1 /*-------------------------------------------------------------------------
2 *
3 * nodeModifyTable.c
4 * routines to handle ModifyTable nodes.
5 *
6 * Portions Copyright (c) 1996-2021, 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 * The ModifyTable node receives input from its outerPlan, which is
23 * the data to insert for INSERT cases, or the changed columns' new
24 * values plus row-locating info for UPDATE cases, or just the
25 * row-locating info for DELETE cases.
26 *
27 * If the query specifies RETURNING, then the ModifyTable returns a
28 * RETURNING tuple after completing each row insert, update, or delete.
29 * It must be called again to continue the operation. Without RETURNING,
30 * we just loop within the node until all the work is done, then
31 * return NULL. This avoids useless call/return overhead.
32 */
33
34 #include "postgres.h"
35
36 #include "access/heapam.h"
37 #include "access/htup_details.h"
38 #include "access/tableam.h"
39 #include "access/xact.h"
40 #include "catalog/catalog.h"
41 #include "commands/trigger.h"
42 #include "executor/execPartition.h"
43 #include "executor/executor.h"
44 #include "executor/nodeModifyTable.h"
45 #include "foreign/fdwapi.h"
46 #include "miscadmin.h"
47 #include "nodes/nodeFuncs.h"
48 #include "rewrite/rewriteHandler.h"
49 #include "storage/bufmgr.h"
50 #include "storage/lmgr.h"
51 #include "utils/builtins.h"
52 #include "utils/datum.h"
53 #include "utils/memutils.h"
54 #include "utils/rel.h"
55
56
57 typedef struct MTTargetRelLookup
58 {
59 Oid relationOid; /* hash key, must be first */
60 int relationIndex; /* rel's index in resultRelInfo[] array */
61 } MTTargetRelLookup;
62
63 static void ExecBatchInsert(ModifyTableState *mtstate,
64 ResultRelInfo *resultRelInfo,
65 TupleTableSlot **slots,
66 TupleTableSlot **planSlots,
67 int numSlots,
68 EState *estate,
69 bool canSetTag);
70 static bool ExecOnConflictUpdate(ModifyTableState *mtstate,
71 ResultRelInfo *resultRelInfo,
72 ItemPointer conflictTid,
73 TupleTableSlot *planSlot,
74 TupleTableSlot *excludedSlot,
75 EState *estate,
76 bool canSetTag,
77 TupleTableSlot **returning);
78 static TupleTableSlot *ExecPrepareTupleRouting(ModifyTableState *mtstate,
79 EState *estate,
80 PartitionTupleRouting *proute,
81 ResultRelInfo *targetRelInfo,
82 TupleTableSlot *slot,
83 ResultRelInfo **partRelInfo);
84
85 /*
86 * Verify that the tuples to be produced by INSERT match the
87 * target relation's rowtype
88 *
89 * We do this to guard against stale plans. If plan invalidation is
90 * functioning properly then we should never get a failure here, but better
91 * safe than sorry. Note that this is called after we have obtained lock
92 * on the target rel, so the rowtype can't change underneath us.
93 *
94 * The plan output is represented by its targetlist, because that makes
95 * handling the dropped-column case easier.
96 *
97 * We used to use this for UPDATE as well, but now the equivalent checks
98 * are done in ExecBuildUpdateProjection.
99 */
100 static void
ExecCheckPlanOutput(Relation resultRel,List * targetList)101 ExecCheckPlanOutput(Relation resultRel, List *targetList)
102 {
103 TupleDesc resultDesc = RelationGetDescr(resultRel);
104 int attno = 0;
105 ListCell *lc;
106
107 foreach(lc, targetList)
108 {
109 TargetEntry *tle = (TargetEntry *) lfirst(lc);
110 Form_pg_attribute attr;
111
112 Assert(!tle->resjunk); /* caller removed junk items already */
113
114 if (attno >= resultDesc->natts)
115 ereport(ERROR,
116 (errcode(ERRCODE_DATATYPE_MISMATCH),
117 errmsg("table row type and query-specified row type do not match"),
118 errdetail("Query has too many columns.")));
119 attr = TupleDescAttr(resultDesc, attno);
120 attno++;
121
122 if (!attr->attisdropped)
123 {
124 /* Normal case: demand type match */
125 if (exprType((Node *) tle->expr) != attr->atttypid)
126 ereport(ERROR,
127 (errcode(ERRCODE_DATATYPE_MISMATCH),
128 errmsg("table row type and query-specified row type do not match"),
129 errdetail("Table has type %s at ordinal position %d, but query expects %s.",
130 format_type_be(attr->atttypid),
131 attno,
132 format_type_be(exprType((Node *) tle->expr)))));
133 }
134 else
135 {
136 /*
137 * For a dropped column, we can't check atttypid (it's likely 0).
138 * In any case the planner has most likely inserted an INT4 null.
139 * What we insist on is just *some* NULL constant.
140 */
141 if (!IsA(tle->expr, Const) ||
142 !((Const *) tle->expr)->constisnull)
143 ereport(ERROR,
144 (errcode(ERRCODE_DATATYPE_MISMATCH),
145 errmsg("table row type and query-specified row type do not match"),
146 errdetail("Query provides a value for a dropped column at ordinal position %d.",
147 attno)));
148 }
149 }
150 if (attno != resultDesc->natts)
151 ereport(ERROR,
152 (errcode(ERRCODE_DATATYPE_MISMATCH),
153 errmsg("table row type and query-specified row type do not match"),
154 errdetail("Query has too few columns.")));
155 }
156
157 /*
158 * ExecProcessReturning --- evaluate a RETURNING list
159 *
160 * resultRelInfo: current result rel
161 * tupleSlot: slot holding tuple actually inserted/updated/deleted
162 * planSlot: slot holding tuple returned by top subplan node
163 *
164 * Note: If tupleSlot is NULL, the FDW should have already provided econtext's
165 * scan tuple.
166 *
167 * Returns a slot holding the result tuple
168 */
169 static TupleTableSlot *
ExecProcessReturning(ResultRelInfo * resultRelInfo,TupleTableSlot * tupleSlot,TupleTableSlot * planSlot)170 ExecProcessReturning(ResultRelInfo *resultRelInfo,
171 TupleTableSlot *tupleSlot,
172 TupleTableSlot *planSlot)
173 {
174 ProjectionInfo *projectReturning = resultRelInfo->ri_projectReturning;
175 ExprContext *econtext = projectReturning->pi_exprContext;
176
177 /* Make tuple and any needed join variables available to ExecProject */
178 if (tupleSlot)
179 econtext->ecxt_scantuple = tupleSlot;
180 econtext->ecxt_outertuple = planSlot;
181
182 /*
183 * RETURNING expressions might reference the tableoid column, so
184 * reinitialize tts_tableOid before evaluating them.
185 */
186 econtext->ecxt_scantuple->tts_tableOid =
187 RelationGetRelid(resultRelInfo->ri_RelationDesc);
188
189 /* Compute the RETURNING expressions */
190 return ExecProject(projectReturning);
191 }
192
193 /*
194 * ExecCheckTupleVisible -- verify tuple is visible
195 *
196 * It would not be consistent with guarantees of the higher isolation levels to
197 * proceed with avoiding insertion (taking speculative insertion's alternative
198 * path) on the basis of another tuple that is not visible to MVCC snapshot.
199 * Check for the need to raise a serialization failure, and do so as necessary.
200 */
201 static void
ExecCheckTupleVisible(EState * estate,Relation rel,TupleTableSlot * slot)202 ExecCheckTupleVisible(EState *estate,
203 Relation rel,
204 TupleTableSlot *slot)
205 {
206 if (!IsolationUsesXactSnapshot())
207 return;
208
209 if (!table_tuple_satisfies_snapshot(rel, slot, estate->es_snapshot))
210 {
211 Datum xminDatum;
212 TransactionId xmin;
213 bool isnull;
214
215 xminDatum = slot_getsysattr(slot, MinTransactionIdAttributeNumber, &isnull);
216 Assert(!isnull);
217 xmin = DatumGetTransactionId(xminDatum);
218
219 /*
220 * We should not raise a serialization failure if the conflict is
221 * against a tuple inserted by our own transaction, even if it's not
222 * visible to our snapshot. (This would happen, for example, if
223 * conflicting keys are proposed for insertion in a single command.)
224 */
225 if (!TransactionIdIsCurrentTransactionId(xmin))
226 ereport(ERROR,
227 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
228 errmsg("could not serialize access due to concurrent update")));
229 }
230 }
231
232 /*
233 * ExecCheckTIDVisible -- convenience variant of ExecCheckTupleVisible()
234 */
235 static void
ExecCheckTIDVisible(EState * estate,ResultRelInfo * relinfo,ItemPointer tid,TupleTableSlot * tempSlot)236 ExecCheckTIDVisible(EState *estate,
237 ResultRelInfo *relinfo,
238 ItemPointer tid,
239 TupleTableSlot *tempSlot)
240 {
241 Relation rel = relinfo->ri_RelationDesc;
242
243 /* Redundantly check isolation level */
244 if (!IsolationUsesXactSnapshot())
245 return;
246
247 if (!table_tuple_fetch_row_version(rel, tid, SnapshotAny, tempSlot))
248 elog(ERROR, "failed to fetch conflicting tuple for ON CONFLICT");
249 ExecCheckTupleVisible(estate, rel, tempSlot);
250 ExecClearTuple(tempSlot);
251 }
252
253 /*
254 * Compute stored generated columns for a tuple
255 */
256 void
ExecComputeStoredGenerated(ResultRelInfo * resultRelInfo,EState * estate,TupleTableSlot * slot,CmdType cmdtype)257 ExecComputeStoredGenerated(ResultRelInfo *resultRelInfo,
258 EState *estate, TupleTableSlot *slot,
259 CmdType cmdtype)
260 {
261 Relation rel = resultRelInfo->ri_RelationDesc;
262 TupleDesc tupdesc = RelationGetDescr(rel);
263 int natts = tupdesc->natts;
264 MemoryContext oldContext;
265 Datum *values;
266 bool *nulls;
267
268 Assert(tupdesc->constr && tupdesc->constr->has_generated_stored);
269
270 /*
271 * If first time through for this result relation, build expression
272 * nodetrees for rel's stored generation expressions. Keep them in the
273 * per-query memory context so they'll survive throughout the query.
274 */
275 if (resultRelInfo->ri_GeneratedExprs == NULL)
276 {
277 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
278
279 resultRelInfo->ri_GeneratedExprs =
280 (ExprState **) palloc(natts * sizeof(ExprState *));
281 resultRelInfo->ri_NumGeneratedNeeded = 0;
282
283 for (int i = 0; i < natts; i++)
284 {
285 if (TupleDescAttr(tupdesc, i)->attgenerated == ATTRIBUTE_GENERATED_STORED)
286 {
287 Expr *expr;
288
289 /*
290 * If it's an update and the current column was not marked as
291 * being updated, then we can skip the computation. But if
292 * there is a BEFORE ROW UPDATE trigger, we cannot skip
293 * because the trigger might affect additional columns.
294 */
295 if (cmdtype == CMD_UPDATE &&
296 !(rel->trigdesc && rel->trigdesc->trig_update_before_row) &&
297 !bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
298 ExecGetExtraUpdatedCols(resultRelInfo, estate)))
299 {
300 resultRelInfo->ri_GeneratedExprs[i] = NULL;
301 continue;
302 }
303
304 expr = (Expr *) build_column_default(rel, i + 1);
305 if (expr == NULL)
306 elog(ERROR, "no generation expression found for column number %d of table \"%s\"",
307 i + 1, RelationGetRelationName(rel));
308
309 resultRelInfo->ri_GeneratedExprs[i] = ExecPrepareExpr(expr, estate);
310 resultRelInfo->ri_NumGeneratedNeeded++;
311 }
312 }
313
314 MemoryContextSwitchTo(oldContext);
315 }
316
317 /*
318 * If no generated columns have been affected by this change, then skip
319 * the rest.
320 */
321 if (resultRelInfo->ri_NumGeneratedNeeded == 0)
322 return;
323
324 oldContext = MemoryContextSwitchTo(GetPerTupleMemoryContext(estate));
325
326 values = palloc(sizeof(*values) * natts);
327 nulls = palloc(sizeof(*nulls) * natts);
328
329 slot_getallattrs(slot);
330 memcpy(nulls, slot->tts_isnull, sizeof(*nulls) * natts);
331
332 for (int i = 0; i < natts; i++)
333 {
334 Form_pg_attribute attr = TupleDescAttr(tupdesc, i);
335
336 if (attr->attgenerated == ATTRIBUTE_GENERATED_STORED &&
337 resultRelInfo->ri_GeneratedExprs[i])
338 {
339 ExprContext *econtext;
340 Datum val;
341 bool isnull;
342
343 econtext = GetPerTupleExprContext(estate);
344 econtext->ecxt_scantuple = slot;
345
346 val = ExecEvalExpr(resultRelInfo->ri_GeneratedExprs[i], econtext, &isnull);
347
348 /*
349 * We must make a copy of val as we have no guarantees about where
350 * memory for a pass-by-reference Datum is located.
351 */
352 if (!isnull)
353 val = datumCopy(val, attr->attbyval, attr->attlen);
354
355 values[i] = val;
356 nulls[i] = isnull;
357 }
358 else
359 {
360 if (!nulls[i])
361 values[i] = datumCopy(slot->tts_values[i], attr->attbyval, attr->attlen);
362 }
363 }
364
365 ExecClearTuple(slot);
366 memcpy(slot->tts_values, values, sizeof(*values) * natts);
367 memcpy(slot->tts_isnull, nulls, sizeof(*nulls) * natts);
368 ExecStoreVirtualTuple(slot);
369 ExecMaterializeSlot(slot);
370
371 MemoryContextSwitchTo(oldContext);
372 }
373
374 /*
375 * ExecInitInsertProjection
376 * Do one-time initialization of projection data for INSERT tuples.
377 *
378 * INSERT queries may need a projection to filter out junk attrs in the tlist.
379 *
380 * This is also a convenient place to verify that the
381 * output of an INSERT matches the target table.
382 */
383 static void
ExecInitInsertProjection(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo)384 ExecInitInsertProjection(ModifyTableState *mtstate,
385 ResultRelInfo *resultRelInfo)
386 {
387 ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
388 Plan *subplan = outerPlan(node);
389 EState *estate = mtstate->ps.state;
390 List *insertTargetList = NIL;
391 bool need_projection = false;
392 ListCell *l;
393
394 /* Extract non-junk columns of the subplan's result tlist. */
395 foreach(l, subplan->targetlist)
396 {
397 TargetEntry *tle = (TargetEntry *) lfirst(l);
398
399 if (!tle->resjunk)
400 insertTargetList = lappend(insertTargetList, tle);
401 else
402 need_projection = true;
403 }
404
405 /*
406 * The junk-free list must produce a tuple suitable for the result
407 * relation.
408 */
409 ExecCheckPlanOutput(resultRelInfo->ri_RelationDesc, insertTargetList);
410
411 /* We'll need a slot matching the table's format. */
412 resultRelInfo->ri_newTupleSlot =
413 table_slot_create(resultRelInfo->ri_RelationDesc,
414 &estate->es_tupleTable);
415
416 /* Build ProjectionInfo if needed (it probably isn't). */
417 if (need_projection)
418 {
419 TupleDesc relDesc = RelationGetDescr(resultRelInfo->ri_RelationDesc);
420
421 /* need an expression context to do the projection */
422 if (mtstate->ps.ps_ExprContext == NULL)
423 ExecAssignExprContext(estate, &mtstate->ps);
424
425 resultRelInfo->ri_projectNew =
426 ExecBuildProjectionInfo(insertTargetList,
427 mtstate->ps.ps_ExprContext,
428 resultRelInfo->ri_newTupleSlot,
429 &mtstate->ps,
430 relDesc);
431 }
432
433 resultRelInfo->ri_projectNewInfoValid = true;
434 }
435
436 /*
437 * ExecInitUpdateProjection
438 * Do one-time initialization of projection data for UPDATE tuples.
439 *
440 * UPDATE always needs a projection, because (1) there's always some junk
441 * attrs, and (2) we may need to merge values of not-updated columns from
442 * the old tuple into the final tuple. In UPDATE, the tuple arriving from
443 * the subplan contains only new values for the changed columns, plus row
444 * identity info in the junk attrs.
445 *
446 * This is "one-time" for any given result rel, but we might touch more than
447 * one result rel in the course of an inherited UPDATE, and each one needs
448 * its own projection due to possible column order variation.
449 *
450 * This is also a convenient place to verify that the output of an UPDATE
451 * matches the target table (ExecBuildUpdateProjection does that).
452 */
453 static void
ExecInitUpdateProjection(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo)454 ExecInitUpdateProjection(ModifyTableState *mtstate,
455 ResultRelInfo *resultRelInfo)
456 {
457 ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
458 Plan *subplan = outerPlan(node);
459 EState *estate = mtstate->ps.state;
460 TupleDesc relDesc = RelationGetDescr(resultRelInfo->ri_RelationDesc);
461 int whichrel;
462 List *updateColnos;
463
464 /*
465 * Usually, mt_lastResultIndex matches the target rel. If it happens not
466 * to, we can get the index the hard way with an integer division.
467 */
468 whichrel = mtstate->mt_lastResultIndex;
469 if (resultRelInfo != mtstate->resultRelInfo + whichrel)
470 {
471 whichrel = resultRelInfo - mtstate->resultRelInfo;
472 Assert(whichrel >= 0 && whichrel < mtstate->mt_nrels);
473 }
474
475 updateColnos = (List *) list_nth(node->updateColnosLists, whichrel);
476
477 /*
478 * For UPDATE, we use the old tuple to fill up missing values in the tuple
479 * produced by the subplan to get the new tuple. We need two slots, both
480 * matching the table's desired format.
481 */
482 resultRelInfo->ri_oldTupleSlot =
483 table_slot_create(resultRelInfo->ri_RelationDesc,
484 &estate->es_tupleTable);
485 resultRelInfo->ri_newTupleSlot =
486 table_slot_create(resultRelInfo->ri_RelationDesc,
487 &estate->es_tupleTable);
488
489 /* need an expression context to do the projection */
490 if (mtstate->ps.ps_ExprContext == NULL)
491 ExecAssignExprContext(estate, &mtstate->ps);
492
493 resultRelInfo->ri_projectNew =
494 ExecBuildUpdateProjection(subplan->targetlist,
495 false, /* subplan did the evaluation */
496 updateColnos,
497 relDesc,
498 mtstate->ps.ps_ExprContext,
499 resultRelInfo->ri_newTupleSlot,
500 &mtstate->ps);
501
502 resultRelInfo->ri_projectNewInfoValid = true;
503 }
504
505 /*
506 * ExecGetInsertNewTuple
507 * This prepares a "new" tuple ready to be inserted into given result
508 * relation, by removing any junk columns of the plan's output tuple
509 * and (if necessary) coercing the tuple to the right tuple format.
510 */
511 static TupleTableSlot *
ExecGetInsertNewTuple(ResultRelInfo * relinfo,TupleTableSlot * planSlot)512 ExecGetInsertNewTuple(ResultRelInfo *relinfo,
513 TupleTableSlot *planSlot)
514 {
515 ProjectionInfo *newProj = relinfo->ri_projectNew;
516 ExprContext *econtext;
517
518 /*
519 * If there's no projection to be done, just make sure the slot is of the
520 * right type for the target rel. If the planSlot is the right type we
521 * can use it as-is, else copy the data into ri_newTupleSlot.
522 */
523 if (newProj == NULL)
524 {
525 if (relinfo->ri_newTupleSlot->tts_ops != planSlot->tts_ops)
526 {
527 ExecCopySlot(relinfo->ri_newTupleSlot, planSlot);
528 return relinfo->ri_newTupleSlot;
529 }
530 else
531 return planSlot;
532 }
533
534 /*
535 * Else project; since the projection output slot is ri_newTupleSlot, this
536 * will also fix any slot-type problem.
537 *
538 * Note: currently, this is dead code, because INSERT cases don't receive
539 * any junk columns so there's never a projection to be done.
540 */
541 econtext = newProj->pi_exprContext;
542 econtext->ecxt_outertuple = planSlot;
543 return ExecProject(newProj);
544 }
545
546 /*
547 * ExecGetUpdateNewTuple
548 * This prepares a "new" tuple by combining an UPDATE subplan's output
549 * tuple (which contains values of changed columns) with unchanged
550 * columns taken from the old tuple.
551 *
552 * The subplan tuple might also contain junk columns, which are ignored.
553 * Note that the projection also ensures we have a slot of the right type.
554 */
555 TupleTableSlot *
ExecGetUpdateNewTuple(ResultRelInfo * relinfo,TupleTableSlot * planSlot,TupleTableSlot * oldSlot)556 ExecGetUpdateNewTuple(ResultRelInfo *relinfo,
557 TupleTableSlot *planSlot,
558 TupleTableSlot *oldSlot)
559 {
560 ProjectionInfo *newProj = relinfo->ri_projectNew;
561 ExprContext *econtext;
562
563 /* Use a few extra Asserts to protect against outside callers */
564 Assert(relinfo->ri_projectNewInfoValid);
565 Assert(planSlot != NULL && !TTS_EMPTY(planSlot));
566 Assert(oldSlot != NULL && !TTS_EMPTY(oldSlot));
567
568 econtext = newProj->pi_exprContext;
569 econtext->ecxt_outertuple = planSlot;
570 econtext->ecxt_scantuple = oldSlot;
571 return ExecProject(newProj);
572 }
573
574
575 /* ----------------------------------------------------------------
576 * ExecInsert
577 *
578 * For INSERT, we have to insert the tuple into the target relation
579 * (or partition thereof) and insert appropriate tuples into the index
580 * relations.
581 *
582 * slot contains the new tuple value to be stored.
583 * planSlot is the output of the ModifyTable's subplan; we use it
584 * to access "junk" columns that are not going to be stored.
585 *
586 * Returns RETURNING result if any, otherwise NULL.
587 *
588 * This may change the currently active tuple conversion map in
589 * mtstate->mt_transition_capture, so the callers must take care to
590 * save the previous value to avoid losing track of it.
591 * ----------------------------------------------------------------
592 */
593 static TupleTableSlot *
ExecInsert(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo,TupleTableSlot * slot,TupleTableSlot * planSlot,EState * estate,bool canSetTag)594 ExecInsert(ModifyTableState *mtstate,
595 ResultRelInfo *resultRelInfo,
596 TupleTableSlot *slot,
597 TupleTableSlot *planSlot,
598 EState *estate,
599 bool canSetTag)
600 {
601 Relation resultRelationDesc;
602 List *recheckIndexes = NIL;
603 TupleTableSlot *result = NULL;
604 TransitionCaptureState *ar_insert_trig_tcs;
605 ModifyTable *node = (ModifyTable *) mtstate->ps.plan;
606 OnConflictAction onconflict = node->onConflictAction;
607 PartitionTupleRouting *proute = mtstate->mt_partition_tuple_routing;
608 MemoryContext oldContext;
609
610 /*
611 * If the input result relation is a partitioned table, find the leaf
612 * partition to insert the tuple into.
613 */
614 if (proute)
615 {
616 ResultRelInfo *partRelInfo;
617
618 slot = ExecPrepareTupleRouting(mtstate, estate, proute,
619 resultRelInfo, slot,
620 &partRelInfo);
621 resultRelInfo = partRelInfo;
622 }
623
624 ExecMaterializeSlot(slot);
625
626 resultRelationDesc = resultRelInfo->ri_RelationDesc;
627
628 /*
629 * Open the table's indexes, if we have not done so already, so that we
630 * can add new index entries for the inserted tuple.
631 */
632 if (resultRelationDesc->rd_rel->relhasindex &&
633 resultRelInfo->ri_IndexRelationDescs == NULL)
634 ExecOpenIndices(resultRelInfo, onconflict != ONCONFLICT_NONE);
635
636 /*
637 * BEFORE ROW INSERT Triggers.
638 *
639 * Note: We fire BEFORE ROW TRIGGERS for every attempted insertion in an
640 * INSERT ... ON CONFLICT statement. We cannot check for constraint
641 * violations before firing these triggers, because they can change the
642 * values to insert. Also, they can run arbitrary user-defined code with
643 * side-effects that we can't cancel by just not inserting the tuple.
644 */
645 if (resultRelInfo->ri_TrigDesc &&
646 resultRelInfo->ri_TrigDesc->trig_insert_before_row)
647 {
648 if (!ExecBRInsertTriggers(estate, resultRelInfo, slot))
649 return NULL; /* "do nothing" */
650 }
651
652 /* INSTEAD OF ROW INSERT Triggers */
653 if (resultRelInfo->ri_TrigDesc &&
654 resultRelInfo->ri_TrigDesc->trig_insert_instead_row)
655 {
656 if (!ExecIRInsertTriggers(estate, resultRelInfo, slot))
657 return NULL; /* "do nothing" */
658 }
659 else if (resultRelInfo->ri_FdwRoutine)
660 {
661 /*
662 * GENERATED expressions might reference the tableoid column, so
663 * (re-)initialize tts_tableOid before evaluating them.
664 */
665 slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
666
667 /*
668 * Compute stored generated columns
669 */
670 if (resultRelationDesc->rd_att->constr &&
671 resultRelationDesc->rd_att->constr->has_generated_stored)
672 ExecComputeStoredGenerated(resultRelInfo, estate, slot,
673 CMD_INSERT);
674
675 /*
676 * If the FDW supports batching, and batching is requested, accumulate
677 * rows and insert them in batches. Otherwise use the per-row inserts.
678 */
679 if (resultRelInfo->ri_BatchSize > 1)
680 {
681 /*
682 * If a certain number of tuples have already been accumulated, or
683 * a tuple has come for a different relation than that for the
684 * accumulated tuples, perform the batch insert
685 */
686 if (resultRelInfo->ri_NumSlots == resultRelInfo->ri_BatchSize)
687 {
688 ExecBatchInsert(mtstate, resultRelInfo,
689 resultRelInfo->ri_Slots,
690 resultRelInfo->ri_PlanSlots,
691 resultRelInfo->ri_NumSlots,
692 estate, canSetTag);
693 resultRelInfo->ri_NumSlots = 0;
694 }
695
696 oldContext = MemoryContextSwitchTo(estate->es_query_cxt);
697
698 if (resultRelInfo->ri_Slots == NULL)
699 {
700 resultRelInfo->ri_Slots = palloc(sizeof(TupleTableSlot *) *
701 resultRelInfo->ri_BatchSize);
702 resultRelInfo->ri_PlanSlots = palloc(sizeof(TupleTableSlot *) *
703 resultRelInfo->ri_BatchSize);
704 }
705
706 /*
707 * Initialize the batch slots. We don't know how many slots will
708 * be needed, so we initialize them as the batch grows, and we
709 * keep them across batches. To mitigate an inefficiency in how
710 * resource owner handles objects with many references (as with
711 * many slots all referencing the same tuple descriptor) we copy
712 * the appropriate tuple descriptor for each slot.
713 */
714 if (resultRelInfo->ri_NumSlots >= resultRelInfo->ri_NumSlotsInitialized)
715 {
716 TupleDesc tdesc = CreateTupleDescCopy(slot->tts_tupleDescriptor);
717 TupleDesc plan_tdesc =
718 CreateTupleDescCopy(planSlot->tts_tupleDescriptor);
719
720 resultRelInfo->ri_Slots[resultRelInfo->ri_NumSlots] =
721 MakeSingleTupleTableSlot(tdesc, slot->tts_ops);
722
723 resultRelInfo->ri_PlanSlots[resultRelInfo->ri_NumSlots] =
724 MakeSingleTupleTableSlot(plan_tdesc, planSlot->tts_ops);
725
726 /* remember how many batch slots we initialized */
727 resultRelInfo->ri_NumSlotsInitialized++;
728 }
729
730 ExecCopySlot(resultRelInfo->ri_Slots[resultRelInfo->ri_NumSlots],
731 slot);
732
733 ExecCopySlot(resultRelInfo->ri_PlanSlots[resultRelInfo->ri_NumSlots],
734 planSlot);
735
736 resultRelInfo->ri_NumSlots++;
737
738 MemoryContextSwitchTo(oldContext);
739
740 return NULL;
741 }
742
743 /*
744 * insert into foreign table: let the FDW do it
745 */
746 slot = resultRelInfo->ri_FdwRoutine->ExecForeignInsert(estate,
747 resultRelInfo,
748 slot,
749 planSlot);
750
751 if (slot == NULL) /* "do nothing" */
752 return NULL;
753
754 /*
755 * AFTER ROW Triggers or RETURNING expressions might reference the
756 * tableoid column, so (re-)initialize tts_tableOid before evaluating
757 * them. (This covers the case where the FDW replaced the slot.)
758 */
759 slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
760 }
761 else
762 {
763 WCOKind wco_kind;
764
765 /*
766 * Constraints and GENERATED expressions might reference the tableoid
767 * column, so (re-)initialize tts_tableOid before evaluating them.
768 */
769 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
770
771 /*
772 * Compute stored generated columns
773 */
774 if (resultRelationDesc->rd_att->constr &&
775 resultRelationDesc->rd_att->constr->has_generated_stored)
776 ExecComputeStoredGenerated(resultRelInfo, estate, slot,
777 CMD_INSERT);
778
779 /*
780 * Check any RLS WITH CHECK policies.
781 *
782 * Normally we should check INSERT policies. But if the insert is the
783 * result of a partition key update that moved the tuple to a new
784 * partition, we should instead check UPDATE policies, because we are
785 * executing policies defined on the target table, and not those
786 * defined on the child partitions.
787 */
788 wco_kind = (mtstate->operation == CMD_UPDATE) ?
789 WCO_RLS_UPDATE_CHECK : WCO_RLS_INSERT_CHECK;
790
791 /*
792 * ExecWithCheckOptions() will skip any WCOs which are not of the kind
793 * we are looking for at this point.
794 */
795 if (resultRelInfo->ri_WithCheckOptions != NIL)
796 ExecWithCheckOptions(wco_kind, resultRelInfo, slot, estate);
797
798 /*
799 * Check the constraints of the tuple.
800 */
801 if (resultRelationDesc->rd_att->constr)
802 ExecConstraints(resultRelInfo, slot, estate);
803
804 /*
805 * Also check the tuple against the partition constraint, if there is
806 * one; except that if we got here via tuple-routing, we don't need to
807 * if there's no BR trigger defined on the partition.
808 */
809 if (resultRelationDesc->rd_rel->relispartition &&
810 (resultRelInfo->ri_RootResultRelInfo == NULL ||
811 (resultRelInfo->ri_TrigDesc &&
812 resultRelInfo->ri_TrigDesc->trig_insert_before_row)))
813 ExecPartitionCheck(resultRelInfo, slot, estate, true);
814
815 if (onconflict != ONCONFLICT_NONE && resultRelInfo->ri_NumIndices > 0)
816 {
817 /* Perform a speculative insertion. */
818 uint32 specToken;
819 ItemPointerData conflictTid;
820 bool specConflict;
821 List *arbiterIndexes;
822
823 arbiterIndexes = resultRelInfo->ri_onConflictArbiterIndexes;
824
825 /*
826 * Do a non-conclusive check for conflicts first.
827 *
828 * We're not holding any locks yet, so this doesn't guarantee that
829 * the later insert won't conflict. But it avoids leaving behind
830 * a lot of canceled speculative insertions, if you run a lot of
831 * INSERT ON CONFLICT statements that do conflict.
832 *
833 * We loop back here if we find a conflict below, either during
834 * the pre-check, or when we re-check after inserting the tuple
835 * speculatively.
836 */
837 vlock:
838 specConflict = false;
839 if (!ExecCheckIndexConstraints(resultRelInfo, slot, estate,
840 &conflictTid, arbiterIndexes))
841 {
842 /* committed conflict tuple found */
843 if (onconflict == ONCONFLICT_UPDATE)
844 {
845 /*
846 * In case of ON CONFLICT DO UPDATE, execute the UPDATE
847 * part. Be prepared to retry if the UPDATE fails because
848 * of another concurrent UPDATE/DELETE to the conflict
849 * tuple.
850 */
851 TupleTableSlot *returning = NULL;
852
853 if (ExecOnConflictUpdate(mtstate, resultRelInfo,
854 &conflictTid, planSlot, slot,
855 estate, canSetTag, &returning))
856 {
857 InstrCountTuples2(&mtstate->ps, 1);
858 return returning;
859 }
860 else
861 goto vlock;
862 }
863 else
864 {
865 /*
866 * In case of ON CONFLICT DO NOTHING, do nothing. However,
867 * verify that the tuple is visible to the executor's MVCC
868 * snapshot at higher isolation levels.
869 *
870 * Using ExecGetReturningSlot() to store the tuple for the
871 * recheck isn't that pretty, but we can't trivially use
872 * the input slot, because it might not be of a compatible
873 * type. As there's no conflicting usage of
874 * ExecGetReturningSlot() in the DO NOTHING case...
875 */
876 Assert(onconflict == ONCONFLICT_NOTHING);
877 ExecCheckTIDVisible(estate, resultRelInfo, &conflictTid,
878 ExecGetReturningSlot(estate, resultRelInfo));
879 InstrCountTuples2(&mtstate->ps, 1);
880 return NULL;
881 }
882 }
883
884 /*
885 * Before we start insertion proper, acquire our "speculative
886 * insertion lock". Others can use that to wait for us to decide
887 * if we're going to go ahead with the insertion, instead of
888 * waiting for the whole transaction to complete.
889 */
890 specToken = SpeculativeInsertionLockAcquire(GetCurrentTransactionId());
891
892 /* insert the tuple, with the speculative token */
893 table_tuple_insert_speculative(resultRelationDesc, slot,
894 estate->es_output_cid,
895 0,
896 NULL,
897 specToken);
898
899 /* insert index entries for tuple */
900 recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
901 slot, estate, false, true,
902 &specConflict,
903 arbiterIndexes);
904
905 /* adjust the tuple's state accordingly */
906 table_tuple_complete_speculative(resultRelationDesc, slot,
907 specToken, !specConflict);
908
909 /*
910 * Wake up anyone waiting for our decision. They will re-check
911 * the tuple, see that it's no longer speculative, and wait on our
912 * XID as if this was a regularly inserted tuple all along. Or if
913 * we killed the tuple, they will see it's dead, and proceed as if
914 * the tuple never existed.
915 */
916 SpeculativeInsertionLockRelease(GetCurrentTransactionId());
917
918 /*
919 * If there was a conflict, start from the beginning. We'll do
920 * the pre-check again, which will now find the conflicting tuple
921 * (unless it aborts before we get there).
922 */
923 if (specConflict)
924 {
925 list_free(recheckIndexes);
926 goto vlock;
927 }
928
929 /* Since there was no insertion conflict, we're done */
930 }
931 else
932 {
933 /* insert the tuple normally */
934 table_tuple_insert(resultRelationDesc, slot,
935 estate->es_output_cid,
936 0, NULL);
937
938 /* insert index entries for tuple */
939 if (resultRelInfo->ri_NumIndices > 0)
940 recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
941 slot, estate, false,
942 false, NULL, NIL);
943 }
944 }
945
946 if (canSetTag)
947 (estate->es_processed)++;
948
949 /*
950 * If this insert is the result of a partition key update that moved the
951 * tuple to a new partition, put this row into the transition NEW TABLE,
952 * if there is one. We need to do this separately for DELETE and INSERT
953 * because they happen on different tables.
954 */
955 ar_insert_trig_tcs = mtstate->mt_transition_capture;
956 if (mtstate->operation == CMD_UPDATE && mtstate->mt_transition_capture
957 && mtstate->mt_transition_capture->tcs_update_new_table)
958 {
959 ExecARUpdateTriggers(estate, resultRelInfo, NULL,
960 NULL,
961 slot,
962 NULL,
963 mtstate->mt_transition_capture);
964
965 /*
966 * We've already captured the NEW TABLE row, so make sure any AR
967 * INSERT trigger fired below doesn't capture it again.
968 */
969 ar_insert_trig_tcs = NULL;
970 }
971
972 /* AFTER ROW INSERT Triggers */
973 ExecARInsertTriggers(estate, resultRelInfo, slot, recheckIndexes,
974 ar_insert_trig_tcs);
975
976 list_free(recheckIndexes);
977
978 /*
979 * Check any WITH CHECK OPTION constraints from parent views. We are
980 * required to do this after testing all constraints and uniqueness
981 * violations per the SQL spec, so we do it after actually inserting the
982 * record into the heap and all indexes.
983 *
984 * ExecWithCheckOptions will elog(ERROR) if a violation is found, so the
985 * tuple will never be seen, if it violates the WITH CHECK OPTION.
986 *
987 * ExecWithCheckOptions() will skip any WCOs which are not of the kind we
988 * are looking for at this point.
989 */
990 if (resultRelInfo->ri_WithCheckOptions != NIL)
991 ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
992
993 /* Process RETURNING if present */
994 if (resultRelInfo->ri_projectReturning)
995 result = ExecProcessReturning(resultRelInfo, slot, planSlot);
996
997 return result;
998 }
999
1000 /* ----------------------------------------------------------------
1001 * ExecBatchInsert
1002 *
1003 * Insert multiple tuples in an efficient way.
1004 * Currently, this handles inserting into a foreign table without
1005 * RETURNING clause.
1006 * ----------------------------------------------------------------
1007 */
1008 static void
ExecBatchInsert(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo,TupleTableSlot ** slots,TupleTableSlot ** planSlots,int numSlots,EState * estate,bool canSetTag)1009 ExecBatchInsert(ModifyTableState *mtstate,
1010 ResultRelInfo *resultRelInfo,
1011 TupleTableSlot **slots,
1012 TupleTableSlot **planSlots,
1013 int numSlots,
1014 EState *estate,
1015 bool canSetTag)
1016 {
1017 int i;
1018 int numInserted = numSlots;
1019 TupleTableSlot *slot = NULL;
1020 TupleTableSlot **rslots;
1021
1022 /*
1023 * insert into foreign table: let the FDW do it
1024 */
1025 rslots = resultRelInfo->ri_FdwRoutine->ExecForeignBatchInsert(estate,
1026 resultRelInfo,
1027 slots,
1028 planSlots,
1029 &numInserted);
1030
1031 for (i = 0; i < numInserted; i++)
1032 {
1033 slot = rslots[i];
1034
1035 /*
1036 * AFTER ROW Triggers or RETURNING expressions might reference the
1037 * tableoid column, so (re-)initialize tts_tableOid before evaluating
1038 * them.
1039 */
1040 slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
1041
1042 /* AFTER ROW INSERT Triggers */
1043 ExecARInsertTriggers(estate, resultRelInfo, slot, NIL,
1044 mtstate->mt_transition_capture);
1045
1046 /*
1047 * Check any WITH CHECK OPTION constraints from parent views. See the
1048 * comment in ExecInsert.
1049 */
1050 if (resultRelInfo->ri_WithCheckOptions != NIL)
1051 ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
1052 }
1053
1054 if (canSetTag && numInserted > 0)
1055 estate->es_processed += numInserted;
1056 }
1057
1058 /* ----------------------------------------------------------------
1059 * ExecDelete
1060 *
1061 * DELETE is like UPDATE, except that we delete the tuple and no
1062 * index modifications are needed.
1063 *
1064 * When deleting from a table, tupleid identifies the tuple to
1065 * delete and oldtuple is NULL. When deleting from a view,
1066 * oldtuple is passed to the INSTEAD OF triggers and identifies
1067 * what to delete, and tupleid is invalid. When deleting from a
1068 * foreign table, tupleid is invalid; the FDW has to figure out
1069 * which row to delete using data from the planSlot. oldtuple is
1070 * passed to foreign table triggers; it is NULL when the foreign
1071 * table has no relevant triggers. We use tupleDeleted to indicate
1072 * whether the tuple is actually deleted, callers can use it to
1073 * decide whether to continue the operation. When this DELETE is a
1074 * part of an UPDATE of partition-key, then the slot returned by
1075 * EvalPlanQual() is passed back using output parameter epqslot.
1076 *
1077 * Returns RETURNING result if any, otherwise NULL.
1078 * ----------------------------------------------------------------
1079 */
1080 static TupleTableSlot *
ExecDelete(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo,ItemPointer tupleid,HeapTuple oldtuple,TupleTableSlot * planSlot,EPQState * epqstate,EState * estate,bool processReturning,bool canSetTag,bool changingPart,bool * tupleDeleted,TupleTableSlot ** epqreturnslot)1081 ExecDelete(ModifyTableState *mtstate,
1082 ResultRelInfo *resultRelInfo,
1083 ItemPointer tupleid,
1084 HeapTuple oldtuple,
1085 TupleTableSlot *planSlot,
1086 EPQState *epqstate,
1087 EState *estate,
1088 bool processReturning,
1089 bool canSetTag,
1090 bool changingPart,
1091 bool *tupleDeleted,
1092 TupleTableSlot **epqreturnslot)
1093 {
1094 Relation resultRelationDesc = resultRelInfo->ri_RelationDesc;
1095 TM_Result result;
1096 TM_FailureData tmfd;
1097 TupleTableSlot *slot = NULL;
1098 TransitionCaptureState *ar_delete_trig_tcs;
1099
1100 if (tupleDeleted)
1101 *tupleDeleted = false;
1102
1103 /* BEFORE ROW DELETE Triggers */
1104 if (resultRelInfo->ri_TrigDesc &&
1105 resultRelInfo->ri_TrigDesc->trig_delete_before_row)
1106 {
1107 bool dodelete;
1108
1109 dodelete = ExecBRDeleteTriggers(estate, epqstate, resultRelInfo,
1110 tupleid, oldtuple, epqreturnslot);
1111
1112 if (!dodelete) /* "do nothing" */
1113 return NULL;
1114 }
1115
1116 /* INSTEAD OF ROW DELETE Triggers */
1117 if (resultRelInfo->ri_TrigDesc &&
1118 resultRelInfo->ri_TrigDesc->trig_delete_instead_row)
1119 {
1120 bool dodelete;
1121
1122 Assert(oldtuple != NULL);
1123 dodelete = ExecIRDeleteTriggers(estate, resultRelInfo, oldtuple);
1124
1125 if (!dodelete) /* "do nothing" */
1126 return NULL;
1127 }
1128 else if (resultRelInfo->ri_FdwRoutine)
1129 {
1130 /*
1131 * delete from foreign table: let the FDW do it
1132 *
1133 * We offer the returning slot as a place to store RETURNING data,
1134 * although the FDW can return some other slot if it wants.
1135 */
1136 slot = ExecGetReturningSlot(estate, resultRelInfo);
1137 slot = resultRelInfo->ri_FdwRoutine->ExecForeignDelete(estate,
1138 resultRelInfo,
1139 slot,
1140 planSlot);
1141
1142 if (slot == NULL) /* "do nothing" */
1143 return NULL;
1144
1145 /*
1146 * RETURNING expressions might reference the tableoid column, so
1147 * (re)initialize tts_tableOid before evaluating them.
1148 */
1149 if (TTS_EMPTY(slot))
1150 ExecStoreAllNullTuple(slot);
1151
1152 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
1153 }
1154 else
1155 {
1156 /*
1157 * delete the tuple
1158 *
1159 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check
1160 * that the row to be deleted is visible to that snapshot, and throw a
1161 * can't-serialize error if not. This is a special-case behavior
1162 * needed for referential integrity updates in transaction-snapshot
1163 * mode transactions.
1164 */
1165 ldelete:;
1166 result = table_tuple_delete(resultRelationDesc, tupleid,
1167 estate->es_output_cid,
1168 estate->es_snapshot,
1169 estate->es_crosscheck_snapshot,
1170 true /* wait for commit */ ,
1171 &tmfd,
1172 changingPart);
1173
1174 switch (result)
1175 {
1176 case TM_SelfModified:
1177
1178 /*
1179 * The target tuple was already updated or deleted by the
1180 * current command, or by a later command in the current
1181 * transaction. The former case is possible in a join DELETE
1182 * where multiple tuples join to the same target tuple. This
1183 * is somewhat questionable, but Postgres has always allowed
1184 * it: we just ignore additional deletion attempts.
1185 *
1186 * The latter case arises if the tuple is modified by a
1187 * command in a BEFORE trigger, or perhaps by a command in a
1188 * volatile function used in the query. In such situations we
1189 * should not ignore the deletion, but it is equally unsafe to
1190 * proceed. We don't want to discard the original DELETE
1191 * while keeping the triggered actions based on its deletion;
1192 * and it would be no better to allow the original DELETE
1193 * while discarding updates that it triggered. The row update
1194 * carries some information that might be important according
1195 * to business rules; so throwing an error is the only safe
1196 * course.
1197 *
1198 * If a trigger actually intends this type of interaction, it
1199 * can re-execute the DELETE and then return NULL to cancel
1200 * the outer delete.
1201 */
1202 if (tmfd.cmax != estate->es_output_cid)
1203 ereport(ERROR,
1204 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
1205 errmsg("tuple to be deleted was already modified by an operation triggered by the current command"),
1206 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
1207
1208 /* Else, already deleted by self; nothing to do */
1209 return NULL;
1210
1211 case TM_Ok:
1212 break;
1213
1214 case TM_Updated:
1215 {
1216 TupleTableSlot *inputslot;
1217 TupleTableSlot *epqslot;
1218
1219 if (IsolationUsesXactSnapshot())
1220 ereport(ERROR,
1221 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1222 errmsg("could not serialize access due to concurrent update")));
1223
1224 /*
1225 * Already know that we're going to need to do EPQ, so
1226 * fetch tuple directly into the right slot.
1227 */
1228 EvalPlanQualBegin(epqstate);
1229 inputslot = EvalPlanQualSlot(epqstate, resultRelationDesc,
1230 resultRelInfo->ri_RangeTableIndex);
1231
1232 result = table_tuple_lock(resultRelationDesc, tupleid,
1233 estate->es_snapshot,
1234 inputslot, estate->es_output_cid,
1235 LockTupleExclusive, LockWaitBlock,
1236 TUPLE_LOCK_FLAG_FIND_LAST_VERSION,
1237 &tmfd);
1238
1239 switch (result)
1240 {
1241 case TM_Ok:
1242 Assert(tmfd.traversed);
1243 epqslot = EvalPlanQual(epqstate,
1244 resultRelationDesc,
1245 resultRelInfo->ri_RangeTableIndex,
1246 inputslot);
1247 if (TupIsNull(epqslot))
1248 /* Tuple not passing quals anymore, exiting... */
1249 return NULL;
1250
1251 /*
1252 * If requested, skip delete and pass back the
1253 * updated row.
1254 */
1255 if (epqreturnslot)
1256 {
1257 *epqreturnslot = epqslot;
1258 return NULL;
1259 }
1260 else
1261 goto ldelete;
1262
1263 case TM_SelfModified:
1264
1265 /*
1266 * This can be reached when following an update
1267 * chain from a tuple updated by another session,
1268 * reaching a tuple that was already updated in
1269 * this transaction. If previously updated by this
1270 * command, ignore the delete, otherwise error
1271 * out.
1272 *
1273 * See also TM_SelfModified response to
1274 * table_tuple_delete() above.
1275 */
1276 if (tmfd.cmax != estate->es_output_cid)
1277 ereport(ERROR,
1278 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
1279 errmsg("tuple to be deleted was already modified by an operation triggered by the current command"),
1280 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
1281 return NULL;
1282
1283 case TM_Deleted:
1284 /* tuple already deleted; nothing to do */
1285 return NULL;
1286
1287 default:
1288
1289 /*
1290 * TM_Invisible should be impossible because we're
1291 * waiting for updated row versions, and would
1292 * already have errored out if the first version
1293 * is invisible.
1294 *
1295 * TM_Updated should be impossible, because we're
1296 * locking the latest version via
1297 * TUPLE_LOCK_FLAG_FIND_LAST_VERSION.
1298 */
1299 elog(ERROR, "unexpected table_tuple_lock status: %u",
1300 result);
1301 return NULL;
1302 }
1303
1304 Assert(false);
1305 break;
1306 }
1307
1308 case TM_Deleted:
1309 if (IsolationUsesXactSnapshot())
1310 ereport(ERROR,
1311 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1312 errmsg("could not serialize access due to concurrent delete")));
1313 /* tuple already deleted; nothing to do */
1314 return NULL;
1315
1316 default:
1317 elog(ERROR, "unrecognized table_tuple_delete status: %u",
1318 result);
1319 return NULL;
1320 }
1321
1322 /*
1323 * Note: Normally one would think that we have to delete index tuples
1324 * associated with the heap tuple now...
1325 *
1326 * ... but in POSTGRES, we have no need to do this because VACUUM will
1327 * take care of it later. We can't delete index tuples immediately
1328 * anyway, since the tuple is still visible to other transactions.
1329 */
1330 }
1331
1332 if (canSetTag)
1333 (estate->es_processed)++;
1334
1335 /* Tell caller that the delete actually happened. */
1336 if (tupleDeleted)
1337 *tupleDeleted = true;
1338
1339 /*
1340 * If this delete is the result of a partition key update that moved the
1341 * tuple to a new partition, put this row into the transition OLD TABLE,
1342 * if there is one. We need to do this separately for DELETE and INSERT
1343 * because they happen on different tables.
1344 */
1345 ar_delete_trig_tcs = mtstate->mt_transition_capture;
1346 if (mtstate->operation == CMD_UPDATE && mtstate->mt_transition_capture
1347 && mtstate->mt_transition_capture->tcs_update_old_table)
1348 {
1349 ExecARUpdateTriggers(estate, resultRelInfo,
1350 tupleid,
1351 oldtuple,
1352 NULL,
1353 NULL,
1354 mtstate->mt_transition_capture);
1355
1356 /*
1357 * We've already captured the NEW TABLE row, so make sure any AR
1358 * DELETE trigger fired below doesn't capture it again.
1359 */
1360 ar_delete_trig_tcs = NULL;
1361 }
1362
1363 /* AFTER ROW DELETE Triggers */
1364 ExecARDeleteTriggers(estate, resultRelInfo, tupleid, oldtuple,
1365 ar_delete_trig_tcs);
1366
1367 /* Process RETURNING if present and if requested */
1368 if (processReturning && resultRelInfo->ri_projectReturning)
1369 {
1370 /*
1371 * We have to put the target tuple into a slot, which means first we
1372 * gotta fetch it. We can use the trigger tuple slot.
1373 */
1374 TupleTableSlot *rslot;
1375
1376 if (resultRelInfo->ri_FdwRoutine)
1377 {
1378 /* FDW must have provided a slot containing the deleted row */
1379 Assert(!TupIsNull(slot));
1380 }
1381 else
1382 {
1383 slot = ExecGetReturningSlot(estate, resultRelInfo);
1384 if (oldtuple != NULL)
1385 {
1386 ExecForceStoreHeapTuple(oldtuple, slot, false);
1387 }
1388 else
1389 {
1390 if (!table_tuple_fetch_row_version(resultRelationDesc, tupleid,
1391 SnapshotAny, slot))
1392 elog(ERROR, "failed to fetch deleted tuple for DELETE RETURNING");
1393 }
1394 }
1395
1396 rslot = ExecProcessReturning(resultRelInfo, slot, planSlot);
1397
1398 /*
1399 * Before releasing the target tuple again, make sure rslot has a
1400 * local copy of any pass-by-reference values.
1401 */
1402 ExecMaterializeSlot(rslot);
1403
1404 ExecClearTuple(slot);
1405
1406 return rslot;
1407 }
1408
1409 return NULL;
1410 }
1411
1412 /*
1413 * ExecCrossPartitionUpdate --- Move an updated tuple to another partition.
1414 *
1415 * This works by first deleting the old tuple from the current partition,
1416 * followed by inserting the new tuple into the root parent table, that is,
1417 * mtstate->rootResultRelInfo. It will be re-routed from there to the
1418 * correct partition.
1419 *
1420 * Returns true if the tuple has been successfully moved, or if it's found
1421 * that the tuple was concurrently deleted so there's nothing more to do
1422 * for the caller.
1423 *
1424 * False is returned if the tuple we're trying to move is found to have been
1425 * concurrently updated. In that case, the caller must to check if the
1426 * updated tuple that's returned in *retry_slot still needs to be re-routed,
1427 * and call this function again or perform a regular update accordingly.
1428 */
1429 static bool
ExecCrossPartitionUpdate(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo,ItemPointer tupleid,HeapTuple oldtuple,TupleTableSlot * slot,TupleTableSlot * planSlot,EPQState * epqstate,bool canSetTag,TupleTableSlot ** retry_slot,TupleTableSlot ** inserted_tuple)1430 ExecCrossPartitionUpdate(ModifyTableState *mtstate,
1431 ResultRelInfo *resultRelInfo,
1432 ItemPointer tupleid, HeapTuple oldtuple,
1433 TupleTableSlot *slot, TupleTableSlot *planSlot,
1434 EPQState *epqstate, bool canSetTag,
1435 TupleTableSlot **retry_slot,
1436 TupleTableSlot **inserted_tuple)
1437 {
1438 EState *estate = mtstate->ps.state;
1439 TupleConversionMap *tupconv_map;
1440 bool tuple_deleted;
1441 TupleTableSlot *epqslot = NULL;
1442
1443 *inserted_tuple = NULL;
1444 *retry_slot = NULL;
1445
1446 /*
1447 * Disallow an INSERT ON CONFLICT DO UPDATE that causes the original row
1448 * to migrate to a different partition. Maybe this can be implemented
1449 * some day, but it seems a fringe feature with little redeeming value.
1450 */
1451 if (((ModifyTable *) mtstate->ps.plan)->onConflictAction == ONCONFLICT_UPDATE)
1452 ereport(ERROR,
1453 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1454 errmsg("invalid ON UPDATE specification"),
1455 errdetail("The result tuple would appear in a different partition than the original tuple.")));
1456
1457 /*
1458 * When an UPDATE is run directly on a leaf partition, simply fail with a
1459 * partition constraint violation error.
1460 */
1461 if (resultRelInfo == mtstate->rootResultRelInfo)
1462 ExecPartitionCheckEmitError(resultRelInfo, slot, estate);
1463
1464 /* Initialize tuple routing info if not already done. */
1465 if (mtstate->mt_partition_tuple_routing == NULL)
1466 {
1467 Relation rootRel = mtstate->rootResultRelInfo->ri_RelationDesc;
1468 MemoryContext oldcxt;
1469
1470 /* Things built here have to last for the query duration. */
1471 oldcxt = MemoryContextSwitchTo(estate->es_query_cxt);
1472
1473 mtstate->mt_partition_tuple_routing =
1474 ExecSetupPartitionTupleRouting(estate, rootRel);
1475
1476 /*
1477 * Before a partition's tuple can be re-routed, it must first be
1478 * converted to the root's format, so we'll need a slot for storing
1479 * such tuples.
1480 */
1481 Assert(mtstate->mt_root_tuple_slot == NULL);
1482 mtstate->mt_root_tuple_slot = table_slot_create(rootRel, NULL);
1483
1484 MemoryContextSwitchTo(oldcxt);
1485 }
1486
1487 /*
1488 * Row movement, part 1. Delete the tuple, but skip RETURNING processing.
1489 * We want to return rows from INSERT.
1490 */
1491 ExecDelete(mtstate, resultRelInfo, tupleid, oldtuple, planSlot,
1492 epqstate, estate,
1493 false, /* processReturning */
1494 false, /* canSetTag */
1495 true, /* changingPart */
1496 &tuple_deleted, &epqslot);
1497
1498 /*
1499 * For some reason if DELETE didn't happen (e.g. trigger prevented it, or
1500 * it was already deleted by self, or it was concurrently deleted by
1501 * another transaction), then we should skip the insert as well;
1502 * otherwise, an UPDATE could cause an increase in the total number of
1503 * rows across all partitions, which is clearly wrong.
1504 *
1505 * For a normal UPDATE, the case where the tuple has been the subject of a
1506 * concurrent UPDATE or DELETE would be handled by the EvalPlanQual
1507 * machinery, but for an UPDATE that we've translated into a DELETE from
1508 * this partition and an INSERT into some other partition, that's not
1509 * available, because CTID chains can't span relation boundaries. We
1510 * mimic the semantics to a limited extent by skipping the INSERT if the
1511 * DELETE fails to find a tuple. This ensures that two concurrent
1512 * attempts to UPDATE the same tuple at the same time can't turn one tuple
1513 * into two, and that an UPDATE of a just-deleted tuple can't resurrect
1514 * it.
1515 */
1516 if (!tuple_deleted)
1517 {
1518 /*
1519 * epqslot will be typically NULL. But when ExecDelete() finds that
1520 * another transaction has concurrently updated the same row, it
1521 * re-fetches the row, skips the delete, and epqslot is set to the
1522 * re-fetched tuple slot. In that case, we need to do all the checks
1523 * again.
1524 */
1525 if (TupIsNull(epqslot))
1526 return true;
1527 else
1528 {
1529 /* Fetch the most recent version of old tuple. */
1530 TupleTableSlot *oldSlot;
1531
1532 /* ... but first, make sure ri_oldTupleSlot is initialized. */
1533 if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
1534 ExecInitUpdateProjection(mtstate, resultRelInfo);
1535 oldSlot = resultRelInfo->ri_oldTupleSlot;
1536 if (!table_tuple_fetch_row_version(resultRelInfo->ri_RelationDesc,
1537 tupleid,
1538 SnapshotAny,
1539 oldSlot))
1540 elog(ERROR, "failed to fetch tuple being updated");
1541 *retry_slot = ExecGetUpdateNewTuple(resultRelInfo, epqslot,
1542 oldSlot);
1543 return false;
1544 }
1545 }
1546
1547 /*
1548 * resultRelInfo is one of the per-relation resultRelInfos. So we should
1549 * convert the tuple into root's tuple descriptor if needed, since
1550 * ExecInsert() starts the search from root.
1551 */
1552 tupconv_map = ExecGetChildToRootMap(resultRelInfo);
1553 if (tupconv_map != NULL)
1554 slot = execute_attr_map_slot(tupconv_map->attrMap,
1555 slot,
1556 mtstate->mt_root_tuple_slot);
1557
1558 /* Tuple routing starts from the root table. */
1559 *inserted_tuple = ExecInsert(mtstate, mtstate->rootResultRelInfo, slot,
1560 planSlot, estate, canSetTag);
1561
1562 /*
1563 * Reset the transition state that may possibly have been written by
1564 * INSERT.
1565 */
1566 if (mtstate->mt_transition_capture)
1567 mtstate->mt_transition_capture->tcs_original_insert_tuple = NULL;
1568
1569 /* We're done moving. */
1570 return true;
1571 }
1572
1573 /* ----------------------------------------------------------------
1574 * ExecUpdate
1575 *
1576 * note: we can't run UPDATE queries with transactions
1577 * off because UPDATEs are actually INSERTs and our
1578 * scan will mistakenly loop forever, updating the tuple
1579 * it just inserted.. This should be fixed but until it
1580 * is, we don't want to get stuck in an infinite loop
1581 * which corrupts your database..
1582 *
1583 * When updating a table, tupleid identifies the tuple to
1584 * update and oldtuple is NULL. When updating a view, oldtuple
1585 * is passed to the INSTEAD OF triggers and identifies what to
1586 * update, and tupleid is invalid. When updating a foreign table,
1587 * tupleid is invalid; the FDW has to figure out which row to
1588 * update using data from the planSlot. oldtuple is passed to
1589 * foreign table triggers; it is NULL when the foreign table has
1590 * no relevant triggers.
1591 *
1592 * slot contains the new tuple value to be stored.
1593 * planSlot is the output of the ModifyTable's subplan; we use it
1594 * to access values from other input tables (for RETURNING),
1595 * row-ID junk columns, etc.
1596 *
1597 * Returns RETURNING result if any, otherwise NULL.
1598 * ----------------------------------------------------------------
1599 */
1600 static TupleTableSlot *
ExecUpdate(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo,ItemPointer tupleid,HeapTuple oldtuple,TupleTableSlot * slot,TupleTableSlot * planSlot,EPQState * epqstate,EState * estate,bool canSetTag)1601 ExecUpdate(ModifyTableState *mtstate,
1602 ResultRelInfo *resultRelInfo,
1603 ItemPointer tupleid,
1604 HeapTuple oldtuple,
1605 TupleTableSlot *slot,
1606 TupleTableSlot *planSlot,
1607 EPQState *epqstate,
1608 EState *estate,
1609 bool canSetTag)
1610 {
1611 Relation resultRelationDesc = resultRelInfo->ri_RelationDesc;
1612 TM_Result result;
1613 TM_FailureData tmfd;
1614 List *recheckIndexes = NIL;
1615
1616 /*
1617 * abort the operation if not running transactions
1618 */
1619 if (IsBootstrapProcessingMode())
1620 elog(ERROR, "cannot UPDATE during bootstrap");
1621
1622 ExecMaterializeSlot(slot);
1623
1624 /*
1625 * Open the table's indexes, if we have not done so already, so that we
1626 * can add new index entries for the updated tuple.
1627 */
1628 if (resultRelationDesc->rd_rel->relhasindex &&
1629 resultRelInfo->ri_IndexRelationDescs == NULL)
1630 ExecOpenIndices(resultRelInfo, false);
1631
1632 /* BEFORE ROW UPDATE Triggers */
1633 if (resultRelInfo->ri_TrigDesc &&
1634 resultRelInfo->ri_TrigDesc->trig_update_before_row)
1635 {
1636 if (!ExecBRUpdateTriggers(estate, epqstate, resultRelInfo,
1637 tupleid, oldtuple, slot))
1638 return NULL; /* "do nothing" */
1639 }
1640
1641 /* INSTEAD OF ROW UPDATE Triggers */
1642 if (resultRelInfo->ri_TrigDesc &&
1643 resultRelInfo->ri_TrigDesc->trig_update_instead_row)
1644 {
1645 if (!ExecIRUpdateTriggers(estate, resultRelInfo,
1646 oldtuple, slot))
1647 return NULL; /* "do nothing" */
1648 }
1649 else if (resultRelInfo->ri_FdwRoutine)
1650 {
1651 /*
1652 * GENERATED expressions might reference the tableoid column, so
1653 * (re-)initialize tts_tableOid before evaluating them.
1654 */
1655 slot->tts_tableOid = RelationGetRelid(resultRelInfo->ri_RelationDesc);
1656
1657 /*
1658 * Compute stored generated columns
1659 */
1660 if (resultRelationDesc->rd_att->constr &&
1661 resultRelationDesc->rd_att->constr->has_generated_stored)
1662 ExecComputeStoredGenerated(resultRelInfo, estate, slot,
1663 CMD_UPDATE);
1664
1665 /*
1666 * update in foreign table: let the FDW do it
1667 */
1668 slot = resultRelInfo->ri_FdwRoutine->ExecForeignUpdate(estate,
1669 resultRelInfo,
1670 slot,
1671 planSlot);
1672
1673 if (slot == NULL) /* "do nothing" */
1674 return NULL;
1675
1676 /*
1677 * AFTER ROW Triggers or RETURNING expressions might reference the
1678 * tableoid column, so (re-)initialize tts_tableOid before evaluating
1679 * them. (This covers the case where the FDW replaced the slot.)
1680 */
1681 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
1682 }
1683 else
1684 {
1685 LockTupleMode lockmode;
1686 bool partition_constraint_failed;
1687 bool update_indexes;
1688
1689 /*
1690 * Constraints and GENERATED expressions might reference the tableoid
1691 * column, so (re-)initialize tts_tableOid before evaluating them.
1692 */
1693 slot->tts_tableOid = RelationGetRelid(resultRelationDesc);
1694
1695 /*
1696 * Compute stored generated columns
1697 */
1698 if (resultRelationDesc->rd_att->constr &&
1699 resultRelationDesc->rd_att->constr->has_generated_stored)
1700 ExecComputeStoredGenerated(resultRelInfo, estate, slot,
1701 CMD_UPDATE);
1702
1703 /*
1704 * Check any RLS UPDATE WITH CHECK policies
1705 *
1706 * If we generate a new candidate tuple after EvalPlanQual testing, we
1707 * must loop back here and recheck any RLS policies and constraints.
1708 * (We don't need to redo triggers, however. If there are any BEFORE
1709 * triggers then trigger.c will have done table_tuple_lock to lock the
1710 * correct tuple, so there's no need to do them again.)
1711 */
1712 lreplace:;
1713
1714 /* ensure slot is independent, consider e.g. EPQ */
1715 ExecMaterializeSlot(slot);
1716
1717 /*
1718 * If partition constraint fails, this row might get moved to another
1719 * partition, in which case we should check the RLS CHECK policy just
1720 * before inserting into the new partition, rather than doing it here.
1721 * This is because a trigger on that partition might again change the
1722 * row. So skip the WCO checks if the partition constraint fails.
1723 */
1724 partition_constraint_failed =
1725 resultRelationDesc->rd_rel->relispartition &&
1726 !ExecPartitionCheck(resultRelInfo, slot, estate, false);
1727
1728 if (!partition_constraint_failed &&
1729 resultRelInfo->ri_WithCheckOptions != NIL)
1730 {
1731 /*
1732 * ExecWithCheckOptions() will skip any WCOs which are not of the
1733 * kind we are looking for at this point.
1734 */
1735 ExecWithCheckOptions(WCO_RLS_UPDATE_CHECK,
1736 resultRelInfo, slot, estate);
1737 }
1738
1739 /*
1740 * If a partition check failed, try to move the row into the right
1741 * partition.
1742 */
1743 if (partition_constraint_failed)
1744 {
1745 TupleTableSlot *inserted_tuple,
1746 *retry_slot;
1747 bool retry;
1748
1749 /*
1750 * ExecCrossPartitionUpdate will first DELETE the row from the
1751 * partition it's currently in and then insert it back into the
1752 * root table, which will re-route it to the correct partition.
1753 * The first part may have to be repeated if it is detected that
1754 * the tuple we're trying to move has been concurrently updated.
1755 */
1756 retry = !ExecCrossPartitionUpdate(mtstate, resultRelInfo, tupleid,
1757 oldtuple, slot, planSlot,
1758 epqstate, canSetTag,
1759 &retry_slot, &inserted_tuple);
1760 if (retry)
1761 {
1762 slot = retry_slot;
1763 goto lreplace;
1764 }
1765
1766 return inserted_tuple;
1767 }
1768
1769 /*
1770 * Check the constraints of the tuple. We've already checked the
1771 * partition constraint above; however, we must still ensure the tuple
1772 * passes all other constraints, so we will call ExecConstraints() and
1773 * have it validate all remaining checks.
1774 */
1775 if (resultRelationDesc->rd_att->constr)
1776 ExecConstraints(resultRelInfo, slot, estate);
1777
1778 /*
1779 * replace the heap tuple
1780 *
1781 * Note: if es_crosscheck_snapshot isn't InvalidSnapshot, we check
1782 * that the row to be updated is visible to that snapshot, and throw a
1783 * can't-serialize error if not. This is a special-case behavior
1784 * needed for referential integrity updates in transaction-snapshot
1785 * mode transactions.
1786 */
1787 result = table_tuple_update(resultRelationDesc, tupleid, slot,
1788 estate->es_output_cid,
1789 estate->es_snapshot,
1790 estate->es_crosscheck_snapshot,
1791 true /* wait for commit */ ,
1792 &tmfd, &lockmode, &update_indexes);
1793
1794 switch (result)
1795 {
1796 case TM_SelfModified:
1797
1798 /*
1799 * The target tuple was already updated or deleted by the
1800 * current command, or by a later command in the current
1801 * transaction. The former case is possible in a join UPDATE
1802 * where multiple tuples join to the same target tuple. This
1803 * is pretty questionable, but Postgres has always allowed it:
1804 * we just execute the first update action and ignore
1805 * additional update attempts.
1806 *
1807 * The latter case arises if the tuple is modified by a
1808 * command in a BEFORE trigger, or perhaps by a command in a
1809 * volatile function used in the query. In such situations we
1810 * should not ignore the update, but it is equally unsafe to
1811 * proceed. We don't want to discard the original UPDATE
1812 * while keeping the triggered actions based on it; and we
1813 * have no principled way to merge this update with the
1814 * previous ones. So throwing an error is the only safe
1815 * course.
1816 *
1817 * If a trigger actually intends this type of interaction, it
1818 * can re-execute the UPDATE (assuming it can figure out how)
1819 * and then return NULL to cancel the outer update.
1820 */
1821 if (tmfd.cmax != estate->es_output_cid)
1822 ereport(ERROR,
1823 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
1824 errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
1825 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
1826
1827 /* Else, already updated by self; nothing to do */
1828 return NULL;
1829
1830 case TM_Ok:
1831 break;
1832
1833 case TM_Updated:
1834 {
1835 TupleTableSlot *inputslot;
1836 TupleTableSlot *epqslot;
1837 TupleTableSlot *oldSlot;
1838
1839 if (IsolationUsesXactSnapshot())
1840 ereport(ERROR,
1841 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1842 errmsg("could not serialize access due to concurrent update")));
1843
1844 /*
1845 * Already know that we're going to need to do EPQ, so
1846 * fetch tuple directly into the right slot.
1847 */
1848 inputslot = EvalPlanQualSlot(epqstate, resultRelationDesc,
1849 resultRelInfo->ri_RangeTableIndex);
1850
1851 result = table_tuple_lock(resultRelationDesc, tupleid,
1852 estate->es_snapshot,
1853 inputslot, estate->es_output_cid,
1854 lockmode, LockWaitBlock,
1855 TUPLE_LOCK_FLAG_FIND_LAST_VERSION,
1856 &tmfd);
1857
1858 switch (result)
1859 {
1860 case TM_Ok:
1861 Assert(tmfd.traversed);
1862
1863 epqslot = EvalPlanQual(epqstate,
1864 resultRelationDesc,
1865 resultRelInfo->ri_RangeTableIndex,
1866 inputslot);
1867 if (TupIsNull(epqslot))
1868 /* Tuple not passing quals anymore, exiting... */
1869 return NULL;
1870
1871 /* Make sure ri_oldTupleSlot is initialized. */
1872 if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
1873 ExecInitUpdateProjection(mtstate, resultRelInfo);
1874
1875 /* Fetch the most recent version of old tuple. */
1876 oldSlot = resultRelInfo->ri_oldTupleSlot;
1877 if (!table_tuple_fetch_row_version(resultRelationDesc,
1878 tupleid,
1879 SnapshotAny,
1880 oldSlot))
1881 elog(ERROR, "failed to fetch tuple being updated");
1882 slot = ExecGetUpdateNewTuple(resultRelInfo,
1883 epqslot, oldSlot);
1884 goto lreplace;
1885
1886 case TM_Deleted:
1887 /* tuple already deleted; nothing to do */
1888 return NULL;
1889
1890 case TM_SelfModified:
1891
1892 /*
1893 * This can be reached when following an update
1894 * chain from a tuple updated by another session,
1895 * reaching a tuple that was already updated in
1896 * this transaction. If previously modified by
1897 * this command, ignore the redundant update,
1898 * otherwise error out.
1899 *
1900 * See also TM_SelfModified response to
1901 * table_tuple_update() above.
1902 */
1903 if (tmfd.cmax != estate->es_output_cid)
1904 ereport(ERROR,
1905 (errcode(ERRCODE_TRIGGERED_DATA_CHANGE_VIOLATION),
1906 errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
1907 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
1908 return NULL;
1909
1910 default:
1911 /* see table_tuple_lock call in ExecDelete() */
1912 elog(ERROR, "unexpected table_tuple_lock status: %u",
1913 result);
1914 return NULL;
1915 }
1916 }
1917
1918 break;
1919
1920 case TM_Deleted:
1921 if (IsolationUsesXactSnapshot())
1922 ereport(ERROR,
1923 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
1924 errmsg("could not serialize access due to concurrent delete")));
1925 /* tuple already deleted; nothing to do */
1926 return NULL;
1927
1928 default:
1929 elog(ERROR, "unrecognized table_tuple_update status: %u",
1930 result);
1931 return NULL;
1932 }
1933
1934 /* insert index entries for tuple if necessary */
1935 if (resultRelInfo->ri_NumIndices > 0 && update_indexes)
1936 recheckIndexes = ExecInsertIndexTuples(resultRelInfo,
1937 slot, estate, true, false,
1938 NULL, NIL);
1939 }
1940
1941 if (canSetTag)
1942 (estate->es_processed)++;
1943
1944 /* AFTER ROW UPDATE Triggers */
1945 ExecARUpdateTriggers(estate, resultRelInfo, tupleid, oldtuple, slot,
1946 recheckIndexes,
1947 mtstate->operation == CMD_INSERT ?
1948 mtstate->mt_oc_transition_capture :
1949 mtstate->mt_transition_capture);
1950
1951 list_free(recheckIndexes);
1952
1953 /*
1954 * Check any WITH CHECK OPTION constraints from parent views. We are
1955 * required to do this after testing all constraints and uniqueness
1956 * violations per the SQL spec, so we do it after actually updating the
1957 * record in the heap and all indexes.
1958 *
1959 * ExecWithCheckOptions() will skip any WCOs which are not of the kind we
1960 * are looking for at this point.
1961 */
1962 if (resultRelInfo->ri_WithCheckOptions != NIL)
1963 ExecWithCheckOptions(WCO_VIEW_CHECK, resultRelInfo, slot, estate);
1964
1965 /* Process RETURNING if present */
1966 if (resultRelInfo->ri_projectReturning)
1967 return ExecProcessReturning(resultRelInfo, slot, planSlot);
1968
1969 return NULL;
1970 }
1971
1972 /*
1973 * ExecOnConflictUpdate --- execute UPDATE of INSERT ON CONFLICT DO UPDATE
1974 *
1975 * Try to lock tuple for update as part of speculative insertion. If
1976 * a qual originating from ON CONFLICT DO UPDATE is satisfied, update
1977 * (but still lock row, even though it may not satisfy estate's
1978 * snapshot).
1979 *
1980 * Returns true if we're done (with or without an update), or false if
1981 * the caller must retry the INSERT from scratch.
1982 */
1983 static bool
ExecOnConflictUpdate(ModifyTableState * mtstate,ResultRelInfo * resultRelInfo,ItemPointer conflictTid,TupleTableSlot * planSlot,TupleTableSlot * excludedSlot,EState * estate,bool canSetTag,TupleTableSlot ** returning)1984 ExecOnConflictUpdate(ModifyTableState *mtstate,
1985 ResultRelInfo *resultRelInfo,
1986 ItemPointer conflictTid,
1987 TupleTableSlot *planSlot,
1988 TupleTableSlot *excludedSlot,
1989 EState *estate,
1990 bool canSetTag,
1991 TupleTableSlot **returning)
1992 {
1993 ExprContext *econtext = mtstate->ps.ps_ExprContext;
1994 Relation relation = resultRelInfo->ri_RelationDesc;
1995 ExprState *onConflictSetWhere = resultRelInfo->ri_onConflict->oc_WhereClause;
1996 TupleTableSlot *existing = resultRelInfo->ri_onConflict->oc_Existing;
1997 TM_FailureData tmfd;
1998 LockTupleMode lockmode;
1999 TM_Result test;
2000 Datum xminDatum;
2001 TransactionId xmin;
2002 bool isnull;
2003
2004 /* Determine lock mode to use */
2005 lockmode = ExecUpdateLockMode(estate, resultRelInfo);
2006
2007 /*
2008 * Lock tuple for update. Don't follow updates when tuple cannot be
2009 * locked without doing so. A row locking conflict here means our
2010 * previous conclusion that the tuple is conclusively committed is not
2011 * true anymore.
2012 */
2013 test = table_tuple_lock(relation, conflictTid,
2014 estate->es_snapshot,
2015 existing, estate->es_output_cid,
2016 lockmode, LockWaitBlock, 0,
2017 &tmfd);
2018 switch (test)
2019 {
2020 case TM_Ok:
2021 /* success! */
2022 break;
2023
2024 case TM_Invisible:
2025
2026 /*
2027 * This can occur when a just inserted tuple is updated again in
2028 * the same command. E.g. because multiple rows with the same
2029 * conflicting key values are inserted.
2030 *
2031 * This is somewhat similar to the ExecUpdate() TM_SelfModified
2032 * case. We do not want to proceed because it would lead to the
2033 * same row being updated a second time in some unspecified order,
2034 * and in contrast to plain UPDATEs there's no historical behavior
2035 * to break.
2036 *
2037 * It is the user's responsibility to prevent this situation from
2038 * occurring. These problems are why SQL-2003 similarly specifies
2039 * that for SQL MERGE, an exception must be raised in the event of
2040 * an attempt to update the same row twice.
2041 */
2042 xminDatum = slot_getsysattr(existing,
2043 MinTransactionIdAttributeNumber,
2044 &isnull);
2045 Assert(!isnull);
2046 xmin = DatumGetTransactionId(xminDatum);
2047
2048 if (TransactionIdIsCurrentTransactionId(xmin))
2049 ereport(ERROR,
2050 (errcode(ERRCODE_CARDINALITY_VIOLATION),
2051 errmsg("ON CONFLICT DO UPDATE command cannot affect row a second time"),
2052 errhint("Ensure that no rows proposed for insertion within the same command have duplicate constrained values.")));
2053
2054 /* This shouldn't happen */
2055 elog(ERROR, "attempted to lock invisible tuple");
2056 break;
2057
2058 case TM_SelfModified:
2059
2060 /*
2061 * This state should never be reached. As a dirty snapshot is used
2062 * to find conflicting tuples, speculative insertion wouldn't have
2063 * seen this row to conflict with.
2064 */
2065 elog(ERROR, "unexpected self-updated tuple");
2066 break;
2067
2068 case TM_Updated:
2069 if (IsolationUsesXactSnapshot())
2070 ereport(ERROR,
2071 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
2072 errmsg("could not serialize access due to concurrent update")));
2073
2074 /*
2075 * As long as we don't support an UPDATE of INSERT ON CONFLICT for
2076 * a partitioned table we shouldn't reach to a case where tuple to
2077 * be lock is moved to another partition due to concurrent update
2078 * of the partition key.
2079 */
2080 Assert(!ItemPointerIndicatesMovedPartitions(&tmfd.ctid));
2081
2082 /*
2083 * Tell caller to try again from the very start.
2084 *
2085 * It does not make sense to use the usual EvalPlanQual() style
2086 * loop here, as the new version of the row might not conflict
2087 * anymore, or the conflicting tuple has actually been deleted.
2088 */
2089 ExecClearTuple(existing);
2090 return false;
2091
2092 case TM_Deleted:
2093 if (IsolationUsesXactSnapshot())
2094 ereport(ERROR,
2095 (errcode(ERRCODE_T_R_SERIALIZATION_FAILURE),
2096 errmsg("could not serialize access due to concurrent delete")));
2097
2098 /* see TM_Updated case */
2099 Assert(!ItemPointerIndicatesMovedPartitions(&tmfd.ctid));
2100 ExecClearTuple(existing);
2101 return false;
2102
2103 default:
2104 elog(ERROR, "unrecognized table_tuple_lock status: %u", test);
2105 }
2106
2107 /* Success, the tuple is locked. */
2108
2109 /*
2110 * Verify that the tuple is visible to our MVCC snapshot if the current
2111 * isolation level mandates that.
2112 *
2113 * It's not sufficient to rely on the check within ExecUpdate() as e.g.
2114 * CONFLICT ... WHERE clause may prevent us from reaching that.
2115 *
2116 * This means we only ever continue when a new command in the current
2117 * transaction could see the row, even though in READ COMMITTED mode the
2118 * tuple will not be visible according to the current statement's
2119 * snapshot. This is in line with the way UPDATE deals with newer tuple
2120 * versions.
2121 */
2122 ExecCheckTupleVisible(estate, relation, existing);
2123
2124 /*
2125 * Make tuple and any needed join variables available to ExecQual and
2126 * ExecProject. The EXCLUDED tuple is installed in ecxt_innertuple, while
2127 * the target's existing tuple is installed in the scantuple. EXCLUDED
2128 * has been made to reference INNER_VAR in setrefs.c, but there is no
2129 * other redirection.
2130 */
2131 econtext->ecxt_scantuple = existing;
2132 econtext->ecxt_innertuple = excludedSlot;
2133 econtext->ecxt_outertuple = NULL;
2134
2135 if (!ExecQual(onConflictSetWhere, econtext))
2136 {
2137 ExecClearTuple(existing); /* see return below */
2138 InstrCountFiltered1(&mtstate->ps, 1);
2139 return true; /* done with the tuple */
2140 }
2141
2142 if (resultRelInfo->ri_WithCheckOptions != NIL)
2143 {
2144 /*
2145 * Check target's existing tuple against UPDATE-applicable USING
2146 * security barrier quals (if any), enforced here as RLS checks/WCOs.
2147 *
2148 * The rewriter creates UPDATE RLS checks/WCOs for UPDATE security
2149 * quals, and stores them as WCOs of "kind" WCO_RLS_CONFLICT_CHECK,
2150 * but that's almost the extent of its special handling for ON
2151 * CONFLICT DO UPDATE.
2152 *
2153 * The rewriter will also have associated UPDATE applicable straight
2154 * RLS checks/WCOs for the benefit of the ExecUpdate() call that
2155 * follows. INSERTs and UPDATEs naturally have mutually exclusive WCO
2156 * kinds, so there is no danger of spurious over-enforcement in the
2157 * INSERT or UPDATE path.
2158 */
2159 ExecWithCheckOptions(WCO_RLS_CONFLICT_CHECK, resultRelInfo,
2160 existing,
2161 mtstate->ps.state);
2162 }
2163
2164 /* Project the new tuple version */
2165 ExecProject(resultRelInfo->ri_onConflict->oc_ProjInfo);
2166
2167 /*
2168 * Note that it is possible that the target tuple has been modified in
2169 * this session, after the above table_tuple_lock. We choose to not error
2170 * out in that case, in line with ExecUpdate's treatment of similar cases.
2171 * This can happen if an UPDATE is triggered from within ExecQual(),
2172 * ExecWithCheckOptions() or ExecProject() above, e.g. by selecting from a
2173 * wCTE in the ON CONFLICT's SET.
2174 */
2175
2176 /* Execute UPDATE with projection */
2177 *returning = ExecUpdate(mtstate, resultRelInfo, conflictTid, NULL,
2178 resultRelInfo->ri_onConflict->oc_ProjSlot,
2179 planSlot,
2180 &mtstate->mt_epqstate, mtstate->ps.state,
2181 canSetTag);
2182
2183 /*
2184 * Clear out existing tuple, as there might not be another conflict among
2185 * the next input rows. Don't want to hold resources till the end of the
2186 * query.
2187 */
2188 ExecClearTuple(existing);
2189 return true;
2190 }
2191
2192
2193 /*
2194 * Process BEFORE EACH STATEMENT triggers
2195 */
2196 static void
fireBSTriggers(ModifyTableState * node)2197 fireBSTriggers(ModifyTableState *node)
2198 {
2199 ModifyTable *plan = (ModifyTable *) node->ps.plan;
2200 ResultRelInfo *resultRelInfo = node->rootResultRelInfo;
2201
2202 switch (node->operation)
2203 {
2204 case CMD_INSERT:
2205 ExecBSInsertTriggers(node->ps.state, resultRelInfo);
2206 if (plan->onConflictAction == ONCONFLICT_UPDATE)
2207 ExecBSUpdateTriggers(node->ps.state,
2208 resultRelInfo);
2209 break;
2210 case CMD_UPDATE:
2211 ExecBSUpdateTriggers(node->ps.state, resultRelInfo);
2212 break;
2213 case CMD_DELETE:
2214 ExecBSDeleteTriggers(node->ps.state, resultRelInfo);
2215 break;
2216 default:
2217 elog(ERROR, "unknown operation");
2218 break;
2219 }
2220 }
2221
2222 /*
2223 * Process AFTER EACH STATEMENT triggers
2224 */
2225 static void
fireASTriggers(ModifyTableState * node)2226 fireASTriggers(ModifyTableState *node)
2227 {
2228 ModifyTable *plan = (ModifyTable *) node->ps.plan;
2229 ResultRelInfo *resultRelInfo = node->rootResultRelInfo;
2230
2231 switch (node->operation)
2232 {
2233 case CMD_INSERT:
2234 if (plan->onConflictAction == ONCONFLICT_UPDATE)
2235 ExecASUpdateTriggers(node->ps.state,
2236 resultRelInfo,
2237 node->mt_oc_transition_capture);
2238 ExecASInsertTriggers(node->ps.state, resultRelInfo,
2239 node->mt_transition_capture);
2240 break;
2241 case CMD_UPDATE:
2242 ExecASUpdateTriggers(node->ps.state, resultRelInfo,
2243 node->mt_transition_capture);
2244 break;
2245 case CMD_DELETE:
2246 ExecASDeleteTriggers(node->ps.state, resultRelInfo,
2247 node->mt_transition_capture);
2248 break;
2249 default:
2250 elog(ERROR, "unknown operation");
2251 break;
2252 }
2253 }
2254
2255 /*
2256 * Set up the state needed for collecting transition tuples for AFTER
2257 * triggers.
2258 */
2259 static void
ExecSetupTransitionCaptureState(ModifyTableState * mtstate,EState * estate)2260 ExecSetupTransitionCaptureState(ModifyTableState *mtstate, EState *estate)
2261 {
2262 ModifyTable *plan = (ModifyTable *) mtstate->ps.plan;
2263 ResultRelInfo *targetRelInfo = mtstate->rootResultRelInfo;
2264
2265 /* Check for transition tables on the directly targeted relation. */
2266 mtstate->mt_transition_capture =
2267 MakeTransitionCaptureState(targetRelInfo->ri_TrigDesc,
2268 RelationGetRelid(targetRelInfo->ri_RelationDesc),
2269 mtstate->operation);
2270 if (plan->operation == CMD_INSERT &&
2271 plan->onConflictAction == ONCONFLICT_UPDATE)
2272 mtstate->mt_oc_transition_capture =
2273 MakeTransitionCaptureState(targetRelInfo->ri_TrigDesc,
2274 RelationGetRelid(targetRelInfo->ri_RelationDesc),
2275 CMD_UPDATE);
2276 }
2277
2278 /*
2279 * ExecPrepareTupleRouting --- prepare for routing one tuple
2280 *
2281 * Determine the partition in which the tuple in slot is to be inserted,
2282 * and return its ResultRelInfo in *partRelInfo. The return value is
2283 * a slot holding the tuple of the partition rowtype.
2284 *
2285 * This also sets the transition table information in mtstate based on the
2286 * selected partition.
2287 */
2288 static TupleTableSlot *
ExecPrepareTupleRouting(ModifyTableState * mtstate,EState * estate,PartitionTupleRouting * proute,ResultRelInfo * targetRelInfo,TupleTableSlot * slot,ResultRelInfo ** partRelInfo)2289 ExecPrepareTupleRouting(ModifyTableState *mtstate,
2290 EState *estate,
2291 PartitionTupleRouting *proute,
2292 ResultRelInfo *targetRelInfo,
2293 TupleTableSlot *slot,
2294 ResultRelInfo **partRelInfo)
2295 {
2296 ResultRelInfo *partrel;
2297 TupleConversionMap *map;
2298
2299 /*
2300 * Lookup the target partition's ResultRelInfo. If ExecFindPartition does
2301 * not find a valid partition for the tuple in 'slot' then an error is
2302 * raised. An error may also be raised if the found partition is not a
2303 * valid target for INSERTs. This is required since a partitioned table
2304 * UPDATE to another partition becomes a DELETE+INSERT.
2305 */
2306 partrel = ExecFindPartition(mtstate, targetRelInfo, proute, slot, estate);
2307
2308 /*
2309 * If we're capturing transition tuples, we might need to convert from the
2310 * partition rowtype to root partitioned table's rowtype. But if there
2311 * are no BEFORE triggers on the partition that could change the tuple, we
2312 * can just remember the original unconverted tuple to avoid a needless
2313 * round trip conversion.
2314 */
2315 if (mtstate->mt_transition_capture != NULL)
2316 {
2317 bool has_before_insert_row_trig;
2318
2319 has_before_insert_row_trig = (partrel->ri_TrigDesc &&
2320 partrel->ri_TrigDesc->trig_insert_before_row);
2321
2322 mtstate->mt_transition_capture->tcs_original_insert_tuple =
2323 !has_before_insert_row_trig ? slot : NULL;
2324 }
2325
2326 /*
2327 * Convert the tuple, if necessary.
2328 */
2329 map = partrel->ri_RootToPartitionMap;
2330 if (map != NULL)
2331 {
2332 TupleTableSlot *new_slot = partrel->ri_PartitionTupleSlot;
2333
2334 slot = execute_attr_map_slot(map->attrMap, slot, new_slot);
2335 }
2336
2337 *partRelInfo = partrel;
2338 return slot;
2339 }
2340
2341 /* ----------------------------------------------------------------
2342 * ExecModifyTable
2343 *
2344 * Perform table modifications as required, and return RETURNING results
2345 * if needed.
2346 * ----------------------------------------------------------------
2347 */
2348 static TupleTableSlot *
ExecModifyTable(PlanState * pstate)2349 ExecModifyTable(PlanState *pstate)
2350 {
2351 ModifyTableState *node = castNode(ModifyTableState, pstate);
2352 EState *estate = node->ps.state;
2353 CmdType operation = node->operation;
2354 ResultRelInfo *resultRelInfo;
2355 PlanState *subplanstate;
2356 TupleTableSlot *slot;
2357 TupleTableSlot *planSlot;
2358 TupleTableSlot *oldSlot;
2359 ItemPointer tupleid;
2360 ItemPointerData tuple_ctid;
2361 HeapTupleData oldtupdata;
2362 HeapTuple oldtuple;
2363 PartitionTupleRouting *proute = node->mt_partition_tuple_routing;
2364 List *relinfos = NIL;
2365 ListCell *lc;
2366
2367 CHECK_FOR_INTERRUPTS();
2368
2369 /*
2370 * This should NOT get called during EvalPlanQual; we should have passed a
2371 * subplan tree to EvalPlanQual, instead. Use a runtime test not just
2372 * Assert because this condition is easy to miss in testing. (Note:
2373 * although ModifyTable should not get executed within an EvalPlanQual
2374 * operation, we do have to allow it to be initialized and shut down in
2375 * case it is within a CTE subplan. Hence this test must be here, not in
2376 * ExecInitModifyTable.)
2377 */
2378 if (estate->es_epq_active != NULL)
2379 elog(ERROR, "ModifyTable should not be called during EvalPlanQual");
2380
2381 /*
2382 * If we've already completed processing, don't try to do more. We need
2383 * this test because ExecPostprocessPlan might call us an extra time, and
2384 * our subplan's nodes aren't necessarily robust against being called
2385 * extra times.
2386 */
2387 if (node->mt_done)
2388 return NULL;
2389
2390 /*
2391 * On first call, fire BEFORE STATEMENT triggers before proceeding.
2392 */
2393 if (node->fireBSTriggers)
2394 {
2395 fireBSTriggers(node);
2396 node->fireBSTriggers = false;
2397 }
2398
2399 /* Preload local variables */
2400 resultRelInfo = node->resultRelInfo + node->mt_lastResultIndex;
2401 subplanstate = outerPlanState(node);
2402
2403 /*
2404 * Fetch rows from subplan, and execute the required table modification
2405 * for each row.
2406 */
2407 for (;;)
2408 {
2409 /*
2410 * Reset the per-output-tuple exprcontext. This is needed because
2411 * triggers expect to use that context as workspace. It's a bit ugly
2412 * to do this below the top level of the plan, however. We might need
2413 * to rethink this later.
2414 */
2415 ResetPerTupleExprContext(estate);
2416
2417 /*
2418 * Reset per-tuple memory context used for processing on conflict and
2419 * returning clauses, to free any expression evaluation storage
2420 * allocated in the previous cycle.
2421 */
2422 if (pstate->ps_ExprContext)
2423 ResetExprContext(pstate->ps_ExprContext);
2424
2425 planSlot = ExecProcNode(subplanstate);
2426
2427 /* No more tuples to process? */
2428 if (TupIsNull(planSlot))
2429 break;
2430
2431 /*
2432 * When there are multiple result relations, each tuple contains a
2433 * junk column that gives the OID of the rel from which it came.
2434 * Extract it and select the correct result relation.
2435 */
2436 if (AttributeNumberIsValid(node->mt_resultOidAttno))
2437 {
2438 Datum datum;
2439 bool isNull;
2440 Oid resultoid;
2441
2442 datum = ExecGetJunkAttribute(planSlot, node->mt_resultOidAttno,
2443 &isNull);
2444 if (isNull)
2445 elog(ERROR, "tableoid is NULL");
2446 resultoid = DatumGetObjectId(datum);
2447
2448 /* If it's not the same as last time, we need to locate the rel */
2449 if (resultoid != node->mt_lastResultOid)
2450 resultRelInfo = ExecLookupResultRelByOid(node, resultoid,
2451 false, true);
2452 }
2453
2454 /*
2455 * If resultRelInfo->ri_usesFdwDirectModify is true, all we need to do
2456 * here is compute the RETURNING expressions.
2457 */
2458 if (resultRelInfo->ri_usesFdwDirectModify)
2459 {
2460 Assert(resultRelInfo->ri_projectReturning);
2461
2462 /*
2463 * A scan slot containing the data that was actually inserted,
2464 * updated or deleted has already been made available to
2465 * ExecProcessReturning by IterateDirectModify, so no need to
2466 * provide it here.
2467 */
2468 slot = ExecProcessReturning(resultRelInfo, NULL, planSlot);
2469
2470 return slot;
2471 }
2472
2473 EvalPlanQualSetSlot(&node->mt_epqstate, planSlot);
2474 slot = planSlot;
2475
2476 tupleid = NULL;
2477 oldtuple = NULL;
2478
2479 /*
2480 * For UPDATE/DELETE, fetch the row identity info for the tuple to be
2481 * updated/deleted. For a heap relation, that's a TID; otherwise we
2482 * may have a wholerow junk attr that carries the old tuple in toto.
2483 * Keep this in step with the part of ExecInitModifyTable that sets up
2484 * ri_RowIdAttNo.
2485 */
2486 if (operation == CMD_UPDATE || operation == CMD_DELETE)
2487 {
2488 char relkind;
2489 Datum datum;
2490 bool isNull;
2491
2492 relkind = resultRelInfo->ri_RelationDesc->rd_rel->relkind;
2493 if (relkind == RELKIND_RELATION ||
2494 relkind == RELKIND_MATVIEW ||
2495 relkind == RELKIND_PARTITIONED_TABLE)
2496 {
2497 /* ri_RowIdAttNo refers to a ctid attribute */
2498 Assert(AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo));
2499 datum = ExecGetJunkAttribute(slot,
2500 resultRelInfo->ri_RowIdAttNo,
2501 &isNull);
2502 /* shouldn't ever get a null result... */
2503 if (isNull)
2504 elog(ERROR, "ctid is NULL");
2505
2506 tupleid = (ItemPointer) DatumGetPointer(datum);
2507 tuple_ctid = *tupleid; /* be sure we don't free ctid!! */
2508 tupleid = &tuple_ctid;
2509 }
2510
2511 /*
2512 * Use the wholerow attribute, when available, to reconstruct the
2513 * old relation tuple. The old tuple serves one or both of two
2514 * purposes: 1) it serves as the OLD tuple for row triggers, 2) it
2515 * provides values for any unchanged columns for the NEW tuple of
2516 * an UPDATE, because the subplan does not produce all the columns
2517 * of the target table.
2518 *
2519 * Note that the wholerow attribute does not carry system columns,
2520 * so foreign table triggers miss seeing those, except that we
2521 * know enough here to set t_tableOid. Quite separately from
2522 * this, the FDW may fetch its own junk attrs to identify the row.
2523 *
2524 * Other relevant relkinds, currently limited to views, always
2525 * have a wholerow attribute.
2526 */
2527 else if (AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
2528 {
2529 datum = ExecGetJunkAttribute(slot,
2530 resultRelInfo->ri_RowIdAttNo,
2531 &isNull);
2532 /* shouldn't ever get a null result... */
2533 if (isNull)
2534 elog(ERROR, "wholerow is NULL");
2535
2536 oldtupdata.t_data = DatumGetHeapTupleHeader(datum);
2537 oldtupdata.t_len =
2538 HeapTupleHeaderGetDatumLength(oldtupdata.t_data);
2539 ItemPointerSetInvalid(&(oldtupdata.t_self));
2540 /* Historically, view triggers see invalid t_tableOid. */
2541 oldtupdata.t_tableOid =
2542 (relkind == RELKIND_VIEW) ? InvalidOid :
2543 RelationGetRelid(resultRelInfo->ri_RelationDesc);
2544
2545 oldtuple = &oldtupdata;
2546 }
2547 else
2548 {
2549 /* Only foreign tables are allowed to omit a row-ID attr */
2550 Assert(relkind == RELKIND_FOREIGN_TABLE);
2551 }
2552 }
2553
2554 switch (operation)
2555 {
2556 case CMD_INSERT:
2557 /* Initialize projection info if first time for this table */
2558 if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
2559 ExecInitInsertProjection(node, resultRelInfo);
2560 slot = ExecGetInsertNewTuple(resultRelInfo, planSlot);
2561 slot = ExecInsert(node, resultRelInfo, slot, planSlot,
2562 estate, node->canSetTag);
2563 break;
2564 case CMD_UPDATE:
2565 /* Initialize projection info if first time for this table */
2566 if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
2567 ExecInitUpdateProjection(node, resultRelInfo);
2568
2569 /*
2570 * Make the new tuple by combining plan's output tuple with
2571 * the old tuple being updated.
2572 */
2573 oldSlot = resultRelInfo->ri_oldTupleSlot;
2574 if (oldtuple != NULL)
2575 {
2576 /* Use the wholerow junk attr as the old tuple. */
2577 ExecForceStoreHeapTuple(oldtuple, oldSlot, false);
2578 }
2579 else
2580 {
2581 /* Fetch the most recent version of old tuple. */
2582 Relation relation = resultRelInfo->ri_RelationDesc;
2583
2584 Assert(tupleid != NULL);
2585 if (!table_tuple_fetch_row_version(relation, tupleid,
2586 SnapshotAny,
2587 oldSlot))
2588 elog(ERROR, "failed to fetch tuple being updated");
2589 }
2590 slot = ExecGetUpdateNewTuple(resultRelInfo, planSlot,
2591 oldSlot);
2592
2593 /* Now apply the update. */
2594 slot = ExecUpdate(node, resultRelInfo, tupleid, oldtuple, slot,
2595 planSlot, &node->mt_epqstate, estate,
2596 node->canSetTag);
2597 break;
2598 case CMD_DELETE:
2599 slot = ExecDelete(node, resultRelInfo, tupleid, oldtuple,
2600 planSlot, &node->mt_epqstate, estate,
2601 true, /* processReturning */
2602 node->canSetTag,
2603 false, /* changingPart */
2604 NULL, NULL);
2605 break;
2606 default:
2607 elog(ERROR, "unknown operation");
2608 break;
2609 }
2610
2611 /*
2612 * If we got a RETURNING result, return it to caller. We'll continue
2613 * the work on next call.
2614 */
2615 if (slot)
2616 return slot;
2617 }
2618
2619 /*
2620 * Insert remaining tuples for batch insert.
2621 */
2622 if (proute)
2623 relinfos = estate->es_tuple_routing_result_relations;
2624 else
2625 relinfos = estate->es_opened_result_relations;
2626
2627 foreach(lc, relinfos)
2628 {
2629 resultRelInfo = lfirst(lc);
2630 if (resultRelInfo->ri_NumSlots > 0)
2631 ExecBatchInsert(node, resultRelInfo,
2632 resultRelInfo->ri_Slots,
2633 resultRelInfo->ri_PlanSlots,
2634 resultRelInfo->ri_NumSlots,
2635 estate, node->canSetTag);
2636 }
2637
2638 /*
2639 * We're done, but fire AFTER STATEMENT triggers before exiting.
2640 */
2641 fireASTriggers(node);
2642
2643 node->mt_done = true;
2644
2645 return NULL;
2646 }
2647
2648 /*
2649 * ExecLookupResultRelByOid
2650 * If the table with given OID is among the result relations to be
2651 * updated by the given ModifyTable node, return its ResultRelInfo.
2652 *
2653 * If not found, return NULL if missing_ok, else raise error.
2654 *
2655 * If update_cache is true, then upon successful lookup, update the node's
2656 * one-element cache. ONLY ExecModifyTable may pass true for this.
2657 */
2658 ResultRelInfo *
ExecLookupResultRelByOid(ModifyTableState * node,Oid resultoid,bool missing_ok,bool update_cache)2659 ExecLookupResultRelByOid(ModifyTableState *node, Oid resultoid,
2660 bool missing_ok, bool update_cache)
2661 {
2662 if (node->mt_resultOidHash)
2663 {
2664 /* Use the pre-built hash table to locate the rel */
2665 MTTargetRelLookup *mtlookup;
2666
2667 mtlookup = (MTTargetRelLookup *)
2668 hash_search(node->mt_resultOidHash, &resultoid, HASH_FIND, NULL);
2669 if (mtlookup)
2670 {
2671 if (update_cache)
2672 {
2673 node->mt_lastResultOid = resultoid;
2674 node->mt_lastResultIndex = mtlookup->relationIndex;
2675 }
2676 return node->resultRelInfo + mtlookup->relationIndex;
2677 }
2678 }
2679 else
2680 {
2681 /* With few target rels, just search the ResultRelInfo array */
2682 for (int ndx = 0; ndx < node->mt_nrels; ndx++)
2683 {
2684 ResultRelInfo *rInfo = node->resultRelInfo + ndx;
2685
2686 if (RelationGetRelid(rInfo->ri_RelationDesc) == resultoid)
2687 {
2688 if (update_cache)
2689 {
2690 node->mt_lastResultOid = resultoid;
2691 node->mt_lastResultIndex = ndx;
2692 }
2693 return rInfo;
2694 }
2695 }
2696 }
2697
2698 if (!missing_ok)
2699 elog(ERROR, "incorrect result relation OID %u", resultoid);
2700 return NULL;
2701 }
2702
2703 /* ----------------------------------------------------------------
2704 * ExecInitModifyTable
2705 * ----------------------------------------------------------------
2706 */
2707 ModifyTableState *
ExecInitModifyTable(ModifyTable * node,EState * estate,int eflags)2708 ExecInitModifyTable(ModifyTable *node, EState *estate, int eflags)
2709 {
2710 ModifyTableState *mtstate;
2711 Plan *subplan = outerPlan(node);
2712 CmdType operation = node->operation;
2713 int nrels = list_length(node->resultRelations);
2714 ResultRelInfo *resultRelInfo;
2715 List *arowmarks;
2716 ListCell *l;
2717 int i;
2718 Relation rel;
2719
2720 /* check for unsupported flags */
2721 Assert(!(eflags & (EXEC_FLAG_BACKWARD | EXEC_FLAG_MARK)));
2722
2723 /*
2724 * create state structure
2725 */
2726 mtstate = makeNode(ModifyTableState);
2727 mtstate->ps.plan = (Plan *) node;
2728 mtstate->ps.state = estate;
2729 mtstate->ps.ExecProcNode = ExecModifyTable;
2730
2731 mtstate->operation = operation;
2732 mtstate->canSetTag = node->canSetTag;
2733 mtstate->mt_done = false;
2734
2735 mtstate->mt_nrels = nrels;
2736 mtstate->resultRelInfo = (ResultRelInfo *)
2737 palloc(nrels * sizeof(ResultRelInfo));
2738
2739 /*----------
2740 * Resolve the target relation. This is the same as:
2741 *
2742 * - the relation for which we will fire FOR STATEMENT triggers,
2743 * - the relation into whose tuple format all captured transition tuples
2744 * must be converted, and
2745 * - the root partitioned table used for tuple routing.
2746 *
2747 * If it's a partitioned table, the root partition doesn't appear
2748 * elsewhere in the plan and its RT index is given explicitly in
2749 * node->rootRelation. Otherwise (i.e. table inheritance) the target
2750 * relation is the first relation in the node->resultRelations list.
2751 *----------
2752 */
2753 if (node->rootRelation > 0)
2754 {
2755 mtstate->rootResultRelInfo = makeNode(ResultRelInfo);
2756 ExecInitResultRelation(estate, mtstate->rootResultRelInfo,
2757 node->rootRelation);
2758 }
2759 else
2760 {
2761 mtstate->rootResultRelInfo = mtstate->resultRelInfo;
2762 ExecInitResultRelation(estate, mtstate->resultRelInfo,
2763 linitial_int(node->resultRelations));
2764 }
2765
2766 /* set up epqstate with dummy subplan data for the moment */
2767 EvalPlanQualInit(&mtstate->mt_epqstate, estate, NULL, NIL, node->epqParam);
2768 mtstate->fireBSTriggers = true;
2769
2770 /*
2771 * Build state for collecting transition tuples. This requires having a
2772 * valid trigger query context, so skip it in explain-only mode.
2773 */
2774 if (!(eflags & EXEC_FLAG_EXPLAIN_ONLY))
2775 ExecSetupTransitionCaptureState(mtstate, estate);
2776
2777 /*
2778 * Open all the result relations and initialize the ResultRelInfo structs.
2779 * (But root relation was initialized above, if it's part of the array.)
2780 * We must do this before initializing the subplan, because direct-modify
2781 * FDWs expect their ResultRelInfos to be available.
2782 */
2783 resultRelInfo = mtstate->resultRelInfo;
2784 i = 0;
2785 foreach(l, node->resultRelations)
2786 {
2787 Index resultRelation = lfirst_int(l);
2788
2789 if (resultRelInfo != mtstate->rootResultRelInfo)
2790 {
2791 ExecInitResultRelation(estate, resultRelInfo, resultRelation);
2792
2793 /*
2794 * For child result relations, store the root result relation
2795 * pointer. We do so for the convenience of places that want to
2796 * look at the query's original target relation but don't have the
2797 * mtstate handy.
2798 */
2799 resultRelInfo->ri_RootResultRelInfo = mtstate->rootResultRelInfo;
2800 }
2801
2802 /* Initialize the usesFdwDirectModify flag */
2803 resultRelInfo->ri_usesFdwDirectModify = bms_is_member(i,
2804 node->fdwDirectModifyPlans);
2805
2806 /*
2807 * Verify result relation is a valid target for the current operation
2808 */
2809 CheckValidResultRel(resultRelInfo, operation);
2810
2811 resultRelInfo++;
2812 i++;
2813 }
2814
2815 /*
2816 * Now we may initialize the subplan.
2817 */
2818 outerPlanState(mtstate) = ExecInitNode(subplan, estate, eflags);
2819
2820 /*
2821 * Do additional per-result-relation initialization.
2822 */
2823 for (i = 0; i < nrels; i++)
2824 {
2825 resultRelInfo = &mtstate->resultRelInfo[i];
2826
2827 /* Let FDWs init themselves for foreign-table result rels */
2828 if (!resultRelInfo->ri_usesFdwDirectModify &&
2829 resultRelInfo->ri_FdwRoutine != NULL &&
2830 resultRelInfo->ri_FdwRoutine->BeginForeignModify != NULL)
2831 {
2832 List *fdw_private = (List *) list_nth(node->fdwPrivLists, i);
2833
2834 resultRelInfo->ri_FdwRoutine->BeginForeignModify(mtstate,
2835 resultRelInfo,
2836 fdw_private,
2837 i,
2838 eflags);
2839 }
2840
2841 /*
2842 * For UPDATE/DELETE, find the appropriate junk attr now, either a
2843 * 'ctid' or 'wholerow' attribute depending on relkind. For foreign
2844 * tables, the FDW might have created additional junk attr(s), but
2845 * those are no concern of ours.
2846 */
2847 if (operation == CMD_UPDATE || operation == CMD_DELETE)
2848 {
2849 char relkind;
2850
2851 relkind = resultRelInfo->ri_RelationDesc->rd_rel->relkind;
2852 if (relkind == RELKIND_RELATION ||
2853 relkind == RELKIND_MATVIEW ||
2854 relkind == RELKIND_PARTITIONED_TABLE)
2855 {
2856 resultRelInfo->ri_RowIdAttNo =
2857 ExecFindJunkAttributeInTlist(subplan->targetlist, "ctid");
2858 if (!AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
2859 elog(ERROR, "could not find junk ctid column");
2860 }
2861 else if (relkind == RELKIND_FOREIGN_TABLE)
2862 {
2863 /*
2864 * When there is a row-level trigger, there should be a
2865 * wholerow attribute. We also require it to be present in
2866 * UPDATE, so we can get the values of unchanged columns.
2867 */
2868 resultRelInfo->ri_RowIdAttNo =
2869 ExecFindJunkAttributeInTlist(subplan->targetlist,
2870 "wholerow");
2871 if (mtstate->operation == CMD_UPDATE &&
2872 !AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
2873 elog(ERROR, "could not find junk wholerow column");
2874 }
2875 else
2876 {
2877 /* Other valid target relkinds must provide wholerow */
2878 resultRelInfo->ri_RowIdAttNo =
2879 ExecFindJunkAttributeInTlist(subplan->targetlist,
2880 "wholerow");
2881 if (!AttributeNumberIsValid(resultRelInfo->ri_RowIdAttNo))
2882 elog(ERROR, "could not find junk wholerow column");
2883 }
2884 }
2885 }
2886
2887 /*
2888 * If this is an inherited update/delete, there will be a junk attribute
2889 * named "tableoid" present in the subplan's targetlist. It will be used
2890 * to identify the result relation for a given tuple to be
2891 * updated/deleted.
2892 */
2893 mtstate->mt_resultOidAttno =
2894 ExecFindJunkAttributeInTlist(subplan->targetlist, "tableoid");
2895 Assert(AttributeNumberIsValid(mtstate->mt_resultOidAttno) || nrels == 1);
2896 mtstate->mt_lastResultOid = InvalidOid; /* force lookup at first tuple */
2897 mtstate->mt_lastResultIndex = 0; /* must be zero if no such attr */
2898
2899 /* Get the root target relation */
2900 rel = mtstate->rootResultRelInfo->ri_RelationDesc;
2901
2902 /*
2903 * Build state for tuple routing if it's a partitioned INSERT. An UPDATE
2904 * might need this too, but only if it actually moves tuples between
2905 * partitions; in that case setup is done by ExecCrossPartitionUpdate.
2906 */
2907 if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE &&
2908 operation == CMD_INSERT)
2909 mtstate->mt_partition_tuple_routing =
2910 ExecSetupPartitionTupleRouting(estate, rel);
2911
2912 /*
2913 * Initialize any WITH CHECK OPTION constraints if needed.
2914 */
2915 resultRelInfo = mtstate->resultRelInfo;
2916 foreach(l, node->withCheckOptionLists)
2917 {
2918 List *wcoList = (List *) lfirst(l);
2919 List *wcoExprs = NIL;
2920 ListCell *ll;
2921
2922 foreach(ll, wcoList)
2923 {
2924 WithCheckOption *wco = (WithCheckOption *) lfirst(ll);
2925 ExprState *wcoExpr = ExecInitQual((List *) wco->qual,
2926 &mtstate->ps);
2927
2928 wcoExprs = lappend(wcoExprs, wcoExpr);
2929 }
2930
2931 resultRelInfo->ri_WithCheckOptions = wcoList;
2932 resultRelInfo->ri_WithCheckOptionExprs = wcoExprs;
2933 resultRelInfo++;
2934 }
2935
2936 /*
2937 * Initialize RETURNING projections if needed.
2938 */
2939 if (node->returningLists)
2940 {
2941 TupleTableSlot *slot;
2942 ExprContext *econtext;
2943
2944 /*
2945 * Initialize result tuple slot and assign its rowtype using the first
2946 * RETURNING list. We assume the rest will look the same.
2947 */
2948 mtstate->ps.plan->targetlist = (List *) linitial(node->returningLists);
2949
2950 /* Set up a slot for the output of the RETURNING projection(s) */
2951 ExecInitResultTupleSlotTL(&mtstate->ps, &TTSOpsVirtual);
2952 slot = mtstate->ps.ps_ResultTupleSlot;
2953
2954 /* Need an econtext too */
2955 if (mtstate->ps.ps_ExprContext == NULL)
2956 ExecAssignExprContext(estate, &mtstate->ps);
2957 econtext = mtstate->ps.ps_ExprContext;
2958
2959 /*
2960 * Build a projection for each result rel.
2961 */
2962 resultRelInfo = mtstate->resultRelInfo;
2963 foreach(l, node->returningLists)
2964 {
2965 List *rlist = (List *) lfirst(l);
2966
2967 resultRelInfo->ri_returningList = rlist;
2968 resultRelInfo->ri_projectReturning =
2969 ExecBuildProjectionInfo(rlist, econtext, slot, &mtstate->ps,
2970 resultRelInfo->ri_RelationDesc->rd_att);
2971 resultRelInfo++;
2972 }
2973 }
2974 else
2975 {
2976 /*
2977 * We still must construct a dummy result tuple type, because InitPlan
2978 * expects one (maybe should change that?).
2979 */
2980 mtstate->ps.plan->targetlist = NIL;
2981 ExecInitResultTypeTL(&mtstate->ps);
2982
2983 mtstate->ps.ps_ExprContext = NULL;
2984 }
2985
2986 /* Set the list of arbiter indexes if needed for ON CONFLICT */
2987 resultRelInfo = mtstate->resultRelInfo;
2988 if (node->onConflictAction != ONCONFLICT_NONE)
2989 {
2990 /* insert may only have one relation, inheritance is not expanded */
2991 Assert(nrels == 1);
2992 resultRelInfo->ri_onConflictArbiterIndexes = node->arbiterIndexes;
2993 }
2994
2995 /*
2996 * If needed, Initialize target list, projection and qual for ON CONFLICT
2997 * DO UPDATE.
2998 */
2999 if (node->onConflictAction == ONCONFLICT_UPDATE)
3000 {
3001 OnConflictSetState *onconfl = makeNode(OnConflictSetState);
3002 ExprContext *econtext;
3003 TupleDesc relationDesc;
3004
3005 /* already exists if created by RETURNING processing above */
3006 if (mtstate->ps.ps_ExprContext == NULL)
3007 ExecAssignExprContext(estate, &mtstate->ps);
3008
3009 econtext = mtstate->ps.ps_ExprContext;
3010 relationDesc = resultRelInfo->ri_RelationDesc->rd_att;
3011
3012 /* create state for DO UPDATE SET operation */
3013 resultRelInfo->ri_onConflict = onconfl;
3014
3015 /* initialize slot for the existing tuple */
3016 onconfl->oc_Existing =
3017 table_slot_create(resultRelInfo->ri_RelationDesc,
3018 &mtstate->ps.state->es_tupleTable);
3019
3020 /*
3021 * Create the tuple slot for the UPDATE SET projection. We want a slot
3022 * of the table's type here, because the slot will be used to insert
3023 * into the table, and for RETURNING processing - which may access
3024 * system attributes.
3025 */
3026 onconfl->oc_ProjSlot =
3027 table_slot_create(resultRelInfo->ri_RelationDesc,
3028 &mtstate->ps.state->es_tupleTable);
3029
3030 /* build UPDATE SET projection state */
3031 onconfl->oc_ProjInfo =
3032 ExecBuildUpdateProjection(node->onConflictSet,
3033 true,
3034 node->onConflictCols,
3035 relationDesc,
3036 econtext,
3037 onconfl->oc_ProjSlot,
3038 &mtstate->ps);
3039
3040 /* initialize state to evaluate the WHERE clause, if any */
3041 if (node->onConflictWhere)
3042 {
3043 ExprState *qualexpr;
3044
3045 qualexpr = ExecInitQual((List *) node->onConflictWhere,
3046 &mtstate->ps);
3047 onconfl->oc_WhereClause = qualexpr;
3048 }
3049 }
3050
3051 /*
3052 * If we have any secondary relations in an UPDATE or DELETE, they need to
3053 * be treated like non-locked relations in SELECT FOR UPDATE, ie, the
3054 * EvalPlanQual mechanism needs to be told about them. Locate the
3055 * relevant ExecRowMarks.
3056 */
3057 arowmarks = NIL;
3058 foreach(l, node->rowMarks)
3059 {
3060 PlanRowMark *rc = lfirst_node(PlanRowMark, l);
3061 ExecRowMark *erm;
3062 ExecAuxRowMark *aerm;
3063
3064 /* ignore "parent" rowmarks; they are irrelevant at runtime */
3065 if (rc->isParent)
3066 continue;
3067
3068 /* Find ExecRowMark and build ExecAuxRowMark */
3069 erm = ExecFindRowMark(estate, rc->rti, false);
3070 aerm = ExecBuildAuxRowMark(erm, subplan->targetlist);
3071 arowmarks = lappend(arowmarks, aerm);
3072 }
3073
3074 EvalPlanQualSetPlan(&mtstate->mt_epqstate, subplan, arowmarks);
3075
3076 /*
3077 * If there are a lot of result relations, use a hash table to speed the
3078 * lookups. If there are not a lot, a simple linear search is faster.
3079 *
3080 * It's not clear where the threshold is, but try 64 for starters. In a
3081 * debugging build, use a small threshold so that we get some test
3082 * coverage of both code paths.
3083 */
3084 #ifdef USE_ASSERT_CHECKING
3085 #define MT_NRELS_HASH 4
3086 #else
3087 #define MT_NRELS_HASH 64
3088 #endif
3089 if (nrels >= MT_NRELS_HASH)
3090 {
3091 HASHCTL hash_ctl;
3092
3093 hash_ctl.keysize = sizeof(Oid);
3094 hash_ctl.entrysize = sizeof(MTTargetRelLookup);
3095 hash_ctl.hcxt = CurrentMemoryContext;
3096 mtstate->mt_resultOidHash =
3097 hash_create("ModifyTable target hash",
3098 nrels, &hash_ctl,
3099 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
3100 for (i = 0; i < nrels; i++)
3101 {
3102 Oid hashkey;
3103 MTTargetRelLookup *mtlookup;
3104 bool found;
3105
3106 resultRelInfo = &mtstate->resultRelInfo[i];
3107 hashkey = RelationGetRelid(resultRelInfo->ri_RelationDesc);
3108 mtlookup = (MTTargetRelLookup *)
3109 hash_search(mtstate->mt_resultOidHash, &hashkey,
3110 HASH_ENTER, &found);
3111 Assert(!found);
3112 mtlookup->relationIndex = i;
3113 }
3114 }
3115 else
3116 mtstate->mt_resultOidHash = NULL;
3117
3118 /*
3119 * Determine if the FDW supports batch insert and determine the batch size
3120 * (a FDW may support batching, but it may be disabled for the
3121 * server/table).
3122 *
3123 * We only do this for INSERT, so that for UPDATE/DELETE the batch size
3124 * remains set to 0.
3125 */
3126 if (operation == CMD_INSERT)
3127 {
3128 /* insert may only have one relation, inheritance is not expanded */
3129 Assert(nrels == 1);
3130 resultRelInfo = mtstate->resultRelInfo;
3131 if (!resultRelInfo->ri_usesFdwDirectModify &&
3132 resultRelInfo->ri_FdwRoutine != NULL &&
3133 resultRelInfo->ri_FdwRoutine->GetForeignModifyBatchSize &&
3134 resultRelInfo->ri_FdwRoutine->ExecForeignBatchInsert)
3135 {
3136 resultRelInfo->ri_BatchSize =
3137 resultRelInfo->ri_FdwRoutine->GetForeignModifyBatchSize(resultRelInfo);
3138 Assert(resultRelInfo->ri_BatchSize >= 1);
3139 }
3140 else
3141 resultRelInfo->ri_BatchSize = 1;
3142 }
3143
3144 /*
3145 * Lastly, if this is not the primary (canSetTag) ModifyTable node, add it
3146 * to estate->es_auxmodifytables so that it will be run to completion by
3147 * ExecPostprocessPlan. (It'd actually work fine to add the primary
3148 * ModifyTable node too, but there's no need.) Note the use of lcons not
3149 * lappend: we need later-initialized ModifyTable nodes to be shut down
3150 * before earlier ones. This ensures that we don't throw away RETURNING
3151 * rows that need to be seen by a later CTE subplan.
3152 */
3153 if (!mtstate->canSetTag)
3154 estate->es_auxmodifytables = lcons(mtstate,
3155 estate->es_auxmodifytables);
3156
3157 return mtstate;
3158 }
3159
3160 /* ----------------------------------------------------------------
3161 * ExecEndModifyTable
3162 *
3163 * Shuts down the plan.
3164 *
3165 * Returns nothing of interest.
3166 * ----------------------------------------------------------------
3167 */
3168 void
ExecEndModifyTable(ModifyTableState * node)3169 ExecEndModifyTable(ModifyTableState *node)
3170 {
3171 int i;
3172
3173 /*
3174 * Allow any FDWs to shut down
3175 */
3176 for (i = 0; i < node->mt_nrels; i++)
3177 {
3178 int j;
3179 ResultRelInfo *resultRelInfo = node->resultRelInfo + i;
3180
3181 if (!resultRelInfo->ri_usesFdwDirectModify &&
3182 resultRelInfo->ri_FdwRoutine != NULL &&
3183 resultRelInfo->ri_FdwRoutine->EndForeignModify != NULL)
3184 resultRelInfo->ri_FdwRoutine->EndForeignModify(node->ps.state,
3185 resultRelInfo);
3186
3187 /*
3188 * Cleanup the initialized batch slots. This only matters for FDWs
3189 * with batching, but the other cases will have ri_NumSlotsInitialized
3190 * == 0.
3191 */
3192 for (j = 0; j < resultRelInfo->ri_NumSlotsInitialized; j++)
3193 {
3194 ExecDropSingleTupleTableSlot(resultRelInfo->ri_Slots[j]);
3195 ExecDropSingleTupleTableSlot(resultRelInfo->ri_PlanSlots[j]);
3196 }
3197 }
3198
3199 /*
3200 * Close all the partitioned tables, leaf partitions, and their indices
3201 * and release the slot used for tuple routing, if set.
3202 */
3203 if (node->mt_partition_tuple_routing)
3204 {
3205 ExecCleanupTupleRouting(node, node->mt_partition_tuple_routing);
3206
3207 if (node->mt_root_tuple_slot)
3208 ExecDropSingleTupleTableSlot(node->mt_root_tuple_slot);
3209 }
3210
3211 /*
3212 * Free the exprcontext
3213 */
3214 ExecFreeExprContext(&node->ps);
3215
3216 /*
3217 * clean out the tuple table
3218 */
3219 if (node->ps.ps_ResultTupleSlot)
3220 ExecClearTuple(node->ps.ps_ResultTupleSlot);
3221
3222 /*
3223 * Terminate EPQ execution if active
3224 */
3225 EvalPlanQualEnd(&node->mt_epqstate);
3226
3227 /*
3228 * shut down subplan
3229 */
3230 ExecEndNode(outerPlanState(node));
3231 }
3232
3233 void
ExecReScanModifyTable(ModifyTableState * node)3234 ExecReScanModifyTable(ModifyTableState *node)
3235 {
3236 /*
3237 * Currently, we don't need to support rescan on ModifyTable nodes. The
3238 * semantics of that would be a bit debatable anyway.
3239 */
3240 elog(ERROR, "ExecReScanModifyTable is not implemented");
3241 }
3242