirst(ll);
2444 ExprState *wcoExpr = ExecInitQual((List *) wco->qual,
2445 &mtstate->ps);
2446
2447 wcoExprs = lappend(wcoExprs, wcoExpr);
2448 }
2449
2450 resultRelInfo->ri_WithCheckOptions = wcoList;
2451 resultRelInfo->ri_WithCheckOptionExprs = wcoExprs;
2452 resultRelInfo++;
2453 i++;
2454 }
2455
2456 /*
2457 * Initialize RETURNING projections if needed.
2458 */
2459 if (node->returningLists)
2460 {
2461 TupleTableSlot *slot;
2462 ExprContext *econtext;
2463
2464 /*
2465 * Initialize result tuple slot and assign its rowtype using the first
2466 * RETURNING list. We assume the rest will look the same.
2467 */
2468 mtstate->ps.plan->targetlist = (List *) linitial(node->returningLists);
2469
2470 /* Set up a slot for the output of the RETURNING projection(s) */
2471 ExecInitResultTupleSlotTL(&mtstate->ps, &TTSOpsVirtual);
2472 slot = mtstate->ps.ps_ResultTupleSlot;
2473
2474 /* Need an econtext too */
2475 if (mtstate->ps.ps_ExprContext == NULL)
2476 ExecAssignExprContext(estate, &mtstate->ps);
2477 econtext = mtstate->ps.ps_ExprContext;
2478
2479 /*
2480 * Build a projection for each result rel.
2481 */
2482 resultRelInfo = mtstate->resultRelInfo;
2483 foreach(l, node->returningLists)
2484 {
2485 List *rlist = (List *) lfirst(l);
2486
2487 resultRelInfo->ri_returningList = rlist;
2488 resultRelInfo->ri_projectReturning =
2489 ExecBuildProjectionInfo(rlist, econtext, slot, &mtstate->ps,
2490 resultRelInfo->ri_RelationDesc->rd_att);
2491 resultRelInfo++;
2492 }
2493 }
2494 else
2495 {
2496 /*
2497 * We still must construct a dummy result tuple type, because InitPlan
2498 * expects one (maybe should change that?).
2499 */
2500 mtstate->ps.plan->targetlist = NIL;
2501 ExecInitResultTypeTL(&mtstate->ps);
2502
2503 mtstate->ps.ps_ExprContext = NULL;
2504 }
2505
2506 /* Set the list of arbiter indexes if needed for ON CONFLICT */
2507 resultRelInfo = mtstate->resultRelInfo;
2508 if (node->onConflictAction != ONCONFLICT_NONE)
2509 resultRelInfo->ri_onConflictArbiterIndexes = node->arbiterIndexes;
2510
2511 /*
2512 * If needed, Initialize target list, projection and qual for ON CONFLICT
2513 * DO UPDATE.
2514 */
2515 if (node->onConflictAction == ONCONFLICT_UPDATE)
2516 {
2517 ExprContext *econtext;
2518 TupleDesc relationDesc;
2519 TupleDesc tupDesc;
2520
2521 /* insert may only have one plan, inheritance is not expanded */
2522 Assert(nplans == 1);
2523
2524 /* already exists if created by RETURNING processing above */
2525 if (mtstate->ps.ps_ExprContext == NULL)
2526 ExecAssignExprContext(estate, &mtstate->ps);
2527
2528 econtext = mtstate->ps.ps_ExprContext;
2529 relationDesc = resultRelInfo->ri_RelationDesc->rd_att;
2530
2531 /* carried forward solely for the benefit of explain */
2532 mtstate->mt_excludedtlist = node->exclRelTlist;
2533
2534 /* create state for DO UPDATE SET operation */
2535 resultRelInfo->ri_onConflict = makeNode(OnConflictSetState);
2536
2537 /* initialize slot for the existing tuple */
2538 resultRelInfo->ri_onConflict->oc_Existing =
2539 table_slot_create(resultRelInfo->ri_RelationDesc,
2540 &mtstate->ps.state->es_tupleTable);
2541
2542 /*
2543 * Create the tuple slot for the UPDATE SET projection. We want a slot
2544 * of the table's type here, because the slot will be used to insert
2545 * into the table, and for RETURNING processing - which may access
2546 * system attributes.
2547 */
2548 tupDesc = ExecTypeFromTL((List *) node->onConflictSet);
2549 resultRelInfo->ri_onConflict->oc_ProjSlot =
2550 ExecInitExtraTupleSlot(mtstate->ps.state, tupDesc,
2551 table_slot_callbacks(resultRelInfo->ri_RelationDesc));
2552
2553 /* build UPDATE SET projection state */
2554 resultRelInfo->ri_onConflict->oc_ProjInfo =
2555 ExecBuildProjectionInfo(node->onConflictSet, econtext,
2556 resultRelInfo->ri_onConflict->oc_ProjSlot,
2557 &mtstate->ps,
2558 relationDesc);
2559
2560 /* initialize state to evaluate the WHERE clause, if any */
2561 if (node->onConflictWhere)
2562 {
2563 ExprState *qualexpr;
2564
2565 qualexpr = ExecInitQual((List *) node->onConflictWhere,
2566 &mtstate->ps);
2567 resultRelInfo->ri_onConflict->oc_WhereClause = qualexpr;
2568 }
2569 }
2570
2571 /*
2572 * If we have any secondary relations in an UPDATE or DELETE, they need to
2573 * be treated like non-locked relations in SELECT FOR UPDATE, ie, the
2574 * EvalPlanQual mechanism needs to be told about them. Locate the
2575 * relevant ExecRowMarks.
2576 */
2577 foreach(l, node->rowMarks)
2578 {
2579 PlanRowMark *rc = lfirst_node(PlanRowMark, l);
2580 ExecRowMark *erm;
2581
2582 /* ignore "parent" rowmarks; they are irrelevant at runtime */
2583 if (rc->isParent)
2584 continue;
2585
2586 /* find ExecRowMark (same for all subplans) */
2587 erm = ExecFindRowMark(estate, rc->rti, false);
2588
2589 /* build ExecAuxRowMark for each subplan */
2590 for (i = 0; i < nplans; i++)
2591 {
2592 ExecAuxRowMark *aerm;
2593
2594 subplan = mtstate->mt_plans[i]->plan;
2595 aerm = ExecBuildAuxRowMark(erm, subplan->targetlist);
2596 mtstate->mt_arowmarks[i] = lappend(mtstate->mt_arowmarks[i], aerm);
2597 }
2598 }
2599
2600 /* select first subplan */
2601 mtstate->mt_whichplan = 0;
2602 subplan = (Plan *) linitial(node->plans);
2603 EvalPlanQualSetPlan(&mtstate->mt_epqstate, subplan,
2604 mtstate->mt_arowmarks[0]);
2605
2606 /*
2607 * Initialize the junk filter(s) if needed. INSERT queries need a filter
2608 * if there are any junk attrs in the tlist. UPDATE and DELETE always
2609 * need a filter, since there's always at least one junk attribute present
2610 * --- no need to look first. Typically, this will be a 'ctid' or
2611 * 'wholerow' attribute, but in the case of a foreign data wrapper it
2612 * might be a set of junk attributes sufficient to identify the remote
2613 * row.
2614 *
2615 * If there are multiple result relations, each one needs its own junk
2616 * filter. Note multiple rels are only possible for UPDATE/DELETE, so we
2617 * can't be fooled by some needing a filter and some not.
2618 *
2619 * This section of code is also a convenient place to verify that the
2620 * output of an INSERT or UPDATE matches the target table(s).
2621 */
2622 {
2623 bool junk_filter_needed = false;
2624
2625 switch (operation)
2626 {
2627 case CMD_INSERT:
2628 foreach(l, subplan->targetlist)
2629 {
2630 TargetEntry *tle = (TargetEntry *) lfirst(l);
2631
2632 if (tle->resjunk)
2633 {
2634 junk_filter_needed = true;
2635 break;
2636 }
2637 }
2638 break;
2639 case CMD_UPDATE:
2640 case CMD_DELETE:
2641 junk_filter_needed = true;
2642 break;
2643 default:
2644 elog(ERROR, "unknown operation");
2645 break;
2646 }
2647
2648 if (junk_filter_needed)
2649 {
2650 resultRelInfo = mtstate->resultRelInfo;
2651 for (i = 0; i < nplans; i++)
2652 {
2653 JunkFilter *j;
2654 TupleTableSlot *junkresslot;
2655
2656 subplan = mtstate->mt_plans[i]->plan;
2657 if (operation == CMD_INSERT || operation == CMD_UPDATE)
2658 ExecCheckPlanOutput(resultRelInfo->ri_RelationDesc,
2659 subplan->targetlist);
2660
2661 junkresslot =
2662 ExecInitExtraTupleSlot(estate, NULL,
2663 table_slot_callbacks(resultRelInfo->ri_RelationDesc));
2664 j = ExecInitJunkFilter(subplan->targetlist,
2665 junkresslot);
2666
2667 if (operation == CMD_UPDATE || operation == CMD_DELETE)
2668 {
2669 /* For UPDATE/DELETE, find the appropriate junk attr now */
2670 char relkind;
2671
2672 relkind = resultRelInfo->ri_RelationDesc->rd_rel->relkind;
2673 if (relkind == RELKIND_RELATION ||
2674 relkind == RELKIND_MATVIEW ||
2675 relkind == RELKIND_PARTITIONED_TABLE)
2676 {
2677 j->jf_junkAttNo = ExecFindJunkAttribute(j, "ctid");
2678 if (!AttributeNumberIsValid(j->jf_junkAttNo))
2679 elog(ERROR, "could not find junk ctid column");
2680 }
2681 else if (relkind == RELKIND_FOREIGN_TABLE)
2682 {
2683 /*
2684 * When there is a row-level trigger, there should be
2685 * a wholerow attribute.
2686 */
2687 j->jf_junkAttNo = ExecFindJunkAttribute(j, "wholerow");
2688 }
2689 else
2690 {
2691 j->jf_junkAttNo = ExecFindJunkAttribute(j, "wholerow");
2692 if (!AttributeNumberIsValid(j->jf_junkAttNo))
2693 elog(ERROR, "could not find junk wholerow column");
2694 }
2695 }
2696
2697 resultRelInfo->ri_junkFilter = j;
2698 resultRelInfo++;
2699 }
2700 }
2701 else
2702 {
2703 if (operation == CMD_INSERT)
2704 ExecCheckPlanOutput(mtstate->resultRelInfo->ri_RelationDesc,
2705 subplan->targetlist);
2706 }
2707 }
2708
2709 /*
2710 * Lastly, if this is not the primary (canSetTag) ModifyTable node, add it
2711 * to estate->es_auxmodifytables so that it will be run to completion by
2712 * ExecPostprocessPlan. (It'd actually work fine to add the primary
2713 * ModifyTable node too, but there's no need.) Note the use of lcons not
2714 * lappend: we need later-initialized ModifyTable nodes to be shut down
2715 * before earlier ones. This ensures that we don't throw away RETURNING
2716 * rows that need to be seen by a later CTE subplan.
2717 */
2718 if (!mtstate->canSetTag)
2719 estate->es_auxmodifytables = lcons(mtstate,
2720 estate->es_auxmodifytables);
2721
2722 return mtstate;
2723}
2724
2725/* ----------------------------------------------------------------
2726 * ExecEndModifyTable
2727 *
2728 * Shuts down the plan.
2729 *
2730 * Returns nothing of interest.
2731 * ----------------------------------------------------------------
2732 */
2733void
2734ExecEndModifyTable(ModifyTableState *node)
2735{
2736 int i;
2737
2738 /*
2739 * Allow any FDWs to shut down
2740 */
2741 for (i = 0; i < node->mt_nplans; i++)
2742 {
2743 ResultRelInfo *resultRelInfo = node->resultRelInfo + i;
2744
2745 if (!resultRelInfo->ri_usesFdwDirectModify &&
2746 resultRelInfo->ri_FdwRoutine != NULL &&
2747 resultRelInfo->ri_FdwRoutine->EndForeignModify != NULL)
2748 resultRelInfo->ri_FdwRoutine->EndForeignModify(node->ps.state,
2749 resultRelInfo);
2750 }
2751
2752 /*
2753 * Close all the partitioned tables, leaf partitions, and their indices
2754 * and release the slot used for tuple routing, if set.
2755 */
2756 if (node->mt_partition_tuple_routing)
2757 {
2758 ExecCleanupTupleRouting(node, node->mt_partition_tuple_routing);
2759
2760 if (node->mt_root_tuple_slot)
2761 ExecDropSingleTupleTableSlot(node->mt_root_tuple_slot);
2762 }
2763
2764 /*
2765 * Free the exprcontext
2766 */
2767 ExecFreeExprContext(&node->ps);
2768
2769 /*
2770 * clean out the tuple table
2771 */
2772 if (node->ps.ps_ResultTupleSlot)
2773 ExecClearTuple(node->ps.ps_ResultTupleSlot);
2774
2775 /*
2776 * Terminate EPQ execution if active
2777 */
2778 EvalPlanQualEnd(&node->mt_epqstate);
2779
2780 /*
2781 * shut down subplans
2782 */
2783 for (i = 0; i < node->mt_nplans; i++)
2784 ExecEndNode(node->mt_plans[i]);
2785}
2786
2787void
2788ExecReScanModifyTable(ModifyTableState *node)
2789{
2790 /*
2791 * Currently, we don't need to support rescan on ModifyTable nodes. The
2792 * semantics of that would be a bit debatable anyway.
2793 */
2794 elog(ERROR, "ExecReScanModifyTable is not implemented");
2795}
2796