Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * relcache.c
4 : * POSTGRES relation descriptor cache code
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/cache/relcache.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : /*
16 : * INTERFACE ROUTINES
17 : * RelationCacheInitialize - initialize relcache (to empty)
18 : * RelationCacheInitializePhase2 - initialize shared-catalog entries
19 : * RelationCacheInitializePhase3 - finish initializing relcache
20 : * RelationIdGetRelation - get a reldesc by relation id
21 : * RelationClose - close an open relation
22 : *
23 : * NOTES
24 : * The following code contains many undocumented hacks. Please be
25 : * careful....
26 : */
27 : #include "postgres.h"
28 :
29 : #include <sys/file.h>
30 : #include <fcntl.h>
31 : #include <unistd.h>
32 :
33 : #include "access/htup_details.h"
34 : #include "access/multixact.h"
35 : #include "access/parallel.h"
36 : #include "access/reloptions.h"
37 : #include "access/sysattr.h"
38 : #include "access/table.h"
39 : #include "access/tableam.h"
40 : #include "access/tupdesc_details.h"
41 : #include "access/xact.h"
42 : #include "catalog/binary_upgrade.h"
43 : #include "catalog/catalog.h"
44 : #include "catalog/indexing.h"
45 : #include "catalog/namespace.h"
46 : #include "catalog/partition.h"
47 : #include "catalog/pg_am.h"
48 : #include "catalog/pg_amproc.h"
49 : #include "catalog/pg_attrdef.h"
50 : #include "catalog/pg_auth_members.h"
51 : #include "catalog/pg_authid.h"
52 : #include "catalog/pg_constraint.h"
53 : #include "catalog/pg_database.h"
54 : #include "catalog/pg_namespace.h"
55 : #include "catalog/pg_opclass.h"
56 : #include "catalog/pg_proc.h"
57 : #include "catalog/pg_publication.h"
58 : #include "catalog/pg_rewrite.h"
59 : #include "catalog/pg_shseclabel.h"
60 : #include "catalog/pg_statistic_ext.h"
61 : #include "catalog/pg_subscription.h"
62 : #include "catalog/pg_tablespace.h"
63 : #include "catalog/pg_trigger.h"
64 : #include "catalog/pg_type.h"
65 : #include "catalog/schemapg.h"
66 : #include "catalog/storage.h"
67 : #include "commands/policy.h"
68 : #include "commands/publicationcmds.h"
69 : #include "commands/trigger.h"
70 : #include "common/int.h"
71 : #include "miscadmin.h"
72 : #include "nodes/makefuncs.h"
73 : #include "nodes/nodeFuncs.h"
74 : #include "optimizer/optimizer.h"
75 : #include "pgstat.h"
76 : #include "rewrite/rewriteDefine.h"
77 : #include "rewrite/rowsecurity.h"
78 : #include "storage/lmgr.h"
79 : #include "storage/smgr.h"
80 : #include "utils/array.h"
81 : #include "utils/builtins.h"
82 : #include "utils/catcache.h"
83 : #include "utils/datum.h"
84 : #include "utils/fmgroids.h"
85 : #include "utils/inval.h"
86 : #include "utils/lsyscache.h"
87 : #include "utils/memutils.h"
88 : #include "utils/relmapper.h"
89 : #include "utils/resowner.h"
90 : #include "utils/snapmgr.h"
91 : #include "utils/syscache.h"
92 :
93 : #define RELCACHE_INIT_FILEMAGIC 0x573266 /* version ID value */
94 :
95 : /*
96 : * Whether to bother checking if relation cache memory needs to be freed
97 : * eagerly. See also RelationBuildDesc() and pg_config_manual.h.
98 : */
99 : #if defined(RECOVER_RELATION_BUILD_MEMORY) && (RECOVER_RELATION_BUILD_MEMORY != 0)
100 : #define MAYBE_RECOVER_RELATION_BUILD_MEMORY 1
101 : #else
102 : #define RECOVER_RELATION_BUILD_MEMORY 0
103 : #ifdef DISCARD_CACHES_ENABLED
104 : #define MAYBE_RECOVER_RELATION_BUILD_MEMORY 1
105 : #endif
106 : #endif
107 :
108 : /*
109 : * hardcoded tuple descriptors, contents generated by genbki.pl
110 : */
111 : static const FormData_pg_attribute Desc_pg_class[Natts_pg_class] = {Schema_pg_class};
112 : static const FormData_pg_attribute Desc_pg_attribute[Natts_pg_attribute] = {Schema_pg_attribute};
113 : static const FormData_pg_attribute Desc_pg_proc[Natts_pg_proc] = {Schema_pg_proc};
114 : static const FormData_pg_attribute Desc_pg_type[Natts_pg_type] = {Schema_pg_type};
115 : static const FormData_pg_attribute Desc_pg_database[Natts_pg_database] = {Schema_pg_database};
116 : static const FormData_pg_attribute Desc_pg_authid[Natts_pg_authid] = {Schema_pg_authid};
117 : static const FormData_pg_attribute Desc_pg_auth_members[Natts_pg_auth_members] = {Schema_pg_auth_members};
118 : static const FormData_pg_attribute Desc_pg_index[Natts_pg_index] = {Schema_pg_index};
119 : static const FormData_pg_attribute Desc_pg_shseclabel[Natts_pg_shseclabel] = {Schema_pg_shseclabel};
120 : static const FormData_pg_attribute Desc_pg_subscription[Natts_pg_subscription] = {Schema_pg_subscription};
121 :
122 : /*
123 : * Hash tables that index the relation cache
124 : *
125 : * We used to index the cache by both name and OID, but now there
126 : * is only an index by OID.
127 : */
128 : typedef struct relidcacheent
129 : {
130 : Oid reloid;
131 : Relation reldesc;
132 : } RelIdCacheEnt;
133 :
134 : static HTAB *RelationIdCache;
135 :
136 : /*
137 : * This flag is false until we have prepared the critical relcache entries
138 : * that are needed to do indexscans on the tables read by relcache building.
139 : */
140 : bool criticalRelcachesBuilt = false;
141 :
142 : /*
143 : * This flag is false until we have prepared the critical relcache entries
144 : * for shared catalogs (which are the tables needed for login).
145 : */
146 : bool criticalSharedRelcachesBuilt = false;
147 :
148 : /*
149 : * This counter counts relcache inval events received since backend startup
150 : * (but only for rels that are actually in cache). Presently, we use it only
151 : * to detect whether data about to be written by write_relcache_init_file()
152 : * might already be obsolete.
153 : */
154 : static long relcacheInvalsReceived = 0L;
155 :
156 : /*
157 : * in_progress_list is a stack of ongoing RelationBuildDesc() calls. CREATE
158 : * INDEX CONCURRENTLY makes catalog changes under ShareUpdateExclusiveLock.
159 : * It critically relies on each backend absorbing those changes no later than
160 : * next transaction start. Hence, RelationBuildDesc() loops until it finishes
161 : * without accepting a relevant invalidation. (Most invalidation consumers
162 : * don't do this.)
163 : */
164 : typedef struct inprogressent
165 : {
166 : Oid reloid; /* OID of relation being built */
167 : bool invalidated; /* whether an invalidation arrived for it */
168 : } InProgressEnt;
169 :
170 : static InProgressEnt *in_progress_list;
171 : static int in_progress_list_len;
172 : static int in_progress_list_maxlen;
173 :
174 : /*
175 : * eoxact_list[] stores the OIDs of relations that (might) need AtEOXact
176 : * cleanup work. This list intentionally has limited size; if it overflows,
177 : * we fall back to scanning the whole hashtable. There is no value in a very
178 : * large list because (1) at some point, a hash_seq_search scan is faster than
179 : * retail lookups, and (2) the value of this is to reduce EOXact work for
180 : * short transactions, which can't have dirtied all that many tables anyway.
181 : * EOXactListAdd() does not bother to prevent duplicate list entries, so the
182 : * cleanup processing must be idempotent.
183 : */
184 : #define MAX_EOXACT_LIST 32
185 : static Oid eoxact_list[MAX_EOXACT_LIST];
186 : static int eoxact_list_len = 0;
187 : static bool eoxact_list_overflowed = false;
188 :
189 : #define EOXactListAdd(rel) \
190 : do { \
191 : if (eoxact_list_len < MAX_EOXACT_LIST) \
192 : eoxact_list[eoxact_list_len++] = (rel)->rd_id; \
193 : else \
194 : eoxact_list_overflowed = true; \
195 : } while (0)
196 :
197 : /*
198 : * EOXactTupleDescArray stores TupleDescs that (might) need AtEOXact
199 : * cleanup work. The array expands as needed; there is no hashtable because
200 : * we don't need to access individual items except at EOXact.
201 : */
202 : static TupleDesc *EOXactTupleDescArray;
203 : static int NextEOXactTupleDescNum = 0;
204 : static int EOXactTupleDescArrayLen = 0;
205 :
206 : /*
207 : * macros to manipulate the lookup hashtable
208 : */
209 : #define RelationCacheInsert(RELATION, replace_allowed) \
210 : do { \
211 : RelIdCacheEnt *hentry; bool found; \
212 : hentry = (RelIdCacheEnt *) hash_search(RelationIdCache, \
213 : &((RELATION)->rd_id), \
214 : HASH_ENTER, &found); \
215 : if (found) \
216 : { \
217 : /* see comments in RelationBuildDesc and RelationBuildLocalRelation */ \
218 : Relation _old_rel = hentry->reldesc; \
219 : Assert(replace_allowed); \
220 : hentry->reldesc = (RELATION); \
221 : if (RelationHasReferenceCountZero(_old_rel)) \
222 : RelationDestroyRelation(_old_rel, false); \
223 : else if (!IsBootstrapProcessingMode()) \
224 : elog(WARNING, "leaking still-referenced relcache entry for \"%s\"", \
225 : RelationGetRelationName(_old_rel)); \
226 : } \
227 : else \
228 : hentry->reldesc = (RELATION); \
229 : } while(0)
230 :
231 : #define RelationIdCacheLookup(ID, RELATION) \
232 : do { \
233 : RelIdCacheEnt *hentry; \
234 : hentry = (RelIdCacheEnt *) hash_search(RelationIdCache, \
235 : &(ID), \
236 : HASH_FIND, NULL); \
237 : if (hentry) \
238 : RELATION = hentry->reldesc; \
239 : else \
240 : RELATION = NULL; \
241 : } while(0)
242 :
243 : #define RelationCacheDelete(RELATION) \
244 : do { \
245 : RelIdCacheEnt *hentry; \
246 : hentry = (RelIdCacheEnt *) hash_search(RelationIdCache, \
247 : &((RELATION)->rd_id), \
248 : HASH_REMOVE, NULL); \
249 : if (hentry == NULL) \
250 : elog(WARNING, "failed to delete relcache entry for OID %u", \
251 : (RELATION)->rd_id); \
252 : } while(0)
253 :
254 :
255 : /*
256 : * Special cache for opclass-related information
257 : *
258 : * Note: only default support procs get cached, ie, those with
259 : * lefttype = righttype = opcintype.
260 : */
261 : typedef struct opclasscacheent
262 : {
263 : Oid opclassoid; /* lookup key: OID of opclass */
264 : bool valid; /* set true after successful fill-in */
265 : StrategyNumber numSupport; /* max # of support procs (from pg_am) */
266 : Oid opcfamily; /* OID of opclass's family */
267 : Oid opcintype; /* OID of opclass's declared input type */
268 : RegProcedure *supportProcs; /* OIDs of support procedures */
269 : } OpClassCacheEnt;
270 :
271 : static HTAB *OpClassCache = NULL;
272 :
273 :
274 : /* non-export function prototypes */
275 :
276 : static void RelationCloseCleanup(Relation relation);
277 : static void RelationDestroyRelation(Relation relation, bool remember_tupdesc);
278 : static void RelationClearRelation(Relation relation, bool rebuild);
279 :
280 : static void RelationReloadIndexInfo(Relation relation);
281 : static void RelationReloadNailed(Relation relation);
282 : static void RelationFlushRelation(Relation relation);
283 : static void RememberToFreeTupleDescAtEOX(TupleDesc td);
284 : #ifdef USE_ASSERT_CHECKING
285 : static void AssertPendingSyncConsistency(Relation relation);
286 : #endif
287 : static void AtEOXact_cleanup(Relation relation, bool isCommit);
288 : static void AtEOSubXact_cleanup(Relation relation, bool isCommit,
289 : SubTransactionId mySubid, SubTransactionId parentSubid);
290 : static bool load_relcache_init_file(bool shared);
291 : static void write_relcache_init_file(bool shared);
292 : static void write_item(const void *data, Size len, FILE *fp);
293 :
294 : static void formrdesc(const char *relationName, Oid relationReltype,
295 : bool isshared, int natts, const FormData_pg_attribute *attrs);
296 :
297 : static HeapTuple ScanPgRelation(Oid targetRelId, bool indexOK, bool force_non_historic);
298 : static Relation AllocateRelationDesc(Form_pg_class relp);
299 : static void RelationParseRelOptions(Relation relation, HeapTuple tuple);
300 : static void RelationBuildTupleDesc(Relation relation);
301 : static Relation RelationBuildDesc(Oid targetRelId, bool insertIt);
302 : static void RelationInitPhysicalAddr(Relation relation);
303 : static void load_critical_index(Oid indexoid, Oid heapoid);
304 : static TupleDesc GetPgClassDescriptor(void);
305 : static TupleDesc GetPgIndexDescriptor(void);
306 : static void AttrDefaultFetch(Relation relation, int ndef);
307 : static int AttrDefaultCmp(const void *a, const void *b);
308 : static void CheckConstraintFetch(Relation relation);
309 : static int CheckConstraintCmp(const void *a, const void *b);
310 : static void InitIndexAmRoutine(Relation relation);
311 : static void IndexSupportInitialize(oidvector *indclass,
312 : RegProcedure *indexSupport,
313 : Oid *opFamily,
314 : Oid *opcInType,
315 : StrategyNumber maxSupportNumber,
316 : AttrNumber maxAttributeNumber);
317 : static OpClassCacheEnt *LookupOpclassInfo(Oid operatorClassOid,
318 : StrategyNumber numSupport);
319 : static void RelationCacheInitFileRemoveInDir(const char *tblspcpath);
320 : static void unlink_initfile(const char *initfilename, int elevel);
321 :
322 :
323 : /*
324 : * ScanPgRelation
325 : *
326 : * This is used by RelationBuildDesc to find a pg_class
327 : * tuple matching targetRelId. The caller must hold at least
328 : * AccessShareLock on the target relid to prevent concurrent-update
329 : * scenarios; it isn't guaranteed that all scans used to build the
330 : * relcache entry will use the same snapshot. If, for example,
331 : * an attribute were to be added after scanning pg_class and before
332 : * scanning pg_attribute, relnatts wouldn't match.
333 : *
334 : * NB: the returned tuple has been copied into palloc'd storage
335 : * and must eventually be freed with heap_freetuple.
336 : */
337 : static HeapTuple
338 1156144 : ScanPgRelation(Oid targetRelId, bool indexOK, bool force_non_historic)
339 : {
340 : HeapTuple pg_class_tuple;
341 : Relation pg_class_desc;
342 : SysScanDesc pg_class_scan;
343 : ScanKeyData key[1];
344 1156144 : Snapshot snapshot = NULL;
345 :
346 : /*
347 : * If something goes wrong during backend startup, we might find ourselves
348 : * trying to read pg_class before we've selected a database. That ain't
349 : * gonna work, so bail out with a useful error message. If this happens,
350 : * it probably means a relcache entry that needs to be nailed isn't.
351 : */
352 1156144 : if (!OidIsValid(MyDatabaseId))
353 0 : elog(FATAL, "cannot read pg_class without having selected a database");
354 :
355 : /*
356 : * form a scan key
357 : */
358 1156144 : ScanKeyInit(&key[0],
359 : Anum_pg_class_oid,
360 : BTEqualStrategyNumber, F_OIDEQ,
361 : ObjectIdGetDatum(targetRelId));
362 :
363 : /*
364 : * Open pg_class and fetch a tuple. Force heap scan if we haven't yet
365 : * built the critical relcache entries (this includes initdb and startup
366 : * without a pg_internal.init file). The caller can also force a heap
367 : * scan by setting indexOK == false.
368 : */
369 1156144 : pg_class_desc = table_open(RelationRelationId, AccessShareLock);
370 :
371 : /*
372 : * The caller might need a tuple that's newer than the one the historic
373 : * snapshot; currently the only case requiring to do so is looking up the
374 : * relfilenumber of non mapped system relations during decoding. That
375 : * snapshot can't change in the midst of a relcache build, so there's no
376 : * need to register the snapshot.
377 : */
378 1156144 : if (force_non_historic)
379 2628 : snapshot = GetNonHistoricCatalogSnapshot(RelationRelationId);
380 :
381 1156144 : pg_class_scan = systable_beginscan(pg_class_desc, ClassOidIndexId,
382 1156144 : indexOK && criticalRelcachesBuilt,
383 : snapshot,
384 : 1, key);
385 :
386 1156138 : pg_class_tuple = systable_getnext(pg_class_scan);
387 :
388 : /*
389 : * Must copy tuple before releasing buffer.
390 : */
391 1156132 : if (HeapTupleIsValid(pg_class_tuple))
392 1156122 : pg_class_tuple = heap_copytuple(pg_class_tuple);
393 :
394 : /* all done */
395 1156132 : systable_endscan(pg_class_scan);
396 1156132 : table_close(pg_class_desc, AccessShareLock);
397 :
398 1156132 : return pg_class_tuple;
399 : }
400 :
401 : /*
402 : * AllocateRelationDesc
403 : *
404 : * This is used to allocate memory for a new relation descriptor
405 : * and initialize the rd_rel field from the given pg_class tuple.
406 : */
407 : static Relation
408 1062068 : AllocateRelationDesc(Form_pg_class relp)
409 : {
410 : Relation relation;
411 : MemoryContext oldcxt;
412 : Form_pg_class relationForm;
413 :
414 : /* Relcache entries must live in CacheMemoryContext */
415 1062068 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
416 :
417 : /*
418 : * allocate and zero space for new relation descriptor
419 : */
420 1062068 : relation = (Relation) palloc0(sizeof(RelationData));
421 :
422 : /* make sure relation is marked as having no open file yet */
423 1062068 : relation->rd_smgr = NULL;
424 :
425 : /*
426 : * Copy the relation tuple form
427 : *
428 : * We only allocate space for the fixed fields, ie, CLASS_TUPLE_SIZE. The
429 : * variable-length fields (relacl, reloptions) are NOT stored in the
430 : * relcache --- there'd be little point in it, since we don't copy the
431 : * tuple's nulls bitmap and hence wouldn't know if the values are valid.
432 : * Bottom line is that relacl *cannot* be retrieved from the relcache. Get
433 : * it from the syscache if you need it. The same goes for the original
434 : * form of reloptions (however, we do store the parsed form of reloptions
435 : * in rd_options).
436 : */
437 1062068 : relationForm = (Form_pg_class) palloc(CLASS_TUPLE_SIZE);
438 :
439 1062068 : memcpy(relationForm, relp, CLASS_TUPLE_SIZE);
440 :
441 : /* initialize relation tuple form */
442 1062068 : relation->rd_rel = relationForm;
443 :
444 : /* and allocate attribute tuple form storage */
445 1062068 : relation->rd_att = CreateTemplateTupleDesc(relationForm->relnatts);
446 : /* which we mark as a reference-counted tupdesc */
447 1062068 : relation->rd_att->tdrefcount = 1;
448 :
449 1062068 : MemoryContextSwitchTo(oldcxt);
450 :
451 1062068 : return relation;
452 : }
453 :
454 : /*
455 : * RelationParseRelOptions
456 : * Convert pg_class.reloptions into pre-parsed rd_options
457 : *
458 : * tuple is the real pg_class tuple (not rd_rel!) for relation
459 : *
460 : * Note: rd_rel and (if an index) rd_indam must be valid already
461 : */
462 : static void
463 1148452 : RelationParseRelOptions(Relation relation, HeapTuple tuple)
464 : {
465 : bytea *options;
466 : amoptions_function amoptsfn;
467 :
468 1148452 : relation->rd_options = NULL;
469 :
470 : /*
471 : * Look up any AM-specific parse function; fall out if relkind should not
472 : * have options.
473 : */
474 1148452 : switch (relation->rd_rel->relkind)
475 : {
476 660590 : case RELKIND_RELATION:
477 : case RELKIND_TOASTVALUE:
478 : case RELKIND_VIEW:
479 : case RELKIND_MATVIEW:
480 : case RELKIND_PARTITIONED_TABLE:
481 660590 : amoptsfn = NULL;
482 660590 : break;
483 478110 : case RELKIND_INDEX:
484 : case RELKIND_PARTITIONED_INDEX:
485 478110 : amoptsfn = relation->rd_indam->amoptions;
486 478110 : break;
487 9752 : default:
488 9752 : return;
489 : }
490 :
491 : /*
492 : * Fetch reloptions from tuple; have to use a hardwired descriptor because
493 : * we might not have any other for pg_class yet (consider executing this
494 : * code for pg_class itself)
495 : */
496 1138700 : options = extractRelOptions(tuple, GetPgClassDescriptor(), amoptsfn);
497 :
498 : /*
499 : * Copy parsed data into CacheMemoryContext. To guard against the
500 : * possibility of leaks in the reloptions code, we want to do the actual
501 : * parsing in the caller's memory context and copy the results into
502 : * CacheMemoryContext after the fact.
503 : */
504 1138700 : if (options)
505 : {
506 26256 : relation->rd_options = MemoryContextAlloc(CacheMemoryContext,
507 13128 : VARSIZE(options));
508 13128 : memcpy(relation->rd_options, options, VARSIZE(options));
509 13128 : pfree(options);
510 : }
511 : }
512 :
513 : /*
514 : * RelationBuildTupleDesc
515 : *
516 : * Form the relation's tuple descriptor from information in
517 : * the pg_attribute, pg_attrdef & pg_constraint system catalogs.
518 : */
519 : static void
520 1062068 : RelationBuildTupleDesc(Relation relation)
521 : {
522 : HeapTuple pg_attribute_tuple;
523 : Relation pg_attribute_desc;
524 : SysScanDesc pg_attribute_scan;
525 : ScanKeyData skey[2];
526 : int need;
527 : TupleConstr *constr;
528 1062068 : AttrMissing *attrmiss = NULL;
529 1062068 : int ndef = 0;
530 :
531 : /* fill rd_att's type ID fields (compare heap.c's AddNewRelationTuple) */
532 1062068 : relation->rd_att->tdtypeid =
533 1062068 : relation->rd_rel->reltype ? relation->rd_rel->reltype : RECORDOID;
534 1062068 : relation->rd_att->tdtypmod = -1; /* just to be sure */
535 :
536 1062068 : constr = (TupleConstr *) MemoryContextAllocZero(CacheMemoryContext,
537 : sizeof(TupleConstr));
538 1062068 : constr->has_not_null = false;
539 1062068 : constr->has_generated_stored = false;
540 :
541 : /*
542 : * Form a scan key that selects only user attributes (attnum > 0).
543 : * (Eliminating system attribute rows at the index level is lots faster
544 : * than fetching them.)
545 : */
546 1062068 : ScanKeyInit(&skey[0],
547 : Anum_pg_attribute_attrelid,
548 : BTEqualStrategyNumber, F_OIDEQ,
549 : ObjectIdGetDatum(RelationGetRelid(relation)));
550 1062068 : ScanKeyInit(&skey[1],
551 : Anum_pg_attribute_attnum,
552 : BTGreaterStrategyNumber, F_INT2GT,
553 : Int16GetDatum(0));
554 :
555 : /*
556 : * Open pg_attribute and begin a scan. Force heap scan if we haven't yet
557 : * built the critical relcache entries (this includes initdb and startup
558 : * without a pg_internal.init file).
559 : */
560 1062068 : pg_attribute_desc = table_open(AttributeRelationId, AccessShareLock);
561 1062068 : pg_attribute_scan = systable_beginscan(pg_attribute_desc,
562 : AttributeRelidNumIndexId,
563 : criticalRelcachesBuilt,
564 : NULL,
565 : 2, skey);
566 :
567 : /*
568 : * add attribute data to relation->rd_att
569 : */
570 1062068 : need = RelationGetNumberOfAttributes(relation);
571 :
572 3667384 : while (HeapTupleIsValid(pg_attribute_tuple = systable_getnext(pg_attribute_scan)))
573 : {
574 : Form_pg_attribute attp;
575 : int attnum;
576 :
577 3666356 : attp = (Form_pg_attribute) GETSTRUCT(pg_attribute_tuple);
578 :
579 3666356 : attnum = attp->attnum;
580 3666356 : if (attnum <= 0 || attnum > RelationGetNumberOfAttributes(relation))
581 0 : elog(ERROR, "invalid attribute number %d for relation \"%s\"",
582 : attp->attnum, RelationGetRelationName(relation));
583 :
584 3666356 : memcpy(TupleDescAttr(relation->rd_att, attnum - 1),
585 : attp,
586 : ATTRIBUTE_FIXED_PART_SIZE);
587 :
588 : /* Update constraint/default info */
589 3666356 : if (attp->attnotnull)
590 1476782 : constr->has_not_null = true;
591 3666356 : if (attp->attgenerated == ATTRIBUTE_GENERATED_STORED)
592 8552 : constr->has_generated_stored = true;
593 3666356 : if (attp->atthasdef)
594 39334 : ndef++;
595 :
596 : /* If the column has a "missing" value, put it in the attrmiss array */
597 3666356 : if (attp->atthasmissing)
598 : {
599 : Datum missingval;
600 : bool missingNull;
601 :
602 : /* Do we have a missing value? */
603 7622 : missingval = heap_getattr(pg_attribute_tuple,
604 : Anum_pg_attribute_attmissingval,
605 : pg_attribute_desc->rd_att,
606 : &missingNull);
607 7622 : if (!missingNull)
608 : {
609 : /* Yes, fetch from the array */
610 : MemoryContext oldcxt;
611 : bool is_null;
612 7622 : int one = 1;
613 : Datum missval;
614 :
615 7622 : if (attrmiss == NULL)
616 : attrmiss = (AttrMissing *)
617 3666 : MemoryContextAllocZero(CacheMemoryContext,
618 3666 : relation->rd_rel->relnatts *
619 : sizeof(AttrMissing));
620 :
621 7622 : missval = array_get_element(missingval,
622 : 1,
623 : &one,
624 : -1,
625 7622 : attp->attlen,
626 7622 : attp->attbyval,
627 7622 : attp->attalign,
628 : &is_null);
629 : Assert(!is_null);
630 7622 : if (attp->attbyval)
631 : {
632 : /* for copy by val just copy the datum direct */
633 4820 : attrmiss[attnum - 1].am_value = missval;
634 : }
635 : else
636 : {
637 : /* otherwise copy in the correct context */
638 2802 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
639 5604 : attrmiss[attnum - 1].am_value = datumCopy(missval,
640 2802 : attp->attbyval,
641 2802 : attp->attlen);
642 2802 : MemoryContextSwitchTo(oldcxt);
643 : }
644 7622 : attrmiss[attnum - 1].am_present = true;
645 : }
646 : }
647 3666356 : need--;
648 3666356 : if (need == 0)
649 1061040 : break;
650 : }
651 :
652 : /*
653 : * end the scan and close the attribute relation
654 : */
655 1062068 : systable_endscan(pg_attribute_scan);
656 1062068 : table_close(pg_attribute_desc, AccessShareLock);
657 :
658 1062068 : if (need != 0)
659 0 : elog(ERROR, "pg_attribute catalog is missing %d attribute(s) for relation OID %u",
660 : need, RelationGetRelid(relation));
661 :
662 : /*
663 : * The attcacheoff values we read from pg_attribute should all be -1
664 : * ("unknown"). Verify this if assert checking is on. They will be
665 : * computed when and if needed during tuple access.
666 : */
667 : #ifdef USE_ASSERT_CHECKING
668 : {
669 : int i;
670 :
671 : for (i = 0; i < RelationGetNumberOfAttributes(relation); i++)
672 : Assert(TupleDescAttr(relation->rd_att, i)->attcacheoff == -1);
673 : }
674 : #endif
675 :
676 : /*
677 : * However, we can easily set the attcacheoff value for the first
678 : * attribute: it must be zero. This eliminates the need for special cases
679 : * for attnum=1 that used to exist in fastgetattr() and index_getattr().
680 : */
681 1062068 : if (RelationGetNumberOfAttributes(relation) > 0)
682 1061040 : TupleDescAttr(relation->rd_att, 0)->attcacheoff = 0;
683 :
684 : /*
685 : * Set up constraint/default info
686 : */
687 1062068 : if (constr->has_not_null ||
688 735716 : constr->has_generated_stored ||
689 726218 : ndef > 0 ||
690 725990 : attrmiss ||
691 725990 : relation->rd_rel->relchecks > 0)
692 : {
693 341356 : relation->rd_att->constr = constr;
694 :
695 341356 : if (ndef > 0) /* DEFAULTs */
696 28096 : AttrDefaultFetch(relation, ndef);
697 : else
698 313260 : constr->num_defval = 0;
699 :
700 341356 : constr->missing = attrmiss;
701 :
702 341356 : if (relation->rd_rel->relchecks > 0) /* CHECKs */
703 11180 : CheckConstraintFetch(relation);
704 : else
705 330176 : constr->num_check = 0;
706 : }
707 : else
708 : {
709 720712 : pfree(constr);
710 720712 : relation->rd_att->constr = NULL;
711 : }
712 1062068 : }
713 :
714 : /*
715 : * RelationBuildRuleLock
716 : *
717 : * Form the relation's rewrite rules from information in
718 : * the pg_rewrite system catalog.
719 : *
720 : * Note: The rule parsetrees are potentially very complex node structures.
721 : * To allow these trees to be freed when the relcache entry is flushed,
722 : * we make a private memory context to hold the RuleLock information for
723 : * each relcache entry that has associated rules. The context is used
724 : * just for rule info, not for any other subsidiary data of the relcache
725 : * entry, because that keeps the update logic in RelationClearRelation()
726 : * manageable. The other subsidiary data structures are simple enough
727 : * to be easy to free explicitly, anyway.
728 : *
729 : * Note: The relation's reloptions must have been extracted first.
730 : */
731 : static void
732 31634 : RelationBuildRuleLock(Relation relation)
733 : {
734 : MemoryContext rulescxt;
735 : MemoryContext oldcxt;
736 : HeapTuple rewrite_tuple;
737 : Relation rewrite_desc;
738 : TupleDesc rewrite_tupdesc;
739 : SysScanDesc rewrite_scan;
740 : ScanKeyData key;
741 : RuleLock *rulelock;
742 : int numlocks;
743 : RewriteRule **rules;
744 : int maxlocks;
745 :
746 : /*
747 : * Make the private context. Assume it'll not contain much data.
748 : */
749 31634 : rulescxt = AllocSetContextCreate(CacheMemoryContext,
750 : "relation rules",
751 : ALLOCSET_SMALL_SIZES);
752 31634 : relation->rd_rulescxt = rulescxt;
753 31634 : MemoryContextCopyAndSetIdentifier(rulescxt,
754 : RelationGetRelationName(relation));
755 :
756 : /*
757 : * allocate an array to hold the rewrite rules (the array is extended if
758 : * necessary)
759 : */
760 31634 : maxlocks = 4;
761 : rules = (RewriteRule **)
762 31634 : MemoryContextAlloc(rulescxt, sizeof(RewriteRule *) * maxlocks);
763 31634 : numlocks = 0;
764 :
765 : /*
766 : * form a scan key
767 : */
768 31634 : ScanKeyInit(&key,
769 : Anum_pg_rewrite_ev_class,
770 : BTEqualStrategyNumber, F_OIDEQ,
771 : ObjectIdGetDatum(RelationGetRelid(relation)));
772 :
773 : /*
774 : * open pg_rewrite and begin a scan
775 : *
776 : * Note: since we scan the rules using RewriteRelRulenameIndexId, we will
777 : * be reading the rules in name order, except possibly during
778 : * emergency-recovery operations (ie, IgnoreSystemIndexes). This in turn
779 : * ensures that rules will be fired in name order.
780 : */
781 31634 : rewrite_desc = table_open(RewriteRelationId, AccessShareLock);
782 31634 : rewrite_tupdesc = RelationGetDescr(rewrite_desc);
783 31634 : rewrite_scan = systable_beginscan(rewrite_desc,
784 : RewriteRelRulenameIndexId,
785 : true, NULL,
786 : 1, &key);
787 :
788 61954 : while (HeapTupleIsValid(rewrite_tuple = systable_getnext(rewrite_scan)))
789 : {
790 30320 : Form_pg_rewrite rewrite_form = (Form_pg_rewrite) GETSTRUCT(rewrite_tuple);
791 : bool isnull;
792 : Datum rule_datum;
793 : char *rule_str;
794 : RewriteRule *rule;
795 : Oid check_as_user;
796 :
797 30320 : rule = (RewriteRule *) MemoryContextAlloc(rulescxt,
798 : sizeof(RewriteRule));
799 :
800 30320 : rule->ruleId = rewrite_form->oid;
801 :
802 30320 : rule->event = rewrite_form->ev_type - '0';
803 30320 : rule->enabled = rewrite_form->ev_enabled;
804 30320 : rule->isInstead = rewrite_form->is_instead;
805 :
806 : /*
807 : * Must use heap_getattr to fetch ev_action and ev_qual. Also, the
808 : * rule strings are often large enough to be toasted. To avoid
809 : * leaking memory in the caller's context, do the detoasting here so
810 : * we can free the detoasted version.
811 : */
812 30320 : rule_datum = heap_getattr(rewrite_tuple,
813 : Anum_pg_rewrite_ev_action,
814 : rewrite_tupdesc,
815 : &isnull);
816 : Assert(!isnull);
817 30320 : rule_str = TextDatumGetCString(rule_datum);
818 30320 : oldcxt = MemoryContextSwitchTo(rulescxt);
819 30320 : rule->actions = (List *) stringToNode(rule_str);
820 30320 : MemoryContextSwitchTo(oldcxt);
821 30320 : pfree(rule_str);
822 :
823 30320 : rule_datum = heap_getattr(rewrite_tuple,
824 : Anum_pg_rewrite_ev_qual,
825 : rewrite_tupdesc,
826 : &isnull);
827 : Assert(!isnull);
828 30320 : rule_str = TextDatumGetCString(rule_datum);
829 30320 : oldcxt = MemoryContextSwitchTo(rulescxt);
830 30320 : rule->qual = (Node *) stringToNode(rule_str);
831 30320 : MemoryContextSwitchTo(oldcxt);
832 30320 : pfree(rule_str);
833 :
834 : /*
835 : * If this is a SELECT rule defining a view, and the view has
836 : * "security_invoker" set, we must perform all permissions checks on
837 : * relations referred to by the rule as the invoking user.
838 : *
839 : * In all other cases (including non-SELECT rules on security invoker
840 : * views), perform the permissions checks as the relation owner.
841 : */
842 30320 : if (rule->event == CMD_SELECT &&
843 27940 : relation->rd_rel->relkind == RELKIND_VIEW &&
844 24620 : RelationHasSecurityInvoker(relation))
845 156 : check_as_user = InvalidOid;
846 : else
847 30164 : check_as_user = relation->rd_rel->relowner;
848 :
849 : /*
850 : * Scan through the rule's actions and set the checkAsUser field on
851 : * all RTEPermissionInfos. We have to look at the qual as well, in
852 : * case it contains sublinks.
853 : *
854 : * The reason for doing this when the rule is loaded, rather than when
855 : * it is stored, is that otherwise ALTER TABLE OWNER would have to
856 : * grovel through stored rules to update checkAsUser fields. Scanning
857 : * the rule tree during load is relatively cheap (compared to
858 : * constructing it in the first place), so we do it here.
859 : */
860 30320 : setRuleCheckAsUser((Node *) rule->actions, check_as_user);
861 30320 : setRuleCheckAsUser(rule->qual, check_as_user);
862 :
863 30320 : if (numlocks >= maxlocks)
864 : {
865 34 : maxlocks *= 2;
866 : rules = (RewriteRule **)
867 34 : repalloc(rules, sizeof(RewriteRule *) * maxlocks);
868 : }
869 30320 : rules[numlocks++] = rule;
870 : }
871 :
872 : /*
873 : * end the scan and close the attribute relation
874 : */
875 31634 : systable_endscan(rewrite_scan);
876 31634 : table_close(rewrite_desc, AccessShareLock);
877 :
878 : /*
879 : * there might not be any rules (if relhasrules is out-of-date)
880 : */
881 31634 : if (numlocks == 0)
882 : {
883 2770 : relation->rd_rules = NULL;
884 2770 : relation->rd_rulescxt = NULL;
885 2770 : MemoryContextDelete(rulescxt);
886 2770 : return;
887 : }
888 :
889 : /*
890 : * form a RuleLock and insert into relation
891 : */
892 28864 : rulelock = (RuleLock *) MemoryContextAlloc(rulescxt, sizeof(RuleLock));
893 28864 : rulelock->numLocks = numlocks;
894 28864 : rulelock->rules = rules;
895 :
896 28864 : relation->rd_rules = rulelock;
897 : }
898 :
899 : /*
900 : * equalRuleLocks
901 : *
902 : * Determine whether two RuleLocks are equivalent
903 : *
904 : * Probably this should be in the rules code someplace...
905 : */
906 : static bool
907 286104 : equalRuleLocks(RuleLock *rlock1, RuleLock *rlock2)
908 : {
909 : int i;
910 :
911 : /*
912 : * As of 7.3 we assume the rule ordering is repeatable, because
913 : * RelationBuildRuleLock should read 'em in a consistent order. So just
914 : * compare corresponding slots.
915 : */
916 286104 : if (rlock1 != NULL)
917 : {
918 1978 : if (rlock2 == NULL)
919 44 : return false;
920 1934 : if (rlock1->numLocks != rlock2->numLocks)
921 40 : return false;
922 3538 : for (i = 0; i < rlock1->numLocks; i++)
923 : {
924 1930 : RewriteRule *rule1 = rlock1->rules[i];
925 1930 : RewriteRule *rule2 = rlock2->rules[i];
926 :
927 1930 : if (rule1->ruleId != rule2->ruleId)
928 0 : return false;
929 1930 : if (rule1->event != rule2->event)
930 0 : return false;
931 1930 : if (rule1->enabled != rule2->enabled)
932 46 : return false;
933 1884 : if (rule1->isInstead != rule2->isInstead)
934 0 : return false;
935 1884 : if (!equal(rule1->qual, rule2->qual))
936 0 : return false;
937 1884 : if (!equal(rule1->actions, rule2->actions))
938 240 : return false;
939 : }
940 : }
941 284126 : else if (rlock2 != NULL)
942 13812 : return false;
943 271922 : return true;
944 : }
945 :
946 : /*
947 : * equalPolicy
948 : *
949 : * Determine whether two policies are equivalent
950 : */
951 : static bool
952 186 : equalPolicy(RowSecurityPolicy *policy1, RowSecurityPolicy *policy2)
953 : {
954 : int i;
955 : Oid *r1,
956 : *r2;
957 :
958 186 : if (policy1 != NULL)
959 : {
960 186 : if (policy2 == NULL)
961 0 : return false;
962 :
963 186 : if (policy1->polcmd != policy2->polcmd)
964 0 : return false;
965 186 : if (policy1->hassublinks != policy2->hassublinks)
966 0 : return false;
967 186 : if (strcmp(policy1->policy_name, policy2->policy_name) != 0)
968 0 : return false;
969 186 : if (ARR_DIMS(policy1->roles)[0] != ARR_DIMS(policy2->roles)[0])
970 0 : return false;
971 :
972 186 : r1 = (Oid *) ARR_DATA_PTR(policy1->roles);
973 186 : r2 = (Oid *) ARR_DATA_PTR(policy2->roles);
974 :
975 372 : for (i = 0; i < ARR_DIMS(policy1->roles)[0]; i++)
976 : {
977 186 : if (r1[i] != r2[i])
978 0 : return false;
979 : }
980 :
981 186 : if (!equal(policy1->qual, policy2->qual))
982 0 : return false;
983 186 : if (!equal(policy1->with_check_qual, policy2->with_check_qual))
984 0 : return false;
985 : }
986 0 : else if (policy2 != NULL)
987 0 : return false;
988 :
989 186 : return true;
990 : }
991 :
992 : /*
993 : * equalRSDesc
994 : *
995 : * Determine whether two RowSecurityDesc's are equivalent
996 : */
997 : static bool
998 286104 : equalRSDesc(RowSecurityDesc *rsdesc1, RowSecurityDesc *rsdesc2)
999 : {
1000 : ListCell *lc,
1001 : *rc;
1002 :
1003 286104 : if (rsdesc1 == NULL && rsdesc2 == NULL)
1004 285676 : return true;
1005 :
1006 428 : if ((rsdesc1 != NULL && rsdesc2 == NULL) ||
1007 278 : (rsdesc1 == NULL && rsdesc2 != NULL))
1008 288 : return false;
1009 :
1010 140 : if (list_length(rsdesc1->policies) != list_length(rsdesc2->policies))
1011 0 : return false;
1012 :
1013 : /* RelationBuildRowSecurity should build policies in order */
1014 326 : forboth(lc, rsdesc1->policies, rc, rsdesc2->policies)
1015 : {
1016 186 : RowSecurityPolicy *l = (RowSecurityPolicy *) lfirst(lc);
1017 186 : RowSecurityPolicy *r = (RowSecurityPolicy *) lfirst(rc);
1018 :
1019 186 : if (!equalPolicy(l, r))
1020 0 : return false;
1021 : }
1022 :
1023 140 : return true;
1024 : }
1025 :
1026 : /*
1027 : * RelationBuildDesc
1028 : *
1029 : * Build a relation descriptor. The caller must hold at least
1030 : * AccessShareLock on the target relid.
1031 : *
1032 : * The new descriptor is inserted into the hash table if insertIt is true.
1033 : *
1034 : * Returns NULL if no pg_class row could be found for the given relid
1035 : * (suggesting we are trying to access a just-deleted relation).
1036 : * Any other error is reported via elog.
1037 : */
1038 : static Relation
1039 1062060 : RelationBuildDesc(Oid targetRelId, bool insertIt)
1040 : {
1041 : int in_progress_offset;
1042 : Relation relation;
1043 : Oid relid;
1044 : HeapTuple pg_class_tuple;
1045 : Form_pg_class relp;
1046 :
1047 : /*
1048 : * This function and its subroutines can allocate a good deal of transient
1049 : * data in CurrentMemoryContext. Traditionally we've just leaked that
1050 : * data, reasoning that the caller's context is at worst of transaction
1051 : * scope, and relcache loads shouldn't happen so often that it's essential
1052 : * to recover transient data before end of statement/transaction. However
1053 : * that's definitely not true when debug_discard_caches is active, and
1054 : * perhaps it's not true in other cases.
1055 : *
1056 : * When debug_discard_caches is active or when forced to by
1057 : * RECOVER_RELATION_BUILD_MEMORY=1, arrange to allocate the junk in a
1058 : * temporary context that we'll free before returning. Make it a child of
1059 : * caller's context so that it will get cleaned up appropriately if we
1060 : * error out partway through.
1061 : */
1062 : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
1063 : MemoryContext tmpcxt = NULL;
1064 : MemoryContext oldcxt = NULL;
1065 :
1066 : if (RECOVER_RELATION_BUILD_MEMORY || debug_discard_caches > 0)
1067 : {
1068 : tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
1069 : "RelationBuildDesc workspace",
1070 : ALLOCSET_DEFAULT_SIZES);
1071 : oldcxt = MemoryContextSwitchTo(tmpcxt);
1072 : }
1073 : #endif
1074 :
1075 : /* Register to catch invalidation messages */
1076 1062060 : if (in_progress_list_len >= in_progress_list_maxlen)
1077 : {
1078 : int allocsize;
1079 :
1080 0 : allocsize = in_progress_list_maxlen * 2;
1081 0 : in_progress_list = repalloc(in_progress_list,
1082 : allocsize * sizeof(*in_progress_list));
1083 0 : in_progress_list_maxlen = allocsize;
1084 : }
1085 1062060 : in_progress_offset = in_progress_list_len++;
1086 1062060 : in_progress_list[in_progress_offset].reloid = targetRelId;
1087 1062078 : retry:
1088 1062078 : in_progress_list[in_progress_offset].invalidated = false;
1089 :
1090 : /*
1091 : * find the tuple in pg_class corresponding to the given relation id
1092 : */
1093 1062078 : pg_class_tuple = ScanPgRelation(targetRelId, true, false);
1094 :
1095 : /*
1096 : * if no such tuple exists, return NULL
1097 : */
1098 1062078 : if (!HeapTupleIsValid(pg_class_tuple))
1099 : {
1100 : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
1101 : if (tmpcxt)
1102 : {
1103 : /* Return to caller's context, and blow away the temporary context */
1104 : MemoryContextSwitchTo(oldcxt);
1105 : MemoryContextDelete(tmpcxt);
1106 : }
1107 : #endif
1108 : Assert(in_progress_offset + 1 == in_progress_list_len);
1109 10 : in_progress_list_len--;
1110 10 : return NULL;
1111 : }
1112 :
1113 : /*
1114 : * get information from the pg_class_tuple
1115 : */
1116 1062068 : relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
1117 1062068 : relid = relp->oid;
1118 : Assert(relid == targetRelId);
1119 :
1120 : /*
1121 : * allocate storage for the relation descriptor, and copy pg_class_tuple
1122 : * to relation->rd_rel.
1123 : */
1124 1062068 : relation = AllocateRelationDesc(relp);
1125 :
1126 : /*
1127 : * initialize the relation's relation id (relation->rd_id)
1128 : */
1129 1062068 : RelationGetRelid(relation) = relid;
1130 :
1131 : /*
1132 : * Normal relations are not nailed into the cache. Since we don't flush
1133 : * new relations, it won't be new. It could be temp though.
1134 : */
1135 1062068 : relation->rd_refcnt = 0;
1136 1062068 : relation->rd_isnailed = false;
1137 1062068 : relation->rd_createSubid = InvalidSubTransactionId;
1138 1062068 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
1139 1062068 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
1140 1062068 : relation->rd_droppedSubid = InvalidSubTransactionId;
1141 1062068 : switch (relation->rd_rel->relpersistence)
1142 : {
1143 1037106 : case RELPERSISTENCE_UNLOGGED:
1144 : case RELPERSISTENCE_PERMANENT:
1145 1037106 : relation->rd_backend = INVALID_PROC_NUMBER;
1146 1037106 : relation->rd_islocaltemp = false;
1147 1037106 : break;
1148 24962 : case RELPERSISTENCE_TEMP:
1149 24962 : if (isTempOrTempToastNamespace(relation->rd_rel->relnamespace))
1150 : {
1151 24928 : relation->rd_backend = ProcNumberForTempRelations();
1152 24928 : relation->rd_islocaltemp = true;
1153 : }
1154 : else
1155 : {
1156 : /*
1157 : * If it's a temp table, but not one of ours, we have to use
1158 : * the slow, grotty method to figure out the owning backend.
1159 : *
1160 : * Note: it's possible that rd_backend gets set to
1161 : * MyProcNumber here, in case we are looking at a pg_class
1162 : * entry left over from a crashed backend that coincidentally
1163 : * had the same ProcNumber we're using. We should *not*
1164 : * consider such a table to be "ours"; this is why we need the
1165 : * separate rd_islocaltemp flag. The pg_class entry will get
1166 : * flushed if/when we clean out the corresponding temp table
1167 : * namespace in preparation for using it.
1168 : */
1169 34 : relation->rd_backend =
1170 34 : GetTempNamespaceProcNumber(relation->rd_rel->relnamespace);
1171 : Assert(relation->rd_backend != INVALID_PROC_NUMBER);
1172 34 : relation->rd_islocaltemp = false;
1173 : }
1174 24962 : break;
1175 0 : default:
1176 0 : elog(ERROR, "invalid relpersistence: %c",
1177 : relation->rd_rel->relpersistence);
1178 : break;
1179 : }
1180 :
1181 : /*
1182 : * initialize the tuple descriptor (relation->rd_att).
1183 : */
1184 1062068 : RelationBuildTupleDesc(relation);
1185 :
1186 : /* foreign key data is not loaded till asked for */
1187 1062068 : relation->rd_fkeylist = NIL;
1188 1062068 : relation->rd_fkeyvalid = false;
1189 :
1190 : /* partitioning data is not loaded till asked for */
1191 1062068 : relation->rd_partkey = NULL;
1192 1062068 : relation->rd_partkeycxt = NULL;
1193 1062068 : relation->rd_partdesc = NULL;
1194 1062068 : relation->rd_partdesc_nodetached = NULL;
1195 1062068 : relation->rd_partdesc_nodetached_xmin = InvalidTransactionId;
1196 1062068 : relation->rd_pdcxt = NULL;
1197 1062068 : relation->rd_pddcxt = NULL;
1198 1062068 : relation->rd_partcheck = NIL;
1199 1062068 : relation->rd_partcheckvalid = false;
1200 1062068 : relation->rd_partcheckcxt = NULL;
1201 :
1202 : /*
1203 : * initialize access method information
1204 : */
1205 1062068 : if (relation->rd_rel->relkind == RELKIND_INDEX ||
1206 660122 : relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
1207 408084 : RelationInitIndexAccessInfo(relation);
1208 653984 : else if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) ||
1209 103972 : relation->rd_rel->relkind == RELKIND_SEQUENCE)
1210 555280 : RelationInitTableAccessMethod(relation);
1211 : else if (relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
1212 : {
1213 : /*
1214 : * Do nothing: access methods are a setting that partitions can
1215 : * inherit.
1216 : */
1217 : }
1218 : else
1219 : Assert(relation->rd_rel->relam == InvalidOid);
1220 :
1221 : /* extract reloptions if any */
1222 1062060 : RelationParseRelOptions(relation, pg_class_tuple);
1223 :
1224 : /*
1225 : * Fetch rules and triggers that affect this relation.
1226 : *
1227 : * Note that RelationBuildRuleLock() relies on this being done after
1228 : * extracting the relation's reloptions.
1229 : */
1230 1062060 : if (relation->rd_rel->relhasrules)
1231 31634 : RelationBuildRuleLock(relation);
1232 : else
1233 : {
1234 1030426 : relation->rd_rules = NULL;
1235 1030426 : relation->rd_rulescxt = NULL;
1236 : }
1237 :
1238 1062060 : if (relation->rd_rel->relhastriggers)
1239 59200 : RelationBuildTriggers(relation);
1240 : else
1241 1002860 : relation->trigdesc = NULL;
1242 :
1243 1062060 : if (relation->rd_rel->relrowsecurity)
1244 1978 : RelationBuildRowSecurity(relation);
1245 : else
1246 1060082 : relation->rd_rsdesc = NULL;
1247 :
1248 : /*
1249 : * initialize the relation lock manager information
1250 : */
1251 1062060 : RelationInitLockInfo(relation); /* see lmgr.c */
1252 :
1253 : /*
1254 : * initialize physical addressing information for the relation
1255 : */
1256 1062060 : RelationInitPhysicalAddr(relation);
1257 :
1258 : /* make sure relation is marked as having no open file yet */
1259 1062060 : relation->rd_smgr = NULL;
1260 :
1261 : /*
1262 : * now we can free the memory allocated for pg_class_tuple
1263 : */
1264 1062060 : heap_freetuple(pg_class_tuple);
1265 :
1266 : /*
1267 : * If an invalidation arrived mid-build, start over. Between here and the
1268 : * end of this function, don't add code that does or reasonably could read
1269 : * system catalogs. That range must be free from invalidation processing
1270 : * for the !insertIt case. For the insertIt case, RelationCacheInsert()
1271 : * will enroll this relation in ordinary relcache invalidation processing,
1272 : */
1273 1062060 : if (in_progress_list[in_progress_offset].invalidated)
1274 : {
1275 18 : RelationDestroyRelation(relation, false);
1276 18 : goto retry;
1277 : }
1278 : Assert(in_progress_offset + 1 == in_progress_list_len);
1279 1062042 : in_progress_list_len--;
1280 :
1281 : /*
1282 : * Insert newly created relation into relcache hash table, if requested.
1283 : *
1284 : * There is one scenario in which we might find a hashtable entry already
1285 : * present, even though our caller failed to find it: if the relation is a
1286 : * system catalog or index that's used during relcache load, we might have
1287 : * recursively created the same relcache entry during the preceding steps.
1288 : * So allow RelationCacheInsert to delete any already-present relcache
1289 : * entry for the same OID. The already-present entry should have refcount
1290 : * zero (else somebody forgot to close it); in the event that it doesn't,
1291 : * we'll elog a WARNING and leak the already-present entry.
1292 : */
1293 1062042 : if (insertIt)
1294 775938 : RelationCacheInsert(relation, true);
1295 :
1296 : /* It's fully valid */
1297 1062042 : relation->rd_isvalid = true;
1298 :
1299 : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
1300 : if (tmpcxt)
1301 : {
1302 : /* Return to caller's context, and blow away the temporary context */
1303 : MemoryContextSwitchTo(oldcxt);
1304 : MemoryContextDelete(tmpcxt);
1305 : }
1306 : #endif
1307 :
1308 1062042 : return relation;
1309 : }
1310 :
1311 : /*
1312 : * Initialize the physical addressing info (RelFileLocator) for a relcache entry
1313 : *
1314 : * Note: at the physical level, relations in the pg_global tablespace must
1315 : * be treated as shared, even if relisshared isn't set. Hence we do not
1316 : * look at relisshared here.
1317 : */
1318 : static void
1319 4465090 : RelationInitPhysicalAddr(Relation relation)
1320 : {
1321 4465090 : RelFileNumber oldnumber = relation->rd_locator.relNumber;
1322 :
1323 : /* these relations kinds never have storage */
1324 4465090 : if (!RELKIND_HAS_STORAGE(relation->rd_rel->relkind))
1325 129672 : return;
1326 :
1327 4335418 : if (relation->rd_rel->reltablespace)
1328 736582 : relation->rd_locator.spcOid = relation->rd_rel->reltablespace;
1329 : else
1330 3598836 : relation->rd_locator.spcOid = MyDatabaseTableSpace;
1331 4335418 : if (relation->rd_locator.spcOid == GLOBALTABLESPACE_OID)
1332 733628 : relation->rd_locator.dbOid = InvalidOid;
1333 : else
1334 3601790 : relation->rd_locator.dbOid = MyDatabaseId;
1335 :
1336 4335418 : if (relation->rd_rel->relfilenode)
1337 : {
1338 : /*
1339 : * Even if we are using a decoding snapshot that doesn't represent the
1340 : * current state of the catalog we need to make sure the filenode
1341 : * points to the current file since the older file will be gone (or
1342 : * truncated). The new file will still contain older rows so lookups
1343 : * in them will work correctly. This wouldn't work correctly if
1344 : * rewrites were allowed to change the schema in an incompatible way,
1345 : * but those are prevented both on catalog tables and on user tables
1346 : * declared as additional catalog tables.
1347 : */
1348 3126576 : if (HistoricSnapshotActive()
1349 3942 : && RelationIsAccessibleInLogicalDecoding(relation)
1350 2628 : && IsTransactionState())
1351 : {
1352 : HeapTuple phys_tuple;
1353 : Form_pg_class physrel;
1354 :
1355 2628 : phys_tuple = ScanPgRelation(RelationGetRelid(relation),
1356 2628 : RelationGetRelid(relation) != ClassOidIndexId,
1357 : true);
1358 2628 : if (!HeapTupleIsValid(phys_tuple))
1359 0 : elog(ERROR, "could not find pg_class entry for %u",
1360 : RelationGetRelid(relation));
1361 2628 : physrel = (Form_pg_class) GETSTRUCT(phys_tuple);
1362 :
1363 2628 : relation->rd_rel->reltablespace = physrel->reltablespace;
1364 2628 : relation->rd_rel->relfilenode = physrel->relfilenode;
1365 2628 : heap_freetuple(phys_tuple);
1366 : }
1367 :
1368 3126576 : relation->rd_locator.relNumber = relation->rd_rel->relfilenode;
1369 : }
1370 : else
1371 : {
1372 : /* Consult the relation mapper */
1373 1208842 : relation->rd_locator.relNumber =
1374 1208842 : RelationMapOidToFilenumber(relation->rd_id,
1375 1208842 : relation->rd_rel->relisshared);
1376 1208842 : if (!RelFileNumberIsValid(relation->rd_locator.relNumber))
1377 0 : elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
1378 : RelationGetRelationName(relation), relation->rd_id);
1379 : }
1380 :
1381 : /*
1382 : * For RelationNeedsWAL() to answer correctly on parallel workers, restore
1383 : * rd_firstRelfilelocatorSubid. No subtransactions start or end while in
1384 : * parallel mode, so the specific SubTransactionId does not matter.
1385 : */
1386 4335418 : if (IsParallelWorker() && oldnumber != relation->rd_locator.relNumber)
1387 : {
1388 38098 : if (RelFileLocatorSkippingWAL(relation->rd_locator))
1389 300 : relation->rd_firstRelfilelocatorSubid = TopSubTransactionId;
1390 : else
1391 37798 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
1392 : }
1393 : }
1394 :
1395 : /*
1396 : * Fill in the IndexAmRoutine for an index relation.
1397 : *
1398 : * relation's rd_amhandler and rd_indexcxt must be valid already.
1399 : */
1400 : static void
1401 2235534 : InitIndexAmRoutine(Relation relation)
1402 : {
1403 : IndexAmRoutine *cached,
1404 : *tmp;
1405 :
1406 : /*
1407 : * Call the amhandler in current, short-lived memory context, just in case
1408 : * it leaks anything (it probably won't, but let's be paranoid).
1409 : */
1410 2235534 : tmp = GetIndexAmRoutine(relation->rd_amhandler);
1411 :
1412 : /* OK, now transfer the data into relation's rd_indexcxt. */
1413 2235534 : cached = (IndexAmRoutine *) MemoryContextAlloc(relation->rd_indexcxt,
1414 : sizeof(IndexAmRoutine));
1415 2235534 : memcpy(cached, tmp, sizeof(IndexAmRoutine));
1416 2235534 : relation->rd_indam = cached;
1417 :
1418 2235534 : pfree(tmp);
1419 2235534 : }
1420 :
1421 : /*
1422 : * Initialize index-access-method support data for an index relation
1423 : */
1424 : void
1425 420884 : RelationInitIndexAccessInfo(Relation relation)
1426 : {
1427 : HeapTuple tuple;
1428 : Form_pg_am aform;
1429 : Datum indcollDatum;
1430 : Datum indclassDatum;
1431 : Datum indoptionDatum;
1432 : bool isnull;
1433 : oidvector *indcoll;
1434 : oidvector *indclass;
1435 : int2vector *indoption;
1436 : MemoryContext indexcxt;
1437 : MemoryContext oldcontext;
1438 : int indnatts;
1439 : int indnkeyatts;
1440 : uint16 amsupport;
1441 :
1442 : /*
1443 : * Make a copy of the pg_index entry for the index. Since pg_index
1444 : * contains variable-length and possibly-null fields, we have to do this
1445 : * honestly rather than just treating it as a Form_pg_index struct.
1446 : */
1447 420884 : tuple = SearchSysCache1(INDEXRELID,
1448 : ObjectIdGetDatum(RelationGetRelid(relation)));
1449 420884 : if (!HeapTupleIsValid(tuple))
1450 0 : elog(ERROR, "cache lookup failed for index %u",
1451 : RelationGetRelid(relation));
1452 420884 : oldcontext = MemoryContextSwitchTo(CacheMemoryContext);
1453 420884 : relation->rd_indextuple = heap_copytuple(tuple);
1454 420884 : relation->rd_index = (Form_pg_index) GETSTRUCT(relation->rd_indextuple);
1455 420884 : MemoryContextSwitchTo(oldcontext);
1456 420884 : ReleaseSysCache(tuple);
1457 :
1458 : /*
1459 : * Look up the index's access method, save the OID of its handler function
1460 : */
1461 : Assert(relation->rd_rel->relam != InvalidOid);
1462 420884 : tuple = SearchSysCache1(AMOID, ObjectIdGetDatum(relation->rd_rel->relam));
1463 420882 : if (!HeapTupleIsValid(tuple))
1464 0 : elog(ERROR, "cache lookup failed for access method %u",
1465 : relation->rd_rel->relam);
1466 420882 : aform = (Form_pg_am) GETSTRUCT(tuple);
1467 420882 : relation->rd_amhandler = aform->amhandler;
1468 420882 : ReleaseSysCache(tuple);
1469 :
1470 420882 : indnatts = RelationGetNumberOfAttributes(relation);
1471 420882 : if (indnatts != IndexRelationGetNumberOfAttributes(relation))
1472 0 : elog(ERROR, "relnatts disagrees with indnatts for index %u",
1473 : RelationGetRelid(relation));
1474 420882 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(relation);
1475 :
1476 : /*
1477 : * Make the private context to hold index access info. The reason we need
1478 : * a context, and not just a couple of pallocs, is so that we won't leak
1479 : * any subsidiary info attached to fmgr lookup records.
1480 : */
1481 420882 : indexcxt = AllocSetContextCreate(CacheMemoryContext,
1482 : "index info",
1483 : ALLOCSET_SMALL_SIZES);
1484 420882 : relation->rd_indexcxt = indexcxt;
1485 420882 : MemoryContextCopyAndSetIdentifier(indexcxt,
1486 : RelationGetRelationName(relation));
1487 :
1488 : /*
1489 : * Now we can fetch the index AM's API struct
1490 : */
1491 420882 : InitIndexAmRoutine(relation);
1492 :
1493 : /*
1494 : * Allocate arrays to hold data. Opclasses are not used for included
1495 : * columns, so allocate them for indnkeyatts only.
1496 : */
1497 420882 : relation->rd_opfamily = (Oid *)
1498 420882 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
1499 420882 : relation->rd_opcintype = (Oid *)
1500 420882 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
1501 :
1502 420882 : amsupport = relation->rd_indam->amsupport;
1503 420882 : if (amsupport > 0)
1504 : {
1505 420882 : int nsupport = indnatts * amsupport;
1506 :
1507 420882 : relation->rd_support = (RegProcedure *)
1508 420882 : MemoryContextAllocZero(indexcxt, nsupport * sizeof(RegProcedure));
1509 420882 : relation->rd_supportinfo = (FmgrInfo *)
1510 420882 : MemoryContextAllocZero(indexcxt, nsupport * sizeof(FmgrInfo));
1511 : }
1512 : else
1513 : {
1514 0 : relation->rd_support = NULL;
1515 0 : relation->rd_supportinfo = NULL;
1516 : }
1517 :
1518 420882 : relation->rd_indcollation = (Oid *)
1519 420882 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
1520 :
1521 420882 : relation->rd_indoption = (int16 *)
1522 420882 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(int16));
1523 :
1524 : /*
1525 : * indcollation cannot be referenced directly through the C struct,
1526 : * because it comes after the variable-width indkey field. Must extract
1527 : * the datum the hard way...
1528 : */
1529 420882 : indcollDatum = fastgetattr(relation->rd_indextuple,
1530 : Anum_pg_index_indcollation,
1531 : GetPgIndexDescriptor(),
1532 : &isnull);
1533 : Assert(!isnull);
1534 420882 : indcoll = (oidvector *) DatumGetPointer(indcollDatum);
1535 420882 : memcpy(relation->rd_indcollation, indcoll->values, indnkeyatts * sizeof(Oid));
1536 :
1537 : /*
1538 : * indclass cannot be referenced directly through the C struct, because it
1539 : * comes after the variable-width indkey field. Must extract the datum
1540 : * the hard way...
1541 : */
1542 420882 : indclassDatum = fastgetattr(relation->rd_indextuple,
1543 : Anum_pg_index_indclass,
1544 : GetPgIndexDescriptor(),
1545 : &isnull);
1546 : Assert(!isnull);
1547 420882 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1548 :
1549 : /*
1550 : * Fill the support procedure OID array, as well as the info about
1551 : * opfamilies and opclass input types. (aminfo and supportinfo are left
1552 : * as zeroes, and are filled on-the-fly when used)
1553 : */
1554 420882 : IndexSupportInitialize(indclass, relation->rd_support,
1555 : relation->rd_opfamily, relation->rd_opcintype,
1556 : amsupport, indnkeyatts);
1557 :
1558 : /*
1559 : * Similarly extract indoption and copy it to the cache entry
1560 : */
1561 420882 : indoptionDatum = fastgetattr(relation->rd_indextuple,
1562 : Anum_pg_index_indoption,
1563 : GetPgIndexDescriptor(),
1564 : &isnull);
1565 : Assert(!isnull);
1566 420882 : indoption = (int2vector *) DatumGetPointer(indoptionDatum);
1567 420882 : memcpy(relation->rd_indoption, indoption->values, indnkeyatts * sizeof(int16));
1568 :
1569 420882 : (void) RelationGetIndexAttOptions(relation, false);
1570 :
1571 : /*
1572 : * expressions, predicate, exclusion caches will be filled later
1573 : */
1574 420876 : relation->rd_indexprs = NIL;
1575 420876 : relation->rd_indpred = NIL;
1576 420876 : relation->rd_exclops = NULL;
1577 420876 : relation->rd_exclprocs = NULL;
1578 420876 : relation->rd_exclstrats = NULL;
1579 420876 : relation->rd_amcache = NULL;
1580 420876 : }
1581 :
1582 : /*
1583 : * IndexSupportInitialize
1584 : * Initializes an index's cached opclass information,
1585 : * given the index's pg_index.indclass entry.
1586 : *
1587 : * Data is returned into *indexSupport, *opFamily, and *opcInType,
1588 : * which are arrays allocated by the caller.
1589 : *
1590 : * The caller also passes maxSupportNumber and maxAttributeNumber, since these
1591 : * indicate the size of the arrays it has allocated --- but in practice these
1592 : * numbers must always match those obtainable from the system catalog entries
1593 : * for the index and access method.
1594 : */
1595 : static void
1596 420882 : IndexSupportInitialize(oidvector *indclass,
1597 : RegProcedure *indexSupport,
1598 : Oid *opFamily,
1599 : Oid *opcInType,
1600 : StrategyNumber maxSupportNumber,
1601 : AttrNumber maxAttributeNumber)
1602 : {
1603 : int attIndex;
1604 :
1605 1129596 : for (attIndex = 0; attIndex < maxAttributeNumber; attIndex++)
1606 : {
1607 : OpClassCacheEnt *opcentry;
1608 :
1609 708714 : if (!OidIsValid(indclass->values[attIndex]))
1610 0 : elog(ERROR, "bogus pg_index tuple");
1611 :
1612 : /* look up the info for this opclass, using a cache */
1613 708714 : opcentry = LookupOpclassInfo(indclass->values[attIndex],
1614 : maxSupportNumber);
1615 :
1616 : /* copy cached data into relcache entry */
1617 708714 : opFamily[attIndex] = opcentry->opcfamily;
1618 708714 : opcInType[attIndex] = opcentry->opcintype;
1619 708714 : if (maxSupportNumber > 0)
1620 708714 : memcpy(&indexSupport[attIndex * maxSupportNumber],
1621 708714 : opcentry->supportProcs,
1622 : maxSupportNumber * sizeof(RegProcedure));
1623 : }
1624 420882 : }
1625 :
1626 : /*
1627 : * LookupOpclassInfo
1628 : *
1629 : * This routine maintains a per-opclass cache of the information needed
1630 : * by IndexSupportInitialize(). This is more efficient than relying on
1631 : * the catalog cache, because we can load all the info about a particular
1632 : * opclass in a single indexscan of pg_amproc.
1633 : *
1634 : * The information from pg_am about expected range of support function
1635 : * numbers is passed in, rather than being looked up, mainly because the
1636 : * caller will have it already.
1637 : *
1638 : * Note there is no provision for flushing the cache. This is OK at the
1639 : * moment because there is no way to ALTER any interesting properties of an
1640 : * existing opclass --- all you can do is drop it, which will result in
1641 : * a useless but harmless dead entry in the cache. To support altering
1642 : * opclass membership (not the same as opfamily membership!), we'd need to
1643 : * be able to flush this cache as well as the contents of relcache entries
1644 : * for indexes.
1645 : */
1646 : static OpClassCacheEnt *
1647 708714 : LookupOpclassInfo(Oid operatorClassOid,
1648 : StrategyNumber numSupport)
1649 : {
1650 : OpClassCacheEnt *opcentry;
1651 : bool found;
1652 : Relation rel;
1653 : SysScanDesc scan;
1654 : ScanKeyData skey[3];
1655 : HeapTuple htup;
1656 : bool indexOK;
1657 :
1658 708714 : if (OpClassCache == NULL)
1659 : {
1660 : /* First time through: initialize the opclass cache */
1661 : HASHCTL ctl;
1662 :
1663 : /* Also make sure CacheMemoryContext exists */
1664 23586 : if (!CacheMemoryContext)
1665 0 : CreateCacheMemoryContext();
1666 :
1667 23586 : ctl.keysize = sizeof(Oid);
1668 23586 : ctl.entrysize = sizeof(OpClassCacheEnt);
1669 23586 : OpClassCache = hash_create("Operator class cache", 64,
1670 : &ctl, HASH_ELEM | HASH_BLOBS);
1671 : }
1672 :
1673 708714 : opcentry = (OpClassCacheEnt *) hash_search(OpClassCache,
1674 : &operatorClassOid,
1675 : HASH_ENTER, &found);
1676 :
1677 708714 : if (!found)
1678 : {
1679 : /* Initialize new entry */
1680 61838 : opcentry->valid = false; /* until known OK */
1681 61838 : opcentry->numSupport = numSupport;
1682 61838 : opcentry->supportProcs = NULL; /* filled below */
1683 : }
1684 : else
1685 : {
1686 : Assert(numSupport == opcentry->numSupport);
1687 : }
1688 :
1689 : /*
1690 : * When aggressively testing cache-flush hazards, we disable the operator
1691 : * class cache and force reloading of the info on each call. This models
1692 : * no real-world behavior, since the cache entries are never invalidated
1693 : * otherwise. However it can be helpful for detecting bugs in the cache
1694 : * loading logic itself, such as reliance on a non-nailed index. Given
1695 : * the limited use-case and the fact that this adds a great deal of
1696 : * expense, we enable it only for high values of debug_discard_caches.
1697 : */
1698 : #ifdef DISCARD_CACHES_ENABLED
1699 : if (debug_discard_caches > 2)
1700 : opcentry->valid = false;
1701 : #endif
1702 :
1703 708714 : if (opcentry->valid)
1704 646876 : return opcentry;
1705 :
1706 : /*
1707 : * Need to fill in new entry. First allocate space, unless we already did
1708 : * so in some previous attempt.
1709 : */
1710 61838 : if (opcentry->supportProcs == NULL && numSupport > 0)
1711 61838 : opcentry->supportProcs = (RegProcedure *)
1712 61838 : MemoryContextAllocZero(CacheMemoryContext,
1713 : numSupport * sizeof(RegProcedure));
1714 :
1715 : /*
1716 : * To avoid infinite recursion during startup, force heap scans if we're
1717 : * looking up info for the opclasses used by the indexes we would like to
1718 : * reference here.
1719 : */
1720 68882 : indexOK = criticalRelcachesBuilt ||
1721 7044 : (operatorClassOid != OID_BTREE_OPS_OID &&
1722 4856 : operatorClassOid != INT2_BTREE_OPS_OID);
1723 :
1724 : /*
1725 : * We have to fetch the pg_opclass row to determine its opfamily and
1726 : * opcintype, which are needed to look up related operators and functions.
1727 : * It'd be convenient to use the syscache here, but that probably doesn't
1728 : * work while bootstrapping.
1729 : */
1730 61838 : ScanKeyInit(&skey[0],
1731 : Anum_pg_opclass_oid,
1732 : BTEqualStrategyNumber, F_OIDEQ,
1733 : ObjectIdGetDatum(operatorClassOid));
1734 61838 : rel = table_open(OperatorClassRelationId, AccessShareLock);
1735 61838 : scan = systable_beginscan(rel, OpclassOidIndexId, indexOK,
1736 : NULL, 1, skey);
1737 :
1738 61838 : if (HeapTupleIsValid(htup = systable_getnext(scan)))
1739 : {
1740 61838 : Form_pg_opclass opclassform = (Form_pg_opclass) GETSTRUCT(htup);
1741 :
1742 61838 : opcentry->opcfamily = opclassform->opcfamily;
1743 61838 : opcentry->opcintype = opclassform->opcintype;
1744 : }
1745 : else
1746 0 : elog(ERROR, "could not find tuple for opclass %u", operatorClassOid);
1747 :
1748 61838 : systable_endscan(scan);
1749 61838 : table_close(rel, AccessShareLock);
1750 :
1751 : /*
1752 : * Scan pg_amproc to obtain support procs for the opclass. We only fetch
1753 : * the default ones (those with lefttype = righttype = opcintype).
1754 : */
1755 61838 : if (numSupport > 0)
1756 : {
1757 61838 : ScanKeyInit(&skey[0],
1758 : Anum_pg_amproc_amprocfamily,
1759 : BTEqualStrategyNumber, F_OIDEQ,
1760 : ObjectIdGetDatum(opcentry->opcfamily));
1761 61838 : ScanKeyInit(&skey[1],
1762 : Anum_pg_amproc_amproclefttype,
1763 : BTEqualStrategyNumber, F_OIDEQ,
1764 : ObjectIdGetDatum(opcentry->opcintype));
1765 61838 : ScanKeyInit(&skey[2],
1766 : Anum_pg_amproc_amprocrighttype,
1767 : BTEqualStrategyNumber, F_OIDEQ,
1768 : ObjectIdGetDatum(opcentry->opcintype));
1769 61838 : rel = table_open(AccessMethodProcedureRelationId, AccessShareLock);
1770 61838 : scan = systable_beginscan(rel, AccessMethodProcedureIndexId, indexOK,
1771 : NULL, 3, skey);
1772 :
1773 257192 : while (HeapTupleIsValid(htup = systable_getnext(scan)))
1774 : {
1775 195354 : Form_pg_amproc amprocform = (Form_pg_amproc) GETSTRUCT(htup);
1776 :
1777 195354 : if (amprocform->amprocnum <= 0 ||
1778 195354 : (StrategyNumber) amprocform->amprocnum > numSupport)
1779 0 : elog(ERROR, "invalid amproc number %d for opclass %u",
1780 : amprocform->amprocnum, operatorClassOid);
1781 :
1782 195354 : opcentry->supportProcs[amprocform->amprocnum - 1] =
1783 195354 : amprocform->amproc;
1784 : }
1785 :
1786 61838 : systable_endscan(scan);
1787 61838 : table_close(rel, AccessShareLock);
1788 : }
1789 :
1790 61838 : opcentry->valid = true;
1791 61838 : return opcentry;
1792 : }
1793 :
1794 : /*
1795 : * Fill in the TableAmRoutine for a relation
1796 : *
1797 : * relation's rd_amhandler must be valid already.
1798 : */
1799 : static void
1800 1691318 : InitTableAmRoutine(Relation relation)
1801 : {
1802 1691318 : relation->rd_tableam = GetTableAmRoutine(relation->rd_amhandler);
1803 1691318 : }
1804 :
1805 : /*
1806 : * Initialize table access method support for a table like relation
1807 : */
1808 : void
1809 1691318 : RelationInitTableAccessMethod(Relation relation)
1810 : {
1811 : HeapTuple tuple;
1812 : Form_pg_am aform;
1813 :
1814 1691318 : if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
1815 : {
1816 : /*
1817 : * Sequences are currently accessed like heap tables, but it doesn't
1818 : * seem prudent to show that in the catalog. So just overwrite it
1819 : * here.
1820 : */
1821 : Assert(relation->rd_rel->relam == InvalidOid);
1822 7002 : relation->rd_amhandler = F_HEAP_TABLEAM_HANDLER;
1823 : }
1824 1684316 : else if (IsCatalogRelation(relation))
1825 : {
1826 : /*
1827 : * Avoid doing a syscache lookup for catalog tables.
1828 : */
1829 : Assert(relation->rd_rel->relam == HEAP_TABLE_AM_OID);
1830 1300708 : relation->rd_amhandler = F_HEAP_TABLEAM_HANDLER;
1831 : }
1832 : else
1833 : {
1834 : /*
1835 : * Look up the table access method, save the OID of its handler
1836 : * function.
1837 : */
1838 : Assert(relation->rd_rel->relam != InvalidOid);
1839 383608 : tuple = SearchSysCache1(AMOID,
1840 383608 : ObjectIdGetDatum(relation->rd_rel->relam));
1841 383608 : if (!HeapTupleIsValid(tuple))
1842 0 : elog(ERROR, "cache lookup failed for access method %u",
1843 : relation->rd_rel->relam);
1844 383608 : aform = (Form_pg_am) GETSTRUCT(tuple);
1845 383608 : relation->rd_amhandler = aform->amhandler;
1846 383608 : ReleaseSysCache(tuple);
1847 : }
1848 :
1849 : /*
1850 : * Now we can fetch the table AM's API struct
1851 : */
1852 1691318 : InitTableAmRoutine(relation);
1853 1691318 : }
1854 :
1855 : /*
1856 : * formrdesc
1857 : *
1858 : * This is a special cut-down version of RelationBuildDesc(),
1859 : * used while initializing the relcache.
1860 : * The relation descriptor is built just from the supplied parameters,
1861 : * without actually looking at any system table entries. We cheat
1862 : * quite a lot since we only need to work for a few basic system
1863 : * catalogs.
1864 : *
1865 : * The catalogs this is used for can't have constraints (except attnotnull),
1866 : * default values, rules, or triggers, since we don't cope with any of that.
1867 : * (Well, actually, this only matters for properties that need to be valid
1868 : * during bootstrap or before RelationCacheInitializePhase3 runs, and none of
1869 : * these properties matter then...)
1870 : *
1871 : * NOTE: we assume we are already switched into CacheMemoryContext.
1872 : */
1873 : static void
1874 24752 : formrdesc(const char *relationName, Oid relationReltype,
1875 : bool isshared,
1876 : int natts, const FormData_pg_attribute *attrs)
1877 : {
1878 : Relation relation;
1879 : int i;
1880 : bool has_not_null;
1881 :
1882 : /*
1883 : * allocate new relation desc, clear all fields of reldesc
1884 : */
1885 24752 : relation = (Relation) palloc0(sizeof(RelationData));
1886 :
1887 : /* make sure relation is marked as having no open file yet */
1888 24752 : relation->rd_smgr = NULL;
1889 :
1890 : /*
1891 : * initialize reference count: 1 because it is nailed in cache
1892 : */
1893 24752 : relation->rd_refcnt = 1;
1894 :
1895 : /*
1896 : * all entries built with this routine are nailed-in-cache; none are for
1897 : * new or temp relations.
1898 : */
1899 24752 : relation->rd_isnailed = true;
1900 24752 : relation->rd_createSubid = InvalidSubTransactionId;
1901 24752 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
1902 24752 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
1903 24752 : relation->rd_droppedSubid = InvalidSubTransactionId;
1904 24752 : relation->rd_backend = INVALID_PROC_NUMBER;
1905 24752 : relation->rd_islocaltemp = false;
1906 :
1907 : /*
1908 : * initialize relation tuple form
1909 : *
1910 : * The data we insert here is pretty incomplete/bogus, but it'll serve to
1911 : * get us launched. RelationCacheInitializePhase3() will read the real
1912 : * data from pg_class and replace what we've done here. Note in
1913 : * particular that relowner is left as zero; this cues
1914 : * RelationCacheInitializePhase3 that the real data isn't there yet.
1915 : */
1916 24752 : relation->rd_rel = (Form_pg_class) palloc0(CLASS_TUPLE_SIZE);
1917 :
1918 24752 : namestrcpy(&relation->rd_rel->relname, relationName);
1919 24752 : relation->rd_rel->relnamespace = PG_CATALOG_NAMESPACE;
1920 24752 : relation->rd_rel->reltype = relationReltype;
1921 :
1922 : /*
1923 : * It's important to distinguish between shared and non-shared relations,
1924 : * even at bootstrap time, to make sure we know where they are stored.
1925 : */
1926 24752 : relation->rd_rel->relisshared = isshared;
1927 24752 : if (isshared)
1928 16000 : relation->rd_rel->reltablespace = GLOBALTABLESPACE_OID;
1929 :
1930 : /* formrdesc is used only for permanent relations */
1931 24752 : relation->rd_rel->relpersistence = RELPERSISTENCE_PERMANENT;
1932 :
1933 : /* ... and they're always populated, too */
1934 24752 : relation->rd_rel->relispopulated = true;
1935 :
1936 24752 : relation->rd_rel->relreplident = REPLICA_IDENTITY_NOTHING;
1937 24752 : relation->rd_rel->relpages = 0;
1938 24752 : relation->rd_rel->reltuples = -1;
1939 24752 : relation->rd_rel->relallvisible = 0;
1940 24752 : relation->rd_rel->relkind = RELKIND_RELATION;
1941 24752 : relation->rd_rel->relnatts = (int16) natts;
1942 24752 : relation->rd_rel->relam = HEAP_TABLE_AM_OID;
1943 :
1944 : /*
1945 : * initialize attribute tuple form
1946 : *
1947 : * Unlike the case with the relation tuple, this data had better be right
1948 : * because it will never be replaced. The data comes from
1949 : * src/include/catalog/ headers via genbki.pl.
1950 : */
1951 24752 : relation->rd_att = CreateTemplateTupleDesc(natts);
1952 24752 : relation->rd_att->tdrefcount = 1; /* mark as refcounted */
1953 :
1954 24752 : relation->rd_att->tdtypeid = relationReltype;
1955 24752 : relation->rd_att->tdtypmod = -1; /* just to be sure */
1956 :
1957 : /*
1958 : * initialize tuple desc info
1959 : */
1960 24752 : has_not_null = false;
1961 478300 : for (i = 0; i < natts; i++)
1962 : {
1963 453548 : memcpy(TupleDescAttr(relation->rd_att, i),
1964 453548 : &attrs[i],
1965 : ATTRIBUTE_FIXED_PART_SIZE);
1966 453548 : has_not_null |= attrs[i].attnotnull;
1967 : /* make sure attcacheoff is valid */
1968 453548 : TupleDescAttr(relation->rd_att, i)->attcacheoff = -1;
1969 : }
1970 :
1971 : /* initialize first attribute's attcacheoff, cf RelationBuildTupleDesc */
1972 24752 : TupleDescAttr(relation->rd_att, 0)->attcacheoff = 0;
1973 :
1974 : /* mark not-null status */
1975 24752 : if (has_not_null)
1976 : {
1977 24752 : TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
1978 :
1979 24752 : constr->has_not_null = true;
1980 24752 : relation->rd_att->constr = constr;
1981 : }
1982 :
1983 : /*
1984 : * initialize relation id from info in att array (my, this is ugly)
1985 : */
1986 24752 : RelationGetRelid(relation) = TupleDescAttr(relation->rd_att, 0)->attrelid;
1987 :
1988 : /*
1989 : * All relations made with formrdesc are mapped. This is necessarily so
1990 : * because there is no other way to know what filenumber they currently
1991 : * have. In bootstrap mode, add them to the initial relation mapper data,
1992 : * specifying that the initial filenumber is the same as the OID.
1993 : */
1994 24752 : relation->rd_rel->relfilenode = InvalidRelFileNumber;
1995 24752 : if (IsBootstrapProcessingMode())
1996 320 : RelationMapUpdateMap(RelationGetRelid(relation),
1997 : RelationGetRelid(relation),
1998 : isshared, true);
1999 :
2000 : /*
2001 : * initialize the relation lock manager information
2002 : */
2003 24752 : RelationInitLockInfo(relation); /* see lmgr.c */
2004 :
2005 : /*
2006 : * initialize physical addressing information for the relation
2007 : */
2008 24752 : RelationInitPhysicalAddr(relation);
2009 :
2010 : /*
2011 : * initialize the table am handler
2012 : */
2013 24752 : relation->rd_rel->relam = HEAP_TABLE_AM_OID;
2014 24752 : relation->rd_tableam = GetHeapamTableAmRoutine();
2015 :
2016 : /*
2017 : * initialize the rel-has-index flag, using hardwired knowledge
2018 : */
2019 24752 : if (IsBootstrapProcessingMode())
2020 : {
2021 : /* In bootstrap mode, we have no indexes */
2022 320 : relation->rd_rel->relhasindex = false;
2023 : }
2024 : else
2025 : {
2026 : /* Otherwise, all the rels formrdesc is used for have indexes */
2027 24432 : relation->rd_rel->relhasindex = true;
2028 : }
2029 :
2030 : /*
2031 : * add new reldesc to relcache
2032 : */
2033 24752 : RelationCacheInsert(relation, false);
2034 :
2035 : /* It's fully valid */
2036 24752 : relation->rd_isvalid = true;
2037 24752 : }
2038 :
2039 :
2040 : /* ----------------------------------------------------------------
2041 : * Relation Descriptor Lookup Interface
2042 : * ----------------------------------------------------------------
2043 : */
2044 :
2045 : /*
2046 : * RelationIdGetRelation
2047 : *
2048 : * Lookup a reldesc by OID; make one if not already in cache.
2049 : *
2050 : * Returns NULL if no pg_class row could be found for the given relid
2051 : * (suggesting we are trying to access a just-deleted relation).
2052 : * Any other error is reported via elog.
2053 : *
2054 : * NB: caller should already have at least AccessShareLock on the
2055 : * relation ID, else there are nasty race conditions.
2056 : *
2057 : * NB: relation ref count is incremented, or set to 1 if new entry.
2058 : * Caller should eventually decrement count. (Usually,
2059 : * that happens by calling RelationClose().)
2060 : */
2061 : Relation
2062 32101012 : RelationIdGetRelation(Oid relationId)
2063 : {
2064 : Relation rd;
2065 :
2066 : /* Make sure we're in an xact, even if this ends up being a cache hit */
2067 : Assert(IsTransactionState());
2068 :
2069 : /*
2070 : * first try to find reldesc in the cache
2071 : */
2072 32101012 : RelationIdCacheLookup(relationId, rd);
2073 :
2074 32101012 : if (RelationIsValid(rd))
2075 : {
2076 : /* return NULL for dropped relations */
2077 31349334 : if (rd->rd_droppedSubid != InvalidSubTransactionId)
2078 : {
2079 : Assert(!rd->rd_isvalid);
2080 4 : return NULL;
2081 : }
2082 :
2083 31349330 : RelationIncrementReferenceCount(rd);
2084 : /* revalidate cache entry if necessary */
2085 31349330 : if (!rd->rd_isvalid)
2086 : {
2087 : /*
2088 : * Indexes only have a limited number of possible schema changes,
2089 : * and we don't want to use the full-blown procedure because it's
2090 : * a headache for indexes that reload itself depends on.
2091 : */
2092 139394 : if (rd->rd_rel->relkind == RELKIND_INDEX ||
2093 117222 : rd->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
2094 22172 : RelationReloadIndexInfo(rd);
2095 : else
2096 117222 : RelationClearRelation(rd, true);
2097 :
2098 : /*
2099 : * Normally entries need to be valid here, but before the relcache
2100 : * has been initialized, not enough infrastructure exists to
2101 : * perform pg_class lookups. The structure of such entries doesn't
2102 : * change, but we still want to update the rd_rel entry. So
2103 : * rd_isvalid = false is left in place for a later lookup.
2104 : */
2105 : Assert(rd->rd_isvalid ||
2106 : (rd->rd_isnailed && !criticalRelcachesBuilt));
2107 : }
2108 31349318 : return rd;
2109 : }
2110 :
2111 : /*
2112 : * no reldesc in the cache, so have RelationBuildDesc() build one and add
2113 : * it.
2114 : */
2115 751678 : rd = RelationBuildDesc(relationId, true);
2116 751676 : if (RelationIsValid(rd))
2117 751666 : RelationIncrementReferenceCount(rd);
2118 751676 : return rd;
2119 : }
2120 :
2121 : /* ----------------------------------------------------------------
2122 : * cache invalidation support routines
2123 : * ----------------------------------------------------------------
2124 : */
2125 :
2126 : /* ResourceOwner callbacks to track relcache references */
2127 : static void ResOwnerReleaseRelation(Datum res);
2128 : static char *ResOwnerPrintRelCache(Datum res);
2129 :
2130 : static const ResourceOwnerDesc relref_resowner_desc =
2131 : {
2132 : .name = "relcache reference",
2133 : .release_phase = RESOURCE_RELEASE_BEFORE_LOCKS,
2134 : .release_priority = RELEASE_PRIO_RELCACHE_REFS,
2135 : .ReleaseResource = ResOwnerReleaseRelation,
2136 : .DebugPrint = ResOwnerPrintRelCache
2137 : };
2138 :
2139 : /* Convenience wrappers over ResourceOwnerRemember/Forget */
2140 : static inline void
2141 46849768 : ResourceOwnerRememberRelationRef(ResourceOwner owner, Relation rel)
2142 : {
2143 46849768 : ResourceOwnerRemember(owner, PointerGetDatum(rel), &relref_resowner_desc);
2144 46849768 : }
2145 : static inline void
2146 46814460 : ResourceOwnerForgetRelationRef(ResourceOwner owner, Relation rel)
2147 : {
2148 46814460 : ResourceOwnerForget(owner, PointerGetDatum(rel), &relref_resowner_desc);
2149 46814460 : }
2150 :
2151 : /*
2152 : * RelationIncrementReferenceCount
2153 : * Increments relation reference count.
2154 : *
2155 : * Note: bootstrap mode has its own weird ideas about relation refcount
2156 : * behavior; we ought to fix it someday, but for now, just disable
2157 : * reference count ownership tracking in bootstrap mode.
2158 : */
2159 : void
2160 47305688 : RelationIncrementReferenceCount(Relation rel)
2161 : {
2162 47305688 : ResourceOwnerEnlarge(CurrentResourceOwner);
2163 47305688 : rel->rd_refcnt += 1;
2164 47305688 : if (!IsBootstrapProcessingMode())
2165 46849768 : ResourceOwnerRememberRelationRef(CurrentResourceOwner, rel);
2166 47305688 : }
2167 :
2168 : /*
2169 : * RelationDecrementReferenceCount
2170 : * Decrements relation reference count.
2171 : */
2172 : void
2173 47270380 : RelationDecrementReferenceCount(Relation rel)
2174 : {
2175 : Assert(rel->rd_refcnt > 0);
2176 47270380 : rel->rd_refcnt -= 1;
2177 47270380 : if (!IsBootstrapProcessingMode())
2178 46814460 : ResourceOwnerForgetRelationRef(CurrentResourceOwner, rel);
2179 47270380 : }
2180 :
2181 : /*
2182 : * RelationClose - close an open relation
2183 : *
2184 : * Actually, we just decrement the refcount.
2185 : *
2186 : * NOTE: if compiled with -DRELCACHE_FORCE_RELEASE then relcache entries
2187 : * will be freed as soon as their refcount goes to zero. In combination
2188 : * with aset.c's CLOBBER_FREED_MEMORY option, this provides a good test
2189 : * to catch references to already-released relcache entries. It slows
2190 : * things down quite a bit, however.
2191 : */
2192 : void
2193 32189568 : RelationClose(Relation relation)
2194 : {
2195 : /* Note: no locking manipulations needed */
2196 32189568 : RelationDecrementReferenceCount(relation);
2197 :
2198 32189568 : RelationCloseCleanup(relation);
2199 32189568 : }
2200 :
2201 : static void
2202 32224876 : RelationCloseCleanup(Relation relation)
2203 : {
2204 : /*
2205 : * If the relation is no longer open in this session, we can clean up any
2206 : * stale partition descriptors it has. This is unlikely, so check to see
2207 : * if there are child contexts before expending a call to mcxt.c.
2208 : */
2209 32224876 : if (RelationHasReferenceCountZero(relation))
2210 : {
2211 19359600 : if (relation->rd_pdcxt != NULL &&
2212 93550 : relation->rd_pdcxt->firstchild != NULL)
2213 3994 : MemoryContextDeleteChildren(relation->rd_pdcxt);
2214 :
2215 19359600 : if (relation->rd_pddcxt != NULL &&
2216 94 : relation->rd_pddcxt->firstchild != NULL)
2217 0 : MemoryContextDeleteChildren(relation->rd_pddcxt);
2218 : }
2219 :
2220 : #ifdef RELCACHE_FORCE_RELEASE
2221 : if (RelationHasReferenceCountZero(relation) &&
2222 : relation->rd_createSubid == InvalidSubTransactionId &&
2223 : relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId)
2224 : RelationClearRelation(relation, false);
2225 : #endif
2226 32224876 : }
2227 :
2228 : /*
2229 : * RelationReloadIndexInfo - reload minimal information for an open index
2230 : *
2231 : * This function is used only for indexes. A relcache inval on an index
2232 : * can mean that its pg_class or pg_index row changed. There are only
2233 : * very limited changes that are allowed to an existing index's schema,
2234 : * so we can update the relcache entry without a complete rebuild; which
2235 : * is fortunate because we can't rebuild an index entry that is "nailed"
2236 : * and/or in active use. We support full replacement of the pg_class row,
2237 : * as well as updates of a few simple fields of the pg_index row.
2238 : *
2239 : * We can't necessarily reread the catalog rows right away; we might be
2240 : * in a failed transaction when we receive the SI notification. If so,
2241 : * RelationClearRelation just marks the entry as invalid by setting
2242 : * rd_isvalid to false. This routine is called to fix the entry when it
2243 : * is next needed.
2244 : *
2245 : * We assume that at the time we are called, we have at least AccessShareLock
2246 : * on the target index. (Note: in the calls from RelationClearRelation,
2247 : * this is legitimate because we know the rel has positive refcount.)
2248 : *
2249 : * If the target index is an index on pg_class or pg_index, we'd better have
2250 : * previously gotten at least AccessShareLock on its underlying catalog,
2251 : * else we are at risk of deadlock against someone trying to exclusive-lock
2252 : * the heap and index in that order. This is ensured in current usage by
2253 : * only applying this to indexes being opened or having positive refcount.
2254 : */
2255 : static void
2256 70040 : RelationReloadIndexInfo(Relation relation)
2257 : {
2258 : bool indexOK;
2259 : HeapTuple pg_class_tuple;
2260 : Form_pg_class relp;
2261 :
2262 : /* Should be called only for invalidated, live indexes */
2263 : Assert((relation->rd_rel->relkind == RELKIND_INDEX ||
2264 : relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX) &&
2265 : !relation->rd_isvalid &&
2266 : relation->rd_droppedSubid == InvalidSubTransactionId);
2267 :
2268 : /* Ensure it's closed at smgr level */
2269 70040 : RelationCloseSmgr(relation);
2270 :
2271 : /* Must free any AM cached data upon relcache flush */
2272 70040 : if (relation->rd_amcache)
2273 0 : pfree(relation->rd_amcache);
2274 70040 : relation->rd_amcache = NULL;
2275 :
2276 : /*
2277 : * If it's a shared index, we might be called before backend startup has
2278 : * finished selecting a database, in which case we have no way to read
2279 : * pg_class yet. However, a shared index can never have any significant
2280 : * schema updates, so it's okay to ignore the invalidation signal. Just
2281 : * mark it valid and return without doing anything more.
2282 : */
2283 70040 : if (relation->rd_rel->relisshared && !criticalRelcachesBuilt)
2284 : {
2285 0 : relation->rd_isvalid = true;
2286 0 : return;
2287 : }
2288 :
2289 : /*
2290 : * Read the pg_class row
2291 : *
2292 : * Don't try to use an indexscan of pg_class_oid_index to reload the info
2293 : * for pg_class_oid_index ...
2294 : */
2295 70040 : indexOK = (RelationGetRelid(relation) != ClassOidIndexId);
2296 70040 : pg_class_tuple = ScanPgRelation(RelationGetRelid(relation), indexOK, false);
2297 70034 : if (!HeapTupleIsValid(pg_class_tuple))
2298 0 : elog(ERROR, "could not find pg_class tuple for index %u",
2299 : RelationGetRelid(relation));
2300 70034 : relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
2301 70034 : memcpy(relation->rd_rel, relp, CLASS_TUPLE_SIZE);
2302 : /* Reload reloptions in case they changed */
2303 70034 : if (relation->rd_options)
2304 510 : pfree(relation->rd_options);
2305 70034 : RelationParseRelOptions(relation, pg_class_tuple);
2306 : /* done with pg_class tuple */
2307 70034 : heap_freetuple(pg_class_tuple);
2308 : /* We must recalculate physical address in case it changed */
2309 70034 : RelationInitPhysicalAddr(relation);
2310 :
2311 : /*
2312 : * For a non-system index, there are fields of the pg_index row that are
2313 : * allowed to change, so re-read that row and update the relcache entry.
2314 : * Most of the info derived from pg_index (such as support function lookup
2315 : * info) cannot change, and indeed the whole point of this routine is to
2316 : * update the relcache entry without clobbering that data; so wholesale
2317 : * replacement is not appropriate.
2318 : */
2319 70034 : if (!IsSystemRelation(relation))
2320 : {
2321 : HeapTuple tuple;
2322 : Form_pg_index index;
2323 :
2324 24140 : tuple = SearchSysCache1(INDEXRELID,
2325 : ObjectIdGetDatum(RelationGetRelid(relation)));
2326 24140 : if (!HeapTupleIsValid(tuple))
2327 0 : elog(ERROR, "cache lookup failed for index %u",
2328 : RelationGetRelid(relation));
2329 24140 : index = (Form_pg_index) GETSTRUCT(tuple);
2330 :
2331 : /*
2332 : * Basically, let's just copy all the bool fields. There are one or
2333 : * two of these that can't actually change in the current code, but
2334 : * it's not worth it to track exactly which ones they are. None of
2335 : * the array fields are allowed to change, though.
2336 : */
2337 24140 : relation->rd_index->indisunique = index->indisunique;
2338 24140 : relation->rd_index->indnullsnotdistinct = index->indnullsnotdistinct;
2339 24140 : relation->rd_index->indisprimary = index->indisprimary;
2340 24140 : relation->rd_index->indisexclusion = index->indisexclusion;
2341 24140 : relation->rd_index->indimmediate = index->indimmediate;
2342 24140 : relation->rd_index->indisclustered = index->indisclustered;
2343 24140 : relation->rd_index->indisvalid = index->indisvalid;
2344 24140 : relation->rd_index->indcheckxmin = index->indcheckxmin;
2345 24140 : relation->rd_index->indisready = index->indisready;
2346 24140 : relation->rd_index->indislive = index->indislive;
2347 24140 : relation->rd_index->indisreplident = index->indisreplident;
2348 :
2349 : /* Copy xmin too, as that is needed to make sense of indcheckxmin */
2350 24140 : HeapTupleHeaderSetXmin(relation->rd_indextuple->t_data,
2351 : HeapTupleHeaderGetXmin(tuple->t_data));
2352 :
2353 24140 : ReleaseSysCache(tuple);
2354 : }
2355 :
2356 : /* Okay, now it's valid again */
2357 70034 : relation->rd_isvalid = true;
2358 : }
2359 :
2360 : /*
2361 : * RelationReloadNailed - reload minimal information for nailed relations.
2362 : *
2363 : * The structure of a nailed relation can never change (which is good, because
2364 : * we rely on knowing their structure to be able to read catalog content). But
2365 : * some parts, e.g. pg_class.relfrozenxid, are still important to have
2366 : * accurate content for. Therefore those need to be reloaded after the arrival
2367 : * of invalidations.
2368 : */
2369 : static void
2370 222592 : RelationReloadNailed(Relation relation)
2371 : {
2372 : Assert(relation->rd_isnailed);
2373 :
2374 : /*
2375 : * Redo RelationInitPhysicalAddr in case it is a mapped relation whose
2376 : * mapping changed.
2377 : */
2378 222592 : RelationInitPhysicalAddr(relation);
2379 :
2380 : /* flag as needing to be revalidated */
2381 222592 : relation->rd_isvalid = false;
2382 :
2383 : /*
2384 : * Can only reread catalog contents if in a transaction. If the relation
2385 : * is currently open (not counting the nailed refcount), do so
2386 : * immediately. Otherwise we've already marked the entry as possibly
2387 : * invalid, and it'll be fixed when next opened.
2388 : */
2389 222592 : if (!IsTransactionState() || relation->rd_refcnt <= 1)
2390 101364 : return;
2391 :
2392 121228 : if (relation->rd_rel->relkind == RELKIND_INDEX)
2393 : {
2394 : /*
2395 : * If it's a nailed-but-not-mapped index, then we need to re-read the
2396 : * pg_class row to see if its relfilenumber changed.
2397 : */
2398 580 : RelationReloadIndexInfo(relation);
2399 : }
2400 : else
2401 : {
2402 : /*
2403 : * Reload a non-index entry. We can't easily do so if relcaches
2404 : * aren't yet built, but that's fine because at that stage the
2405 : * attributes that need to be current (like relfrozenxid) aren't yet
2406 : * accessed. To ensure the entry will later be revalidated, we leave
2407 : * it in invalid state, but allow use (cf. RelationIdGetRelation()).
2408 : */
2409 120648 : if (criticalRelcachesBuilt)
2410 : {
2411 : HeapTuple pg_class_tuple;
2412 : Form_pg_class relp;
2413 :
2414 : /*
2415 : * NB: Mark the entry as valid before starting to scan, to avoid
2416 : * self-recursion when re-building pg_class.
2417 : */
2418 21398 : relation->rd_isvalid = true;
2419 :
2420 21398 : pg_class_tuple = ScanPgRelation(RelationGetRelid(relation),
2421 : true, false);
2422 21392 : relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
2423 21392 : memcpy(relation->rd_rel, relp, CLASS_TUPLE_SIZE);
2424 21392 : heap_freetuple(pg_class_tuple);
2425 :
2426 : /*
2427 : * Again mark as valid, to protect against concurrently arriving
2428 : * invalidations.
2429 : */
2430 21392 : relation->rd_isvalid = true;
2431 : }
2432 : }
2433 : }
2434 :
2435 : /*
2436 : * RelationDestroyRelation
2437 : *
2438 : * Physically delete a relation cache entry and all subsidiary data.
2439 : * Caller must already have unhooked the entry from the hash table.
2440 : */
2441 : static void
2442 993526 : RelationDestroyRelation(Relation relation, bool remember_tupdesc)
2443 : {
2444 : Assert(RelationHasReferenceCountZero(relation));
2445 :
2446 : /*
2447 : * Make sure smgr and lower levels close the relation's files, if they
2448 : * weren't closed already. (This was probably done by caller, but let's
2449 : * just be real sure.)
2450 : */
2451 993526 : RelationCloseSmgr(relation);
2452 :
2453 : /* break mutual link with stats entry */
2454 993526 : pgstat_unlink_relation(relation);
2455 :
2456 : /*
2457 : * Free all the subsidiary data structures of the relcache entry, then the
2458 : * entry itself.
2459 : */
2460 993526 : if (relation->rd_rel)
2461 993526 : pfree(relation->rd_rel);
2462 : /* can't use DecrTupleDescRefCount here */
2463 : Assert(relation->rd_att->tdrefcount > 0);
2464 993526 : if (--relation->rd_att->tdrefcount == 0)
2465 : {
2466 : /*
2467 : * If we Rebuilt a relcache entry during a transaction then its
2468 : * possible we did that because the TupDesc changed as the result of
2469 : * an ALTER TABLE that ran at less than AccessExclusiveLock. It's
2470 : * possible someone copied that TupDesc, in which case the copy would
2471 : * point to free'd memory. So if we rebuild an entry we keep the
2472 : * TupDesc around until end of transaction, to be safe.
2473 : */
2474 990638 : if (remember_tupdesc)
2475 19474 : RememberToFreeTupleDescAtEOX(relation->rd_att);
2476 : else
2477 971164 : FreeTupleDesc(relation->rd_att);
2478 : }
2479 993526 : FreeTriggerDesc(relation->trigdesc);
2480 993526 : list_free_deep(relation->rd_fkeylist);
2481 993526 : list_free(relation->rd_indexlist);
2482 993526 : list_free(relation->rd_statlist);
2483 993526 : bms_free(relation->rd_keyattr);
2484 993526 : bms_free(relation->rd_pkattr);
2485 993526 : bms_free(relation->rd_idattr);
2486 993526 : bms_free(relation->rd_hotblockingattr);
2487 993526 : bms_free(relation->rd_summarizedattr);
2488 993526 : if (relation->rd_pubdesc)
2489 6344 : pfree(relation->rd_pubdesc);
2490 993526 : if (relation->rd_options)
2491 8348 : pfree(relation->rd_options);
2492 993526 : if (relation->rd_indextuple)
2493 311246 : pfree(relation->rd_indextuple);
2494 993526 : if (relation->rd_amcache)
2495 0 : pfree(relation->rd_amcache);
2496 993526 : if (relation->rd_fdwroutine)
2497 264 : pfree(relation->rd_fdwroutine);
2498 993526 : if (relation->rd_indexcxt)
2499 311246 : MemoryContextDelete(relation->rd_indexcxt);
2500 993526 : if (relation->rd_rulescxt)
2501 21182 : MemoryContextDelete(relation->rd_rulescxt);
2502 993526 : if (relation->rd_rsdesc)
2503 1808 : MemoryContextDelete(relation->rd_rsdesc->rscxt);
2504 993526 : if (relation->rd_partkeycxt)
2505 16042 : MemoryContextDelete(relation->rd_partkeycxt);
2506 993526 : if (relation->rd_pdcxt)
2507 15526 : MemoryContextDelete(relation->rd_pdcxt);
2508 993526 : if (relation->rd_pddcxt)
2509 60 : MemoryContextDelete(relation->rd_pddcxt);
2510 993526 : if (relation->rd_partcheckcxt)
2511 2906 : MemoryContextDelete(relation->rd_partcheckcxt);
2512 993526 : pfree(relation);
2513 993526 : }
2514 :
2515 : /*
2516 : * RelationClearRelation
2517 : *
2518 : * Physically blow away a relation cache entry, or reset it and rebuild
2519 : * it from scratch (that is, from catalog entries). The latter path is
2520 : * used when we are notified of a change to an open relation (one with
2521 : * refcount > 0).
2522 : *
2523 : * NB: when rebuilding, we'd better hold some lock on the relation,
2524 : * else the catalog data we need to read could be changing under us.
2525 : * Also, a rel to be rebuilt had better have refcnt > 0. This is because
2526 : * a sinval reset could happen while we're accessing the catalogs, and
2527 : * the rel would get blown away underneath us by RelationCacheInvalidate
2528 : * if it has zero refcnt.
2529 : *
2530 : * The "rebuild" parameter is redundant in current usage because it has
2531 : * to match the relation's refcnt status, but we keep it as a crosscheck
2532 : * that we're doing what the caller expects.
2533 : */
2534 : static void
2535 1265226 : RelationClearRelation(Relation relation, bool rebuild)
2536 : {
2537 : /*
2538 : * As per notes above, a rel to be rebuilt MUST have refcnt > 0; while of
2539 : * course it would be an equally bad idea to blow away one with nonzero
2540 : * refcnt, since that would leave someone somewhere with a dangling
2541 : * pointer. All callers are expected to have verified that this holds.
2542 : */
2543 : Assert(rebuild ?
2544 : !RelationHasReferenceCountZero(relation) :
2545 : RelationHasReferenceCountZero(relation));
2546 :
2547 : /*
2548 : * Make sure smgr and lower levels close the relation's files, if they
2549 : * weren't closed already. If the relation is not getting deleted, the
2550 : * next smgr access should reopen the files automatically. This ensures
2551 : * that the low-level file access state is updated after, say, a vacuum
2552 : * truncation.
2553 : */
2554 1265226 : RelationCloseSmgr(relation);
2555 :
2556 : /* Free AM cached data, if any */
2557 1265226 : if (relation->rd_amcache)
2558 77534 : pfree(relation->rd_amcache);
2559 1265226 : relation->rd_amcache = NULL;
2560 :
2561 : /*
2562 : * Treat nailed-in system relations separately, they always need to be
2563 : * accessible, so we can't blow them away.
2564 : */
2565 1265226 : if (relation->rd_isnailed)
2566 : {
2567 222592 : RelationReloadNailed(relation);
2568 222586 : return;
2569 : }
2570 :
2571 : /* Mark it invalid until we've finished rebuild */
2572 1042634 : relation->rd_isvalid = false;
2573 :
2574 : /* See RelationForgetRelation(). */
2575 1042634 : if (relation->rd_droppedSubid != InvalidSubTransactionId)
2576 1760 : return;
2577 :
2578 : /*
2579 : * Even non-system indexes should not be blown away if they are open and
2580 : * have valid index support information. This avoids problems with active
2581 : * use of the index support information. As with nailed indexes, we
2582 : * re-read the pg_class row to handle possible physical relocation of the
2583 : * index, and we check for pg_index updates too.
2584 : */
2585 1040874 : if ((relation->rd_rel->relkind == RELKIND_INDEX ||
2586 662932 : relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX) &&
2587 388950 : relation->rd_refcnt > 0 &&
2588 77706 : relation->rd_indexcxt != NULL)
2589 : {
2590 47288 : if (IsTransactionState())
2591 47288 : RelationReloadIndexInfo(relation);
2592 47288 : return;
2593 : }
2594 :
2595 : /*
2596 : * If we're really done with the relcache entry, blow it away. But if
2597 : * someone is still using it, reconstruct the whole deal without moving
2598 : * the physical RelationData record (so that the someone's pointer is
2599 : * still valid).
2600 : */
2601 993586 : if (!rebuild)
2602 : {
2603 : /* Remove it from the hash table */
2604 707404 : RelationCacheDelete(relation);
2605 :
2606 : /* And release storage */
2607 707404 : RelationDestroyRelation(relation, false);
2608 : }
2609 286182 : else if (!IsTransactionState())
2610 : {
2611 : /*
2612 : * If we're not inside a valid transaction, we can't do any catalog
2613 : * access so it's not possible to rebuild yet. Just exit, leaving
2614 : * rd_isvalid = false so that the rebuild will occur when the entry is
2615 : * next opened.
2616 : *
2617 : * Note: it's possible that we come here during subtransaction abort,
2618 : * and the reason for wanting to rebuild is that the rel is open in
2619 : * the outer transaction. In that case it might seem unsafe to not
2620 : * rebuild immediately, since whatever code has the rel already open
2621 : * will keep on using the relcache entry as-is. However, in such a
2622 : * case the outer transaction should be holding a lock that's
2623 : * sufficient to prevent any significant change in the rel's schema,
2624 : * so the existing entry contents should be good enough for its
2625 : * purposes; at worst we might be behind on statistics updates or the
2626 : * like. (See also CheckTableNotInUse() and its callers.) These same
2627 : * remarks also apply to the cases above where we exit without having
2628 : * done RelationReloadIndexInfo() yet.
2629 : */
2630 72 : return;
2631 : }
2632 : else
2633 : {
2634 : /*
2635 : * Our strategy for rebuilding an open relcache entry is to build a
2636 : * new entry from scratch, swap its contents with the old entry, and
2637 : * finally delete the new entry (along with any infrastructure swapped
2638 : * over from the old entry). This is to avoid trouble in case an
2639 : * error causes us to lose control partway through. The old entry
2640 : * will still be marked !rd_isvalid, so we'll try to rebuild it again
2641 : * on next access. Meanwhile it's not any less valid than it was
2642 : * before, so any code that might expect to continue accessing it
2643 : * isn't hurt by the rebuild failure. (Consider for example a
2644 : * subtransaction that ALTERs a table and then gets canceled partway
2645 : * through the cache entry rebuild. The outer transaction should
2646 : * still see the not-modified cache entry as valid.) The worst
2647 : * consequence of an error is leaking the necessarily-unreferenced new
2648 : * entry, and this shouldn't happen often enough for that to be a big
2649 : * problem.
2650 : *
2651 : * When rebuilding an open relcache entry, we must preserve ref count,
2652 : * rd_*Subid, and rd_toastoid state. Also attempt to preserve the
2653 : * pg_class entry (rd_rel), tupledesc, rewrite-rule, partition key,
2654 : * and partition descriptor substructures in place, because various
2655 : * places assume that these structures won't move while they are
2656 : * working with an open relcache entry. (Note: the refcount
2657 : * mechanism for tupledescs might someday allow us to remove this hack
2658 : * for the tupledesc.)
2659 : *
2660 : * Note that this process does not touch CurrentResourceOwner; which
2661 : * is good because whatever ref counts the entry may have do not
2662 : * necessarily belong to that resource owner.
2663 : */
2664 : Relation newrel;
2665 286110 : Oid save_relid = RelationGetRelid(relation);
2666 : bool keep_tupdesc;
2667 : bool keep_rules;
2668 : bool keep_policies;
2669 : bool keep_partkey;
2670 :
2671 : /* Build temporary entry, but don't link it into hashtable */
2672 286110 : newrel = RelationBuildDesc(save_relid, false);
2673 :
2674 : /*
2675 : * Between here and the end of the swap, don't add code that does or
2676 : * reasonably could read system catalogs. That range must be free
2677 : * from invalidation processing. See RelationBuildDesc() manipulation
2678 : * of in_progress_list.
2679 : */
2680 :
2681 286104 : if (newrel == NULL)
2682 : {
2683 : /*
2684 : * We can validly get here, if we're using a historic snapshot in
2685 : * which a relation, accessed from outside logical decoding, is
2686 : * still invisible. In that case it's fine to just mark the
2687 : * relation as invalid and return - it'll fully get reloaded by
2688 : * the cache reset at the end of logical decoding (or at the next
2689 : * access). During normal processing we don't want to ignore this
2690 : * case as it shouldn't happen there, as explained below.
2691 : */
2692 0 : if (HistoricSnapshotActive())
2693 0 : return;
2694 :
2695 : /*
2696 : * This shouldn't happen as dropping a relation is intended to be
2697 : * impossible if still referenced (cf. CheckTableNotInUse()). But
2698 : * if we get here anyway, we can't just delete the relcache entry,
2699 : * as it possibly could get accessed later (as e.g. the error
2700 : * might get trapped and handled via a subtransaction rollback).
2701 : */
2702 0 : elog(ERROR, "relation %u deleted while still in use", save_relid);
2703 : }
2704 :
2705 : /*
2706 : * If we were to, again, have cases of the relkind of a relcache entry
2707 : * changing, we would need to ensure that pgstats does not get
2708 : * confused.
2709 : */
2710 : Assert(relation->rd_rel->relkind == newrel->rd_rel->relkind);
2711 :
2712 286104 : keep_tupdesc = equalTupleDescs(relation->rd_att, newrel->rd_att);
2713 286104 : keep_rules = equalRuleLocks(relation->rd_rules, newrel->rd_rules);
2714 286104 : keep_policies = equalRSDesc(relation->rd_rsdesc, newrel->rd_rsdesc);
2715 : /* partkey is immutable once set up, so we can always keep it */
2716 286104 : keep_partkey = (relation->rd_partkey != NULL);
2717 :
2718 : /*
2719 : * Perform swapping of the relcache entry contents. Within this
2720 : * process the old entry is momentarily invalid, so there *must* be no
2721 : * possibility of CHECK_FOR_INTERRUPTS within this sequence. Do it in
2722 : * all-in-line code for safety.
2723 : *
2724 : * Since the vast majority of fields should be swapped, our method is
2725 : * to swap the whole structures and then re-swap those few fields we
2726 : * didn't want swapped.
2727 : */
2728 : #define SWAPFIELD(fldtype, fldname) \
2729 : do { \
2730 : fldtype _tmp = newrel->fldname; \
2731 : newrel->fldname = relation->fldname; \
2732 : relation->fldname = _tmp; \
2733 : } while (0)
2734 :
2735 : /* swap all Relation struct fields */
2736 : {
2737 : RelationData tmpstruct;
2738 :
2739 286104 : memcpy(&tmpstruct, newrel, sizeof(RelationData));
2740 286104 : memcpy(newrel, relation, sizeof(RelationData));
2741 286104 : memcpy(relation, &tmpstruct, sizeof(RelationData));
2742 : }
2743 :
2744 : /* rd_smgr must not be swapped, due to back-links from smgr level */
2745 286104 : SWAPFIELD(SMgrRelation, rd_smgr);
2746 : /* rd_refcnt must be preserved */
2747 286104 : SWAPFIELD(int, rd_refcnt);
2748 : /* isnailed shouldn't change */
2749 : Assert(newrel->rd_isnailed == relation->rd_isnailed);
2750 : /* creation sub-XIDs must be preserved */
2751 286104 : SWAPFIELD(SubTransactionId, rd_createSubid);
2752 286104 : SWAPFIELD(SubTransactionId, rd_newRelfilelocatorSubid);
2753 286104 : SWAPFIELD(SubTransactionId, rd_firstRelfilelocatorSubid);
2754 286104 : SWAPFIELD(SubTransactionId, rd_droppedSubid);
2755 : /* un-swap rd_rel pointers, swap contents instead */
2756 286104 : SWAPFIELD(Form_pg_class, rd_rel);
2757 : /* ... but actually, we don't have to update newrel->rd_rel */
2758 286104 : memcpy(relation->rd_rel, newrel->rd_rel, CLASS_TUPLE_SIZE);
2759 : /* preserve old tupledesc, rules, policies if no logical change */
2760 286104 : if (keep_tupdesc)
2761 266414 : SWAPFIELD(TupleDesc, rd_att);
2762 286104 : if (keep_rules)
2763 : {
2764 271922 : SWAPFIELD(RuleLock *, rd_rules);
2765 271922 : SWAPFIELD(MemoryContext, rd_rulescxt);
2766 : }
2767 286104 : if (keep_policies)
2768 285816 : SWAPFIELD(RowSecurityDesc *, rd_rsdesc);
2769 : /* toast OID override must be preserved */
2770 286104 : SWAPFIELD(Oid, rd_toastoid);
2771 : /* pgstat_info / enabled must be preserved */
2772 286104 : SWAPFIELD(struct PgStat_TableStatus *, pgstat_info);
2773 286104 : SWAPFIELD(bool, pgstat_enabled);
2774 : /* preserve old partition key if we have one */
2775 286104 : if (keep_partkey)
2776 : {
2777 12648 : SWAPFIELD(PartitionKey, rd_partkey);
2778 12648 : SWAPFIELD(MemoryContext, rd_partkeycxt);
2779 : }
2780 286104 : if (newrel->rd_pdcxt != NULL || newrel->rd_pddcxt != NULL)
2781 : {
2782 : /*
2783 : * We are rebuilding a partitioned relation with a non-zero
2784 : * reference count, so we must keep the old partition descriptor
2785 : * around, in case there's a PartitionDirectory with a pointer to
2786 : * it. This means we can't free the old rd_pdcxt yet. (This is
2787 : * necessary because RelationGetPartitionDesc hands out direct
2788 : * pointers to the relcache's data structure, unlike our usual
2789 : * practice which is to hand out copies. We'd have the same
2790 : * problem with rd_partkey, except that we always preserve that
2791 : * once created.)
2792 : *
2793 : * To ensure that it's not leaked completely, re-attach it to the
2794 : * new reldesc, or make it a child of the new reldesc's rd_pdcxt
2795 : * in the unlikely event that there is one already. (Compare hack
2796 : * in RelationBuildPartitionDesc.) RelationClose will clean up
2797 : * any such contexts once the reference count reaches zero.
2798 : *
2799 : * In the case where the reference count is zero, this code is not
2800 : * reached, which should be OK because in that case there should
2801 : * be no PartitionDirectory with a pointer to the old entry.
2802 : *
2803 : * Note that newrel and relation have already been swapped, so the
2804 : * "old" partition descriptor is actually the one hanging off of
2805 : * newrel.
2806 : */
2807 10300 : relation->rd_partdesc = NULL; /* ensure rd_partdesc is invalid */
2808 10300 : relation->rd_partdesc_nodetached = NULL;
2809 10300 : relation->rd_partdesc_nodetached_xmin = InvalidTransactionId;
2810 10300 : if (relation->rd_pdcxt != NULL) /* probably never happens */
2811 0 : MemoryContextSetParent(newrel->rd_pdcxt, relation->rd_pdcxt);
2812 : else
2813 10300 : relation->rd_pdcxt = newrel->rd_pdcxt;
2814 10300 : if (relation->rd_pddcxt != NULL)
2815 0 : MemoryContextSetParent(newrel->rd_pddcxt, relation->rd_pddcxt);
2816 : else
2817 10300 : relation->rd_pddcxt = newrel->rd_pddcxt;
2818 : /* drop newrel's pointers so we don't destroy it below */
2819 10300 : newrel->rd_partdesc = NULL;
2820 10300 : newrel->rd_partdesc_nodetached = NULL;
2821 10300 : newrel->rd_partdesc_nodetached_xmin = InvalidTransactionId;
2822 10300 : newrel->rd_pdcxt = NULL;
2823 10300 : newrel->rd_pddcxt = NULL;
2824 : }
2825 :
2826 : #undef SWAPFIELD
2827 :
2828 : /* And now we can throw away the temporary entry */
2829 286104 : RelationDestroyRelation(newrel, !keep_tupdesc);
2830 : }
2831 : }
2832 :
2833 : /*
2834 : * RelationFlushRelation
2835 : *
2836 : * Rebuild the relation if it is open (refcount > 0), else blow it away.
2837 : * This is used when we receive a cache invalidation event for the rel.
2838 : */
2839 : static void
2840 611740 : RelationFlushRelation(Relation relation)
2841 : {
2842 611740 : if (relation->rd_createSubid != InvalidSubTransactionId ||
2843 360644 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
2844 : {
2845 : /*
2846 : * New relcache entries are always rebuilt, not flushed; else we'd
2847 : * forget the "new" status of the relation. Ditto for the
2848 : * new-relfilenumber status.
2849 : *
2850 : * The rel could have zero refcnt here, so temporarily increment the
2851 : * refcnt to ensure it's safe to rebuild it. We can assume that the
2852 : * current transaction has some lock on the rel already.
2853 : */
2854 264158 : RelationIncrementReferenceCount(relation);
2855 264158 : RelationClearRelation(relation, true);
2856 264152 : RelationDecrementReferenceCount(relation);
2857 : }
2858 : else
2859 : {
2860 : /*
2861 : * Pre-existing rels can be dropped from the relcache if not open.
2862 : */
2863 347582 : bool rebuild = !RelationHasReferenceCountZero(relation);
2864 :
2865 347582 : RelationClearRelation(relation, rebuild);
2866 : }
2867 611734 : }
2868 :
2869 : /*
2870 : * RelationForgetRelation - caller reports that it dropped the relation
2871 : */
2872 : void
2873 65212 : RelationForgetRelation(Oid rid)
2874 : {
2875 : Relation relation;
2876 :
2877 65212 : RelationIdCacheLookup(rid, relation);
2878 :
2879 65212 : if (!PointerIsValid(relation))
2880 0 : return; /* not in cache, nothing to do */
2881 :
2882 65212 : if (!RelationHasReferenceCountZero(relation))
2883 0 : elog(ERROR, "relation %u is still open", rid);
2884 :
2885 : Assert(relation->rd_droppedSubid == InvalidSubTransactionId);
2886 65212 : if (relation->rd_createSubid != InvalidSubTransactionId ||
2887 64360 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
2888 : {
2889 : /*
2890 : * In the event of subtransaction rollback, we must not forget
2891 : * rd_*Subid. Mark the entry "dropped" so RelationClearRelation()
2892 : * invalidates it in lieu of destroying it. (If we're in a top
2893 : * transaction, we could opt to destroy the entry.)
2894 : */
2895 880 : relation->rd_droppedSubid = GetCurrentSubTransactionId();
2896 : }
2897 :
2898 65212 : RelationClearRelation(relation, false);
2899 : }
2900 :
2901 : /*
2902 : * RelationCacheInvalidateEntry
2903 : *
2904 : * This routine is invoked for SI cache flush messages.
2905 : *
2906 : * Any relcache entry matching the relid must be flushed. (Note: caller has
2907 : * already determined that the relid belongs to our database or is a shared
2908 : * relation.)
2909 : *
2910 : * We used to skip local relations, on the grounds that they could
2911 : * not be targets of cross-backend SI update messages; but it seems
2912 : * safer to process them, so that our *own* SI update messages will
2913 : * have the same effects during CommandCounterIncrement for both
2914 : * local and nonlocal relations.
2915 : */
2916 : void
2917 2185134 : RelationCacheInvalidateEntry(Oid relationId)
2918 : {
2919 : Relation relation;
2920 :
2921 2185134 : RelationIdCacheLookup(relationId, relation);
2922 :
2923 2185134 : if (PointerIsValid(relation))
2924 : {
2925 611740 : relcacheInvalsReceived++;
2926 611740 : RelationFlushRelation(relation);
2927 : }
2928 : else
2929 : {
2930 : int i;
2931 :
2932 1597186 : for (i = 0; i < in_progress_list_len; i++)
2933 23792 : if (in_progress_list[i].reloid == relationId)
2934 16 : in_progress_list[i].invalidated = true;
2935 : }
2936 2185128 : }
2937 :
2938 : /*
2939 : * RelationCacheInvalidate
2940 : * Blow away cached relation descriptors that have zero reference counts,
2941 : * and rebuild those with positive reference counts. Also reset the smgr
2942 : * relation cache and re-read relation mapping data.
2943 : *
2944 : * Apart from debug_discard_caches, this is currently used only to recover
2945 : * from SI message buffer overflow, so we do not touch relations having
2946 : * new-in-transaction relfilenumbers; they cannot be targets of cross-backend
2947 : * SI updates (and our own updates now go through a separate linked list
2948 : * that isn't limited by the SI message buffer size).
2949 : *
2950 : * We do this in two phases: the first pass deletes deletable items, and
2951 : * the second one rebuilds the rebuildable items. This is essential for
2952 : * safety, because hash_seq_search only copes with concurrent deletion of
2953 : * the element it is currently visiting. If a second SI overflow were to
2954 : * occur while we are walking the table, resulting in recursive entry to
2955 : * this routine, we could crash because the inner invocation blows away
2956 : * the entry next to be visited by the outer scan. But this way is OK,
2957 : * because (a) during the first pass we won't process any more SI messages,
2958 : * so hash_seq_search will complete safely; (b) during the second pass we
2959 : * only hold onto pointers to nondeletable entries.
2960 : *
2961 : * The two-phase approach also makes it easy to update relfilenumbers for
2962 : * mapped relations before we do anything else, and to ensure that the
2963 : * second pass processes nailed-in-cache items before other nondeletable
2964 : * items. This should ensure that system catalogs are up to date before
2965 : * we attempt to use them to reload information about other open relations.
2966 : *
2967 : * After those two phases of work having immediate effects, we normally
2968 : * signal any RelationBuildDesc() on the stack to start over. However, we
2969 : * don't do this if called as part of debug_discard_caches. Otherwise,
2970 : * RelationBuildDesc() would become an infinite loop.
2971 : */
2972 : void
2973 4316 : RelationCacheInvalidate(bool debug_discard)
2974 : {
2975 : HASH_SEQ_STATUS status;
2976 : RelIdCacheEnt *idhentry;
2977 : Relation relation;
2978 4316 : List *rebuildFirstList = NIL;
2979 4316 : List *rebuildList = NIL;
2980 : ListCell *l;
2981 : int i;
2982 :
2983 : /*
2984 : * Reload relation mapping data before starting to reconstruct cache.
2985 : */
2986 4316 : RelationMapInvalidateAll();
2987 :
2988 : /* Phase 1 */
2989 4316 : hash_seq_init(&status, RelationIdCache);
2990 :
2991 470970 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
2992 : {
2993 466654 : relation = idhentry->reldesc;
2994 :
2995 : /*
2996 : * Ignore new relations; no other backend will manipulate them before
2997 : * we commit. Likewise, before replacing a relation's relfilelocator,
2998 : * we shall have acquired AccessExclusiveLock and drained any
2999 : * applicable pending invalidations.
3000 : */
3001 466654 : if (relation->rd_createSubid != InvalidSubTransactionId ||
3002 466612 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
3003 48 : continue;
3004 :
3005 466606 : relcacheInvalsReceived++;
3006 :
3007 466606 : if (RelationHasReferenceCountZero(relation))
3008 : {
3009 : /* Delete this entry immediately */
3010 : Assert(!relation->rd_isnailed);
3011 372790 : RelationClearRelation(relation, false);
3012 : }
3013 : else
3014 : {
3015 : /*
3016 : * If it's a mapped relation, immediately update its rd_locator in
3017 : * case its relfilenumber changed. We must do this during phase 1
3018 : * in case the relation is consulted during rebuild of other
3019 : * relcache entries in phase 2. It's safe since consulting the
3020 : * map doesn't involve any access to relcache entries.
3021 : */
3022 93816 : if (RelationIsMapped(relation))
3023 72438 : RelationInitPhysicalAddr(relation);
3024 :
3025 : /*
3026 : * Add this entry to list of stuff to rebuild in second pass.
3027 : * pg_class goes to the front of rebuildFirstList while
3028 : * pg_class_oid_index goes to the back of rebuildFirstList, so
3029 : * they are done first and second respectively. Other nailed
3030 : * relations go to the front of rebuildList, so they'll be done
3031 : * next in no particular order; and everything else goes to the
3032 : * back of rebuildList.
3033 : */
3034 93816 : if (RelationGetRelid(relation) == RelationRelationId)
3035 4236 : rebuildFirstList = lcons(relation, rebuildFirstList);
3036 89580 : else if (RelationGetRelid(relation) == ClassOidIndexId)
3037 4236 : rebuildFirstList = lappend(rebuildFirstList, relation);
3038 85344 : else if (relation->rd_isnailed)
3039 85144 : rebuildList = lcons(relation, rebuildList);
3040 : else
3041 200 : rebuildList = lappend(rebuildList, relation);
3042 : }
3043 : }
3044 :
3045 : /*
3046 : * We cannot destroy the SMgrRelations as there might still be references
3047 : * to them, but close the underlying file descriptors.
3048 : */
3049 4316 : smgrreleaseall();
3050 :
3051 : /* Phase 2: rebuild the items found to need rebuild in phase 1 */
3052 12788 : foreach(l, rebuildFirstList)
3053 : {
3054 8472 : relation = (Relation) lfirst(l);
3055 8472 : RelationClearRelation(relation, true);
3056 : }
3057 4316 : list_free(rebuildFirstList);
3058 89660 : foreach(l, rebuildList)
3059 : {
3060 85344 : relation = (Relation) lfirst(l);
3061 85344 : RelationClearRelation(relation, true);
3062 : }
3063 4316 : list_free(rebuildList);
3064 :
3065 4316 : if (!debug_discard)
3066 : /* Any RelationBuildDesc() on the stack must start over. */
3067 4322 : for (i = 0; i < in_progress_list_len; i++)
3068 6 : in_progress_list[i].invalidated = true;
3069 4316 : }
3070 :
3071 : static void
3072 19474 : RememberToFreeTupleDescAtEOX(TupleDesc td)
3073 : {
3074 19474 : if (EOXactTupleDescArray == NULL)
3075 : {
3076 : MemoryContext oldcxt;
3077 :
3078 10578 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
3079 :
3080 10578 : EOXactTupleDescArray = (TupleDesc *) palloc(16 * sizeof(TupleDesc));
3081 10578 : EOXactTupleDescArrayLen = 16;
3082 10578 : NextEOXactTupleDescNum = 0;
3083 10578 : MemoryContextSwitchTo(oldcxt);
3084 : }
3085 8896 : else if (NextEOXactTupleDescNum >= EOXactTupleDescArrayLen)
3086 : {
3087 34 : int32 newlen = EOXactTupleDescArrayLen * 2;
3088 :
3089 : Assert(EOXactTupleDescArrayLen > 0);
3090 :
3091 34 : EOXactTupleDescArray = (TupleDesc *) repalloc(EOXactTupleDescArray,
3092 : newlen * sizeof(TupleDesc));
3093 34 : EOXactTupleDescArrayLen = newlen;
3094 : }
3095 :
3096 19474 : EOXactTupleDescArray[NextEOXactTupleDescNum++] = td;
3097 19474 : }
3098 :
3099 : #ifdef USE_ASSERT_CHECKING
3100 : static void
3101 : AssertPendingSyncConsistency(Relation relation)
3102 : {
3103 : bool relcache_verdict =
3104 : RelationIsPermanent(relation) &&
3105 : ((relation->rd_createSubid != InvalidSubTransactionId &&
3106 : RELKIND_HAS_STORAGE(relation->rd_rel->relkind)) ||
3107 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId);
3108 :
3109 : Assert(relcache_verdict == RelFileLocatorSkippingWAL(relation->rd_locator));
3110 :
3111 : if (relation->rd_droppedSubid != InvalidSubTransactionId)
3112 : Assert(!relation->rd_isvalid &&
3113 : (relation->rd_createSubid != InvalidSubTransactionId ||
3114 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId));
3115 : }
3116 :
3117 : /*
3118 : * AssertPendingSyncs_RelationCache
3119 : *
3120 : * Assert that relcache.c and storage.c agree on whether to skip WAL.
3121 : */
3122 : void
3123 : AssertPendingSyncs_RelationCache(void)
3124 : {
3125 : HASH_SEQ_STATUS status;
3126 : LOCALLOCK *locallock;
3127 : Relation *rels;
3128 : int maxrels;
3129 : int nrels;
3130 : RelIdCacheEnt *idhentry;
3131 : int i;
3132 :
3133 : /*
3134 : * Open every relation that this transaction has locked. If, for some
3135 : * relation, storage.c is skipping WAL and relcache.c is not skipping WAL,
3136 : * a CommandCounterIncrement() typically yields a local invalidation
3137 : * message that destroys the relcache entry. By recreating such entries
3138 : * here, we detect the problem.
3139 : */
3140 : PushActiveSnapshot(GetTransactionSnapshot());
3141 : maxrels = 1;
3142 : rels = palloc(maxrels * sizeof(*rels));
3143 : nrels = 0;
3144 : hash_seq_init(&status, GetLockMethodLocalHash());
3145 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3146 : {
3147 : Oid relid;
3148 : Relation r;
3149 :
3150 : if (locallock->nLocks <= 0)
3151 : continue;
3152 : if ((LockTagType) locallock->tag.lock.locktag_type !=
3153 : LOCKTAG_RELATION)
3154 : continue;
3155 : relid = ObjectIdGetDatum(locallock->tag.lock.locktag_field2);
3156 : r = RelationIdGetRelation(relid);
3157 : if (!RelationIsValid(r))
3158 : continue;
3159 : if (nrels >= maxrels)
3160 : {
3161 : maxrels *= 2;
3162 : rels = repalloc(rels, maxrels * sizeof(*rels));
3163 : }
3164 : rels[nrels++] = r;
3165 : }
3166 :
3167 : hash_seq_init(&status, RelationIdCache);
3168 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3169 : AssertPendingSyncConsistency(idhentry->reldesc);
3170 :
3171 : for (i = 0; i < nrels; i++)
3172 : RelationClose(rels[i]);
3173 : PopActiveSnapshot();
3174 : }
3175 : #endif
3176 :
3177 : /*
3178 : * AtEOXact_RelationCache
3179 : *
3180 : * Clean up the relcache at main-transaction commit or abort.
3181 : *
3182 : * Note: this must be called *before* processing invalidation messages.
3183 : * In the case of abort, we don't want to try to rebuild any invalidated
3184 : * cache entries (since we can't safely do database accesses). Therefore
3185 : * we must reset refcnts before handling pending invalidations.
3186 : *
3187 : * As of PostgreSQL 8.1, relcache refcnts should get released by the
3188 : * ResourceOwner mechanism. This routine just does a debugging
3189 : * cross-check that no pins remain. However, we also need to do special
3190 : * cleanup when the current transaction created any relations or made use
3191 : * of forced index lists.
3192 : */
3193 : void
3194 566958 : AtEOXact_RelationCache(bool isCommit)
3195 : {
3196 : HASH_SEQ_STATUS status;
3197 : RelIdCacheEnt *idhentry;
3198 : int i;
3199 :
3200 : /*
3201 : * Forget in_progress_list. This is relevant when we're aborting due to
3202 : * an error during RelationBuildDesc().
3203 : */
3204 : Assert(in_progress_list_len == 0 || !isCommit);
3205 566958 : in_progress_list_len = 0;
3206 :
3207 : /*
3208 : * Unless the eoxact_list[] overflowed, we only need to examine the rels
3209 : * listed in it. Otherwise fall back on a hash_seq_search scan.
3210 : *
3211 : * For simplicity, eoxact_list[] entries are not deleted till end of
3212 : * top-level transaction, even though we could remove them at
3213 : * subtransaction end in some cases, or remove relations from the list if
3214 : * they are cleared for other reasons. Therefore we should expect the
3215 : * case that list entries are not found in the hashtable; if not, there's
3216 : * nothing to do for them.
3217 : */
3218 566958 : if (eoxact_list_overflowed)
3219 : {
3220 90 : hash_seq_init(&status, RelationIdCache);
3221 26708 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3222 : {
3223 26618 : AtEOXact_cleanup(idhentry->reldesc, isCommit);
3224 : }
3225 : }
3226 : else
3227 : {
3228 675202 : for (i = 0; i < eoxact_list_len; i++)
3229 : {
3230 108334 : idhentry = (RelIdCacheEnt *) hash_search(RelationIdCache,
3231 108334 : &eoxact_list[i],
3232 : HASH_FIND,
3233 : NULL);
3234 108334 : if (idhentry != NULL)
3235 106292 : AtEOXact_cleanup(idhentry->reldesc, isCommit);
3236 : }
3237 : }
3238 :
3239 566958 : if (EOXactTupleDescArrayLen > 0)
3240 : {
3241 : Assert(EOXactTupleDescArray != NULL);
3242 30052 : for (i = 0; i < NextEOXactTupleDescNum; i++)
3243 19474 : FreeTupleDesc(EOXactTupleDescArray[i]);
3244 10578 : pfree(EOXactTupleDescArray);
3245 10578 : EOXactTupleDescArray = NULL;
3246 : }
3247 :
3248 : /* Now we're out of the transaction and can clear the lists */
3249 566958 : eoxact_list_len = 0;
3250 566958 : eoxact_list_overflowed = false;
3251 566958 : NextEOXactTupleDescNum = 0;
3252 566958 : EOXactTupleDescArrayLen = 0;
3253 566958 : }
3254 :
3255 : /*
3256 : * AtEOXact_cleanup
3257 : *
3258 : * Clean up a single rel at main-transaction commit or abort
3259 : *
3260 : * NB: this processing must be idempotent, because EOXactListAdd() doesn't
3261 : * bother to prevent duplicate entries in eoxact_list[].
3262 : */
3263 : static void
3264 132910 : AtEOXact_cleanup(Relation relation, bool isCommit)
3265 : {
3266 132910 : bool clear_relcache = false;
3267 :
3268 : /*
3269 : * The relcache entry's ref count should be back to its normal
3270 : * not-in-a-transaction state: 0 unless it's nailed in cache.
3271 : *
3272 : * In bootstrap mode, this is NOT true, so don't check it --- the
3273 : * bootstrap code expects relations to stay open across start/commit
3274 : * transaction calls. (That seems bogus, but it's not worth fixing.)
3275 : *
3276 : * Note: ideally this check would be applied to every relcache entry, not
3277 : * just those that have eoxact work to do. But it's not worth forcing a
3278 : * scan of the whole relcache just for this. (Moreover, doing so would
3279 : * mean that assert-enabled testing never tests the hash_search code path
3280 : * above, which seems a bad idea.)
3281 : */
3282 : #ifdef USE_ASSERT_CHECKING
3283 : if (!IsBootstrapProcessingMode())
3284 : {
3285 : int expected_refcnt;
3286 :
3287 : expected_refcnt = relation->rd_isnailed ? 1 : 0;
3288 : Assert(relation->rd_refcnt == expected_refcnt);
3289 : }
3290 : #endif
3291 :
3292 : /*
3293 : * Is the relation live after this transaction ends?
3294 : *
3295 : * During commit, clear the relcache entry if it is preserved after
3296 : * relation drop, in order not to orphan the entry. During rollback,
3297 : * clear the relcache entry if the relation is created in the current
3298 : * transaction since it isn't interesting any longer once we are out of
3299 : * the transaction.
3300 : */
3301 132910 : clear_relcache =
3302 : (isCommit ?
3303 132910 : relation->rd_droppedSubid != InvalidSubTransactionId :
3304 4132 : relation->rd_createSubid != InvalidSubTransactionId);
3305 :
3306 : /*
3307 : * Since we are now out of the transaction, reset the subids to zero. That
3308 : * also lets RelationClearRelation() drop the relcache entry.
3309 : */
3310 132910 : relation->rd_createSubid = InvalidSubTransactionId;
3311 132910 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3312 132910 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3313 132910 : relation->rd_droppedSubid = InvalidSubTransactionId;
3314 :
3315 132910 : if (clear_relcache)
3316 : {
3317 4322 : if (RelationHasReferenceCountZero(relation))
3318 : {
3319 4322 : RelationClearRelation(relation, false);
3320 4322 : return;
3321 : }
3322 : else
3323 : {
3324 : /*
3325 : * Hmm, somewhere there's a (leaked?) reference to the relation.
3326 : * We daren't remove the entry for fear of dereferencing a
3327 : * dangling pointer later. Bleat, and mark it as not belonging to
3328 : * the current transaction. Hopefully it'll get cleaned up
3329 : * eventually. This must be just a WARNING to avoid
3330 : * error-during-error-recovery loops.
3331 : */
3332 0 : elog(WARNING, "cannot remove relcache entry for \"%s\" because it has nonzero refcount",
3333 : RelationGetRelationName(relation));
3334 : }
3335 : }
3336 : }
3337 :
3338 : /*
3339 : * AtEOSubXact_RelationCache
3340 : *
3341 : * Clean up the relcache at sub-transaction commit or abort.
3342 : *
3343 : * Note: this must be called *before* processing invalidation messages.
3344 : */
3345 : void
3346 18110 : AtEOSubXact_RelationCache(bool isCommit, SubTransactionId mySubid,
3347 : SubTransactionId parentSubid)
3348 : {
3349 : HASH_SEQ_STATUS status;
3350 : RelIdCacheEnt *idhentry;
3351 : int i;
3352 :
3353 : /*
3354 : * Forget in_progress_list. This is relevant when we're aborting due to
3355 : * an error during RelationBuildDesc(). We don't commit subtransactions
3356 : * during RelationBuildDesc().
3357 : */
3358 : Assert(in_progress_list_len == 0 || !isCommit);
3359 18110 : in_progress_list_len = 0;
3360 :
3361 : /*
3362 : * Unless the eoxact_list[] overflowed, we only need to examine the rels
3363 : * listed in it. Otherwise fall back on a hash_seq_search scan. Same
3364 : * logic as in AtEOXact_RelationCache.
3365 : */
3366 18110 : if (eoxact_list_overflowed)
3367 : {
3368 0 : hash_seq_init(&status, RelationIdCache);
3369 0 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3370 : {
3371 0 : AtEOSubXact_cleanup(idhentry->reldesc, isCommit,
3372 : mySubid, parentSubid);
3373 : }
3374 : }
3375 : else
3376 : {
3377 27486 : for (i = 0; i < eoxact_list_len; i++)
3378 : {
3379 9376 : idhentry = (RelIdCacheEnt *) hash_search(RelationIdCache,
3380 9376 : &eoxact_list[i],
3381 : HASH_FIND,
3382 : NULL);
3383 9376 : if (idhentry != NULL)
3384 8432 : AtEOSubXact_cleanup(idhentry->reldesc, isCommit,
3385 : mySubid, parentSubid);
3386 : }
3387 : }
3388 :
3389 : /* Don't reset the list; we still need more cleanup later */
3390 18110 : }
3391 :
3392 : /*
3393 : * AtEOSubXact_cleanup
3394 : *
3395 : * Clean up a single rel at subtransaction commit or abort
3396 : *
3397 : * NB: this processing must be idempotent, because EOXactListAdd() doesn't
3398 : * bother to prevent duplicate entries in eoxact_list[].
3399 : */
3400 : static void
3401 8432 : AtEOSubXact_cleanup(Relation relation, bool isCommit,
3402 : SubTransactionId mySubid, SubTransactionId parentSubid)
3403 : {
3404 : /*
3405 : * Is it a relation created in the current subtransaction?
3406 : *
3407 : * During subcommit, mark it as belonging to the parent, instead, as long
3408 : * as it has not been dropped. Otherwise simply delete the relcache entry.
3409 : * --- it isn't interesting any longer.
3410 : */
3411 8432 : if (relation->rd_createSubid == mySubid)
3412 : {
3413 : /*
3414 : * Valid rd_droppedSubid means the corresponding relation is dropped
3415 : * but the relcache entry is preserved for at-commit pending sync. We
3416 : * need to drop it explicitly here not to make the entry orphan.
3417 : */
3418 : Assert(relation->rd_droppedSubid == mySubid ||
3419 : relation->rd_droppedSubid == InvalidSubTransactionId);
3420 198 : if (isCommit && relation->rd_droppedSubid == InvalidSubTransactionId)
3421 74 : relation->rd_createSubid = parentSubid;
3422 124 : else if (RelationHasReferenceCountZero(relation))
3423 : {
3424 : /* allow the entry to be removed */
3425 124 : relation->rd_createSubid = InvalidSubTransactionId;
3426 124 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3427 124 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3428 124 : relation->rd_droppedSubid = InvalidSubTransactionId;
3429 124 : RelationClearRelation(relation, false);
3430 124 : return;
3431 : }
3432 : else
3433 : {
3434 : /*
3435 : * Hmm, somewhere there's a (leaked?) reference to the relation.
3436 : * We daren't remove the entry for fear of dereferencing a
3437 : * dangling pointer later. Bleat, and transfer it to the parent
3438 : * subtransaction so we can try again later. This must be just a
3439 : * WARNING to avoid error-during-error-recovery loops.
3440 : */
3441 0 : relation->rd_createSubid = parentSubid;
3442 0 : elog(WARNING, "cannot remove relcache entry for \"%s\" because it has nonzero refcount",
3443 : RelationGetRelationName(relation));
3444 : }
3445 : }
3446 :
3447 : /*
3448 : * Likewise, update or drop any new-relfilenumber-in-subtransaction record
3449 : * or drop record.
3450 : */
3451 8308 : if (relation->rd_newRelfilelocatorSubid == mySubid)
3452 : {
3453 140 : if (isCommit)
3454 72 : relation->rd_newRelfilelocatorSubid = parentSubid;
3455 : else
3456 68 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3457 : }
3458 :
3459 8308 : if (relation->rd_firstRelfilelocatorSubid == mySubid)
3460 : {
3461 96 : if (isCommit)
3462 32 : relation->rd_firstRelfilelocatorSubid = parentSubid;
3463 : else
3464 64 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3465 : }
3466 :
3467 8308 : if (relation->rd_droppedSubid == mySubid)
3468 : {
3469 20 : if (isCommit)
3470 2 : relation->rd_droppedSubid = parentSubid;
3471 : else
3472 18 : relation->rd_droppedSubid = InvalidSubTransactionId;
3473 : }
3474 : }
3475 :
3476 :
3477 : /*
3478 : * RelationBuildLocalRelation
3479 : * Build a relcache entry for an about-to-be-created relation,
3480 : * and enter it into the relcache.
3481 : */
3482 : Relation
3483 118818 : RelationBuildLocalRelation(const char *relname,
3484 : Oid relnamespace,
3485 : TupleDesc tupDesc,
3486 : Oid relid,
3487 : Oid accessmtd,
3488 : RelFileNumber relfilenumber,
3489 : Oid reltablespace,
3490 : bool shared_relation,
3491 : bool mapped_relation,
3492 : char relpersistence,
3493 : char relkind)
3494 : {
3495 : Relation rel;
3496 : MemoryContext oldcxt;
3497 118818 : int natts = tupDesc->natts;
3498 : int i;
3499 : bool has_not_null;
3500 : bool nailit;
3501 :
3502 : Assert(natts >= 0);
3503 :
3504 : /*
3505 : * check for creation of a rel that must be nailed in cache.
3506 : *
3507 : * XXX this list had better match the relations specially handled in
3508 : * RelationCacheInitializePhase2/3.
3509 : */
3510 118818 : switch (relid)
3511 : {
3512 560 : case DatabaseRelationId:
3513 : case AuthIdRelationId:
3514 : case AuthMemRelationId:
3515 : case RelationRelationId:
3516 : case AttributeRelationId:
3517 : case ProcedureRelationId:
3518 : case TypeRelationId:
3519 560 : nailit = true;
3520 560 : break;
3521 118258 : default:
3522 118258 : nailit = false;
3523 118258 : break;
3524 : }
3525 :
3526 : /*
3527 : * check that hardwired list of shared rels matches what's in the
3528 : * bootstrap .bki file. If you get a failure here during initdb, you
3529 : * probably need to fix IsSharedRelation() to match whatever you've done
3530 : * to the set of shared relations.
3531 : */
3532 118818 : if (shared_relation != IsSharedRelation(relid))
3533 0 : elog(ERROR, "shared_relation flag for \"%s\" does not match IsSharedRelation(%u)",
3534 : relname, relid);
3535 :
3536 : /* Shared relations had better be mapped, too */
3537 : Assert(mapped_relation || !shared_relation);
3538 :
3539 : /*
3540 : * switch to the cache context to create the relcache entry.
3541 : */
3542 118818 : if (!CacheMemoryContext)
3543 0 : CreateCacheMemoryContext();
3544 :
3545 118818 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
3546 :
3547 : /*
3548 : * allocate a new relation descriptor and fill in basic state fields.
3549 : */
3550 118818 : rel = (Relation) palloc0(sizeof(RelationData));
3551 :
3552 : /* make sure relation is marked as having no open file yet */
3553 118818 : rel->rd_smgr = NULL;
3554 :
3555 : /* mark it nailed if appropriate */
3556 118818 : rel->rd_isnailed = nailit;
3557 :
3558 118818 : rel->rd_refcnt = nailit ? 1 : 0;
3559 :
3560 : /* it's being created in this transaction */
3561 118818 : rel->rd_createSubid = GetCurrentSubTransactionId();
3562 118818 : rel->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3563 118818 : rel->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3564 118818 : rel->rd_droppedSubid = InvalidSubTransactionId;
3565 :
3566 : /*
3567 : * create a new tuple descriptor from the one passed in. We do this
3568 : * partly to copy it into the cache context, and partly because the new
3569 : * relation can't have any defaults or constraints yet; they have to be
3570 : * added in later steps, because they require additions to multiple system
3571 : * catalogs. We can copy attnotnull constraints here, however.
3572 : */
3573 118818 : rel->rd_att = CreateTupleDescCopy(tupDesc);
3574 118818 : rel->rd_att->tdrefcount = 1; /* mark as refcounted */
3575 118818 : has_not_null = false;
3576 505518 : for (i = 0; i < natts; i++)
3577 : {
3578 386700 : Form_pg_attribute satt = TupleDescAttr(tupDesc, i);
3579 386700 : Form_pg_attribute datt = TupleDescAttr(rel->rd_att, i);
3580 :
3581 386700 : datt->attidentity = satt->attidentity;
3582 386700 : datt->attgenerated = satt->attgenerated;
3583 386700 : datt->attnotnull = satt->attnotnull;
3584 386700 : has_not_null |= satt->attnotnull;
3585 : }
3586 :
3587 118818 : if (has_not_null)
3588 : {
3589 17852 : TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
3590 :
3591 17852 : constr->has_not_null = true;
3592 17852 : rel->rd_att->constr = constr;
3593 : }
3594 :
3595 : /*
3596 : * initialize relation tuple form (caller may add/override data later)
3597 : */
3598 118818 : rel->rd_rel = (Form_pg_class) palloc0(CLASS_TUPLE_SIZE);
3599 :
3600 118818 : namestrcpy(&rel->rd_rel->relname, relname);
3601 118818 : rel->rd_rel->relnamespace = relnamespace;
3602 :
3603 118818 : rel->rd_rel->relkind = relkind;
3604 118818 : rel->rd_rel->relnatts = natts;
3605 118818 : rel->rd_rel->reltype = InvalidOid;
3606 : /* needed when bootstrapping: */
3607 118818 : rel->rd_rel->relowner = BOOTSTRAP_SUPERUSERID;
3608 :
3609 : /* set up persistence and relcache fields dependent on it */
3610 118818 : rel->rd_rel->relpersistence = relpersistence;
3611 118818 : switch (relpersistence)
3612 : {
3613 113054 : case RELPERSISTENCE_UNLOGGED:
3614 : case RELPERSISTENCE_PERMANENT:
3615 113054 : rel->rd_backend = INVALID_PROC_NUMBER;
3616 113054 : rel->rd_islocaltemp = false;
3617 113054 : break;
3618 5764 : case RELPERSISTENCE_TEMP:
3619 : Assert(isTempOrTempToastNamespace(relnamespace));
3620 5764 : rel->rd_backend = ProcNumberForTempRelations();
3621 5764 : rel->rd_islocaltemp = true;
3622 5764 : break;
3623 0 : default:
3624 0 : elog(ERROR, "invalid relpersistence: %c", relpersistence);
3625 : break;
3626 : }
3627 :
3628 : /* if it's a materialized view, it's not populated initially */
3629 118818 : if (relkind == RELKIND_MATVIEW)
3630 438 : rel->rd_rel->relispopulated = false;
3631 : else
3632 118380 : rel->rd_rel->relispopulated = true;
3633 :
3634 : /* set replica identity -- system catalogs and non-tables don't have one */
3635 118818 : if (!IsCatalogNamespace(relnamespace) &&
3636 64786 : (relkind == RELKIND_RELATION ||
3637 64348 : relkind == RELKIND_MATVIEW ||
3638 : relkind == RELKIND_PARTITIONED_TABLE))
3639 38160 : rel->rd_rel->relreplident = REPLICA_IDENTITY_DEFAULT;
3640 : else
3641 80658 : rel->rd_rel->relreplident = REPLICA_IDENTITY_NOTHING;
3642 :
3643 : /*
3644 : * Insert relation physical and logical identifiers (OIDs) into the right
3645 : * places. For a mapped relation, we set relfilenumber to zero and rely
3646 : * on RelationInitPhysicalAddr to consult the map.
3647 : */
3648 118818 : rel->rd_rel->relisshared = shared_relation;
3649 :
3650 118818 : RelationGetRelid(rel) = relid;
3651 :
3652 505518 : for (i = 0; i < natts; i++)
3653 386700 : TupleDescAttr(rel->rd_att, i)->attrelid = relid;
3654 :
3655 118818 : rel->rd_rel->reltablespace = reltablespace;
3656 :
3657 118818 : if (mapped_relation)
3658 : {
3659 5522 : rel->rd_rel->relfilenode = InvalidRelFileNumber;
3660 : /* Add it to the active mapping information */
3661 5522 : RelationMapUpdateMap(relid, relfilenumber, shared_relation, true);
3662 : }
3663 : else
3664 113296 : rel->rd_rel->relfilenode = relfilenumber;
3665 :
3666 118818 : RelationInitLockInfo(rel); /* see lmgr.c */
3667 :
3668 118818 : RelationInitPhysicalAddr(rel);
3669 :
3670 118818 : rel->rd_rel->relam = accessmtd;
3671 :
3672 : /*
3673 : * RelationInitTableAccessMethod will do syscache lookups, so we mustn't
3674 : * run it in CacheMemoryContext. Fortunately, the remaining steps don't
3675 : * require a long-lived current context.
3676 : */
3677 118818 : MemoryContextSwitchTo(oldcxt);
3678 :
3679 118818 : if (RELKIND_HAS_TABLE_AM(relkind) || relkind == RELKIND_SEQUENCE)
3680 56294 : RelationInitTableAccessMethod(rel);
3681 :
3682 : /*
3683 : * Okay to insert into the relcache hash table.
3684 : *
3685 : * Ordinarily, there should certainly not be an existing hash entry for
3686 : * the same OID; but during bootstrap, when we create a "real" relcache
3687 : * entry for one of the bootstrap relations, we'll be overwriting the
3688 : * phony one created with formrdesc. So allow that to happen for nailed
3689 : * rels.
3690 : */
3691 118818 : RelationCacheInsert(rel, nailit);
3692 :
3693 : /*
3694 : * Flag relation as needing eoxact cleanup (to clear rd_createSubid). We
3695 : * can't do this before storing relid in it.
3696 : */
3697 118818 : EOXactListAdd(rel);
3698 :
3699 : /* It's fully valid */
3700 118818 : rel->rd_isvalid = true;
3701 :
3702 : /*
3703 : * Caller expects us to pin the returned entry.
3704 : */
3705 118818 : RelationIncrementReferenceCount(rel);
3706 :
3707 118818 : return rel;
3708 : }
3709 :
3710 :
3711 : /*
3712 : * RelationSetNewRelfilenumber
3713 : *
3714 : * Assign a new relfilenumber (physical file name), and possibly a new
3715 : * persistence setting, to the relation.
3716 : *
3717 : * This allows a full rewrite of the relation to be done with transactional
3718 : * safety (since the filenumber assignment can be rolled back). Note however
3719 : * that there is no simple way to access the relation's old data for the
3720 : * remainder of the current transaction. This limits the usefulness to cases
3721 : * such as TRUNCATE or rebuilding an index from scratch.
3722 : *
3723 : * Caller must already hold exclusive lock on the relation.
3724 : */
3725 : void
3726 11622 : RelationSetNewRelfilenumber(Relation relation, char persistence)
3727 : {
3728 : RelFileNumber newrelfilenumber;
3729 : Relation pg_class;
3730 : HeapTuple tuple;
3731 : Form_pg_class classform;
3732 11622 : MultiXactId minmulti = InvalidMultiXactId;
3733 11622 : TransactionId freezeXid = InvalidTransactionId;
3734 : RelFileLocator newrlocator;
3735 :
3736 11622 : if (!IsBinaryUpgrade)
3737 : {
3738 : /* Allocate a new relfilenumber */
3739 11582 : newrelfilenumber = GetNewRelFileNumber(relation->rd_rel->reltablespace,
3740 : NULL, persistence);
3741 : }
3742 40 : else if (relation->rd_rel->relkind == RELKIND_INDEX)
3743 : {
3744 20 : if (!OidIsValid(binary_upgrade_next_index_pg_class_relfilenumber))
3745 0 : ereport(ERROR,
3746 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3747 : errmsg("index relfilenumber value not set when in binary upgrade mode")));
3748 :
3749 20 : newrelfilenumber = binary_upgrade_next_index_pg_class_relfilenumber;
3750 20 : binary_upgrade_next_index_pg_class_relfilenumber = InvalidOid;
3751 : }
3752 20 : else if (relation->rd_rel->relkind == RELKIND_RELATION)
3753 : {
3754 20 : if (!OidIsValid(binary_upgrade_next_heap_pg_class_relfilenumber))
3755 0 : ereport(ERROR,
3756 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3757 : errmsg("heap relfilenumber value not set when in binary upgrade mode")));
3758 :
3759 20 : newrelfilenumber = binary_upgrade_next_heap_pg_class_relfilenumber;
3760 20 : binary_upgrade_next_heap_pg_class_relfilenumber = InvalidOid;
3761 : }
3762 : else
3763 0 : ereport(ERROR,
3764 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3765 : errmsg("unexpected request for new relfilenumber in binary upgrade mode")));
3766 :
3767 : /*
3768 : * Get a writable copy of the pg_class tuple for the given relation.
3769 : */
3770 11622 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
3771 :
3772 11622 : tuple = SearchSysCacheCopy1(RELOID,
3773 : ObjectIdGetDatum(RelationGetRelid(relation)));
3774 11622 : if (!HeapTupleIsValid(tuple))
3775 0 : elog(ERROR, "could not find tuple for relation %u",
3776 : RelationGetRelid(relation));
3777 11622 : classform = (Form_pg_class) GETSTRUCT(tuple);
3778 :
3779 : /*
3780 : * Schedule unlinking of the old storage at transaction commit, except
3781 : * when performing a binary upgrade, when we must do it immediately.
3782 : */
3783 11622 : if (IsBinaryUpgrade)
3784 : {
3785 : SMgrRelation srel;
3786 :
3787 : /*
3788 : * During a binary upgrade, we use this code path to ensure that
3789 : * pg_largeobject and its index have the same relfilenumbers as in the
3790 : * old cluster. This is necessary because pg_upgrade treats
3791 : * pg_largeobject like a user table, not a system table. It is however
3792 : * possible that a table or index may need to end up with the same
3793 : * relfilenumber in the new cluster as what it had in the old cluster.
3794 : * Hence, we can't wait until commit time to remove the old storage.
3795 : *
3796 : * In general, this function needs to have transactional semantics,
3797 : * and removing the old storage before commit time surely isn't.
3798 : * However, it doesn't really matter, because if a binary upgrade
3799 : * fails at this stage, the new cluster will need to be recreated
3800 : * anyway.
3801 : */
3802 40 : srel = smgropen(relation->rd_locator, relation->rd_backend);
3803 40 : smgrdounlinkall(&srel, 1, false);
3804 40 : smgrclose(srel);
3805 : }
3806 : else
3807 : {
3808 : /* Not a binary upgrade, so just schedule it to happen later. */
3809 11582 : RelationDropStorage(relation);
3810 : }
3811 :
3812 : /*
3813 : * Create storage for the main fork of the new relfilenumber. If it's a
3814 : * table-like object, call into the table AM to do so, which'll also
3815 : * create the table's init fork if needed.
3816 : *
3817 : * NOTE: If relevant for the AM, any conflict in relfilenumber value will
3818 : * be caught here, if GetNewRelFileNumber messes up for any reason.
3819 : */
3820 11622 : newrlocator = relation->rd_locator;
3821 11622 : newrlocator.relNumber = newrelfilenumber;
3822 :
3823 11622 : if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind))
3824 : {
3825 4194 : table_relation_set_new_filelocator(relation, &newrlocator,
3826 : persistence,
3827 : &freezeXid, &minmulti);
3828 : }
3829 7428 : else if (RELKIND_HAS_STORAGE(relation->rd_rel->relkind))
3830 7428 : {
3831 : /* handle these directly, at least for now */
3832 : SMgrRelation srel;
3833 :
3834 7428 : srel = RelationCreateStorage(newrlocator, persistence, true);
3835 7428 : smgrclose(srel);
3836 : }
3837 : else
3838 : {
3839 : /* we shouldn't be called for anything else */
3840 0 : elog(ERROR, "relation \"%s\" does not have storage",
3841 : RelationGetRelationName(relation));
3842 : }
3843 :
3844 : /*
3845 : * If we're dealing with a mapped index, pg_class.relfilenode doesn't
3846 : * change; instead we have to send the update to the relation mapper.
3847 : *
3848 : * For mapped indexes, we don't actually change the pg_class entry at all;
3849 : * this is essential when reindexing pg_class itself. That leaves us with
3850 : * possibly-inaccurate values of relpages etc, but those will be fixed up
3851 : * later.
3852 : */
3853 11622 : if (RelationIsMapped(relation))
3854 : {
3855 : /* This case is only supported for indexes */
3856 : Assert(relation->rd_rel->relkind == RELKIND_INDEX);
3857 :
3858 : /* Since we're not updating pg_class, these had better not change */
3859 : Assert(classform->relfrozenxid == freezeXid);
3860 : Assert(classform->relminmxid == minmulti);
3861 : Assert(classform->relpersistence == persistence);
3862 :
3863 : /*
3864 : * In some code paths it's possible that the tuple update we'd
3865 : * otherwise do here is the only thing that would assign an XID for
3866 : * the current transaction. However, we must have an XID to delete
3867 : * files, so make sure one is assigned.
3868 : */
3869 850 : (void) GetCurrentTransactionId();
3870 :
3871 : /* Do the deed */
3872 850 : RelationMapUpdateMap(RelationGetRelid(relation),
3873 : newrelfilenumber,
3874 850 : relation->rd_rel->relisshared,
3875 : false);
3876 :
3877 : /* Since we're not updating pg_class, must trigger inval manually */
3878 850 : CacheInvalidateRelcache(relation);
3879 : }
3880 : else
3881 : {
3882 : /* Normal case, update the pg_class entry */
3883 10772 : classform->relfilenode = newrelfilenumber;
3884 :
3885 : /* relpages etc. never change for sequences */
3886 10772 : if (relation->rd_rel->relkind != RELKIND_SEQUENCE)
3887 : {
3888 10514 : classform->relpages = 0; /* it's empty until further notice */
3889 10514 : classform->reltuples = -1;
3890 10514 : classform->relallvisible = 0;
3891 : }
3892 10772 : classform->relfrozenxid = freezeXid;
3893 10772 : classform->relminmxid = minmulti;
3894 10772 : classform->relpersistence = persistence;
3895 :
3896 10772 : CatalogTupleUpdate(pg_class, &tuple->t_self, tuple);
3897 : }
3898 :
3899 11622 : heap_freetuple(tuple);
3900 :
3901 11622 : table_close(pg_class, RowExclusiveLock);
3902 :
3903 : /*
3904 : * Make the pg_class row change or relation map change visible. This will
3905 : * cause the relcache entry to get updated, too.
3906 : */
3907 11622 : CommandCounterIncrement();
3908 :
3909 11622 : RelationAssumeNewRelfilelocator(relation);
3910 11622 : }
3911 :
3912 : /*
3913 : * RelationAssumeNewRelfilelocator
3914 : *
3915 : * Code that modifies pg_class.reltablespace or pg_class.relfilenode must call
3916 : * this. The call shall precede any code that might insert WAL records whose
3917 : * replay would modify bytes in the new RelFileLocator, and the call shall follow
3918 : * any WAL modifying bytes in the prior RelFileLocator. See struct RelationData.
3919 : * Ideally, call this as near as possible to the CommandCounterIncrement()
3920 : * that makes the pg_class change visible (before it or after it); that
3921 : * minimizes the chance of future development adding a forbidden WAL insertion
3922 : * between RelationAssumeNewRelfilelocator() and CommandCounterIncrement().
3923 : */
3924 : void
3925 13694 : RelationAssumeNewRelfilelocator(Relation relation)
3926 : {
3927 13694 : relation->rd_newRelfilelocatorSubid = GetCurrentSubTransactionId();
3928 13694 : if (relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId)
3929 13606 : relation->rd_firstRelfilelocatorSubid = relation->rd_newRelfilelocatorSubid;
3930 :
3931 : /* Flag relation as needing eoxact cleanup (to clear these fields) */
3932 13694 : EOXactListAdd(relation);
3933 13694 : }
3934 :
3935 :
3936 : /*
3937 : * RelationCacheInitialize
3938 : *
3939 : * This initializes the relation descriptor cache. At the time
3940 : * that this is invoked, we can't do database access yet (mainly
3941 : * because the transaction subsystem is not up); all we are doing
3942 : * is making an empty cache hashtable. This must be done before
3943 : * starting the initialization transaction, because otherwise
3944 : * AtEOXact_RelationCache would crash if that transaction aborts
3945 : * before we can get the relcache set up.
3946 : */
3947 :
3948 : #define INITRELCACHESIZE 400
3949 :
3950 : void
3951 25898 : RelationCacheInitialize(void)
3952 : {
3953 : HASHCTL ctl;
3954 : int allocsize;
3955 :
3956 : /*
3957 : * make sure cache memory context exists
3958 : */
3959 25898 : if (!CacheMemoryContext)
3960 25898 : CreateCacheMemoryContext();
3961 :
3962 : /*
3963 : * create hashtable that indexes the relcache
3964 : */
3965 25898 : ctl.keysize = sizeof(Oid);
3966 25898 : ctl.entrysize = sizeof(RelIdCacheEnt);
3967 25898 : RelationIdCache = hash_create("Relcache by OID", INITRELCACHESIZE,
3968 : &ctl, HASH_ELEM | HASH_BLOBS);
3969 :
3970 : /*
3971 : * reserve enough in_progress_list slots for many cases
3972 : */
3973 25898 : allocsize = 4;
3974 25898 : in_progress_list =
3975 25898 : MemoryContextAlloc(CacheMemoryContext,
3976 : allocsize * sizeof(*in_progress_list));
3977 25898 : in_progress_list_maxlen = allocsize;
3978 :
3979 : /*
3980 : * relation mapper needs to be initialized too
3981 : */
3982 25898 : RelationMapInitialize();
3983 25898 : }
3984 :
3985 : /*
3986 : * RelationCacheInitializePhase2
3987 : *
3988 : * This is called to prepare for access to shared catalogs during startup.
3989 : * We must at least set up nailed reldescs for pg_database, pg_authid,
3990 : * pg_auth_members, and pg_shseclabel. Ideally we'd like to have reldescs
3991 : * for their indexes, too. We attempt to load this information from the
3992 : * shared relcache init file. If that's missing or broken, just make
3993 : * phony entries for the catalogs themselves.
3994 : * RelationCacheInitializePhase3 will clean up as needed.
3995 : */
3996 : void
3997 25898 : RelationCacheInitializePhase2(void)
3998 : {
3999 : MemoryContext oldcxt;
4000 :
4001 : /*
4002 : * relation mapper needs initialized too
4003 : */
4004 25898 : RelationMapInitializePhase2();
4005 :
4006 : /*
4007 : * In bootstrap mode, the shared catalogs aren't there yet anyway, so do
4008 : * nothing.
4009 : */
4010 25898 : if (IsBootstrapProcessingMode())
4011 80 : return;
4012 :
4013 : /*
4014 : * switch to cache memory context
4015 : */
4016 25818 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4017 :
4018 : /*
4019 : * Try to load the shared relcache cache file. If unsuccessful, bootstrap
4020 : * the cache with pre-made descriptors for the critical shared catalogs.
4021 : */
4022 25818 : if (!load_relcache_init_file(true))
4023 : {
4024 3200 : formrdesc("pg_database", DatabaseRelation_Rowtype_Id, true,
4025 : Natts_pg_database, Desc_pg_database);
4026 3200 : formrdesc("pg_authid", AuthIdRelation_Rowtype_Id, true,
4027 : Natts_pg_authid, Desc_pg_authid);
4028 3200 : formrdesc("pg_auth_members", AuthMemRelation_Rowtype_Id, true,
4029 : Natts_pg_auth_members, Desc_pg_auth_members);
4030 3200 : formrdesc("pg_shseclabel", SharedSecLabelRelation_Rowtype_Id, true,
4031 : Natts_pg_shseclabel, Desc_pg_shseclabel);
4032 3200 : formrdesc("pg_subscription", SubscriptionRelation_Rowtype_Id, true,
4033 : Natts_pg_subscription, Desc_pg_subscription);
4034 :
4035 : #define NUM_CRITICAL_SHARED_RELS 5 /* fix if you change list above */
4036 : }
4037 :
4038 25818 : MemoryContextSwitchTo(oldcxt);
4039 : }
4040 :
4041 : /*
4042 : * RelationCacheInitializePhase3
4043 : *
4044 : * This is called as soon as the catcache and transaction system
4045 : * are functional and we have determined MyDatabaseId. At this point
4046 : * we can actually read data from the database's system catalogs.
4047 : * We first try to read pre-computed relcache entries from the local
4048 : * relcache init file. If that's missing or broken, make phony entries
4049 : * for the minimum set of nailed-in-cache relations. Then (unless
4050 : * bootstrapping) make sure we have entries for the critical system
4051 : * indexes. Once we've done all this, we have enough infrastructure to
4052 : * open any system catalog or use any catcache. The last step is to
4053 : * rewrite the cache files if needed.
4054 : */
4055 : void
4056 23588 : RelationCacheInitializePhase3(void)
4057 : {
4058 : HASH_SEQ_STATUS status;
4059 : RelIdCacheEnt *idhentry;
4060 : MemoryContext oldcxt;
4061 23588 : bool needNewCacheFile = !criticalSharedRelcachesBuilt;
4062 :
4063 : /*
4064 : * relation mapper needs initialized too
4065 : */
4066 23588 : RelationMapInitializePhase3();
4067 :
4068 : /*
4069 : * switch to cache memory context
4070 : */
4071 23588 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4072 :
4073 : /*
4074 : * Try to load the local relcache cache file. If unsuccessful, bootstrap
4075 : * the cache with pre-made descriptors for the critical "nailed-in" system
4076 : * catalogs.
4077 : */
4078 23588 : if (IsBootstrapProcessingMode() ||
4079 23508 : !load_relcache_init_file(false))
4080 : {
4081 2188 : needNewCacheFile = true;
4082 :
4083 2188 : formrdesc("pg_class", RelationRelation_Rowtype_Id, false,
4084 : Natts_pg_class, Desc_pg_class);
4085 2188 : formrdesc("pg_attribute", AttributeRelation_Rowtype_Id, false,
4086 : Natts_pg_attribute, Desc_pg_attribute);
4087 2188 : formrdesc("pg_proc", ProcedureRelation_Rowtype_Id, false,
4088 : Natts_pg_proc, Desc_pg_proc);
4089 2188 : formrdesc("pg_type", TypeRelation_Rowtype_Id, false,
4090 : Natts_pg_type, Desc_pg_type);
4091 :
4092 : #define NUM_CRITICAL_LOCAL_RELS 4 /* fix if you change list above */
4093 : }
4094 :
4095 23588 : MemoryContextSwitchTo(oldcxt);
4096 :
4097 : /* In bootstrap mode, the faked-up formrdesc info is all we'll have */
4098 23588 : if (IsBootstrapProcessingMode())
4099 80 : return;
4100 :
4101 : /*
4102 : * If we didn't get the critical system indexes loaded into relcache, do
4103 : * so now. These are critical because the catcache and/or opclass cache
4104 : * depend on them for fetches done during relcache load. Thus, we have an
4105 : * infinite-recursion problem. We can break the recursion by doing
4106 : * heapscans instead of indexscans at certain key spots. To avoid hobbling
4107 : * performance, we only want to do that until we have the critical indexes
4108 : * loaded into relcache. Thus, the flag criticalRelcachesBuilt is used to
4109 : * decide whether to do heapscan or indexscan at the key spots, and we set
4110 : * it true after we've loaded the critical indexes.
4111 : *
4112 : * The critical indexes are marked as "nailed in cache", partly to make it
4113 : * easy for load_relcache_init_file to count them, but mainly because we
4114 : * cannot flush and rebuild them once we've set criticalRelcachesBuilt to
4115 : * true. (NOTE: perhaps it would be possible to reload them by
4116 : * temporarily setting criticalRelcachesBuilt to false again. For now,
4117 : * though, we just nail 'em in.)
4118 : *
4119 : * RewriteRelRulenameIndexId and TriggerRelidNameIndexId are not critical
4120 : * in the same way as the others, because the critical catalogs don't
4121 : * (currently) have any rules or triggers, and so these indexes can be
4122 : * rebuilt without inducing recursion. However they are used during
4123 : * relcache load when a rel does have rules or triggers, so we choose to
4124 : * nail them for performance reasons.
4125 : */
4126 23508 : if (!criticalRelcachesBuilt)
4127 : {
4128 2108 : load_critical_index(ClassOidIndexId,
4129 : RelationRelationId);
4130 2108 : load_critical_index(AttributeRelidNumIndexId,
4131 : AttributeRelationId);
4132 2108 : load_critical_index(IndexRelidIndexId,
4133 : IndexRelationId);
4134 2108 : load_critical_index(OpclassOidIndexId,
4135 : OperatorClassRelationId);
4136 2108 : load_critical_index(AccessMethodProcedureIndexId,
4137 : AccessMethodProcedureRelationId);
4138 2108 : load_critical_index(RewriteRelRulenameIndexId,
4139 : RewriteRelationId);
4140 2108 : load_critical_index(TriggerRelidNameIndexId,
4141 : TriggerRelationId);
4142 :
4143 : #define NUM_CRITICAL_LOCAL_INDEXES 7 /* fix if you change list above */
4144 :
4145 2108 : criticalRelcachesBuilt = true;
4146 : }
4147 :
4148 : /*
4149 : * Process critical shared indexes too.
4150 : *
4151 : * DatabaseNameIndexId isn't critical for relcache loading, but rather for
4152 : * initial lookup of MyDatabaseId, without which we'll never find any
4153 : * non-shared catalogs at all. Autovacuum calls InitPostgres with a
4154 : * database OID, so it instead depends on DatabaseOidIndexId. We also
4155 : * need to nail up some indexes on pg_authid and pg_auth_members for use
4156 : * during client authentication. SharedSecLabelObjectIndexId isn't
4157 : * critical for the core system, but authentication hooks might be
4158 : * interested in it.
4159 : */
4160 23508 : if (!criticalSharedRelcachesBuilt)
4161 : {
4162 1586 : load_critical_index(DatabaseNameIndexId,
4163 : DatabaseRelationId);
4164 1586 : load_critical_index(DatabaseOidIndexId,
4165 : DatabaseRelationId);
4166 1586 : load_critical_index(AuthIdRolnameIndexId,
4167 : AuthIdRelationId);
4168 1586 : load_critical_index(AuthIdOidIndexId,
4169 : AuthIdRelationId);
4170 1586 : load_critical_index(AuthMemMemRoleIndexId,
4171 : AuthMemRelationId);
4172 1586 : load_critical_index(SharedSecLabelObjectIndexId,
4173 : SharedSecLabelRelationId);
4174 :
4175 : #define NUM_CRITICAL_SHARED_INDEXES 6 /* fix if you change list above */
4176 :
4177 1586 : criticalSharedRelcachesBuilt = true;
4178 : }
4179 :
4180 : /*
4181 : * Now, scan all the relcache entries and update anything that might be
4182 : * wrong in the results from formrdesc or the relcache cache file. If we
4183 : * faked up relcache entries using formrdesc, then read the real pg_class
4184 : * rows and replace the fake entries with them. Also, if any of the
4185 : * relcache entries have rules, triggers, or security policies, load that
4186 : * info the hard way since it isn't recorded in the cache file.
4187 : *
4188 : * Whenever we access the catalogs to read data, there is a possibility of
4189 : * a shared-inval cache flush causing relcache entries to be removed.
4190 : * Since hash_seq_search only guarantees to still work after the *current*
4191 : * entry is removed, it's unsafe to continue the hashtable scan afterward.
4192 : * We handle this by restarting the scan from scratch after each access.
4193 : * This is theoretically O(N^2), but the number of entries that actually
4194 : * need to be fixed is small enough that it doesn't matter.
4195 : */
4196 23508 : hash_seq_init(&status, RelationIdCache);
4197 :
4198 3237408 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
4199 : {
4200 3213900 : Relation relation = idhentry->reldesc;
4201 3213900 : bool restart = false;
4202 :
4203 : /*
4204 : * Make sure *this* entry doesn't get flushed while we work with it.
4205 : */
4206 3213900 : RelationIncrementReferenceCount(relation);
4207 :
4208 : /*
4209 : * If it's a faked-up entry, read the real pg_class tuple.
4210 : */
4211 3213900 : if (relation->rd_rel->relowner == InvalidOid)
4212 : {
4213 : HeapTuple htup;
4214 : Form_pg_class relp;
4215 :
4216 16358 : htup = SearchSysCache1(RELOID,
4217 : ObjectIdGetDatum(RelationGetRelid(relation)));
4218 16358 : if (!HeapTupleIsValid(htup))
4219 0 : ereport(FATAL,
4220 : errcode(ERRCODE_UNDEFINED_OBJECT),
4221 : errmsg_internal("cache lookup failed for relation %u",
4222 : RelationGetRelid(relation)));
4223 16358 : relp = (Form_pg_class) GETSTRUCT(htup);
4224 :
4225 : /*
4226 : * Copy tuple to relation->rd_rel. (See notes in
4227 : * AllocateRelationDesc())
4228 : */
4229 16358 : memcpy((char *) relation->rd_rel, (char *) relp, CLASS_TUPLE_SIZE);
4230 :
4231 : /* Update rd_options while we have the tuple */
4232 16358 : if (relation->rd_options)
4233 0 : pfree(relation->rd_options);
4234 16358 : RelationParseRelOptions(relation, htup);
4235 :
4236 : /*
4237 : * Check the values in rd_att were set up correctly. (We cannot
4238 : * just copy them over now: formrdesc must have set up the rd_att
4239 : * data correctly to start with, because it may already have been
4240 : * copied into one or more catcache entries.)
4241 : */
4242 : Assert(relation->rd_att->tdtypeid == relp->reltype);
4243 : Assert(relation->rd_att->tdtypmod == -1);
4244 :
4245 16358 : ReleaseSysCache(htup);
4246 :
4247 : /* relowner had better be OK now, else we'll loop forever */
4248 16358 : if (relation->rd_rel->relowner == InvalidOid)
4249 0 : elog(ERROR, "invalid relowner in pg_class entry for \"%s\"",
4250 : RelationGetRelationName(relation));
4251 :
4252 16358 : restart = true;
4253 : }
4254 :
4255 : /*
4256 : * Fix data that isn't saved in relcache cache file.
4257 : *
4258 : * relhasrules or relhastriggers could possibly be wrong or out of
4259 : * date. If we don't actually find any rules or triggers, clear the
4260 : * local copy of the flag so that we don't get into an infinite loop
4261 : * here. We don't make any attempt to fix the pg_class entry, though.
4262 : */
4263 3213900 : if (relation->rd_rel->relhasrules && relation->rd_rules == NULL)
4264 : {
4265 0 : RelationBuildRuleLock(relation);
4266 0 : if (relation->rd_rules == NULL)
4267 0 : relation->rd_rel->relhasrules = false;
4268 0 : restart = true;
4269 : }
4270 3213900 : if (relation->rd_rel->relhastriggers && relation->trigdesc == NULL)
4271 : {
4272 0 : RelationBuildTriggers(relation);
4273 0 : if (relation->trigdesc == NULL)
4274 0 : relation->rd_rel->relhastriggers = false;
4275 0 : restart = true;
4276 : }
4277 :
4278 : /*
4279 : * Re-load the row security policies if the relation has them, since
4280 : * they are not preserved in the cache. Note that we can never NOT
4281 : * have a policy while relrowsecurity is true,
4282 : * RelationBuildRowSecurity will create a single default-deny policy
4283 : * if there is no policy defined in pg_policy.
4284 : */
4285 3213900 : if (relation->rd_rel->relrowsecurity && relation->rd_rsdesc == NULL)
4286 : {
4287 0 : RelationBuildRowSecurity(relation);
4288 :
4289 : Assert(relation->rd_rsdesc != NULL);
4290 0 : restart = true;
4291 : }
4292 :
4293 : /* Reload tableam data if needed */
4294 3213900 : if (relation->rd_tableam == NULL &&
4295 1970336 : (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) || relation->rd_rel->relkind == RELKIND_SEQUENCE))
4296 : {
4297 0 : RelationInitTableAccessMethod(relation);
4298 : Assert(relation->rd_tableam != NULL);
4299 :
4300 0 : restart = true;
4301 : }
4302 :
4303 : /* Release hold on the relation */
4304 3213900 : RelationDecrementReferenceCount(relation);
4305 :
4306 : /* Now, restart the hashtable scan if needed */
4307 3213900 : if (restart)
4308 : {
4309 16358 : hash_seq_term(&status);
4310 16358 : hash_seq_init(&status, RelationIdCache);
4311 : }
4312 : }
4313 :
4314 : /*
4315 : * Lastly, write out new relcache cache files if needed. We don't bother
4316 : * to distinguish cases where only one of the two needs an update.
4317 : */
4318 23508 : if (needNewCacheFile)
4319 : {
4320 : /*
4321 : * Force all the catcaches to finish initializing and thereby open the
4322 : * catalogs and indexes they use. This will preload the relcache with
4323 : * entries for all the most important system catalogs and indexes, so
4324 : * that the init files will be most useful for future backends.
4325 : */
4326 2156 : InitCatalogCachePhase2();
4327 :
4328 : /* now write the files */
4329 2152 : write_relcache_init_file(true);
4330 2152 : write_relcache_init_file(false);
4331 : }
4332 : }
4333 :
4334 : /*
4335 : * Load one critical system index into the relcache
4336 : *
4337 : * indexoid is the OID of the target index, heapoid is the OID of the catalog
4338 : * it belongs to.
4339 : */
4340 : static void
4341 24272 : load_critical_index(Oid indexoid, Oid heapoid)
4342 : {
4343 : Relation ird;
4344 :
4345 : /*
4346 : * We must lock the underlying catalog before locking the index to avoid
4347 : * deadlock, since RelationBuildDesc might well need to read the catalog,
4348 : * and if anyone else is exclusive-locking this catalog and index they'll
4349 : * be doing it in that order.
4350 : */
4351 24272 : LockRelationOid(heapoid, AccessShareLock);
4352 24272 : LockRelationOid(indexoid, AccessShareLock);
4353 24272 : ird = RelationBuildDesc(indexoid, true);
4354 24272 : if (ird == NULL)
4355 0 : ereport(PANIC,
4356 : errcode(ERRCODE_DATA_CORRUPTED),
4357 : errmsg_internal("could not open critical system index %u", indexoid));
4358 24272 : ird->rd_isnailed = true;
4359 24272 : ird->rd_refcnt = 1;
4360 24272 : UnlockRelationOid(indexoid, AccessShareLock);
4361 24272 : UnlockRelationOid(heapoid, AccessShareLock);
4362 :
4363 24272 : (void) RelationGetIndexAttOptions(ird, false);
4364 24272 : }
4365 :
4366 : /*
4367 : * GetPgClassDescriptor -- get a predefined tuple descriptor for pg_class
4368 : * GetPgIndexDescriptor -- get a predefined tuple descriptor for pg_index
4369 : *
4370 : * We need this kluge because we have to be able to access non-fixed-width
4371 : * fields of pg_class and pg_index before we have the standard catalog caches
4372 : * available. We use predefined data that's set up in just the same way as
4373 : * the bootstrapped reldescs used by formrdesc(). The resulting tupdesc is
4374 : * not 100% kosher: it does not have the correct rowtype OID in tdtypeid, nor
4375 : * does it have a TupleConstr field. But it's good enough for the purpose of
4376 : * extracting fields.
4377 : */
4378 : static TupleDesc
4379 47174 : BuildHardcodedDescriptor(int natts, const FormData_pg_attribute *attrs)
4380 : {
4381 : TupleDesc result;
4382 : MemoryContext oldcxt;
4383 : int i;
4384 :
4385 47174 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4386 :
4387 47174 : result = CreateTemplateTupleDesc(natts);
4388 47174 : result->tdtypeid = RECORDOID; /* not right, but we don't care */
4389 47174 : result->tdtypmod = -1;
4390 :
4391 1320884 : for (i = 0; i < natts; i++)
4392 : {
4393 1273710 : memcpy(TupleDescAttr(result, i), &attrs[i], ATTRIBUTE_FIXED_PART_SIZE);
4394 : /* make sure attcacheoff is valid */
4395 1273710 : TupleDescAttr(result, i)->attcacheoff = -1;
4396 : }
4397 :
4398 : /* initialize first attribute's attcacheoff, cf RelationBuildTupleDesc */
4399 47174 : TupleDescAttr(result, 0)->attcacheoff = 0;
4400 :
4401 : /* Note: we don't bother to set up a TupleConstr entry */
4402 :
4403 47174 : MemoryContextSwitchTo(oldcxt);
4404 :
4405 47174 : return result;
4406 : }
4407 :
4408 : static TupleDesc
4409 1138700 : GetPgClassDescriptor(void)
4410 : {
4411 : static TupleDesc pgclassdesc = NULL;
4412 :
4413 : /* Already done? */
4414 1138700 : if (pgclassdesc == NULL)
4415 23588 : pgclassdesc = BuildHardcodedDescriptor(Natts_pg_class,
4416 : Desc_pg_class);
4417 :
4418 1138700 : return pgclassdesc;
4419 : }
4420 :
4421 : static TupleDesc
4422 1317870 : GetPgIndexDescriptor(void)
4423 : {
4424 : static TupleDesc pgindexdesc = NULL;
4425 :
4426 : /* Already done? */
4427 1317870 : if (pgindexdesc == NULL)
4428 23586 : pgindexdesc = BuildHardcodedDescriptor(Natts_pg_index,
4429 : Desc_pg_index);
4430 :
4431 1317870 : return pgindexdesc;
4432 : }
4433 :
4434 : /*
4435 : * Load any default attribute value definitions for the relation.
4436 : *
4437 : * ndef is the number of attributes that were marked atthasdef.
4438 : *
4439 : * Note: we don't make it a hard error to be missing some pg_attrdef records.
4440 : * We can limp along as long as nothing needs to use the default value. Code
4441 : * that fails to find an expected AttrDefault record should throw an error.
4442 : */
4443 : static void
4444 28096 : AttrDefaultFetch(Relation relation, int ndef)
4445 : {
4446 : AttrDefault *attrdef;
4447 : Relation adrel;
4448 : SysScanDesc adscan;
4449 : ScanKeyData skey;
4450 : HeapTuple htup;
4451 28096 : int found = 0;
4452 :
4453 : /* Allocate array with room for as many entries as expected */
4454 : attrdef = (AttrDefault *)
4455 28096 : MemoryContextAllocZero(CacheMemoryContext,
4456 : ndef * sizeof(AttrDefault));
4457 :
4458 : /* Search pg_attrdef for relevant entries */
4459 28096 : ScanKeyInit(&skey,
4460 : Anum_pg_attrdef_adrelid,
4461 : BTEqualStrategyNumber, F_OIDEQ,
4462 : ObjectIdGetDatum(RelationGetRelid(relation)));
4463 :
4464 28096 : adrel = table_open(AttrDefaultRelationId, AccessShareLock);
4465 28096 : adscan = systable_beginscan(adrel, AttrDefaultIndexId, true,
4466 : NULL, 1, &skey);
4467 :
4468 67430 : while (HeapTupleIsValid(htup = systable_getnext(adscan)))
4469 : {
4470 39334 : Form_pg_attrdef adform = (Form_pg_attrdef) GETSTRUCT(htup);
4471 : Datum val;
4472 : bool isnull;
4473 :
4474 : /* protect limited size of array */
4475 39334 : if (found >= ndef)
4476 : {
4477 0 : elog(WARNING, "unexpected pg_attrdef record found for attribute %d of relation \"%s\"",
4478 : adform->adnum, RelationGetRelationName(relation));
4479 0 : break;
4480 : }
4481 :
4482 39334 : val = fastgetattr(htup,
4483 : Anum_pg_attrdef_adbin,
4484 : adrel->rd_att, &isnull);
4485 39334 : if (isnull)
4486 0 : elog(WARNING, "null adbin for attribute %d of relation \"%s\"",
4487 : adform->adnum, RelationGetRelationName(relation));
4488 : else
4489 : {
4490 : /* detoast and convert to cstring in caller's context */
4491 39334 : char *s = TextDatumGetCString(val);
4492 :
4493 39334 : attrdef[found].adnum = adform->adnum;
4494 39334 : attrdef[found].adbin = MemoryContextStrdup(CacheMemoryContext, s);
4495 39334 : pfree(s);
4496 39334 : found++;
4497 : }
4498 : }
4499 :
4500 28096 : systable_endscan(adscan);
4501 28096 : table_close(adrel, AccessShareLock);
4502 :
4503 28096 : if (found != ndef)
4504 0 : elog(WARNING, "%d pg_attrdef record(s) missing for relation \"%s\"",
4505 : ndef - found, RelationGetRelationName(relation));
4506 :
4507 : /*
4508 : * Sort the AttrDefault entries by adnum, for the convenience of
4509 : * equalTupleDescs(). (Usually, they already will be in order, but this
4510 : * might not be so if systable_getnext isn't using an index.)
4511 : */
4512 28096 : if (found > 1)
4513 5804 : qsort(attrdef, found, sizeof(AttrDefault), AttrDefaultCmp);
4514 :
4515 : /* Install array only after it's fully valid */
4516 28096 : relation->rd_att->constr->defval = attrdef;
4517 28096 : relation->rd_att->constr->num_defval = found;
4518 28096 : }
4519 :
4520 : /*
4521 : * qsort comparator to sort AttrDefault entries by adnum
4522 : */
4523 : static int
4524 11238 : AttrDefaultCmp(const void *a, const void *b)
4525 : {
4526 11238 : const AttrDefault *ada = (const AttrDefault *) a;
4527 11238 : const AttrDefault *adb = (const AttrDefault *) b;
4528 :
4529 11238 : return pg_cmp_s16(ada->adnum, adb->adnum);
4530 : }
4531 :
4532 : /*
4533 : * Load any check constraints for the relation.
4534 : *
4535 : * As with defaults, if we don't find the expected number of them, just warn
4536 : * here. The executor should throw an error if an INSERT/UPDATE is attempted.
4537 : */
4538 : static void
4539 11180 : CheckConstraintFetch(Relation relation)
4540 : {
4541 : ConstrCheck *check;
4542 11180 : int ncheck = relation->rd_rel->relchecks;
4543 : Relation conrel;
4544 : SysScanDesc conscan;
4545 : ScanKeyData skey[1];
4546 : HeapTuple htup;
4547 11180 : int found = 0;
4548 :
4549 : /* Allocate array with room for as many entries as expected */
4550 : check = (ConstrCheck *)
4551 11180 : MemoryContextAllocZero(CacheMemoryContext,
4552 : ncheck * sizeof(ConstrCheck));
4553 :
4554 : /* Search pg_constraint for relevant entries */
4555 11180 : ScanKeyInit(&skey[0],
4556 : Anum_pg_constraint_conrelid,
4557 : BTEqualStrategyNumber, F_OIDEQ,
4558 : ObjectIdGetDatum(RelationGetRelid(relation)));
4559 :
4560 11180 : conrel = table_open(ConstraintRelationId, AccessShareLock);
4561 11180 : conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
4562 : NULL, 1, skey);
4563 :
4564 34870 : while (HeapTupleIsValid(htup = systable_getnext(conscan)))
4565 : {
4566 23690 : Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(htup);
4567 : Datum val;
4568 : bool isnull;
4569 :
4570 : /* We want check constraints only */
4571 23690 : if (conform->contype != CONSTRAINT_CHECK)
4572 9388 : continue;
4573 :
4574 : /* protect limited size of array */
4575 14302 : if (found >= ncheck)
4576 : {
4577 0 : elog(WARNING, "unexpected pg_constraint record found for relation \"%s\"",
4578 : RelationGetRelationName(relation));
4579 0 : break;
4580 : }
4581 :
4582 14302 : check[found].ccvalid = conform->convalidated;
4583 14302 : check[found].ccnoinherit = conform->connoinherit;
4584 28604 : check[found].ccname = MemoryContextStrdup(CacheMemoryContext,
4585 14302 : NameStr(conform->conname));
4586 :
4587 : /* Grab and test conbin is actually set */
4588 14302 : val = fastgetattr(htup,
4589 : Anum_pg_constraint_conbin,
4590 : conrel->rd_att, &isnull);
4591 14302 : if (isnull)
4592 0 : elog(WARNING, "null conbin for relation \"%s\"",
4593 : RelationGetRelationName(relation));
4594 : else
4595 : {
4596 : /* detoast and convert to cstring in caller's context */
4597 14302 : char *s = TextDatumGetCString(val);
4598 :
4599 14302 : check[found].ccbin = MemoryContextStrdup(CacheMemoryContext, s);
4600 14302 : pfree(s);
4601 14302 : found++;
4602 : }
4603 : }
4604 :
4605 11180 : systable_endscan(conscan);
4606 11180 : table_close(conrel, AccessShareLock);
4607 :
4608 11180 : if (found != ncheck)
4609 0 : elog(WARNING, "%d pg_constraint record(s) missing for relation \"%s\"",
4610 : ncheck - found, RelationGetRelationName(relation));
4611 :
4612 : /*
4613 : * Sort the records by name. This ensures that CHECKs are applied in a
4614 : * deterministic order, and it also makes equalTupleDescs() faster.
4615 : */
4616 11180 : if (found > 1)
4617 2578 : qsort(check, found, sizeof(ConstrCheck), CheckConstraintCmp);
4618 :
4619 : /* Install array only after it's fully valid */
4620 11180 : relation->rd_att->constr->check = check;
4621 11180 : relation->rd_att->constr->num_check = found;
4622 11180 : }
4623 :
4624 : /*
4625 : * qsort comparator to sort ConstrCheck entries by name
4626 : */
4627 : static int
4628 3122 : CheckConstraintCmp(const void *a, const void *b)
4629 : {
4630 3122 : const ConstrCheck *ca = (const ConstrCheck *) a;
4631 3122 : const ConstrCheck *cb = (const ConstrCheck *) b;
4632 :
4633 3122 : return strcmp(ca->ccname, cb->ccname);
4634 : }
4635 :
4636 : /*
4637 : * RelationGetFKeyList -- get a list of foreign key info for the relation
4638 : *
4639 : * Returns a list of ForeignKeyCacheInfo structs, one per FK constraining
4640 : * the given relation. This data is a direct copy of relevant fields from
4641 : * pg_constraint. The list items are in no particular order.
4642 : *
4643 : * CAUTION: the returned list is part of the relcache's data, and could
4644 : * vanish in a relcache entry reset. Callers must inspect or copy it
4645 : * before doing anything that might trigger a cache flush, such as
4646 : * system catalog accesses. copyObject() can be used if desired.
4647 : * (We define it this way because current callers want to filter and
4648 : * modify the list entries anyway, so copying would be a waste of time.)
4649 : */
4650 : List *
4651 198924 : RelationGetFKeyList(Relation relation)
4652 : {
4653 : List *result;
4654 : Relation conrel;
4655 : SysScanDesc conscan;
4656 : ScanKeyData skey;
4657 : HeapTuple htup;
4658 : List *oldlist;
4659 : MemoryContext oldcxt;
4660 :
4661 : /* Quick exit if we already computed the list. */
4662 198924 : if (relation->rd_fkeyvalid)
4663 1176 : return relation->rd_fkeylist;
4664 :
4665 : /* Fast path: non-partitioned tables without triggers can't have FKs */
4666 197748 : if (!relation->rd_rel->relhastriggers &&
4667 194720 : relation->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
4668 185166 : return NIL;
4669 :
4670 : /*
4671 : * We build the list we intend to return (in the caller's context) while
4672 : * doing the scan. After successfully completing the scan, we copy that
4673 : * list into the relcache entry. This avoids cache-context memory leakage
4674 : * if we get some sort of error partway through.
4675 : */
4676 12582 : result = NIL;
4677 :
4678 : /* Prepare to scan pg_constraint for entries having conrelid = this rel. */
4679 12582 : ScanKeyInit(&skey,
4680 : Anum_pg_constraint_conrelid,
4681 : BTEqualStrategyNumber, F_OIDEQ,
4682 : ObjectIdGetDatum(RelationGetRelid(relation)));
4683 :
4684 12582 : conrel = table_open(ConstraintRelationId, AccessShareLock);
4685 12582 : conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
4686 : NULL, 1, &skey);
4687 :
4688 19368 : while (HeapTupleIsValid(htup = systable_getnext(conscan)))
4689 : {
4690 6786 : Form_pg_constraint constraint = (Form_pg_constraint) GETSTRUCT(htup);
4691 : ForeignKeyCacheInfo *info;
4692 :
4693 : /* consider only foreign keys */
4694 6786 : if (constraint->contype != CONSTRAINT_FOREIGN)
4695 4080 : continue;
4696 :
4697 2706 : info = makeNode(ForeignKeyCacheInfo);
4698 2706 : info->conoid = constraint->oid;
4699 2706 : info->conrelid = constraint->conrelid;
4700 2706 : info->confrelid = constraint->confrelid;
4701 :
4702 2706 : DeconstructFkConstraintRow(htup, &info->nkeys,
4703 2706 : info->conkey,
4704 2706 : info->confkey,
4705 2706 : info->conpfeqop,
4706 : NULL, NULL, NULL, NULL);
4707 :
4708 : /* Add FK's node to the result list */
4709 2706 : result = lappend(result, info);
4710 : }
4711 :
4712 12582 : systable_endscan(conscan);
4713 12582 : table_close(conrel, AccessShareLock);
4714 :
4715 : /* Now save a copy of the completed list in the relcache entry. */
4716 12582 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4717 12582 : oldlist = relation->rd_fkeylist;
4718 12582 : relation->rd_fkeylist = copyObject(result);
4719 12582 : relation->rd_fkeyvalid = true;
4720 12582 : MemoryContextSwitchTo(oldcxt);
4721 :
4722 : /* Don't leak the old list, if there is one */
4723 12582 : list_free_deep(oldlist);
4724 :
4725 12582 : return result;
4726 : }
4727 :
4728 : /*
4729 : * RelationGetIndexList -- get a list of OIDs of indexes on this relation
4730 : *
4731 : * The index list is created only if someone requests it. We scan pg_index
4732 : * to find relevant indexes, and add the list to the relcache entry so that
4733 : * we won't have to compute it again. Note that shared cache inval of a
4734 : * relcache entry will delete the old list and set rd_indexvalid to false,
4735 : * so that we must recompute the index list on next request. This handles
4736 : * creation or deletion of an index.
4737 : *
4738 : * Indexes that are marked not indislive are omitted from the returned list.
4739 : * Such indexes are expected to be dropped momentarily, and should not be
4740 : * touched at all by any caller of this function.
4741 : *
4742 : * The returned list is guaranteed to be sorted in order by OID. This is
4743 : * needed by the executor, since for index types that we obtain exclusive
4744 : * locks on when updating the index, all backends must lock the indexes in
4745 : * the same order or we will get deadlocks (see ExecOpenIndices()). Any
4746 : * consistent ordering would do, but ordering by OID is easy.
4747 : *
4748 : * Since shared cache inval causes the relcache's copy of the list to go away,
4749 : * we return a copy of the list palloc'd in the caller's context. The caller
4750 : * may list_free() the returned list after scanning it. This is necessary
4751 : * since the caller will typically be doing syscache lookups on the relevant
4752 : * indexes, and syscache lookup could cause SI messages to be processed!
4753 : *
4754 : * In exactly the same way, we update rd_pkindex, which is the OID of the
4755 : * relation's primary key index if any, else InvalidOid; and rd_replidindex,
4756 : * which is the pg_class OID of an index to be used as the relation's
4757 : * replication identity index, or InvalidOid if there is no such index.
4758 : */
4759 : List *
4760 1989068 : RelationGetIndexList(Relation relation)
4761 : {
4762 : Relation indrel;
4763 : SysScanDesc indscan;
4764 : ScanKeyData skey;
4765 : HeapTuple htup;
4766 : List *result;
4767 : List *oldlist;
4768 1989068 : char replident = relation->rd_rel->relreplident;
4769 1989068 : Oid pkeyIndex = InvalidOid;
4770 1989068 : Oid candidateIndex = InvalidOid;
4771 1989068 : bool pkdeferrable = false;
4772 : MemoryContext oldcxt;
4773 :
4774 : /* Quick exit if we already computed the list. */
4775 1989068 : if (relation->rd_indexvalid)
4776 1821546 : return list_copy(relation->rd_indexlist);
4777 :
4778 : /*
4779 : * We build the list we intend to return (in the caller's context) while
4780 : * doing the scan. After successfully completing the scan, we copy that
4781 : * list into the relcache entry. This avoids cache-context memory leakage
4782 : * if we get some sort of error partway through.
4783 : */
4784 167522 : result = NIL;
4785 :
4786 : /* Prepare to scan pg_index for entries having indrelid = this rel. */
4787 167522 : ScanKeyInit(&skey,
4788 : Anum_pg_index_indrelid,
4789 : BTEqualStrategyNumber, F_OIDEQ,
4790 : ObjectIdGetDatum(RelationGetRelid(relation)));
4791 :
4792 167522 : indrel = table_open(IndexRelationId, AccessShareLock);
4793 167522 : indscan = systable_beginscan(indrel, IndexIndrelidIndexId, true,
4794 : NULL, 1, &skey);
4795 :
4796 393786 : while (HeapTupleIsValid(htup = systable_getnext(indscan)))
4797 : {
4798 226264 : Form_pg_index index = (Form_pg_index) GETSTRUCT(htup);
4799 :
4800 : /*
4801 : * Ignore any indexes that are currently being dropped. This will
4802 : * prevent them from being searched, inserted into, or considered in
4803 : * HOT-safety decisions. It's unsafe to touch such an index at all
4804 : * since its catalog entries could disappear at any instant.
4805 : */
4806 226264 : if (!index->indislive)
4807 46 : continue;
4808 :
4809 : /* add index's OID to result list */
4810 226218 : result = lappend_oid(result, index->indexrelid);
4811 :
4812 : /*
4813 : * Non-unique or predicate indexes aren't interesting for either oid
4814 : * indexes or replication identity indexes, so don't check them.
4815 : * Deferred ones are not useful for replication identity either; but
4816 : * we do include them if they are PKs.
4817 : */
4818 226218 : if (!index->indisunique ||
4819 180396 : !heap_attisnull(htup, Anum_pg_index_indpred, NULL))
4820 45976 : continue;
4821 :
4822 : /*
4823 : * Remember primary key index, if any. We do this only if the index
4824 : * is valid; but if the table is partitioned, then we do it even if
4825 : * it's invalid.
4826 : *
4827 : * The reason for returning invalid primary keys for foreign tables is
4828 : * because of pg_dump of NOT NULL constraints, and the fact that PKs
4829 : * remain marked invalid until the partitions' PKs are attached to it.
4830 : * If we make rd_pkindex invalid, then the attnotnull flag is reset
4831 : * after the PK is created, which causes the ALTER INDEX ATTACH
4832 : * PARTITION to fail with 'column ... is not marked NOT NULL'. With
4833 : * this, dropconstraint_internal() will believe that the columns must
4834 : * not have attnotnull reset, so the PKs-on-partitions can be attached
4835 : * correctly, until finally the PK-on-parent is marked valid.
4836 : *
4837 : * Also, this doesn't harm anything, because rd_pkindex is not a
4838 : * "real" index anyway, but a RELKIND_PARTITIONED_INDEX.
4839 : */
4840 180242 : if (index->indisprimary &&
4841 112282 : (index->indisvalid ||
4842 30 : relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE))
4843 : {
4844 112282 : pkeyIndex = index->indexrelid;
4845 112282 : pkdeferrable = !index->indimmediate;
4846 : }
4847 :
4848 180242 : if (!index->indimmediate)
4849 168 : continue;
4850 :
4851 180074 : if (!index->indisvalid)
4852 84 : continue;
4853 :
4854 : /* remember explicitly chosen replica index */
4855 179990 : if (index->indisreplident)
4856 530 : candidateIndex = index->indexrelid;
4857 : }
4858 :
4859 167522 : systable_endscan(indscan);
4860 :
4861 167522 : table_close(indrel, AccessShareLock);
4862 :
4863 : /* Sort the result list into OID order, per API spec. */
4864 167522 : list_sort(result, list_oid_cmp);
4865 :
4866 : /* Now save a copy of the completed list in the relcache entry. */
4867 167522 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4868 167522 : oldlist = relation->rd_indexlist;
4869 167522 : relation->rd_indexlist = list_copy(result);
4870 167522 : relation->rd_pkindex = pkeyIndex;
4871 167522 : relation->rd_ispkdeferrable = pkdeferrable;
4872 167522 : if (replident == REPLICA_IDENTITY_DEFAULT && OidIsValid(pkeyIndex) && !pkdeferrable)
4873 24724 : relation->rd_replidindex = pkeyIndex;
4874 142798 : else if (replident == REPLICA_IDENTITY_INDEX && OidIsValid(candidateIndex))
4875 530 : relation->rd_replidindex = candidateIndex;
4876 : else
4877 142268 : relation->rd_replidindex = InvalidOid;
4878 167522 : relation->rd_indexvalid = true;
4879 167522 : MemoryContextSwitchTo(oldcxt);
4880 :
4881 : /* Don't leak the old list, if there is one */
4882 167522 : list_free(oldlist);
4883 :
4884 167522 : return result;
4885 : }
4886 :
4887 : /*
4888 : * RelationGetStatExtList
4889 : * get a list of OIDs of statistics objects on this relation
4890 : *
4891 : * The statistics list is created only if someone requests it, in a way
4892 : * similar to RelationGetIndexList(). We scan pg_statistic_ext to find
4893 : * relevant statistics, and add the list to the relcache entry so that we
4894 : * won't have to compute it again. Note that shared cache inval of a
4895 : * relcache entry will delete the old list and set rd_statvalid to 0,
4896 : * so that we must recompute the statistics list on next request. This
4897 : * handles creation or deletion of a statistics object.
4898 : *
4899 : * The returned list is guaranteed to be sorted in order by OID, although
4900 : * this is not currently needed.
4901 : *
4902 : * Since shared cache inval causes the relcache's copy of the list to go away,
4903 : * we return a copy of the list palloc'd in the caller's context. The caller
4904 : * may list_free() the returned list after scanning it. This is necessary
4905 : * since the caller will typically be doing syscache lookups on the relevant
4906 : * statistics, and syscache lookup could cause SI messages to be processed!
4907 : */
4908 : List *
4909 414710 : RelationGetStatExtList(Relation relation)
4910 : {
4911 : Relation indrel;
4912 : SysScanDesc indscan;
4913 : ScanKeyData skey;
4914 : HeapTuple htup;
4915 : List *result;
4916 : List *oldlist;
4917 : MemoryContext oldcxt;
4918 :
4919 : /* Quick exit if we already computed the list. */
4920 414710 : if (relation->rd_statvalid != 0)
4921 311856 : return list_copy(relation->rd_statlist);
4922 :
4923 : /*
4924 : * We build the list we intend to return (in the caller's context) while
4925 : * doing the scan. After successfully completing the scan, we copy that
4926 : * list into the relcache entry. This avoids cache-context memory leakage
4927 : * if we get some sort of error partway through.
4928 : */
4929 102854 : result = NIL;
4930 :
4931 : /*
4932 : * Prepare to scan pg_statistic_ext for entries having stxrelid = this
4933 : * rel.
4934 : */
4935 102854 : ScanKeyInit(&skey,
4936 : Anum_pg_statistic_ext_stxrelid,
4937 : BTEqualStrategyNumber, F_OIDEQ,
4938 : ObjectIdGetDatum(RelationGetRelid(relation)));
4939 :
4940 102854 : indrel = table_open(StatisticExtRelationId, AccessShareLock);
4941 102854 : indscan = systable_beginscan(indrel, StatisticExtRelidIndexId, true,
4942 : NULL, 1, &skey);
4943 :
4944 103222 : while (HeapTupleIsValid(htup = systable_getnext(indscan)))
4945 : {
4946 368 : Oid oid = ((Form_pg_statistic_ext) GETSTRUCT(htup))->oid;
4947 :
4948 368 : result = lappend_oid(result, oid);
4949 : }
4950 :
4951 102854 : systable_endscan(indscan);
4952 :
4953 102854 : table_close(indrel, AccessShareLock);
4954 :
4955 : /* Sort the result list into OID order, per API spec. */
4956 102854 : list_sort(result, list_oid_cmp);
4957 :
4958 : /* Now save a copy of the completed list in the relcache entry. */
4959 102854 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4960 102854 : oldlist = relation->rd_statlist;
4961 102854 : relation->rd_statlist = list_copy(result);
4962 :
4963 102854 : relation->rd_statvalid = true;
4964 102854 : MemoryContextSwitchTo(oldcxt);
4965 :
4966 : /* Don't leak the old list, if there is one */
4967 102854 : list_free(oldlist);
4968 :
4969 102854 : return result;
4970 : }
4971 :
4972 : /*
4973 : * RelationGetPrimaryKeyIndex -- get OID of the relation's primary key index
4974 : *
4975 : * Returns InvalidOid if there is no such index, or if the primary key is
4976 : * DEFERRABLE.
4977 : */
4978 : Oid
4979 338 : RelationGetPrimaryKeyIndex(Relation relation)
4980 : {
4981 : List *ilist;
4982 :
4983 338 : if (!relation->rd_indexvalid)
4984 : {
4985 : /* RelationGetIndexList does the heavy lifting. */
4986 0 : ilist = RelationGetIndexList(relation);
4987 0 : list_free(ilist);
4988 : Assert(relation->rd_indexvalid);
4989 : }
4990 :
4991 338 : return relation->rd_ispkdeferrable ? InvalidOid : relation->rd_pkindex;
4992 : }
4993 :
4994 : /*
4995 : * RelationGetReplicaIndex -- get OID of the relation's replica identity index
4996 : *
4997 : * Returns InvalidOid if there is no such index.
4998 : */
4999 : Oid
5000 317642 : RelationGetReplicaIndex(Relation relation)
5001 : {
5002 : List *ilist;
5003 :
5004 317642 : if (!relation->rd_indexvalid)
5005 : {
5006 : /* RelationGetIndexList does the heavy lifting. */
5007 4752 : ilist = RelationGetIndexList(relation);
5008 4752 : list_free(ilist);
5009 : Assert(relation->rd_indexvalid);
5010 : }
5011 :
5012 317642 : return relation->rd_replidindex;
5013 : }
5014 :
5015 : /*
5016 : * RelationGetIndexExpressions -- get the index expressions for an index
5017 : *
5018 : * We cache the result of transforming pg_index.indexprs into a node tree.
5019 : * If the rel is not an index or has no expressional columns, we return NIL.
5020 : * Otherwise, the returned tree is copied into the caller's memory context.
5021 : * (We don't want to return a pointer to the relcache copy, since it could
5022 : * disappear due to relcache invalidation.)
5023 : */
5024 : List *
5025 3632210 : RelationGetIndexExpressions(Relation relation)
5026 : {
5027 : List *result;
5028 : Datum exprsDatum;
5029 : bool isnull;
5030 : char *exprsString;
5031 : MemoryContext oldcxt;
5032 :
5033 : /* Quick exit if we already computed the result. */
5034 3632210 : if (relation->rd_indexprs)
5035 3330 : return copyObject(relation->rd_indexprs);
5036 :
5037 : /* Quick exit if there is nothing to do. */
5038 7257760 : if (relation->rd_indextuple == NULL ||
5039 3628880 : heap_attisnull(relation->rd_indextuple, Anum_pg_index_indexprs, NULL))
5040 3627406 : return NIL;
5041 :
5042 : /*
5043 : * We build the tree we intend to return in the caller's context. After
5044 : * successfully completing the work, we copy it into the relcache entry.
5045 : * This avoids problems if we get some sort of error partway through.
5046 : */
5047 1474 : exprsDatum = heap_getattr(relation->rd_indextuple,
5048 : Anum_pg_index_indexprs,
5049 : GetPgIndexDescriptor(),
5050 : &isnull);
5051 : Assert(!isnull);
5052 1474 : exprsString = TextDatumGetCString(exprsDatum);
5053 1474 : result = (List *) stringToNode(exprsString);
5054 1474 : pfree(exprsString);
5055 :
5056 : /*
5057 : * Run the expressions through eval_const_expressions. This is not just an
5058 : * optimization, but is necessary, because the planner will be comparing
5059 : * them to similarly-processed qual clauses, and may fail to detect valid
5060 : * matches without this. We must not use canonicalize_qual, however,
5061 : * since these aren't qual expressions.
5062 : */
5063 1474 : result = (List *) eval_const_expressions(NULL, (Node *) result);
5064 :
5065 : /* May as well fix opfuncids too */
5066 1474 : fix_opfuncids((Node *) result);
5067 :
5068 : /* Now save a copy of the completed tree in the relcache entry. */
5069 1474 : oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
5070 1474 : relation->rd_indexprs = copyObject(result);
5071 1474 : MemoryContextSwitchTo(oldcxt);
5072 :
5073 1474 : return result;
5074 : }
5075 :
5076 : /*
5077 : * RelationGetDummyIndexExpressions -- get dummy expressions for an index
5078 : *
5079 : * Return a list of dummy expressions (just Const nodes) with the same
5080 : * types/typmods/collations as the index's real expressions. This is
5081 : * useful in situations where we don't want to run any user-defined code.
5082 : */
5083 : List *
5084 244 : RelationGetDummyIndexExpressions(Relation relation)
5085 : {
5086 : List *result;
5087 : Datum exprsDatum;
5088 : bool isnull;
5089 : char *exprsString;
5090 : List *rawExprs;
5091 : ListCell *lc;
5092 :
5093 : /* Quick exit if there is nothing to do. */
5094 488 : if (relation->rd_indextuple == NULL ||
5095 244 : heap_attisnull(relation->rd_indextuple, Anum_pg_index_indexprs, NULL))
5096 226 : return NIL;
5097 :
5098 : /* Extract raw node tree(s) from index tuple. */
5099 18 : exprsDatum = heap_getattr(relation->rd_indextuple,
5100 : Anum_pg_index_indexprs,
5101 : GetPgIndexDescriptor(),
5102 : &isnull);
5103 : Assert(!isnull);
5104 18 : exprsString = TextDatumGetCString(exprsDatum);
5105 18 : rawExprs = (List *) stringToNode(exprsString);
5106 18 : pfree(exprsString);
5107 :
5108 : /* Construct null Consts; the typlen and typbyval are arbitrary. */
5109 18 : result = NIL;
5110 36 : foreach(lc, rawExprs)
5111 : {
5112 18 : Node *rawExpr = (Node *) lfirst(lc);
5113 :
5114 18 : result = lappend(result,
5115 18 : makeConst(exprType(rawExpr),
5116 : exprTypmod(rawExpr),
5117 : exprCollation(rawExpr),
5118 : 1,
5119 : (Datum) 0,
5120 : true,
5121 : true));
5122 : }
5123 :
5124 18 : return result;
5125 : }
5126 :
5127 : /*
5128 : * RelationGetIndexPredicate -- get the index predicate for an index
5129 : *
5130 : * We cache the result of transforming pg_index.indpred into an implicit-AND
5131 : * node tree (suitable for use in planning).
5132 : * If the rel is not an index or has no predicate, we return NIL.
5133 : * Otherwise, the returned tree is copied into the caller's memory context.
5134 : * (We don't want to return a pointer to the relcache copy, since it could
5135 : * disappear due to relcache invalidation.)
5136 : */
5137 : List *
5138 3632102 : RelationGetIndexPredicate(Relation relation)
5139 : {
5140 : List *result;
5141 : Datum predDatum;
5142 : bool isnull;
5143 : char *predString;
5144 : MemoryContext oldcxt;
5145 :
5146 : /* Quick exit if we already computed the result. */
5147 3632102 : if (relation->rd_indpred)
5148 1262 : return copyObject(relation->rd_indpred);
5149 :
5150 : /* Quick exit if there is nothing to do. */
5151 7261680 : if (relation->rd_indextuple == NULL ||
5152 3630840 : heap_attisnull(relation->rd_indextuple, Anum_pg_index_indpred, NULL))
5153 3629928 : return NIL;
5154 :
5155 : /*
5156 : * We build the tree we intend to return in the caller's context. After
5157 : * successfully completing the work, we copy it into the relcache entry.
5158 : * This avoids problems if we get some sort of error partway through.
5159 : */
5160 912 : predDatum = heap_getattr(relation->rd_indextuple,
5161 : Anum_pg_index_indpred,
5162 : GetPgIndexDescriptor(),
5163 : &isnull);
5164 : Assert(!isnull);
5165 912 : predString = TextDatumGetCString(predDatum);
5166 912 : result = (List *) stringToNode(predString);
5167 912 : pfree(predString);
5168 :
5169 : /*
5170 : * Run the expression through const-simplification and canonicalization.
5171 : * This is not just an optimization, but is necessary, because the planner
5172 : * will be comparing it to similarly-processed qual clauses, and may fail
5173 : * to detect valid matches without this. This must match the processing
5174 : * done to qual clauses in preprocess_expression()! (We can skip the
5175 : * stuff involving subqueries, however, since we don't allow any in index
5176 : * predicates.)
5177 : */
5178 912 : result = (List *) eval_const_expressions(NULL, (Node *) result);
5179 :
5180 912 : result = (List *) canonicalize_qual((Expr *) result, false);
5181 :
5182 : /* Also convert to implicit-AND format */
5183 912 : result = make_ands_implicit((Expr *) result);
5184 :
5185 : /* May as well fix opfuncids too */
5186 912 : fix_opfuncids((Node *) result);
5187 :
5188 : /* Now save a copy of the completed tree in the relcache entry. */
5189 912 : oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
5190 912 : relation->rd_indpred = copyObject(result);
5191 912 : MemoryContextSwitchTo(oldcxt);
5192 :
5193 912 : return result;
5194 : }
5195 :
5196 : /*
5197 : * RelationGetIndexAttrBitmap -- get a bitmap of index attribute numbers
5198 : *
5199 : * The result has a bit set for each attribute used anywhere in the index
5200 : * definitions of all the indexes on this relation. (This includes not only
5201 : * simple index keys, but attributes used in expressions and partial-index
5202 : * predicates.)
5203 : *
5204 : * Depending on attrKind, a bitmap covering attnums for certain columns is
5205 : * returned:
5206 : * INDEX_ATTR_BITMAP_KEY Columns in non-partial unique indexes not
5207 : * in expressions (i.e., usable for FKs)
5208 : * INDEX_ATTR_BITMAP_PRIMARY_KEY Columns in the table's primary key
5209 : * (beware: even if PK is deferrable!)
5210 : * INDEX_ATTR_BITMAP_IDENTITY_KEY Columns in the table's replica identity
5211 : * index (empty if FULL)
5212 : * INDEX_ATTR_BITMAP_HOT_BLOCKING Columns that block updates from being HOT
5213 : * INDEX_ATTR_BITMAP_SUMMARIZED Columns included in summarizing indexes
5214 : *
5215 : * Attribute numbers are offset by FirstLowInvalidHeapAttributeNumber so that
5216 : * we can include system attributes (e.g., OID) in the bitmap representation.
5217 : *
5218 : * Deferred indexes are considered for the primary key, but not for replica
5219 : * identity.
5220 : *
5221 : * Caller had better hold at least RowExclusiveLock on the target relation
5222 : * to ensure it is safe (deadlock-free) for us to take locks on the relation's
5223 : * indexes. Note that since the introduction of CREATE INDEX CONCURRENTLY,
5224 : * that lock level doesn't guarantee a stable set of indexes, so we have to
5225 : * be prepared to retry here in case of a change in the set of indexes.
5226 : *
5227 : * The returned result is palloc'd in the caller's memory context and should
5228 : * be bms_free'd when not needed anymore.
5229 : */
5230 : Bitmapset *
5231 2613472 : RelationGetIndexAttrBitmap(Relation relation, IndexAttrBitmapKind attrKind)
5232 : {
5233 : Bitmapset *uindexattrs; /* columns in unique indexes */
5234 : Bitmapset *pkindexattrs; /* columns in the primary index */
5235 : Bitmapset *idindexattrs; /* columns in the replica identity */
5236 : Bitmapset *hotblockingattrs; /* columns with HOT blocking indexes */
5237 : Bitmapset *summarizedattrs; /* columns with summarizing indexes */
5238 : List *indexoidlist;
5239 : List *newindexoidlist;
5240 : Oid relpkindex;
5241 : Oid relreplindex;
5242 : ListCell *l;
5243 : MemoryContext oldcxt;
5244 :
5245 : /* Quick exit if we already computed the result. */
5246 2613472 : if (relation->rd_attrsvalid)
5247 : {
5248 2157368 : switch (attrKind)
5249 : {
5250 523252 : case INDEX_ATTR_BITMAP_KEY:
5251 523252 : return bms_copy(relation->rd_keyattr);
5252 494 : case INDEX_ATTR_BITMAP_PRIMARY_KEY:
5253 494 : return bms_copy(relation->rd_pkattr);
5254 611224 : case INDEX_ATTR_BITMAP_IDENTITY_KEY:
5255 611224 : return bms_copy(relation->rd_idattr);
5256 505658 : case INDEX_ATTR_BITMAP_HOT_BLOCKING:
5257 505658 : return bms_copy(relation->rd_hotblockingattr);
5258 516740 : case INDEX_ATTR_BITMAP_SUMMARIZED:
5259 516740 : return bms_copy(relation->rd_summarizedattr);
5260 0 : default:
5261 0 : elog(ERROR, "unknown attrKind %u", attrKind);
5262 : }
5263 : }
5264 :
5265 : /* Fast path if definitely no indexes */
5266 456104 : if (!RelationGetForm(relation)->relhasindex)
5267 426054 : return NULL;
5268 :
5269 : /*
5270 : * Get cached list of index OIDs. If we have to start over, we do so here.
5271 : */
5272 30050 : restart:
5273 30050 : indexoidlist = RelationGetIndexList(relation);
5274 :
5275 : /* Fall out if no indexes (but relhasindex was set) */
5276 30050 : if (indexoidlist == NIL)
5277 14568 : return NULL;
5278 :
5279 : /*
5280 : * Copy the rd_pkindex and rd_replidindex values computed by
5281 : * RelationGetIndexList before proceeding. This is needed because a
5282 : * relcache flush could occur inside index_open below, resetting the
5283 : * fields managed by RelationGetIndexList. We need to do the work with
5284 : * stable values of these fields.
5285 : */
5286 15482 : relpkindex = relation->rd_pkindex;
5287 15482 : relreplindex = relation->rd_replidindex;
5288 :
5289 : /*
5290 : * For each index, add referenced attributes to indexattrs.
5291 : *
5292 : * Note: we consider all indexes returned by RelationGetIndexList, even if
5293 : * they are not indisready or indisvalid. This is important because an
5294 : * index for which CREATE INDEX CONCURRENTLY has just started must be
5295 : * included in HOT-safety decisions (see README.HOT). If a DROP INDEX
5296 : * CONCURRENTLY is far enough along that we should ignore the index, it
5297 : * won't be returned at all by RelationGetIndexList.
5298 : */
5299 15482 : uindexattrs = NULL;
5300 15482 : pkindexattrs = NULL;
5301 15482 : idindexattrs = NULL;
5302 15482 : hotblockingattrs = NULL;
5303 15482 : summarizedattrs = NULL;
5304 41892 : foreach(l, indexoidlist)
5305 : {
5306 26410 : Oid indexOid = lfirst_oid(l);
5307 : Relation indexDesc;
5308 : Datum datum;
5309 : bool isnull;
5310 : Node *indexExpressions;
5311 : Node *indexPredicate;
5312 : int i;
5313 : bool isKey; /* candidate key */
5314 : bool isPK; /* primary key */
5315 : bool isIDKey; /* replica identity index */
5316 : Bitmapset **attrs;
5317 :
5318 26410 : indexDesc = index_open(indexOid, AccessShareLock);
5319 :
5320 : /*
5321 : * Extract index expressions and index predicate. Note: Don't use
5322 : * RelationGetIndexExpressions()/RelationGetIndexPredicate(), because
5323 : * those might run constant expressions evaluation, which needs a
5324 : * snapshot, which we might not have here. (Also, it's probably more
5325 : * sound to collect the bitmaps before any transformations that might
5326 : * eliminate columns, but the practical impact of this is limited.)
5327 : */
5328 :
5329 26410 : datum = heap_getattr(indexDesc->rd_indextuple, Anum_pg_index_indexprs,
5330 : GetPgIndexDescriptor(), &isnull);
5331 26410 : if (!isnull)
5332 128 : indexExpressions = stringToNode(TextDatumGetCString(datum));
5333 : else
5334 26282 : indexExpressions = NULL;
5335 :
5336 26410 : datum = heap_getattr(indexDesc->rd_indextuple, Anum_pg_index_indpred,
5337 : GetPgIndexDescriptor(), &isnull);
5338 26410 : if (!isnull)
5339 126 : indexPredicate = stringToNode(TextDatumGetCString(datum));
5340 : else
5341 26284 : indexPredicate = NULL;
5342 :
5343 : /* Can this index be referenced by a foreign key? */
5344 20834 : isKey = indexDesc->rd_index->indisunique &&
5345 47244 : indexExpressions == NULL &&
5346 : indexPredicate == NULL;
5347 :
5348 : /* Is this a primary key? */
5349 26410 : isPK = (indexOid == relpkindex);
5350 :
5351 : /* Is this index the configured (or default) replica identity? */
5352 26410 : isIDKey = (indexOid == relreplindex);
5353 :
5354 : /*
5355 : * If the index is summarizing, it doesn't block HOT updates, but we
5356 : * may still need to update it (if the attributes were modified). So
5357 : * decide which bitmap we'll update in the following loop.
5358 : */
5359 26410 : if (indexDesc->rd_indam->amsummarizing)
5360 56 : attrs = &summarizedattrs;
5361 : else
5362 26354 : attrs = &hotblockingattrs;
5363 :
5364 : /* Collect simple attribute references */
5365 66838 : for (i = 0; i < indexDesc->rd_index->indnatts; i++)
5366 : {
5367 40428 : int attrnum = indexDesc->rd_index->indkey.values[i];
5368 :
5369 : /*
5370 : * Since we have covering indexes with non-key columns, we must
5371 : * handle them accurately here. non-key columns must be added into
5372 : * hotblockingattrs or summarizedattrs, since they are in index,
5373 : * and update shouldn't miss them.
5374 : *
5375 : * Summarizing indexes do not block HOT, but do need to be updated
5376 : * when the column value changes, thus require a separate
5377 : * attribute bitmapset.
5378 : *
5379 : * Obviously, non-key columns couldn't be referenced by foreign
5380 : * key or identity key. Hence we do not include them into
5381 : * uindexattrs, pkindexattrs and idindexattrs bitmaps.
5382 : */
5383 40428 : if (attrnum != 0)
5384 : {
5385 40300 : *attrs = bms_add_member(*attrs,
5386 : attrnum - FirstLowInvalidHeapAttributeNumber);
5387 :
5388 40300 : if (isKey && i < indexDesc->rd_index->indnkeyatts)
5389 30600 : uindexattrs = bms_add_member(uindexattrs,
5390 : attrnum - FirstLowInvalidHeapAttributeNumber);
5391 :
5392 40300 : if (isPK && i < indexDesc->rd_index->indnkeyatts)
5393 16292 : pkindexattrs = bms_add_member(pkindexattrs,
5394 : attrnum - FirstLowInvalidHeapAttributeNumber);
5395 :
5396 40300 : if (isIDKey && i < indexDesc->rd_index->indnkeyatts)
5397 5964 : idindexattrs = bms_add_member(idindexattrs,
5398 : attrnum - FirstLowInvalidHeapAttributeNumber);
5399 : }
5400 : }
5401 :
5402 : /* Collect all attributes used in expressions, too */
5403 26410 : pull_varattnos(indexExpressions, 1, attrs);
5404 :
5405 : /* Collect all attributes in the index predicate, too */
5406 26410 : pull_varattnos(indexPredicate, 1, attrs);
5407 :
5408 26410 : index_close(indexDesc, AccessShareLock);
5409 : }
5410 :
5411 : /*
5412 : * During one of the index_opens in the above loop, we might have received
5413 : * a relcache flush event on this relcache entry, which might have been
5414 : * signaling a change in the rel's index list. If so, we'd better start
5415 : * over to ensure we deliver up-to-date attribute bitmaps.
5416 : */
5417 15482 : newindexoidlist = RelationGetIndexList(relation);
5418 15482 : if (equal(indexoidlist, newindexoidlist) &&
5419 15482 : relpkindex == relation->rd_pkindex &&
5420 15482 : relreplindex == relation->rd_replidindex)
5421 : {
5422 : /* Still the same index set, so proceed */
5423 15482 : list_free(newindexoidlist);
5424 15482 : list_free(indexoidlist);
5425 : }
5426 : else
5427 : {
5428 : /* Gotta do it over ... might as well not leak memory */
5429 0 : list_free(newindexoidlist);
5430 0 : list_free(indexoidlist);
5431 0 : bms_free(uindexattrs);
5432 0 : bms_free(pkindexattrs);
5433 0 : bms_free(idindexattrs);
5434 0 : bms_free(hotblockingattrs);
5435 0 : bms_free(summarizedattrs);
5436 :
5437 0 : goto restart;
5438 : }
5439 :
5440 : /* Don't leak the old values of these bitmaps, if any */
5441 15482 : relation->rd_attrsvalid = false;
5442 15482 : bms_free(relation->rd_keyattr);
5443 15482 : relation->rd_keyattr = NULL;
5444 15482 : bms_free(relation->rd_pkattr);
5445 15482 : relation->rd_pkattr = NULL;
5446 15482 : bms_free(relation->rd_idattr);
5447 15482 : relation->rd_idattr = NULL;
5448 15482 : bms_free(relation->rd_hotblockingattr);
5449 15482 : relation->rd_hotblockingattr = NULL;
5450 15482 : bms_free(relation->rd_summarizedattr);
5451 15482 : relation->rd_summarizedattr = NULL;
5452 :
5453 : /*
5454 : * Now save copies of the bitmaps in the relcache entry. We intentionally
5455 : * set rd_attrsvalid last, because that's the one that signals validity of
5456 : * the values; if we run out of memory before making that copy, we won't
5457 : * leave the relcache entry looking like the other ones are valid but
5458 : * empty.
5459 : */
5460 15482 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5461 15482 : relation->rd_keyattr = bms_copy(uindexattrs);
5462 15482 : relation->rd_pkattr = bms_copy(pkindexattrs);
5463 15482 : relation->rd_idattr = bms_copy(idindexattrs);
5464 15482 : relation->rd_hotblockingattr = bms_copy(hotblockingattrs);
5465 15482 : relation->rd_summarizedattr = bms_copy(summarizedattrs);
5466 15482 : relation->rd_attrsvalid = true;
5467 15482 : MemoryContextSwitchTo(oldcxt);
5468 :
5469 : /* We return our original working copy for caller to play with */
5470 15482 : switch (attrKind)
5471 : {
5472 994 : case INDEX_ATTR_BITMAP_KEY:
5473 994 : return uindexattrs;
5474 2274 : case INDEX_ATTR_BITMAP_PRIMARY_KEY:
5475 2274 : return pkindexattrs;
5476 1132 : case INDEX_ATTR_BITMAP_IDENTITY_KEY:
5477 1132 : return idindexattrs;
5478 11082 : case INDEX_ATTR_BITMAP_HOT_BLOCKING:
5479 11082 : return hotblockingattrs;
5480 0 : case INDEX_ATTR_BITMAP_SUMMARIZED:
5481 0 : return summarizedattrs;
5482 0 : default:
5483 0 : elog(ERROR, "unknown attrKind %u", attrKind);
5484 : return NULL;
5485 : }
5486 : }
5487 :
5488 : /*
5489 : * RelationGetIdentityKeyBitmap -- get a bitmap of replica identity attribute
5490 : * numbers
5491 : *
5492 : * A bitmap of index attribute numbers for the configured replica identity
5493 : * index is returned.
5494 : *
5495 : * See also comments of RelationGetIndexAttrBitmap().
5496 : *
5497 : * This is a special purpose function used during logical replication. Here,
5498 : * unlike RelationGetIndexAttrBitmap(), we don't acquire a lock on the required
5499 : * index as we build the cache entry using a historic snapshot and all the
5500 : * later changes are absorbed while decoding WAL. Due to this reason, we don't
5501 : * need to retry here in case of a change in the set of indexes.
5502 : */
5503 : Bitmapset *
5504 542 : RelationGetIdentityKeyBitmap(Relation relation)
5505 : {
5506 542 : Bitmapset *idindexattrs = NULL; /* columns in the replica identity */
5507 : Relation indexDesc;
5508 : int i;
5509 : Oid replidindex;
5510 : MemoryContext oldcxt;
5511 :
5512 : /* Quick exit if we already computed the result */
5513 542 : if (relation->rd_idattr != NULL)
5514 90 : return bms_copy(relation->rd_idattr);
5515 :
5516 : /* Fast path if definitely no indexes */
5517 452 : if (!RelationGetForm(relation)->relhasindex)
5518 124 : return NULL;
5519 :
5520 : /* Historic snapshot must be set. */
5521 : Assert(HistoricSnapshotActive());
5522 :
5523 328 : replidindex = RelationGetReplicaIndex(relation);
5524 :
5525 : /* Fall out if there is no replica identity index */
5526 328 : if (!OidIsValid(replidindex))
5527 4 : return NULL;
5528 :
5529 : /* Look up the description for the replica identity index */
5530 324 : indexDesc = RelationIdGetRelation(replidindex);
5531 :
5532 324 : if (!RelationIsValid(indexDesc))
5533 0 : elog(ERROR, "could not open relation with OID %u",
5534 : relation->rd_replidindex);
5535 :
5536 : /* Add referenced attributes to idindexattrs */
5537 656 : for (i = 0; i < indexDesc->rd_index->indnatts; i++)
5538 : {
5539 332 : int attrnum = indexDesc->rd_index->indkey.values[i];
5540 :
5541 : /*
5542 : * We don't include non-key columns into idindexattrs bitmaps. See
5543 : * RelationGetIndexAttrBitmap.
5544 : */
5545 332 : if (attrnum != 0)
5546 : {
5547 332 : if (i < indexDesc->rd_index->indnkeyatts)
5548 330 : idindexattrs = bms_add_member(idindexattrs,
5549 : attrnum - FirstLowInvalidHeapAttributeNumber);
5550 : }
5551 : }
5552 :
5553 324 : RelationClose(indexDesc);
5554 :
5555 : /* Don't leak the old values of these bitmaps, if any */
5556 324 : bms_free(relation->rd_idattr);
5557 324 : relation->rd_idattr = NULL;
5558 :
5559 : /* Now save copy of the bitmap in the relcache entry */
5560 324 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5561 324 : relation->rd_idattr = bms_copy(idindexattrs);
5562 324 : MemoryContextSwitchTo(oldcxt);
5563 :
5564 : /* We return our original working copy for caller to play with */
5565 324 : return idindexattrs;
5566 : }
5567 :
5568 : /*
5569 : * RelationGetExclusionInfo -- get info about index's exclusion constraint
5570 : *
5571 : * This should be called only for an index that is known to have an associated
5572 : * exclusion constraint or primary key/unique constraint using WITHOUT
5573 : * OVERLAPS.
5574 :
5575 : * It returns arrays (palloc'd in caller's context) of the exclusion operator
5576 : * OIDs, their underlying functions' OIDs, and their strategy numbers in the
5577 : * index's opclasses. We cache all this information since it requires a fair
5578 : * amount of work to get.
5579 : */
5580 : void
5581 1302 : RelationGetExclusionInfo(Relation indexRelation,
5582 : Oid **operators,
5583 : Oid **procs,
5584 : uint16 **strategies)
5585 : {
5586 : int indnkeyatts;
5587 : Oid *ops;
5588 : Oid *funcs;
5589 : uint16 *strats;
5590 : Relation conrel;
5591 : SysScanDesc conscan;
5592 : ScanKeyData skey[1];
5593 : HeapTuple htup;
5594 : bool found;
5595 : MemoryContext oldcxt;
5596 : int i;
5597 :
5598 1302 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(indexRelation);
5599 :
5600 : /* Allocate result space in caller context */
5601 1302 : *operators = ops = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5602 1302 : *procs = funcs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5603 1302 : *strategies = strats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
5604 :
5605 : /* Quick exit if we have the data cached already */
5606 1302 : if (indexRelation->rd_exclstrats != NULL)
5607 : {
5608 900 : memcpy(ops, indexRelation->rd_exclops, sizeof(Oid) * indnkeyatts);
5609 900 : memcpy(funcs, indexRelation->rd_exclprocs, sizeof(Oid) * indnkeyatts);
5610 900 : memcpy(strats, indexRelation->rd_exclstrats, sizeof(uint16) * indnkeyatts);
5611 900 : return;
5612 : }
5613 :
5614 : /*
5615 : * Search pg_constraint for the constraint associated with the index. To
5616 : * make this not too painfully slow, we use the index on conrelid; that
5617 : * will hold the parent relation's OID not the index's own OID.
5618 : *
5619 : * Note: if we wanted to rely on the constraint name matching the index's
5620 : * name, we could just do a direct lookup using pg_constraint's unique
5621 : * index. For the moment it doesn't seem worth requiring that.
5622 : */
5623 402 : ScanKeyInit(&skey[0],
5624 : Anum_pg_constraint_conrelid,
5625 : BTEqualStrategyNumber, F_OIDEQ,
5626 402 : ObjectIdGetDatum(indexRelation->rd_index->indrelid));
5627 :
5628 402 : conrel = table_open(ConstraintRelationId, AccessShareLock);
5629 402 : conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
5630 : NULL, 1, skey);
5631 402 : found = false;
5632 :
5633 1208 : while (HeapTupleIsValid(htup = systable_getnext(conscan)))
5634 : {
5635 806 : Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(htup);
5636 : Datum val;
5637 : bool isnull;
5638 : ArrayType *arr;
5639 : int nelem;
5640 :
5641 : /* We want the exclusion constraint owning the index */
5642 806 : if ((conform->contype != CONSTRAINT_EXCLUSION &&
5643 576 : !(conform->conperiod && (
5644 408 : conform->contype == CONSTRAINT_PRIMARY
5645 128 : || conform->contype == CONSTRAINT_UNIQUE))) ||
5646 534 : conform->conindid != RelationGetRelid(indexRelation))
5647 404 : continue;
5648 :
5649 : /* There should be only one */
5650 402 : if (found)
5651 0 : elog(ERROR, "unexpected exclusion constraint record found for rel %s",
5652 : RelationGetRelationName(indexRelation));
5653 402 : found = true;
5654 :
5655 : /* Extract the operator OIDS from conexclop */
5656 402 : val = fastgetattr(htup,
5657 : Anum_pg_constraint_conexclop,
5658 : conrel->rd_att, &isnull);
5659 402 : if (isnull)
5660 0 : elog(ERROR, "null conexclop for rel %s",
5661 : RelationGetRelationName(indexRelation));
5662 :
5663 402 : arr = DatumGetArrayTypeP(val); /* ensure not toasted */
5664 402 : nelem = ARR_DIMS(arr)[0];
5665 402 : if (ARR_NDIM(arr) != 1 ||
5666 402 : nelem != indnkeyatts ||
5667 402 : ARR_HASNULL(arr) ||
5668 402 : ARR_ELEMTYPE(arr) != OIDOID)
5669 0 : elog(ERROR, "conexclop is not a 1-D Oid array");
5670 :
5671 402 : memcpy(ops, ARR_DATA_PTR(arr), sizeof(Oid) * indnkeyatts);
5672 : }
5673 :
5674 402 : systable_endscan(conscan);
5675 402 : table_close(conrel, AccessShareLock);
5676 :
5677 402 : if (!found)
5678 0 : elog(ERROR, "exclusion constraint record missing for rel %s",
5679 : RelationGetRelationName(indexRelation));
5680 :
5681 : /* We need the func OIDs and strategy numbers too */
5682 1120 : for (i = 0; i < indnkeyatts; i++)
5683 : {
5684 718 : funcs[i] = get_opcode(ops[i]);
5685 1436 : strats[i] = get_op_opfamily_strategy(ops[i],
5686 718 : indexRelation->rd_opfamily[i]);
5687 : /* shouldn't fail, since it was checked at index creation */
5688 718 : if (strats[i] == InvalidStrategy)
5689 0 : elog(ERROR, "could not find strategy for operator %u in family %u",
5690 : ops[i], indexRelation->rd_opfamily[i]);
5691 : }
5692 :
5693 : /* Save a copy of the results in the relcache entry. */
5694 402 : oldcxt = MemoryContextSwitchTo(indexRelation->rd_indexcxt);
5695 402 : indexRelation->rd_exclops = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5696 402 : indexRelation->rd_exclprocs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5697 402 : indexRelation->rd_exclstrats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
5698 402 : memcpy(indexRelation->rd_exclops, ops, sizeof(Oid) * indnkeyatts);
5699 402 : memcpy(indexRelation->rd_exclprocs, funcs, sizeof(Oid) * indnkeyatts);
5700 402 : memcpy(indexRelation->rd_exclstrats, strats, sizeof(uint16) * indnkeyatts);
5701 402 : MemoryContextSwitchTo(oldcxt);
5702 : }
5703 :
5704 : /*
5705 : * Get the publication information for the given relation.
5706 : *
5707 : * Traverse all the publications which the relation is in to get the
5708 : * publication actions and validate the row filter expressions for such
5709 : * publications if any. We consider the row filter expression as invalid if it
5710 : * references any column which is not part of REPLICA IDENTITY.
5711 : *
5712 : * To avoid fetching the publication information repeatedly, we cache the
5713 : * publication actions and row filter validation information.
5714 : */
5715 : void
5716 171752 : RelationBuildPublicationDesc(Relation relation, PublicationDesc *pubdesc)
5717 : {
5718 : List *puboids;
5719 : ListCell *lc;
5720 : MemoryContext oldcxt;
5721 : Oid schemaid;
5722 171752 : List *ancestors = NIL;
5723 171752 : Oid relid = RelationGetRelid(relation);
5724 :
5725 : /*
5726 : * If not publishable, it publishes no actions. (pgoutput_change() will
5727 : * ignore it.)
5728 : */
5729 171752 : if (!is_publishable_relation(relation))
5730 : {
5731 5766 : memset(pubdesc, 0, sizeof(PublicationDesc));
5732 5766 : pubdesc->rf_valid_for_update = true;
5733 5766 : pubdesc->rf_valid_for_delete = true;
5734 5766 : pubdesc->cols_valid_for_update = true;
5735 5766 : pubdesc->cols_valid_for_delete = true;
5736 5766 : return;
5737 : }
5738 :
5739 165986 : if (relation->rd_pubdesc)
5740 : {
5741 158276 : memcpy(pubdesc, relation->rd_pubdesc, sizeof(PublicationDesc));
5742 158276 : return;
5743 : }
5744 :
5745 7710 : memset(pubdesc, 0, sizeof(PublicationDesc));
5746 7710 : pubdesc->rf_valid_for_update = true;
5747 7710 : pubdesc->rf_valid_for_delete = true;
5748 7710 : pubdesc->cols_valid_for_update = true;
5749 7710 : pubdesc->cols_valid_for_delete = true;
5750 :
5751 : /* Fetch the publication membership info. */
5752 7710 : puboids = GetRelationPublications(relid);
5753 7710 : schemaid = RelationGetNamespace(relation);
5754 7710 : puboids = list_concat_unique_oid(puboids, GetSchemaPublications(schemaid));
5755 :
5756 7710 : if (relation->rd_rel->relispartition)
5757 : {
5758 : /* Add publications that the ancestors are in too. */
5759 1764 : ancestors = get_partition_ancestors(relid);
5760 :
5761 4112 : foreach(lc, ancestors)
5762 : {
5763 2348 : Oid ancestor = lfirst_oid(lc);
5764 :
5765 2348 : puboids = list_concat_unique_oid(puboids,
5766 2348 : GetRelationPublications(ancestor));
5767 2348 : schemaid = get_rel_namespace(ancestor);
5768 2348 : puboids = list_concat_unique_oid(puboids,
5769 2348 : GetSchemaPublications(schemaid));
5770 : }
5771 : }
5772 7710 : puboids = list_concat_unique_oid(puboids, GetAllTablesPublications());
5773 :
5774 8354 : foreach(lc, puboids)
5775 : {
5776 812 : Oid pubid = lfirst_oid(lc);
5777 : HeapTuple tup;
5778 : Form_pg_publication pubform;
5779 :
5780 812 : tup = SearchSysCache1(PUBLICATIONOID, ObjectIdGetDatum(pubid));
5781 :
5782 812 : if (!HeapTupleIsValid(tup))
5783 0 : elog(ERROR, "cache lookup failed for publication %u", pubid);
5784 :
5785 812 : pubform = (Form_pg_publication) GETSTRUCT(tup);
5786 :
5787 812 : pubdesc->pubactions.pubinsert |= pubform->pubinsert;
5788 812 : pubdesc->pubactions.pubupdate |= pubform->pubupdate;
5789 812 : pubdesc->pubactions.pubdelete |= pubform->pubdelete;
5790 812 : pubdesc->pubactions.pubtruncate |= pubform->pubtruncate;
5791 :
5792 : /*
5793 : * Check if all columns referenced in the filter expression are part
5794 : * of the REPLICA IDENTITY index or not.
5795 : *
5796 : * If the publication is FOR ALL TABLES then it means the table has no
5797 : * row filters and we can skip the validation.
5798 : */
5799 812 : if (!pubform->puballtables &&
5800 1328 : (pubform->pubupdate || pubform->pubdelete) &&
5801 662 : pub_rf_contains_invalid_column(pubid, relation, ancestors,
5802 662 : pubform->pubviaroot))
5803 : {
5804 60 : if (pubform->pubupdate)
5805 60 : pubdesc->rf_valid_for_update = false;
5806 60 : if (pubform->pubdelete)
5807 60 : pubdesc->rf_valid_for_delete = false;
5808 : }
5809 :
5810 : /*
5811 : * Check if all columns are part of the REPLICA IDENTITY index or not.
5812 : *
5813 : * If the publication is FOR ALL TABLES then it means the table has no
5814 : * column list and we can skip the validation.
5815 : */
5816 812 : if (!pubform->puballtables &&
5817 1328 : (pubform->pubupdate || pubform->pubdelete) &&
5818 662 : pub_collist_contains_invalid_column(pubid, relation, ancestors,
5819 662 : pubform->pubviaroot))
5820 : {
5821 108 : if (pubform->pubupdate)
5822 108 : pubdesc->cols_valid_for_update = false;
5823 108 : if (pubform->pubdelete)
5824 108 : pubdesc->cols_valid_for_delete = false;
5825 : }
5826 :
5827 812 : ReleaseSysCache(tup);
5828 :
5829 : /*
5830 : * If we know everything is replicated and the row filter is invalid
5831 : * for update and delete, there is no point to check for other
5832 : * publications.
5833 : */
5834 812 : if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
5835 806 : pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
5836 794 : !pubdesc->rf_valid_for_update && !pubdesc->rf_valid_for_delete)
5837 60 : break;
5838 :
5839 : /*
5840 : * If we know everything is replicated and the column list is invalid
5841 : * for update and delete, there is no point to check for other
5842 : * publications.
5843 : */
5844 752 : if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
5845 746 : pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
5846 734 : !pubdesc->cols_valid_for_update && !pubdesc->cols_valid_for_delete)
5847 108 : break;
5848 : }
5849 :
5850 7710 : if (relation->rd_pubdesc)
5851 : {
5852 0 : pfree(relation->rd_pubdesc);
5853 0 : relation->rd_pubdesc = NULL;
5854 : }
5855 :
5856 : /* Now save copy of the descriptor in the relcache entry. */
5857 7710 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5858 7710 : relation->rd_pubdesc = palloc(sizeof(PublicationDesc));
5859 7710 : memcpy(relation->rd_pubdesc, pubdesc, sizeof(PublicationDesc));
5860 7710 : MemoryContextSwitchTo(oldcxt);
5861 : }
5862 :
5863 : static bytea **
5864 1043514 : CopyIndexAttOptions(bytea **srcopts, int natts)
5865 : {
5866 1043514 : bytea **opts = palloc(sizeof(*opts) * natts);
5867 :
5868 2949354 : for (int i = 0; i < natts; i++)
5869 : {
5870 1905840 : bytea *opt = srcopts[i];
5871 :
5872 1997508 : opts[i] = !opt ? NULL : (bytea *)
5873 91668 : DatumGetPointer(datumCopy(PointerGetDatum(opt), false, -1));
5874 : }
5875 :
5876 1043514 : return opts;
5877 : }
5878 :
5879 : /*
5880 : * RelationGetIndexAttOptions
5881 : * get AM/opclass-specific options for an index parsed into a binary form
5882 : */
5883 : bytea **
5884 2008924 : RelationGetIndexAttOptions(Relation relation, bool copy)
5885 : {
5886 : MemoryContext oldcxt;
5887 2008924 : bytea **opts = relation->rd_opcoptions;
5888 2008924 : Oid relid = RelationGetRelid(relation);
5889 2008924 : int natts = RelationGetNumberOfAttributes(relation); /* XXX
5890 : * IndexRelationGetNumberOfKeyAttributes */
5891 : int i;
5892 :
5893 : /* Try to copy cached options. */
5894 2008924 : if (opts)
5895 1588042 : return copy ? CopyIndexAttOptions(opts, natts) : opts;
5896 :
5897 : /* Get and parse opclass options. */
5898 420882 : opts = palloc0(sizeof(*opts) * natts);
5899 :
5900 1131056 : for (i = 0; i < natts; i++)
5901 : {
5902 710180 : if (criticalRelcachesBuilt && relid != AttributeRelidNumIndexId)
5903 : {
5904 660616 : Datum attoptions = get_attoptions(relid, i + 1);
5905 :
5906 660616 : opts[i] = index_opclass_options(relation, i + 1, attoptions, false);
5907 :
5908 660610 : if (attoptions != (Datum) 0)
5909 284 : pfree(DatumGetPointer(attoptions));
5910 : }
5911 : }
5912 :
5913 : /* Copy parsed options to the cache. */
5914 420876 : oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
5915 420876 : relation->rd_opcoptions = CopyIndexAttOptions(opts, natts);
5916 420876 : MemoryContextSwitchTo(oldcxt);
5917 :
5918 420876 : if (copy)
5919 0 : return opts;
5920 :
5921 1131050 : for (i = 0; i < natts; i++)
5922 : {
5923 710174 : if (opts[i])
5924 1680 : pfree(opts[i]);
5925 : }
5926 :
5927 420876 : pfree(opts);
5928 :
5929 420876 : return relation->rd_opcoptions;
5930 : }
5931 :
5932 : /*
5933 : * Routines to support ereport() reports of relation-related errors
5934 : *
5935 : * These could have been put into elog.c, but it seems like a module layering
5936 : * violation to have elog.c calling relcache or syscache stuff --- and we
5937 : * definitely don't want elog.h including rel.h. So we put them here.
5938 : */
5939 :
5940 : /*
5941 : * errtable --- stores schema_name and table_name of a table
5942 : * within the current errordata.
5943 : */
5944 : int
5945 3270 : errtable(Relation rel)
5946 : {
5947 3270 : err_generic_string(PG_DIAG_SCHEMA_NAME,
5948 3270 : get_namespace_name(RelationGetNamespace(rel)));
5949 3270 : err_generic_string(PG_DIAG_TABLE_NAME, RelationGetRelationName(rel));
5950 :
5951 3270 : return 0; /* return value does not matter */
5952 : }
5953 :
5954 : /*
5955 : * errtablecol --- stores schema_name, table_name and column_name
5956 : * of a table column within the current errordata.
5957 : *
5958 : * The column is specified by attribute number --- for most callers, this is
5959 : * easier and less error-prone than getting the column name for themselves.
5960 : */
5961 : int
5962 416 : errtablecol(Relation rel, int attnum)
5963 : {
5964 416 : TupleDesc reldesc = RelationGetDescr(rel);
5965 : const char *colname;
5966 :
5967 : /* Use reldesc if it's a user attribute, else consult the catalogs */
5968 416 : if (attnum > 0 && attnum <= reldesc->natts)
5969 416 : colname = NameStr(TupleDescAttr(reldesc, attnum - 1)->attname);
5970 : else
5971 0 : colname = get_attname(RelationGetRelid(rel), attnum, false);
5972 :
5973 416 : return errtablecolname(rel, colname);
5974 : }
5975 :
5976 : /*
5977 : * errtablecolname --- stores schema_name, table_name and column_name
5978 : * of a table column within the current errordata, where the column name is
5979 : * given directly rather than extracted from the relation's catalog data.
5980 : *
5981 : * Don't use this directly unless errtablecol() is inconvenient for some
5982 : * reason. This might possibly be needed during intermediate states in ALTER
5983 : * TABLE, for instance.
5984 : */
5985 : int
5986 416 : errtablecolname(Relation rel, const char *colname)
5987 : {
5988 416 : errtable(rel);
5989 416 : err_generic_string(PG_DIAG_COLUMN_NAME, colname);
5990 :
5991 416 : return 0; /* return value does not matter */
5992 : }
5993 :
5994 : /*
5995 : * errtableconstraint --- stores schema_name, table_name and constraint_name
5996 : * of a table-related constraint within the current errordata.
5997 : */
5998 : int
5999 2352 : errtableconstraint(Relation rel, const char *conname)
6000 : {
6001 2352 : errtable(rel);
6002 2352 : err_generic_string(PG_DIAG_CONSTRAINT_NAME, conname);
6003 :
6004 2352 : return 0; /* return value does not matter */
6005 : }
6006 :
6007 :
6008 : /*
6009 : * load_relcache_init_file, write_relcache_init_file
6010 : *
6011 : * In late 1992, we started regularly having databases with more than
6012 : * a thousand classes in them. With this number of classes, it became
6013 : * critical to do indexed lookups on the system catalogs.
6014 : *
6015 : * Bootstrapping these lookups is very hard. We want to be able to
6016 : * use an index on pg_attribute, for example, but in order to do so,
6017 : * we must have read pg_attribute for the attributes in the index,
6018 : * which implies that we need to use the index.
6019 : *
6020 : * In order to get around the problem, we do the following:
6021 : *
6022 : * + When the database system is initialized (at initdb time), we
6023 : * don't use indexes. We do sequential scans.
6024 : *
6025 : * + When the backend is started up in normal mode, we load an image
6026 : * of the appropriate relation descriptors, in internal format,
6027 : * from an initialization file in the data/base/... directory.
6028 : *
6029 : * + If the initialization file isn't there, then we create the
6030 : * relation descriptors using sequential scans and write 'em to
6031 : * the initialization file for use by subsequent backends.
6032 : *
6033 : * As of Postgres 9.0, there is one local initialization file in each
6034 : * database, plus one shared initialization file for shared catalogs.
6035 : *
6036 : * We could dispense with the initialization files and just build the
6037 : * critical reldescs the hard way on every backend startup, but that
6038 : * slows down backend startup noticeably.
6039 : *
6040 : * We can in fact go further, and save more relcache entries than
6041 : * just the ones that are absolutely critical; this allows us to speed
6042 : * up backend startup by not having to build such entries the hard way.
6043 : * Presently, all the catalog and index entries that are referred to
6044 : * by catcaches are stored in the initialization files.
6045 : *
6046 : * The same mechanism that detects when catcache and relcache entries
6047 : * need to be invalidated (due to catalog updates) also arranges to
6048 : * unlink the initialization files when the contents may be out of date.
6049 : * The files will then be rebuilt during the next backend startup.
6050 : */
6051 :
6052 : /*
6053 : * load_relcache_init_file -- attempt to load cache from the shared
6054 : * or local cache init file
6055 : *
6056 : * If successful, return true and set criticalRelcachesBuilt or
6057 : * criticalSharedRelcachesBuilt to true.
6058 : * If not successful, return false.
6059 : *
6060 : * NOTE: we assume we are already switched into CacheMemoryContext.
6061 : */
6062 : static bool
6063 49326 : load_relcache_init_file(bool shared)
6064 : {
6065 : FILE *fp;
6066 : char initfilename[MAXPGPATH];
6067 : Relation *rels;
6068 : int relno,
6069 : num_rels,
6070 : max_rels,
6071 : nailed_rels,
6072 : nailed_indexes,
6073 : magic;
6074 : int i;
6075 :
6076 49326 : if (shared)
6077 25818 : snprintf(initfilename, sizeof(initfilename), "global/%s",
6078 : RELCACHE_INIT_FILENAME);
6079 : else
6080 23508 : snprintf(initfilename, sizeof(initfilename), "%s/%s",
6081 : DatabasePath, RELCACHE_INIT_FILENAME);
6082 :
6083 49326 : fp = AllocateFile(initfilename, PG_BINARY_R);
6084 49326 : if (fp == NULL)
6085 5308 : return false;
6086 :
6087 : /*
6088 : * Read the index relcache entries from the file. Note we will not enter
6089 : * any of them into the cache if the read fails partway through; this
6090 : * helps to guard against broken init files.
6091 : */
6092 44018 : max_rels = 100;
6093 44018 : rels = (Relation *) palloc(max_rels * sizeof(Relation));
6094 44018 : num_rels = 0;
6095 44018 : nailed_rels = nailed_indexes = 0;
6096 :
6097 : /* check for correct magic number (compatible version) */
6098 44018 : if (fread(&magic, 1, sizeof(magic), fp) != sizeof(magic))
6099 0 : goto read_failed;
6100 44018 : if (magic != RELCACHE_INIT_FILEMAGIC)
6101 0 : goto read_failed;
6102 :
6103 44018 : for (relno = 0;; relno++)
6104 2894396 : {
6105 : Size len;
6106 : size_t nread;
6107 : Relation rel;
6108 : Form_pg_class relform;
6109 : bool has_not_null;
6110 :
6111 : /* first read the relation descriptor length */
6112 2938414 : nread = fread(&len, 1, sizeof(len), fp);
6113 2938414 : if (nread != sizeof(len))
6114 : {
6115 44018 : if (nread == 0)
6116 44018 : break; /* end of file */
6117 0 : goto read_failed;
6118 : }
6119 :
6120 : /* safety check for incompatible relcache layout */
6121 2894396 : if (len != sizeof(RelationData))
6122 0 : goto read_failed;
6123 :
6124 : /* allocate another relcache header */
6125 2894396 : if (num_rels >= max_rels)
6126 : {
6127 21400 : max_rels *= 2;
6128 21400 : rels = (Relation *) repalloc(rels, max_rels * sizeof(Relation));
6129 : }
6130 :
6131 2894396 : rel = rels[num_rels++] = (Relation) palloc(len);
6132 :
6133 : /* then, read the Relation structure */
6134 2894396 : if (fread(rel, 1, len, fp) != len)
6135 0 : goto read_failed;
6136 :
6137 : /* next read the relation tuple form */
6138 2894396 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6139 0 : goto read_failed;
6140 :
6141 2894396 : relform = (Form_pg_class) palloc(len);
6142 2894396 : if (fread(relform, 1, len, fp) != len)
6143 0 : goto read_failed;
6144 :
6145 2894396 : rel->rd_rel = relform;
6146 :
6147 : /* initialize attribute tuple forms */
6148 2894396 : rel->rd_att = CreateTemplateTupleDesc(relform->relnatts);
6149 2894396 : rel->rd_att->tdrefcount = 1; /* mark as refcounted */
6150 :
6151 2894396 : rel->rd_att->tdtypeid = relform->reltype ? relform->reltype : RECORDOID;
6152 2894396 : rel->rd_att->tdtypmod = -1; /* just to be sure */
6153 :
6154 : /* next read all the attribute tuple form data entries */
6155 2894396 : has_not_null = false;
6156 17021016 : for (i = 0; i < relform->relnatts; i++)
6157 : {
6158 14126620 : Form_pg_attribute attr = TupleDescAttr(rel->rd_att, i);
6159 :
6160 14126620 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6161 0 : goto read_failed;
6162 14126620 : if (len != ATTRIBUTE_FIXED_PART_SIZE)
6163 0 : goto read_failed;
6164 14126620 : if (fread(attr, 1, len, fp) != len)
6165 0 : goto read_failed;
6166 :
6167 14126620 : has_not_null |= attr->attnotnull;
6168 : }
6169 :
6170 : /* next read the access method specific field */
6171 2894396 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6172 0 : goto read_failed;
6173 2894396 : if (len > 0)
6174 : {
6175 0 : rel->rd_options = palloc(len);
6176 0 : if (fread(rel->rd_options, 1, len, fp) != len)
6177 0 : goto read_failed;
6178 0 : if (len != VARSIZE(rel->rd_options))
6179 0 : goto read_failed; /* sanity check */
6180 : }
6181 : else
6182 : {
6183 2894396 : rel->rd_options = NULL;
6184 : }
6185 :
6186 : /* mark not-null status */
6187 2894396 : if (has_not_null)
6188 : {
6189 1079744 : TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
6190 :
6191 1079744 : constr->has_not_null = true;
6192 1079744 : rel->rd_att->constr = constr;
6193 : }
6194 :
6195 : /*
6196 : * If it's an index, there's more to do. Note we explicitly ignore
6197 : * partitioned indexes here.
6198 : */
6199 2894396 : if (rel->rd_rel->relkind == RELKIND_INDEX)
6200 : {
6201 : MemoryContext indexcxt;
6202 : Oid *opfamily;
6203 : Oid *opcintype;
6204 : RegProcedure *support;
6205 : int nsupport;
6206 : int16 *indoption;
6207 : Oid *indcollation;
6208 :
6209 : /* Count nailed indexes to ensure we have 'em all */
6210 1814652 : if (rel->rd_isnailed)
6211 285508 : nailed_indexes++;
6212 :
6213 : /* read the pg_index tuple */
6214 1814652 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6215 0 : goto read_failed;
6216 :
6217 1814652 : rel->rd_indextuple = (HeapTuple) palloc(len);
6218 1814652 : if (fread(rel->rd_indextuple, 1, len, fp) != len)
6219 0 : goto read_failed;
6220 :
6221 : /* Fix up internal pointers in the tuple -- see heap_copytuple */
6222 1814652 : rel->rd_indextuple->t_data = (HeapTupleHeader) ((char *) rel->rd_indextuple + HEAPTUPLESIZE);
6223 1814652 : rel->rd_index = (Form_pg_index) GETSTRUCT(rel->rd_indextuple);
6224 :
6225 : /*
6226 : * prepare index info context --- parameters should match
6227 : * RelationInitIndexAccessInfo
6228 : */
6229 1814652 : indexcxt = AllocSetContextCreate(CacheMemoryContext,
6230 : "index info",
6231 : ALLOCSET_SMALL_SIZES);
6232 1814652 : rel->rd_indexcxt = indexcxt;
6233 1814652 : MemoryContextCopyAndSetIdentifier(indexcxt,
6234 : RelationGetRelationName(rel));
6235 :
6236 : /*
6237 : * Now we can fetch the index AM's API struct. (We can't store
6238 : * that in the init file, since it contains function pointers that
6239 : * might vary across server executions. Fortunately, it should be
6240 : * safe to call the amhandler even while bootstrapping indexes.)
6241 : */
6242 1814652 : InitIndexAmRoutine(rel);
6243 :
6244 : /* read the vector of opfamily OIDs */
6245 1814652 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6246 0 : goto read_failed;
6247 :
6248 1814652 : opfamily = (Oid *) MemoryContextAlloc(indexcxt, len);
6249 1814652 : if (fread(opfamily, 1, len, fp) != len)
6250 0 : goto read_failed;
6251 :
6252 1814652 : rel->rd_opfamily = opfamily;
6253 :
6254 : /* read the vector of opcintype OIDs */
6255 1814652 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6256 0 : goto read_failed;
6257 :
6258 1814652 : opcintype = (Oid *) MemoryContextAlloc(indexcxt, len);
6259 1814652 : if (fread(opcintype, 1, len, fp) != len)
6260 0 : goto read_failed;
6261 :
6262 1814652 : rel->rd_opcintype = opcintype;
6263 :
6264 : /* read the vector of support procedure OIDs */
6265 1814652 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6266 0 : goto read_failed;
6267 1814652 : support = (RegProcedure *) MemoryContextAlloc(indexcxt, len);
6268 1814652 : if (fread(support, 1, len, fp) != len)
6269 0 : goto read_failed;
6270 :
6271 1814652 : rel->rd_support = support;
6272 :
6273 : /* read the vector of collation OIDs */
6274 1814652 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6275 0 : goto read_failed;
6276 :
6277 1814652 : indcollation = (Oid *) MemoryContextAlloc(indexcxt, len);
6278 1814652 : if (fread(indcollation, 1, len, fp) != len)
6279 0 : goto read_failed;
6280 :
6281 1814652 : rel->rd_indcollation = indcollation;
6282 :
6283 : /* read the vector of indoption values */
6284 1814652 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6285 0 : goto read_failed;
6286 :
6287 1814652 : indoption = (int16 *) MemoryContextAlloc(indexcxt, len);
6288 1814652 : if (fread(indoption, 1, len, fp) != len)
6289 0 : goto read_failed;
6290 :
6291 1814652 : rel->rd_indoption = indoption;
6292 :
6293 : /* read the vector of opcoptions values */
6294 1814652 : rel->rd_opcoptions = (bytea **)
6295 1814652 : MemoryContextAllocZero(indexcxt, sizeof(*rel->rd_opcoptions) * relform->relnatts);
6296 :
6297 4814830 : for (i = 0; i < relform->relnatts; i++)
6298 : {
6299 3000178 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6300 0 : goto read_failed;
6301 :
6302 3000178 : if (len > 0)
6303 : {
6304 0 : rel->rd_opcoptions[i] = (bytea *) MemoryContextAlloc(indexcxt, len);
6305 0 : if (fread(rel->rd_opcoptions[i], 1, len, fp) != len)
6306 0 : goto read_failed;
6307 : }
6308 : }
6309 :
6310 : /* set up zeroed fmgr-info vector */
6311 1814652 : nsupport = relform->relnatts * rel->rd_indam->amsupport;
6312 1814652 : rel->rd_supportinfo = (FmgrInfo *)
6313 1814652 : MemoryContextAllocZero(indexcxt, nsupport * sizeof(FmgrInfo));
6314 : }
6315 : else
6316 : {
6317 : /* Count nailed rels to ensure we have 'em all */
6318 1079744 : if (rel->rd_isnailed)
6319 198690 : nailed_rels++;
6320 :
6321 : /* Load table AM data */
6322 1079744 : if (RELKIND_HAS_TABLE_AM(rel->rd_rel->relkind) || rel->rd_rel->relkind == RELKIND_SEQUENCE)
6323 1079744 : RelationInitTableAccessMethod(rel);
6324 :
6325 : Assert(rel->rd_index == NULL);
6326 : Assert(rel->rd_indextuple == NULL);
6327 : Assert(rel->rd_indexcxt == NULL);
6328 : Assert(rel->rd_indam == NULL);
6329 : Assert(rel->rd_opfamily == NULL);
6330 : Assert(rel->rd_opcintype == NULL);
6331 : Assert(rel->rd_support == NULL);
6332 : Assert(rel->rd_supportinfo == NULL);
6333 : Assert(rel->rd_indoption == NULL);
6334 : Assert(rel->rd_indcollation == NULL);
6335 : Assert(rel->rd_opcoptions == NULL);
6336 : }
6337 :
6338 : /*
6339 : * Rules and triggers are not saved (mainly because the internal
6340 : * format is complex and subject to change). They must be rebuilt if
6341 : * needed by RelationCacheInitializePhase3. This is not expected to
6342 : * be a big performance hit since few system catalogs have such. Ditto
6343 : * for RLS policy data, partition info, index expressions, predicates,
6344 : * exclusion info, and FDW info.
6345 : */
6346 2894396 : rel->rd_rules = NULL;
6347 2894396 : rel->rd_rulescxt = NULL;
6348 2894396 : rel->trigdesc = NULL;
6349 2894396 : rel->rd_rsdesc = NULL;
6350 2894396 : rel->rd_partkey = NULL;
6351 2894396 : rel->rd_partkeycxt = NULL;
6352 2894396 : rel->rd_partdesc = NULL;
6353 2894396 : rel->rd_partdesc_nodetached = NULL;
6354 2894396 : rel->rd_partdesc_nodetached_xmin = InvalidTransactionId;
6355 2894396 : rel->rd_pdcxt = NULL;
6356 2894396 : rel->rd_pddcxt = NULL;
6357 2894396 : rel->rd_partcheck = NIL;
6358 2894396 : rel->rd_partcheckvalid = false;
6359 2894396 : rel->rd_partcheckcxt = NULL;
6360 2894396 : rel->rd_indexprs = NIL;
6361 2894396 : rel->rd_indpred = NIL;
6362 2894396 : rel->rd_exclops = NULL;
6363 2894396 : rel->rd_exclprocs = NULL;
6364 2894396 : rel->rd_exclstrats = NULL;
6365 2894396 : rel->rd_fdwroutine = NULL;
6366 :
6367 : /*
6368 : * Reset transient-state fields in the relcache entry
6369 : */
6370 2894396 : rel->rd_smgr = NULL;
6371 2894396 : if (rel->rd_isnailed)
6372 484198 : rel->rd_refcnt = 1;
6373 : else
6374 2410198 : rel->rd_refcnt = 0;
6375 2894396 : rel->rd_indexvalid = false;
6376 2894396 : rel->rd_indexlist = NIL;
6377 2894396 : rel->rd_pkindex = InvalidOid;
6378 2894396 : rel->rd_replidindex = InvalidOid;
6379 2894396 : rel->rd_attrsvalid = false;
6380 2894396 : rel->rd_keyattr = NULL;
6381 2894396 : rel->rd_pkattr = NULL;
6382 2894396 : rel->rd_idattr = NULL;
6383 2894396 : rel->rd_pubdesc = NULL;
6384 2894396 : rel->rd_statvalid = false;
6385 2894396 : rel->rd_statlist = NIL;
6386 2894396 : rel->rd_fkeyvalid = false;
6387 2894396 : rel->rd_fkeylist = NIL;
6388 2894396 : rel->rd_createSubid = InvalidSubTransactionId;
6389 2894396 : rel->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
6390 2894396 : rel->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
6391 2894396 : rel->rd_droppedSubid = InvalidSubTransactionId;
6392 2894396 : rel->rd_amcache = NULL;
6393 2894396 : rel->pgstat_info = NULL;
6394 :
6395 : /*
6396 : * Recompute lock and physical addressing info. This is needed in
6397 : * case the pg_internal.init file was copied from some other database
6398 : * by CREATE DATABASE.
6399 : */
6400 2894396 : RelationInitLockInfo(rel);
6401 2894396 : RelationInitPhysicalAddr(rel);
6402 : }
6403 :
6404 : /*
6405 : * We reached the end of the init file without apparent problem. Did we
6406 : * get the right number of nailed items? This is a useful crosscheck in
6407 : * case the set of critical rels or indexes changes. However, that should
6408 : * not happen in a normally-running system, so let's bleat if it does.
6409 : *
6410 : * For the shared init file, we're called before client authentication is
6411 : * done, which means that elog(WARNING) will go only to the postmaster
6412 : * log, where it's easily missed. To ensure that developers notice bad
6413 : * values of NUM_CRITICAL_SHARED_RELS/NUM_CRITICAL_SHARED_INDEXES, we put
6414 : * an Assert(false) there.
6415 : */
6416 44018 : if (shared)
6417 : {
6418 22618 : if (nailed_rels != NUM_CRITICAL_SHARED_RELS ||
6419 : nailed_indexes != NUM_CRITICAL_SHARED_INDEXES)
6420 : {
6421 0 : elog(WARNING, "found %d nailed shared rels and %d nailed shared indexes in init file, but expected %d and %d respectively",
6422 : nailed_rels, nailed_indexes,
6423 : NUM_CRITICAL_SHARED_RELS, NUM_CRITICAL_SHARED_INDEXES);
6424 : /* Make sure we get developers' attention about this */
6425 : Assert(false);
6426 : /* In production builds, recover by bootstrapping the relcache */
6427 0 : goto read_failed;
6428 : }
6429 : }
6430 : else
6431 : {
6432 21400 : if (nailed_rels != NUM_CRITICAL_LOCAL_RELS ||
6433 : nailed_indexes != NUM_CRITICAL_LOCAL_INDEXES)
6434 : {
6435 0 : elog(WARNING, "found %d nailed rels and %d nailed indexes in init file, but expected %d and %d respectively",
6436 : nailed_rels, nailed_indexes,
6437 : NUM_CRITICAL_LOCAL_RELS, NUM_CRITICAL_LOCAL_INDEXES);
6438 : /* We don't need an Assert() in this case */
6439 0 : goto read_failed;
6440 : }
6441 : }
6442 :
6443 : /*
6444 : * OK, all appears well.
6445 : *
6446 : * Now insert all the new relcache entries into the cache.
6447 : */
6448 2938414 : for (relno = 0; relno < num_rels; relno++)
6449 : {
6450 2894396 : RelationCacheInsert(rels[relno], false);
6451 : }
6452 :
6453 44018 : pfree(rels);
6454 44018 : FreeFile(fp);
6455 :
6456 44018 : if (shared)
6457 22618 : criticalSharedRelcachesBuilt = true;
6458 : else
6459 21400 : criticalRelcachesBuilt = true;
6460 44018 : return true;
6461 :
6462 : /*
6463 : * init file is broken, so do it the hard way. We don't bother trying to
6464 : * free the clutter we just allocated; it's not in the relcache so it
6465 : * won't hurt.
6466 : */
6467 0 : read_failed:
6468 0 : pfree(rels);
6469 0 : FreeFile(fp);
6470 :
6471 0 : return false;
6472 : }
6473 :
6474 : /*
6475 : * Write out a new initialization file with the current contents
6476 : * of the relcache (either shared rels or local rels, as indicated).
6477 : */
6478 : static void
6479 4304 : write_relcache_init_file(bool shared)
6480 : {
6481 : FILE *fp;
6482 : char tempfilename[MAXPGPATH];
6483 : char finalfilename[MAXPGPATH];
6484 : int magic;
6485 : HASH_SEQ_STATUS status;
6486 : RelIdCacheEnt *idhentry;
6487 : int i;
6488 :
6489 : /*
6490 : * If we have already received any relcache inval events, there's no
6491 : * chance of succeeding so we may as well skip the whole thing.
6492 : */
6493 4304 : if (relcacheInvalsReceived != 0L)
6494 24 : return;
6495 :
6496 : /*
6497 : * We must write a temporary file and rename it into place. Otherwise,
6498 : * another backend starting at about the same time might crash trying to
6499 : * read the partially-complete file.
6500 : */
6501 4280 : if (shared)
6502 : {
6503 2140 : snprintf(tempfilename, sizeof(tempfilename), "global/%s.%d",
6504 : RELCACHE_INIT_FILENAME, MyProcPid);
6505 2140 : snprintf(finalfilename, sizeof(finalfilename), "global/%s",
6506 : RELCACHE_INIT_FILENAME);
6507 : }
6508 : else
6509 : {
6510 2140 : snprintf(tempfilename, sizeof(tempfilename), "%s/%s.%d",
6511 : DatabasePath, RELCACHE_INIT_FILENAME, MyProcPid);
6512 2140 : snprintf(finalfilename, sizeof(finalfilename), "%s/%s",
6513 : DatabasePath, RELCACHE_INIT_FILENAME);
6514 : }
6515 :
6516 4280 : unlink(tempfilename); /* in case it exists w/wrong permissions */
6517 :
6518 4280 : fp = AllocateFile(tempfilename, PG_BINARY_W);
6519 4280 : if (fp == NULL)
6520 : {
6521 : /*
6522 : * We used to consider this a fatal error, but we might as well
6523 : * continue with backend startup ...
6524 : */
6525 0 : ereport(WARNING,
6526 : (errcode_for_file_access(),
6527 : errmsg("could not create relation-cache initialization file \"%s\": %m",
6528 : tempfilename),
6529 : errdetail("Continuing anyway, but there's something wrong.")));
6530 0 : return;
6531 : }
6532 :
6533 : /*
6534 : * Write a magic number to serve as a file version identifier. We can
6535 : * change the magic number whenever the relcache layout changes.
6536 : */
6537 4280 : magic = RELCACHE_INIT_FILEMAGIC;
6538 4280 : if (fwrite(&magic, 1, sizeof(magic), fp) != sizeof(magic))
6539 0 : ereport(FATAL,
6540 : errcode_for_file_access(),
6541 : errmsg_internal("could not write init file: %m"));
6542 :
6543 : /*
6544 : * Write all the appropriate reldescs (in no particular order).
6545 : */
6546 4280 : hash_seq_init(&status, RelationIdCache);
6547 :
6548 577800 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
6549 : {
6550 573520 : Relation rel = idhentry->reldesc;
6551 573520 : Form_pg_class relform = rel->rd_rel;
6552 :
6553 : /* ignore if not correct group */
6554 573520 : if (relform->relisshared != shared)
6555 286760 : continue;
6556 :
6557 : /*
6558 : * Ignore if not supposed to be in init file. We can allow any shared
6559 : * relation that's been loaded so far to be in the shared init file,
6560 : * but unshared relations must be ones that should be in the local
6561 : * file per RelationIdIsInInitFile. (Note: if you want to change the
6562 : * criterion for rels to be kept in the init file, see also inval.c.
6563 : * The reason for filtering here is to be sure that we don't put
6564 : * anything into the local init file for which a relcache inval would
6565 : * not cause invalidation of that init file.)
6566 : */
6567 286760 : if (!shared && !RelationIdIsInInitFile(RelationGetRelid(rel)))
6568 : {
6569 : /* Nailed rels had better get stored. */
6570 : Assert(!rel->rd_isnailed);
6571 0 : continue;
6572 : }
6573 :
6574 : /* first write the relcache entry proper */
6575 286760 : write_item(rel, sizeof(RelationData), fp);
6576 :
6577 : /* next write the relation tuple form */
6578 286760 : write_item(relform, CLASS_TUPLE_SIZE, fp);
6579 :
6580 : /* next, do all the attribute tuple form data entries */
6581 1688460 : for (i = 0; i < relform->relnatts; i++)
6582 : {
6583 1401700 : write_item(TupleDescAttr(rel->rd_att, i),
6584 : ATTRIBUTE_FIXED_PART_SIZE, fp);
6585 : }
6586 :
6587 : /* next, do the access method specific field */
6588 286760 : write_item(rel->rd_options,
6589 286760 : (rel->rd_options ? VARSIZE(rel->rd_options) : 0),
6590 : fp);
6591 :
6592 : /*
6593 : * If it's an index, there's more to do. Note we explicitly ignore
6594 : * partitioned indexes here.
6595 : */
6596 286760 : if (rel->rd_rel->relkind == RELKIND_INDEX)
6597 : {
6598 : /* write the pg_index tuple */
6599 : /* we assume this was created by heap_copytuple! */
6600 179760 : write_item(rel->rd_indextuple,
6601 179760 : HEAPTUPLESIZE + rel->rd_indextuple->t_len,
6602 : fp);
6603 :
6604 : /* write the vector of opfamily OIDs */
6605 179760 : write_item(rel->rd_opfamily,
6606 179760 : relform->relnatts * sizeof(Oid),
6607 : fp);
6608 :
6609 : /* write the vector of opcintype OIDs */
6610 179760 : write_item(rel->rd_opcintype,
6611 179760 : relform->relnatts * sizeof(Oid),
6612 : fp);
6613 :
6614 : /* write the vector of support procedure OIDs */
6615 179760 : write_item(rel->rd_support,
6616 179760 : relform->relnatts * (rel->rd_indam->amsupport * sizeof(RegProcedure)),
6617 : fp);
6618 :
6619 : /* write the vector of collation OIDs */
6620 179760 : write_item(rel->rd_indcollation,
6621 179760 : relform->relnatts * sizeof(Oid),
6622 : fp);
6623 :
6624 : /* write the vector of indoption values */
6625 179760 : write_item(rel->rd_indoption,
6626 179760 : relform->relnatts * sizeof(int16),
6627 : fp);
6628 :
6629 : Assert(rel->rd_opcoptions);
6630 :
6631 : /* write the vector of opcoptions values */
6632 477220 : for (i = 0; i < relform->relnatts; i++)
6633 : {
6634 297460 : bytea *opt = rel->rd_opcoptions[i];
6635 :
6636 297460 : write_item(opt, opt ? VARSIZE(opt) : 0, fp);
6637 : }
6638 : }
6639 : }
6640 :
6641 4280 : if (FreeFile(fp))
6642 0 : ereport(FATAL,
6643 : errcode_for_file_access(),
6644 : errmsg_internal("could not write init file: %m"));
6645 :
6646 : /*
6647 : * Now we have to check whether the data we've so painstakingly
6648 : * accumulated is already obsolete due to someone else's just-committed
6649 : * catalog changes. If so, we just delete the temp file and leave it to
6650 : * the next backend to try again. (Our own relcache entries will be
6651 : * updated by SI message processing, but we can't be sure whether what we
6652 : * wrote out was up-to-date.)
6653 : *
6654 : * This mustn't run concurrently with the code that unlinks an init file
6655 : * and sends SI messages, so grab a serialization lock for the duration.
6656 : */
6657 4280 : LWLockAcquire(RelCacheInitLock, LW_EXCLUSIVE);
6658 :
6659 : /* Make sure we have seen all incoming SI messages */
6660 4280 : AcceptInvalidationMessages();
6661 :
6662 : /*
6663 : * If we have received any SI relcache invals since backend start, assume
6664 : * we may have written out-of-date data.
6665 : */
6666 4280 : if (relcacheInvalsReceived == 0L)
6667 : {
6668 : /*
6669 : * OK, rename the temp file to its final name, deleting any
6670 : * previously-existing init file.
6671 : *
6672 : * Note: a failure here is possible under Cygwin, if some other
6673 : * backend is holding open an unlinked-but-not-yet-gone init file. So
6674 : * treat this as a noncritical failure; just remove the useless temp
6675 : * file on failure.
6676 : */
6677 4280 : if (rename(tempfilename, finalfilename) < 0)
6678 0 : unlink(tempfilename);
6679 : }
6680 : else
6681 : {
6682 : /* Delete the already-obsolete temp file */
6683 0 : unlink(tempfilename);
6684 : }
6685 :
6686 4280 : LWLockRelease(RelCacheInitLock);
6687 : }
6688 :
6689 : /* write a chunk of data preceded by its length */
6690 : static void
6691 3638000 : write_item(const void *data, Size len, FILE *fp)
6692 : {
6693 3638000 : if (fwrite(&len, 1, sizeof(len), fp) != sizeof(len))
6694 0 : ereport(FATAL,
6695 : errcode_for_file_access(),
6696 : errmsg_internal("could not write init file: %m"));
6697 3638000 : if (len > 0 && fwrite(data, 1, len, fp) != len)
6698 0 : ereport(FATAL,
6699 : errcode_for_file_access(),
6700 : errmsg_internal("could not write init file: %m"));
6701 3638000 : }
6702 :
6703 : /*
6704 : * Determine whether a given relation (identified by OID) is one of the ones
6705 : * we should store in a relcache init file.
6706 : *
6707 : * We must cache all nailed rels, and for efficiency we should cache every rel
6708 : * that supports a syscache. The former set is almost but not quite a subset
6709 : * of the latter. The special cases are relations where
6710 : * RelationCacheInitializePhase2/3 chooses to nail for efficiency reasons, but
6711 : * which do not support any syscache.
6712 : */
6713 : bool
6714 1887100 : RelationIdIsInInitFile(Oid relationId)
6715 : {
6716 1887100 : if (relationId == SharedSecLabelRelationId ||
6717 1884420 : relationId == TriggerRelidNameIndexId ||
6718 1884334 : relationId == DatabaseNameIndexId ||
6719 : relationId == SharedSecLabelObjectIndexId)
6720 : {
6721 : /*
6722 : * If this Assert fails, we don't need the applicable special case
6723 : * anymore.
6724 : */
6725 : Assert(!RelationSupportsSysCache(relationId));
6726 2878 : return true;
6727 : }
6728 1884222 : return RelationSupportsSysCache(relationId);
6729 : }
6730 :
6731 : /*
6732 : * Invalidate (remove) the init file during commit of a transaction that
6733 : * changed one or more of the relation cache entries that are kept in the
6734 : * local init file.
6735 : *
6736 : * To be safe against concurrent inspection or rewriting of the init file,
6737 : * we must take RelCacheInitLock, then remove the old init file, then send
6738 : * the SI messages that include relcache inval for such relations, and then
6739 : * release RelCacheInitLock. This serializes the whole affair against
6740 : * write_relcache_init_file, so that we can be sure that any other process
6741 : * that's concurrently trying to create a new init file won't move an
6742 : * already-stale version into place after we unlink. Also, because we unlink
6743 : * before sending the SI messages, a backend that's currently starting cannot
6744 : * read the now-obsolete init file and then miss the SI messages that will
6745 : * force it to update its relcache entries. (This works because the backend
6746 : * startup sequence gets into the sinval array before trying to load the init
6747 : * file.)
6748 : *
6749 : * We take the lock and do the unlink in RelationCacheInitFilePreInvalidate,
6750 : * then release the lock in RelationCacheInitFilePostInvalidate. Caller must
6751 : * send any pending SI messages between those calls.
6752 : */
6753 : void
6754 15328 : RelationCacheInitFilePreInvalidate(void)
6755 : {
6756 : char localinitfname[MAXPGPATH];
6757 : char sharedinitfname[MAXPGPATH];
6758 :
6759 15328 : if (DatabasePath)
6760 15328 : snprintf(localinitfname, sizeof(localinitfname), "%s/%s",
6761 : DatabasePath, RELCACHE_INIT_FILENAME);
6762 15328 : snprintf(sharedinitfname, sizeof(sharedinitfname), "global/%s",
6763 : RELCACHE_INIT_FILENAME);
6764 :
6765 15328 : LWLockAcquire(RelCacheInitLock, LW_EXCLUSIVE);
6766 :
6767 : /*
6768 : * The files might not be there if no backend has been started since the
6769 : * last removal. But complain about failures other than ENOENT with
6770 : * ERROR. Fortunately, it's not too late to abort the transaction if we
6771 : * can't get rid of the would-be-obsolete init file.
6772 : */
6773 15328 : if (DatabasePath)
6774 15328 : unlink_initfile(localinitfname, ERROR);
6775 15328 : unlink_initfile(sharedinitfname, ERROR);
6776 15328 : }
6777 :
6778 : void
6779 15328 : RelationCacheInitFilePostInvalidate(void)
6780 : {
6781 15328 : LWLockRelease(RelCacheInitLock);
6782 15328 : }
6783 :
6784 : /*
6785 : * Remove the init files during postmaster startup.
6786 : *
6787 : * We used to keep the init files across restarts, but that is unsafe in PITR
6788 : * scenarios, and even in simple crash-recovery cases there are windows for
6789 : * the init files to become out-of-sync with the database. So now we just
6790 : * remove them during startup and expect the first backend launch to rebuild
6791 : * them. Of course, this has to happen in each database of the cluster.
6792 : */
6793 : void
6794 1520 : RelationCacheInitFileRemove(void)
6795 : {
6796 1520 : const char *tblspcdir = "pg_tblspc";
6797 : DIR *dir;
6798 : struct dirent *de;
6799 : char path[MAXPGPATH + 10 + sizeof(TABLESPACE_VERSION_DIRECTORY)];
6800 :
6801 1520 : snprintf(path, sizeof(path), "global/%s",
6802 : RELCACHE_INIT_FILENAME);
6803 1520 : unlink_initfile(path, LOG);
6804 :
6805 : /* Scan everything in the default tablespace */
6806 1520 : RelationCacheInitFileRemoveInDir("base");
6807 :
6808 : /* Scan the tablespace link directory to find non-default tablespaces */
6809 1520 : dir = AllocateDir(tblspcdir);
6810 :
6811 4652 : while ((de = ReadDirExtended(dir, tblspcdir, LOG)) != NULL)
6812 : {
6813 3132 : if (strspn(de->d_name, "0123456789") == strlen(de->d_name))
6814 : {
6815 : /* Scan the tablespace dir for per-database dirs */
6816 92 : snprintf(path, sizeof(path), "%s/%s/%s",
6817 92 : tblspcdir, de->d_name, TABLESPACE_VERSION_DIRECTORY);
6818 92 : RelationCacheInitFileRemoveInDir(path);
6819 : }
6820 : }
6821 :
6822 1520 : FreeDir(dir);
6823 1520 : }
6824 :
6825 : /* Process one per-tablespace directory for RelationCacheInitFileRemove */
6826 : static void
6827 1612 : RelationCacheInitFileRemoveInDir(const char *tblspcpath)
6828 : {
6829 : DIR *dir;
6830 : struct dirent *de;
6831 : char initfilename[MAXPGPATH * 2];
6832 :
6833 : /* Scan the tablespace directory to find per-database directories */
6834 1612 : dir = AllocateDir(tblspcpath);
6835 :
6836 9776 : while ((de = ReadDirExtended(dir, tblspcpath, LOG)) != NULL)
6837 : {
6838 8164 : if (strspn(de->d_name, "0123456789") == strlen(de->d_name))
6839 : {
6840 : /* Try to remove the init file in each database */
6841 4810 : snprintf(initfilename, sizeof(initfilename), "%s/%s/%s",
6842 4810 : tblspcpath, de->d_name, RELCACHE_INIT_FILENAME);
6843 4810 : unlink_initfile(initfilename, LOG);
6844 : }
6845 : }
6846 :
6847 1612 : FreeDir(dir);
6848 1612 : }
6849 :
6850 : static void
6851 36986 : unlink_initfile(const char *initfilename, int elevel)
6852 : {
6853 36986 : if (unlink(initfilename) < 0)
6854 : {
6855 : /* It might not be there, but log any error other than ENOENT */
6856 35240 : if (errno != ENOENT)
6857 0 : ereport(elevel,
6858 : (errcode_for_file_access(),
6859 : errmsg("could not remove cache file \"%s\": %m",
6860 : initfilename)));
6861 : }
6862 36986 : }
6863 :
6864 : /*
6865 : * ResourceOwner callbacks
6866 : */
6867 : static char *
6868 0 : ResOwnerPrintRelCache(Datum res)
6869 : {
6870 0 : Relation rel = (Relation) DatumGetPointer(res);
6871 :
6872 0 : return psprintf("relation \"%s\"", RelationGetRelationName(rel));
6873 : }
6874 :
6875 : static void
6876 35308 : ResOwnerReleaseRelation(Datum res)
6877 : {
6878 35308 : Relation rel = (Relation) DatumGetPointer(res);
6879 :
6880 : /*
6881 : * This reference has already been removed from the resource owner, so
6882 : * just decrement reference count without calling
6883 : * ResourceOwnerForgetRelationRef.
6884 : */
6885 : Assert(rel->rd_refcnt > 0);
6886 35308 : rel->rd_refcnt -= 1;
6887 :
6888 35308 : RelationCloseCleanup((Relation) res);
6889 35308 : }
|