| blob | 6f5a5262c7373b585db3e62eb34a714281a8efc9 |
1 /*-------------------------------------------------------------------------
2 *
3 * trigger.c
4 * PostgreSQL TRIGGERs support code.
5 *
6 * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
8 *
9 * IDENTIFICATION
10 * src/backend/commands/trigger.c
11 *
12 *-------------------------------------------------------------------------
13 */
69 /* GUC variables */
72 /* How many levels deep into trigger execution are we? */
75 /* Local function prototypes */
86 ItemPointer tid,
87 LockTupleMode lockmode,
99 MemoryContext per_tuple_context);
112 /*
113 * Create a trigger. Returns the address of the created trigger.
114 *
115 * queryString is the source text of the CREATE TRIGGER command.
116 * This must be supplied if a whenClause is specified, else it can be NULL.
117 *
118 * relOid, if nonzero, is the relation on which the trigger should be
119 * created. If zero, the name provided in the statement will be looked up.
120 *
121 * refRelOid, if nonzero, is the relation to which the constraint trigger
122 * refers. If zero, the constraint relation name provided in the statement
123 * will be looked up as needed.
124 *
125 * constraintOid, if nonzero, says that this trigger is being created
126 * internally to implement that constraint. A suitable pg_depend entry will
127 * be made to link the trigger to that constraint. constraintOid is zero when
128 * executing a user-entered CREATE TRIGGER command. (For CREATE CONSTRAINT
129 * TRIGGER, we build a pg_constraint entry internally.)
130 *
131 * indexOid, if nonzero, is the OID of an index associated with the constraint.
132 * We do nothing with this except store it into pg_trigger.tgconstrindid;
133 * but when creating a trigger for a deferrable unique constraint on a
134 * partitioned table, its children are looked up. Note we don't cope with
135 * invalid indexes in that case.
136 *
137 * funcoid, if nonzero, is the OID of the function to invoke. When this is
138 * given, stmt->funcname is ignored.
139 *
140 * parentTriggerOid, if nonzero, is a trigger that begets this one; so that
141 * if that trigger is dropped, this one should be too. There are two cases
142 * when a nonzero value is passed for this: 1) when this function recurses to
143 * create the trigger on partitions, 2) when creating child foreign key
144 * triggers; see CreateFKCheckTrigger() and createForeignKeyActionTriggers().
145 *
146 * If whenClause is passed, it is an already-transformed expression for
147 * WHEN. In this case, we ignore any that may come in stmt->whenClause.
148 *
149 * If isInternal is true then this is an internally-generated trigger.
150 * This argument sets the tgisinternal field of the pg_trigger entry, and
151 * if true causes us to modify the given trigger name to ensure uniqueness.
152 *
153 * When isInternal is not true we require ACL_TRIGGER permissions on the
154 * relation, as well as ACL_EXECUTE on the trigger function. For internal
155 * triggers the caller must apply any required permission checks.
156 *
157 * When called on partitioned tables, this function recurses to create the
158 * trigger on all the partitions, except if isInternal is true, in which
159 * case caller is expected to execute recursion on its own. in_partition
160 * indicates such a recursive call; outside callers should pass "false"
161 * (but see CloneRowTriggersToPartition).
162 */
163 ObjectAddress
168 {
169 return
174 }
176 /*
177 * Like the above; additionally the firing condition
178 * (always/origin/replica/disabled) can be specified.
179 */
180 ObjectAddress
186 {
187 int16 tgtype;
195 Relation rel;
196 AclResult aclresult;
197 Relation tgrel;
198 Relation pgrel;
200 Oid funcrettype;
205 ObjectAddress myself,
206 referenced;
217 else
220 /*
221 * Triggers must be on tables or views, and there are additional
222 * relation-type-specific restrictions.
223 */
225 {
226 /* Tables can't have INSTEAD OF triggers */
234 }
236 {
237 /* Partitioned tables can't have INSTEAD OF triggers */
246 /*
247 * FOR EACH ROW triggers have further restrictions
248 */
250 {
251 /*
252 * Disallow use of transition tables.
253 *
254 * Note that we have another restriction about transition tables
255 * in partitions; search for 'has_superclass' below for an
256 * explanation. The check here is just to protect from the fact
257 * that if we allowed it here, the creation would succeed for a
258 * partitioned table with no partitions, but would be blocked by
259 * the other restriction when the first partition was created,
260 * which is very unfriendly behavior.
261 */
268 }
269 }
271 {
272 /*
273 * Views can have INSTEAD OF triggers (which we check below are
274 * row-level), or statement-level BEFORE/AFTER triggers.
275 */
282 /* Disallow TRUNCATE triggers on VIEWs */
289 }
291 {
300 /*
301 * We disallow constraint triggers to protect the assumption that
302 * triggers on FKs can't be deferred. See notes with AfterTriggers
303 * data structures, below.
304 */
311 }
312 else
326 {
327 /*
328 * We must take a lock on the target relation to protect against
329 * concurrent drop. It's not clear that AccessShareLock is strong
330 * enough, but we certainly need at least that much... otherwise, we
331 * might end up creating a pg_constraint entry referencing a
332 * nonexistent table.
333 */
335 {
338 }
342 }
344 /* permission checks */
346 {
348 ACL_TRIGGER);
354 {
356 ACL_TRIGGER);
360 }
361 }
363 /*
364 * When called on a partitioned table to create a FOR EACH ROW trigger
365 * that's not internal, we create one trigger for each partition, too.
366 *
367 * For that, we'd better hold lock on all of them ahead of time.
368 */
375 /* Compute tgtype */
382 /* Disallow ROW-level TRUNCATE triggers */
388 /* INSTEAD triggers must be row-level, and can't have WHEN or columns */
390 {
403 }
405 /*
406 * We don't yet support naming ROW transition variables, but the parser
407 * recognizes the syntax so we can give a nicer message here.
408 *
409 * Per standard, REFERENCING TABLE names are only allowed on AFTER
410 * triggers. Per standard, REFERENCING ROW names are not allowed with FOR
411 * EACH STATEMENT. Per standard, each OLD/NEW, ROW/TABLE permutation is
412 * only allowed once. Per standard, OLD may not be specified when
413 * creating a trigger only for INSERT, and NEW may not be specified when
414 * creating a trigger only for DELETE.
415 *
416 * Notice that the standard allows an AFTER ... FOR EACH ROW trigger to
417 * reference both ROW and TABLE transition data.
418 */
420 {
425 {
434 /*
435 * Because of the above test, we omit further ROW-related testing
436 * below. If we later allow naming OLD and NEW ROW variables,
437 * adjustments will be needed below.
438 */
454 /*
455 * We currently don't allow row-level triggers with transition
456 * tables on partition or inheritance children. Such triggers
457 * would somehow need to see tuples converted to the format of the
458 * table they're attached to, and it's not clear which subset of
459 * tuples each child should see. See also the prohibitions in
460 * ATExecAttachPartition() and ATExecAddInherit().
461 */
463 {
464 /* Use appropriate error message. */
469 else
473 }
485 /*
486 * We currently don't allow multi-event triggers ("INSERT OR
487 * UPDATE") with transition tables, because it's not clear how to
488 * handle INSERT ... ON CONFLICT statements which can fire both
489 * INSERT and UPDATE triggers. We show the inserted tuples to
490 * INSERT triggers and the updated tuples to UPDATE triggers, but
491 * it's not yet clear what INSERT OR UPDATE trigger should see.
492 * This restriction could be lifted if we can decide on the right
493 * semantics in a later release.
494 */
502 /*
503 * We currently don't allow column-specific triggers with
504 * transition tables. Per spec, that seems to require
505 * accumulating separate transition tables for each combination of
506 * columns, which is a lot of work for a rather marginal feature.
507 */
513 /*
514 * We disallow constraint triggers with transition tables, to
515 * protect the assumption that such triggers can't be deferred.
516 * See notes with AfterTriggers data structures, below.
517 *
518 * Currently this is enforced by the grammar, so just Assert here.
519 */
523 {
536 }
537 else
538 {
551 }
552 }
559 }
561 /*
562 * Parse the WHEN clause, if any and we weren't passed an already
563 * transformed one.
564 *
565 * Note that as a side effect, we fill whenRtable when parsing. If we got
566 * an already parsed clause, this does not occur, which is what we want --
567 * no point in adding redundant dependencies below.
568 */
570 {
576 /* Set up a pstate to parse with */
580 /*
581 * Set up nsitems for OLD and NEW references.
582 *
583 * 'OLD' must always have varno equal to 1 and 'NEW' equal to 2.
584 */
586 AccessShareLock,
591 AccessShareLock,
596 /* Transform expression. Copy to be sure we don't modify original */
599 EXPR_KIND_TRIGGER_WHEN,
601 /* we have to fix its collations too */
604 /*
605 * Check for disallowed references to OLD/NEW.
606 *
607 * NB: pull_var_clause is okay here only because we don't allow
608 * subselects in WHEN clauses; it would fail to examine the contents
609 * of subselects.
610 */
613 {
617 {
629 /* system columns are okay here */
667 /* can't happen without add_missing_from, so just elog */
670 }
671 }
673 /* we'll need the rtable for recordDependencyOnExpr */
679 }
681 {
685 }
686 else
687 {
690 }
692 /*
693 * Find and validate the trigger function.
694 */
698 {
703 }
711 /*
712 * Scan pg_trigger to see if there is already a trigger of the same name.
713 * Skip this for internally generated triggers, since we'll modify the
714 * name to be unique below.
715 *
716 * NOTE that this is cool only because we have ShareRowExclusiveLock on
717 * the relation, so the trigger set won't be changing underneath us.
718 */
721 {
723 SysScanDesc tgscan;
726 Anum_pg_trigger_tgrelid,
731 Anum_pg_trigger_tgname,
738 /* There should be at most one matching tuple */
740 {
748 /* copy the tuple to use in CatalogTupleUpdate() */
750 }
752 }
755 {
756 /* Generate the OID for the new trigger. */
758 Anum_pg_trigger_oid);
759 }
760 else
761 {
762 /*
763 * If OR REPLACE was specified, we'll replace the old trigger;
764 * otherwise complain about the duplicate name.
765 */
772 /*
773 * An internal trigger or a child trigger (isClone) cannot be replaced
774 * by a user-defined trigger. However, skip this test when
775 * in_partition, because then we're recursing from a partitioned table
776 * and the check was made at the parent level.
777 */
785 /*
786 * It is not allowed to replace with a constraint trigger; gram.y
787 * should have enforced this already.
788 */
791 /*
792 * It is not allowed to replace an existing constraint trigger,
793 * either. (The reason for these restrictions is partly that it seems
794 * difficult to deal with pending trigger events in such cases, and
795 * partly that the command might imply changing the constraint's
796 * properties as well, which doesn't seem nice.)
797 */
803 }
805 /*
806 * If it's a user-entered CREATE CONSTRAINT TRIGGER command, make a
807 * corresponding pg_constraint entry.
808 */
810 {
811 /* Internal callers should have made their own constraints */
815 CONSTRAINT_TRIGGER,
827 NULL,
828 NULL,
829 NULL,
830 NULL,
834 NULL,
839 NULL,
844 }
846 /*
847 * If trigger is internally generated, modify the provided trigger name to
848 * ensure uniqueness by appending the trigger OID. (Callers will usually
849 * supply a simple constant trigger name in these cases.)
850 */
852 {
856 }
857 else
858 {
859 /* user-defined trigger; use the specified trigger name as-is */
861 }
863 /*
864 * Build the new pg_trigger tuple.
865 *
866 * When we're creating a trigger in a partition, we mark it as internal,
867 * even though we don't do the isInternal magic in this function. This
868 * makes the triggers in partitions identical to the ones in the
869 * partitioned tables, except that they are marked internal.
870 */
889 {
896 {
901 {
903 len++;
904 }
905 }
909 {
914 {
918 }
920 }
924 }
925 else
926 {
930 }
932 /* build column number array if it's a column-specific trigger */
936 else
937 {
943 {
945 int16 attnum;
948 /* Lookup column name. System columns are not allowed */
956 /* Check for duplicates */
958 {
963 name)));
964 }
967 }
968 }
972 /* set tgqual if trigger has WHEN clause */
975 else
981 else
986 else
989 /*
990 * Insert or replace tuple in pg_trigger.
991 */
993 {
996 }
997 else
998 {
999 HeapTuple newtup;
1004 }
1017 /*
1018 * Update relation's pg_class entry; if necessary; and if not, send an SI
1019 * message to make other backends (and this one) rebuild relcache entries.
1020 */
1028 {
1034 }
1035 else
1041 /*
1042 * If we're replacing a trigger, flush all the old dependencies before
1043 * recording new ones.
1044 */
1048 /*
1049 * Record dependencies for trigger. Always place a normal dependency on
1050 * the function.
1051 */
1062 {
1063 /*
1064 * Internally-generated trigger for a constraint, so make it an
1065 * internal dependency of the constraint. We can skip depending on
1066 * the relation(s), as there'll be an indirect dependency via the
1067 * constraint.
1068 */
1073 }
1074 else
1075 {
1076 /*
1077 * User CREATE TRIGGER, so place dependencies. We make trigger be
1078 * auto-dropped if its relation is dropped or if the FK relation is
1079 * dropped. (Auto drop is compatible with our pre-7.3 behavior.)
1080 */
1087 {
1092 }
1093 /* Not possible to have an index dependency in this case */
1096 /*
1097 * If it's a user-specified constraint trigger, make the constraint
1098 * internally dependent on the trigger instead of vice versa.
1099 */
1101 {
1106 }
1108 /*
1109 * If it's a partition trigger, create the partition dependencies.
1110 */
1112 {
1117 }
1118 }
1120 /* If column-specific trigger, add normal dependencies on columns */
1122 {
1128 {
1131 }
1132 }
1134 /*
1135 * If it has a WHEN clause, add dependencies on objects mentioned in the
1136 * expression (eg, functions, as well as any columns used).
1137 */
1140 DEPENDENCY_NORMAL);
1142 /* Post creation hook for new trigger */
1144 isInternal);
1146 /*
1147 * Lastly, create the trigger on child relations, if needed.
1148 */
1150 {
1153 MemoryContext oldcxt,
1154 perChildCxt;
1158 ALLOCSET_SMALL_SIZES);
1160 /*
1161 * We don't currently expect to be called with a valid indexOid. If
1162 * that ever changes then we'll need to write code here to find the
1163 * corresponding child index.
1164 */
1169 /* Iterate to create the trigger on each existing partition */
1171 {
1173 Relation childTbl;
1178 /*
1179 * Initialize our fabricated parse node by copying the original
1180 * one, then resetting fields that we pass separately.
1181 */
1186 /* If there is a WHEN clause, create a modified copy of it */
1204 }
1208 }
1210 /* Keep lock on target rel until end of xact */
1214 }
1216 /*
1217 * TriggerSetParentTrigger
1218 * Set a partition's trigger as child of its parent trigger,
1219 * or remove the linkage if parentTrigId is InvalidOid.
1220 *
1221 * This updates the constraint's pg_trigger row to show it as inherited, and
1222 * adds PARTITION dependencies to prevent the trigger from being deleted
1223 * on its own. Alternatively, reverse that.
1224 */
1225 void
1227 Oid childTrigId,
1228 Oid parentTrigId,
1229 Oid childTableId)
1230 {
1231 SysScanDesc tgscan;
1233 Form_pg_trigger trigForm;
1234 HeapTuple tuple,
1235 newtup;
1236 ObjectAddress depender;
1237 ObjectAddress referenced;
1239 /*
1240 * Find the trigger to delete.
1241 */
1243 Anum_pg_trigger_oid,
1256 {
1257 /* don't allow setting parent for a constraint that already has one */
1260 childTrigId);
1273 }
1274 else
1275 {
1281 TriggerRelationId,
1282 DEPENDENCY_PARTITION_PRI);
1284 RelationRelationId,
1285 DEPENDENCY_PARTITION_SEC);
1286 }
1290 }
1293 /*
1294 * Guts of trigger deletion.
1295 */
1296 void
1298 {
1299 Relation tgrel;
1300 SysScanDesc tgscan;
1302 HeapTuple tup;
1303 Oid relid;
1304 Relation rel;
1308 /*
1309 * Find the trigger to delete.
1310 */
1312 Anum_pg_trigger_oid,
1323 /*
1324 * Open and exclusive-lock the relation the trigger belongs to.
1325 */
1346 /*
1347 * Delete the pg_trigger tuple.
1348 */
1354 /*
1355 * We do not bother to try to determine whether any other triggers remain,
1356 * which would be needed in order to decide whether it's safe to clear the
1357 * relation's relhastriggers. (In any case, there might be a concurrent
1358 * process adding new triggers.) Instead, just force a relcache inval to
1359 * make other backends (and this one too!) rebuild their relcache entries.
1360 * There's no great harm in leaving relhastriggers true even if there are
1361 * no triggers left.
1362 */
1365 /* Keep lock on trigger's rel until end of xact */
1367 }
1369 /*
1370 * get_trigger_oid - Look up a trigger by name to find its OID.
1371 *
1372 * If missing_ok is false, throw an error if trigger not found. If
1373 * true, just return InvalidOid.
1374 */
1375 Oid
1377 {
1378 Relation tgrel;
1380 SysScanDesc tgscan;
1381 HeapTuple tup;
1382 Oid oid;
1384 /*
1385 * Find the trigger, verify permissions, set up object address
1386 */
1390 Anum_pg_trigger_tgrelid,
1394 Anum_pg_trigger_tgname,
1404 {
1411 }
1412 else
1413 {
1415 }
1420 }
1422 /*
1423 * Perform permissions and integrity checks before acquiring a relation lock.
1424 */
1425 static void
1428 {
1429 HeapTuple tuple;
1430 Form_pg_class form;
1437 /* only tables and views can have triggers */
1447 /* you must own the table to rename one of its triggers */
1457 }
1459 /*
1460 * renametrig - changes the name of a trigger on a relation
1461 *
1462 * trigger name is changed in trigger catalog.
1463 * No record of the previous name is kept.
1464 *
1465 * get proper relrelation from relation catalog (if not arg)
1466 * scan trigger catalog
1467 * for name conflict (within rel)
1468 * for original trigger (if not arg)
1469 * modify tgname in trigger tuple
1470 * update row in catalog
1471 */
1472 ObjectAddress
1474 {
1475 Oid tgoid;
1476 Relation targetrel;
1477 Relation tgrel;
1478 HeapTuple tuple;
1479 SysScanDesc tgscan;
1481 Oid relid;
1482 ObjectAddress address;
1484 /*
1485 * Look up name, check permissions, and acquire lock (which we will NOT
1486 * release until end of transaction).
1487 */
1490 RangeVarCallbackForRenameTrigger,
1491 NULL);
1493 /* Have lock already, so just need to build relcache entry. */
1496 /*
1497 * On partitioned tables, this operation recurses to partitions. Lock all
1498 * tables upfront.
1499 */
1505 /*
1506 * Search for the trigger to modify.
1507 */
1509 Anum_pg_trigger_tgrelid,
1513 Anum_pg_trigger_tgname,
1519 {
1520 Form_pg_trigger trigform;
1525 /*
1526 * If the trigger descends from a trigger on a parent partitioned
1527 * table, reject the rename. We don't allow a trigger in a partition
1528 * to differ in name from that of its parent: that would lead to an
1529 * inconsistency that pg_dump would not reproduce.
1530 */
1539 /* Rename the trigger on this relation ... */
1543 /* ... and if it is partitioned, recurse to its partitions */
1545 {
1549 {
1554 }
1555 }
1556 }
1557 else
1558 {
1563 }
1571 /*
1572 * Close rel, but keep exclusive lock!
1573 */
1577 }
1579 /*
1580 * Subroutine for renametrig -- perform the actual work of renaming one
1581 * trigger on one table.
1582 *
1583 * If the trigger has a name different from the expected one, raise a
1584 * NOTICE about it.
1585 */
1586 static void
1589 {
1590 HeapTuple tuple;
1591 Form_pg_trigger tgform;
1593 SysScanDesc tgscan;
1595 /* If the trigger already has the new name, nothing to do. */
1600 /*
1601 * Before actually trying the rename, search for triggers with the same
1602 * name. The update would fail with an ugly message in that case, and it
1603 * is better to throw a nicer error.
1604 */
1606 Anum_pg_trigger_tgrelid,
1610 Anum_pg_trigger_tgname,
1622 /*
1623 * The target name is free; update the existing pg_trigger tuple with it.
1624 */
1628 /*
1629 * If the trigger has a name different from what we expected, let the user
1630 * know. (We can proceed anyway, since we must have reached here following
1631 * a tgparentid link.)
1632 */
1645 /*
1646 * Invalidate relation's relcache entry so that other backends (and this
1647 * one too!) are sent SI message to make them rebuild relcache entries.
1648 * (Ideally this should happen automatically...)
1649 */
1651 }
1653 /*
1654 * Subroutine for renametrig -- Helper for recursing to partitions when
1655 * renaming triggers on a partitioned table.
1656 */
1657 static void
1660 {
1661 SysScanDesc tgscan;
1662 ScanKeyData key;
1663 HeapTuple tuple;
1665 /*
1666 * Given a relation and the OID of a trigger on parent relation, find the
1667 * corresponding trigger in the child and rename that trigger to the given
1668 * name.
1669 */
1671 Anum_pg_trigger_tgrelid,
1677 {
1679 Relation partitionRel;
1686 /* Rename the trigger on this partition */
1689 /* And if this relation is partitioned, recurse to its partitions */
1691 {
1696 {
1701 }
1702 }
1705 /* There should be at most one matching tuple */
1707 }
1709 }
1711 /*
1712 * EnableDisableTrigger()
1713 *
1714 * Called by ALTER TABLE ENABLE/DISABLE [ REPLICA | ALWAYS ] TRIGGER
1715 * to change 'tgenabled' field for the specified trigger(s)
1716 *
1717 * rel: relation to process (caller must hold suitable lock on it)
1718 * tgname: trigger to process, or NULL to scan all triggers
1719 * fires_when: new value for tgenabled field. In addition to generic
1720 * enablement/disablement, this also defines when the trigger
1721 * should be fired in session replication roles.
1722 * skip_system: if true, skip "system" triggers (constraint triggers)
1723 * recurse: if true, recurse to partitions
1724 *
1725 * Caller should have checked permissions for the table; here we also
1726 * enforce that superuser privilege is required to alter the state of
1727 * system triggers
1728 */
1729 void
1732 LOCKMODE lockmode)
1733 {
1734 Relation tgrel;
1737 SysScanDesc tgscan;
1738 HeapTuple tuple;
1742 /* Scan the relevant entries in pg_triggers */
1746 Anum_pg_trigger_tgrelid,
1750 {
1752 Anum_pg_trigger_tgname,
1756 }
1757 else
1766 {
1770 {
1771 /* system trigger ... ok to process? */
1779 }
1784 {
1785 /* need to change this one ... make a copy to scribble on */
1796 }
1798 /*
1799 * When altering FOR EACH ROW triggers on a partitioned table, do the
1800 * same on the partitions as well, unless ONLY is specified.
1801 *
1802 * Note that we recurse even if we didn't change the trigger above,
1803 * because the partitions' copy of the trigger may have a different
1804 * value of tgenabled than the parent's trigger and thus might need to
1805 * be changed.
1806 */
1810 {
1815 {
1816 Relation part;
1821 lockmode);
1823 }
1824 }
1828 }
1840 /*
1841 * If we changed anything, broadcast a SI inval message to force each
1842 * backend (including our own!) to rebuild relation's relcache entry.
1843 * Otherwise they will fail to apply the change promptly.
1844 */
1847 }
1850 /*
1851 * Build trigger data to attach to the given relcache entry.
1852 *
1853 * Note that trigger data attached to a relcache entry must be stored in
1854 * CacheMemoryContext to ensure it survives as long as the relcache entry.
1855 * But we should be running in a less long-lived working context. To avoid
1856 * leaking cache memory if this routine fails partway through, we build a
1857 * temporary TriggerDesc in working memory and then copy the completed
1858 * structure into cache memory.
1859 */
1860 void
1862 {
1867 Relation tgrel;
1868 ScanKeyData skey;
1869 SysScanDesc tgscan;
1870 HeapTuple htup;
1871 MemoryContext oldContext;
1874 /*
1875 * Allocate a working array to hold the triggers (the array is extended if
1876 * necessary)
1877 */
1882 /*
1883 * Note: since we scan the triggers using TriggerRelidNameIndexId, we will
1884 * be reading the triggers in name order, except possibly during
1885 * emergency-recovery operations (ie, IgnoreSystemIndexes). This in turn
1886 * ensures that triggers will be fired in name order.
1887 */
1889 Anum_pg_trigger_tgrelid,
1898 {
1901 Datum datum;
1905 {
1908 }
1925 /* tgattr is first var-width field, so OK to access directly */
1928 {
1932 }
1933 else
1936 {
1941 Anum_pg_trigger_tgargs,
1949 {
1952 }
1953 }
1954 else
1962 else
1970 else
1977 else
1980 numtrigs++;
1981 }
1986 /* There might not be any triggers */
1988 {
1991 }
1993 /* Build trigdesc */
2000 /* Copy completed trigdesc into cache storage */
2005 /* Release working memory */
2007 }
2009 /*
2010 * Update the TriggerDesc's hint flags to include the specified trigger
2011 */
2012 static void
2014 {
2062 /* there are no row-level truncate triggers */
2082 }
2084 /*
2085 * Copy a TriggerDesc data structure.
2086 *
2087 * The copy is allocated in the current memory context.
2088 */
2089 TriggerDesc *
2091 {
2108 {
2111 {
2118 }
2120 {
2122 int16 j;
2128 }
2135 trigger++;
2136 }
2139 }
2141 /*
2142 * Free a TriggerDesc data structure.
2143 */
2144 void
2146 {
2155 {
2160 {
2164 }
2171 trigger++;
2172 }
2175 }
2177 /*
2178 * Compare two TriggerDesc structures for logical equality.
2179 */
2180 #ifdef NOT_USED
2181 bool
2183 {
2185 j;
2187 /*
2188 * We need not examine the hint flags, just the trigger array itself; if
2189 * we have the same triggers with the same types, the flags should match.
2190 *
2191 * As of 7.3 we assume trigger set ordering is significant in the
2192 * comparison; so we just compare corresponding slots of the two sets.
2193 *
2194 * Note: comparing the stringToNode forms of the WHEN clauses means that
2195 * parse column locations will affect the result. This is okay as long as
2196 * this function is only used for detecting exact equality, as for example
2197 * in checking for staleness of a cache entry.
2198 */
2200 {
2206 {
2263 }
2264 }
2268 }
2271 /*
2272 * Check if there is a row-level trigger with transition tables that prevents
2273 * a table from becoming an inheritance child or partition. Return the name
2274 * of the first such incompatible trigger, or NULL if there is none.
2275 */
2278 {
2280 {
2284 {
2289 }
2290 }
2293 }
2295 /*
2296 * Call a trigger function.
2297 *
2298 * trigdata: trigger descriptor.
2299 * tgindx: trigger's index in finfo and instr arrays.
2300 * finfo: array of cached trigger function call information.
2301 * instr: optional array of EXPLAIN ANALYZE instrumentation state.
2302 * per_tuple_context: memory context to execute the function in.
2303 *
2304 * Returns the tuple (or NULL) as returned by the function.
2305 */
2306 static HeapTuple
2311 MemoryContext per_tuple_context)
2312 {
2314 PgStat_FunctionCallUsage fcusage;
2315 Datum result;
2316 MemoryContext oldContext;
2318 /*
2319 * Protect against code paths that may fail to initialize transition table
2320 * info.
2321 */
2332 /*
2333 * We cache fmgr lookup info, to avoid making the lookup again on each
2334 * call.
2335 */
2341 /*
2342 * If doing EXPLAIN ANALYZE, start charging time to this trigger.
2343 */
2347 /*
2348 * Do the function evaluation in the per-tuple memory context, so that
2349 * leaked memory will be reclaimed once per tuple. Note in particular that
2350 * any new tuple created by the trigger function will live till the end of
2351 * the tuple cycle.
2352 */
2355 /*
2356 * Call the function, passing no arguments but setting a context.
2357 */
2363 MyTriggerDepth++;
2365 {
2367 }
2369 {
2370 MyTriggerDepth--;
2371 }
2378 /*
2379 * Trigger protocol allows function to return a null pointer, but NOT to
2380 * set the isnull result flag.
2381 */
2388 /*
2389 * If doing EXPLAIN ANALYZE, stop charging time to this trigger, and count
2390 * one "tuple returned" (really the number of firings).
2391 */
2396 }
2398 void
2400 {
2412 /* no-op if we already fired BS triggers in this context */
2414 CMD_INSERT))
2419 TRIGGER_EVENT_BEFORE;
2422 {
2424 HeapTuple newtuple;
2427 TRIGGER_TYPE_STATEMENT,
2428 TRIGGER_TYPE_BEFORE,
2429 TRIGGER_TYPE_INSERT))
2437 i,
2446 }
2447 }
2449 void
2452 {
2457 TRIGGER_EVENT_INSERT,
2460 }
2462 bool
2465 {
2474 TRIGGER_EVENT_ROW |
2475 TRIGGER_EVENT_BEFORE;
2478 {
2480 HeapTuple oldtuple;
2483 TRIGGER_TYPE_ROW,
2484 TRIGGER_TYPE_BEFORE,
2485 TRIGGER_TYPE_INSERT))
2498 i,
2503 {
2507 }
2509 {
2512 /*
2513 * After a tuple in a partition goes through a trigger, the user
2514 * could have changed the partition key enough that the tuple no
2515 * longer fits the partition. Verify that.
2516 */
2521 errmsg("moving row to another partition during a BEFORE FOR EACH ROW trigger is not supported"),
2530 /* signal tuple should be re-fetched if used */
2532 }
2533 }
2536 }
2538 void
2542 {
2548 TRIGGER_EVENT_INSERT,
2551 transition_capture,
2553 }
2555 bool
2558 {
2567 TRIGGER_EVENT_ROW |
2568 TRIGGER_EVENT_INSTEAD;
2571 {
2573 HeapTuple oldtuple;
2576 TRIGGER_TYPE_ROW,
2577 TRIGGER_TYPE_INSTEAD,
2578 TRIGGER_TYPE_INSERT))
2591 i,
2596 {
2600 }
2602 {
2608 /* signal tuple should be re-fetched if used */
2610 }
2611 }
2614 }
2616 void
2618 {
2630 /* no-op if we already fired BS triggers in this context */
2632 CMD_DELETE))
2637 TRIGGER_EVENT_BEFORE;
2640 {
2642 HeapTuple newtuple;
2645 TRIGGER_TYPE_STATEMENT,
2646 TRIGGER_TYPE_BEFORE,
2647 TRIGGER_TYPE_DELETE))
2655 i,
2664 }
2665 }
2667 void
2670 {
2675 TRIGGER_EVENT_DELETE,
2678 }
2680 /*
2681 * Execute BEFORE ROW DELETE triggers.
2682 *
2683 * True indicates caller can proceed with the delete. False indicates caller
2684 * need to suppress the delete and additionally if requested, we need to pass
2685 * back the concurrently updated tuple if any.
2686 */
2687 bool
2690 ItemPointer tupleid,
2691 HeapTuple fdw_trigtuple,
2693 {
2698 HeapTuple trigtuple;
2704 {
2709 NULL))
2712 /*
2713 * If the tuple was concurrently updated and the caller of this
2714 * function requested for the updated tuple, skip the trigger
2715 * execution.
2716 */
2718 {
2721 }
2724 }
2725 else
2726 {
2729 }
2733 TRIGGER_EVENT_ROW |
2734 TRIGGER_EVENT_BEFORE;
2737 {
2738 HeapTuple newtuple;
2742 TRIGGER_TYPE_ROW,
2743 TRIGGER_TYPE_BEFORE,
2744 TRIGGER_TYPE_DELETE))
2754 i,
2759 {
2762 }
2765 }
2770 }
2772 /*
2773 * Note: is_crosspart_update must be true if the DELETE is being performed
2774 * as part of a cross-partition update.
2775 */
2776 void
2779 ItemPointer tupleid,
2780 HeapTuple fdw_trigtuple,
2783 {
2788 {
2794 NULL,
2795 relinfo,
2796 tupleid,
2797 LockTupleExclusive,
2798 slot,
2799 NULL,
2800 NULL);
2801 else
2805 TRIGGER_EVENT_DELETE,
2807 transition_capture,
2808 is_crosspart_update);
2809 }
2810 }
2812 bool
2814 HeapTuple trigtuple)
2815 {
2823 TRIGGER_EVENT_ROW |
2824 TRIGGER_EVENT_INSTEAD;
2830 {
2831 HeapTuple rettuple;
2835 TRIGGER_TYPE_ROW,
2836 TRIGGER_TYPE_INSTEAD,
2837 TRIGGER_TYPE_DELETE))
2847 i,
2855 }
2857 }
2859 void
2861 {
2874 /* no-op if we already fired BS triggers in this context */
2876 CMD_UPDATE))
2879 /* statement-level triggers operate on the parent table */
2886 TRIGGER_EVENT_BEFORE;
2890 {
2892 HeapTuple newtuple;
2895 TRIGGER_TYPE_STATEMENT,
2896 TRIGGER_TYPE_BEFORE,
2897 TRIGGER_TYPE_UPDATE))
2905 i,
2914 }
2915 }
2917 void
2920 {
2923 /* statement-level triggers operate on the parent table */
2928 TRIGGER_EVENT_UPDATE,
2931 transition_capture,
2933 }
2935 bool
2938 ItemPointer tupleid,
2939 HeapTuple fdw_trigtuple,
2942 {
2946 HeapTuple trigtuple;
2952 LockTupleMode lockmode;
2954 /* Determine lock mode to use */
2959 {
2962 /* get a copy of the on-disk tuple we are planning to update */
2965 tmfd))
2968 /*
2969 * In READ COMMITTED isolation level it's possible that target tuple
2970 * was changed due to concurrent update. In that case we have a raw
2971 * subplan output tuple in epqslot_candidate, and need to form a new
2972 * insertable tuple using ExecGetUpdateNewTuple to replace the one we
2973 * received in newslot. Neither we nor our callers have any further
2974 * interest in the passed-in tuple, so it's okay to overwrite newslot
2975 * with the newer data.
2976 *
2977 * (Typically, newslot was also generated by ExecGetUpdateNewTuple, so
2978 * that epqslot_clean will be that same slot and the copy step below
2979 * is not needed.)
2980 */
2982 {
2986 oldslot);
2990 }
2993 }
2994 else
2995 {
2998 }
3002 TRIGGER_EVENT_ROW |
3003 TRIGGER_EVENT_BEFORE;
3008 {
3010 HeapTuple oldtuple;
3013 TRIGGER_TYPE_ROW,
3014 TRIGGER_TYPE_BEFORE,
3015 TRIGGER_TYPE_UPDATE))
3030 i,
3036 {
3042 }
3044 {
3047 /*
3048 * If the tuple returned by the trigger / being stored, is the old
3049 * row version, and the heap tuple passed to the trigger was
3050 * allocated locally, materialize the slot. Otherwise we might
3051 * free it while still referenced by the slot.
3052 */
3059 /* signal tuple should be re-fetched if used */
3061 }
3062 }
3067 }
3069 /*
3070 * Note: 'src_partinfo' and 'dst_partinfo', when non-NULL, refer to the source
3071 * and destination partitions, respectively, of a cross-partition update of
3072 * the root partitioned table mentioned in the query, given by 'relinfo'.
3073 * 'tupleid' in that case refers to the ctid of the "old" tuple in the source
3074 * partition, and 'newslot' contains the "new" tuple in the destination
3075 * partition. This interface allows to support the requirements of
3076 * ExecCrossPartitionUpdateForeignKey(); is_crosspart_update must be true in
3077 * that case.
3078 */
3079 void
3083 ItemPointer tupleid,
3084 HeapTuple fdw_trigtuple,
3089 {
3096 {
3097 /*
3098 * Note: if the UPDATE is converted into a DELETE+INSERT as part of
3099 * update-partition-key operation, then this function is also called
3100 * separately for DELETE and INSERT to capture transition table rows.
3101 * In such case, either old tuple or new tuple can be NULL.
3102 */
3114 NULL,
3115 tupsrc,
3116 tupleid,
3117 LockTupleExclusive,
3118 oldslot,
3119 NULL,
3120 NULL);
3123 else
3128 TRIGGER_EVENT_UPDATE,
3132 transition_capture,
3133 is_crosspart_update);
3134 }
3135 }
3137 bool
3140 {
3150 TRIGGER_EVENT_ROW |
3151 TRIGGER_EVENT_INSTEAD;
3157 {
3159 HeapTuple oldtuple;
3162 TRIGGER_TYPE_ROW,
3163 TRIGGER_TYPE_INSTEAD,
3164 TRIGGER_TYPE_UPDATE))
3180 i,
3185 {
3187 }
3189 {
3195 /* signal tuple should be re-fetched if used */
3197 }
3198 }
3201 }
3203 void
3205 {
3219 TRIGGER_EVENT_BEFORE;
3223 {
3225 HeapTuple newtuple;
3228 TRIGGER_TYPE_STATEMENT,
3229 TRIGGER_TYPE_BEFORE,
3230 TRIGGER_TYPE_TRUNCATE))
3238 i,
3247 }
3248 }
3250 void
3252 {
3258 TRIGGER_EVENT_TRUNCATE,
3261 }
3264 /*
3265 * Fetch tuple into "oldslot", dealing with locking and EPQ if necessary
3266 */
3267 static bool
3271 ItemPointer tid,
3272 LockTupleMode lockmode,
3276 {
3280 {
3281 TM_Result test;
3282 TM_FailureData tmfd;
3287 /* caller must pass an epqstate if EvalPlanQual is possible */
3290 /*
3291 * lock tuple for update
3292 */
3298 lockflags,
3301 /* Let the caller know about the status of this operation */
3306 {
3309 /*
3310 * The target tuple was already updated or deleted by the
3311 * current command, or by a later command in the current
3312 * transaction. We ignore the tuple in the former case, and
3313 * throw error in the latter case, for the same reasons
3314 * enumerated in ExecUpdate and ExecDelete in
3315 * nodeModifyTable.c.
3316 */
3320 errmsg("tuple to be updated was already modified by an operation triggered by the current command"),
3321 errhint("Consider using an AFTER trigger instead of a BEFORE trigger to propagate changes to other rows.")));
3323 /* treat it as deleted; do not process */
3328 {
3330 relation,
3332 oldslot);
3334 /*
3335 * If PlanQual failed for updated tuple - we must not
3336 * process this tuple!
3337 */
3339 {
3342 }
3343 }
3359 /* tuple was deleted */
3369 }
3370 }
3371 else
3372 {
3373 /*
3374 * We expect the tuple to be present, thus very simple error handling
3375 * suffices.
3376 */
3378 oldslot))
3380 }
3383 }
3385 /*
3386 * Is trigger enabled to fire?
3387 */
3388 static bool
3393 {
3394 /* Check replication-role-dependent enable state */
3396 {
3400 }
3402 {
3406 }
3408 /*
3409 * Check for column-specific trigger (only possible for UPDATE, and in
3410 * fact we *must* ignore tgattr for other event types)
3411 */
3413 {
3419 {
3421 modifiedCols))
3422 {
3425 }
3426 }
3429 }
3431 /* Check for WHEN clause */
3433 {
3436 MemoryContext oldContext;
3441 /*
3442 * trigger is an element of relinfo->ri_TrigDesc->triggers[]; find the
3443 * matching element of relinfo->ri_TrigWhenExprs[]
3444 */
3448 /*
3449 * If first time through for this WHEN expression, build expression
3450 * nodetrees for it. Keep them in the per-query memory context so
3451 * they'll survive throughout the query.
3452 */
3454 {
3459 /* Change references to OLD and NEW to INNER_VAR and OUTER_VAR */
3462 /* ExecPrepareQual wants implicit-AND form */
3466 }
3468 /*
3469 * We will use the EState's per-tuple context for evaluating WHEN
3470 * expressions (creating it if it's not already there).
3471 */
3474 /*
3475 * Finally evaluate the expression, making the old and/or new tuples
3476 * available as INNER_VAR/OUTER_VAR respectively.
3477 */
3482 }
3485 }
3488 /* ----------
3489 * After-trigger stuff
3490 *
3491 * The AfterTriggersData struct holds data about pending AFTER trigger events
3492 * during the current transaction tree. (BEFORE triggers are fired
3493 * immediately so we don't need any persistent state about them.) The struct
3494 * and most of its subsidiary data are kept in TopTransactionContext; however
3495 * some data that can be discarded sooner appears in the CurTransactionContext
3496 * of the relevant subtransaction. Also, the individual event records are
3497 * kept in a separate sub-context of TopTransactionContext. This is done
3498 * mainly so that it's easy to tell from a memory context dump how much space
3499 * is being eaten by trigger events.
3500 *
3501 * Because the list of pending events can grow large, we go to some
3502 * considerable effort to minimize per-event memory consumption. The event
3503 * records are grouped into chunks and common data for similar events in the
3504 * same chunk is only stored once.
3505 *
3506 * XXX We need to be able to save the per-event data in a file if it grows too
3507 * large.
3508 * ----------
3509 */
3511 /* Per-trigger SET CONSTRAINT status */
3513 {
3514 Oid sct_tgoid;
3520 /*
3521 * SET CONSTRAINT intra-transaction status.
3522 *
3523 * We make this a single palloc'd object so it can be copied and freed easily.
3524 *
3525 * all_isset and all_isdeferred are used to keep track
3526 * of SET CONSTRAINTS ALL {DEFERRED, IMMEDIATE}.
3527 *
3528 * trigstates[] stores per-trigger tgisdeferred settings.
3529 */
3531 {
3542 /*
3543 * Per-trigger-event data
3544 *
3545 * The actual per-event data, AfterTriggerEventData, includes DONE/IN_PROGRESS
3546 * status bits, up to two tuple CTIDs, and optionally two OIDs of partitions.
3547 * Each event record also has an associated AfterTriggerSharedData that is
3548 * shared across all instances of similar events within a "chunk".
3549 *
3550 * For row-level triggers, we arrange not to waste storage on unneeded ctid
3551 * fields. Updates of regular tables use two; inserts and deletes of regular
3552 * tables use one; foreign tables always use zero and save the tuple(s) to a
3553 * tuplestore. AFTER_TRIGGER_FDW_FETCH directs AfterTriggerExecute() to
3554 * retrieve a fresh tuple or pair of tuples from that tuplestore, while
3555 * AFTER_TRIGGER_FDW_REUSE directs it to use the most-recently-retrieved
3556 * tuple(s). This permits storing tuples once regardless of the number of
3557 * row-level triggers on a foreign table.
3558 *
3559 * When updates on partitioned tables cause rows to move between partitions,
3560 * the OIDs of both partitions are stored too, so that the tuples can be
3561 * fetched; such entries are marked AFTER_TRIGGER_CP_UPDATE (for "cross-
3562 * partition update").
3563 *
3564 * Note that we need triggers on foreign tables to be fired in exactly the
3565 * order they were queued, so that the tuples come out of the tuplestore in
3566 * the right order. To ensure that, we forbid deferrable (constraint)
3567 * triggers on foreign tables. This also ensures that such triggers do not
3568 * get deferred into outer trigger query levels, meaning that it's okay to
3569 * destroy the tuplestore at the end of the query level.
3570 *
3571 * Statement-level triggers always bear AFTER_TRIGGER_1CTID, though they
3572 * require no ctid field. We lack the flag bit space to neatly represent that
3573 * distinct case, and it seems unlikely to be worth much trouble.
3574 *
3575 * Note: ats_firing_id is initially zero and is set to something else when
3576 * AFTER_TRIGGER_IN_PROGRESS is set. It indicates which trigger firing
3577 * cycle the trigger will be fired in (or was fired in, if DONE is set).
3578 * Although this is mutable state, we can keep it in AfterTriggerSharedData
3579 * because all instances of the same type of event in a given event list will
3580 * be fired at the same time, if they were queued between the same firing
3581 * cycles. So we need only ensure that ats_firing_id is zero when attaching
3582 * a new event to an existing AfterTriggerSharedData record.
3583 */
3587 #define AFTER_TRIGGER_DONE 0x80000000
3588 #define AFTER_TRIGGER_IN_PROGRESS 0x40000000
3589 /* bits describing the size and tuple sources of this event */
3590 #define AFTER_TRIGGER_FDW_REUSE 0x00000000
3591 #define AFTER_TRIGGER_FDW_FETCH 0x20000000
3592 #define AFTER_TRIGGER_1CTID 0x10000000
3593 #define AFTER_TRIGGER_2CTID 0x30000000
3594 #define AFTER_TRIGGER_CP_UPDATE 0x08000000
3595 #define AFTER_TRIGGER_TUP_BITS 0x38000000
3599 {
3611 {
3616 /*
3617 * During a cross-partition update of a partitioned table, we also store
3618 * the OIDs of source and destination partitions that are needed to fetch
3619 * the old (ctid1) and the new tuple (ctid2) from, respectively.
3620 */
3621 Oid ate_src_part;
3622 Oid ate_dst_part;
3625 /* AfterTriggerEventData, minus ate_src_part, ate_dst_part */
3627 {
3628 TriggerFlags ate_flags;
3629 ItemPointerData ate_ctid1;
3630 ItemPointerData ate_ctid2;
3633 /* AfterTriggerEventData, minus ate_*_part and ate_ctid2 */
3635 {
3640 /* AfterTriggerEventData, minus ate_*_part, ate_ctid1 and ate_ctid2 */
3642 {
3646 #define SizeofTriggerEvent(evt) \
3647 (((evt)->ate_flags & AFTER_TRIGGER_TUP_BITS) == AFTER_TRIGGER_CP_UPDATE ? \
3648 sizeof(AfterTriggerEventData) : \
3649 (((evt)->ate_flags & AFTER_TRIGGER_TUP_BITS) == AFTER_TRIGGER_2CTID ? \
3650 sizeof(AfterTriggerEventDataNoOids) : \
3651 (((evt)->ate_flags & AFTER_TRIGGER_TUP_BITS) == AFTER_TRIGGER_1CTID ? \
3652 sizeof(AfterTriggerEventDataOneCtid) : \
3653 sizeof(AfterTriggerEventDataZeroCtids))))
3655 #define GetTriggerSharedData(evt) \
3656 ((AfterTriggerShared) ((char *) (evt) + ((evt)->ate_flags & AFTER_TRIGGER_OFFSET)))
3658 /*
3659 * To avoid palloc overhead, we keep trigger events in arrays in successively-
3660 * larger chunks (a slightly more sophisticated version of an expansible
3661 * array). The space between CHUNK_DATA_START and freeptr is occupied by
3662 * AfterTriggerEventData records; the space between endfree and endptr is
3663 * occupied by AfterTriggerSharedData records.
3664 */
3666 {
3671 /* event data follows here */
3674 #define CHUNK_DATA_START(cptr) ((char *) (cptr) + MAXALIGN(sizeof(AfterTriggerEventChunk)))
3676 /* A list of events */
3678 {
3684 /* Macros to help in iterating over a list of events */
3685 #define for_each_chunk(cptr, evtlist) \
3686 for (cptr = (evtlist).head; cptr != NULL; cptr = cptr->next)
3687 #define for_each_event(eptr, cptr) \
3688 for (eptr = (AfterTriggerEvent) CHUNK_DATA_START(cptr); \
3689 (char *) eptr < (cptr)->freeptr; \
3690 eptr = (AfterTriggerEvent) (((char *) eptr) + SizeofTriggerEvent(eptr)))
3691 /* Use this if no special per-chunk processing is needed */
3692 #define for_each_event_chunk(eptr, cptr, evtlist) \
3693 for_each_chunk(cptr, evtlist) for_each_event(eptr, cptr)
3695 /* Macros for iterating from a start point that might not be list start */
3696 #define for_each_chunk_from(cptr) \
3697 for (; cptr != NULL; cptr = cptr->next)
3698 #define for_each_event_from(eptr, cptr) \
3699 for (; \
3700 (char *) eptr < (cptr)->freeptr; \
3701 eptr = (AfterTriggerEvent) (((char *) eptr) + SizeofTriggerEvent(eptr)))
3704 /*
3705 * All per-transaction data for the AFTER TRIGGERS module.
3706 *
3707 * AfterTriggersData has the following fields:
3708 *
3709 * firing_counter is incremented for each call of afterTriggerInvokeEvents.
3710 * We mark firable events with the current firing cycle's ID so that we can
3711 * tell which ones to work on. This ensures sane behavior if a trigger
3712 * function chooses to do SET CONSTRAINTS: the inner SET CONSTRAINTS will
3713 * only fire those events that weren't already scheduled for firing.
3714 *
3715 * state keeps track of the transaction-local effects of SET CONSTRAINTS.
3716 * This is saved and restored across failed subtransactions.
3717 *
3718 * events is the current list of deferred events. This is global across
3719 * all subtransactions of the current transaction. In a subtransaction
3720 * abort, we know that the events added by the subtransaction are at the
3721 * end of the list, so it is relatively easy to discard them. The event
3722 * list chunks themselves are stored in event_cxt.
3723 *
3724 * query_depth is the current depth of nested AfterTriggerBeginQuery calls
3725 * (-1 when the stack is empty).
3726 *
3727 * query_stack[query_depth] is the per-query-level data, including these fields:
3728 *
3729 * events is a list of AFTER trigger events queued by the current query.
3730 * None of these are valid until the matching AfterTriggerEndQuery call
3731 * occurs. At that point we fire immediate-mode triggers, and append any
3732 * deferred events to the main events list.
3733 *
3734 * fdw_tuplestore is a tuplestore containing the foreign-table tuples
3735 * needed by events queued by the current query. (Note: we use just one
3736 * tuplestore even though more than one foreign table might be involved.
3737 * This is okay because tuplestores don't really care what's in the tuples
3738 * they store; but it's possible that someday it'd break.)
3739 *
3740 * tables is a List of AfterTriggersTableData structs for target tables
3741 * of the current query (see below).
3742 *
3743 * maxquerydepth is just the allocated length of query_stack.
3744 *
3745 * trans_stack holds per-subtransaction data, including these fields:
3746 *
3747 * state is NULL or a pointer to a saved copy of the SET CONSTRAINTS
3748 * state data. Each subtransaction level that modifies that state first
3749 * saves a copy, which we use to restore the state if we abort.
3750 *
3751 * events is a copy of the events head/tail pointers,
3752 * which we use to restore those values during subtransaction abort.
3753 *
3754 * query_depth is the subtransaction-start-time value of query_depth,
3755 * which we similarly use to clean up at subtransaction abort.
3756 *
3757 * firing_counter is the subtransaction-start-time value of firing_counter.
3758 * We use this to recognize which deferred triggers were fired (or marked
3759 * for firing) within an aborted subtransaction.
3760 *
3761 * We use GetCurrentTransactionNestLevel() to determine the correct array
3762 * index in trans_stack. maxtransdepth is the number of allocated entries in
3763 * trans_stack. (By not keeping our own stack pointer, we can avoid trouble
3764 * in cases where errors during subxact abort cause multiple invocations
3765 * of AfterTriggerEndSubXact() at the same nesting depth.)
3766 *
3767 * We create an AfterTriggersTableData struct for each target table of the
3768 * current query, and each operation mode (INSERT/UPDATE/DELETE), that has
3769 * either transition tables or statement-level triggers. This is used to
3770 * hold the relevant transition tables, as well as info tracking whether
3771 * we already queued the statement triggers. (We use that info to prevent
3772 * firing the same statement triggers more than once per statement, or really
3773 * once per transition table set.) These structs, along with the transition
3774 * table tuplestores, live in the (sub)transaction's CurTransactionContext.
3775 * That's sufficient lifespan because we don't allow transition tables to be
3776 * used by deferrable triggers, so they only need to survive until
3777 * AfterTriggerEndQuery.
3778 */
3784 {
3790 /* per-query-level data: */
3795 /* per-subtransaction-level data: */
3800 struct AfterTriggersQueryData
3801 {
3805 };
3807 struct AfterTriggersTransData
3808 {
3809 /* these fields are just for resetting at subtrans abort: */
3814 };
3816 struct AfterTriggersTableData
3817 {
3818 /* relid + cmdType form the lookup key for these structs: */
3826 /*
3827 * We maintain separate transition tables for UPDATE/INSERT/DELETE since
3828 * MERGE can run all three actions in a single statement. Note that UPDATE
3829 * needs both old and new transition tables whereas INSERT needs only new,
3830 * and DELETE needs only old.
3831 */
3833 /* "old" transition table for UPDATE, if any */
3835 /* "new" transition table for UPDATE, if any */
3837 /* "old" transition table for DELETE, if any */
3839 /* "new" transition table for INSERT, if any */
3843 };
3848 AfterTriggerEvent event,
3855 MemoryContext per_tuple_context,
3859 CmdType cmdType);
3861 TupleDesc tupdesc);
3880 /*
3881 * Get the FDW tuplestore for the current trigger query level, creating it
3882 * if necessary.
3883 */
3886 {
3891 {
3892 MemoryContext oldcxt;
3893 ResourceOwner saveResourceOwner;
3895 /*
3896 * Make the tuplestore valid until end of subtransaction. We really
3897 * only need it until AfterTriggerEndQuery().
3898 */
3909 }
3912 }
3914 /* ----------
3915 * afterTriggerCheckState()
3916 *
3917 * Returns true if the trigger event is actually in state DEFERRED.
3918 * ----------
3919 */
3920 static bool
3922 {
3927 /*
3928 * For not-deferrable triggers (i.e. normal AFTER ROW triggers and
3929 * constraints declared NOT DEFERRABLE), the state is always false.
3930 */
3934 /*
3935 * If constraint state exists, SET CONSTRAINTS might have been executed
3936 * either for this trigger or for all triggers.
3937 */
3939 {
3940 /* Check for SET CONSTRAINTS for this specific trigger. */
3942 {
3945 }
3947 /* Check for SET CONSTRAINTS ALL. */
3950 }
3952 /*
3953 * Otherwise return the default state for the trigger.
3954 */
3956 }
3959 /* ----------
3960 * afterTriggerAddEvent()
3961 *
3962 * Add a new trigger event to the specified queue.
3963 * The passed-in event data is copied.
3964 * ----------
3965 */
3966 static void
3969 {
3973 AfterTriggerShared newshared;
3974 AfterTriggerEvent newevent;
3976 /*
3977 * If empty list or not enough room in the tail chunk, make a new chunk.
3978 * We assume here that a new shared record will always be needed.
3979 */
3983 {
3984 Size chunksize;
3986 /* Create event context if we didn't already */
3991 ALLOCSET_DEFAULT_SIZES);
3993 /*
3994 * Chunk size starts at 1KB and is allowed to increase up to 1MB.
3995 * These numbers are fairly arbitrary, though there is a hard limit at
3996 * AFTER_TRIGGER_OFFSET; else we couldn't link event records to their
3997 * shared records using the available space in ate_flags. Another
3998 * constraint is that if the chunk size gets too huge, the search loop
3999 * below would get slow given a (not too common) usage pattern with
4000 * many distinct event types in a chunk. Therefore, we double the
4001 * preceding chunk size only if there weren't too many shared records
4002 * in the preceding chunk; otherwise we halve it. This gives us some
4003 * ability to adapt to the actual usage pattern of the current query
4004 * while still having large chunk sizes in typical usage. All chunk
4005 * sizes used should be MAXALIGN multiples, to ensure that the shared
4006 * records will be aligned safely.
4007 */
4008 #define MIN_CHUNK_SIZE 1024
4009 #define MAX_CHUNK_SIZE (1024*1024)
4011 #if MAX_CHUNK_SIZE > (AFTER_TRIGGER_OFFSET+1)
4012 #error MAX_CHUNK_SIZE must not exceed AFTER_TRIGGER_OFFSET
4013 #endif
4017 else
4018 {
4019 /* preceding chunk size... */
4021 /* check number of shared records in preceding chunk */
4025 else
4028 }
4037 else
4040 /* events->tailfree is now out of sync, but we'll fix it below */
4041 }
4043 /*
4044 * Try to locate a matching shared-data record already in the chunk. If
4045 * none, make a new one.
4046 */
4049 newshared--)
4050 {
4057 }
4059 {
4063 }
4065 /* Insert the data */
4068 /* ... and link the new event to its shared record */
4074 }
4076 /* ----------
4077 * afterTriggerFreeEventList()
4078 *
4079 * Free all the event storage in the given list.
4080 * ----------
4081 */
4082 static void
4084 {
4088 {
4091 }
4094 }
4096 /* ----------
4097 * afterTriggerRestoreEventList()
4098 *
4099 * Restore an event list to its prior length, removing all the events
4100 * added since it had the value old_events.
4101 * ----------
4102 */
4103 static void
4106 {
4111 {
4112 /* restoring to a completely empty state, so free everything */
4114 }
4115 else
4116 {
4118 /* free any chunks after the last one we want to keep */
4120 {
4123 }
4124 /* and clean up the tail chunk to be the right length */
4128 /*
4129 * We don't make any effort to remove now-unused shared data records.
4130 * They might still be useful, anyway.
4131 */
4132 }
4133 }
4135 /* ----------
4136 * afterTriggerDeleteHeadEventChunk()
4137 *
4138 * Remove the first chunk of events from the query level's event list.
4139 * Keep any event list pointers elsewhere in the query level's data
4140 * structures in sync.
4141 * ----------
4142 */
4143 static void
4145 {
4151 /*
4152 * First, update any pointers in the per-table data, so that they won't be
4153 * dangling. Resetting obsoleted pointers to NULL will make
4154 * cancel_prior_stmt_triggers start from the list head, which is fine.
4155 */
4157 {
4162 {
4166 }
4167 }
4169 /* Now we can flush the head chunk */
4172 }
4175 /* ----------
4176 * AfterTriggerExecute()
4177 *
4178 * Fetch the required tuples back from the heap and fire one
4179 * single trigger function.
4180 *
4181 * Frequently, this will be fired many times in a row for triggers of
4182 * a single relation. Therefore, we cache the open relation and provide
4183 * fmgr lookup cache space at the caller level. (For triggers fired at
4184 * the end of a query, we can even piggyback on the executor's state.)
4185 *
4186 * When fired for a cross-partition update of a partitioned table, the old
4187 * tuple is fetched using 'src_relInfo' (the source leaf partition) and
4188 * the new tuple using 'dst_relInfo' (the destination leaf partition), though
4189 * both are converted into the root partitioned table's format before passing
4190 * to the trigger function.
4191 *
4192 * event: event currently being fired.
4193 * relInfo: result relation for event.
4194 * src_relInfo: source partition of a cross-partition update
4195 * dst_relInfo: its destination partition
4196 * trigdesc: working copy of rel's trigger info.
4197 * finfo: array of fmgr lookup cache entries (one per trigger in trigdesc).
4198 * instr: array of EXPLAIN ANALYZE instrumentation nodes (one per trigger),
4199 * or NULL if no instrumentation is wanted.
4200 * per_tuple_context: memory context to call trigger function in.
4201 * trig_tuple_slot1: scratch slot for tg_trigtuple (foreign tables only)
4202 * trig_tuple_slot2: scratch slot for tg_newtuple (foreign tables only)
4203 * ----------
4204 */
4205 static void
4207 AfterTriggerEvent event,
4213 MemoryContext per_tuple_context,
4216 {
4223 HeapTuple rettuple;
4228 /*
4229 * Locate trigger in trigdesc.
4230 */
4232 {
4234 {
4237 }
4238 }
4242 /*
4243 * If doing EXPLAIN ANALYZE, start charging time to this trigger. We want
4244 * to include time spent re-fetching tuples in the trigger cost.
4245 */
4249 /*
4250 * Fetch the required tuple(s).
4251 */
4253 {
4255 {
4259 trig_tuple_slot1))
4263 TRIGGER_EVENT_UPDATE &&
4265 trig_tuple_slot2))
4267 }
4268 /* fall through */
4271 /*
4272 * Store tuple in the slot so that tg_trigtuple does not reference
4273 * tuplestore memory. (It is formally possible for the trigger
4274 * function to queue trigger events that add to the same
4275 * tuplestore, which can push other tuples out of memory.) The
4276 * distinction is academic, because we start with a minimal tuple
4277 * that is stored as a heap tuple, constructed in different memory
4278 * context, in the slot anyway.
4279 */
4285 TRIGGER_EVENT_UPDATE)
4286 {
4290 }
4291 else
4292 {
4294 }
4299 {
4301 src_relInfo);
4305 SnapshotAny,
4306 src_slot))
4309 /*
4310 * Store the tuple fetched from the source partition into the
4311 * target (root partitioned) table slot, converting if needed.
4312 */
4314 {
4319 {
4321 src_slot,
4323 }
4324 else
4326 }
4327 else
4331 }
4332 else
4333 {
4335 }
4337 /* don't touch ctid2 if not there */
4341 {
4343 dst_relInfo);
4347 SnapshotAny,
4348 dst_slot))
4351 /*
4352 * Store the tuple fetched from the destination partition into
4353 * the target (root partitioned) table slot, converting if
4354 * needed.
4355 */
4357 {
4362 {
4364 dst_slot,
4366 }
4367 else
4369 }
4370 else
4374 }
4375 else
4376 {
4378 }
4379 }
4381 /*
4382 * Set up the tuplestore information to let the trigger have access to
4383 * transition tables. When we first make a transition table available to
4384 * a trigger, mark it "closed" so that it cannot change anymore. If any
4385 * additional events of the same type get queued in the current trigger
4386 * query level, they'll go into new transition tables.
4387 */
4390 {
4392 {
4395 else
4398 }
4401 {
4404 else
4407 }
4408 }
4410 /*
4411 * Setup the remaining trigger information
4412 */
4422 /*
4423 * Call the trigger and throw away any possibly returned updated tuple.
4424 * (Don't let ExecCallTriggerFunc measure EXPLAIN time.)
4425 */
4427 tgindx,
4428 finfo,
4429 NULL,
4430 per_tuple_context);
4436 /*
4437 * Release resources
4438 */
4444 /* don't clear slots' contents if foreign table */
4446 {
4451 }
4453 /*
4454 * If doing EXPLAIN ANALYZE, stop charging time to this trigger, and count
4455 * one "tuple returned" (really the number of firings).
4456 */
4459 }
4462 /*
4463 * afterTriggerMarkEvents()
4464 *
4465 * Scan the given event list for not yet invoked events. Mark the ones
4466 * that can be invoked now with the current firing ID.
4467 *
4468 * If move_list isn't NULL, events that are not to be invoked now are
4469 * transferred to move_list.
4470 *
4471 * When immediate_only is true, do not invoke currently-deferred triggers.
4472 * (This will be false only at main transaction exit.)
4473 *
4474 * Returns true if any invokable events were found.
4475 */
4476 static bool
4480 {
4483 AfterTriggerEvent event;
4487 {
4493 {
4494 /*
4495 * This trigger hasn't been called or scheduled yet. Check if we
4496 * should call it now.
4497 */
4499 {
4501 }
4502 else
4503 {
4504 /*
4505 * Mark it as to be fired in this firing cycle.
4506 */
4510 }
4511 }
4513 /*
4514 * If it's deferred, move it to move_list, if requested.
4515 */
4517 {
4519 /* add it to move_list */
4521 /* mark original copy "done" so we don't do it again */
4523 }
4524 }
4526 /*
4527 * We could allow deferred triggers if, before the end of the
4528 * security-restricted operation, we were to verify that a SET CONSTRAINTS
4529 * ... IMMEDIATE has fired all such triggers. For now, don't bother.
4530 */
4537 }
4539 /*
4540 * afterTriggerInvokeEvents()
4541 *
4542 * Scan the given event list for events that are marked as to be fired
4543 * in the current firing cycle, and fire them.
4544 *
4545 * If estate isn't NULL, we use its result relation info to avoid repeated
4546 * openings and closing of trigger target relations. If it is NULL, we
4547 * make one locally to cache the info in case there are multiple trigger
4548 * events per rel.
4549 *
4550 * When delete_ok is true, it's safe to delete fully-processed events.
4551 * (We are not very tense about that: we simply reset a chunk to be empty
4552 * if all its events got fired. The objective here is just to avoid useless
4553 * rescanning of events when a trigger queues new events during transaction
4554 * end, so it's not necessary to worry much about the case where only
4555 * some events are fired.)
4556 *
4557 * Returns true if no unfired events remain in the list (this allows us
4558 * to avoid repeating afterTriggerMarkEvents).
4559 */
4560 static bool
4562 CommandId firing_id,
4565 {
4568 MemoryContext per_tuple_context;
4578 /* Make a local EState if need be */
4580 {
4583 }
4585 /* Make a per-tuple memory context for trigger function calls */
4586 per_tuple_context =
4589 ALLOCSET_DEFAULT_SIZES);
4592 {
4593 AfterTriggerEvent event;
4597 {
4600 /*
4601 * Is it one for me to fire?
4602 */
4605 {
4609 /*
4610 * So let's fire it... but first, find the correct relation if
4611 * this is not the same relation as before.
4612 */
4614 {
4616 NULL);
4618 /* Catch calls with insufficient relcache refcounting */
4624 {
4628 }
4630 {
4635 }
4639 }
4641 /*
4642 * Look up source and destination partition result rels of a
4643 * cross-partition update event.
4644 */
4646 AFTER_TRIGGER_CP_UPDATE)
4647 {
4652 rInfo);
4655 rInfo);
4656 }
4657 else
4660 /*
4661 * Fire it. Note that the AFTER_TRIGGER_IN_PROGRESS flag is
4662 * still set, so recursive examinations of the event list
4663 * won't try to re-fire it.
4664 */
4670 /*
4671 * Mark the event as done.
4672 */
4675 }
4677 {
4678 /* something remains to be done */
4680 }
4681 }
4683 /* Clear the chunk if delete_ok and nothing left of interest */
4685 {
4689 /*
4690 * If it's last chunk, must sync event list's tailfree too. Note
4691 * that delete_ok must NOT be passed as true if there could be
4692 * additional AfterTriggerEventList values pointing at this event
4693 * list, since we'd fail to fix their copies of tailfree.
4694 */
4697 }
4698 }
4700 {
4703 }
4705 /* Release working resources */
4709 {
4713 }
4716 }
4719 /*
4720 * GetAfterTriggersTableData
4721 *
4722 * Find or create an AfterTriggersTableData struct for the specified
4723 * trigger event (relation + operation type). Ignore existing structs
4724 * marked "closed"; we don't want to put any additional tuples into them,
4725 * nor change their stmt-triggers-fired state.
4726 *
4727 * Note: the AfterTriggersTableData list is allocated in the current
4728 * (sub)transaction's CurTransactionContext. This is OK because
4729 * we don't need it to live past AfterTriggerEndQuery.
4730 */
4733 {
4736 MemoryContext oldcxt;
4739 /* Caller should have ensured query_depth is OK. */
4745 {
4750 }
4762 }
4764 /*
4765 * Returns a TupleTableSlot suitable for holding the tuples to be put
4766 * into AfterTriggersTableData's transition table tuplestores.
4767 */
4770 TupleDesc tupdesc)
4771 {
4772 /* Create it if not already done. */
4774 {
4775 MemoryContext oldcxt;
4777 /*
4778 * We only need this slot only until AfterTriggerEndQuery, but making
4779 * it last till end-of-subxact is good enough. It'll be freed by
4780 * AfterTriggerFreeQuery().
4781 */
4785 }
4788 }
4790 /*
4791 * MakeTransitionCaptureState
4792 *
4793 * Make a TransitionCaptureState object for the given TriggerDesc, target
4794 * relation, and operation type. The TCS object holds all the state needed
4795 * to decide whether to capture tuples in transition tables.
4796 *
4797 * If there are no triggers in 'trigdesc' that request relevant transition
4798 * tables, then return NULL.
4799 *
4800 * The resulting object can be passed to the ExecAR* functions. When
4801 * dealing with child tables, the caller can set tcs_original_insert_tuple
4802 * to avoid having to reconstruct the original tuple in the root table's
4803 * format.
4804 *
4805 * Note that we copy the flags from a parent table into this struct (rather
4806 * than subsequently using the relation's TriggerDesc directly) so that we can
4807 * use it to control collection of transition tuples from child tables.
4808 *
4809 * Per SQL spec, all operations of the same kind (INSERT/UPDATE/DELETE)
4810 * on the same table during one query should share one transition table.
4811 * Therefore, the Tuplestores are owned by an AfterTriggersTableData struct
4812 * looked up using the table OID + CmdType, and are merely referenced by
4813 * the TransitionCaptureState objects we hand out to callers.
4814 */
4815 TransitionCaptureState *
4817 {
4820 need_new_upd,
4821 need_old_del,
4822 need_new_ins;
4824 MemoryContext oldcxt;
4825 ResourceOwner saveResourceOwner;
4830 /* Detect which table(s) we need. */
4832 {
4854 /* keep compiler quiet */
4857 }
4861 /* Check state, like AfterTriggerSaveEvent. */
4865 /* Be sure we have enough space to record events at this query depth. */
4869 /*
4870 * Find or create an AfterTriggersTableData struct to hold the
4871 * tuplestore(s). If there's a matching struct but it's marked closed,
4872 * ignore it; we need a newer one.
4873 *
4874 * Note: the AfterTriggersTableData list, as well as the tuplestores, are
4875 * allocated in the current (sub)transaction's CurTransactionContext, and
4876 * the tuplestores are managed by the (sub)transaction's resource owner.
4877 * This is sufficient lifespan because we do not allow triggers using
4878 * transition tables to be deferrable; they will be fired during
4879 * AfterTriggerEndQuery, after which it's okay to delete the data.
4880 */
4883 /* Now create required tuplestore(s), if we don't have them already. */
4900 /* Now build the TransitionCaptureState struct, in caller's context */
4909 }
4912 /* ----------
4913 * AfterTriggerBeginXact()
4914 *
4915 * Called at transaction start (either BEGIN or implicit for single
4916 * statement outside of transaction block).
4917 * ----------
4918 */
4919 void
4921 {
4922 /*
4923 * Initialize after-trigger state structure to empty
4924 */
4928 /*
4929 * Verify that there is no leftover state remaining. If these assertions
4930 * trip, it means that AfterTriggerEndXact wasn't called or didn't clean
4931 * up properly.
4932 */
4940 }
4943 /* ----------
4944 * AfterTriggerBeginQuery()
4945 *
4946 * Called just before we start processing a single query within a
4947 * transaction (or subtransaction). Most of the real work gets deferred
4948 * until somebody actually tries to queue a trigger event.
4949 * ----------
4950 */
4951 void
4953 {
4954 /* Increase the query stack depth */
4956 }
4959 /* ----------
4960 * AfterTriggerEndQuery()
4961 *
4962 * Called after one query has been completely processed. At this time
4963 * we invoke all AFTER IMMEDIATE trigger events queued by the query, and
4964 * transfer deferred trigger events to the global deferred-trigger list.
4965 *
4966 * Note that this must be called BEFORE closing down the executor
4967 * with ExecutorEnd, because we make use of the EState's info about
4968 * target relations. Normally it is called from ExecutorFinish.
4969 * ----------
4970 */
4971 void
4973 {
4976 /* Must be inside a query, too */
4979 /*
4980 * If we never even got as far as initializing the event stack, there
4981 * certainly won't be any events, so exit quickly.
4982 */
4984 {
4987 }
4989 /*
4990 * Process all immediate-mode triggers queued by the query, and move the
4991 * deferred ones to the main list of deferred events.
4992 *
4993 * Notice that we decide which ones will be fired, and put the deferred
4994 * ones on the main list, before anything is actually fired. This ensures
4995 * reasonably sane behavior if a trigger function does SET CONSTRAINTS ...
4996 * IMMEDIATE: all events we have decided to defer will be available for it
4997 * to fire.
4998 *
4999 * We loop in case a trigger queues more events at the same query level.
5000 * Ordinary trigger functions, including all PL/pgSQL trigger functions,
5001 * will instead fire any triggers in a dedicated query level. Foreign key
5002 * enforcement triggers do add to the current query level, thanks to their
5003 * passing fire_triggers = false to SPI_execute_snapshot(). Other
5004 * C-language triggers might do likewise.
5005 *
5006 * If we find no firable events, we don't have to increment
5007 * firing_counter.
5008 */
5012 {
5014 {
5021 /*
5022 * Firing a trigger could result in query_stack being repalloc'd,
5023 * so we must recalculate qs after each afterTriggerInvokeEvents
5024 * call. Furthermore, it's unsafe to pass delete_ok = true here,
5025 * because that could cause afterTriggerInvokeEvents to try to
5026 * access qs->events after the stack has been repalloc'd.
5027 */
5030 /*
5031 * We'll need to scan the events list again. To reduce the cost
5032 * of doing so, get rid of completely-fired chunks. We know that
5033 * all events were marked IN_PROGRESS or DONE at the conclusion of
5034 * afterTriggerMarkEvents, so any still-interesting events must
5035 * have been added after that, and so must be in the chunk that
5036 * was then the tail chunk, or in later chunks. So, zap all
5037 * chunks before oldtail. This is approximately the same set of
5038 * events we would have gotten rid of by passing delete_ok = true.
5039 */
5043 }
5044 else
5046 }
5048 /* Release query-level-local storage, including tuplestores if any */
5052 }
5055 /*
5056 * AfterTriggerFreeQuery
5057 * Release subsidiary storage for a trigger query level.
5058 * This includes closing down tuplestores.
5059 * Note: it's important for this to be safe if interrupted by an error
5060 * and then called again for the same query level.
5061 */
5062 static void
5064 {
5069 /* Drop the trigger events */
5072 /* Drop FDW tuplestore if any */
5078 /* Release per-table subsidiary storage */
5081 {
5102 }
5104 /*
5105 * Now free the AfterTriggersTableData structs and list cells. Reset list
5106 * pointer first; if list_free_deep somehow gets an error, better to leak
5107 * that storage than have an infinite loop.
5108 */
5111 }
5114 /* ----------
5115 * AfterTriggerFireDeferred()
5116 *
5117 * Called just before the current transaction is committed. At this
5118 * time we invoke all pending DEFERRED triggers.
5119 *
5120 * It is possible for other modules to queue additional deferred triggers
5121 * during pre-commit processing; therefore xact.c may have to call this
5122 * multiple times.
5123 * ----------
5124 */
5125 void
5127 {
5131 /* Must not be inside a query */
5134 /*
5135 * If there are any triggers to fire, make sure we have set a snapshot for
5136 * them to use. (Since PortalRunUtility doesn't set a snap for COMMIT, we
5137 * can't assume ActiveSnapshot is valid on entry.)
5138 */
5141 {
5144 }
5146 /*
5147 * Run all the remaining triggers. Loop until they are all gone, in case
5148 * some trigger queues more for us to do.
5149 */
5151 {
5156 }
5158 /*
5159 * We don't bother freeing the event list, since it will go away anyway
5160 * (and more efficiently than via pfree) in AfterTriggerEndXact.
5161 */
5165 }
5168 /* ----------
5169 * AfterTriggerEndXact()
5170 *
5171 * The current transaction is finishing.
5172 *
5173 * Any unfired triggers are canceled so we simply throw
5174 * away anything we know.
5175 *
5176 * Note: it is possible for this to be called repeatedly in case of
5177 * error during transaction abort; therefore, do not complain if
5178 * already closed down.
5179 * ----------
5180 */
5181 void
5183 {
5184 /*
5185 * Forget the pending-events list.
5186 *
5187 * Since all the info is in TopTransactionContext or children thereof, we
5188 * don't really need to do anything to reclaim memory. However, the
5189 * pending-events list could be large, and so it's useful to discard it as
5190 * soon as possible --- especially if we are aborting because we ran out
5191 * of memory for the list!
5192 */
5194 {
5200 }
5202 /*
5203 * Forget any subtransaction state as well. Since this can't be very
5204 * large, we let the eventual reset of TopTransactionContext free the
5205 * memory instead of doing it here.
5206 */
5211 /*
5212 * Forget the query stack and constraint-related state information. As
5213 * with the subtransaction state information, we don't bother freeing the
5214 * memory here.
5215 */
5220 /* No more afterTriggers manipulation until next transaction starts. */
5222 }
5224 /*
5225 * AfterTriggerBeginSubXact()
5226 *
5227 * Start a subtransaction.
5228 */
5229 void
5231 {
5234 /*
5235 * Allocate more space in the trans_stack if needed. (Note: because the
5236 * minimum nest level of a subtransaction is 2, we waste the first couple
5237 * entries of the array; not worth the notational effort to avoid it.)
5238 */
5240 {
5242 {
5243 /* Arbitrarily initialize for max of 8 subtransaction levels */
5248 }
5249 else
5250 {
5251 /* repalloc will keep the stack in the same context */
5258 }
5259 }
5261 /*
5262 * Push the current information into the stack. The SET CONSTRAINTS state
5263 * is not saved until/unless changed. Likewise, we don't make a
5264 * per-subtransaction event context until needed.
5265 */
5270 }
5272 /*
5273 * AfterTriggerEndSubXact()
5274 *
5275 * The current subtransaction is ending.
5276 */
5277 void
5279 {
5281 SetConstraintState state;
5282 AfterTriggerEvent event;
5284 CommandId subxact_firing_id;
5286 /*
5287 * Pop the prior state if needed.
5288 */
5290 {
5292 /* If we saved a prior state, we don't need it anymore */
5296 /* this avoids double pfree if error later: */
5300 }
5301 else
5302 {
5303 /*
5304 * Aborting. It is possible subxact start failed before calling
5305 * AfterTriggerBeginSubXact, in which case we mustn't risk touching
5306 * trans_stack levels that aren't there.
5307 */
5311 /*
5312 * Release query-level storage for queries being aborted, and restore
5313 * query_depth to its pre-subxact value. This assumes that a
5314 * subtransaction will not add events to query levels started in a
5315 * earlier transaction state.
5316 */
5318 {
5322 }
5326 /*
5327 * Restore the global deferred-event list to its former length,
5328 * discarding any events queued by the subxact.
5329 */
5333 /*
5334 * Restore the trigger state. If the saved state is NULL, then this
5335 * subxact didn't save it, so it doesn't need restoring.
5336 */
5339 {
5342 }
5343 /* this avoids double pfree if error later: */
5346 /*
5347 * Scan for any remaining deferred events that were marked DONE or IN
5348 * PROGRESS by this subxact or a child, and un-mark them. We can
5349 * recognize such events because they have a firing ID greater than or
5350 * equal to the firing_counter value we saved at subtransaction start.
5351 * (This essentially assumes that the current subxact includes all
5352 * subxacts started after it.)
5353 */
5356 {
5361 {
5365 }
5366 }
5367 }
5368 }
5370 /*
5371 * Get the transition table for the given event and depending on whether we are
5372 * processing the old or the new tuple.
5373 */
5379 {
5386 /*
5387 * For INSERT events NEW should be non-NULL, for DELETE events OLD should
5388 * be non-NULL, whereas for UPDATE events normally both OLD and NEW are
5389 * non-NULL. But for UPDATE events fired for capturing transition tuples
5390 * during UPDATE partition-key row movement, OLD is NULL when the event is
5391 * for a row being inserted, whereas NEW is NULL when the event is for a
5392 * row being deleted.
5393 */
5400 {
5406 }
5408 {
5414 }
5417 }
5419 /*
5420 * Add the given heap tuple to the given tuplestore, applying the conversion
5421 * map if necessary.
5422 *
5423 * If original_insert_tuple is given, we can add that tuple without conversion.
5424 */
5425 static void
5432 {
5435 /*
5436 * Nothing needs to be done if we don't have a tuplestore.
5437 */
5444 {
5451 }
5452 else
5454 }
5456 /* ----------
5457 * AfterTriggerEnlargeQueryState()
5458 *
5459 * Prepare the necessary state so that we can record AFTER trigger events
5460 * queued by a query. It is allowed to have nested queries within a
5461 * (sub)transaction, so we need to have separate state for each query
5462 * nesting level.
5463 * ----------
5464 */
5465 static void
5467 {
5473 {
5480 }
5481 else
5482 {
5483 /* repalloc will keep the stack in the same context */
5492 }
5494 /* Initialize new array entries to empty */
5496 {
5506 }
5507 }
5509 /*
5510 * Create an empty SetConstraintState with room for numalloc trigstates
5511 */
5512 static SetConstraintState
5514 {
5515 SetConstraintState state;
5517 /* Behave sanely with numalloc == 0 */
5521 /*
5522 * We assume that zeroing will correctly initialize the state values.
5523 */
5532 }
5534 /*
5535 * Copy a SetConstraintState
5536 */
5537 static SetConstraintState
5539 {
5540 SetConstraintState state;
5551 }
5553 /*
5554 * Add a per-trigger item to a SetConstraintState. Returns possibly-changed
5555 * pointer to the state object (it will change if we have to repalloc).
5556 */
5557 static SetConstraintState
5560 {
5562 {
5572 }
5579 }
5581 /* ----------
5582 * AfterTriggerSetState()
5583 *
5584 * Execute the SET CONSTRAINTS ... utility command.
5585 * ----------
5586 */
5587 void
5589 {
5592 /* If we haven't already done so, initialize our state. */
5596 /*
5597 * If in a subtransaction, and we didn't save the current state already,
5598 * save it so it can be restored if the subtransaction aborts.
5599 */
5602 {
5605 }
5607 /*
5608 * Handle SET CONSTRAINTS ALL ...
5609 */
5611 {
5612 /*
5613 * Forget any previous SET CONSTRAINTS commands in this transaction.
5614 */
5617 /*
5618 * Set the per-transaction ALL state to known.
5619 */
5622 }
5623 else
5624 {
5625 Relation conrel;
5626 Relation tgrel;
5631 /*
5632 * Handle SET CONSTRAINTS constraint-name [, ...]
5633 *
5634 * First, identify all the named constraints and make a list of their
5635 * OIDs. Since, unlike the SQL spec, we allow multiple constraints of
5636 * the same name within a schema, the specifications are not
5637 * necessarily unique. Our strategy is to target all matching
5638 * constraints within the first search-path schema that has any
5639 * matches, but disregard matches in schemas beyond the first match.
5640 * (This is a bit odd but it's the historical behavior.)
5641 *
5642 * A constraint in a partitioned table may have corresponding
5643 * constraints in the partitions. Grab those too.
5644 */
5648 {
5655 {
5662 }
5664 /*
5665 * If we're given the schema name with the constraint, look only
5666 * in that schema. If given a bare constraint name, use the
5667 * search path to find the first matching constraint.
5668 */
5670 {
5675 }
5676 else
5677 {
5679 }
5683 {
5685 SysScanDesc conscan;
5687 HeapTuple tup;
5690 Anum_pg_constraint_conname,
5694 Anum_pg_constraint_connamespace,
5702 {
5713 }
5717 /*
5718 * Once we've found a matching constraint we do not search
5719 * later parts of the search path.
5720 */
5723 }
5727 /*
5728 * Not found ?
5729 */
5735 }
5737 /*
5738 * Scan for any possible descendants of the constraints. We append
5739 * whatever we find to the same list that we're scanning; this has the
5740 * effect that we create new scans for those, too, so if there are
5741 * further descendents, we'll also catch them.
5742 */
5744 {
5746 ScanKeyData key;
5747 SysScanDesc scan;
5748 HeapTuple tuple;
5751 Anum_pg_constraint_conparentid,
5758 {
5762 }
5765 }
5769 /*
5770 * Now, locate the trigger(s) implementing each of these constraints,
5771 * and make a list of their OIDs.
5772 */
5776 {
5778 ScanKeyData skey;
5779 SysScanDesc tgscan;
5780 HeapTuple htup;
5783 Anum_pg_trigger_tgconstraint,
5791 {
5794 /*
5795 * Silently skip triggers that are marked as non-deferrable in
5796 * pg_trigger. This is not an error condition, since a
5797 * deferrable RI constraint may have some non-deferrable
5798 * actions.
5799 */
5802 }
5805 }
5809 /*
5810 * Now we can set the trigger states of individual triggers for this
5811 * xact.
5812 */
5814 {
5821 {
5823 {
5827 }
5828 }
5830 {
5833 }
5834 }
5835 }
5837 /*
5838 * SQL99 requires that when a constraint is set to IMMEDIATE, any deferred
5839 * checks against that constraint must be made when the SET CONSTRAINTS
5840 * command is executed -- i.e. the effects of the SET CONSTRAINTS command
5841 * apply retroactively. We've updated the constraints state, so scan the
5842 * list of previously deferred events to fire any that have now become
5843 * immediate.
5844 *
5845 * Obviously, if this was SET ... DEFERRED then it can't have converted
5846 * any unfired events to immediate, so we need do nothing in that case.
5847 */
5849 {
5854 {
5857 /*
5858 * Make sure a snapshot has been established in case trigger
5859 * functions need one. Note that we avoid setting a snapshot if
5860 * we don't find at least one trigger that has to be fired now.
5861 * This is so that BEGIN; SET CONSTRAINTS ...; SET TRANSACTION
5862 * ISOLATION LEVEL SERIALIZABLE; ... works properly. (If we are
5863 * at the start of a transaction it's not possible for any trigger
5864 * events to be queued yet.)
5865 */
5867 {
5870 }
5872 /*
5873 * We can delete fired events if we are at top transaction level,
5874 * but we'd better not if inside a subtransaction, since the
5875 * subtransaction could later get rolled back.
5876 */
5880 }
5884 }
5885 }
5887 /* ----------
5888 * AfterTriggerPendingOnRel()
5889 * Test to see if there are any pending after-trigger events for rel.
5890 *
5891 * This is used by TRUNCATE, CLUSTER, ALTER TABLE, etc to detect whether
5892 * it is unsafe to perform major surgery on a relation. Note that only
5893 * local pending events are examined. We assume that having exclusive lock
5894 * on a rel guarantees there are no unserviced events in other backends ---
5895 * but having a lock does not prevent there being such events in our own.
5896 *
5897 * In some scenarios it'd be reasonable to remove pending events (more
5898 * specifically, mark them DONE by the current subxact) but without a lot
5899 * of knowledge of the trigger semantics we can't do this in general.
5900 * ----------
5901 */
5902 bool
5904 {
5905 AfterTriggerEvent event;
5909 /* Scan queued events */
5911 {
5914 /*
5915 * We can ignore completed events. (Even if a DONE flag is rolled
5916 * back by subxact abort, it's OK because the effects of the TRUNCATE
5917 * or whatever must get rolled back too.)
5918 */
5924 }
5926 /*
5927 * Also scan events queued by incomplete queries. This could only matter
5928 * if TRUNCATE/etc is executed by a function or trigger within an updating
5929 * query on the same relation, which is pretty perverse, but let's check.
5930 */
5931 for (depth = 0; depth <= afterTriggers.query_depth && depth < afterTriggers.maxquerydepth; depth++)
5932 {
5934 {
5942 }
5943 }
5946 }
5948 /* ----------
5949 * AfterTriggerSaveEvent()
5950 *
5951 * Called by ExecA[RS]...Triggers() to queue up the triggers that should
5952 * be fired for an event.
5953 *
5954 * NOTE: this is called whenever there are any triggers associated with
5955 * the event (even if they are disabled). This function decides which
5956 * triggers actually need to be queued. It is also called after each row,
5957 * even if there are no triggers for that event, if there are any AFTER
5958 * STATEMENT triggers for the statement which use transition tables, so that
5959 * the transition tuplestores can be built. Furthermore, if the transition
5960 * capture is happening for UPDATEd rows being moved to another partition due
5961 * to the partition-key being changed, then this function is called once when
5962 * the row is deleted (to capture OLD row), and once when the row is inserted
5963 * into another partition (to capture NEW row). This is done separately because
5964 * DELETE and INSERT happen on different tables.
5965 *
5966 * Transition tuplestores are built now, rather than when events are pulled
5967 * off of the queue because AFTER ROW triggers are allowed to select from the
5968 * transition tables for the statement.
5969 *
5970 * This contains special support to queue the update events for the case where
5971 * a partitioned table undergoing a cross-partition update may have foreign
5972 * keys pointing into it. Normally, a partitioned table's row triggers are
5973 * not fired because the leaf partition(s) which are modified as a result of
5974 * the operation on the partitioned table contain the same triggers which are
5975 * fired instead. But that general scheme can cause problematic behavior with
5976 * foreign key triggers during cross-partition updates, which are implemented
5977 * as DELETE on the source partition followed by INSERT into the destination
5978 * partition. Specifically, firing DELETE triggers would lead to the wrong
5979 * foreign key action to be enforced considering that the original command is
5980 * UPDATE; in this case, this function is called with relinfo as the
5981 * partitioned table, and src_partinfo and dst_partinfo referring to the
5982 * source and target leaf partitions, respectively.
5983 *
5984 * is_crosspart_update is true either when a DELETE event is fired on the
5985 * source partition (which is to be ignored) or an UPDATE event is fired on
5986 * the root partitioned table.
5987 * ----------
5988 */
5989 static void
5998 {
6001 AfterTriggerEventData new_event;
6002 AfterTriggerSharedData new_shared;
6009 /*
6010 * Check state. We use a normal test not Assert because it is possible to
6011 * reach here in the wrong state given misconfigured RI triggers, in
6012 * particular deferring a cascade action trigger.
6013 */
6017 /* Be sure we have enough space to record events at this query depth. */
6021 /*
6022 * If the directly named relation has any triggers with transition tables,
6023 * then we need to capture transition tuples.
6024 */
6026 {
6029 /*
6030 * Capture the old tuple in the appropriate transition table based on
6031 * the event.
6032 */
6034 {
6038 oldslot,
6039 NULL,
6040 transition_capture);
6043 }
6045 /*
6046 * Capture the new tuple in the appropriate transition table based on
6047 * the event.
6048 */
6050 {
6054 NULL,
6055 newslot,
6056 transition_capture);
6059 }
6061 /*
6062 * If transition tables are the only reason we're here, return. As
6063 * mentioned above, we can also be here during update tuple routing in
6064 * presence of transition tables, in which case this function is
6065 * called separately for OLD and NEW, so we expect exactly one of them
6066 * to be NULL.
6067 */
6074 }
6076 /*
6077 * We normally don't see partitioned tables here for row level triggers
6078 * except in the special case of a cross-partition update. In that case,
6079 * nodeModifyTable.c:ExecCrossPartitionUpdateForeignKey() calls here to
6080 * queue an update event on the root target partitioned table, also
6081 * passing the source and destination partitions and their tuples.
6082 */
6089 /*
6090 * Validate the event code and collect the associated tuple CTIDs.
6091 *
6092 * The event code will be used both as a bitmask and an array offset, so
6093 * validation is important to make sure we don't walk off the edge of our
6094 * arrays.
6095 *
6096 * Also, if we're considering statement-level triggers, check whether we
6097 * already queued a set of them for this event, and cancel the prior set
6098 * if so. This preserves the behavior that statement-level triggers fire
6099 * just once per statement and fire after row-level triggers.
6100 */
6102 {
6106 {
6111 }
6112 else
6113 {
6120 }
6125 {
6130 }
6131 else
6132 {
6139 }
6144 {
6150 /*
6151 * Also remember the OIDs of partitions to fetch these tuples
6152 * out of later in AfterTriggerExecute().
6153 */
6155 {
6161 }
6162 }
6163 else
6164 {
6171 }
6184 }
6186 /* Determine flags */
6188 {
6190 {
6193 else
6195 }
6196 else
6198 }
6200 /* else, we'll initialize ate_flags for each trigger */
6204 /*
6205 * Must convert/copy the source and destination partition tuples into the
6206 * root partitioned table's format/slot, because the processing in the
6207 * loop below expects both oldslot and newslot tuples to be in that form.
6208 */
6210 {
6218 oldslot,
6219 rootslot);
6220 else
6227 newslot,
6228 rootslot);
6229 else
6231 }
6234 {
6238 tgtype_level,
6239 TRIGGER_TYPE_AFTER,
6240 tgtype_event))
6247 {
6249 {
6252 }
6253 else
6254 /* subsequent event for the same tuple */
6256 }
6258 /*
6259 * If the trigger is a foreign key enforcement trigger, there are
6260 * certain cases where we can skip queueing the event because we can
6261 * tell by inspection that the FK constraint will still pass. There
6262 * are also some cases during cross-partition updates of a partitioned
6263 * table where queuing the event can be skipped.
6264 */
6266 {
6268 {
6271 /*
6272 * For cross-partitioned updates of partitioned PK table,
6273 * skip the event fired by the component delete on the
6274 * source leaf partition unless the constraint originates
6275 * in the partition itself (!tgisclone), because the
6276 * update event that will be fired on the root
6277 * (partitioned) target table will be used to perform the
6278 * necessary foreign key enforcement action.
6279 */
6285 /* Update or delete on trigger's PK table */
6288 {
6289 /* skip queuing this event */
6291 }
6296 /*
6297 * Update on trigger's FK table. We can skip the update
6298 * event fired on a partitioned table during a
6299 * cross-partition of that table, because the insert event
6300 * that is fired on the destination leaf partition would
6301 * suffice to perform the necessary foreign key check.
6302 * Moreover, RI_FKey_fk_upd_check_required() expects to be
6303 * passed a tuple that contains system attributes, most of
6304 * which are not present in the virtual slot belonging to
6305 * a partitioned table.
6306 */
6310 {
6311 /* skip queuing this event */
6313 }
6318 /*
6319 * Not an FK trigger. No need to queue the update event
6320 * fired during a cross-partitioned update of a
6321 * partitioned table, because the same row trigger must be
6322 * present in the leaf partition(s) that are affected as
6323 * part of this update and the events fired on them are
6324 * queued instead.
6325 */
6330 }
6331 }
6333 /*
6334 * If the trigger is a deferred unique constraint check trigger, only
6335 * queue it if the unique constraint was potentially violated, which
6336 * we know from index insertion time.
6337 */
6339 {
6342 }
6344 /*
6345 * Fill in event structure and add it to the current query's queue.
6346 * Note we set ats_table to NULL whenever this trigger doesn't use
6347 * transition tables, to improve sharability of the shared event data.
6348 */
6360 else
6366 }
6368 /*
6369 * Finally, spool any foreign tuple(s). The tuplestore squashes them to
6370 * minimal tuples, so this loses any system columns. The executor lost
6371 * those columns before us, for an unrelated reason, so this is fine.
6372 */
6374 {
6379 }
6380 }
6382 /*
6383 * Detect whether we already queued BEFORE STATEMENT triggers for the given
6384 * relation + operation, and set the flag so the next call will report "true".
6385 */
6386 static bool
6388 {
6392 /* Check state, like AfterTriggerSaveEvent. */
6396 /* Be sure we have enough space to record events at this query depth. */
6400 /*
6401 * We keep this state in the AfterTriggersTableData that also holds
6402 * transition tables for the relation + operation. In this way, if we are
6403 * forced to make a new set of transition tables because more tuples get
6404 * entered after we've already fired triggers, we will allow a new set of
6405 * statement triggers to get queued.
6406 */
6411 }
6413 /*
6414 * If we previously queued a set of AFTER STATEMENT triggers for the given
6415 * relation + operation, and they've not been fired yet, cancel them. The
6416 * caller will queue a fresh set that's after any row-level triggers that may
6417 * have been queued by the current sub-statement, preserving (as much as
6418 * possible) the property that AFTER ROW triggers fire before AFTER STATEMENT
6419 * triggers, and that the latter only fire once. This deals with the
6420 * situation where several FK enforcement triggers sequentially queue triggers
6421 * for the same table into the same trigger query level. We can't fully
6422 * prevent odd behavior though: if there are AFTER ROW triggers taking
6423 * transition tables, we don't want to change the transition tables once the
6424 * first such trigger has seen them. In such a case, any additional events
6425 * will result in creating new transition tables and allowing new firings of
6426 * statement triggers.
6427 *
6428 * This also saves the current event list location so that a later invocation
6429 * of this function can cheaply find the triggers we're about to queue and
6430 * cancel them.
6431 */
6432 static void
6434 {
6438 /*
6439 * We keep this state in the AfterTriggersTableData that also holds
6440 * transition tables for the relation + operation. In this way, if we are
6441 * forced to make a new set of transition tables because more tuples get
6442 * entered after we've already fired triggers, we will allow a new set of
6443 * statement triggers to get queued without canceling the old ones.
6444 */
6448 {
6449 /*
6450 * We want to start scanning from the tail location that existed just
6451 * before we inserted any statement triggers. But the events list
6452 * might've been entirely empty then, in which case scan from the
6453 * current head.
6454 */
6455 AfterTriggerEvent event;
6459 {
6462 }
6463 else
6464 {
6467 }
6470 {
6474 {
6477 /*
6478 * Exit loop when we reach events that aren't AS triggers for
6479 * the target relation.
6480 */
6489 /* OK, mark it DONE */
6492 }
6493 /* signal we must reinitialize event ptr for next chunk */
6495 }
6496 }
6497 done:
6499 /* In any case, save current insertion point for next time */
6502 }
6504 /*
6505 * GUC assign_hook for session_replication_role
6506 */
6507 void
6509 {
6510 /*
6511 * Must flush the plan cache when changing replication role; but don't
6512 * flush unnecessarily.
6513 */
6516 }
6518 /*
6519 * SQL function pg_trigger_depth()
6520 */
6521 Datum
6523 {
6525 }