Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * relcache.c
4 : * POSTGRES relation descriptor cache code
5 : *
6 : * Portions Copyright (c) 1996-2025, 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 RelationInvalidateRelation(Relation relation);
279 : static void RelationClearRelation(Relation relation);
280 : static void RelationRebuildRelation(Relation relation);
281 :
282 : static void RelationReloadIndexInfo(Relation relation);
283 : static void RelationReloadNailed(Relation relation);
284 : static void RelationFlushRelation(Relation relation);
285 : static void RememberToFreeTupleDescAtEOX(TupleDesc td);
286 : #ifdef USE_ASSERT_CHECKING
287 : static void AssertPendingSyncConsistency(Relation relation);
288 : #endif
289 : static void AtEOXact_cleanup(Relation relation, bool isCommit);
290 : static void AtEOSubXact_cleanup(Relation relation, bool isCommit,
291 : SubTransactionId mySubid, SubTransactionId parentSubid);
292 : static bool load_relcache_init_file(bool shared);
293 : static void write_relcache_init_file(bool shared);
294 : static void write_item(const void *data, Size len, FILE *fp);
295 :
296 : static void formrdesc(const char *relationName, Oid relationReltype,
297 : bool isshared, int natts, const FormData_pg_attribute *attrs);
298 :
299 : static HeapTuple ScanPgRelation(Oid targetRelId, bool indexOK, bool force_non_historic);
300 : static Relation AllocateRelationDesc(Form_pg_class relp);
301 : static void RelationParseRelOptions(Relation relation, HeapTuple tuple);
302 : static void RelationBuildTupleDesc(Relation relation);
303 : static Relation RelationBuildDesc(Oid targetRelId, bool insertIt);
304 : static void RelationInitPhysicalAddr(Relation relation);
305 : static void load_critical_index(Oid indexoid, Oid heapoid);
306 : static TupleDesc GetPgClassDescriptor(void);
307 : static TupleDesc GetPgIndexDescriptor(void);
308 : static void AttrDefaultFetch(Relation relation, int ndef);
309 : static int AttrDefaultCmp(const void *a, const void *b);
310 : static void CheckNNConstraintFetch(Relation relation);
311 : static int CheckConstraintCmp(const void *a, const void *b);
312 : static void InitIndexAmRoutine(Relation relation);
313 : static void IndexSupportInitialize(oidvector *indclass,
314 : RegProcedure *indexSupport,
315 : Oid *opFamily,
316 : Oid *opcInType,
317 : StrategyNumber maxSupportNumber,
318 : AttrNumber maxAttributeNumber);
319 : static OpClassCacheEnt *LookupOpclassInfo(Oid operatorClassOid,
320 : StrategyNumber numSupport);
321 : static void RelationCacheInitFileRemoveInDir(const char *tblspcpath);
322 : static void unlink_initfile(const char *initfilename, int elevel);
323 :
324 :
325 : /*
326 : * ScanPgRelation
327 : *
328 : * This is used by RelationBuildDesc to find a pg_class
329 : * tuple matching targetRelId. The caller must hold at least
330 : * AccessShareLock on the target relid to prevent concurrent-update
331 : * scenarios; it isn't guaranteed that all scans used to build the
332 : * relcache entry will use the same snapshot. If, for example,
333 : * an attribute were to be added after scanning pg_class and before
334 : * scanning pg_attribute, relnatts wouldn't match.
335 : *
336 : * NB: the returned tuple has been copied into palloc'd storage
337 : * and must eventually be freed with heap_freetuple.
338 : */
339 : static HeapTuple
340 1918432 : ScanPgRelation(Oid targetRelId, bool indexOK, bool force_non_historic)
341 : {
342 : HeapTuple pg_class_tuple;
343 : Relation pg_class_desc;
344 : SysScanDesc pg_class_scan;
345 : ScanKeyData key[1];
346 1918432 : Snapshot snapshot = NULL;
347 :
348 : /*
349 : * If something goes wrong during backend startup, we might find ourselves
350 : * trying to read pg_class before we've selected a database. That ain't
351 : * gonna work, so bail out with a useful error message. If this happens,
352 : * it probably means a relcache entry that needs to be nailed isn't.
353 : */
354 1918432 : if (!OidIsValid(MyDatabaseId))
355 0 : elog(FATAL, "cannot read pg_class without having selected a database");
356 :
357 : /*
358 : * form a scan key
359 : */
360 1918432 : ScanKeyInit(&key[0],
361 : Anum_pg_class_oid,
362 : BTEqualStrategyNumber, F_OIDEQ,
363 : ObjectIdGetDatum(targetRelId));
364 :
365 : /*
366 : * Open pg_class and fetch a tuple. Force heap scan if we haven't yet
367 : * built the critical relcache entries (this includes initdb and startup
368 : * without a pg_internal.init file). The caller can also force a heap
369 : * scan by setting indexOK == false.
370 : */
371 1918432 : pg_class_desc = table_open(RelationRelationId, AccessShareLock);
372 :
373 : /*
374 : * The caller might need a tuple that's newer than what's visible to the
375 : * historic snapshot; currently the only case requiring to do so is
376 : * looking up the relfilenumber of non mapped system relations during
377 : * decoding.
378 : */
379 1918432 : if (force_non_historic)
380 3274 : snapshot = RegisterSnapshot(GetNonHistoricCatalogSnapshot(RelationRelationId));
381 :
382 1918432 : pg_class_scan = systable_beginscan(pg_class_desc, ClassOidIndexId,
383 1918432 : indexOK && criticalRelcachesBuilt,
384 : snapshot,
385 : 1, key);
386 :
387 1918426 : pg_class_tuple = systable_getnext(pg_class_scan);
388 :
389 : /*
390 : * Must copy tuple before releasing buffer.
391 : */
392 1918418 : if (HeapTupleIsValid(pg_class_tuple))
393 1918408 : pg_class_tuple = heap_copytuple(pg_class_tuple);
394 :
395 : /* all done */
396 1918418 : systable_endscan(pg_class_scan);
397 :
398 1918418 : if (snapshot)
399 3274 : UnregisterSnapshot(snapshot);
400 :
401 1918418 : table_close(pg_class_desc, AccessShareLock);
402 :
403 1918418 : return pg_class_tuple;
404 : }
405 :
406 : /*
407 : * AllocateRelationDesc
408 : *
409 : * This is used to allocate memory for a new relation descriptor
410 : * and initialize the rd_rel field from the given pg_class tuple.
411 : */
412 : static Relation
413 1767428 : AllocateRelationDesc(Form_pg_class relp)
414 : {
415 : Relation relation;
416 : MemoryContext oldcxt;
417 : Form_pg_class relationForm;
418 :
419 : /* Relcache entries must live in CacheMemoryContext */
420 1767428 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
421 :
422 : /*
423 : * allocate and zero space for new relation descriptor
424 : */
425 1767428 : relation = (Relation) palloc0(sizeof(RelationData));
426 :
427 : /* make sure relation is marked as having no open file yet */
428 1767428 : relation->rd_smgr = NULL;
429 :
430 : /*
431 : * Copy the relation tuple form
432 : *
433 : * We only allocate space for the fixed fields, ie, CLASS_TUPLE_SIZE. The
434 : * variable-length fields (relacl, reloptions) are NOT stored in the
435 : * relcache --- there'd be little point in it, since we don't copy the
436 : * tuple's nulls bitmap and hence wouldn't know if the values are valid.
437 : * Bottom line is that relacl *cannot* be retrieved from the relcache. Get
438 : * it from the syscache if you need it. The same goes for the original
439 : * form of reloptions (however, we do store the parsed form of reloptions
440 : * in rd_options).
441 : */
442 1767428 : relationForm = (Form_pg_class) palloc(CLASS_TUPLE_SIZE);
443 :
444 1767428 : memcpy(relationForm, relp, CLASS_TUPLE_SIZE);
445 :
446 : /* initialize relation tuple form */
447 1767428 : relation->rd_rel = relationForm;
448 :
449 : /* and allocate attribute tuple form storage */
450 1767428 : relation->rd_att = CreateTemplateTupleDesc(relationForm->relnatts);
451 : /* which we mark as a reference-counted tupdesc */
452 1767428 : relation->rd_att->tdrefcount = 1;
453 :
454 1767428 : MemoryContextSwitchTo(oldcxt);
455 :
456 1767428 : return relation;
457 : }
458 :
459 : /*
460 : * RelationParseRelOptions
461 : * Convert pg_class.reloptions into pre-parsed rd_options
462 : *
463 : * tuple is the real pg_class tuple (not rd_rel!) for relation
464 : *
465 : * Note: rd_rel and (if an index) rd_indam must be valid already
466 : */
467 : static void
468 1910416 : RelationParseRelOptions(Relation relation, HeapTuple tuple)
469 : {
470 : bytea *options;
471 : amoptions_function amoptsfn;
472 :
473 1910416 : relation->rd_options = NULL;
474 :
475 : /*
476 : * Look up any AM-specific parse function; fall out if relkind should not
477 : * have options.
478 : */
479 1910416 : switch (relation->rd_rel->relkind)
480 : {
481 1039956 : case RELKIND_RELATION:
482 : case RELKIND_TOASTVALUE:
483 : case RELKIND_VIEW:
484 : case RELKIND_MATVIEW:
485 : case RELKIND_PARTITIONED_TABLE:
486 1039956 : amoptsfn = NULL;
487 1039956 : break;
488 851476 : case RELKIND_INDEX:
489 : case RELKIND_PARTITIONED_INDEX:
490 851476 : amoptsfn = relation->rd_indam->amoptions;
491 851476 : break;
492 18984 : default:
493 18984 : return;
494 : }
495 :
496 : /*
497 : * Fetch reloptions from tuple; have to use a hardwired descriptor because
498 : * we might not have any other for pg_class yet (consider executing this
499 : * code for pg_class itself)
500 : */
501 1891432 : options = extractRelOptions(tuple, GetPgClassDescriptor(), amoptsfn);
502 :
503 : /*
504 : * Copy parsed data into CacheMemoryContext. To guard against the
505 : * possibility of leaks in the reloptions code, we want to do the actual
506 : * parsing in the caller's memory context and copy the results into
507 : * CacheMemoryContext after the fact.
508 : */
509 1891432 : if (options)
510 : {
511 39104 : relation->rd_options = MemoryContextAlloc(CacheMemoryContext,
512 19552 : VARSIZE(options));
513 19552 : memcpy(relation->rd_options, options, VARSIZE(options));
514 19552 : pfree(options);
515 : }
516 : }
517 :
518 : /*
519 : * RelationBuildTupleDesc
520 : *
521 : * Form the relation's tuple descriptor from information in
522 : * the pg_attribute, pg_attrdef & pg_constraint system catalogs.
523 : */
524 : static void
525 1767428 : RelationBuildTupleDesc(Relation relation)
526 : {
527 : HeapTuple pg_attribute_tuple;
528 : Relation pg_attribute_desc;
529 : SysScanDesc pg_attribute_scan;
530 : ScanKeyData skey[2];
531 : int need;
532 : TupleConstr *constr;
533 1767428 : AttrMissing *attrmiss = NULL;
534 1767428 : int ndef = 0;
535 :
536 : /* fill rd_att's type ID fields (compare heap.c's AddNewRelationTuple) */
537 1767428 : relation->rd_att->tdtypeid =
538 1767428 : relation->rd_rel->reltype ? relation->rd_rel->reltype : RECORDOID;
539 1767428 : relation->rd_att->tdtypmod = -1; /* just to be sure */
540 :
541 1767428 : constr = (TupleConstr *) MemoryContextAllocZero(CacheMemoryContext,
542 : sizeof(TupleConstr));
543 :
544 : /*
545 : * Form a scan key that selects only user attributes (attnum > 0).
546 : * (Eliminating system attribute rows at the index level is lots faster
547 : * than fetching them.)
548 : */
549 1767428 : ScanKeyInit(&skey[0],
550 : Anum_pg_attribute_attrelid,
551 : BTEqualStrategyNumber, F_OIDEQ,
552 : ObjectIdGetDatum(RelationGetRelid(relation)));
553 1767428 : ScanKeyInit(&skey[1],
554 : Anum_pg_attribute_attnum,
555 : BTGreaterStrategyNumber, F_INT2GT,
556 : Int16GetDatum(0));
557 :
558 : /*
559 : * Open pg_attribute and begin a scan. Force heap scan if we haven't yet
560 : * built the critical relcache entries (this includes initdb and startup
561 : * without a pg_internal.init file).
562 : */
563 1767428 : pg_attribute_desc = table_open(AttributeRelationId, AccessShareLock);
564 1767428 : pg_attribute_scan = systable_beginscan(pg_attribute_desc,
565 : AttributeRelidNumIndexId,
566 : criticalRelcachesBuilt,
567 : NULL,
568 : 2, skey);
569 :
570 : /*
571 : * add attribute data to relation->rd_att
572 : */
573 1767428 : need = RelationGetNumberOfAttributes(relation);
574 :
575 6128244 : while (HeapTupleIsValid(pg_attribute_tuple = systable_getnext(pg_attribute_scan)))
576 : {
577 : Form_pg_attribute attp;
578 : int attnum;
579 :
580 6117756 : attp = (Form_pg_attribute) GETSTRUCT(pg_attribute_tuple);
581 :
582 6117756 : attnum = attp->attnum;
583 6117756 : if (attnum <= 0 || attnum > RelationGetNumberOfAttributes(relation))
584 0 : elog(ERROR, "invalid attribute number %d for relation \"%s\"",
585 : attp->attnum, RelationGetRelationName(relation));
586 :
587 6117756 : memcpy(TupleDescAttr(relation->rd_att, attnum - 1),
588 : attp,
589 : ATTRIBUTE_FIXED_PART_SIZE);
590 :
591 6117756 : populate_compact_attribute(relation->rd_att, attnum - 1);
592 :
593 : /* Update constraint/default info */
594 6117756 : if (attp->attnotnull)
595 2508850 : constr->has_not_null = true;
596 6117756 : if (attp->attgenerated == ATTRIBUTE_GENERATED_STORED)
597 10680 : constr->has_generated_stored = true;
598 6117756 : if (attp->attgenerated == ATTRIBUTE_GENERATED_VIRTUAL)
599 7458 : constr->has_generated_virtual = true;
600 6117756 : if (attp->atthasdef)
601 53218 : ndef++;
602 :
603 : /* If the column has a "missing" value, put it in the attrmiss array */
604 6117756 : if (attp->atthasmissing)
605 : {
606 : Datum missingval;
607 : bool missingNull;
608 :
609 : /* Do we have a missing value? */
610 7898 : missingval = heap_getattr(pg_attribute_tuple,
611 : Anum_pg_attribute_attmissingval,
612 : pg_attribute_desc->rd_att,
613 : &missingNull);
614 7898 : if (!missingNull)
615 : {
616 : /* Yes, fetch from the array */
617 : MemoryContext oldcxt;
618 : bool is_null;
619 7898 : int one = 1;
620 : Datum missval;
621 :
622 7898 : if (attrmiss == NULL)
623 : attrmiss = (AttrMissing *)
624 3758 : MemoryContextAllocZero(CacheMemoryContext,
625 3758 : relation->rd_rel->relnatts *
626 : sizeof(AttrMissing));
627 :
628 7898 : missval = array_get_element(missingval,
629 : 1,
630 : &one,
631 : -1,
632 7898 : attp->attlen,
633 7898 : attp->attbyval,
634 7898 : attp->attalign,
635 : &is_null);
636 : Assert(!is_null);
637 7898 : if (attp->attbyval)
638 : {
639 : /* for copy by val just copy the datum direct */
640 4950 : attrmiss[attnum - 1].am_value = missval;
641 : }
642 : else
643 : {
644 : /* otherwise copy in the correct context */
645 2948 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
646 5896 : attrmiss[attnum - 1].am_value = datumCopy(missval,
647 2948 : attp->attbyval,
648 2948 : attp->attlen);
649 2948 : MemoryContextSwitchTo(oldcxt);
650 : }
651 7898 : attrmiss[attnum - 1].am_present = true;
652 : }
653 : }
654 6117756 : need--;
655 6117756 : if (need == 0)
656 1756940 : break;
657 : }
658 :
659 : /*
660 : * end the scan and close the attribute relation
661 : */
662 1767428 : systable_endscan(pg_attribute_scan);
663 1767428 : table_close(pg_attribute_desc, AccessShareLock);
664 :
665 1767428 : if (need != 0)
666 0 : elog(ERROR, "pg_attribute catalog is missing %d attribute(s) for relation OID %u",
667 : need, RelationGetRelid(relation));
668 :
669 : /*
670 : * We can easily set the attcacheoff value for the first attribute: it
671 : * must be zero. This eliminates the need for special cases for attnum=1
672 : * that used to exist in fastgetattr() and index_getattr().
673 : */
674 1767428 : if (RelationGetNumberOfAttributes(relation) > 0)
675 1756940 : TupleDescCompactAttr(relation->rd_att, 0)->attcacheoff = 0;
676 :
677 : /*
678 : * Set up constraint/default info
679 : */
680 1767428 : if (constr->has_not_null ||
681 1246978 : constr->has_generated_stored ||
682 1243140 : constr->has_generated_virtual ||
683 1236486 : ndef > 0 ||
684 1236438 : attrmiss ||
685 1236438 : relation->rd_rel->relchecks > 0)
686 536742 : {
687 536742 : bool is_catalog = IsCatalogRelation(relation);
688 :
689 536742 : relation->rd_att->constr = constr;
690 :
691 536742 : if (ndef > 0) /* DEFAULTs */
692 37538 : AttrDefaultFetch(relation, ndef);
693 : else
694 499204 : constr->num_defval = 0;
695 :
696 536742 : constr->missing = attrmiss;
697 :
698 : /* CHECK and NOT NULLs */
699 536742 : if (relation->rd_rel->relchecks > 0 ||
700 523766 : (!is_catalog && constr->has_not_null))
701 175010 : CheckNNConstraintFetch(relation);
702 :
703 : /*
704 : * Any not-null constraint that wasn't marked invalid by
705 : * CheckNNConstraintFetch must necessarily be valid; make it so in the
706 : * CompactAttribute array.
707 : */
708 536742 : if (!is_catalog)
709 : {
710 638052 : for (int i = 0; i < relation->rd_rel->relnatts; i++)
711 : {
712 : CompactAttribute *attr;
713 :
714 452930 : attr = TupleDescCompactAttr(relation->rd_att, i);
715 :
716 452930 : if (attr->attnullability == ATTNULLABLE_UNKNOWN)
717 249286 : attr->attnullability = ATTNULLABLE_VALID;
718 : else
719 : Assert(attr->attnullability == ATTNULLABLE_INVALID ||
720 : attr->attnullability == ATTNULLABLE_UNRESTRICTED);
721 : }
722 : }
723 :
724 536742 : if (relation->rd_rel->relchecks == 0)
725 523766 : constr->num_check = 0;
726 : }
727 : else
728 : {
729 1230686 : pfree(constr);
730 1230686 : relation->rd_att->constr = NULL;
731 : }
732 1767428 : }
733 :
734 : /*
735 : * RelationBuildRuleLock
736 : *
737 : * Form the relation's rewrite rules from information in
738 : * the pg_rewrite system catalog.
739 : *
740 : * Note: The rule parsetrees are potentially very complex node structures.
741 : * To allow these trees to be freed when the relcache entry is flushed,
742 : * we make a private memory context to hold the RuleLock information for
743 : * each relcache entry that has associated rules. The context is used
744 : * just for rule info, not for any other subsidiary data of the relcache
745 : * entry, because that keeps the update logic in RelationRebuildRelation()
746 : * manageable. The other subsidiary data structures are simple enough
747 : * to be easy to free explicitly, anyway.
748 : *
749 : * Note: The relation's reloptions must have been extracted first.
750 : */
751 : static void
752 40178 : RelationBuildRuleLock(Relation relation)
753 : {
754 : MemoryContext rulescxt;
755 : MemoryContext oldcxt;
756 : HeapTuple rewrite_tuple;
757 : Relation rewrite_desc;
758 : TupleDesc rewrite_tupdesc;
759 : SysScanDesc rewrite_scan;
760 : ScanKeyData key;
761 : RuleLock *rulelock;
762 : int numlocks;
763 : RewriteRule **rules;
764 : int maxlocks;
765 :
766 : /*
767 : * Make the private context. Assume it'll not contain much data.
768 : */
769 40178 : rulescxt = AllocSetContextCreate(CacheMemoryContext,
770 : "relation rules",
771 : ALLOCSET_SMALL_SIZES);
772 40178 : relation->rd_rulescxt = rulescxt;
773 40178 : MemoryContextCopyAndSetIdentifier(rulescxt,
774 : RelationGetRelationName(relation));
775 :
776 : /*
777 : * allocate an array to hold the rewrite rules (the array is extended if
778 : * necessary)
779 : */
780 40178 : maxlocks = 4;
781 : rules = (RewriteRule **)
782 40178 : MemoryContextAlloc(rulescxt, sizeof(RewriteRule *) * maxlocks);
783 40178 : numlocks = 0;
784 :
785 : /*
786 : * form a scan key
787 : */
788 40178 : ScanKeyInit(&key,
789 : Anum_pg_rewrite_ev_class,
790 : BTEqualStrategyNumber, F_OIDEQ,
791 : ObjectIdGetDatum(RelationGetRelid(relation)));
792 :
793 : /*
794 : * open pg_rewrite and begin a scan
795 : *
796 : * Note: since we scan the rules using RewriteRelRulenameIndexId, we will
797 : * be reading the rules in name order, except possibly during
798 : * emergency-recovery operations (ie, IgnoreSystemIndexes). This in turn
799 : * ensures that rules will be fired in name order.
800 : */
801 40178 : rewrite_desc = table_open(RewriteRelationId, AccessShareLock);
802 40178 : rewrite_tupdesc = RelationGetDescr(rewrite_desc);
803 40178 : rewrite_scan = systable_beginscan(rewrite_desc,
804 : RewriteRelRulenameIndexId,
805 : true, NULL,
806 : 1, &key);
807 :
808 80272 : while (HeapTupleIsValid(rewrite_tuple = systable_getnext(rewrite_scan)))
809 : {
810 40094 : Form_pg_rewrite rewrite_form = (Form_pg_rewrite) GETSTRUCT(rewrite_tuple);
811 : bool isnull;
812 : Datum rule_datum;
813 : char *rule_str;
814 : RewriteRule *rule;
815 : Oid check_as_user;
816 :
817 40094 : rule = (RewriteRule *) MemoryContextAlloc(rulescxt,
818 : sizeof(RewriteRule));
819 :
820 40094 : rule->ruleId = rewrite_form->oid;
821 :
822 40094 : rule->event = rewrite_form->ev_type - '0';
823 40094 : rule->enabled = rewrite_form->ev_enabled;
824 40094 : rule->isInstead = rewrite_form->is_instead;
825 :
826 : /*
827 : * Must use heap_getattr to fetch ev_action and ev_qual. Also, the
828 : * rule strings are often large enough to be toasted. To avoid
829 : * leaking memory in the caller's context, do the detoasting here so
830 : * we can free the detoasted version.
831 : */
832 40094 : rule_datum = heap_getattr(rewrite_tuple,
833 : Anum_pg_rewrite_ev_action,
834 : rewrite_tupdesc,
835 : &isnull);
836 : Assert(!isnull);
837 40094 : rule_str = TextDatumGetCString(rule_datum);
838 40094 : oldcxt = MemoryContextSwitchTo(rulescxt);
839 40094 : rule->actions = (List *) stringToNode(rule_str);
840 40094 : MemoryContextSwitchTo(oldcxt);
841 40094 : pfree(rule_str);
842 :
843 40094 : rule_datum = heap_getattr(rewrite_tuple,
844 : Anum_pg_rewrite_ev_qual,
845 : rewrite_tupdesc,
846 : &isnull);
847 : Assert(!isnull);
848 40094 : rule_str = TextDatumGetCString(rule_datum);
849 40094 : oldcxt = MemoryContextSwitchTo(rulescxt);
850 40094 : rule->qual = (Node *) stringToNode(rule_str);
851 40094 : MemoryContextSwitchTo(oldcxt);
852 40094 : pfree(rule_str);
853 :
854 : /*
855 : * If this is a SELECT rule defining a view, and the view has
856 : * "security_invoker" set, we must perform all permissions checks on
857 : * relations referred to by the rule as the invoking user.
858 : *
859 : * In all other cases (including non-SELECT rules on security invoker
860 : * views), perform the permissions checks as the relation owner.
861 : */
862 40094 : if (rule->event == CMD_SELECT &&
863 36172 : relation->rd_rel->relkind == RELKIND_VIEW &&
864 31974 : RelationHasSecurityInvoker(relation))
865 168 : check_as_user = InvalidOid;
866 : else
867 39926 : check_as_user = relation->rd_rel->relowner;
868 :
869 : /*
870 : * Scan through the rule's actions and set the checkAsUser field on
871 : * all RTEPermissionInfos. We have to look at the qual as well, in
872 : * case it contains sublinks.
873 : *
874 : * The reason for doing this when the rule is loaded, rather than when
875 : * it is stored, is that otherwise ALTER TABLE OWNER would have to
876 : * grovel through stored rules to update checkAsUser fields. Scanning
877 : * the rule tree during load is relatively cheap (compared to
878 : * constructing it in the first place), so we do it here.
879 : */
880 40094 : setRuleCheckAsUser((Node *) rule->actions, check_as_user);
881 40094 : setRuleCheckAsUser(rule->qual, check_as_user);
882 :
883 40094 : if (numlocks >= maxlocks)
884 : {
885 44 : maxlocks *= 2;
886 : rules = (RewriteRule **)
887 44 : repalloc(rules, sizeof(RewriteRule *) * maxlocks);
888 : }
889 40094 : rules[numlocks++] = rule;
890 : }
891 :
892 : /*
893 : * end the scan and close the attribute relation
894 : */
895 40178 : systable_endscan(rewrite_scan);
896 40178 : table_close(rewrite_desc, AccessShareLock);
897 :
898 : /*
899 : * there might not be any rules (if relhasrules is out-of-date)
900 : */
901 40178 : if (numlocks == 0)
902 : {
903 2900 : relation->rd_rules = NULL;
904 2900 : relation->rd_rulescxt = NULL;
905 2900 : MemoryContextDelete(rulescxt);
906 2900 : return;
907 : }
908 :
909 : /*
910 : * form a RuleLock and insert into relation
911 : */
912 37278 : rulelock = (RuleLock *) MemoryContextAlloc(rulescxt, sizeof(RuleLock));
913 37278 : rulelock->numLocks = numlocks;
914 37278 : rulelock->rules = rules;
915 :
916 37278 : relation->rd_rules = rulelock;
917 : }
918 :
919 : /*
920 : * equalRuleLocks
921 : *
922 : * Determine whether two RuleLocks are equivalent
923 : *
924 : * Probably this should be in the rules code someplace...
925 : */
926 : static bool
927 439776 : equalRuleLocks(RuleLock *rlock1, RuleLock *rlock2)
928 : {
929 : int i;
930 :
931 : /*
932 : * As of 7.3 we assume the rule ordering is repeatable, because
933 : * RelationBuildRuleLock should read 'em in a consistent order. So just
934 : * compare corresponding slots.
935 : */
936 439776 : if (rlock1 != NULL)
937 : {
938 2824 : if (rlock2 == NULL)
939 68 : return false;
940 2756 : if (rlock1->numLocks != rlock2->numLocks)
941 6 : return false;
942 5252 : for (i = 0; i < rlock1->numLocks; i++)
943 : {
944 2798 : RewriteRule *rule1 = rlock1->rules[i];
945 2798 : RewriteRule *rule2 = rlock2->rules[i];
946 :
947 2798 : if (rule1->ruleId != rule2->ruleId)
948 0 : return false;
949 2798 : if (rule1->event != rule2->event)
950 0 : return false;
951 2798 : if (rule1->enabled != rule2->enabled)
952 52 : return false;
953 2746 : if (rule1->isInstead != rule2->isInstead)
954 0 : return false;
955 2746 : if (!equal(rule1->qual, rule2->qual))
956 0 : return false;
957 2746 : if (!equal(rule1->actions, rule2->actions))
958 244 : return false;
959 : }
960 : }
961 436952 : else if (rlock2 != NULL)
962 17508 : return false;
963 421898 : return true;
964 : }
965 :
966 : /*
967 : * equalPolicy
968 : *
969 : * Determine whether two policies are equivalent
970 : */
971 : static bool
972 228 : equalPolicy(RowSecurityPolicy *policy1, RowSecurityPolicy *policy2)
973 : {
974 : int i;
975 : Oid *r1,
976 : *r2;
977 :
978 228 : if (policy1 != NULL)
979 : {
980 228 : if (policy2 == NULL)
981 0 : return false;
982 :
983 228 : if (policy1->polcmd != policy2->polcmd)
984 0 : return false;
985 228 : if (policy1->hassublinks != policy2->hassublinks)
986 0 : return false;
987 228 : if (strcmp(policy1->policy_name, policy2->policy_name) != 0)
988 0 : return false;
989 228 : if (ARR_DIMS(policy1->roles)[0] != ARR_DIMS(policy2->roles)[0])
990 0 : return false;
991 :
992 228 : r1 = (Oid *) ARR_DATA_PTR(policy1->roles);
993 228 : r2 = (Oid *) ARR_DATA_PTR(policy2->roles);
994 :
995 456 : for (i = 0; i < ARR_DIMS(policy1->roles)[0]; i++)
996 : {
997 228 : if (r1[i] != r2[i])
998 0 : return false;
999 : }
1000 :
1001 228 : if (!equal(policy1->qual, policy2->qual))
1002 0 : return false;
1003 228 : if (!equal(policy1->with_check_qual, policy2->with_check_qual))
1004 0 : return false;
1005 : }
1006 0 : else if (policy2 != NULL)
1007 0 : return false;
1008 :
1009 228 : return true;
1010 : }
1011 :
1012 : /*
1013 : * equalRSDesc
1014 : *
1015 : * Determine whether two RowSecurityDesc's are equivalent
1016 : */
1017 : static bool
1018 439776 : equalRSDesc(RowSecurityDesc *rsdesc1, RowSecurityDesc *rsdesc2)
1019 : {
1020 : ListCell *lc,
1021 : *rc;
1022 :
1023 439776 : if (rsdesc1 == NULL && rsdesc2 == NULL)
1024 439282 : return true;
1025 :
1026 494 : if ((rsdesc1 != NULL && rsdesc2 == NULL) ||
1027 308 : (rsdesc1 == NULL && rsdesc2 != NULL))
1028 318 : return false;
1029 :
1030 176 : if (list_length(rsdesc1->policies) != list_length(rsdesc2->policies))
1031 6 : return false;
1032 :
1033 : /* RelationBuildRowSecurity should build policies in order */
1034 398 : forboth(lc, rsdesc1->policies, rc, rsdesc2->policies)
1035 : {
1036 228 : RowSecurityPolicy *l = (RowSecurityPolicy *) lfirst(lc);
1037 228 : RowSecurityPolicy *r = (RowSecurityPolicy *) lfirst(rc);
1038 :
1039 228 : if (!equalPolicy(l, r))
1040 0 : return false;
1041 : }
1042 :
1043 170 : return true;
1044 : }
1045 :
1046 : /*
1047 : * RelationBuildDesc
1048 : *
1049 : * Build a relation descriptor. The caller must hold at least
1050 : * AccessShareLock on the target relid.
1051 : *
1052 : * The new descriptor is inserted into the hash table if insertIt is true.
1053 : *
1054 : * Returns NULL if no pg_class row could be found for the given relid
1055 : * (suggesting we are trying to access a just-deleted relation).
1056 : * Any other error is reported via elog.
1057 : */
1058 : static Relation
1059 1767406 : RelationBuildDesc(Oid targetRelId, bool insertIt)
1060 : {
1061 : int in_progress_offset;
1062 : Relation relation;
1063 : Oid relid;
1064 : HeapTuple pg_class_tuple;
1065 : Form_pg_class relp;
1066 :
1067 : /*
1068 : * This function and its subroutines can allocate a good deal of transient
1069 : * data in CurrentMemoryContext. Traditionally we've just leaked that
1070 : * data, reasoning that the caller's context is at worst of transaction
1071 : * scope, and relcache loads shouldn't happen so often that it's essential
1072 : * to recover transient data before end of statement/transaction. However
1073 : * that's definitely not true when debug_discard_caches is active, and
1074 : * perhaps it's not true in other cases.
1075 : *
1076 : * When debug_discard_caches is active or when forced to by
1077 : * RECOVER_RELATION_BUILD_MEMORY=1, arrange to allocate the junk in a
1078 : * temporary context that we'll free before returning. Make it a child of
1079 : * caller's context so that it will get cleaned up appropriately if we
1080 : * error out partway through.
1081 : */
1082 : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
1083 : MemoryContext tmpcxt = NULL;
1084 : MemoryContext oldcxt = NULL;
1085 :
1086 : if (RECOVER_RELATION_BUILD_MEMORY || debug_discard_caches > 0)
1087 : {
1088 : tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
1089 : "RelationBuildDesc workspace",
1090 : ALLOCSET_DEFAULT_SIZES);
1091 : oldcxt = MemoryContextSwitchTo(tmpcxt);
1092 : }
1093 : #endif
1094 :
1095 : /* Register to catch invalidation messages */
1096 1767406 : if (in_progress_list_len >= in_progress_list_maxlen)
1097 : {
1098 : int allocsize;
1099 :
1100 38 : allocsize = in_progress_list_maxlen * 2;
1101 38 : in_progress_list = repalloc(in_progress_list,
1102 : allocsize * sizeof(*in_progress_list));
1103 38 : in_progress_list_maxlen = allocsize;
1104 : }
1105 1767406 : in_progress_offset = in_progress_list_len++;
1106 1767406 : in_progress_list[in_progress_offset].reloid = targetRelId;
1107 1767440 : retry:
1108 1767440 : in_progress_list[in_progress_offset].invalidated = false;
1109 :
1110 : /*
1111 : * find the tuple in pg_class corresponding to the given relation id
1112 : */
1113 1767440 : pg_class_tuple = ScanPgRelation(targetRelId, true, false);
1114 :
1115 : /*
1116 : * if no such tuple exists, return NULL
1117 : */
1118 1767438 : if (!HeapTupleIsValid(pg_class_tuple))
1119 : {
1120 : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
1121 : if (tmpcxt)
1122 : {
1123 : /* Return to caller's context, and blow away the temporary context */
1124 : MemoryContextSwitchTo(oldcxt);
1125 : MemoryContextDelete(tmpcxt);
1126 : }
1127 : #endif
1128 : Assert(in_progress_offset + 1 == in_progress_list_len);
1129 10 : in_progress_list_len--;
1130 10 : return NULL;
1131 : }
1132 :
1133 : /*
1134 : * get information from the pg_class_tuple
1135 : */
1136 1767428 : relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
1137 1767428 : relid = relp->oid;
1138 : Assert(relid == targetRelId);
1139 :
1140 : /*
1141 : * allocate storage for the relation descriptor, and copy pg_class_tuple
1142 : * to relation->rd_rel.
1143 : */
1144 1767428 : relation = AllocateRelationDesc(relp);
1145 :
1146 : /*
1147 : * initialize the relation's relation id (relation->rd_id)
1148 : */
1149 1767428 : RelationGetRelid(relation) = relid;
1150 :
1151 : /*
1152 : * Normal relations are not nailed into the cache. Since we don't flush
1153 : * new relations, it won't be new. It could be temp though.
1154 : */
1155 1767428 : relation->rd_refcnt = 0;
1156 1767428 : relation->rd_isnailed = false;
1157 1767428 : relation->rd_createSubid = InvalidSubTransactionId;
1158 1767428 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
1159 1767428 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
1160 1767428 : relation->rd_droppedSubid = InvalidSubTransactionId;
1161 1767428 : switch (relation->rd_rel->relpersistence)
1162 : {
1163 1738258 : case RELPERSISTENCE_UNLOGGED:
1164 : case RELPERSISTENCE_PERMANENT:
1165 1738258 : relation->rd_backend = INVALID_PROC_NUMBER;
1166 1738258 : relation->rd_islocaltemp = false;
1167 1738258 : break;
1168 29170 : case RELPERSISTENCE_TEMP:
1169 29170 : if (isTempOrTempToastNamespace(relation->rd_rel->relnamespace))
1170 : {
1171 29134 : relation->rd_backend = ProcNumberForTempRelations();
1172 29134 : relation->rd_islocaltemp = true;
1173 : }
1174 : else
1175 : {
1176 : /*
1177 : * If it's a temp table, but not one of ours, we have to use
1178 : * the slow, grotty method to figure out the owning backend.
1179 : *
1180 : * Note: it's possible that rd_backend gets set to
1181 : * MyProcNumber here, in case we are looking at a pg_class
1182 : * entry left over from a crashed backend that coincidentally
1183 : * had the same ProcNumber we're using. We should *not*
1184 : * consider such a table to be "ours"; this is why we need the
1185 : * separate rd_islocaltemp flag. The pg_class entry will get
1186 : * flushed if/when we clean out the corresponding temp table
1187 : * namespace in preparation for using it.
1188 : */
1189 36 : relation->rd_backend =
1190 36 : GetTempNamespaceProcNumber(relation->rd_rel->relnamespace);
1191 : Assert(relation->rd_backend != INVALID_PROC_NUMBER);
1192 36 : relation->rd_islocaltemp = false;
1193 : }
1194 29170 : break;
1195 0 : default:
1196 0 : elog(ERROR, "invalid relpersistence: %c",
1197 : relation->rd_rel->relpersistence);
1198 : break;
1199 : }
1200 :
1201 : /*
1202 : * initialize the tuple descriptor (relation->rd_att).
1203 : */
1204 1767428 : RelationBuildTupleDesc(relation);
1205 :
1206 : /* foreign key data is not loaded till asked for */
1207 1767428 : relation->rd_fkeylist = NIL;
1208 1767428 : relation->rd_fkeyvalid = false;
1209 :
1210 : /* partitioning data is not loaded till asked for */
1211 1767428 : relation->rd_partkey = NULL;
1212 1767428 : relation->rd_partkeycxt = NULL;
1213 1767428 : relation->rd_partdesc = NULL;
1214 1767428 : relation->rd_partdesc_nodetached = NULL;
1215 1767428 : relation->rd_partdesc_nodetached_xmin = InvalidTransactionId;
1216 1767428 : relation->rd_pdcxt = NULL;
1217 1767428 : relation->rd_pddcxt = NULL;
1218 1767428 : relation->rd_partcheck = NIL;
1219 1767428 : relation->rd_partcheckvalid = false;
1220 1767428 : relation->rd_partcheckcxt = NULL;
1221 :
1222 : /*
1223 : * initialize access method information
1224 : */
1225 1767428 : if (relation->rd_rel->relkind == RELKIND_INDEX ||
1226 1037830 : relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
1227 735970 : RelationInitIndexAccessInfo(relation);
1228 1031458 : else if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) ||
1229 131502 : relation->rd_rel->relkind == RELKIND_SEQUENCE)
1230 906592 : RelationInitTableAccessMethod(relation);
1231 : else if (relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
1232 : {
1233 : /*
1234 : * Do nothing: access methods are a setting that partitions can
1235 : * inherit.
1236 : */
1237 : }
1238 : else
1239 : Assert(relation->rd_rel->relam == InvalidOid);
1240 :
1241 : /* extract reloptions if any */
1242 1767420 : RelationParseRelOptions(relation, pg_class_tuple);
1243 :
1244 : /*
1245 : * Fetch rules and triggers that affect this relation.
1246 : *
1247 : * Note that RelationBuildRuleLock() relies on this being done after
1248 : * extracting the relation's reloptions.
1249 : */
1250 1767420 : if (relation->rd_rel->relhasrules)
1251 40178 : RelationBuildRuleLock(relation);
1252 : else
1253 : {
1254 1727242 : relation->rd_rules = NULL;
1255 1727242 : relation->rd_rulescxt = NULL;
1256 : }
1257 :
1258 1767420 : if (relation->rd_rel->relhastriggers)
1259 63378 : RelationBuildTriggers(relation);
1260 : else
1261 1704042 : relation->trigdesc = NULL;
1262 :
1263 1767420 : if (relation->rd_rel->relrowsecurity)
1264 2138 : RelationBuildRowSecurity(relation);
1265 : else
1266 1765282 : relation->rd_rsdesc = NULL;
1267 :
1268 : /*
1269 : * initialize the relation lock manager information
1270 : */
1271 1767420 : RelationInitLockInfo(relation); /* see lmgr.c */
1272 :
1273 : /*
1274 : * initialize physical addressing information for the relation
1275 : */
1276 1767420 : RelationInitPhysicalAddr(relation);
1277 :
1278 : /* make sure relation is marked as having no open file yet */
1279 1767420 : relation->rd_smgr = NULL;
1280 :
1281 : /*
1282 : * now we can free the memory allocated for pg_class_tuple
1283 : */
1284 1767420 : heap_freetuple(pg_class_tuple);
1285 :
1286 : /*
1287 : * If an invalidation arrived mid-build, start over. Between here and the
1288 : * end of this function, don't add code that does or reasonably could read
1289 : * system catalogs. That range must be free from invalidation processing
1290 : * for the !insertIt case. For the insertIt case, RelationCacheInsert()
1291 : * will enroll this relation in ordinary relcache invalidation processing,
1292 : */
1293 1767420 : if (in_progress_list[in_progress_offset].invalidated)
1294 : {
1295 34 : RelationDestroyRelation(relation, false);
1296 34 : goto retry;
1297 : }
1298 : Assert(in_progress_offset + 1 == in_progress_list_len);
1299 1767386 : in_progress_list_len--;
1300 :
1301 : /*
1302 : * Insert newly created relation into relcache hash table, if requested.
1303 : *
1304 : * There is one scenario in which we might find a hashtable entry already
1305 : * present, even though our caller failed to find it: if the relation is a
1306 : * system catalog or index that's used during relcache load, we might have
1307 : * recursively created the same relcache entry during the preceding steps.
1308 : * So allow RelationCacheInsert to delete any already-present relcache
1309 : * entry for the same OID. The already-present entry should have refcount
1310 : * zero (else somebody forgot to close it); in the event that it doesn't,
1311 : * we'll elog a WARNING and leak the already-present entry.
1312 : */
1313 1767386 : if (insertIt)
1314 1327610 : RelationCacheInsert(relation, true);
1315 :
1316 : /* It's fully valid */
1317 1767386 : relation->rd_isvalid = true;
1318 :
1319 : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
1320 : if (tmpcxt)
1321 : {
1322 : /* Return to caller's context, and blow away the temporary context */
1323 : MemoryContextSwitchTo(oldcxt);
1324 : MemoryContextDelete(tmpcxt);
1325 : }
1326 : #endif
1327 :
1328 1767386 : return relation;
1329 : }
1330 :
1331 : /*
1332 : * Initialize the physical addressing info (RelFileLocator) for a relcache entry
1333 : *
1334 : * Note: at the physical level, relations in the pg_global tablespace must
1335 : * be treated as shared, even if relisshared isn't set. Hence we do not
1336 : * look at relisshared here.
1337 : */
1338 : static void
1339 6217084 : RelationInitPhysicalAddr(Relation relation)
1340 : {
1341 6217084 : RelFileNumber oldnumber = relation->rd_locator.relNumber;
1342 :
1343 : /* these relations kinds never have storage */
1344 6217084 : if (!RELKIND_HAS_STORAGE(relation->rd_rel->relkind))
1345 163864 : return;
1346 :
1347 6053220 : if (relation->rd_rel->reltablespace)
1348 959740 : relation->rd_locator.spcOid = relation->rd_rel->reltablespace;
1349 : else
1350 5093480 : relation->rd_locator.spcOid = MyDatabaseTableSpace;
1351 6053220 : if (relation->rd_locator.spcOid == GLOBALTABLESPACE_OID)
1352 956386 : relation->rd_locator.dbOid = InvalidOid;
1353 : else
1354 5096834 : relation->rd_locator.dbOid = MyDatabaseId;
1355 :
1356 6053220 : if (relation->rd_rel->relfilenode)
1357 : {
1358 : /*
1359 : * Even if we are using a decoding snapshot that doesn't represent the
1360 : * current state of the catalog we need to make sure the filenode
1361 : * points to the current file since the older file will be gone (or
1362 : * truncated). The new file will still contain older rows so lookups
1363 : * in them will work correctly. This wouldn't work correctly if
1364 : * rewrites were allowed to change the schema in an incompatible way,
1365 : * but those are prevented both on catalog tables and on user tables
1366 : * declared as additional catalog tables.
1367 : */
1368 4501858 : if (HistoricSnapshotActive()
1369 4796 : && RelationIsAccessibleInLogicalDecoding(relation)
1370 3274 : && IsTransactionState())
1371 : {
1372 : HeapTuple phys_tuple;
1373 : Form_pg_class physrel;
1374 :
1375 3274 : phys_tuple = ScanPgRelation(RelationGetRelid(relation),
1376 3274 : RelationGetRelid(relation) != ClassOidIndexId,
1377 : true);
1378 3274 : if (!HeapTupleIsValid(phys_tuple))
1379 0 : elog(ERROR, "could not find pg_class entry for %u",
1380 : RelationGetRelid(relation));
1381 3274 : physrel = (Form_pg_class) GETSTRUCT(phys_tuple);
1382 :
1383 3274 : relation->rd_rel->reltablespace = physrel->reltablespace;
1384 3274 : relation->rd_rel->relfilenode = physrel->relfilenode;
1385 3274 : heap_freetuple(phys_tuple);
1386 : }
1387 :
1388 4501858 : relation->rd_locator.relNumber = relation->rd_rel->relfilenode;
1389 : }
1390 : else
1391 : {
1392 : /* Consult the relation mapper */
1393 1551362 : relation->rd_locator.relNumber =
1394 1551362 : RelationMapOidToFilenumber(relation->rd_id,
1395 1551362 : relation->rd_rel->relisshared);
1396 1551362 : if (!RelFileNumberIsValid(relation->rd_locator.relNumber))
1397 0 : elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
1398 : RelationGetRelationName(relation), relation->rd_id);
1399 : }
1400 :
1401 : /*
1402 : * For RelationNeedsWAL() to answer correctly on parallel workers, restore
1403 : * rd_firstRelfilelocatorSubid. No subtransactions start or end while in
1404 : * parallel mode, so the specific SubTransactionId does not matter.
1405 : */
1406 6053220 : if (IsParallelWorker() && oldnumber != relation->rd_locator.relNumber)
1407 : {
1408 62568 : if (RelFileLocatorSkippingWAL(relation->rd_locator))
1409 304 : relation->rd_firstRelfilelocatorSubid = TopSubTransactionId;
1410 : else
1411 62264 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
1412 : }
1413 : }
1414 :
1415 : /*
1416 : * Fill in the IndexAmRoutine for an index relation.
1417 : *
1418 : * relation's rd_amhandler and rd_indexcxt must be valid already.
1419 : */
1420 : static void
1421 3212502 : InitIndexAmRoutine(Relation relation)
1422 : {
1423 : IndexAmRoutine *cached,
1424 : *tmp;
1425 :
1426 : /*
1427 : * Call the amhandler in current, short-lived memory context, just in case
1428 : * it leaks anything (it probably won't, but let's be paranoid).
1429 : */
1430 3212502 : tmp = GetIndexAmRoutine(relation->rd_amhandler);
1431 :
1432 : /* OK, now transfer the data into relation's rd_indexcxt. */
1433 3212502 : cached = (IndexAmRoutine *) MemoryContextAlloc(relation->rd_indexcxt,
1434 : sizeof(IndexAmRoutine));
1435 3212502 : memcpy(cached, tmp, sizeof(IndexAmRoutine));
1436 3212502 : relation->rd_indam = cached;
1437 :
1438 3212502 : pfree(tmp);
1439 3212502 : }
1440 :
1441 : /*
1442 : * Initialize index-access-method support data for an index relation
1443 : */
1444 : void
1445 752188 : RelationInitIndexAccessInfo(Relation relation)
1446 : {
1447 : HeapTuple tuple;
1448 : Form_pg_am aform;
1449 : Datum indcollDatum;
1450 : Datum indclassDatum;
1451 : Datum indoptionDatum;
1452 : bool isnull;
1453 : oidvector *indcoll;
1454 : oidvector *indclass;
1455 : int2vector *indoption;
1456 : MemoryContext indexcxt;
1457 : MemoryContext oldcontext;
1458 : int indnatts;
1459 : int indnkeyatts;
1460 : uint16 amsupport;
1461 :
1462 : /*
1463 : * Make a copy of the pg_index entry for the index. Since pg_index
1464 : * contains variable-length and possibly-null fields, we have to do this
1465 : * honestly rather than just treating it as a Form_pg_index struct.
1466 : */
1467 752188 : tuple = SearchSysCache1(INDEXRELID,
1468 : ObjectIdGetDatum(RelationGetRelid(relation)));
1469 752188 : if (!HeapTupleIsValid(tuple))
1470 0 : elog(ERROR, "cache lookup failed for index %u",
1471 : RelationGetRelid(relation));
1472 752188 : oldcontext = MemoryContextSwitchTo(CacheMemoryContext);
1473 752188 : relation->rd_indextuple = heap_copytuple(tuple);
1474 752188 : relation->rd_index = (Form_pg_index) GETSTRUCT(relation->rd_indextuple);
1475 752188 : MemoryContextSwitchTo(oldcontext);
1476 752188 : ReleaseSysCache(tuple);
1477 :
1478 : /*
1479 : * Look up the index's access method, save the OID of its handler function
1480 : */
1481 : Assert(relation->rd_rel->relam != InvalidOid);
1482 752188 : tuple = SearchSysCache1(AMOID, ObjectIdGetDatum(relation->rd_rel->relam));
1483 752186 : if (!HeapTupleIsValid(tuple))
1484 0 : elog(ERROR, "cache lookup failed for access method %u",
1485 : relation->rd_rel->relam);
1486 752186 : aform = (Form_pg_am) GETSTRUCT(tuple);
1487 752186 : relation->rd_amhandler = aform->amhandler;
1488 752186 : ReleaseSysCache(tuple);
1489 :
1490 752186 : indnatts = RelationGetNumberOfAttributes(relation);
1491 752186 : if (indnatts != IndexRelationGetNumberOfAttributes(relation))
1492 0 : elog(ERROR, "relnatts disagrees with indnatts for index %u",
1493 : RelationGetRelid(relation));
1494 752186 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(relation);
1495 :
1496 : /*
1497 : * Make the private context to hold index access info. The reason we need
1498 : * a context, and not just a couple of pallocs, is so that we won't leak
1499 : * any subsidiary info attached to fmgr lookup records.
1500 : */
1501 752186 : indexcxt = AllocSetContextCreate(CacheMemoryContext,
1502 : "index info",
1503 : ALLOCSET_SMALL_SIZES);
1504 752186 : relation->rd_indexcxt = indexcxt;
1505 752186 : MemoryContextCopyAndSetIdentifier(indexcxt,
1506 : RelationGetRelationName(relation));
1507 :
1508 : /*
1509 : * Now we can fetch the index AM's API struct
1510 : */
1511 752186 : InitIndexAmRoutine(relation);
1512 :
1513 : /*
1514 : * Allocate arrays to hold data. Opclasses are not used for included
1515 : * columns, so allocate them for indnkeyatts only.
1516 : */
1517 752186 : relation->rd_opfamily = (Oid *)
1518 752186 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
1519 752186 : relation->rd_opcintype = (Oid *)
1520 752186 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
1521 :
1522 752186 : amsupport = relation->rd_indam->amsupport;
1523 752186 : if (amsupport > 0)
1524 : {
1525 752186 : int nsupport = indnatts * amsupport;
1526 :
1527 752186 : relation->rd_support = (RegProcedure *)
1528 752186 : MemoryContextAllocZero(indexcxt, nsupport * sizeof(RegProcedure));
1529 752186 : relation->rd_supportinfo = (FmgrInfo *)
1530 752186 : MemoryContextAllocZero(indexcxt, nsupport * sizeof(FmgrInfo));
1531 : }
1532 : else
1533 : {
1534 0 : relation->rd_support = NULL;
1535 0 : relation->rd_supportinfo = NULL;
1536 : }
1537 :
1538 752186 : relation->rd_indcollation = (Oid *)
1539 752186 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
1540 :
1541 752186 : relation->rd_indoption = (int16 *)
1542 752186 : MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(int16));
1543 :
1544 : /*
1545 : * indcollation cannot be referenced directly through the C struct,
1546 : * because it comes after the variable-width indkey field. Must extract
1547 : * the datum the hard way...
1548 : */
1549 752186 : indcollDatum = fastgetattr(relation->rd_indextuple,
1550 : Anum_pg_index_indcollation,
1551 : GetPgIndexDescriptor(),
1552 : &isnull);
1553 : Assert(!isnull);
1554 752186 : indcoll = (oidvector *) DatumGetPointer(indcollDatum);
1555 752186 : memcpy(relation->rd_indcollation, indcoll->values, indnkeyatts * sizeof(Oid));
1556 :
1557 : /*
1558 : * indclass cannot be referenced directly through the C struct, because it
1559 : * comes after the variable-width indkey field. Must extract the datum
1560 : * the hard way...
1561 : */
1562 752186 : indclassDatum = fastgetattr(relation->rd_indextuple,
1563 : Anum_pg_index_indclass,
1564 : GetPgIndexDescriptor(),
1565 : &isnull);
1566 : Assert(!isnull);
1567 752186 : indclass = (oidvector *) DatumGetPointer(indclassDatum);
1568 :
1569 : /*
1570 : * Fill the support procedure OID array, as well as the info about
1571 : * opfamilies and opclass input types. (aminfo and supportinfo are left
1572 : * as zeroes, and are filled on-the-fly when used)
1573 : */
1574 752186 : IndexSupportInitialize(indclass, relation->rd_support,
1575 : relation->rd_opfamily, relation->rd_opcintype,
1576 : amsupport, indnkeyatts);
1577 :
1578 : /*
1579 : * Similarly extract indoption and copy it to the cache entry
1580 : */
1581 752186 : indoptionDatum = fastgetattr(relation->rd_indextuple,
1582 : Anum_pg_index_indoption,
1583 : GetPgIndexDescriptor(),
1584 : &isnull);
1585 : Assert(!isnull);
1586 752186 : indoption = (int2vector *) DatumGetPointer(indoptionDatum);
1587 752186 : memcpy(relation->rd_indoption, indoption->values, indnkeyatts * sizeof(int16));
1588 :
1589 752186 : (void) RelationGetIndexAttOptions(relation, false);
1590 :
1591 : /*
1592 : * expressions, predicate, exclusion caches will be filled later
1593 : */
1594 752180 : relation->rd_indexprs = NIL;
1595 752180 : relation->rd_indpred = NIL;
1596 752180 : relation->rd_exclops = NULL;
1597 752180 : relation->rd_exclprocs = NULL;
1598 752180 : relation->rd_exclstrats = NULL;
1599 752180 : relation->rd_amcache = NULL;
1600 752180 : }
1601 :
1602 : /*
1603 : * IndexSupportInitialize
1604 : * Initializes an index's cached opclass information,
1605 : * given the index's pg_index.indclass entry.
1606 : *
1607 : * Data is returned into *indexSupport, *opFamily, and *opcInType,
1608 : * which are arrays allocated by the caller.
1609 : *
1610 : * The caller also passes maxSupportNumber and maxAttributeNumber, since these
1611 : * indicate the size of the arrays it has allocated --- but in practice these
1612 : * numbers must always match those obtainable from the system catalog entries
1613 : * for the index and access method.
1614 : */
1615 : static void
1616 752186 : IndexSupportInitialize(oidvector *indclass,
1617 : RegProcedure *indexSupport,
1618 : Oid *opFamily,
1619 : Oid *opcInType,
1620 : StrategyNumber maxSupportNumber,
1621 : AttrNumber maxAttributeNumber)
1622 : {
1623 : int attIndex;
1624 :
1625 2048478 : for (attIndex = 0; attIndex < maxAttributeNumber; attIndex++)
1626 : {
1627 : OpClassCacheEnt *opcentry;
1628 :
1629 1296292 : if (!OidIsValid(indclass->values[attIndex]))
1630 0 : elog(ERROR, "bogus pg_index tuple");
1631 :
1632 : /* look up the info for this opclass, using a cache */
1633 1296292 : opcentry = LookupOpclassInfo(indclass->values[attIndex],
1634 : maxSupportNumber);
1635 :
1636 : /* copy cached data into relcache entry */
1637 1296292 : opFamily[attIndex] = opcentry->opcfamily;
1638 1296292 : opcInType[attIndex] = opcentry->opcintype;
1639 1296292 : if (maxSupportNumber > 0)
1640 1296292 : memcpy(&indexSupport[attIndex * maxSupportNumber],
1641 1296292 : opcentry->supportProcs,
1642 : maxSupportNumber * sizeof(RegProcedure));
1643 : }
1644 752186 : }
1645 :
1646 : /*
1647 : * LookupOpclassInfo
1648 : *
1649 : * This routine maintains a per-opclass cache of the information needed
1650 : * by IndexSupportInitialize(). This is more efficient than relying on
1651 : * the catalog cache, because we can load all the info about a particular
1652 : * opclass in a single indexscan of pg_amproc.
1653 : *
1654 : * The information from pg_am about expected range of support function
1655 : * numbers is passed in, rather than being looked up, mainly because the
1656 : * caller will have it already.
1657 : *
1658 : * Note there is no provision for flushing the cache. This is OK at the
1659 : * moment because there is no way to ALTER any interesting properties of an
1660 : * existing opclass --- all you can do is drop it, which will result in
1661 : * a useless but harmless dead entry in the cache. To support altering
1662 : * opclass membership (not the same as opfamily membership!), we'd need to
1663 : * be able to flush this cache as well as the contents of relcache entries
1664 : * for indexes.
1665 : */
1666 : static OpClassCacheEnt *
1667 1296292 : LookupOpclassInfo(Oid operatorClassOid,
1668 : StrategyNumber numSupport)
1669 : {
1670 : OpClassCacheEnt *opcentry;
1671 : bool found;
1672 : Relation rel;
1673 : SysScanDesc scan;
1674 : ScanKeyData skey[3];
1675 : HeapTuple htup;
1676 : bool indexOK;
1677 :
1678 1296292 : if (OpClassCache == NULL)
1679 : {
1680 : /* First time through: initialize the opclass cache */
1681 : HASHCTL ctl;
1682 :
1683 : /* Also make sure CacheMemoryContext exists */
1684 31872 : if (!CacheMemoryContext)
1685 0 : CreateCacheMemoryContext();
1686 :
1687 31872 : ctl.keysize = sizeof(Oid);
1688 31872 : ctl.entrysize = sizeof(OpClassCacheEnt);
1689 31872 : OpClassCache = hash_create("Operator class cache", 64,
1690 : &ctl, HASH_ELEM | HASH_BLOBS);
1691 : }
1692 :
1693 1296292 : opcentry = (OpClassCacheEnt *) hash_search(OpClassCache,
1694 : &operatorClassOid,
1695 : HASH_ENTER, &found);
1696 :
1697 1296292 : if (!found)
1698 : {
1699 : /* Initialize new entry */
1700 102564 : opcentry->valid = false; /* until known OK */
1701 102564 : opcentry->numSupport = numSupport;
1702 102564 : opcentry->supportProcs = NULL; /* filled below */
1703 : }
1704 : else
1705 : {
1706 : Assert(numSupport == opcentry->numSupport);
1707 : }
1708 :
1709 : /*
1710 : * When aggressively testing cache-flush hazards, we disable the operator
1711 : * class cache and force reloading of the info on each call. This models
1712 : * no real-world behavior, since the cache entries are never invalidated
1713 : * otherwise. However it can be helpful for detecting bugs in the cache
1714 : * loading logic itself, such as reliance on a non-nailed index. Given
1715 : * the limited use-case and the fact that this adds a great deal of
1716 : * expense, we enable it only for high values of debug_discard_caches.
1717 : */
1718 : #ifdef DISCARD_CACHES_ENABLED
1719 : if (debug_discard_caches > 2)
1720 : opcentry->valid = false;
1721 : #endif
1722 :
1723 1296292 : if (opcentry->valid)
1724 1193728 : return opcentry;
1725 :
1726 : /*
1727 : * Need to fill in new entry. First allocate space, unless we already did
1728 : * so in some previous attempt.
1729 : */
1730 102564 : if (opcentry->supportProcs == NULL && numSupport > 0)
1731 102564 : opcentry->supportProcs = (RegProcedure *)
1732 102564 : MemoryContextAllocZero(CacheMemoryContext,
1733 : numSupport * sizeof(RegProcedure));
1734 :
1735 : /*
1736 : * To avoid infinite recursion during startup, force heap scans if we're
1737 : * looking up info for the opclasses used by the indexes we would like to
1738 : * reference here.
1739 : */
1740 113644 : indexOK = criticalRelcachesBuilt ||
1741 11080 : (operatorClassOid != OID_BTREE_OPS_OID &&
1742 7590 : operatorClassOid != INT2_BTREE_OPS_OID);
1743 :
1744 : /*
1745 : * We have to fetch the pg_opclass row to determine its opfamily and
1746 : * opcintype, which are needed to look up related operators and functions.
1747 : * It'd be convenient to use the syscache here, but that probably doesn't
1748 : * work while bootstrapping.
1749 : */
1750 102564 : ScanKeyInit(&skey[0],
1751 : Anum_pg_opclass_oid,
1752 : BTEqualStrategyNumber, F_OIDEQ,
1753 : ObjectIdGetDatum(operatorClassOid));
1754 102564 : rel = table_open(OperatorClassRelationId, AccessShareLock);
1755 102564 : scan = systable_beginscan(rel, OpclassOidIndexId, indexOK,
1756 : NULL, 1, skey);
1757 :
1758 102564 : if (HeapTupleIsValid(htup = systable_getnext(scan)))
1759 : {
1760 102564 : Form_pg_opclass opclassform = (Form_pg_opclass) GETSTRUCT(htup);
1761 :
1762 102564 : opcentry->opcfamily = opclassform->opcfamily;
1763 102564 : opcentry->opcintype = opclassform->opcintype;
1764 : }
1765 : else
1766 0 : elog(ERROR, "could not find tuple for opclass %u", operatorClassOid);
1767 :
1768 102564 : systable_endscan(scan);
1769 102564 : table_close(rel, AccessShareLock);
1770 :
1771 : /*
1772 : * Scan pg_amproc to obtain support procs for the opclass. We only fetch
1773 : * the default ones (those with lefttype = righttype = opcintype).
1774 : */
1775 102564 : if (numSupport > 0)
1776 : {
1777 102564 : ScanKeyInit(&skey[0],
1778 : Anum_pg_amproc_amprocfamily,
1779 : BTEqualStrategyNumber, F_OIDEQ,
1780 : ObjectIdGetDatum(opcentry->opcfamily));
1781 102564 : ScanKeyInit(&skey[1],
1782 : Anum_pg_amproc_amproclefttype,
1783 : BTEqualStrategyNumber, F_OIDEQ,
1784 : ObjectIdGetDatum(opcentry->opcintype));
1785 102564 : ScanKeyInit(&skey[2],
1786 : Anum_pg_amproc_amprocrighttype,
1787 : BTEqualStrategyNumber, F_OIDEQ,
1788 : ObjectIdGetDatum(opcentry->opcintype));
1789 102564 : rel = table_open(AccessMethodProcedureRelationId, AccessShareLock);
1790 102564 : scan = systable_beginscan(rel, AccessMethodProcedureIndexId, indexOK,
1791 : NULL, 3, skey);
1792 :
1793 490148 : while (HeapTupleIsValid(htup = systable_getnext(scan)))
1794 : {
1795 387584 : Form_pg_amproc amprocform = (Form_pg_amproc) GETSTRUCT(htup);
1796 :
1797 387584 : if (amprocform->amprocnum <= 0 ||
1798 387584 : (StrategyNumber) amprocform->amprocnum > numSupport)
1799 0 : elog(ERROR, "invalid amproc number %d for opclass %u",
1800 : amprocform->amprocnum, operatorClassOid);
1801 :
1802 387584 : opcentry->supportProcs[amprocform->amprocnum - 1] =
1803 387584 : amprocform->amproc;
1804 : }
1805 :
1806 102564 : systable_endscan(scan);
1807 102564 : table_close(rel, AccessShareLock);
1808 : }
1809 :
1810 102564 : opcentry->valid = true;
1811 102564 : return opcentry;
1812 : }
1813 :
1814 : /*
1815 : * Fill in the TableAmRoutine for a relation
1816 : *
1817 : * relation's rd_amhandler must be valid already.
1818 : */
1819 : static void
1820 2428536 : InitTableAmRoutine(Relation relation)
1821 : {
1822 2428536 : relation->rd_tableam = GetTableAmRoutine(relation->rd_amhandler);
1823 2428536 : }
1824 :
1825 : /*
1826 : * Initialize table access method support for a table like relation
1827 : */
1828 : void
1829 2428536 : RelationInitTableAccessMethod(Relation relation)
1830 : {
1831 : HeapTuple tuple;
1832 : Form_pg_am aform;
1833 :
1834 2428536 : if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
1835 : {
1836 : /*
1837 : * Sequences are currently accessed like heap tables, but it doesn't
1838 : * seem prudent to show that in the catalog. So just overwrite it
1839 : * here.
1840 : */
1841 : Assert(relation->rd_rel->relam == InvalidOid);
1842 8558 : relation->rd_amhandler = F_HEAP_TABLEAM_HANDLER;
1843 : }
1844 2419978 : else if (IsCatalogRelation(relation))
1845 : {
1846 : /*
1847 : * Avoid doing a syscache lookup for catalog tables.
1848 : */
1849 : Assert(relation->rd_rel->relam == HEAP_TABLE_AM_OID);
1850 1930556 : relation->rd_amhandler = F_HEAP_TABLEAM_HANDLER;
1851 : }
1852 : else
1853 : {
1854 : /*
1855 : * Look up the table access method, save the OID of its handler
1856 : * function.
1857 : */
1858 : Assert(relation->rd_rel->relam != InvalidOid);
1859 489422 : tuple = SearchSysCache1(AMOID,
1860 489422 : ObjectIdGetDatum(relation->rd_rel->relam));
1861 489422 : if (!HeapTupleIsValid(tuple))
1862 0 : elog(ERROR, "cache lookup failed for access method %u",
1863 : relation->rd_rel->relam);
1864 489422 : aform = (Form_pg_am) GETSTRUCT(tuple);
1865 489422 : relation->rd_amhandler = aform->amhandler;
1866 489422 : ReleaseSysCache(tuple);
1867 : }
1868 :
1869 : /*
1870 : * Now we can fetch the table AM's API struct
1871 : */
1872 2428536 : InitTableAmRoutine(relation);
1873 2428536 : }
1874 :
1875 : /*
1876 : * formrdesc
1877 : *
1878 : * This is a special cut-down version of RelationBuildDesc(),
1879 : * used while initializing the relcache.
1880 : * The relation descriptor is built just from the supplied parameters,
1881 : * without actually looking at any system table entries. We cheat
1882 : * quite a lot since we only need to work for a few basic system
1883 : * catalogs.
1884 : *
1885 : * The catalogs this is used for can't have constraints (except attnotnull),
1886 : * default values, rules, or triggers, since we don't cope with any of that.
1887 : * (Well, actually, this only matters for properties that need to be valid
1888 : * during bootstrap or before RelationCacheInitializePhase3 runs, and none of
1889 : * these properties matter then...)
1890 : *
1891 : * NOTE: we assume we are already switched into CacheMemoryContext.
1892 : */
1893 : static void
1894 37208 : formrdesc(const char *relationName, Oid relationReltype,
1895 : bool isshared,
1896 : int natts, const FormData_pg_attribute *attrs)
1897 : {
1898 : Relation relation;
1899 : int i;
1900 : bool has_not_null;
1901 :
1902 : /*
1903 : * allocate new relation desc, clear all fields of reldesc
1904 : */
1905 37208 : relation = (Relation) palloc0(sizeof(RelationData));
1906 :
1907 : /* make sure relation is marked as having no open file yet */
1908 37208 : relation->rd_smgr = NULL;
1909 :
1910 : /*
1911 : * initialize reference count: 1 because it is nailed in cache
1912 : */
1913 37208 : relation->rd_refcnt = 1;
1914 :
1915 : /*
1916 : * all entries built with this routine are nailed-in-cache; none are for
1917 : * new or temp relations.
1918 : */
1919 37208 : relation->rd_isnailed = true;
1920 37208 : relation->rd_createSubid = InvalidSubTransactionId;
1921 37208 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
1922 37208 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
1923 37208 : relation->rd_droppedSubid = InvalidSubTransactionId;
1924 37208 : relation->rd_backend = INVALID_PROC_NUMBER;
1925 37208 : relation->rd_islocaltemp = false;
1926 :
1927 : /*
1928 : * initialize relation tuple form
1929 : *
1930 : * The data we insert here is pretty incomplete/bogus, but it'll serve to
1931 : * get us launched. RelationCacheInitializePhase3() will read the real
1932 : * data from pg_class and replace what we've done here. Note in
1933 : * particular that relowner is left as zero; this cues
1934 : * RelationCacheInitializePhase3 that the real data isn't there yet.
1935 : */
1936 37208 : relation->rd_rel = (Form_pg_class) palloc0(CLASS_TUPLE_SIZE);
1937 :
1938 37208 : namestrcpy(&relation->rd_rel->relname, relationName);
1939 37208 : relation->rd_rel->relnamespace = PG_CATALOG_NAMESPACE;
1940 37208 : relation->rd_rel->reltype = relationReltype;
1941 :
1942 : /*
1943 : * It's important to distinguish between shared and non-shared relations,
1944 : * even at bootstrap time, to make sure we know where they are stored.
1945 : */
1946 37208 : relation->rd_rel->relisshared = isshared;
1947 37208 : if (isshared)
1948 23240 : relation->rd_rel->reltablespace = GLOBALTABLESPACE_OID;
1949 :
1950 : /* formrdesc is used only for permanent relations */
1951 37208 : relation->rd_rel->relpersistence = RELPERSISTENCE_PERMANENT;
1952 :
1953 : /* ... and they're always populated, too */
1954 37208 : relation->rd_rel->relispopulated = true;
1955 :
1956 37208 : relation->rd_rel->relreplident = REPLICA_IDENTITY_NOTHING;
1957 37208 : relation->rd_rel->relpages = 0;
1958 37208 : relation->rd_rel->reltuples = -1;
1959 37208 : relation->rd_rel->relallvisible = 0;
1960 37208 : relation->rd_rel->relallfrozen = 0;
1961 37208 : relation->rd_rel->relkind = RELKIND_RELATION;
1962 37208 : relation->rd_rel->relnatts = (int16) natts;
1963 :
1964 : /*
1965 : * initialize attribute tuple form
1966 : *
1967 : * Unlike the case with the relation tuple, this data had better be right
1968 : * because it will never be replaced. The data comes from
1969 : * src/include/catalog/ headers via genbki.pl.
1970 : */
1971 37208 : relation->rd_att = CreateTemplateTupleDesc(natts);
1972 37208 : relation->rd_att->tdrefcount = 1; /* mark as refcounted */
1973 :
1974 37208 : relation->rd_att->tdtypeid = relationReltype;
1975 37208 : relation->rd_att->tdtypmod = -1; /* just to be sure */
1976 :
1977 : /*
1978 : * initialize tuple desc info
1979 : */
1980 37208 : has_not_null = false;
1981 733594 : for (i = 0; i < natts; i++)
1982 : {
1983 1393168 : memcpy(TupleDescAttr(relation->rd_att, i),
1984 696404 : &attrs[i],
1985 : ATTRIBUTE_FIXED_PART_SIZE);
1986 696764 : has_not_null |= attrs[i].attnotnull;
1987 :
1988 696764 : populate_compact_attribute(relation->rd_att, i);
1989 : }
1990 :
1991 : /* initialize first attribute's attcacheoff, cf RelationBuildTupleDesc */
1992 37190 : TupleDescCompactAttr(relation->rd_att, 0)->attcacheoff = 0;
1993 :
1994 : /* mark not-null status */
1995 37190 : if (has_not_null)
1996 : {
1997 37190 : TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
1998 :
1999 37190 : constr->has_not_null = true;
2000 37190 : relation->rd_att->constr = constr;
2001 : }
2002 :
2003 : /*
2004 : * initialize relation id from info in att array (my, this is ugly)
2005 : */
2006 37190 : RelationGetRelid(relation) = TupleDescAttr(relation->rd_att, 0)->attrelid;
2007 :
2008 : /*
2009 : * All relations made with formrdesc are mapped. This is necessarily so
2010 : * because there is no other way to know what filenumber they currently
2011 : * have. In bootstrap mode, add them to the initial relation mapper data,
2012 : * specifying that the initial filenumber is the same as the OID.
2013 : */
2014 37190 : relation->rd_rel->relfilenode = InvalidRelFileNumber;
2015 37190 : if (IsBootstrapProcessingMode())
2016 408 : RelationMapUpdateMap(RelationGetRelid(relation),
2017 : RelationGetRelid(relation),
2018 : isshared, true);
2019 :
2020 : /*
2021 : * initialize the relation lock manager information
2022 : */
2023 37190 : RelationInitLockInfo(relation); /* see lmgr.c */
2024 :
2025 : /*
2026 : * initialize physical addressing information for the relation
2027 : */
2028 37190 : RelationInitPhysicalAddr(relation);
2029 :
2030 : /*
2031 : * initialize the table am handler
2032 : */
2033 37190 : relation->rd_rel->relam = HEAP_TABLE_AM_OID;
2034 37190 : relation->rd_tableam = GetHeapamTableAmRoutine();
2035 :
2036 : /*
2037 : * initialize the rel-has-index flag, using hardwired knowledge
2038 : */
2039 37190 : if (IsBootstrapProcessingMode())
2040 : {
2041 : /* In bootstrap mode, we have no indexes */
2042 408 : relation->rd_rel->relhasindex = false;
2043 : }
2044 : else
2045 : {
2046 : /* Otherwise, all the rels formrdesc is used for have indexes */
2047 36782 : relation->rd_rel->relhasindex = true;
2048 : }
2049 :
2050 : /*
2051 : * add new reldesc to relcache
2052 : */
2053 37190 : RelationCacheInsert(relation, false);
2054 :
2055 : /* It's fully valid */
2056 37190 : relation->rd_isvalid = true;
2057 37190 : }
2058 :
2059 : #ifdef USE_ASSERT_CHECKING
2060 : /*
2061 : * AssertCouldGetRelation
2062 : *
2063 : * Check safety of calling RelationIdGetRelation().
2064 : *
2065 : * In code that reads catalogs in the event of a cache miss, call this
2066 : * before checking the cache.
2067 : */
2068 : void
2069 : AssertCouldGetRelation(void)
2070 : {
2071 : Assert(IsTransactionState());
2072 : AssertBufferLocksPermitCatalogRead();
2073 : }
2074 : #endif
2075 :
2076 :
2077 : /* ----------------------------------------------------------------
2078 : * Relation Descriptor Lookup Interface
2079 : * ----------------------------------------------------------------
2080 : */
2081 :
2082 : /*
2083 : * RelationIdGetRelation
2084 : *
2085 : * Lookup a reldesc by OID; make one if not already in cache.
2086 : *
2087 : * Returns NULL if no pg_class row could be found for the given relid
2088 : * (suggesting we are trying to access a just-deleted relation).
2089 : * Any other error is reported via elog.
2090 : *
2091 : * NB: caller should already have at least AccessShareLock on the
2092 : * relation ID, else there are nasty race conditions.
2093 : *
2094 : * NB: relation ref count is incremented, or set to 1 if new entry.
2095 : * Caller should eventually decrement count. (Usually,
2096 : * that happens by calling RelationClose().)
2097 : */
2098 : Relation
2099 44752480 : RelationIdGetRelation(Oid relationId)
2100 : {
2101 : Relation rd;
2102 :
2103 44752480 : AssertCouldGetRelation();
2104 :
2105 : /*
2106 : * first try to find reldesc in the cache
2107 : */
2108 44752480 : RelationIdCacheLookup(relationId, rd);
2109 :
2110 44752480 : if (RelationIsValid(rd))
2111 : {
2112 : /* return NULL for dropped relations */
2113 43465038 : if (rd->rd_droppedSubid != InvalidSubTransactionId)
2114 : {
2115 : Assert(!rd->rd_isvalid);
2116 4 : return NULL;
2117 : }
2118 :
2119 43465034 : RelationIncrementReferenceCount(rd);
2120 : /* revalidate cache entry if necessary */
2121 43465034 : if (!rd->rd_isvalid)
2122 : {
2123 175310 : RelationRebuildRelation(rd);
2124 :
2125 : /*
2126 : * Normally entries need to be valid here, but before the relcache
2127 : * has been initialized, not enough infrastructure exists to
2128 : * perform pg_class lookups. The structure of such entries doesn't
2129 : * change, but we still want to update the rd_rel entry. So
2130 : * rd_isvalid = false is left in place for a later lookup.
2131 : */
2132 : Assert(rd->rd_isvalid ||
2133 : (rd->rd_isnailed && !criticalRelcachesBuilt));
2134 : }
2135 43465022 : return rd;
2136 : }
2137 :
2138 : /*
2139 : * no reldesc in the cache, so have RelationBuildDesc() build one and add
2140 : * it.
2141 : */
2142 1287442 : rd = RelationBuildDesc(relationId, true);
2143 1287440 : if (RelationIsValid(rd))
2144 1287430 : RelationIncrementReferenceCount(rd);
2145 1287440 : return rd;
2146 : }
2147 :
2148 : /* ----------------------------------------------------------------
2149 : * cache invalidation support routines
2150 : * ----------------------------------------------------------------
2151 : */
2152 :
2153 : /* ResourceOwner callbacks to track relcache references */
2154 : static void ResOwnerReleaseRelation(Datum res);
2155 : static char *ResOwnerPrintRelCache(Datum res);
2156 :
2157 : static const ResourceOwnerDesc relref_resowner_desc =
2158 : {
2159 : .name = "relcache reference",
2160 : .release_phase = RESOURCE_RELEASE_BEFORE_LOCKS,
2161 : .release_priority = RELEASE_PRIO_RELCACHE_REFS,
2162 : .ReleaseResource = ResOwnerReleaseRelation,
2163 : .DebugPrint = ResOwnerPrintRelCache
2164 : };
2165 :
2166 : /* Convenience wrappers over ResourceOwnerRemember/Forget */
2167 : static inline void
2168 65905034 : ResourceOwnerRememberRelationRef(ResourceOwner owner, Relation rel)
2169 : {
2170 65905034 : ResourceOwnerRemember(owner, PointerGetDatum(rel), &relref_resowner_desc);
2171 65905034 : }
2172 : static inline void
2173 65861322 : ResourceOwnerForgetRelationRef(ResourceOwner owner, Relation rel)
2174 : {
2175 65861322 : ResourceOwnerForget(owner, PointerGetDatum(rel), &relref_resowner_desc);
2176 65861322 : }
2177 :
2178 : /*
2179 : * RelationIncrementReferenceCount
2180 : * Increments relation reference count.
2181 : *
2182 : * Note: bootstrap mode has its own weird ideas about relation refcount
2183 : * behavior; we ought to fix it someday, but for now, just disable
2184 : * reference count ownership tracking in bootstrap mode.
2185 : */
2186 : void
2187 66479702 : RelationIncrementReferenceCount(Relation rel)
2188 : {
2189 66479702 : ResourceOwnerEnlarge(CurrentResourceOwner);
2190 66479702 : rel->rd_refcnt += 1;
2191 66479702 : if (!IsBootstrapProcessingMode())
2192 65905034 : ResourceOwnerRememberRelationRef(CurrentResourceOwner, rel);
2193 66479702 : }
2194 :
2195 : /*
2196 : * RelationDecrementReferenceCount
2197 : * Decrements relation reference count.
2198 : */
2199 : void
2200 66435990 : RelationDecrementReferenceCount(Relation rel)
2201 : {
2202 : Assert(rel->rd_refcnt > 0);
2203 66435990 : rel->rd_refcnt -= 1;
2204 66435990 : if (!IsBootstrapProcessingMode())
2205 65861322 : ResourceOwnerForgetRelationRef(CurrentResourceOwner, rel);
2206 66435990 : }
2207 :
2208 : /*
2209 : * RelationClose - close an open relation
2210 : *
2211 : * Actually, we just decrement the refcount.
2212 : *
2213 : * NOTE: if compiled with -DRELCACHE_FORCE_RELEASE then relcache entries
2214 : * will be freed as soon as their refcount goes to zero. In combination
2215 : * with aset.c's CLOBBER_FREED_MEMORY option, this provides a good test
2216 : * to catch references to already-released relcache entries. It slows
2217 : * things down quite a bit, however.
2218 : */
2219 : void
2220 44855804 : RelationClose(Relation relation)
2221 : {
2222 : /* Note: no locking manipulations needed */
2223 44855804 : RelationDecrementReferenceCount(relation);
2224 :
2225 44855804 : RelationCloseCleanup(relation);
2226 44855804 : }
2227 :
2228 : static void
2229 44899516 : RelationCloseCleanup(Relation relation)
2230 : {
2231 : /*
2232 : * If the relation is no longer open in this session, we can clean up any
2233 : * stale partition descriptors it has. This is unlikely, so check to see
2234 : * if there are child contexts before expending a call to mcxt.c.
2235 : */
2236 44899516 : if (RelationHasReferenceCountZero(relation))
2237 : {
2238 25165690 : if (relation->rd_pdcxt != NULL &&
2239 104126 : relation->rd_pdcxt->firstchild != NULL)
2240 4524 : MemoryContextDeleteChildren(relation->rd_pdcxt);
2241 :
2242 25165690 : if (relation->rd_pddcxt != NULL &&
2243 108 : relation->rd_pddcxt->firstchild != NULL)
2244 0 : MemoryContextDeleteChildren(relation->rd_pddcxt);
2245 : }
2246 :
2247 : #ifdef RELCACHE_FORCE_RELEASE
2248 : if (RelationHasReferenceCountZero(relation) &&
2249 : relation->rd_createSubid == InvalidSubTransactionId &&
2250 : relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId)
2251 : RelationClearRelation(relation);
2252 : #endif
2253 44899516 : }
2254 :
2255 : /*
2256 : * RelationReloadIndexInfo - reload minimal information for an open index
2257 : *
2258 : * This function is used only for indexes. A relcache inval on an index
2259 : * can mean that its pg_class or pg_index row changed. There are only
2260 : * very limited changes that are allowed to an existing index's schema,
2261 : * so we can update the relcache entry without a complete rebuild; which
2262 : * is fortunate because we can't rebuild an index entry that is "nailed"
2263 : * and/or in active use. We support full replacement of the pg_class row,
2264 : * as well as updates of a few simple fields of the pg_index row.
2265 : *
2266 : * We assume that at the time we are called, we have at least AccessShareLock
2267 : * on the target index.
2268 : *
2269 : * If the target index is an index on pg_class or pg_index, we'd better have
2270 : * previously gotten at least AccessShareLock on its underlying catalog,
2271 : * else we are at risk of deadlock against someone trying to exclusive-lock
2272 : * the heap and index in that order. This is ensured in current usage by
2273 : * only applying this to indexes being opened or having positive refcount.
2274 : */
2275 : static void
2276 115520 : RelationReloadIndexInfo(Relation relation)
2277 : {
2278 : bool indexOK;
2279 : HeapTuple pg_class_tuple;
2280 : Form_pg_class relp;
2281 :
2282 : /* Should be called only for invalidated, live indexes */
2283 : Assert((relation->rd_rel->relkind == RELKIND_INDEX ||
2284 : relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX) &&
2285 : !relation->rd_isvalid &&
2286 : relation->rd_droppedSubid == InvalidSubTransactionId);
2287 :
2288 : /*
2289 : * If it's a shared index, we might be called before backend startup has
2290 : * finished selecting a database, in which case we have no way to read
2291 : * pg_class yet. However, a shared index can never have any significant
2292 : * schema updates, so it's okay to mostly ignore the invalidation signal.
2293 : * Its physical relfilenumber might've changed, but that's all. Update
2294 : * the physical relfilenumber, mark it valid and return without doing
2295 : * anything more.
2296 : */
2297 115520 : if (relation->rd_rel->relisshared && !criticalRelcachesBuilt)
2298 : {
2299 0 : RelationInitPhysicalAddr(relation);
2300 0 : relation->rd_isvalid = true;
2301 0 : return;
2302 : }
2303 :
2304 : /*
2305 : * Read the pg_class row
2306 : *
2307 : * Don't try to use an indexscan of pg_class_oid_index to reload the info
2308 : * for pg_class_oid_index ...
2309 : */
2310 115520 : indexOK = (RelationGetRelid(relation) != ClassOidIndexId);
2311 115520 : pg_class_tuple = ScanPgRelation(RelationGetRelid(relation), indexOK, false);
2312 115514 : if (!HeapTupleIsValid(pg_class_tuple))
2313 0 : elog(ERROR, "could not find pg_class tuple for index %u",
2314 : RelationGetRelid(relation));
2315 115514 : relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
2316 115514 : memcpy(relation->rd_rel, relp, CLASS_TUPLE_SIZE);
2317 : /* Reload reloptions in case they changed */
2318 115514 : if (relation->rd_options)
2319 1150 : pfree(relation->rd_options);
2320 115514 : RelationParseRelOptions(relation, pg_class_tuple);
2321 : /* done with pg_class tuple */
2322 115514 : heap_freetuple(pg_class_tuple);
2323 : /* We must recalculate physical address in case it changed */
2324 115514 : RelationInitPhysicalAddr(relation);
2325 :
2326 : /*
2327 : * For a non-system index, there are fields of the pg_index row that are
2328 : * allowed to change, so re-read that row and update the relcache entry.
2329 : * Most of the info derived from pg_index (such as support function lookup
2330 : * info) cannot change, and indeed the whole point of this routine is to
2331 : * update the relcache entry without clobbering that data; so wholesale
2332 : * replacement is not appropriate.
2333 : */
2334 115514 : if (!IsSystemRelation(relation))
2335 : {
2336 : HeapTuple tuple;
2337 : Form_pg_index index;
2338 :
2339 42510 : tuple = SearchSysCache1(INDEXRELID,
2340 : ObjectIdGetDatum(RelationGetRelid(relation)));
2341 42510 : if (!HeapTupleIsValid(tuple))
2342 0 : elog(ERROR, "cache lookup failed for index %u",
2343 : RelationGetRelid(relation));
2344 42510 : index = (Form_pg_index) GETSTRUCT(tuple);
2345 :
2346 : /*
2347 : * Basically, let's just copy all the bool fields. There are one or
2348 : * two of these that can't actually change in the current code, but
2349 : * it's not worth it to track exactly which ones they are. None of
2350 : * the array fields are allowed to change, though.
2351 : */
2352 42510 : relation->rd_index->indisunique = index->indisunique;
2353 42510 : relation->rd_index->indnullsnotdistinct = index->indnullsnotdistinct;
2354 42510 : relation->rd_index->indisprimary = index->indisprimary;
2355 42510 : relation->rd_index->indisexclusion = index->indisexclusion;
2356 42510 : relation->rd_index->indimmediate = index->indimmediate;
2357 42510 : relation->rd_index->indisclustered = index->indisclustered;
2358 42510 : relation->rd_index->indisvalid = index->indisvalid;
2359 42510 : relation->rd_index->indcheckxmin = index->indcheckxmin;
2360 42510 : relation->rd_index->indisready = index->indisready;
2361 42510 : relation->rd_index->indislive = index->indislive;
2362 42510 : relation->rd_index->indisreplident = index->indisreplident;
2363 :
2364 : /* Copy xmin too, as that is needed to make sense of indcheckxmin */
2365 42510 : HeapTupleHeaderSetXmin(relation->rd_indextuple->t_data,
2366 42510 : HeapTupleHeaderGetXmin(tuple->t_data));
2367 :
2368 42510 : ReleaseSysCache(tuple);
2369 : }
2370 :
2371 : /* Okay, now it's valid again */
2372 115514 : relation->rd_isvalid = true;
2373 : }
2374 :
2375 : /*
2376 : * RelationReloadNailed - reload minimal information for nailed relations.
2377 : *
2378 : * The structure of a nailed relation can never change (which is good, because
2379 : * we rely on knowing their structure to be able to read catalog content). But
2380 : * some parts, e.g. pg_class.relfrozenxid, are still important to have
2381 : * accurate content for. Therefore those need to be reloaded after the arrival
2382 : * of invalidations.
2383 : */
2384 : static void
2385 162076 : RelationReloadNailed(Relation relation)
2386 : {
2387 : /* Should be called only for invalidated, nailed relations */
2388 : Assert(!relation->rd_isvalid);
2389 : Assert(relation->rd_isnailed);
2390 : /* nailed indexes are handled by RelationReloadIndexInfo() */
2391 : Assert(relation->rd_rel->relkind == RELKIND_RELATION);
2392 162076 : AssertCouldGetRelation();
2393 :
2394 : /*
2395 : * Redo RelationInitPhysicalAddr in case it is a mapped relation whose
2396 : * mapping changed.
2397 : */
2398 162076 : RelationInitPhysicalAddr(relation);
2399 :
2400 : /*
2401 : * Reload a non-index entry. We can't easily do so if relcaches aren't
2402 : * yet built, but that's fine because at that stage the attributes that
2403 : * need to be current (like relfrozenxid) aren't yet accessed. To ensure
2404 : * the entry will later be revalidated, we leave it in invalid state, but
2405 : * allow use (cf. RelationIdGetRelation()).
2406 : */
2407 162076 : if (criticalRelcachesBuilt)
2408 : {
2409 : HeapTuple pg_class_tuple;
2410 : Form_pg_class relp;
2411 :
2412 : /*
2413 : * NB: Mark the entry as valid before starting to scan, to avoid
2414 : * self-recursion when re-building pg_class.
2415 : */
2416 32198 : relation->rd_isvalid = true;
2417 :
2418 32198 : pg_class_tuple = ScanPgRelation(RelationGetRelid(relation),
2419 : true, false);
2420 32192 : relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
2421 32192 : memcpy(relation->rd_rel, relp, CLASS_TUPLE_SIZE);
2422 32192 : heap_freetuple(pg_class_tuple);
2423 :
2424 : /*
2425 : * Again mark as valid, to protect against concurrently arriving
2426 : * invalidations.
2427 : */
2428 32192 : relation->rd_isvalid = true;
2429 : }
2430 162070 : }
2431 :
2432 : /*
2433 : * RelationDestroyRelation
2434 : *
2435 : * Physically delete a relation cache entry and all subsidiary data.
2436 : * Caller must already have unhooked the entry from the hash table.
2437 : */
2438 : static void
2439 1249648 : RelationDestroyRelation(Relation relation, bool remember_tupdesc)
2440 : {
2441 : Assert(RelationHasReferenceCountZero(relation));
2442 :
2443 : /*
2444 : * Make sure smgr and lower levels close the relation's files, if they
2445 : * weren't closed already. (This was probably done by caller, but let's
2446 : * just be real sure.)
2447 : */
2448 1249648 : RelationCloseSmgr(relation);
2449 :
2450 : /* break mutual link with stats entry */
2451 1249648 : pgstat_unlink_relation(relation);
2452 :
2453 : /*
2454 : * Free all the subsidiary data structures of the relcache entry, then the
2455 : * entry itself.
2456 : */
2457 1249648 : if (relation->rd_rel)
2458 1249648 : pfree(relation->rd_rel);
2459 : /* can't use DecrTupleDescRefCount here */
2460 : Assert(relation->rd_att->tdrefcount > 0);
2461 1249648 : if (--relation->rd_att->tdrefcount == 0)
2462 : {
2463 : /*
2464 : * If we Rebuilt a relcache entry during a transaction then its
2465 : * possible we did that because the TupDesc changed as the result of
2466 : * an ALTER TABLE that ran at less than AccessExclusiveLock. It's
2467 : * possible someone copied that TupDesc, in which case the copy would
2468 : * point to free'd memory. So if we rebuild an entry we keep the
2469 : * TupDesc around until end of transaction, to be safe.
2470 : */
2471 1246460 : if (remember_tupdesc)
2472 21876 : RememberToFreeTupleDescAtEOX(relation->rd_att);
2473 : else
2474 1224584 : FreeTupleDesc(relation->rd_att);
2475 : }
2476 1249648 : FreeTriggerDesc(relation->trigdesc);
2477 1249648 : list_free_deep(relation->rd_fkeylist);
2478 1249648 : list_free(relation->rd_indexlist);
2479 1249648 : list_free(relation->rd_statlist);
2480 1249648 : bms_free(relation->rd_keyattr);
2481 1249648 : bms_free(relation->rd_pkattr);
2482 1249648 : bms_free(relation->rd_idattr);
2483 1249648 : bms_free(relation->rd_hotblockingattr);
2484 1249648 : bms_free(relation->rd_summarizedattr);
2485 1249648 : if (relation->rd_pubdesc)
2486 6594 : pfree(relation->rd_pubdesc);
2487 1249648 : if (relation->rd_options)
2488 11134 : pfree(relation->rd_options);
2489 1249648 : if (relation->rd_indextuple)
2490 337002 : pfree(relation->rd_indextuple);
2491 1249648 : if (relation->rd_amcache)
2492 0 : pfree(relation->rd_amcache);
2493 1249648 : if (relation->rd_fdwroutine)
2494 278 : pfree(relation->rd_fdwroutine);
2495 1249648 : if (relation->rd_indexcxt)
2496 337002 : MemoryContextDelete(relation->rd_indexcxt);
2497 1249648 : if (relation->rd_rulescxt)
2498 26188 : MemoryContextDelete(relation->rd_rulescxt);
2499 1249648 : if (relation->rd_rsdesc)
2500 1928 : MemoryContextDelete(relation->rd_rsdesc->rscxt);
2501 1249648 : if (relation->rd_partkeycxt)
2502 16690 : MemoryContextDelete(relation->rd_partkeycxt);
2503 1249648 : if (relation->rd_pdcxt)
2504 16148 : MemoryContextDelete(relation->rd_pdcxt);
2505 1249648 : if (relation->rd_pddcxt)
2506 60 : MemoryContextDelete(relation->rd_pddcxt);
2507 1249648 : if (relation->rd_partcheckcxt)
2508 3382 : MemoryContextDelete(relation->rd_partcheckcxt);
2509 1249648 : pfree(relation);
2510 1249648 : }
2511 :
2512 : /*
2513 : * RelationInvalidateRelation - mark a relation cache entry as invalid
2514 : *
2515 : * An entry that's marked as invalid will be reloaded on next access.
2516 : */
2517 : static void
2518 1657036 : RelationInvalidateRelation(Relation relation)
2519 : {
2520 : /*
2521 : * Make sure smgr and lower levels close the relation's files, if they
2522 : * weren't closed already. If the relation is not getting deleted, the
2523 : * next smgr access should reopen the files automatically. This ensures
2524 : * that the low-level file access state is updated after, say, a vacuum
2525 : * truncation.
2526 : */
2527 1657036 : RelationCloseSmgr(relation);
2528 :
2529 : /* Free AM cached data, if any */
2530 1657036 : if (relation->rd_amcache)
2531 69188 : pfree(relation->rd_amcache);
2532 1657036 : relation->rd_amcache = NULL;
2533 :
2534 1657036 : relation->rd_isvalid = false;
2535 1657036 : }
2536 :
2537 : /*
2538 : * RelationClearRelation - physically blow away a relation cache entry
2539 : *
2540 : * The caller must ensure that the entry is no longer needed, i.e. its
2541 : * reference count is zero. Also, the rel or its storage must not be created
2542 : * in the current transaction (rd_createSubid and rd_firstRelfilelocatorSubid
2543 : * must not be set).
2544 : */
2545 : static void
2546 809834 : RelationClearRelation(Relation relation)
2547 : {
2548 : Assert(RelationHasReferenceCountZero(relation));
2549 : Assert(!relation->rd_isnailed);
2550 :
2551 : /*
2552 : * Relations created in the same transaction must never be removed, see
2553 : * RelationFlushRelation.
2554 : */
2555 : Assert(relation->rd_createSubid == InvalidSubTransactionId);
2556 : Assert(relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId);
2557 : Assert(relation->rd_droppedSubid == InvalidSubTransactionId);
2558 :
2559 : /* first mark it as invalid */
2560 809834 : RelationInvalidateRelation(relation);
2561 :
2562 : /* Remove it from the hash table */
2563 809834 : RelationCacheDelete(relation);
2564 :
2565 : /* And release storage */
2566 809834 : RelationDestroyRelation(relation, false);
2567 809834 : }
2568 :
2569 : /*
2570 : * RelationRebuildRelation - rebuild a relation cache entry in place
2571 : *
2572 : * Reset and rebuild a relation cache entry from scratch (that is, from
2573 : * catalog entries). This is used when we are notified of a change to an open
2574 : * relation (one with refcount > 0). The entry is reconstructed without
2575 : * moving the physical RelationData record, so that the refcount holder's
2576 : * pointer is still valid.
2577 : *
2578 : * NB: when rebuilding, we'd better hold some lock on the relation, else the
2579 : * catalog data we need to read could be changing under us. Also, a rel to be
2580 : * rebuilt had better have refcnt > 0. This is because a sinval reset could
2581 : * happen while we're accessing the catalogs, and the rel would get blown away
2582 : * underneath us by RelationCacheInvalidate if it has zero refcnt.
2583 : */
2584 : static void
2585 717382 : RelationRebuildRelation(Relation relation)
2586 : {
2587 : Assert(!RelationHasReferenceCountZero(relation));
2588 717382 : AssertCouldGetRelation();
2589 : /* there is no reason to ever rebuild a dropped relation */
2590 : Assert(relation->rd_droppedSubid == InvalidSubTransactionId);
2591 :
2592 : /* Close and mark it as invalid until we've finished the rebuild */
2593 717382 : RelationInvalidateRelation(relation);
2594 :
2595 : /*
2596 : * Indexes only have a limited number of possible schema changes, and we
2597 : * don't want to use the full-blown procedure because it's a headache for
2598 : * indexes that reload itself depends on.
2599 : *
2600 : * As an exception, use the full procedure if the index access info hasn't
2601 : * been initialized yet. Index creation relies on that: it first builds
2602 : * the relcache entry with RelationBuildLocalRelation(), creates the
2603 : * pg_index tuple only after that, and then relies on
2604 : * CommandCounterIncrement to load the pg_index contents.
2605 : */
2606 717382 : if ((relation->rd_rel->relkind == RELKIND_INDEX ||
2607 574492 : relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX) &&
2608 148396 : relation->rd_indexcxt != NULL)
2609 : {
2610 115520 : RelationReloadIndexInfo(relation);
2611 115514 : return;
2612 : }
2613 : /* Nailed relations are handled separately. */
2614 601862 : else if (relation->rd_isnailed)
2615 : {
2616 162076 : RelationReloadNailed(relation);
2617 162070 : return;
2618 : }
2619 : else
2620 : {
2621 : /*
2622 : * Our strategy for rebuilding an open relcache entry is to build a
2623 : * new entry from scratch, swap its contents with the old entry, and
2624 : * finally delete the new entry (along with any infrastructure swapped
2625 : * over from the old entry). This is to avoid trouble in case an
2626 : * error causes us to lose control partway through. The old entry
2627 : * will still be marked !rd_isvalid, so we'll try to rebuild it again
2628 : * on next access. Meanwhile it's not any less valid than it was
2629 : * before, so any code that might expect to continue accessing it
2630 : * isn't hurt by the rebuild failure. (Consider for example a
2631 : * subtransaction that ALTERs a table and then gets canceled partway
2632 : * through the cache entry rebuild. The outer transaction should
2633 : * still see the not-modified cache entry as valid.) The worst
2634 : * consequence of an error is leaking the necessarily-unreferenced new
2635 : * entry, and this shouldn't happen often enough for that to be a big
2636 : * problem.
2637 : *
2638 : * When rebuilding an open relcache entry, we must preserve ref count,
2639 : * rd_*Subid, and rd_toastoid state. Also attempt to preserve the
2640 : * pg_class entry (rd_rel), tupledesc, rewrite-rule, partition key,
2641 : * and partition descriptor substructures in place, because various
2642 : * places assume that these structures won't move while they are
2643 : * working with an open relcache entry. (Note: the refcount
2644 : * mechanism for tupledescs might someday allow us to remove this hack
2645 : * for the tupledesc.)
2646 : *
2647 : * Note that this process does not touch CurrentResourceOwner; which
2648 : * is good because whatever ref counts the entry may have do not
2649 : * necessarily belong to that resource owner.
2650 : */
2651 : Relation newrel;
2652 439786 : Oid save_relid = RelationGetRelid(relation);
2653 : bool keep_tupdesc;
2654 : bool keep_rules;
2655 : bool keep_policies;
2656 : bool keep_partkey;
2657 :
2658 : /* Build temporary entry, but don't link it into hashtable */
2659 439786 : newrel = RelationBuildDesc(save_relid, false);
2660 :
2661 : /*
2662 : * Between here and the end of the swap, don't add code that does or
2663 : * reasonably could read system catalogs. That range must be free
2664 : * from invalidation processing. See RelationBuildDesc() manipulation
2665 : * of in_progress_list.
2666 : */
2667 :
2668 439780 : if (newrel == NULL)
2669 : {
2670 : /*
2671 : * We can validly get here, if we're using a historic snapshot in
2672 : * which a relation, accessed from outside logical decoding, is
2673 : * still invisible. In that case it's fine to just mark the
2674 : * relation as invalid and return - it'll fully get reloaded by
2675 : * the cache reset at the end of logical decoding (or at the next
2676 : * access). During normal processing we don't want to ignore this
2677 : * case as it shouldn't happen there, as explained below.
2678 : */
2679 0 : if (HistoricSnapshotActive())
2680 0 : return;
2681 :
2682 : /*
2683 : * This shouldn't happen as dropping a relation is intended to be
2684 : * impossible if still referenced (cf. CheckTableNotInUse()). But
2685 : * if we get here anyway, we can't just delete the relcache entry,
2686 : * as it possibly could get accessed later (as e.g. the error
2687 : * might get trapped and handled via a subtransaction rollback).
2688 : */
2689 0 : elog(ERROR, "relation %u deleted while still in use", save_relid);
2690 : }
2691 :
2692 : /*
2693 : * If we were to, again, have cases of the relkind of a relcache entry
2694 : * changing, we would need to ensure that pgstats does not get
2695 : * confused.
2696 : */
2697 : Assert(relation->rd_rel->relkind == newrel->rd_rel->relkind);
2698 :
2699 439780 : keep_tupdesc = equalTupleDescs(relation->rd_att, newrel->rd_att);
2700 439780 : keep_rules = equalRuleLocks(relation->rd_rules, newrel->rd_rules);
2701 439780 : keep_policies = equalRSDesc(relation->rd_rsdesc, newrel->rd_rsdesc);
2702 : /* partkey is immutable once set up, so we can always keep it */
2703 439780 : keep_partkey = (relation->rd_partkey != NULL);
2704 :
2705 : /*
2706 : * Perform swapping of the relcache entry contents. Within this
2707 : * process the old entry is momentarily invalid, so there *must* be no
2708 : * possibility of CHECK_FOR_INTERRUPTS within this sequence. Do it in
2709 : * all-in-line code for safety.
2710 : *
2711 : * Since the vast majority of fields should be swapped, our method is
2712 : * to swap the whole structures and then re-swap those few fields we
2713 : * didn't want swapped.
2714 : */
2715 : #define SWAPFIELD(fldtype, fldname) \
2716 : do { \
2717 : fldtype _tmp = newrel->fldname; \
2718 : newrel->fldname = relation->fldname; \
2719 : relation->fldname = _tmp; \
2720 : } while (0)
2721 :
2722 : /* swap all Relation struct fields */
2723 : {
2724 : RelationData tmpstruct;
2725 :
2726 439780 : memcpy(&tmpstruct, newrel, sizeof(RelationData));
2727 439780 : memcpy(newrel, relation, sizeof(RelationData));
2728 439780 : memcpy(relation, &tmpstruct, sizeof(RelationData));
2729 : }
2730 :
2731 : /* rd_smgr must not be swapped, due to back-links from smgr level */
2732 439780 : SWAPFIELD(SMgrRelation, rd_smgr);
2733 : /* rd_refcnt must be preserved */
2734 439780 : SWAPFIELD(int, rd_refcnt);
2735 : /* isnailed shouldn't change */
2736 : Assert(newrel->rd_isnailed == relation->rd_isnailed);
2737 : /* creation sub-XIDs must be preserved */
2738 439780 : SWAPFIELD(SubTransactionId, rd_createSubid);
2739 439780 : SWAPFIELD(SubTransactionId, rd_newRelfilelocatorSubid);
2740 439780 : SWAPFIELD(SubTransactionId, rd_firstRelfilelocatorSubid);
2741 439780 : SWAPFIELD(SubTransactionId, rd_droppedSubid);
2742 : /* un-swap rd_rel pointers, swap contents instead */
2743 439780 : SWAPFIELD(Form_pg_class, rd_rel);
2744 : /* ... but actually, we don't have to update newrel->rd_rel */
2745 439780 : memcpy(relation->rd_rel, newrel->rd_rel, CLASS_TUPLE_SIZE);
2746 : /* preserve old tupledesc, rules, policies if no logical change */
2747 439780 : if (keep_tupdesc)
2748 417638 : SWAPFIELD(TupleDesc, rd_att);
2749 439780 : if (keep_rules)
2750 : {
2751 421902 : SWAPFIELD(RuleLock *, rd_rules);
2752 421902 : SWAPFIELD(MemoryContext, rd_rulescxt);
2753 : }
2754 439780 : if (keep_policies)
2755 439456 : SWAPFIELD(RowSecurityDesc *, rd_rsdesc);
2756 : /* toast OID override must be preserved */
2757 439780 : SWAPFIELD(Oid, rd_toastoid);
2758 : /* pgstat_info / enabled must be preserved */
2759 439780 : SWAPFIELD(struct PgStat_TableStatus *, pgstat_info);
2760 439780 : SWAPFIELD(bool, pgstat_enabled);
2761 : /* preserve old partition key if we have one */
2762 439780 : if (keep_partkey)
2763 : {
2764 15476 : SWAPFIELD(PartitionKey, rd_partkey);
2765 15476 : SWAPFIELD(MemoryContext, rd_partkeycxt);
2766 : }
2767 439780 : if (newrel->rd_pdcxt != NULL || newrel->rd_pddcxt != NULL)
2768 : {
2769 : /*
2770 : * We are rebuilding a partitioned relation with a non-zero
2771 : * reference count, so we must keep the old partition descriptor
2772 : * around, in case there's a PartitionDirectory with a pointer to
2773 : * it. This means we can't free the old rd_pdcxt yet. (This is
2774 : * necessary because RelationGetPartitionDesc hands out direct
2775 : * pointers to the relcache's data structure, unlike our usual
2776 : * practice which is to hand out copies. We'd have the same
2777 : * problem with rd_partkey, except that we always preserve that
2778 : * once created.)
2779 : *
2780 : * To ensure that it's not leaked completely, re-attach it to the
2781 : * new reldesc, or make it a child of the new reldesc's rd_pdcxt
2782 : * in the unlikely event that there is one already. (Compare hack
2783 : * in RelationBuildPartitionDesc.) RelationClose will clean up
2784 : * any such contexts once the reference count reaches zero.
2785 : *
2786 : * In the case where the reference count is zero, this code is not
2787 : * reached, which should be OK because in that case there should
2788 : * be no PartitionDirectory with a pointer to the old entry.
2789 : *
2790 : * Note that newrel and relation have already been swapped, so the
2791 : * "old" partition descriptor is actually the one hanging off of
2792 : * newrel.
2793 : */
2794 12074 : relation->rd_partdesc = NULL; /* ensure rd_partdesc is invalid */
2795 12074 : relation->rd_partdesc_nodetached = NULL;
2796 12074 : relation->rd_partdesc_nodetached_xmin = InvalidTransactionId;
2797 12074 : if (relation->rd_pdcxt != NULL) /* probably never happens */
2798 0 : MemoryContextSetParent(newrel->rd_pdcxt, relation->rd_pdcxt);
2799 : else
2800 12074 : relation->rd_pdcxt = newrel->rd_pdcxt;
2801 12074 : if (relation->rd_pddcxt != NULL)
2802 0 : MemoryContextSetParent(newrel->rd_pddcxt, relation->rd_pddcxt);
2803 : else
2804 12074 : relation->rd_pddcxt = newrel->rd_pddcxt;
2805 : /* drop newrel's pointers so we don't destroy it below */
2806 12074 : newrel->rd_partdesc = NULL;
2807 12074 : newrel->rd_partdesc_nodetached = NULL;
2808 12074 : newrel->rd_partdesc_nodetached_xmin = InvalidTransactionId;
2809 12074 : newrel->rd_pdcxt = NULL;
2810 12074 : newrel->rd_pddcxt = NULL;
2811 : }
2812 :
2813 : #undef SWAPFIELD
2814 :
2815 : /* And now we can throw away the temporary entry */
2816 439780 : RelationDestroyRelation(newrel, !keep_tupdesc);
2817 : }
2818 : }
2819 :
2820 : /*
2821 : * RelationFlushRelation
2822 : *
2823 : * Rebuild the relation if it is open (refcount > 0), else blow it away.
2824 : * This is used when we receive a cache invalidation event for the rel.
2825 : */
2826 : static void
2827 905224 : RelationFlushRelation(Relation relation)
2828 : {
2829 905224 : if (relation->rd_createSubid != InvalidSubTransactionId ||
2830 558170 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
2831 : {
2832 : /*
2833 : * New relcache entries are always rebuilt, not flushed; else we'd
2834 : * forget the "new" status of the relation. Ditto for the
2835 : * new-relfilenumber status.
2836 : */
2837 364830 : if (IsTransactionState() && relation->rd_droppedSubid == InvalidSubTransactionId)
2838 : {
2839 : /*
2840 : * The rel could have zero refcnt here, so temporarily increment
2841 : * the refcnt to ensure it's safe to rebuild it. We can assume
2842 : * that the current transaction has some lock on the rel already.
2843 : */
2844 362942 : RelationIncrementReferenceCount(relation);
2845 362942 : RelationRebuildRelation(relation);
2846 362936 : RelationDecrementReferenceCount(relation);
2847 : }
2848 : else
2849 1888 : RelationInvalidateRelation(relation);
2850 : }
2851 : else
2852 : {
2853 : /*
2854 : * Pre-existing rels can be dropped from the relcache if not open.
2855 : *
2856 : * If the entry is in use, rebuild it if possible. If we're not
2857 : * inside a valid transaction, we can't do any catalog access so it's
2858 : * not possible to rebuild yet. Just mark it as invalid in that case,
2859 : * so that the rebuild will occur when the entry is next opened.
2860 : *
2861 : * Note: it's possible that we come here during subtransaction abort,
2862 : * and the reason for wanting to rebuild is that the rel is open in
2863 : * the outer transaction. In that case it might seem unsafe to not
2864 : * rebuild immediately, since whatever code has the rel already open
2865 : * will keep on using the relcache entry as-is. However, in such a
2866 : * case the outer transaction should be holding a lock that's
2867 : * sufficient to prevent any significant change in the rel's schema,
2868 : * so the existing entry contents should be good enough for its
2869 : * purposes; at worst we might be behind on statistics updates or the
2870 : * like. (See also CheckTableNotInUse() and its callers.)
2871 : */
2872 540394 : if (RelationHasReferenceCountZero(relation))
2873 333778 : RelationClearRelation(relation);
2874 206616 : else if (!IsTransactionState())
2875 22714 : RelationInvalidateRelation(relation);
2876 183902 : else if (relation->rd_isnailed && relation->rd_refcnt == 1)
2877 : {
2878 : /*
2879 : * A nailed relation with refcnt == 1 is unused. We cannot clear
2880 : * it, but there's also no need no need to rebuild it immediately.
2881 : */
2882 4976 : RelationInvalidateRelation(relation);
2883 : }
2884 : else
2885 178926 : RelationRebuildRelation(relation);
2886 : }
2887 905218 : }
2888 :
2889 : /*
2890 : * RelationForgetRelation - caller reports that it dropped the relation
2891 : */
2892 : void
2893 71652 : RelationForgetRelation(Oid rid)
2894 : {
2895 : Relation relation;
2896 :
2897 71652 : RelationIdCacheLookup(rid, relation);
2898 :
2899 71652 : if (!PointerIsValid(relation))
2900 0 : return; /* not in cache, nothing to do */
2901 :
2902 71652 : if (!RelationHasReferenceCountZero(relation))
2903 0 : elog(ERROR, "relation %u is still open", rid);
2904 :
2905 : Assert(relation->rd_droppedSubid == InvalidSubTransactionId);
2906 71652 : if (relation->rd_createSubid != InvalidSubTransactionId ||
2907 70066 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
2908 : {
2909 : /*
2910 : * In the event of subtransaction rollback, we must not forget
2911 : * rd_*Subid. Mark the entry "dropped" and invalidate it, instead of
2912 : * destroying it right away. (If we're in a top transaction, we could
2913 : * opt to destroy the entry.)
2914 : */
2915 1614 : relation->rd_droppedSubid = GetCurrentSubTransactionId();
2916 1614 : RelationInvalidateRelation(relation);
2917 : }
2918 : else
2919 70038 : RelationClearRelation(relation);
2920 : }
2921 :
2922 : /*
2923 : * RelationCacheInvalidateEntry
2924 : *
2925 : * This routine is invoked for SI cache flush messages.
2926 : *
2927 : * Any relcache entry matching the relid must be flushed. (Note: caller has
2928 : * already determined that the relid belongs to our database or is a shared
2929 : * relation.)
2930 : *
2931 : * We used to skip local relations, on the grounds that they could
2932 : * not be targets of cross-backend SI update messages; but it seems
2933 : * safer to process them, so that our *own* SI update messages will
2934 : * have the same effects during CommandCounterIncrement for both
2935 : * local and nonlocal relations.
2936 : */
2937 : void
2938 3077098 : RelationCacheInvalidateEntry(Oid relationId)
2939 : {
2940 : Relation relation;
2941 :
2942 3077098 : RelationIdCacheLookup(relationId, relation);
2943 :
2944 3077098 : if (PointerIsValid(relation))
2945 : {
2946 905224 : relcacheInvalsReceived++;
2947 905224 : RelationFlushRelation(relation);
2948 : }
2949 : else
2950 : {
2951 : int i;
2952 :
2953 2218480 : for (i = 0; i < in_progress_list_len; i++)
2954 46606 : if (in_progress_list[i].reloid == relationId)
2955 34 : in_progress_list[i].invalidated = true;
2956 : }
2957 3077092 : }
2958 :
2959 : /*
2960 : * RelationCacheInvalidate
2961 : * Blow away cached relation descriptors that have zero reference counts,
2962 : * and rebuild those with positive reference counts. Also reset the smgr
2963 : * relation cache and re-read relation mapping data.
2964 : *
2965 : * Apart from debug_discard_caches, this is currently used only to recover
2966 : * from SI message buffer overflow, so we do not touch relations having
2967 : * new-in-transaction relfilenumbers; they cannot be targets of cross-backend
2968 : * SI updates (and our own updates now go through a separate linked list
2969 : * that isn't limited by the SI message buffer size).
2970 : *
2971 : * We do this in two phases: the first pass deletes deletable items, and
2972 : * the second one rebuilds the rebuildable items. This is essential for
2973 : * safety, because hash_seq_search only copes with concurrent deletion of
2974 : * the element it is currently visiting. If a second SI overflow were to
2975 : * occur while we are walking the table, resulting in recursive entry to
2976 : * this routine, we could crash because the inner invocation blows away
2977 : * the entry next to be visited by the outer scan. But this way is OK,
2978 : * because (a) during the first pass we won't process any more SI messages,
2979 : * so hash_seq_search will complete safely; (b) during the second pass we
2980 : * only hold onto pointers to nondeletable entries.
2981 : *
2982 : * The two-phase approach also makes it easy to update relfilenumbers for
2983 : * mapped relations before we do anything else, and to ensure that the
2984 : * second pass processes nailed-in-cache items before other nondeletable
2985 : * items. This should ensure that system catalogs are up to date before
2986 : * we attempt to use them to reload information about other open relations.
2987 : *
2988 : * After those two phases of work having immediate effects, we normally
2989 : * signal any RelationBuildDesc() on the stack to start over. However, we
2990 : * don't do this if called as part of debug_discard_caches. Otherwise,
2991 : * RelationBuildDesc() would become an infinite loop.
2992 : */
2993 : void
2994 4596 : RelationCacheInvalidate(bool debug_discard)
2995 : {
2996 : HASH_SEQ_STATUS status;
2997 : RelIdCacheEnt *idhentry;
2998 : Relation relation;
2999 4596 : List *rebuildFirstList = NIL;
3000 4596 : List *rebuildList = NIL;
3001 : ListCell *l;
3002 : int i;
3003 :
3004 : /*
3005 : * Reload relation mapping data before starting to reconstruct cache.
3006 : */
3007 4596 : RelationMapInvalidateAll();
3008 :
3009 : /* Phase 1 */
3010 4596 : hash_seq_init(&status, RelationIdCache);
3011 :
3012 503848 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3013 : {
3014 499252 : relation = idhentry->reldesc;
3015 :
3016 : /*
3017 : * Ignore new relations; no other backend will manipulate them before
3018 : * we commit. Likewise, before replacing a relation's relfilelocator,
3019 : * we shall have acquired AccessExclusiveLock and drained any
3020 : * applicable pending invalidations.
3021 : */
3022 499252 : if (relation->rd_createSubid != InvalidSubTransactionId ||
3023 499144 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
3024 120 : continue;
3025 :
3026 499132 : relcacheInvalsReceived++;
3027 :
3028 499132 : if (RelationHasReferenceCountZero(relation))
3029 : {
3030 : /* Delete this entry immediately */
3031 400300 : RelationClearRelation(relation);
3032 : }
3033 : else
3034 : {
3035 : /*
3036 : * If it's a mapped relation, immediately update its rd_locator in
3037 : * case its relfilenumber changed. We must do this during phase 1
3038 : * in case the relation is consulted during rebuild of other
3039 : * relcache entries in phase 2. It's safe since consulting the
3040 : * map doesn't involve any access to relcache entries.
3041 : */
3042 98832 : if (RelationIsMapped(relation))
3043 : {
3044 76410 : RelationCloseSmgr(relation);
3045 76410 : RelationInitPhysicalAddr(relation);
3046 : }
3047 :
3048 : /*
3049 : * Add this entry to list of stuff to rebuild in second pass.
3050 : * pg_class goes to the front of rebuildFirstList while
3051 : * pg_class_oid_index goes to the back of rebuildFirstList, so
3052 : * they are done first and second respectively. Other nailed
3053 : * relations go to the front of rebuildList, so they'll be done
3054 : * next in no particular order; and everything else goes to the
3055 : * back of rebuildList.
3056 : */
3057 98832 : if (RelationGetRelid(relation) == RelationRelationId)
3058 4452 : rebuildFirstList = lcons(relation, rebuildFirstList);
3059 94380 : else if (RelationGetRelid(relation) == ClassOidIndexId)
3060 4452 : rebuildFirstList = lappend(rebuildFirstList, relation);
3061 89928 : else if (relation->rd_isnailed)
3062 89760 : rebuildList = lcons(relation, rebuildList);
3063 : else
3064 168 : rebuildList = lappend(rebuildList, relation);
3065 : }
3066 : }
3067 :
3068 : /*
3069 : * We cannot destroy the SMgrRelations as there might still be references
3070 : * to them, but close the underlying file descriptors.
3071 : */
3072 4596 : smgrreleaseall();
3073 :
3074 : /*
3075 : * Phase 2: rebuild (or invalidate) the items found to need rebuild in
3076 : * phase 1
3077 : */
3078 13500 : foreach(l, rebuildFirstList)
3079 : {
3080 8904 : relation = (Relation) lfirst(l);
3081 8904 : if (!IsTransactionState() || (relation->rd_isnailed && relation->rd_refcnt == 1))
3082 8902 : RelationInvalidateRelation(relation);
3083 : else
3084 2 : RelationRebuildRelation(relation);
3085 : }
3086 4596 : list_free(rebuildFirstList);
3087 94524 : foreach(l, rebuildList)
3088 : {
3089 89928 : relation = (Relation) lfirst(l);
3090 89928 : if (!IsTransactionState() || (relation->rd_isnailed && relation->rd_refcnt == 1))
3091 89726 : RelationInvalidateRelation(relation);
3092 : else
3093 202 : RelationRebuildRelation(relation);
3094 : }
3095 4596 : list_free(rebuildList);
3096 :
3097 4596 : if (!debug_discard)
3098 : /* Any RelationBuildDesc() on the stack must start over. */
3099 4600 : for (i = 0; i < in_progress_list_len; i++)
3100 4 : in_progress_list[i].invalidated = true;
3101 4596 : }
3102 :
3103 : static void
3104 21876 : RememberToFreeTupleDescAtEOX(TupleDesc td)
3105 : {
3106 21876 : if (EOXactTupleDescArray == NULL)
3107 : {
3108 : MemoryContext oldcxt;
3109 :
3110 12514 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
3111 :
3112 12514 : EOXactTupleDescArray = (TupleDesc *) palloc(16 * sizeof(TupleDesc));
3113 12514 : EOXactTupleDescArrayLen = 16;
3114 12514 : NextEOXactTupleDescNum = 0;
3115 12514 : MemoryContextSwitchTo(oldcxt);
3116 : }
3117 9362 : else if (NextEOXactTupleDescNum >= EOXactTupleDescArrayLen)
3118 : {
3119 62 : int32 newlen = EOXactTupleDescArrayLen * 2;
3120 :
3121 : Assert(EOXactTupleDescArrayLen > 0);
3122 :
3123 62 : EOXactTupleDescArray = (TupleDesc *) repalloc(EOXactTupleDescArray,
3124 : newlen * sizeof(TupleDesc));
3125 62 : EOXactTupleDescArrayLen = newlen;
3126 : }
3127 :
3128 21876 : EOXactTupleDescArray[NextEOXactTupleDescNum++] = td;
3129 21876 : }
3130 :
3131 : #ifdef USE_ASSERT_CHECKING
3132 : static void
3133 : AssertPendingSyncConsistency(Relation relation)
3134 : {
3135 : bool relcache_verdict =
3136 : RelationIsPermanent(relation) &&
3137 : ((relation->rd_createSubid != InvalidSubTransactionId &&
3138 : RELKIND_HAS_STORAGE(relation->rd_rel->relkind)) ||
3139 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId);
3140 :
3141 : Assert(relcache_verdict == RelFileLocatorSkippingWAL(relation->rd_locator));
3142 :
3143 : if (relation->rd_droppedSubid != InvalidSubTransactionId)
3144 : Assert(!relation->rd_isvalid &&
3145 : (relation->rd_createSubid != InvalidSubTransactionId ||
3146 : relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId));
3147 : }
3148 :
3149 : /*
3150 : * AssertPendingSyncs_RelationCache
3151 : *
3152 : * Assert that relcache.c and storage.c agree on whether to skip WAL.
3153 : */
3154 : void
3155 : AssertPendingSyncs_RelationCache(void)
3156 : {
3157 : HASH_SEQ_STATUS status;
3158 : LOCALLOCK *locallock;
3159 : Relation *rels;
3160 : int maxrels;
3161 : int nrels;
3162 : RelIdCacheEnt *idhentry;
3163 : int i;
3164 :
3165 : /*
3166 : * Open every relation that this transaction has locked. If, for some
3167 : * relation, storage.c is skipping WAL and relcache.c is not skipping WAL,
3168 : * a CommandCounterIncrement() typically yields a local invalidation
3169 : * message that destroys the relcache entry. By recreating such entries
3170 : * here, we detect the problem.
3171 : */
3172 : PushActiveSnapshot(GetTransactionSnapshot());
3173 : maxrels = 1;
3174 : rels = palloc(maxrels * sizeof(*rels));
3175 : nrels = 0;
3176 : hash_seq_init(&status, GetLockMethodLocalHash());
3177 : while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
3178 : {
3179 : Oid relid;
3180 : Relation r;
3181 :
3182 : if (locallock->nLocks <= 0)
3183 : continue;
3184 : if ((LockTagType) locallock->tag.lock.locktag_type !=
3185 : LOCKTAG_RELATION)
3186 : continue;
3187 : relid = ObjectIdGetDatum(locallock->tag.lock.locktag_field2);
3188 : r = RelationIdGetRelation(relid);
3189 : if (!RelationIsValid(r))
3190 : continue;
3191 : if (nrels >= maxrels)
3192 : {
3193 : maxrels *= 2;
3194 : rels = repalloc(rels, maxrels * sizeof(*rels));
3195 : }
3196 : rels[nrels++] = r;
3197 : }
3198 :
3199 : hash_seq_init(&status, RelationIdCache);
3200 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3201 : AssertPendingSyncConsistency(idhentry->reldesc);
3202 :
3203 : for (i = 0; i < nrels; i++)
3204 : RelationClose(rels[i]);
3205 : PopActiveSnapshot();
3206 : }
3207 : #endif
3208 :
3209 : /*
3210 : * AtEOXact_RelationCache
3211 : *
3212 : * Clean up the relcache at main-transaction commit or abort.
3213 : *
3214 : * Note: this must be called *before* processing invalidation messages.
3215 : * In the case of abort, we don't want to try to rebuild any invalidated
3216 : * cache entries (since we can't safely do database accesses). Therefore
3217 : * we must reset refcnts before handling pending invalidations.
3218 : *
3219 : * As of PostgreSQL 8.1, relcache refcnts should get released by the
3220 : * ResourceOwner mechanism. This routine just does a debugging
3221 : * cross-check that no pins remain. However, we also need to do special
3222 : * cleanup when the current transaction created any relations or made use
3223 : * of forced index lists.
3224 : */
3225 : void
3226 1116832 : AtEOXact_RelationCache(bool isCommit)
3227 : {
3228 : HASH_SEQ_STATUS status;
3229 : RelIdCacheEnt *idhentry;
3230 : int i;
3231 :
3232 : /*
3233 : * Forget in_progress_list. This is relevant when we're aborting due to
3234 : * an error during RelationBuildDesc().
3235 : */
3236 : Assert(in_progress_list_len == 0 || !isCommit);
3237 1116832 : in_progress_list_len = 0;
3238 :
3239 : /*
3240 : * Unless the eoxact_list[] overflowed, we only need to examine the rels
3241 : * listed in it. Otherwise fall back on a hash_seq_search scan.
3242 : *
3243 : * For simplicity, eoxact_list[] entries are not deleted till end of
3244 : * top-level transaction, even though we could remove them at
3245 : * subtransaction end in some cases, or remove relations from the list if
3246 : * they are cleared for other reasons. Therefore we should expect the
3247 : * case that list entries are not found in the hashtable; if not, there's
3248 : * nothing to do for them.
3249 : */
3250 1116832 : if (eoxact_list_overflowed)
3251 : {
3252 164 : hash_seq_init(&status, RelationIdCache);
3253 44498 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3254 : {
3255 44334 : AtEOXact_cleanup(idhentry->reldesc, isCommit);
3256 : }
3257 : }
3258 : else
3259 : {
3260 1240046 : for (i = 0; i < eoxact_list_len; i++)
3261 : {
3262 123378 : idhentry = (RelIdCacheEnt *) hash_search(RelationIdCache,
3263 123378 : &eoxact_list[i],
3264 : HASH_FIND,
3265 : NULL);
3266 123378 : if (idhentry != NULL)
3267 121150 : AtEOXact_cleanup(idhentry->reldesc, isCommit);
3268 : }
3269 : }
3270 :
3271 1116832 : if (EOXactTupleDescArrayLen > 0)
3272 : {
3273 : Assert(EOXactTupleDescArray != NULL);
3274 34390 : for (i = 0; i < NextEOXactTupleDescNum; i++)
3275 21876 : FreeTupleDesc(EOXactTupleDescArray[i]);
3276 12514 : pfree(EOXactTupleDescArray);
3277 12514 : EOXactTupleDescArray = NULL;
3278 : }
3279 :
3280 : /* Now we're out of the transaction and can clear the lists */
3281 1116832 : eoxact_list_len = 0;
3282 1116832 : eoxact_list_overflowed = false;
3283 1116832 : NextEOXactTupleDescNum = 0;
3284 1116832 : EOXactTupleDescArrayLen = 0;
3285 1116832 : }
3286 :
3287 : /*
3288 : * AtEOXact_cleanup
3289 : *
3290 : * Clean up a single rel at main-transaction commit or abort
3291 : *
3292 : * NB: this processing must be idempotent, because EOXactListAdd() doesn't
3293 : * bother to prevent duplicate entries in eoxact_list[].
3294 : */
3295 : static void
3296 165484 : AtEOXact_cleanup(Relation relation, bool isCommit)
3297 : {
3298 165484 : bool clear_relcache = false;
3299 :
3300 : /*
3301 : * The relcache entry's ref count should be back to its normal
3302 : * not-in-a-transaction state: 0 unless it's nailed in cache.
3303 : *
3304 : * In bootstrap mode, this is NOT true, so don't check it --- the
3305 : * bootstrap code expects relations to stay open across start/commit
3306 : * transaction calls. (That seems bogus, but it's not worth fixing.)
3307 : *
3308 : * Note: ideally this check would be applied to every relcache entry, not
3309 : * just those that have eoxact work to do. But it's not worth forcing a
3310 : * scan of the whole relcache just for this. (Moreover, doing so would
3311 : * mean that assert-enabled testing never tests the hash_search code path
3312 : * above, which seems a bad idea.)
3313 : */
3314 : #ifdef USE_ASSERT_CHECKING
3315 : if (!IsBootstrapProcessingMode())
3316 : {
3317 : int expected_refcnt;
3318 :
3319 : expected_refcnt = relation->rd_isnailed ? 1 : 0;
3320 : Assert(relation->rd_refcnt == expected_refcnt);
3321 : }
3322 : #endif
3323 :
3324 : /*
3325 : * Is the relation live after this transaction ends?
3326 : *
3327 : * During commit, clear the relcache entry if it is preserved after
3328 : * relation drop, in order not to orphan the entry. During rollback,
3329 : * clear the relcache entry if the relation is created in the current
3330 : * transaction since it isn't interesting any longer once we are out of
3331 : * the transaction.
3332 : */
3333 165484 : clear_relcache =
3334 : (isCommit ?
3335 165484 : relation->rd_droppedSubid != InvalidSubTransactionId :
3336 4790 : relation->rd_createSubid != InvalidSubTransactionId);
3337 :
3338 : /*
3339 : * Since we are now out of the transaction, reset the subids to zero. That
3340 : * also lets RelationClearRelation() drop the relcache entry.
3341 : */
3342 165484 : relation->rd_createSubid = InvalidSubTransactionId;
3343 165484 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3344 165484 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3345 165484 : relation->rd_droppedSubid = InvalidSubTransactionId;
3346 :
3347 165484 : if (clear_relcache)
3348 : {
3349 5594 : if (RelationHasReferenceCountZero(relation))
3350 : {
3351 5594 : RelationClearRelation(relation);
3352 5594 : return;
3353 : }
3354 : else
3355 : {
3356 : /*
3357 : * Hmm, somewhere there's a (leaked?) reference to the relation.
3358 : * We daren't remove the entry for fear of dereferencing a
3359 : * dangling pointer later. Bleat, and mark it as not belonging to
3360 : * the current transaction. Hopefully it'll get cleaned up
3361 : * eventually. This must be just a WARNING to avoid
3362 : * error-during-error-recovery loops.
3363 : */
3364 0 : elog(WARNING, "cannot remove relcache entry for \"%s\" because it has nonzero refcount",
3365 : RelationGetRelationName(relation));
3366 : }
3367 : }
3368 : }
3369 :
3370 : /*
3371 : * AtEOSubXact_RelationCache
3372 : *
3373 : * Clean up the relcache at sub-transaction commit or abort.
3374 : *
3375 : * Note: this must be called *before* processing invalidation messages.
3376 : */
3377 : void
3378 20024 : AtEOSubXact_RelationCache(bool isCommit, SubTransactionId mySubid,
3379 : SubTransactionId parentSubid)
3380 : {
3381 : HASH_SEQ_STATUS status;
3382 : RelIdCacheEnt *idhentry;
3383 : int i;
3384 :
3385 : /*
3386 : * Forget in_progress_list. This is relevant when we're aborting due to
3387 : * an error during RelationBuildDesc(). We don't commit subtransactions
3388 : * during RelationBuildDesc().
3389 : */
3390 : Assert(in_progress_list_len == 0 || !isCommit);
3391 20024 : in_progress_list_len = 0;
3392 :
3393 : /*
3394 : * Unless the eoxact_list[] overflowed, we only need to examine the rels
3395 : * listed in it. Otherwise fall back on a hash_seq_search scan. Same
3396 : * logic as in AtEOXact_RelationCache.
3397 : */
3398 20024 : if (eoxact_list_overflowed)
3399 : {
3400 0 : hash_seq_init(&status, RelationIdCache);
3401 0 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
3402 : {
3403 0 : AtEOSubXact_cleanup(idhentry->reldesc, isCommit,
3404 : mySubid, parentSubid);
3405 : }
3406 : }
3407 : else
3408 : {
3409 29456 : for (i = 0; i < eoxact_list_len; i++)
3410 : {
3411 9432 : idhentry = (RelIdCacheEnt *) hash_search(RelationIdCache,
3412 9432 : &eoxact_list[i],
3413 : HASH_FIND,
3414 : NULL);
3415 9432 : if (idhentry != NULL)
3416 8488 : AtEOSubXact_cleanup(idhentry->reldesc, isCommit,
3417 : mySubid, parentSubid);
3418 : }
3419 : }
3420 :
3421 : /* Don't reset the list; we still need more cleanup later */
3422 20024 : }
3423 :
3424 : /*
3425 : * AtEOSubXact_cleanup
3426 : *
3427 : * Clean up a single rel at subtransaction commit or abort
3428 : *
3429 : * NB: this processing must be idempotent, because EOXactListAdd() doesn't
3430 : * bother to prevent duplicate entries in eoxact_list[].
3431 : */
3432 : static void
3433 8488 : AtEOSubXact_cleanup(Relation relation, bool isCommit,
3434 : SubTransactionId mySubid, SubTransactionId parentSubid)
3435 : {
3436 : /*
3437 : * Is it a relation created in the current subtransaction?
3438 : *
3439 : * During subcommit, mark it as belonging to the parent, instead, as long
3440 : * as it has not been dropped. Otherwise simply delete the relcache entry.
3441 : * --- it isn't interesting any longer.
3442 : */
3443 8488 : if (relation->rd_createSubid == mySubid)
3444 : {
3445 : /*
3446 : * Valid rd_droppedSubid means the corresponding relation is dropped
3447 : * but the relcache entry is preserved for at-commit pending sync. We
3448 : * need to drop it explicitly here not to make the entry orphan.
3449 : */
3450 : Assert(relation->rd_droppedSubid == mySubid ||
3451 : relation->rd_droppedSubid == InvalidSubTransactionId);
3452 198 : if (isCommit && relation->rd_droppedSubid == InvalidSubTransactionId)
3453 74 : relation->rd_createSubid = parentSubid;
3454 124 : else if (RelationHasReferenceCountZero(relation))
3455 : {
3456 : /* allow the entry to be removed */
3457 124 : relation->rd_createSubid = InvalidSubTransactionId;
3458 124 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3459 124 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3460 124 : relation->rd_droppedSubid = InvalidSubTransactionId;
3461 124 : RelationClearRelation(relation);
3462 124 : return;
3463 : }
3464 : else
3465 : {
3466 : /*
3467 : * Hmm, somewhere there's a (leaked?) reference to the relation.
3468 : * We daren't remove the entry for fear of dereferencing a
3469 : * dangling pointer later. Bleat, and transfer it to the parent
3470 : * subtransaction so we can try again later. This must be just a
3471 : * WARNING to avoid error-during-error-recovery loops.
3472 : */
3473 0 : relation->rd_createSubid = parentSubid;
3474 0 : elog(WARNING, "cannot remove relcache entry for \"%s\" because it has nonzero refcount",
3475 : RelationGetRelationName(relation));
3476 : }
3477 : }
3478 :
3479 : /*
3480 : * Likewise, update or drop any new-relfilenumber-in-subtransaction record
3481 : * or drop record.
3482 : */
3483 8364 : if (relation->rd_newRelfilelocatorSubid == mySubid)
3484 : {
3485 130 : if (isCommit)
3486 68 : relation->rd_newRelfilelocatorSubid = parentSubid;
3487 : else
3488 62 : relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3489 : }
3490 :
3491 8364 : if (relation->rd_firstRelfilelocatorSubid == mySubid)
3492 : {
3493 90 : if (isCommit)
3494 30 : relation->rd_firstRelfilelocatorSubid = parentSubid;
3495 : else
3496 60 : relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3497 : }
3498 :
3499 8364 : if (relation->rd_droppedSubid == mySubid)
3500 : {
3501 20 : if (isCommit)
3502 2 : relation->rd_droppedSubid = parentSubid;
3503 : else
3504 18 : relation->rd_droppedSubid = InvalidSubTransactionId;
3505 : }
3506 : }
3507 :
3508 :
3509 : /*
3510 : * RelationBuildLocalRelation
3511 : * Build a relcache entry for an about-to-be-created relation,
3512 : * and enter it into the relcache.
3513 : */
3514 : Relation
3515 139552 : RelationBuildLocalRelation(const char *relname,
3516 : Oid relnamespace,
3517 : TupleDesc tupDesc,
3518 : Oid relid,
3519 : Oid accessmtd,
3520 : RelFileNumber relfilenumber,
3521 : Oid reltablespace,
3522 : bool shared_relation,
3523 : bool mapped_relation,
3524 : char relpersistence,
3525 : char relkind)
3526 : {
3527 : Relation rel;
3528 : MemoryContext oldcxt;
3529 139552 : int natts = tupDesc->natts;
3530 : int i;
3531 : bool has_not_null;
3532 : bool nailit;
3533 :
3534 : Assert(natts >= 0);
3535 :
3536 : /*
3537 : * check for creation of a rel that must be nailed in cache.
3538 : *
3539 : * XXX this list had better match the relations specially handled in
3540 : * RelationCacheInitializePhase2/3.
3541 : */
3542 139552 : switch (relid)
3543 : {
3544 714 : case DatabaseRelationId:
3545 : case AuthIdRelationId:
3546 : case AuthMemRelationId:
3547 : case RelationRelationId:
3548 : case AttributeRelationId:
3549 : case ProcedureRelationId:
3550 : case TypeRelationId:
3551 714 : nailit = true;
3552 714 : break;
3553 138838 : default:
3554 138838 : nailit = false;
3555 138838 : break;
3556 : }
3557 :
3558 : /*
3559 : * check that hardwired list of shared rels matches what's in the
3560 : * bootstrap .bki file. If you get a failure here during initdb, you
3561 : * probably need to fix IsSharedRelation() to match whatever you've done
3562 : * to the set of shared relations.
3563 : */
3564 139552 : if (shared_relation != IsSharedRelation(relid))
3565 0 : elog(ERROR, "shared_relation flag for \"%s\" does not match IsSharedRelation(%u)",
3566 : relname, relid);
3567 :
3568 : /* Shared relations had better be mapped, too */
3569 : Assert(mapped_relation || !shared_relation);
3570 :
3571 : /*
3572 : * switch to the cache context to create the relcache entry.
3573 : */
3574 139552 : if (!CacheMemoryContext)
3575 0 : CreateCacheMemoryContext();
3576 :
3577 139552 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
3578 :
3579 : /*
3580 : * allocate a new relation descriptor and fill in basic state fields.
3581 : */
3582 139552 : rel = (Relation) palloc0(sizeof(RelationData));
3583 :
3584 : /* make sure relation is marked as having no open file yet */
3585 139552 : rel->rd_smgr = NULL;
3586 :
3587 : /* mark it nailed if appropriate */
3588 139552 : rel->rd_isnailed = nailit;
3589 :
3590 139552 : rel->rd_refcnt = nailit ? 1 : 0;
3591 :
3592 : /* it's being created in this transaction */
3593 139552 : rel->rd_createSubid = GetCurrentSubTransactionId();
3594 139552 : rel->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
3595 139552 : rel->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
3596 139552 : rel->rd_droppedSubid = InvalidSubTransactionId;
3597 :
3598 : /*
3599 : * create a new tuple descriptor from the one passed in. We do this
3600 : * partly to copy it into the cache context, and partly because the new
3601 : * relation can't have any defaults or constraints yet; they have to be
3602 : * added in later steps, because they require additions to multiple system
3603 : * catalogs. We can copy attnotnull constraints here, however.
3604 : */
3605 139552 : rel->rd_att = CreateTupleDescCopy(tupDesc);
3606 139552 : rel->rd_att->tdrefcount = 1; /* mark as refcounted */
3607 139552 : has_not_null = false;
3608 604456 : for (i = 0; i < natts; i++)
3609 : {
3610 464904 : Form_pg_attribute satt = TupleDescAttr(tupDesc, i);
3611 464904 : Form_pg_attribute datt = TupleDescAttr(rel->rd_att, i);
3612 :
3613 464904 : datt->attidentity = satt->attidentity;
3614 464904 : datt->attgenerated = satt->attgenerated;
3615 464904 : datt->attnotnull = satt->attnotnull;
3616 464904 : has_not_null |= satt->attnotnull;
3617 464904 : populate_compact_attribute(rel->rd_att, i);
3618 :
3619 464904 : if (satt->attnotnull)
3620 : {
3621 77392 : CompactAttribute *scatt = TupleDescCompactAttr(tupDesc, i);
3622 77392 : CompactAttribute *dcatt = TupleDescCompactAttr(rel->rd_att, i);
3623 :
3624 77392 : dcatt->attnullability = scatt->attnullability;
3625 : }
3626 : }
3627 :
3628 139552 : if (has_not_null)
3629 : {
3630 21116 : TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
3631 :
3632 21116 : constr->has_not_null = true;
3633 21116 : rel->rd_att->constr = constr;
3634 : }
3635 :
3636 : /*
3637 : * initialize relation tuple form (caller may add/override data later)
3638 : */
3639 139552 : rel->rd_rel = (Form_pg_class) palloc0(CLASS_TUPLE_SIZE);
3640 :
3641 139552 : namestrcpy(&rel->rd_rel->relname, relname);
3642 139552 : rel->rd_rel->relnamespace = relnamespace;
3643 :
3644 139552 : rel->rd_rel->relkind = relkind;
3645 139552 : rel->rd_rel->relnatts = natts;
3646 139552 : rel->rd_rel->reltype = InvalidOid;
3647 : /* needed when bootstrapping: */
3648 139552 : rel->rd_rel->relowner = BOOTSTRAP_SUPERUSERID;
3649 :
3650 : /* set up persistence and relcache fields dependent on it */
3651 139552 : rel->rd_rel->relpersistence = relpersistence;
3652 139552 : switch (relpersistence)
3653 : {
3654 133414 : case RELPERSISTENCE_UNLOGGED:
3655 : case RELPERSISTENCE_PERMANENT:
3656 133414 : rel->rd_backend = INVALID_PROC_NUMBER;
3657 133414 : rel->rd_islocaltemp = false;
3658 133414 : break;
3659 6138 : case RELPERSISTENCE_TEMP:
3660 : Assert(isTempOrTempToastNamespace(relnamespace));
3661 6138 : rel->rd_backend = ProcNumberForTempRelations();
3662 6138 : rel->rd_islocaltemp = true;
3663 6138 : break;
3664 0 : default:
3665 0 : elog(ERROR, "invalid relpersistence: %c", relpersistence);
3666 : break;
3667 : }
3668 :
3669 : /* if it's a materialized view, it's not populated initially */
3670 139552 : if (relkind == RELKIND_MATVIEW)
3671 478 : rel->rd_rel->relispopulated = false;
3672 : else
3673 139074 : rel->rd_rel->relispopulated = true;
3674 :
3675 : /* set replica identity -- system catalogs and non-tables don't have one */
3676 139552 : if (!IsCatalogNamespace(relnamespace) &&
3677 76056 : (relkind == RELKIND_RELATION ||
3678 75578 : relkind == RELKIND_MATVIEW ||
3679 : relkind == RELKIND_PARTITIONED_TABLE))
3680 41746 : rel->rd_rel->relreplident = REPLICA_IDENTITY_DEFAULT;
3681 : else
3682 97806 : rel->rd_rel->relreplident = REPLICA_IDENTITY_NOTHING;
3683 :
3684 : /*
3685 : * Insert relation physical and logical identifiers (OIDs) into the right
3686 : * places. For a mapped relation, we set relfilenumber to zero and rely
3687 : * on RelationInitPhysicalAddr to consult the map.
3688 : */
3689 139552 : rel->rd_rel->relisshared = shared_relation;
3690 :
3691 139552 : RelationGetRelid(rel) = relid;
3692 :
3693 604456 : for (i = 0; i < natts; i++)
3694 464904 : TupleDescAttr(rel->rd_att, i)->attrelid = relid;
3695 :
3696 139552 : rel->rd_rel->reltablespace = reltablespace;
3697 :
3698 139552 : if (mapped_relation)
3699 : {
3700 6596 : rel->rd_rel->relfilenode = InvalidRelFileNumber;
3701 : /* Add it to the active mapping information */
3702 6596 : RelationMapUpdateMap(relid, relfilenumber, shared_relation, true);
3703 : }
3704 : else
3705 132956 : rel->rd_rel->relfilenode = relfilenumber;
3706 :
3707 139552 : RelationInitLockInfo(rel); /* see lmgr.c */
3708 :
3709 139552 : RelationInitPhysicalAddr(rel);
3710 :
3711 139552 : rel->rd_rel->relam = accessmtd;
3712 :
3713 : /*
3714 : * RelationInitTableAccessMethod will do syscache lookups, so we mustn't
3715 : * run it in CacheMemoryContext. Fortunately, the remaining steps don't
3716 : * require a long-lived current context.
3717 : */
3718 139552 : MemoryContextSwitchTo(oldcxt);
3719 :
3720 139552 : if (RELKIND_HAS_TABLE_AM(relkind) || relkind == RELKIND_SEQUENCE)
3721 63338 : RelationInitTableAccessMethod(rel);
3722 :
3723 : /*
3724 : * Leave index access method uninitialized, because the pg_index row has
3725 : * not been inserted at this stage of index creation yet. The cache
3726 : * invalidation after pg_index row has been inserted will initialize it.
3727 : */
3728 :
3729 : /*
3730 : * Okay to insert into the relcache hash table.
3731 : *
3732 : * Ordinarily, there should certainly not be an existing hash entry for
3733 : * the same OID; but during bootstrap, when we create a "real" relcache
3734 : * entry for one of the bootstrap relations, we'll be overwriting the
3735 : * phony one created with formrdesc. So allow that to happen for nailed
3736 : * rels.
3737 : */
3738 139552 : RelationCacheInsert(rel, nailit);
3739 :
3740 : /*
3741 : * Flag relation as needing eoxact cleanup (to clear rd_createSubid). We
3742 : * can't do this before storing relid in it.
3743 : */
3744 139552 : EOXactListAdd(rel);
3745 :
3746 : /* It's fully valid */
3747 139552 : rel->rd_isvalid = true;
3748 :
3749 : /*
3750 : * Caller expects us to pin the returned entry.
3751 : */
3752 139552 : RelationIncrementReferenceCount(rel);
3753 :
3754 139552 : return rel;
3755 : }
3756 :
3757 :
3758 : /*
3759 : * RelationSetNewRelfilenumber
3760 : *
3761 : * Assign a new relfilenumber (physical file name), and possibly a new
3762 : * persistence setting, to the relation.
3763 : *
3764 : * This allows a full rewrite of the relation to be done with transactional
3765 : * safety (since the filenumber assignment can be rolled back). Note however
3766 : * that there is no simple way to access the relation's old data for the
3767 : * remainder of the current transaction. This limits the usefulness to cases
3768 : * such as TRUNCATE or rebuilding an index from scratch.
3769 : *
3770 : * Caller must already hold exclusive lock on the relation.
3771 : */
3772 : void
3773 14938 : RelationSetNewRelfilenumber(Relation relation, char persistence)
3774 : {
3775 : RelFileNumber newrelfilenumber;
3776 : Relation pg_class;
3777 : ItemPointerData otid;
3778 : HeapTuple tuple;
3779 : Form_pg_class classform;
3780 14938 : MultiXactId minmulti = InvalidMultiXactId;
3781 14938 : TransactionId freezeXid = InvalidTransactionId;
3782 : RelFileLocator newrlocator;
3783 :
3784 14938 : if (!IsBinaryUpgrade)
3785 : {
3786 : /* Allocate a new relfilenumber */
3787 14842 : newrelfilenumber = GetNewRelFileNumber(relation->rd_rel->reltablespace,
3788 : NULL, persistence);
3789 : }
3790 96 : else if (relation->rd_rel->relkind == RELKIND_INDEX)
3791 : {
3792 48 : if (!OidIsValid(binary_upgrade_next_index_pg_class_relfilenumber))
3793 0 : ereport(ERROR,
3794 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3795 : errmsg("index relfilenumber value not set when in binary upgrade mode")));
3796 :
3797 48 : newrelfilenumber = binary_upgrade_next_index_pg_class_relfilenumber;
3798 48 : binary_upgrade_next_index_pg_class_relfilenumber = InvalidOid;
3799 : }
3800 48 : else if (relation->rd_rel->relkind == RELKIND_RELATION)
3801 : {
3802 48 : if (!OidIsValid(binary_upgrade_next_heap_pg_class_relfilenumber))
3803 0 : ereport(ERROR,
3804 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3805 : errmsg("heap relfilenumber value not set when in binary upgrade mode")));
3806 :
3807 48 : newrelfilenumber = binary_upgrade_next_heap_pg_class_relfilenumber;
3808 48 : binary_upgrade_next_heap_pg_class_relfilenumber = InvalidOid;
3809 : }
3810 : else
3811 0 : ereport(ERROR,
3812 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
3813 : errmsg("unexpected request for new relfilenumber in binary upgrade mode")));
3814 :
3815 : /*
3816 : * Get a writable copy of the pg_class tuple for the given relation.
3817 : */
3818 14938 : pg_class = table_open(RelationRelationId, RowExclusiveLock);
3819 :
3820 14938 : tuple = SearchSysCacheLockedCopy1(RELOID,
3821 : ObjectIdGetDatum(RelationGetRelid(relation)));
3822 14938 : if (!HeapTupleIsValid(tuple))
3823 0 : elog(ERROR, "could not find tuple for relation %u",
3824 : RelationGetRelid(relation));
3825 14938 : otid = tuple->t_self;
3826 14938 : classform = (Form_pg_class) GETSTRUCT(tuple);
3827 :
3828 : /*
3829 : * Schedule unlinking of the old storage at transaction commit, except
3830 : * when performing a binary upgrade, when we must do it immediately.
3831 : */
3832 14938 : if (IsBinaryUpgrade)
3833 : {
3834 : SMgrRelation srel;
3835 :
3836 : /*
3837 : * During a binary upgrade, we use this code path to ensure that
3838 : * pg_largeobject and its index have the same relfilenumbers as in the
3839 : * old cluster. This is necessary because pg_upgrade treats
3840 : * pg_largeobject like a user table, not a system table. It is however
3841 : * possible that a table or index may need to end up with the same
3842 : * relfilenumber in the new cluster as what it had in the old cluster.
3843 : * Hence, we can't wait until commit time to remove the old storage.
3844 : *
3845 : * In general, this function needs to have transactional semantics,
3846 : * and removing the old storage before commit time surely isn't.
3847 : * However, it doesn't really matter, because if a binary upgrade
3848 : * fails at this stage, the new cluster will need to be recreated
3849 : * anyway.
3850 : */
3851 96 : srel = smgropen(relation->rd_locator, relation->rd_backend);
3852 96 : smgrdounlinkall(&srel, 1, false);
3853 96 : smgrclose(srel);
3854 : }
3855 : else
3856 : {
3857 : /* Not a binary upgrade, so just schedule it to happen later. */
3858 14842 : RelationDropStorage(relation);
3859 : }
3860 :
3861 : /*
3862 : * Create storage for the main fork of the new relfilenumber. If it's a
3863 : * table-like object, call into the table AM to do so, which'll also
3864 : * create the table's init fork if needed.
3865 : *
3866 : * NOTE: If relevant for the AM, any conflict in relfilenumber value will
3867 : * be caught here, if GetNewRelFileNumber messes up for any reason.
3868 : */
3869 14938 : newrlocator = relation->rd_locator;
3870 14938 : newrlocator.relNumber = newrelfilenumber;
3871 :
3872 14938 : if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind))
3873 : {
3874 6448 : table_relation_set_new_filelocator(relation, &newrlocator,
3875 : persistence,
3876 : &freezeXid, &minmulti);
3877 : }
3878 8490 : else if (RELKIND_HAS_STORAGE(relation->rd_rel->relkind))
3879 8490 : {
3880 : /* handle these directly, at least for now */
3881 : SMgrRelation srel;
3882 :
3883 8490 : srel = RelationCreateStorage(newrlocator, persistence, true);
3884 8490 : smgrclose(srel);
3885 : }
3886 : else
3887 : {
3888 : /* we shouldn't be called for anything else */
3889 0 : elog(ERROR, "relation \"%s\" does not have storage",
3890 : RelationGetRelationName(relation));
3891 : }
3892 :
3893 : /*
3894 : * If we're dealing with a mapped index, pg_class.relfilenode doesn't
3895 : * change; instead we have to send the update to the relation mapper.
3896 : *
3897 : * For mapped indexes, we don't actually change the pg_class entry at all;
3898 : * this is essential when reindexing pg_class itself. That leaves us with
3899 : * possibly-inaccurate values of relpages etc, but those will be fixed up
3900 : * later.
3901 : */
3902 14938 : if (RelationIsMapped(relation))
3903 : {
3904 : /* This case is only supported for indexes */
3905 : Assert(relation->rd_rel->relkind == RELKIND_INDEX);
3906 :
3907 : /* Since we're not updating pg_class, these had better not change */
3908 : Assert(classform->relfrozenxid == freezeXid);
3909 : Assert(classform->relminmxid == minmulti);
3910 : Assert(classform->relpersistence == persistence);
3911 :
3912 : /*
3913 : * In some code paths it's possible that the tuple update we'd
3914 : * otherwise do here is the only thing that would assign an XID for
3915 : * the current transaction. However, we must have an XID to delete
3916 : * files, so make sure one is assigned.
3917 : */
3918 908 : (void) GetCurrentTransactionId();
3919 :
3920 : /* Do the deed */
3921 908 : RelationMapUpdateMap(RelationGetRelid(relation),
3922 : newrelfilenumber,
3923 908 : relation->rd_rel->relisshared,
3924 : false);
3925 :
3926 : /* Since we're not updating pg_class, must trigger inval manually */
3927 908 : CacheInvalidateRelcache(relation);
3928 : }
3929 : else
3930 : {
3931 : /* Normal case, update the pg_class entry */
3932 14030 : classform->relfilenode = newrelfilenumber;
3933 :
3934 : /* relpages etc. never change for sequences */
3935 14030 : if (relation->rd_rel->relkind != RELKIND_SEQUENCE)
3936 : {
3937 13748 : classform->relpages = 0; /* it's empty until further notice */
3938 13748 : classform->reltuples = -1;
3939 13748 : classform->relallvisible = 0;
3940 13748 : classform->relallfrozen = 0;
3941 : }
3942 14030 : classform->relfrozenxid = freezeXid;
3943 14030 : classform->relminmxid = minmulti;
3944 14030 : classform->relpersistence = persistence;
3945 :
3946 14030 : CatalogTupleUpdate(pg_class, &otid, tuple);
3947 : }
3948 :
3949 14938 : UnlockTuple(pg_class, &otid, InplaceUpdateTupleLock);
3950 14938 : heap_freetuple(tuple);
3951 :
3952 14938 : table_close(pg_class, RowExclusiveLock);
3953 :
3954 : /*
3955 : * Make the pg_class row change or relation map change visible. This will
3956 : * cause the relcache entry to get updated, too.
3957 : */
3958 14938 : CommandCounterIncrement();
3959 :
3960 14938 : RelationAssumeNewRelfilelocator(relation);
3961 14938 : }
3962 :
3963 : /*
3964 : * RelationAssumeNewRelfilelocator
3965 : *
3966 : * Code that modifies pg_class.reltablespace or pg_class.relfilenode must call
3967 : * this. The call shall precede any code that might insert WAL records whose
3968 : * replay would modify bytes in the new RelFileLocator, and the call shall follow
3969 : * any WAL modifying bytes in the prior RelFileLocator. See struct RelationData.
3970 : * Ideally, call this as near as possible to the CommandCounterIncrement()
3971 : * that makes the pg_class change visible (before it or after it); that
3972 : * minimizes the chance of future development adding a forbidden WAL insertion
3973 : * between RelationAssumeNewRelfilelocator() and CommandCounterIncrement().
3974 : */
3975 : void
3976 17504 : RelationAssumeNewRelfilelocator(Relation relation)
3977 : {
3978 17504 : relation->rd_newRelfilelocatorSubid = GetCurrentSubTransactionId();
3979 17504 : if (relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId)
3980 17414 : relation->rd_firstRelfilelocatorSubid = relation->rd_newRelfilelocatorSubid;
3981 :
3982 : /* Flag relation as needing eoxact cleanup (to clear these fields) */
3983 17504 : EOXactListAdd(relation);
3984 17504 : }
3985 :
3986 :
3987 : /*
3988 : * RelationCacheInitialize
3989 : *
3990 : * This initializes the relation descriptor cache. At the time
3991 : * that this is invoked, we can't do database access yet (mainly
3992 : * because the transaction subsystem is not up); all we are doing
3993 : * is making an empty cache hashtable. This must be done before
3994 : * starting the initialization transaction, because otherwise
3995 : * AtEOXact_RelationCache would crash if that transaction aborts
3996 : * before we can get the relcache set up.
3997 : */
3998 :
3999 : #define INITRELCACHESIZE 400
4000 :
4001 : void
4002 34516 : RelationCacheInitialize(void)
4003 : {
4004 : HASHCTL ctl;
4005 : int allocsize;
4006 :
4007 : /*
4008 : * make sure cache memory context exists
4009 : */
4010 34516 : if (!CacheMemoryContext)
4011 34516 : CreateCacheMemoryContext();
4012 :
4013 : /*
4014 : * create hashtable that indexes the relcache
4015 : */
4016 34516 : ctl.keysize = sizeof(Oid);
4017 34516 : ctl.entrysize = sizeof(RelIdCacheEnt);
4018 34516 : RelationIdCache = hash_create("Relcache by OID", INITRELCACHESIZE,
4019 : &ctl, HASH_ELEM | HASH_BLOBS);
4020 :
4021 : /*
4022 : * reserve enough in_progress_list slots for many cases
4023 : */
4024 34516 : allocsize = 4;
4025 34516 : in_progress_list =
4026 34516 : MemoryContextAlloc(CacheMemoryContext,
4027 : allocsize * sizeof(*in_progress_list));
4028 34516 : in_progress_list_maxlen = allocsize;
4029 :
4030 : /*
4031 : * relation mapper needs to be initialized too
4032 : */
4033 34516 : RelationMapInitialize();
4034 34516 : }
4035 :
4036 : /*
4037 : * RelationCacheInitializePhase2
4038 : *
4039 : * This is called to prepare for access to shared catalogs during startup.
4040 : * We must at least set up nailed reldescs for pg_database, pg_authid,
4041 : * pg_auth_members, and pg_shseclabel. Ideally we'd like to have reldescs
4042 : * for their indexes, too. We attempt to load this information from the
4043 : * shared relcache init file. If that's missing or broken, just make
4044 : * phony entries for the catalogs themselves.
4045 : * RelationCacheInitializePhase3 will clean up as needed.
4046 : */
4047 : void
4048 34516 : RelationCacheInitializePhase2(void)
4049 : {
4050 : MemoryContext oldcxt;
4051 :
4052 : /*
4053 : * relation mapper needs initialized too
4054 : */
4055 34516 : RelationMapInitializePhase2();
4056 :
4057 : /*
4058 : * In bootstrap mode, the shared catalogs aren't there yet anyway, so do
4059 : * nothing.
4060 : */
4061 34516 : if (IsBootstrapProcessingMode())
4062 102 : return;
4063 :
4064 : /*
4065 : * switch to cache memory context
4066 : */
4067 34414 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4068 :
4069 : /*
4070 : * Try to load the shared relcache cache file. If unsuccessful, bootstrap
4071 : * the cache with pre-made descriptors for the critical shared catalogs.
4072 : */
4073 34414 : if (!load_relcache_init_file(true))
4074 : {
4075 4648 : formrdesc("pg_database", DatabaseRelation_Rowtype_Id, true,
4076 : Natts_pg_database, Desc_pg_database);
4077 4648 : formrdesc("pg_authid", AuthIdRelation_Rowtype_Id, true,
4078 : Natts_pg_authid, Desc_pg_authid);
4079 4648 : formrdesc("pg_auth_members", AuthMemRelation_Rowtype_Id, true,
4080 : Natts_pg_auth_members, Desc_pg_auth_members);
4081 4648 : formrdesc("pg_shseclabel", SharedSecLabelRelation_Rowtype_Id, true,
4082 : Natts_pg_shseclabel, Desc_pg_shseclabel);
4083 4648 : formrdesc("pg_subscription", SubscriptionRelation_Rowtype_Id, true,
4084 : Natts_pg_subscription, Desc_pg_subscription);
4085 :
4086 : #define NUM_CRITICAL_SHARED_RELS 5 /* fix if you change list above */
4087 : }
4088 :
4089 34414 : MemoryContextSwitchTo(oldcxt);
4090 : }
4091 :
4092 : /*
4093 : * RelationCacheInitializePhase3
4094 : *
4095 : * This is called as soon as the catcache and transaction system
4096 : * are functional and we have determined MyDatabaseId. At this point
4097 : * we can actually read data from the database's system catalogs.
4098 : * We first try to read pre-computed relcache entries from the local
4099 : * relcache init file. If that's missing or broken, make phony entries
4100 : * for the minimum set of nailed-in-cache relations. Then (unless
4101 : * bootstrapping) make sure we have entries for the critical system
4102 : * indexes. Once we've done all this, we have enough infrastructure to
4103 : * open any system catalog or use any catcache. The last step is to
4104 : * rewrite the cache files if needed.
4105 : */
4106 : void
4107 31880 : RelationCacheInitializePhase3(void)
4108 : {
4109 : HASH_SEQ_STATUS status;
4110 : RelIdCacheEnt *idhentry;
4111 : MemoryContext oldcxt;
4112 31880 : bool needNewCacheFile = !criticalSharedRelcachesBuilt;
4113 :
4114 : /*
4115 : * relation mapper needs initialized too
4116 : */
4117 31880 : RelationMapInitializePhase3();
4118 :
4119 : /*
4120 : * switch to cache memory context
4121 : */
4122 31880 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4123 :
4124 : /*
4125 : * Try to load the local relcache cache file. If unsuccessful, bootstrap
4126 : * the cache with pre-made descriptors for the critical "nailed-in" system
4127 : * catalogs.
4128 : */
4129 31880 : if (IsBootstrapProcessingMode() ||
4130 31778 : !load_relcache_init_file(false))
4131 : {
4132 3492 : needNewCacheFile = true;
4133 :
4134 3492 : formrdesc("pg_class", RelationRelation_Rowtype_Id, false,
4135 : Natts_pg_class, Desc_pg_class);
4136 3492 : formrdesc("pg_attribute", AttributeRelation_Rowtype_Id, false,
4137 : Natts_pg_attribute, Desc_pg_attribute);
4138 3492 : formrdesc("pg_proc", ProcedureRelation_Rowtype_Id, false,
4139 : Natts_pg_proc, Desc_pg_proc);
4140 3492 : formrdesc("pg_type", TypeRelation_Rowtype_Id, false,
4141 : Natts_pg_type, Desc_pg_type);
4142 :
4143 : #define NUM_CRITICAL_LOCAL_RELS 4 /* fix if you change list above */
4144 : }
4145 :
4146 31880 : MemoryContextSwitchTo(oldcxt);
4147 :
4148 : /* In bootstrap mode, the faked-up formrdesc info is all we'll have */
4149 31880 : if (IsBootstrapProcessingMode())
4150 102 : return;
4151 :
4152 : /*
4153 : * If we didn't get the critical system indexes loaded into relcache, do
4154 : * so now. These are critical because the catcache and/or opclass cache
4155 : * depend on them for fetches done during relcache load. Thus, we have an
4156 : * infinite-recursion problem. We can break the recursion by doing
4157 : * heapscans instead of indexscans at certain key spots. To avoid hobbling
4158 : * performance, we only want to do that until we have the critical indexes
4159 : * loaded into relcache. Thus, the flag criticalRelcachesBuilt is used to
4160 : * decide whether to do heapscan or indexscan at the key spots, and we set
4161 : * it true after we've loaded the critical indexes.
4162 : *
4163 : * The critical indexes are marked as "nailed in cache", partly to make it
4164 : * easy for load_relcache_init_file to count them, but mainly because we
4165 : * cannot flush and rebuild them once we've set criticalRelcachesBuilt to
4166 : * true. (NOTE: perhaps it would be possible to reload them by
4167 : * temporarily setting criticalRelcachesBuilt to false again. For now,
4168 : * though, we just nail 'em in.)
4169 : *
4170 : * RewriteRelRulenameIndexId and TriggerRelidNameIndexId are not critical
4171 : * in the same way as the others, because the critical catalogs don't
4172 : * (currently) have any rules or triggers, and so these indexes can be
4173 : * rebuilt without inducing recursion. However they are used during
4174 : * relcache load when a rel does have rules or triggers, so we choose to
4175 : * nail them for performance reasons.
4176 : */
4177 31778 : if (!criticalRelcachesBuilt)
4178 : {
4179 3390 : load_critical_index(ClassOidIndexId,
4180 : RelationRelationId);
4181 3390 : load_critical_index(AttributeRelidNumIndexId,
4182 : AttributeRelationId);
4183 3390 : load_critical_index(IndexRelidIndexId,
4184 : IndexRelationId);
4185 3390 : load_critical_index(OpclassOidIndexId,
4186 : OperatorClassRelationId);
4187 3388 : load_critical_index(AccessMethodProcedureIndexId,
4188 : AccessMethodProcedureRelationId);
4189 3388 : load_critical_index(RewriteRelRulenameIndexId,
4190 : RewriteRelationId);
4191 3388 : load_critical_index(TriggerRelidNameIndexId,
4192 : TriggerRelationId);
4193 :
4194 : #define NUM_CRITICAL_LOCAL_INDEXES 7 /* fix if you change list above */
4195 :
4196 3388 : criticalRelcachesBuilt = true;
4197 : }
4198 :
4199 : /*
4200 : * Process critical shared indexes too.
4201 : *
4202 : * DatabaseNameIndexId isn't critical for relcache loading, but rather for
4203 : * initial lookup of MyDatabaseId, without which we'll never find any
4204 : * non-shared catalogs at all. Autovacuum calls InitPostgres with a
4205 : * database OID, so it instead depends on DatabaseOidIndexId. We also
4206 : * need to nail up some indexes on pg_authid and pg_auth_members for use
4207 : * during client authentication. SharedSecLabelObjectIndexId isn't
4208 : * critical for the core system, but authentication hooks might be
4209 : * interested in it.
4210 : */
4211 31776 : if (!criticalSharedRelcachesBuilt)
4212 : {
4213 2790 : load_critical_index(DatabaseNameIndexId,
4214 : DatabaseRelationId);
4215 2790 : load_critical_index(DatabaseOidIndexId,
4216 : DatabaseRelationId);
4217 2790 : load_critical_index(AuthIdRolnameIndexId,
4218 : AuthIdRelationId);
4219 2790 : load_critical_index(AuthIdOidIndexId,
4220 : AuthIdRelationId);
4221 2790 : load_critical_index(AuthMemMemRoleIndexId,
4222 : AuthMemRelationId);
4223 2790 : load_critical_index(SharedSecLabelObjectIndexId,
4224 : SharedSecLabelRelationId);
4225 :
4226 : #define NUM_CRITICAL_SHARED_INDEXES 6 /* fix if you change list above */
4227 :
4228 2790 : criticalSharedRelcachesBuilt = true;
4229 : }
4230 :
4231 : /*
4232 : * Now, scan all the relcache entries and update anything that might be
4233 : * wrong in the results from formrdesc or the relcache cache file. If we
4234 : * faked up relcache entries using formrdesc, then read the real pg_class
4235 : * rows and replace the fake entries with them. Also, if any of the
4236 : * relcache entries have rules, triggers, or security policies, load that
4237 : * info the hard way since it isn't recorded in the cache file.
4238 : *
4239 : * Whenever we access the catalogs to read data, there is a possibility of
4240 : * a shared-inval cache flush causing relcache entries to be removed.
4241 : * Since hash_seq_search only guarantees to still work after the *current*
4242 : * entry is removed, it's unsafe to continue the hashtable scan afterward.
4243 : * We handle this by restarting the scan from scratch after each access.
4244 : * This is theoretically O(N^2), but the number of entries that actually
4245 : * need to be fixed is small enough that it doesn't matter.
4246 : */
4247 31776 : hash_seq_init(&status, RelationIdCache);
4248 :
4249 4642592 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
4250 : {
4251 4579040 : Relation relation = idhentry->reldesc;
4252 4579040 : bool restart = false;
4253 :
4254 : /*
4255 : * Make sure *this* entry doesn't get flushed while we work with it.
4256 : */
4257 4579040 : RelationIncrementReferenceCount(relation);
4258 :
4259 : /*
4260 : * If it's a faked-up entry, read the real pg_class tuple.
4261 : */
4262 4579040 : if (relation->rd_rel->relowner == InvalidOid)
4263 : {
4264 : HeapTuple htup;
4265 : Form_pg_class relp;
4266 :
4267 27500 : htup = SearchSysCache1(RELOID,
4268 : ObjectIdGetDatum(RelationGetRelid(relation)));
4269 27500 : if (!HeapTupleIsValid(htup))
4270 0 : ereport(FATAL,
4271 : errcode(ERRCODE_UNDEFINED_OBJECT),
4272 : errmsg_internal("cache lookup failed for relation %u",
4273 : RelationGetRelid(relation)));
4274 27500 : relp = (Form_pg_class) GETSTRUCT(htup);
4275 :
4276 : /*
4277 : * Copy tuple to relation->rd_rel. (See notes in
4278 : * AllocateRelationDesc())
4279 : */
4280 27500 : memcpy((char *) relation->rd_rel, (char *) relp, CLASS_TUPLE_SIZE);
4281 :
4282 : /* Update rd_options while we have the tuple */
4283 27500 : if (relation->rd_options)
4284 0 : pfree(relation->rd_options);
4285 27500 : RelationParseRelOptions(relation, htup);
4286 :
4287 : /*
4288 : * Check the values in rd_att were set up correctly. (We cannot
4289 : * just copy them over now: formrdesc must have set up the rd_att
4290 : * data correctly to start with, because it may already have been
4291 : * copied into one or more catcache entries.)
4292 : */
4293 : Assert(relation->rd_att->tdtypeid == relp->reltype);
4294 : Assert(relation->rd_att->tdtypmod == -1);
4295 :
4296 27500 : ReleaseSysCache(htup);
4297 :
4298 : /* relowner had better be OK now, else we'll loop forever */
4299 27500 : if (relation->rd_rel->relowner == InvalidOid)
4300 0 : elog(ERROR, "invalid relowner in pg_class entry for \"%s\"",
4301 : RelationGetRelationName(relation));
4302 :
4303 27500 : restart = true;
4304 : }
4305 :
4306 : /*
4307 : * Fix data that isn't saved in relcache cache file.
4308 : *
4309 : * relhasrules or relhastriggers could possibly be wrong or out of
4310 : * date. If we don't actually find any rules or triggers, clear the
4311 : * local copy of the flag so that we don't get into an infinite loop
4312 : * here. We don't make any attempt to fix the pg_class entry, though.
4313 : */
4314 4579040 : if (relation->rd_rel->relhasrules && relation->rd_rules == NULL)
4315 : {
4316 0 : RelationBuildRuleLock(relation);
4317 0 : if (relation->rd_rules == NULL)
4318 0 : relation->rd_rel->relhasrules = false;
4319 0 : restart = true;
4320 : }
4321 4579040 : if (relation->rd_rel->relhastriggers && relation->trigdesc == NULL)
4322 : {
4323 0 : RelationBuildTriggers(relation);
4324 0 : if (relation->trigdesc == NULL)
4325 0 : relation->rd_rel->relhastriggers = false;
4326 0 : restart = true;
4327 : }
4328 :
4329 : /*
4330 : * Re-load the row security policies if the relation has them, since
4331 : * they are not preserved in the cache. Note that we can never NOT
4332 : * have a policy while relrowsecurity is true,
4333 : * RelationBuildRowSecurity will create a single default-deny policy
4334 : * if there is no policy defined in pg_policy.
4335 : */
4336 4579040 : if (relation->rd_rel->relrowsecurity && relation->rd_rsdesc == NULL)
4337 : {
4338 0 : RelationBuildRowSecurity(relation);
4339 :
4340 : Assert(relation->rd_rsdesc != NULL);
4341 0 : restart = true;
4342 : }
4343 :
4344 : /* Reload tableam data if needed */
4345 4579040 : if (relation->rd_tableam == NULL &&
4346 2799872 : (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) || relation->rd_rel->relkind == RELKIND_SEQUENCE))
4347 : {
4348 0 : RelationInitTableAccessMethod(relation);
4349 : Assert(relation->rd_tableam != NULL);
4350 :
4351 0 : restart = true;
4352 : }
4353 :
4354 : /* Release hold on the relation */
4355 4579040 : RelationDecrementReferenceCount(relation);
4356 :
4357 : /* Now, restart the hashtable scan if needed */
4358 4579040 : if (restart)
4359 : {
4360 27500 : hash_seq_term(&status);
4361 27500 : hash_seq_init(&status, RelationIdCache);
4362 : }
4363 : }
4364 :
4365 : /*
4366 : * Lastly, write out new relcache cache files if needed. We don't bother
4367 : * to distinguish cases where only one of the two needs an update.
4368 : */
4369 31776 : if (needNewCacheFile)
4370 : {
4371 : /*
4372 : * Force all the catcaches to finish initializing and thereby open the
4373 : * catalogs and indexes they use. This will preload the relcache with
4374 : * entries for all the most important system catalogs and indexes, so
4375 : * that the init files will be most useful for future backends.
4376 : */
4377 3906 : InitCatalogCachePhase2();
4378 :
4379 : /* now write the files */
4380 3902 : write_relcache_init_file(true);
4381 3902 : write_relcache_init_file(false);
4382 : }
4383 : }
4384 :
4385 : /*
4386 : * Load one critical system index into the relcache
4387 : *
4388 : * indexoid is the OID of the target index, heapoid is the OID of the catalog
4389 : * it belongs to.
4390 : */
4391 : static void
4392 40464 : load_critical_index(Oid indexoid, Oid heapoid)
4393 : {
4394 : Relation ird;
4395 :
4396 : /*
4397 : * We must lock the underlying catalog before locking the index to avoid
4398 : * deadlock, since RelationBuildDesc might well need to read the catalog,
4399 : * and if anyone else is exclusive-locking this catalog and index they'll
4400 : * be doing it in that order.
4401 : */
4402 40464 : LockRelationOid(heapoid, AccessShareLock);
4403 40464 : LockRelationOid(indexoid, AccessShareLock);
4404 40464 : ird = RelationBuildDesc(indexoid, true);
4405 40462 : if (ird == NULL)
4406 0 : ereport(PANIC,
4407 : errcode(ERRCODE_DATA_CORRUPTED),
4408 : errmsg_internal("could not open critical system index %u", indexoid));
4409 40462 : ird->rd_isnailed = true;
4410 40462 : ird->rd_refcnt = 1;
4411 40462 : UnlockRelationOid(indexoid, AccessShareLock);
4412 40462 : UnlockRelationOid(heapoid, AccessShareLock);
4413 :
4414 40462 : (void) RelationGetIndexAttOptions(ird, false);
4415 40462 : }
4416 :
4417 : /*
4418 : * GetPgClassDescriptor -- get a predefined tuple descriptor for pg_class
4419 : * GetPgIndexDescriptor -- get a predefined tuple descriptor for pg_index
4420 : *
4421 : * We need this kluge because we have to be able to access non-fixed-width
4422 : * fields of pg_class and pg_index before we have the standard catalog caches
4423 : * available. We use predefined data that's set up in just the same way as
4424 : * the bootstrapped reldescs used by formrdesc(). The resulting tupdesc is
4425 : * not 100% kosher: it does not have the correct rowtype OID in tdtypeid, nor
4426 : * does it have a TupleConstr field. But it's good enough for the purpose of
4427 : * extracting fields.
4428 : */
4429 : static TupleDesc
4430 63758 : BuildHardcodedDescriptor(int natts, const FormData_pg_attribute *attrs)
4431 : {
4432 : TupleDesc result;
4433 : MemoryContext oldcxt;
4434 : int i;
4435 :
4436 63758 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4437 :
4438 63758 : result = CreateTemplateTupleDesc(natts);
4439 63758 : result->tdtypeid = RECORDOID; /* not right, but we don't care */
4440 63758 : result->tdtypmod = -1;
4441 :
4442 1817116 : for (i = 0; i < natts; i++)
4443 : {
4444 1753358 : memcpy(TupleDescAttr(result, i), &attrs[i], ATTRIBUTE_FIXED_PART_SIZE);
4445 :
4446 1753358 : populate_compact_attribute(result, i);
4447 : }
4448 :
4449 : /* initialize first attribute's attcacheoff, cf RelationBuildTupleDesc */
4450 63758 : TupleDescCompactAttr(result, 0)->attcacheoff = 0;
4451 :
4452 : /* Note: we don't bother to set up a TupleConstr entry */
4453 :
4454 63758 : MemoryContextSwitchTo(oldcxt);
4455 :
4456 63758 : return result;
4457 : }
4458 :
4459 : static TupleDesc
4460 1891740 : GetPgClassDescriptor(void)
4461 : {
4462 : static TupleDesc pgclassdesc = NULL;
4463 :
4464 : /* Already done? */
4465 1891740 : if (pgclassdesc == NULL)
4466 31880 : pgclassdesc = BuildHardcodedDescriptor(Natts_pg_class,
4467 : Desc_pg_class);
4468 :
4469 1891740 : return pgclassdesc;
4470 : }
4471 :
4472 : static TupleDesc
4473 2311926 : GetPgIndexDescriptor(void)
4474 : {
4475 : static TupleDesc pgindexdesc = NULL;
4476 :
4477 : /* Already done? */
4478 2311926 : if (pgindexdesc == NULL)
4479 31878 : pgindexdesc = BuildHardcodedDescriptor(Natts_pg_index,
4480 : Desc_pg_index);
4481 :
4482 2311926 : return pgindexdesc;
4483 : }
4484 :
4485 : /*
4486 : * Load any default attribute value definitions for the relation.
4487 : *
4488 : * ndef is the number of attributes that were marked atthasdef.
4489 : *
4490 : * Note: we don't make it a hard error to be missing some pg_attrdef records.
4491 : * We can limp along as long as nothing needs to use the default value. Code
4492 : * that fails to find an expected AttrDefault record should throw an error.
4493 : */
4494 : static void
4495 37538 : AttrDefaultFetch(Relation relation, int ndef)
4496 : {
4497 : AttrDefault *attrdef;
4498 : Relation adrel;
4499 : SysScanDesc adscan;
4500 : ScanKeyData skey;
4501 : HeapTuple htup;
4502 37538 : int found = 0;
4503 :
4504 : /* Allocate array with room for as many entries as expected */
4505 : attrdef = (AttrDefault *)
4506 37538 : MemoryContextAllocZero(CacheMemoryContext,
4507 : ndef * sizeof(AttrDefault));
4508 :
4509 : /* Search pg_attrdef for relevant entries */
4510 37538 : ScanKeyInit(&skey,
4511 : Anum_pg_attrdef_adrelid,
4512 : BTEqualStrategyNumber, F_OIDEQ,
4513 : ObjectIdGetDatum(RelationGetRelid(relation)));
4514 :
4515 37538 : adrel = table_open(AttrDefaultRelationId, AccessShareLock);
4516 37538 : adscan = systable_beginscan(adrel, AttrDefaultIndexId, true,
4517 : NULL, 1, &skey);
4518 :
4519 90756 : while (HeapTupleIsValid(htup = systable_getnext(adscan)))
4520 : {
4521 53218 : Form_pg_attrdef adform = (Form_pg_attrdef) GETSTRUCT(htup);
4522 : Datum val;
4523 : bool isnull;
4524 :
4525 : /* protect limited size of array */
4526 53218 : if (found >= ndef)
4527 : {
4528 0 : elog(WARNING, "unexpected pg_attrdef record found for attribute %d of relation \"%s\"",
4529 : adform->adnum, RelationGetRelationName(relation));
4530 0 : break;
4531 : }
4532 :
4533 53218 : val = fastgetattr(htup,
4534 : Anum_pg_attrdef_adbin,
4535 : adrel->rd_att, &isnull);
4536 53218 : if (isnull)
4537 0 : elog(WARNING, "null adbin for attribute %d of relation \"%s\"",
4538 : adform->adnum, RelationGetRelationName(relation));
4539 : else
4540 : {
4541 : /* detoast and convert to cstring in caller's context */
4542 53218 : char *s = TextDatumGetCString(val);
4543 :
4544 53218 : attrdef[found].adnum = adform->adnum;
4545 53218 : attrdef[found].adbin = MemoryContextStrdup(CacheMemoryContext, s);
4546 53218 : pfree(s);
4547 53218 : found++;
4548 : }
4549 : }
4550 :
4551 37538 : systable_endscan(adscan);
4552 37538 : table_close(adrel, AccessShareLock);
4553 :
4554 37538 : if (found != ndef)
4555 0 : elog(WARNING, "%d pg_attrdef record(s) missing for relation \"%s\"",
4556 : ndef - found, RelationGetRelationName(relation));
4557 :
4558 : /*
4559 : * Sort the AttrDefault entries by adnum, for the convenience of
4560 : * equalTupleDescs(). (Usually, they already will be in order, but this
4561 : * might not be so if systable_getnext isn't using an index.)
4562 : */
4563 37538 : if (found > 1)
4564 8812 : qsort(attrdef, found, sizeof(AttrDefault), AttrDefaultCmp);
4565 :
4566 : /* Install array only after it's fully valid */
4567 37538 : relation->rd_att->constr->defval = attrdef;
4568 37538 : relation->rd_att->constr->num_defval = found;
4569 37538 : }
4570 :
4571 : /*
4572 : * qsort comparator to sort AttrDefault entries by adnum
4573 : */
4574 : static int
4575 15680 : AttrDefaultCmp(const void *a, const void *b)
4576 : {
4577 15680 : const AttrDefault *ada = (const AttrDefault *) a;
4578 15680 : const AttrDefault *adb = (const AttrDefault *) b;
4579 :
4580 15680 : return pg_cmp_s16(ada->adnum, adb->adnum);
4581 : }
4582 :
4583 : /*
4584 : * Load any check constraints for the relation, and update not-null validity
4585 : * of invalid constraints.
4586 : *
4587 : * As with defaults, if we don't find the expected number of them, just warn
4588 : * here. The executor should throw an error if an INSERT/UPDATE is attempted.
4589 : */
4590 : static void
4591 175010 : CheckNNConstraintFetch(Relation relation)
4592 : {
4593 : ConstrCheck *check;
4594 175010 : int ncheck = relation->rd_rel->relchecks;
4595 : Relation conrel;
4596 : SysScanDesc conscan;
4597 : ScanKeyData skey[1];
4598 : HeapTuple htup;
4599 175010 : int found = 0;
4600 :
4601 : /* Allocate array with room for as many entries as expected, if needed */
4602 175010 : if (ncheck > 0)
4603 : check = (ConstrCheck *)
4604 12976 : MemoryContextAllocZero(CacheMemoryContext,
4605 : ncheck * sizeof(ConstrCheck));
4606 : else
4607 162034 : check = NULL;
4608 :
4609 : /* Search pg_constraint for relevant entries */
4610 175010 : ScanKeyInit(&skey[0],
4611 : Anum_pg_constraint_conrelid,
4612 : BTEqualStrategyNumber, F_OIDEQ,
4613 : ObjectIdGetDatum(RelationGetRelid(relation)));
4614 :
4615 175010 : conrel = table_open(ConstraintRelationId, AccessShareLock);
4616 175010 : conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
4617 : NULL, 1, skey);
4618 :
4619 506450 : while (HeapTupleIsValid(htup = systable_getnext(conscan)))
4620 : {
4621 331440 : Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(htup);
4622 : Datum val;
4623 : bool isnull;
4624 :
4625 : /*
4626 : * If this is a not-null constraint, then only look at it if it's
4627 : * invalid, and if so, mark the TupleDesc entry as known invalid.
4628 : * Otherwise move on. We'll mark any remaining columns that are still
4629 : * in UNKNOWN state as known valid later. This allows us not to have
4630 : * to extract the attnum from this constraint tuple in the vast
4631 : * majority of cases.
4632 : */
4633 331440 : if (conform->contype == CONSTRAINT_NOTNULL)
4634 : {
4635 187138 : if (!conform->convalidated)
4636 : {
4637 : AttrNumber attnum;
4638 :
4639 860 : attnum = extractNotNullColumn(htup);
4640 : Assert(relation->rd_att->compact_attrs[attnum - 1].attnullability ==
4641 : ATTNULLABLE_UNKNOWN);
4642 860 : relation->rd_att->compact_attrs[attnum - 1].attnullability =
4643 : ATTNULLABLE_INVALID;
4644 : }
4645 :
4646 311522 : continue;
4647 : }
4648 :
4649 : /* For what follows, consider check constraints only */
4650 144302 : if (conform->contype != CONSTRAINT_CHECK)
4651 124384 : continue;
4652 :
4653 : /* protect limited size of array */
4654 19918 : if (found >= ncheck)
4655 : {
4656 0 : elog(WARNING, "unexpected pg_constraint record found for relation \"%s\"",
4657 : RelationGetRelationName(relation));
4658 0 : break;
4659 : }
4660 :
4661 19918 : check[found].ccenforced = conform->conenforced;
4662 19918 : check[found].ccvalid = conform->convalidated;
4663 19918 : check[found].ccnoinherit = conform->connoinherit;
4664 39836 : check[found].ccname = MemoryContextStrdup(CacheMemoryContext,
4665 19918 : NameStr(conform->conname));
4666 :
4667 : /* Grab and test conbin is actually set */
4668 19918 : val = fastgetattr(htup,
4669 : Anum_pg_constraint_conbin,
4670 : conrel->rd_att, &isnull);
4671 19918 : if (isnull)
4672 0 : elog(WARNING, "null conbin for relation \"%s\"",
4673 : RelationGetRelationName(relation));
4674 : else
4675 : {
4676 : /* detoast and convert to cstring in caller's context */
4677 19918 : char *s = TextDatumGetCString(val);
4678 :
4679 19918 : check[found].ccbin = MemoryContextStrdup(CacheMemoryContext, s);
4680 19918 : pfree(s);
4681 19918 : found++;
4682 : }
4683 : }
4684 :
4685 175010 : systable_endscan(conscan);
4686 175010 : table_close(conrel, AccessShareLock);
4687 :
4688 175010 : if (found != ncheck)
4689 0 : elog(WARNING, "%d pg_constraint record(s) missing for relation \"%s\"",
4690 : ncheck - found, RelationGetRelationName(relation));
4691 :
4692 : /*
4693 : * Sort the records by name. This ensures that CHECKs are applied in a
4694 : * deterministic order, and it also makes equalTupleDescs() faster.
4695 : */
4696 175010 : if (found > 1)
4697 3774 : qsort(check, found, sizeof(ConstrCheck), CheckConstraintCmp);
4698 :
4699 : /* Install array only after it's fully valid */
4700 175010 : relation->rd_att->constr->check = check;
4701 175010 : relation->rd_att->constr->num_check = found;
4702 175010 : }
4703 :
4704 : /*
4705 : * qsort comparator to sort ConstrCheck entries by name
4706 : */
4707 : static int
4708 6942 : CheckConstraintCmp(const void *a, const void *b)
4709 : {
4710 6942 : const ConstrCheck *ca = (const ConstrCheck *) a;
4711 6942 : const ConstrCheck *cb = (const ConstrCheck *) b;
4712 :
4713 6942 : return strcmp(ca->ccname, cb->ccname);
4714 : }
4715 :
4716 : /*
4717 : * RelationGetFKeyList -- get a list of foreign key info for the relation
4718 : *
4719 : * Returns a list of ForeignKeyCacheInfo structs, one per FK constraining
4720 : * the given relation. This data is a direct copy of relevant fields from
4721 : * pg_constraint. The list items are in no particular order.
4722 : *
4723 : * CAUTION: the returned list is part of the relcache's data, and could
4724 : * vanish in a relcache entry reset. Callers must inspect or copy it
4725 : * before doing anything that might trigger a cache flush, such as
4726 : * system catalog accesses. copyObject() can be used if desired.
4727 : * (We define it this way because current callers want to filter and
4728 : * modify the list entries anyway, so copying would be a waste of time.)
4729 : */
4730 : List *
4731 230310 : RelationGetFKeyList(Relation relation)
4732 : {
4733 : List *result;
4734 : Relation conrel;
4735 : SysScanDesc conscan;
4736 : ScanKeyData skey;
4737 : HeapTuple htup;
4738 : List *oldlist;
4739 : MemoryContext oldcxt;
4740 :
4741 : /* Quick exit if we already computed the list. */
4742 230310 : if (relation->rd_fkeyvalid)
4743 184054 : return relation->rd_fkeylist;
4744 :
4745 : /*
4746 : * We build the list we intend to return (in the caller's context) while
4747 : * doing the scan. After successfully completing the scan, we copy that
4748 : * list into the relcache entry. This avoids cache-context memory leakage
4749 : * if we get some sort of error partway through.
4750 : */
4751 46256 : result = NIL;
4752 :
4753 : /* Prepare to scan pg_constraint for entries having conrelid = this rel. */
4754 46256 : ScanKeyInit(&skey,
4755 : Anum_pg_constraint_conrelid,
4756 : BTEqualStrategyNumber, F_OIDEQ,
4757 : ObjectIdGetDatum(RelationGetRelid(relation)));
4758 :
4759 46256 : conrel = table_open(ConstraintRelationId, AccessShareLock);
4760 46256 : conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
4761 : NULL, 1, &skey);
4762 :
4763 144904 : while (HeapTupleIsValid(htup = systable_getnext(conscan)))
4764 : {
4765 98648 : Form_pg_constraint constraint = (Form_pg_constraint) GETSTRUCT(htup);
4766 : ForeignKeyCacheInfo *info;
4767 :
4768 : /* consider only foreign keys */
4769 98648 : if (constraint->contype != CONSTRAINT_FOREIGN)
4770 94864 : continue;
4771 :
4772 3784 : info = makeNode(ForeignKeyCacheInfo);
4773 3784 : info->conoid = constraint->oid;
4774 3784 : info->conrelid = constraint->conrelid;
4775 3784 : info->confrelid = constraint->confrelid;
4776 3784 : info->conenforced = constraint->conenforced;
4777 :
4778 3784 : DeconstructFkConstraintRow(htup, &info->nkeys,
4779 3784 : info->conkey,
4780 3784 : info->confkey,
4781 3784 : info->conpfeqop,
4782 : NULL, NULL, NULL, NULL);
4783 :
4784 : /* Add FK's node to the result list */
4785 3784 : result = lappend(result, info);
4786 : }
4787 :
4788 46256 : systable_endscan(conscan);
4789 46256 : table_close(conrel, AccessShareLock);
4790 :
4791 : /* Now save a copy of the completed list in the relcache entry. */
4792 46256 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4793 46256 : oldlist = relation->rd_fkeylist;
4794 46256 : relation->rd_fkeylist = copyObject(result);
4795 46256 : relation->rd_fkeyvalid = true;
4796 46256 : MemoryContextSwitchTo(oldcxt);
4797 :
4798 : /* Don't leak the old list, if there is one */
4799 46256 : list_free_deep(oldlist);
4800 :
4801 46256 : return result;
4802 : }
4803 :
4804 : /*
4805 : * RelationGetIndexList -- get a list of OIDs of indexes on this relation
4806 : *
4807 : * The index list is created only if someone requests it. We scan pg_index
4808 : * to find relevant indexes, and add the list to the relcache entry so that
4809 : * we won't have to compute it again. Note that shared cache inval of a
4810 : * relcache entry will delete the old list and set rd_indexvalid to false,
4811 : * so that we must recompute the index list on next request. This handles
4812 : * creation or deletion of an index.
4813 : *
4814 : * Indexes that are marked not indislive are omitted from the returned list.
4815 : * Such indexes are expected to be dropped momentarily, and should not be
4816 : * touched at all by any caller of this function.
4817 : *
4818 : * The returned list is guaranteed to be sorted in order by OID. This is
4819 : * needed by the executor, since for index types that we obtain exclusive
4820 : * locks on when updating the index, all backends must lock the indexes in
4821 : * the same order or we will get deadlocks (see ExecOpenIndices()). Any
4822 : * consistent ordering would do, but ordering by OID is easy.
4823 : *
4824 : * Since shared cache inval causes the relcache's copy of the list to go away,
4825 : * we return a copy of the list palloc'd in the caller's context. The caller
4826 : * may list_free() the returned list after scanning it. This is necessary
4827 : * since the caller will typically be doing syscache lookups on the relevant
4828 : * indexes, and syscache lookup could cause SI messages to be processed!
4829 : *
4830 : * In exactly the same way, we update rd_pkindex, which is the OID of the
4831 : * relation's primary key index if any, else InvalidOid; and rd_replidindex,
4832 : * which is the pg_class OID of an index to be used as the relation's
4833 : * replication identity index, or InvalidOid if there is no such index.
4834 : */
4835 : List *
4836 2525926 : RelationGetIndexList(Relation relation)
4837 : {
4838 : Relation indrel;
4839 : SysScanDesc indscan;
4840 : ScanKeyData skey;
4841 : HeapTuple htup;
4842 : List *result;
4843 : List *oldlist;
4844 2525926 : char replident = relation->rd_rel->relreplident;
4845 2525926 : Oid pkeyIndex = InvalidOid;
4846 2525926 : Oid candidateIndex = InvalidOid;
4847 2525926 : bool pkdeferrable = false;
4848 : MemoryContext oldcxt;
4849 :
4850 : /* Quick exit if we already computed the list. */
4851 2525926 : if (relation->rd_indexvalid)
4852 2109772 : return list_copy(relation->rd_indexlist);
4853 :
4854 : /*
4855 : * We build the list we intend to return (in the caller's context) while
4856 : * doing the scan. After successfully completing the scan, we copy that
4857 : * list into the relcache entry. This avoids cache-context memory leakage
4858 : * if we get some sort of error partway through.
4859 : */
4860 416154 : result = NIL;
4861 :
4862 : /* Prepare to scan pg_index for entries having indrelid = this rel. */
4863 416154 : ScanKeyInit(&skey,
4864 : Anum_pg_index_indrelid,
4865 : BTEqualStrategyNumber, F_OIDEQ,
4866 : ObjectIdGetDatum(RelationGetRelid(relation)));
4867 :
4868 416154 : indrel = table_open(IndexRelationId, AccessShareLock);
4869 416154 : indscan = systable_beginscan(indrel, IndexIndrelidIndexId, true,
4870 : NULL, 1, &skey);
4871 :
4872 1034350 : while (HeapTupleIsValid(htup = systable_getnext(indscan)))
4873 : {
4874 618196 : Form_pg_index index = (Form_pg_index) GETSTRUCT(htup);
4875 :
4876 : /*
4877 : * Ignore any indexes that are currently being dropped. This will
4878 : * prevent them from being searched, inserted into, or considered in
4879 : * HOT-safety decisions. It's unsafe to touch such an index at all
4880 : * since its catalog entries could disappear at any instant.
4881 : */
4882 618196 : if (!index->indislive)
4883 42 : continue;
4884 :
4885 : /* add index's OID to result list */
4886 618154 : result = lappend_oid(result, index->indexrelid);
4887 :
4888 : /*
4889 : * Non-unique or predicate indexes aren't interesting for either oid
4890 : * indexes or replication identity indexes, so don't check them.
4891 : * Deferred ones are not useful for replication identity either; but
4892 : * we do include them if they are PKs.
4893 : */
4894 618154 : if (!index->indisunique ||
4895 535502 : !heap_attisnull(htup, Anum_pg_index_indpred, NULL))
4896 82814 : continue;
4897 :
4898 : /*
4899 : * Remember primary key index, if any. For regular tables we do this
4900 : * only if the index is valid; but for partitioned tables, then we do
4901 : * it even if it's invalid.
4902 : *
4903 : * The reason for returning invalid primary keys for partitioned
4904 : * tables is that we need it to prevent drop of not-null constraints
4905 : * that may underlie such a primary key, which is only a problem for
4906 : * partitioned tables.
4907 : */
4908 535340 : if (index->indisprimary &&
4909 349346 : (index->indisvalid ||
4910 12 : relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE))
4911 : {
4912 349346 : pkeyIndex = index->indexrelid;
4913 349346 : pkdeferrable = !index->indimmediate;
4914 : }
4915 :
4916 535340 : if (!index->indimmediate)
4917 140 : continue;
4918 :
4919 535200 : if (!index->indisvalid)
4920 66 : continue;
4921 :
4922 : /* remember explicitly chosen replica index */
4923 535134 : if (index->indisreplident)
4924 552 : candidateIndex = index->indexrelid;
4925 : }
4926 :
4927 416154 : systable_endscan(indscan);
4928 :
4929 416154 : table_close(indrel, AccessShareLock);
4930 :
4931 : /* Sort the result list into OID order, per API spec. */
4932 416154 : list_sort(result, list_oid_cmp);
4933 :
4934 : /* Now save a copy of the completed list in the relcache entry. */
4935 416154 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
4936 416154 : oldlist = relation->rd_indexlist;
4937 416154 : relation->rd_indexlist = list_copy(result);
4938 416154 : relation->rd_pkindex = pkeyIndex;
4939 416154 : relation->rd_ispkdeferrable = pkdeferrable;
4940 416154 : if (replident == REPLICA_IDENTITY_DEFAULT && OidIsValid(pkeyIndex) && !pkdeferrable)
4941 26322 : relation->rd_replidindex = pkeyIndex;
4942 389832 : else if (replident == REPLICA_IDENTITY_INDEX && OidIsValid(candidateIndex))
4943 552 : relation->rd_replidindex = candidateIndex;
4944 : else
4945 389280 : relation->rd_replidindex = InvalidOid;
4946 416154 : relation->rd_indexvalid = true;
4947 416154 : MemoryContextSwitchTo(oldcxt);
4948 :
4949 : /* Don't leak the old list, if there is one */
4950 416154 : list_free(oldlist);
4951 :
4952 416154 : return result;
4953 : }
4954 :
4955 : /*
4956 : * RelationGetStatExtList
4957 : * get a list of OIDs of statistics objects on this relation
4958 : *
4959 : * The statistics list is created only if someone requests it, in a way
4960 : * similar to RelationGetIndexList(). We scan pg_statistic_ext to find
4961 : * relevant statistics, and add the list to the relcache entry so that we
4962 : * won't have to compute it again. Note that shared cache inval of a
4963 : * relcache entry will delete the old list and set rd_statvalid to 0,
4964 : * so that we must recompute the statistics list on next request. This
4965 : * handles creation or deletion of a statistics object.
4966 : *
4967 : * The returned list is guaranteed to be sorted in order by OID, although
4968 : * this is not currently needed.
4969 : *
4970 : * Since shared cache inval causes the relcache's copy of the list to go away,
4971 : * we return a copy of the list palloc'd in the caller's context. The caller
4972 : * may list_free() the returned list after scanning it. This is necessary
4973 : * since the caller will typically be doing syscache lookups on the relevant
4974 : * statistics, and syscache lookup could cause SI messages to be processed!
4975 : */
4976 : List *
4977 472956 : RelationGetStatExtList(Relation relation)
4978 : {
4979 : Relation indrel;
4980 : SysScanDesc indscan;
4981 : ScanKeyData skey;
4982 : HeapTuple htup;
4983 : List *result;
4984 : List *oldlist;
4985 : MemoryContext oldcxt;
4986 :
4987 : /* Quick exit if we already computed the list. */
4988 472956 : if (relation->rd_statvalid != 0)
4989 354100 : return list_copy(relation->rd_statlist);
4990 :
4991 : /*
4992 : * We build the list we intend to return (in the caller's context) while
4993 : * doing the scan. After successfully completing the scan, we copy that
4994 : * list into the relcache entry. This avoids cache-context memory leakage
4995 : * if we get some sort of error partway through.
4996 : */
4997 118856 : result = NIL;
4998 :
4999 : /*
5000 : * Prepare to scan pg_statistic_ext for entries having stxrelid = this
5001 : * rel.
5002 : */
5003 118856 : ScanKeyInit(&skey,
5004 : Anum_pg_statistic_ext_stxrelid,
5005 : BTEqualStrategyNumber, F_OIDEQ,
5006 : ObjectIdGetDatum(RelationGetRelid(relation)));
5007 :
5008 118856 : indrel = table_open(StatisticExtRelationId, AccessShareLock);
5009 118856 : indscan = systable_beginscan(indrel, StatisticExtRelidIndexId, true,
5010 : NULL, 1, &skey);
5011 :
5012 119266 : while (HeapTupleIsValid(htup = systable_getnext(indscan)))
5013 : {
5014 410 : Oid oid = ((Form_pg_statistic_ext) GETSTRUCT(htup))->oid;
5015 :
5016 410 : result = lappend_oid(result, oid);
5017 : }
5018 :
5019 118856 : systable_endscan(indscan);
5020 :
5021 118856 : table_close(indrel, AccessShareLock);
5022 :
5023 : /* Sort the result list into OID order, per API spec. */
5024 118856 : list_sort(result, list_oid_cmp);
5025 :
5026 : /* Now save a copy of the completed list in the relcache entry. */
5027 118856 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5028 118856 : oldlist = relation->rd_statlist;
5029 118856 : relation->rd_statlist = list_copy(result);
5030 :
5031 118856 : relation->rd_statvalid = true;
5032 118856 : MemoryContextSwitchTo(oldcxt);
5033 :
5034 : /* Don't leak the old list, if there is one */
5035 118856 : list_free(oldlist);
5036 :
5037 118856 : return result;
5038 : }
5039 :
5040 : /*
5041 : * RelationGetPrimaryKeyIndex -- get OID of the relation's primary key index
5042 : *
5043 : * Returns InvalidOid if there is no such index, or if the primary key is
5044 : * DEFERRABLE and the caller isn't OK with that.
5045 : */
5046 : Oid
5047 440 : RelationGetPrimaryKeyIndex(Relation relation, bool deferrable_ok)
5048 : {
5049 : List *ilist;
5050 :
5051 440 : if (!relation->rd_indexvalid)
5052 : {
5053 : /* RelationGetIndexList does the heavy lifting. */
5054 18 : ilist = RelationGetIndexList(relation);
5055 18 : list_free(ilist);
5056 : Assert(relation->rd_indexvalid);
5057 : }
5058 :
5059 440 : if (deferrable_ok)
5060 18 : return relation->rd_pkindex;
5061 422 : else if (relation->rd_ispkdeferrable)
5062 0 : return InvalidOid;
5063 422 : return relation->rd_pkindex;
5064 : }
5065 :
5066 : /*
5067 : * RelationGetReplicaIndex -- get OID of the relation's replica identity index
5068 : *
5069 : * Returns InvalidOid if there is no such index.
5070 : */
5071 : Oid
5072 318746 : RelationGetReplicaIndex(Relation relation)
5073 : {
5074 : List *ilist;
5075 :
5076 318746 : if (!relation->rd_indexvalid)
5077 : {
5078 : /* RelationGetIndexList does the heavy lifting. */
5079 5100 : ilist = RelationGetIndexList(relation);
5080 5100 : list_free(ilist);
5081 : Assert(relation->rd_indexvalid);
5082 : }
5083 :
5084 318746 : return relation->rd_replidindex;
5085 : }
5086 :
5087 : /*
5088 : * RelationGetIndexExpressions -- get the index expressions for an index
5089 : *
5090 : * We cache the result of transforming pg_index.indexprs into a node tree.
5091 : * If the rel is not an index or has no expressional columns, we return NIL.
5092 : * Otherwise, the returned tree is copied into the caller's memory context.
5093 : * (We don't want to return a pointer to the relcache copy, since it could
5094 : * disappear due to relcache invalidation.)
5095 : */
5096 : List *
5097 4348494 : RelationGetIndexExpressions(Relation relation)
5098 : {
5099 : List *result;
5100 : Datum exprsDatum;
5101 : bool isnull;
5102 : char *exprsString;
5103 : MemoryContext oldcxt;
5104 :
5105 : /* Quick exit if we already computed the result. */
5106 4348494 : if (relation->rd_indexprs)
5107 3458 : return copyObject(relation->rd_indexprs);
5108 :
5109 : /* Quick exit if there is nothing to do. */
5110 8690072 : if (relation->rd_indextuple == NULL ||
5111 4345036 : heap_attisnull(relation->rd_indextuple, Anum_pg_index_indexprs, NULL))
5112 4343336 : return NIL;
5113 :
5114 : /*
5115 : * We build the tree we intend to return in the caller's context. After
5116 : * successfully completing the work, we copy it into the relcache entry.
5117 : * This avoids problems if we get some sort of error partway through.
5118 : */
5119 1700 : exprsDatum = heap_getattr(relation->rd_indextuple,
5120 : Anum_pg_index_indexprs,
5121 : GetPgIndexDescriptor(),
5122 : &isnull);
5123 : Assert(!isnull);
5124 1700 : exprsString = TextDatumGetCString(exprsDatum);
5125 1700 : result = (List *) stringToNode(exprsString);
5126 1700 : pfree(exprsString);
5127 :
5128 : /*
5129 : * Run the expressions through eval_const_expressions. This is not just an
5130 : * optimization, but is necessary, because the planner will be comparing
5131 : * them to similarly-processed qual clauses, and may fail to detect valid
5132 : * matches without this. We must not use canonicalize_qual, however,
5133 : * since these aren't qual expressions.
5134 : */
5135 1700 : result = (List *) eval_const_expressions(NULL, (Node *) result);
5136 :
5137 : /* May as well fix opfuncids too */
5138 1700 : fix_opfuncids((Node *) result);
5139 :
5140 : /* Now save a copy of the completed tree in the relcache entry. */
5141 1700 : oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
5142 1700 : relation->rd_indexprs = copyObject(result);
5143 1700 : MemoryContextSwitchTo(oldcxt);
5144 :
5145 1700 : return result;
5146 : }
5147 :
5148 : /*
5149 : * RelationGetDummyIndexExpressions -- get dummy expressions for an index
5150 : *
5151 : * Return a list of dummy expressions (just Const nodes) with the same
5152 : * types/typmods/collations as the index's real expressions. This is
5153 : * useful in situations where we don't want to run any user-defined code.
5154 : */
5155 : List *
5156 258 : RelationGetDummyIndexExpressions(Relation relation)
5157 : {
5158 : List *result;
5159 : Datum exprsDatum;
5160 : bool isnull;
5161 : char *exprsString;
5162 : List *rawExprs;
5163 : ListCell *lc;
5164 :
5165 : /* Quick exit if there is nothing to do. */
5166 516 : if (relation->rd_indextuple == NULL ||
5167 258 : heap_attisnull(relation->rd_indextuple, Anum_pg_index_indexprs, NULL))
5168 204 : return NIL;
5169 :
5170 : /* Extract raw node tree(s) from index tuple. */
5171 54 : exprsDatum = heap_getattr(relation->rd_indextuple,
5172 : Anum_pg_index_indexprs,
5173 : GetPgIndexDescriptor(),
5174 : &isnull);
5175 : Assert(!isnull);
5176 54 : exprsString = TextDatumGetCString(exprsDatum);
5177 54 : rawExprs = (List *) stringToNode(exprsString);
5178 54 : pfree(exprsString);
5179 :
5180 : /* Construct null Consts; the typlen and typbyval are arbitrary. */
5181 54 : result = NIL;
5182 108 : foreach(lc, rawExprs)
5183 : {
5184 54 : Node *rawExpr = (Node *) lfirst(lc);
5185 :
5186 54 : result = lappend(result,
5187 54 : makeConst(exprType(rawExpr),
5188 : exprTypmod(rawExpr),
5189 : exprCollation(rawExpr),
5190 : 1,
5191 : (Datum) 0,
5192 : true,
5193 : true));
5194 : }
5195 :
5196 54 : return result;
5197 : }
5198 :
5199 : /*
5200 : * RelationGetIndexPredicate -- get the index predicate for an index
5201 : *
5202 : * We cache the result of transforming pg_index.indpred into an implicit-AND
5203 : * node tree (suitable for use in planning).
5204 : * If the rel is not an index or has no predicate, we return NIL.
5205 : * Otherwise, the returned tree is copied into the caller's memory context.
5206 : * (We don't want to return a pointer to the relcache copy, since it could
5207 : * disappear due to relcache invalidation.)
5208 : */
5209 : List *
5210 4348328 : RelationGetIndexPredicate(Relation relation)
5211 : {
5212 : List *result;
5213 : Datum predDatum;
5214 : bool isnull;
5215 : char *predString;
5216 : MemoryContext oldcxt;
5217 :
5218 : /* Quick exit if we already computed the result. */
5219 4348328 : if (relation->rd_indpred)
5220 1284 : return copyObject(relation->rd_indpred);
5221 :
5222 : /* Quick exit if there is nothing to do. */
5223 8694088 : if (relation->rd_indextuple == NULL ||
5224 4347044 : heap_attisnull(relation->rd_indextuple, Anum_pg_index_indpred, NULL))
5225 4346070 : return NIL;
5226 :
5227 : /*
5228 : * We build the tree we intend to return in the caller's context. After
5229 : * successfully completing the work, we copy it into the relcache entry.
5230 : * This avoids problems if we get some sort of error partway through.
5231 : */
5232 974 : predDatum = heap_getattr(relation->rd_indextuple,
5233 : Anum_pg_index_indpred,
5234 : GetPgIndexDescriptor(),
5235 : &isnull);
5236 : Assert(!isnull);
5237 974 : predString = TextDatumGetCString(predDatum);
5238 974 : result = (List *) stringToNode(predString);
5239 974 : pfree(predString);
5240 :
5241 : /*
5242 : * Run the expression through const-simplification and canonicalization.
5243 : * This is not just an optimization, but is necessary, because the planner
5244 : * will be comparing it to similarly-processed qual clauses, and may fail
5245 : * to detect valid matches without this. This must match the processing
5246 : * done to qual clauses in preprocess_expression()! (We can skip the
5247 : * stuff involving subqueries, however, since we don't allow any in index
5248 : * predicates.)
5249 : */
5250 974 : result = (List *) eval_const_expressions(NULL, (Node *) result);
5251 :
5252 974 : result = (List *) canonicalize_qual((Expr *) result, false);
5253 :
5254 : /* Also convert to implicit-AND format */
5255 974 : result = make_ands_implicit((Expr *) result);
5256 :
5257 : /* May as well fix opfuncids too */
5258 974 : fix_opfuncids((Node *) result);
5259 :
5260 : /* Now save a copy of the completed tree in the relcache entry. */
5261 974 : oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
5262 974 : relation->rd_indpred = copyObject(result);
5263 974 : MemoryContextSwitchTo(oldcxt);
5264 :
5265 974 : return result;
5266 : }
5267 :
5268 : /*
5269 : * RelationGetIndexAttrBitmap -- get a bitmap of index attribute numbers
5270 : *
5271 : * The result has a bit set for each attribute used anywhere in the index
5272 : * definitions of all the indexes on this relation. (This includes not only
5273 : * simple index keys, but attributes used in expressions and partial-index
5274 : * predicates.)
5275 : *
5276 : * Depending on attrKind, a bitmap covering attnums for certain columns is
5277 : * returned:
5278 : * INDEX_ATTR_BITMAP_KEY Columns in non-partial unique indexes not
5279 : * in expressions (i.e., usable for FKs)
5280 : * INDEX_ATTR_BITMAP_PRIMARY_KEY Columns in the table's primary key
5281 : * (beware: even if PK is deferrable!)
5282 : * INDEX_ATTR_BITMAP_IDENTITY_KEY Columns in the table's replica identity
5283 : * index (empty if FULL)
5284 : * INDEX_ATTR_BITMAP_HOT_BLOCKING Columns that block updates from being HOT
5285 : * INDEX_ATTR_BITMAP_SUMMARIZED Columns included in summarizing indexes
5286 : *
5287 : * Attribute numbers are offset by FirstLowInvalidHeapAttributeNumber so that
5288 : * we can include system attributes (e.g., OID) in the bitmap representation.
5289 : *
5290 : * Deferred indexes are considered for the primary key, but not for replica
5291 : * identity.
5292 : *
5293 : * Caller had better hold at least RowExclusiveLock on the target relation
5294 : * to ensure it is safe (deadlock-free) for us to take locks on the relation's
5295 : * indexes. Note that since the introduction of CREATE INDEX CONCURRENTLY,
5296 : * that lock level doesn't guarantee a stable set of indexes, so we have to
5297 : * be prepared to retry here in case of a change in the set of indexes.
5298 : *
5299 : * The returned result is palloc'd in the caller's memory context and should
5300 : * be bms_free'd when not needed anymore.
5301 : */
5302 : Bitmapset *
5303 2586498 : RelationGetIndexAttrBitmap(Relation relation, IndexAttrBitmapKind attrKind)
5304 : {
5305 : Bitmapset *uindexattrs; /* columns in unique indexes */
5306 : Bitmapset *pkindexattrs; /* columns in the primary index */
5307 : Bitmapset *idindexattrs; /* columns in the replica identity */
5308 : Bitmapset *hotblockingattrs; /* columns with HOT blocking indexes */
5309 : Bitmapset *summarizedattrs; /* columns with summarizing indexes */
5310 : List *indexoidlist;
5311 : List *newindexoidlist;
5312 : Oid relpkindex;
5313 : Oid relreplindex;
5314 : ListCell *l;
5315 : MemoryContext oldcxt;
5316 :
5317 : /* Quick exit if we already computed the result. */
5318 2586498 : if (relation->rd_attrsvalid)
5319 : {
5320 2181982 : switch (attrKind)
5321 : {
5322 529788 : case INDEX_ATTR_BITMAP_KEY:
5323 529788 : return bms_copy(relation->rd_keyattr);
5324 72 : case INDEX_ATTR_BITMAP_PRIMARY_KEY:
5325 72 : return bms_copy(relation->rd_pkattr);
5326 617542 : case INDEX_ATTR_BITMAP_IDENTITY_KEY:
5327 617542 : return bms_copy(relation->rd_idattr);
5328 511312 : case INDEX_ATTR_BITMAP_HOT_BLOCKING:
5329 511312 : return bms_copy(relation->rd_hotblockingattr);
5330 523268 : case INDEX_ATTR_BITMAP_SUMMARIZED:
5331 523268 : return bms_copy(relation->rd_summarizedattr);
5332 0 : default:
5333 0 : elog(ERROR, "unknown attrKind %u", attrKind);
5334 : }
5335 : }
5336 :
5337 : /* Fast path if definitely no indexes */
5338 404516 : if (!RelationGetForm(relation)->relhasindex)
5339 389152 : return NULL;
5340 :
5341 : /*
5342 : * Get cached list of index OIDs. If we have to start over, we do so here.
5343 : */
5344 15364 : restart:
5345 15364 : indexoidlist = RelationGetIndexList(relation);
5346 :
5347 : /* Fall out if no indexes (but relhasindex was set) */
5348 15364 : if (indexoidlist == NIL)
5349 1134 : return NULL;
5350 :
5351 : /*
5352 : * Copy the rd_pkindex and rd_replidindex values computed by
5353 : * RelationGetIndexList before proceeding. This is needed because a
5354 : * relcache flush could occur inside index_open below, resetting the
5355 : * fields managed by RelationGetIndexList. We need to do the work with
5356 : * stable values of these fields.
5357 : */
5358 14230 : relpkindex = relation->rd_pkindex;
5359 14230 : relreplindex = relation->rd_replidindex;
5360 :
5361 : /*
5362 : * For each index, add referenced attributes to indexattrs.
5363 : *
5364 : * Note: we consider all indexes returned by RelationGetIndexList, even if
5365 : * they are not indisready or indisvalid. This is important because an
5366 : * index for which CREATE INDEX CONCURRENTLY has just started must be
5367 : * included in HOT-safety decisions (see README.HOT). If a DROP INDEX
5368 : * CONCURRENTLY is far enough along that we should ignore the index, it
5369 : * won't be returned at all by RelationGetIndexList.
5370 : */
5371 14230 : uindexattrs = NULL;
5372 14230 : pkindexattrs = NULL;
5373 14230 : idindexattrs = NULL;
5374 14230 : hotblockingattrs = NULL;
5375 14230 : summarizedattrs = NULL;
5376 40268 : foreach(l, indexoidlist)
5377 : {
5378 26038 : Oid indexOid = lfirst_oid(l);
5379 : Relation indexDesc;
5380 : Datum datum;
5381 : bool isnull;
5382 : Node *indexExpressions;
5383 : Node *indexPredicate;
5384 : int i;
5385 : bool isKey; /* candidate key */
5386 : bool isPK; /* primary key */
5387 : bool isIDKey; /* replica identity index */
5388 : Bitmapset **attrs;
5389 :
5390 26038 : indexDesc = index_open(indexOid, AccessShareLock);
5391 :
5392 : /*
5393 : * Extract index expressions and index predicate. Note: Don't use
5394 : * RelationGetIndexExpressions()/RelationGetIndexPredicate(), because
5395 : * those might run constant expressions evaluation, which needs a
5396 : * snapshot, which we might not have here. (Also, it's probably more
5397 : * sound to collect the bitmaps before any transformations that might
5398 : * eliminate columns, but the practical impact of this is limited.)
5399 : */
5400 :
5401 26038 : datum = heap_getattr(indexDesc->rd_indextuple, Anum_pg_index_indexprs,
5402 : GetPgIndexDescriptor(), &isnull);
5403 26038 : if (!isnull)
5404 38 : indexExpressions = stringToNode(TextDatumGetCString(datum));
5405 : else
5406 26000 : indexExpressions = NULL;
5407 :
5408 26038 : datum = heap_getattr(indexDesc->rd_indextuple, Anum_pg_index_indpred,
5409 : GetPgIndexDescriptor(), &isnull);
5410 26038 : if (!isnull)
5411 102 : indexPredicate = stringToNode(TextDatumGetCString(datum));
5412 : else
5413 25936 : indexPredicate = NULL;
5414 :
5415 : /* Can this index be referenced by a foreign key? */
5416 20604 : isKey = indexDesc->rd_index->indisunique &&
5417 46642 : indexExpressions == NULL &&
5418 : indexPredicate == NULL;
5419 :
5420 : /* Is this a primary key? */
5421 26038 : isPK = (indexOid == relpkindex);
5422 :
5423 : /* Is this index the configured (or default) replica identity? */
5424 26038 : isIDKey = (indexOid == relreplindex);
5425 :
5426 : /*
5427 : * If the index is summarizing, it doesn't block HOT updates, but we
5428 : * may still need to update it (if the attributes were modified). So
5429 : * decide which bitmap we'll update in the following loop.
5430 : */
5431 26038 : if (indexDesc->rd_indam->amsummarizing)
5432 78 : attrs = &summarizedattrs;
5433 : else
5434 25960 : attrs = &hotblockingattrs;
5435 :
5436 : /* Collect simple attribute references */
5437 66898 : for (i = 0; i < indexDesc->rd_index->indnatts; i++)
5438 : {
5439 40860 : int attrnum = indexDesc->rd_index->indkey.values[i];
5440 :
5441 : /*
5442 : * Since we have covering indexes with non-key columns, we must
5443 : * handle them accurately here. non-key columns must be added into
5444 : * hotblockingattrs or summarizedattrs, since they are in index,
5445 : * and update shouldn't miss them.
5446 : *
5447 : * Summarizing indexes do not block HOT, but do need to be updated
5448 : * when the column value changes, thus require a separate
5449 : * attribute bitmapset.
5450 : *
5451 : * Obviously, non-key columns couldn't be referenced by foreign
5452 : * key or identity key. Hence we do not include them into
5453 : * uindexattrs, pkindexattrs and idindexattrs bitmaps.
5454 : */
5455 40860 : if (attrnum != 0)
5456 : {
5457 40822 : *attrs = bms_add_member(*attrs,
5458 : attrnum - FirstLowInvalidHeapAttributeNumber);
5459 :
5460 40822 : if (isKey && i < indexDesc->rd_index->indnkeyatts)
5461 30742 : uindexattrs = bms_add_member(uindexattrs,
5462 : attrnum - FirstLowInvalidHeapAttributeNumber);
5463 :
5464 40822 : if (isPK && i < indexDesc->rd_index->indnkeyatts)
5465 15532 : pkindexattrs = bms_add_member(pkindexattrs,
5466 : attrnum - FirstLowInvalidHeapAttributeNumber);
5467 :
5468 40822 : if (isIDKey && i < indexDesc->rd_index->indnkeyatts)
5469 4248 : idindexattrs = bms_add_member(idindexattrs,
5470 : attrnum - FirstLowInvalidHeapAttributeNumber);
5471 : }
5472 : }
5473 :
5474 : /* Collect all attributes used in expressions, too */
5475 26038 : pull_varattnos(indexExpressions, 1, attrs);
5476 :
5477 : /* Collect all attributes in the index predicate, too */
5478 26038 : pull_varattnos(indexPredicate, 1, attrs);
5479 :
5480 26038 : index_close(indexDesc, AccessShareLock);
5481 : }
5482 :
5483 : /*
5484 : * During one of the index_opens in the above loop, we might have received
5485 : * a relcache flush event on this relcache entry, which might have been
5486 : * signaling a change in the rel's index list. If so, we'd better start
5487 : * over to ensure we deliver up-to-date attribute bitmaps.
5488 : */
5489 14230 : newindexoidlist = RelationGetIndexList(relation);
5490 14230 : if (equal(indexoidlist, newindexoidlist) &&
5491 14230 : relpkindex == relation->rd_pkindex &&
5492 14230 : relreplindex == relation->rd_replidindex)
5493 : {
5494 : /* Still the same index set, so proceed */
5495 14230 : list_free(newindexoidlist);
5496 14230 : list_free(indexoidlist);
5497 : }
5498 : else
5499 : {
5500 : /* Gotta do it over ... might as well not leak memory */
5501 0 : list_free(newindexoidlist);
5502 0 : list_free(indexoidlist);
5503 0 : bms_free(uindexattrs);
5504 0 : bms_free(pkindexattrs);
5505 0 : bms_free(idindexattrs);
5506 0 : bms_free(hotblockingattrs);
5507 0 : bms_free(summarizedattrs);
5508 :
5509 0 : goto restart;
5510 : }
5511 :
5512 : /* Don't leak the old values of these bitmaps, if any */
5513 14230 : relation->rd_attrsvalid = false;
5514 14230 : bms_free(relation->rd_keyattr);
5515 14230 : relation->rd_keyattr = NULL;
5516 14230 : bms_free(relation->rd_pkattr);
5517 14230 : relation->rd_pkattr = NULL;
5518 14230 : bms_free(relation->rd_idattr);
5519 14230 : relation->rd_idattr = NULL;
5520 14230 : bms_free(relation->rd_hotblockingattr);
5521 14230 : relation->rd_hotblockingattr = NULL;
5522 14230 : bms_free(relation->rd_summarizedattr);
5523 14230 : relation->rd_summarizedattr = NULL;
5524 :
5525 : /*
5526 : * Now save copies of the bitmaps in the relcache entry. We intentionally
5527 : * set rd_attrsvalid last, because that's the one that signals validity of
5528 : * the values; if we run out of memory before making that copy, we won't
5529 : * leave the relcache entry looking like the other ones are valid but
5530 : * empty.
5531 : */
5532 14230 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5533 14230 : relation->rd_keyattr = bms_copy(uindexattrs);
5534 14230 : relation->rd_pkattr = bms_copy(pkindexattrs);
5535 14230 : relation->rd_idattr = bms_copy(idindexattrs);
5536 14230 : relation->rd_hotblockingattr = bms_copy(hotblockingattrs);
5537 14230 : relation->rd_summarizedattr = bms_copy(summarizedattrs);
5538 14230 : relation->rd_attrsvalid = true;
5539 14230 : MemoryContextSwitchTo(oldcxt);
5540 :
5541 : /* We return our original working copy for caller to play with */
5542 14230 : switch (attrKind)
5543 : {
5544 958 : case INDEX_ATTR_BITMAP_KEY:
5545 958 : return uindexattrs;
5546 54 : case INDEX_ATTR_BITMAP_PRIMARY_KEY:
5547 54 : return pkindexattrs;
5548 1262 : case INDEX_ATTR_BITMAP_IDENTITY_KEY:
5549 1262 : return idindexattrs;
5550 11956 : case INDEX_ATTR_BITMAP_HOT_BLOCKING:
5551 11956 : return hotblockingattrs;
5552 0 : case INDEX_ATTR_BITMAP_SUMMARIZED:
5553 0 : return summarizedattrs;
5554 0 : default:
5555 0 : elog(ERROR, "unknown attrKind %u", attrKind);
5556 : return NULL;
5557 : }
5558 : }
5559 :
5560 : /*
5561 : * RelationGetIdentityKeyBitmap -- get a bitmap of replica identity attribute
5562 : * numbers
5563 : *
5564 : * A bitmap of index attribute numbers for the configured replica identity
5565 : * index is returned.
5566 : *
5567 : * See also comments of RelationGetIndexAttrBitmap().
5568 : *
5569 : * This is a special purpose function used during logical replication. Here,
5570 : * unlike RelationGetIndexAttrBitmap(), we don't acquire a lock on the required
5571 : * index as we build the cache entry using a historic snapshot and all the
5572 : * later changes are absorbed while decoding WAL. Due to this reason, we don't
5573 : * need to retry here in case of a change in the set of indexes.
5574 : */
5575 : Bitmapset *
5576 610 : RelationGetIdentityKeyBitmap(Relation relation)
5577 : {
5578 610 : Bitmapset *idindexattrs = NULL; /* columns in the replica identity */
5579 : Relation indexDesc;
5580 : int i;
5581 : Oid replidindex;
5582 : MemoryContext oldcxt;
5583 :
5584 : /* Quick exit if we already computed the result */
5585 610 : if (relation->rd_idattr != NULL)
5586 94 : return bms_copy(relation->rd_idattr);
5587 :
5588 : /* Fast path if definitely no indexes */
5589 516 : if (!RelationGetForm(relation)->relhasindex)
5590 128 : return NULL;
5591 :
5592 : /* Historic snapshot must be set. */
5593 : Assert(HistoricSnapshotActive());
5594 :
5595 388 : replidindex = RelationGetReplicaIndex(relation);
5596 :
5597 : /* Fall out if there is no replica identity index */
5598 388 : if (!OidIsValid(replidindex))
5599 10 : return NULL;
5600 :
5601 : /* Look up the description for the replica identity index */
5602 378 : indexDesc = RelationIdGetRelation(replidindex);
5603 :
5604 378 : if (!RelationIsValid(indexDesc))
5605 0 : elog(ERROR, "could not open relation with OID %u",
5606 : relation->rd_replidindex);
5607 :
5608 : /* Add referenced attributes to idindexattrs */
5609 770 : for (i = 0; i < indexDesc->rd_index->indnatts; i++)
5610 : {
5611 392 : int attrnum = indexDesc->rd_index->indkey.values[i];
5612 :
5613 : /*
5614 : * We don't include non-key columns into idindexattrs bitmaps. See
5615 : * RelationGetIndexAttrBitmap.
5616 : */
5617 392 : if (attrnum != 0)
5618 : {
5619 392 : if (i < indexDesc->rd_index->indnkeyatts)
5620 390 : idindexattrs = bms_add_member(idindexattrs,
5621 : attrnum - FirstLowInvalidHeapAttributeNumber);
5622 : }
5623 : }
5624 :
5625 378 : RelationClose(indexDesc);
5626 :
5627 : /* Don't leak the old values of these bitmaps, if any */
5628 378 : bms_free(relation->rd_idattr);
5629 378 : relation->rd_idattr = NULL;
5630 :
5631 : /* Now save copy of the bitmap in the relcache entry */
5632 378 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5633 378 : relation->rd_idattr = bms_copy(idindexattrs);
5634 378 : MemoryContextSwitchTo(oldcxt);
5635 :
5636 : /* We return our original working copy for caller to play with */
5637 378 : return idindexattrs;
5638 : }
5639 :
5640 : /*
5641 : * RelationGetExclusionInfo -- get info about index's exclusion constraint
5642 : *
5643 : * This should be called only for an index that is known to have an associated
5644 : * exclusion constraint or primary key/unique constraint using WITHOUT
5645 : * OVERLAPS.
5646 :
5647 : * It returns arrays (palloc'd in caller's context) of the exclusion operator
5648 : * OIDs, their underlying functions' OIDs, and their strategy numbers in the
5649 : * index's opclasses. We cache all this information since it requires a fair
5650 : * amount of work to get.
5651 : */
5652 : void
5653 2184 : RelationGetExclusionInfo(Relation indexRelation,
5654 : Oid **operators,
5655 : Oid **procs,
5656 : uint16 **strategies)
5657 : {
5658 : int indnkeyatts;
5659 : Oid *ops;
5660 : Oid *funcs;
5661 : uint16 *strats;
5662 : Relation conrel;
5663 : SysScanDesc conscan;
5664 : ScanKeyData skey[1];
5665 : HeapTuple htup;
5666 : bool found;
5667 : MemoryContext oldcxt;
5668 : int i;
5669 :
5670 2184 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(indexRelation);
5671 :
5672 : /* Allocate result space in caller context */
5673 2184 : *operators = ops = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5674 2184 : *procs = funcs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5675 2184 : *strategies = strats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
5676 :
5677 : /* Quick exit if we have the data cached already */
5678 2184 : if (indexRelation->rd_exclstrats != NULL)
5679 : {
5680 1552 : memcpy(ops, indexRelation->rd_exclops, sizeof(Oid) * indnkeyatts);
5681 1552 : memcpy(funcs, indexRelation->rd_exclprocs, sizeof(Oid) * indnkeyatts);
5682 1552 : memcpy(strats, indexRelation->rd_exclstrats, sizeof(uint16) * indnkeyatts);
5683 1552 : return;
5684 : }
5685 :
5686 : /*
5687 : * Search pg_constraint for the constraint associated with the index. To
5688 : * make this not too painfully slow, we use the index on conrelid; that
5689 : * will hold the parent relation's OID not the index's own OID.
5690 : *
5691 : * Note: if we wanted to rely on the constraint name matching the index's
5692 : * name, we could just do a direct lookup using pg_constraint's unique
5693 : * index. For the moment it doesn't seem worth requiring that.
5694 : */
5695 632 : ScanKeyInit(&skey[0],
5696 : Anum_pg_constraint_conrelid,
5697 : BTEqualStrategyNumber, F_OIDEQ,
5698 632 : ObjectIdGetDatum(indexRelation->rd_index->indrelid));
5699 :
5700 632 : conrel = table_open(ConstraintRelationId, AccessShareLock);
5701 632 : conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
5702 : NULL, 1, skey);
5703 632 : found = false;
5704 :
5705 2504 : while (HeapTupleIsValid(htup = systable_getnext(conscan)))
5706 : {
5707 1872 : Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(htup);
5708 : Datum val;
5709 : bool isnull;
5710 : ArrayType *arr;
5711 : int nelem;
5712 :
5713 : /* We want the exclusion constraint owning the index */
5714 1872 : if ((conform->contype != CONSTRAINT_EXCLUSION &&
5715 1642 : !(conform->conperiod && (conform->contype == CONSTRAINT_PRIMARY
5716 230 : || conform->contype == CONSTRAINT_UNIQUE))) ||
5717 764 : conform->conindid != RelationGetRelid(indexRelation))
5718 1240 : continue;
5719 :
5720 : /* There should be only one */
5721 632 : if (found)
5722 0 : elog(ERROR, "unexpected exclusion constraint record found for rel %s",
5723 : RelationGetRelationName(indexRelation));
5724 632 : found = true;
5725 :
5726 : /* Extract the operator OIDS from conexclop */
5727 632 : val = fastgetattr(htup,
5728 : Anum_pg_constraint_conexclop,
5729 : conrel->rd_att, &isnull);
5730 632 : if (isnull)
5731 0 : elog(ERROR, "null conexclop for rel %s",
5732 : RelationGetRelationName(indexRelation));
5733 :
5734 632 : arr = DatumGetArrayTypeP(val); /* ensure not toasted */
5735 632 : nelem = ARR_DIMS(arr)[0];
5736 632 : if (ARR_NDIM(arr) != 1 ||
5737 632 : nelem != indnkeyatts ||
5738 632 : ARR_HASNULL(arr) ||
5739 632 : ARR_ELEMTYPE(arr) != OIDOID)
5740 0 : elog(ERROR, "conexclop is not a 1-D Oid array");
5741 :
5742 632 : memcpy(ops, ARR_DATA_PTR(arr), sizeof(Oid) * indnkeyatts);
5743 : }
5744 :
5745 632 : systable_endscan(conscan);
5746 632 : table_close(conrel, AccessShareLock);
5747 :
5748 632 : if (!found)
5749 0 : elog(ERROR, "exclusion constraint record missing for rel %s",
5750 : RelationGetRelationName(indexRelation));
5751 :
5752 : /* We need the func OIDs and strategy numbers too */
5753 1810 : for (i = 0; i < indnkeyatts; i++)
5754 : {
5755 1178 : funcs[i] = get_opcode(ops[i]);
5756 2356 : strats[i] = get_op_opfamily_strategy(ops[i],
5757 1178 : indexRelation->rd_opfamily[i]);
5758 : /* shouldn't fail, since it was checked at index creation */
5759 1178 : if (strats[i] == InvalidStrategy)
5760 0 : elog(ERROR, "could not find strategy for operator %u in family %u",
5761 : ops[i], indexRelation->rd_opfamily[i]);
5762 : }
5763 :
5764 : /* Save a copy of the results in the relcache entry. */
5765 632 : oldcxt = MemoryContextSwitchTo(indexRelation->rd_indexcxt);
5766 632 : indexRelation->rd_exclops = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5767 632 : indexRelation->rd_exclprocs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
5768 632 : indexRelation->rd_exclstrats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
5769 632 : memcpy(indexRelation->rd_exclops, ops, sizeof(Oid) * indnkeyatts);
5770 632 : memcpy(indexRelation->rd_exclprocs, funcs, sizeof(Oid) * indnkeyatts);
5771 632 : memcpy(indexRelation->rd_exclstrats, strats, sizeof(uint16) * indnkeyatts);
5772 632 : MemoryContextSwitchTo(oldcxt);
5773 : }
5774 :
5775 : /*
5776 : * Get the publication information for the given relation.
5777 : *
5778 : * Traverse all the publications which the relation is in to get the
5779 : * publication actions and validate:
5780 : * 1. The row filter expressions for such publications if any. We consider the
5781 : * row filter expression as invalid if it references any column which is not
5782 : * part of REPLICA IDENTITY.
5783 : * 2. The column list for such publication if any. We consider the column list
5784 : * invalid if REPLICA IDENTITY contains any column that is not part of it.
5785 : * 3. The generated columns of the relation for such publications. We consider
5786 : * any reference of an unpublished generated column in REPLICA IDENTITY as
5787 : * invalid.
5788 : *
5789 : * To avoid fetching the publication information repeatedly, we cache the
5790 : * publication actions, row filter validation information, column list
5791 : * validation information, and generated column validation information.
5792 : */
5793 : void
5794 172532 : RelationBuildPublicationDesc(Relation relation, PublicationDesc *pubdesc)
5795 : {
5796 : List *puboids;
5797 : ListCell *lc;
5798 : MemoryContext oldcxt;
5799 : Oid schemaid;
5800 172532 : List *ancestors = NIL;
5801 172532 : Oid relid = RelationGetRelid(relation);
5802 :
5803 : /*
5804 : * If not publishable, it publishes no actions. (pgoutput_change() will
5805 : * ignore it.)
5806 : */
5807 172532 : if (!is_publishable_relation(relation))
5808 : {
5809 5782 : memset(pubdesc, 0, sizeof(PublicationDesc));
5810 5782 : pubdesc->rf_valid_for_update = true;
5811 5782 : pubdesc->rf_valid_for_delete = true;
5812 5782 : pubdesc->cols_valid_for_update = true;
5813 5782 : pubdesc->cols_valid_for_delete = true;
5814 5782 : pubdesc->gencols_valid_for_update = true;
5815 5782 : pubdesc->gencols_valid_for_delete = true;
5816 5782 : return;
5817 : }
5818 :
5819 166750 : if (relation->rd_pubdesc)
5820 : {
5821 158690 : memcpy(pubdesc, relation->rd_pubdesc, sizeof(PublicationDesc));
5822 158690 : return;
5823 : }
5824 :
5825 8060 : memset(pubdesc, 0, sizeof(PublicationDesc));
5826 8060 : pubdesc->rf_valid_for_update = true;
5827 8060 : pubdesc->rf_valid_for_delete = true;
5828 8060 : pubdesc->cols_valid_for_update = true;
5829 8060 : pubdesc->cols_valid_for_delete = true;
5830 8060 : pubdesc->gencols_valid_for_update = true;
5831 8060 : pubdesc->gencols_valid_for_delete = true;
5832 :
5833 : /* Fetch the publication membership info. */
5834 8060 : puboids = GetRelationPublications(relid);
5835 8060 : schemaid = RelationGetNamespace(relation);
5836 8060 : puboids = list_concat_unique_oid(puboids, GetSchemaPublications(schemaid));
5837 :
5838 8060 : if (relation->rd_rel->relispartition)
5839 : {
5840 : /* Add publications that the ancestors are in too. */
5841 1900 : ancestors = get_partition_ancestors(relid);
5842 :
5843 4432 : foreach(lc, ancestors)
5844 : {
5845 2532 : Oid ancestor = lfirst_oid(lc);
5846 :
5847 2532 : puboids = list_concat_unique_oid(puboids,
5848 2532 : GetRelationPublications(ancestor));
5849 2532 : schemaid = get_rel_namespace(ancestor);
5850 2532 : puboids = list_concat_unique_oid(puboids,
5851 2532 : GetSchemaPublications(schemaid));
5852 : }
5853 : }
5854 8060 : puboids = list_concat_unique_oid(puboids, GetAllTablesPublications());
5855 :
5856 8776 : foreach(lc, puboids)
5857 : {
5858 908 : Oid pubid = lfirst_oid(lc);
5859 : HeapTuple tup;
5860 : Form_pg_publication pubform;
5861 : bool invalid_column_list;
5862 : bool invalid_gen_col;
5863 :
5864 908 : tup = SearchSysCache1(PUBLICATIONOID, ObjectIdGetDatum(pubid));
5865 :
5866 908 : if (!HeapTupleIsValid(tup))
5867 0 : elog(ERROR, "cache lookup failed for publication %u", pubid);
5868 :
5869 908 : pubform = (Form_pg_publication) GETSTRUCT(tup);
5870 :
5871 908 : pubdesc->pubactions.pubinsert |= pubform->pubinsert;
5872 908 : pubdesc->pubactions.pubupdate |= pubform->pubupdate;
5873 908 : pubdesc->pubactions.pubdelete |= pubform->pubdelete;
5874 908 : pubdesc->pubactions.pubtruncate |= pubform->pubtruncate;
5875 :
5876 : /*
5877 : * Check if all columns referenced in the filter expression are part
5878 : * of the REPLICA IDENTITY index or not.
5879 : *
5880 : * If the publication is FOR ALL TABLES then it means the table has no
5881 : * row filters and we can skip the validation.
5882 : */
5883 908 : if (!pubform->puballtables &&
5884 1428 : (pubform->pubupdate || pubform->pubdelete) &&
5885 712 : pub_rf_contains_invalid_column(pubid, relation, ancestors,
5886 712 : pubform->pubviaroot))
5887 : {
5888 60 : if (pubform->pubupdate)
5889 60 : pubdesc->rf_valid_for_update = false;
5890 60 : if (pubform->pubdelete)
5891 60 : pubdesc->rf_valid_for_delete = false;
5892 : }
5893 :
5894 : /*
5895 : * Check if all columns are part of the REPLICA IDENTITY index or not.
5896 : *
5897 : * Check if all generated columns included in the REPLICA IDENTITY are
5898 : * published.
5899 : */
5900 1812 : if ((pubform->pubupdate || pubform->pubdelete) &&
5901 904 : pub_contains_invalid_column(pubid, relation, ancestors,
5902 904 : pubform->pubviaroot,
5903 904 : pubform->pubgencols,
5904 : &invalid_column_list,
5905 : &invalid_gen_col))
5906 : {
5907 132 : if (pubform->pubupdate)
5908 : {
5909 132 : pubdesc->cols_valid_for_update = !invalid_column_list;
5910 132 : pubdesc->gencols_valid_for_update = !invalid_gen_col;
5911 : }
5912 :
5913 132 : if (pubform->pubdelete)
5914 : {
5915 132 : pubdesc->cols_valid_for_delete = !invalid_column_list;
5916 132 : pubdesc->gencols_valid_for_delete = !invalid_gen_col;
5917 : }
5918 : }
5919 :
5920 908 : ReleaseSysCache(tup);
5921 :
5922 : /*
5923 : * If we know everything is replicated and the row filter is invalid
5924 : * for update and delete, there is no point to check for other
5925 : * publications.
5926 : */
5927 908 : if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
5928 902 : pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
5929 890 : !pubdesc->rf_valid_for_update && !pubdesc->rf_valid_for_delete)
5930 192 : break;
5931 :
5932 : /*
5933 : * If we know everything is replicated and the column list is invalid
5934 : * for update and delete, there is no point to check for other
5935 : * publications.
5936 : */
5937 848 : if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
5938 842 : pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
5939 830 : !pubdesc->cols_valid_for_update && !pubdesc->cols_valid_for_delete)
5940 108 : break;
5941 :
5942 : /*
5943 : * If we know everything is replicated and replica identity has an
5944 : * unpublished generated column, there is no point to check for other
5945 : * publications.
5946 : */
5947 740 : if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
5948 734 : pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
5949 722 : !pubdesc->gencols_valid_for_update &&
5950 24 : !pubdesc->gencols_valid_for_delete)
5951 24 : break;
5952 : }
5953 :
5954 8060 : if (relation->rd_pubdesc)
5955 : {
5956 0 : pfree(relation->rd_pubdesc);
5957 0 : relation->rd_pubdesc = NULL;
5958 : }
5959 :
5960 : /* Now save copy of the descriptor in the relcache entry. */
5961 8060 : oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
5962 8060 : relation->rd_pubdesc = palloc(sizeof(PublicationDesc));
5963 8060 : memcpy(relation->rd_pubdesc, pubdesc, sizeof(PublicationDesc));
5964 8060 : MemoryContextSwitchTo(oldcxt);
5965 : }
5966 :
5967 : static bytea **
5968 1493128 : CopyIndexAttOptions(bytea **srcopts, int natts)
5969 : {
5970 1493128 : bytea **opts = palloc(sizeof(*opts) * natts);
5971 :
5972 4192464 : for (int i = 0; i < natts; i++)
5973 : {
5974 2699336 : bytea *opt = srcopts[i];
5975 :
5976 2791434 : opts[i] = !opt ? NULL : (bytea *)
5977 92098 : DatumGetPointer(datumCopy(PointerGetDatum(opt), false, -1));
5978 : }
5979 :
5980 1493128 : return opts;
5981 : }
5982 :
5983 : /*
5984 : * RelationGetIndexAttOptions
5985 : * get AM/opclass-specific options for an index parsed into a binary form
5986 : */
5987 : bytea **
5988 2704404 : RelationGetIndexAttOptions(Relation relation, bool copy)
5989 : {
5990 : MemoryContext oldcxt;
5991 2704404 : bytea **opts = relation->rd_opcoptions;
5992 2704404 : Oid relid = RelationGetRelid(relation);
5993 2704404 : int natts = RelationGetNumberOfAttributes(relation); /* XXX
5994 : * IndexRelationGetNumberOfKeyAttributes */
5995 : int i;
5996 :
5997 : /* Try to copy cached options. */
5998 2704404 : if (opts)
5999 1952004 : return copy ? CopyIndexAttOptions(opts, natts) : opts;
6000 :
6001 : /* Get and parse opclass options. */
6002 752400 : opts = palloc0(sizeof(*opts) * natts);
6003 :
6004 2050614 : for (i = 0; i < natts; i++)
6005 : {
6006 1298220 : if (criticalRelcachesBuilt && relid != AttributeRelidNumIndexId)
6007 : {
6008 1226118 : Datum attoptions = get_attoptions(relid, i + 1);
6009 :
6010 1226118 : opts[i] = index_opclass_options(relation, i + 1, attoptions, false);
6011 :
6012 1226112 : if (attoptions != (Datum) 0)
6013 292 : pfree(DatumGetPointer(attoptions));
6014 : }
6015 : }
6016 :
6017 : /* Copy parsed options to the cache. */
6018 752394 : oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
6019 752394 : relation->rd_opcoptions = CopyIndexAttOptions(opts, natts);
6020 752394 : MemoryContextSwitchTo(oldcxt);
6021 :
6022 752394 : if (copy)
6023 0 : return opts;
6024 :
6025 2050608 : for (i = 0; i < natts; i++)
6026 : {
6027 1298214 : if (opts[i])
6028 2102 : pfree(opts[i]);
6029 : }
6030 :
6031 752394 : pfree(opts);
6032 :
6033 752394 : return relation->rd_opcoptions;
6034 : }
6035 :
6036 : /*
6037 : * Routines to support ereport() reports of relation-related errors
6038 : *
6039 : * These could have been put into elog.c, but it seems like a module layering
6040 : * violation to have elog.c calling relcache or syscache stuff --- and we
6041 : * definitely don't want elog.h including rel.h. So we put them here.
6042 : */
6043 :
6044 : /*
6045 : * errtable --- stores schema_name and table_name of a table
6046 : * within the current errordata.
6047 : */
6048 : int
6049 3548 : errtable(Relation rel)
6050 : {
6051 3548 : err_generic_string(PG_DIAG_SCHEMA_NAME,
6052 3548 : get_namespace_name(RelationGetNamespace(rel)));
6053 3548 : err_generic_string(PG_DIAG_TABLE_NAME, RelationGetRelationName(rel));
6054 :
6055 3548 : return 0; /* return value does not matter */
6056 : }
6057 :
6058 : /*
6059 : * errtablecol --- stores schema_name, table_name and column_name
6060 : * of a table column within the current errordata.
6061 : *
6062 : * The column is specified by attribute number --- for most callers, this is
6063 : * easier and less error-prone than getting the column name for themselves.
6064 : */
6065 : int
6066 548 : errtablecol(Relation rel, int attnum)
6067 : {
6068 548 : TupleDesc reldesc = RelationGetDescr(rel);
6069 : const char *colname;
6070 :
6071 : /* Use reldesc if it's a user attribute, else consult the catalogs */
6072 548 : if (attnum > 0 && attnum <= reldesc->natts)
6073 548 : colname = NameStr(TupleDescAttr(reldesc, attnum - 1)->attname);
6074 : else
6075 0 : colname = get_attname(RelationGetRelid(rel), attnum, false);
6076 :
6077 548 : return errtablecolname(rel, colname);
6078 : }
6079 :
6080 : /*
6081 : * errtablecolname --- stores schema_name, table_name and column_name
6082 : * of a table column within the current errordata, where the column name is
6083 : * given directly rather than extracted from the relation's catalog data.
6084 : *
6085 : * Don't use this directly unless errtablecol() is inconvenient for some
6086 : * reason. This might possibly be needed during intermediate states in ALTER
6087 : * TABLE, for instance.
6088 : */
6089 : int
6090 548 : errtablecolname(Relation rel, const char *colname)
6091 : {
6092 548 : errtable(rel);
6093 548 : err_generic_string(PG_DIAG_COLUMN_NAME, colname);
6094 :
6095 548 : return 0; /* return value does not matter */
6096 : }
6097 :
6098 : /*
6099 : * errtableconstraint --- stores schema_name, table_name and constraint_name
6100 : * of a table-related constraint within the current errordata.
6101 : */
6102 : int
6103 2492 : errtableconstraint(Relation rel, const char *conname)
6104 : {
6105 2492 : errtable(rel);
6106 2492 : err_generic_string(PG_DIAG_CONSTRAINT_NAME, conname);
6107 :
6108 2492 : return 0; /* return value does not matter */
6109 : }
6110 :
6111 :
6112 : /*
6113 : * load_relcache_init_file, write_relcache_init_file
6114 : *
6115 : * In late 1992, we started regularly having databases with more than
6116 : * a thousand classes in them. With this number of classes, it became
6117 : * critical to do indexed lookups on the system catalogs.
6118 : *
6119 : * Bootstrapping these lookups is very hard. We want to be able to
6120 : * use an index on pg_attribute, for example, but in order to do so,
6121 : * we must have read pg_attribute for the attributes in the index,
6122 : * which implies that we need to use the index.
6123 : *
6124 : * In order to get around the problem, we do the following:
6125 : *
6126 : * + When the database system is initialized (at initdb time), we
6127 : * don't use indexes. We do sequential scans.
6128 : *
6129 : * + When the backend is started up in normal mode, we load an image
6130 : * of the appropriate relation descriptors, in internal format,
6131 : * from an initialization file in the data/base/... directory.
6132 : *
6133 : * + If the initialization file isn't there, then we create the
6134 : * relation descriptors using sequential scans and write 'em to
6135 : * the initialization file for use by subsequent backends.
6136 : *
6137 : * As of Postgres 9.0, there is one local initialization file in each
6138 : * database, plus one shared initialization file for shared catalogs.
6139 : *
6140 : * We could dispense with the initialization files and just build the
6141 : * critical reldescs the hard way on every backend startup, but that
6142 : * slows down backend startup noticeably.
6143 : *
6144 : * We can in fact go further, and save more relcache entries than
6145 : * just the ones that are absolutely critical; this allows us to speed
6146 : * up backend startup by not having to build such entries the hard way.
6147 : * Presently, all the catalog and index entries that are referred to
6148 : * by catcaches are stored in the initialization files.
6149 : *
6150 : * The same mechanism that detects when catcache and relcache entries
6151 : * need to be invalidated (due to catalog updates) also arranges to
6152 : * unlink the initialization files when the contents may be out of date.
6153 : * The files will then be rebuilt during the next backend startup.
6154 : */
6155 :
6156 : /*
6157 : * load_relcache_init_file -- attempt to load cache from the shared
6158 : * or local cache init file
6159 : *
6160 : * If successful, return true and set criticalRelcachesBuilt or
6161 : * criticalSharedRelcachesBuilt to true.
6162 : * If not successful, return false.
6163 : *
6164 : * NOTE: we assume we are already switched into CacheMemoryContext.
6165 : */
6166 : static bool
6167 66196 : load_relcache_init_file(bool shared)
6168 : {
6169 : FILE *fp;
6170 : char initfilename[MAXPGPATH];
6171 : Relation *rels;
6172 : int relno,
6173 : num_rels,
6174 : max_rels,
6175 : nailed_rels,
6176 : nailed_indexes,
6177 : magic;
6178 : int i;
6179 :
6180 66196 : if (shared)
6181 34416 : snprintf(initfilename, sizeof(initfilename), "global/%s",
6182 : RELCACHE_INIT_FILENAME);
6183 : else
6184 31780 : snprintf(initfilename, sizeof(initfilename), "%s/%s",
6185 : DatabasePath, RELCACHE_INIT_FILENAME);
6186 :
6187 66196 : fp = AllocateFile(initfilename, PG_BINARY_R);
6188 66196 : if (fp == NULL)
6189 8038 : return false;
6190 :
6191 : /*
6192 : * Read the index relcache entries from the file. Note we will not enter
6193 : * any of them into the cache if the read fails partway through; this
6194 : * helps to guard against broken init files.
6195 : */
6196 58158 : max_rels = 100;
6197 58158 : rels = (Relation *) palloc(max_rels * sizeof(Relation));
6198 58158 : num_rels = 0;
6199 58158 : nailed_rels = nailed_indexes = 0;
6200 :
6201 : /* check for correct magic number (compatible version) */
6202 58158 : if (fread(&magic, 1, sizeof(magic), fp) != sizeof(magic))
6203 0 : goto read_failed;
6204 58158 : if (magic != RELCACHE_INIT_FILEMAGIC)
6205 0 : goto read_failed;
6206 :
6207 58158 : for (relno = 0;; relno++)
6208 3919746 : {
6209 : Size len;
6210 : size_t nread;
6211 : Relation rel;
6212 : Form_pg_class relform;
6213 : bool has_not_null;
6214 :
6215 : /* first read the relation descriptor length */
6216 3977904 : nread = fread(&len, 1, sizeof(len), fp);
6217 3977904 : if (nread != sizeof(len))
6218 : {
6219 58158 : if (nread == 0)
6220 58158 : break; /* end of file */
6221 0 : goto read_failed;
6222 : }
6223 :
6224 : /* safety check for incompatible relcache layout */
6225 3919746 : if (len != sizeof(RelationData))
6226 0 : goto read_failed;
6227 :
6228 : /* allocate another relcache header */
6229 3919746 : if (num_rels >= max_rels)
6230 : {
6231 28390 : max_rels *= 2;
6232 28390 : rels = (Relation *) repalloc(rels, max_rels * sizeof(Relation));
6233 : }
6234 :
6235 3919746 : rel = rels[num_rels++] = (Relation) palloc(len);
6236 :
6237 : /* then, read the Relation structure */
6238 3919746 : if (fread(rel, 1, len, fp) != len)
6239 0 : goto read_failed;
6240 :
6241 : /* next read the relation tuple form */
6242 3919746 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6243 0 : goto read_failed;
6244 :
6245 3919746 : relform = (Form_pg_class) palloc(len);
6246 3919746 : if (fread(relform, 1, len, fp) != len)
6247 0 : goto read_failed;
6248 :
6249 3919746 : rel->rd_rel = relform;
6250 :
6251 : /* initialize attribute tuple forms */
6252 3919746 : rel->rd_att = CreateTemplateTupleDesc(relform->relnatts);
6253 3919746 : rel->rd_att->tdrefcount = 1; /* mark as refcounted */
6254 :
6255 3919746 : rel->rd_att->tdtypeid = relform->reltype ? relform->reltype : RECORDOID;
6256 3919746 : rel->rd_att->tdtypmod = -1; /* just to be sure */
6257 :
6258 : /* next read all the attribute tuple form data entries */
6259 3919746 : has_not_null = false;
6260 22979896 : for (i = 0; i < relform->relnatts; i++)
6261 : {
6262 19060150 : Form_pg_attribute attr = TupleDescAttr(rel->rd_att, i);
6263 :
6264 19060150 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6265 0 : goto read_failed;
6266 19060150 : if (len != ATTRIBUTE_FIXED_PART_SIZE)
6267 0 : goto read_failed;
6268 19060150 : if (fread(attr, 1, len, fp) != len)
6269 0 : goto read_failed;
6270 :
6271 19060150 : has_not_null |= attr->attnotnull;
6272 :
6273 19060150 : populate_compact_attribute(rel->rd_att, i);
6274 : }
6275 :
6276 : /* next read the access method specific field */
6277 3919746 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6278 0 : goto read_failed;
6279 3919746 : if (len > 0)
6280 : {
6281 0 : rel->rd_options = palloc(len);
6282 0 : if (fread(rel->rd_options, 1, len, fp) != len)
6283 0 : goto read_failed;
6284 0 : if (len != VARSIZE(rel->rd_options))
6285 0 : goto read_failed; /* sanity check */
6286 : }
6287 : else
6288 : {
6289 3919746 : rel->rd_options = NULL;
6290 : }
6291 :
6292 : /* mark not-null status */
6293 3919746 : if (has_not_null)
6294 : {
6295 1458914 : TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
6296 :
6297 1458914 : constr->has_not_null = true;
6298 1458914 : rel->rd_att->constr = constr;
6299 : }
6300 :
6301 : /*
6302 : * If it's an index, there's more to do. Note we explicitly ignore
6303 : * partitioned indexes here.
6304 : */
6305 3919746 : if (rel->rd_rel->relkind == RELKIND_INDEX)
6306 : {
6307 : MemoryContext indexcxt;
6308 : Oid *opfamily;
6309 : Oid *opcintype;
6310 : RegProcedure *support;
6311 : int nsupport;
6312 : int16 *indoption;
6313 : Oid *indcollation;
6314 :
6315 : /* Count nailed indexes to ensure we have 'em all */
6316 2460832 : if (rel->rd_isnailed)
6317 377338 : nailed_indexes++;
6318 :
6319 : /* read the pg_index tuple */
6320 2460832 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6321 0 : goto read_failed;
6322 :
6323 2460832 : rel->rd_indextuple = (HeapTuple) palloc(len);
6324 2460832 : if (fread(rel->rd_indextuple, 1, len, fp) != len)
6325 0 : goto read_failed;
6326 :
6327 : /* Fix up internal pointers in the tuple -- see heap_copytuple */
6328 2460832 : rel->rd_indextuple->t_data = (HeapTupleHeader) ((char *) rel->rd_indextuple + HEAPTUPLESIZE);
6329 2460832 : rel->rd_index = (Form_pg_index) GETSTRUCT(rel->rd_indextuple);
6330 :
6331 : /*
6332 : * prepare index info context --- parameters should match
6333 : * RelationInitIndexAccessInfo
6334 : */
6335 2460832 : indexcxt = AllocSetContextCreate(CacheMemoryContext,
6336 : "index info",
6337 : ALLOCSET_SMALL_SIZES);
6338 2460832 : rel->rd_indexcxt = indexcxt;
6339 2460832 : MemoryContextCopyAndSetIdentifier(indexcxt,
6340 : RelationGetRelationName(rel));
6341 :
6342 : /*
6343 : * Now we can fetch the index AM's API struct. (We can't store
6344 : * that in the init file, since it contains function pointers that
6345 : * might vary across server executions. Fortunately, it should be
6346 : * safe to call the amhandler even while bootstrapping indexes.)
6347 : */
6348 2460832 : InitIndexAmRoutine(rel);
6349 :
6350 : /* read the vector of opfamily OIDs */
6351 2460832 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6352 0 : goto read_failed;
6353 :
6354 2460832 : opfamily = (Oid *) MemoryContextAlloc(indexcxt, len);
6355 2460832 : if (fread(opfamily, 1, len, fp) != len)
6356 0 : goto read_failed;
6357 :
6358 2460832 : rel->rd_opfamily = opfamily;
6359 :
6360 : /* read the vector of opcintype OIDs */
6361 2460832 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6362 0 : goto read_failed;
6363 :
6364 2460832 : opcintype = (Oid *) MemoryContextAlloc(indexcxt, len);
6365 2460832 : if (fread(opcintype, 1, len, fp) != len)
6366 0 : goto read_failed;
6367 :
6368 2460832 : rel->rd_opcintype = opcintype;
6369 :
6370 : /* read the vector of support procedure OIDs */
6371 2460832 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6372 0 : goto read_failed;
6373 2460832 : support = (RegProcedure *) MemoryContextAlloc(indexcxt, len);
6374 2460832 : if (fread(support, 1, len, fp) != len)
6375 0 : goto read_failed;
6376 :
6377 2460832 : rel->rd_support = support;
6378 :
6379 : /* read the vector of collation OIDs */
6380 2460832 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6381 0 : goto read_failed;
6382 :
6383 2460832 : indcollation = (Oid *) MemoryContextAlloc(indexcxt, len);
6384 2460832 : if (fread(indcollation, 1, len, fp) != len)
6385 0 : goto read_failed;
6386 :
6387 2460832 : rel->rd_indcollation = indcollation;
6388 :
6389 : /* read the vector of indoption values */
6390 2460832 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6391 0 : goto read_failed;
6392 :
6393 2460832 : indoption = (int16 *) MemoryContextAlloc(indexcxt, len);
6394 2460832 : if (fread(indoption, 1, len, fp) != len)
6395 0 : goto read_failed;
6396 :
6397 2460832 : rel->rd_indoption = indoption;
6398 :
6399 : /* read the vector of opcoptions values */
6400 2460832 : rel->rd_opcoptions = (bytea **)
6401 2460832 : MemoryContextAllocZero(indexcxt, sizeof(*rel->rd_opcoptions) * relform->relnatts);
6402 :
6403 6492760 : for (i = 0; i < relform->relnatts; i++)
6404 : {
6405 4031928 : if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
6406 0 : goto read_failed;
6407 :
6408 4031928 : if (len > 0)
6409 : {
6410 0 : rel->rd_opcoptions[i] = (bytea *) MemoryContextAlloc(indexcxt, len);
6411 0 : if (fread(rel->rd_opcoptions[i], 1, len, fp) != len)
6412 0 : goto read_failed;
6413 : }
6414 : }
6415 :
6416 : /* set up zeroed fmgr-info vector */
6417 2460832 : nsupport = relform->relnatts * rel->rd_indam->amsupport;
6418 2460832 : rel->rd_supportinfo = (FmgrInfo *)
6419 2460832 : MemoryContextAllocZero(indexcxt, nsupport * sizeof(FmgrInfo));
6420 : }
6421 : else
6422 : {
6423 : /* Count nailed rels to ensure we have 'em all */
6424 1458914 : if (rel->rd_isnailed)
6425 262400 : nailed_rels++;
6426 :
6427 : /* Load table AM data */
6428 1458914 : if (RELKIND_HAS_TABLE_AM(rel->rd_rel->relkind) || rel->rd_rel->relkind == RELKIND_SEQUENCE)
6429 1458914 : RelationInitTableAccessMethod(rel);
6430 :
6431 : Assert(rel->rd_index == NULL);
6432 : Assert(rel->rd_indextuple == NULL);
6433 : Assert(rel->rd_indexcxt == NULL);
6434 : Assert(rel->rd_indam == NULL);
6435 : Assert(rel->rd_opfamily == NULL);
6436 : Assert(rel->rd_opcintype == NULL);
6437 : Assert(rel->rd_support == NULL);
6438 : Assert(rel->rd_supportinfo == NULL);
6439 : Assert(rel->rd_indoption == NULL);
6440 : Assert(rel->rd_indcollation == NULL);
6441 : Assert(rel->rd_opcoptions == NULL);
6442 : }
6443 :
6444 : /*
6445 : * Rules and triggers are not saved (mainly because the internal
6446 : * format is complex and subject to change). They must be rebuilt if
6447 : * needed by RelationCacheInitializePhase3. This is not expected to
6448 : * be a big performance hit since few system catalogs have such. Ditto
6449 : * for RLS policy data, partition info, index expressions, predicates,
6450 : * exclusion info, and FDW info.
6451 : */
6452 3919746 : rel->rd_rules = NULL;
6453 3919746 : rel->rd_rulescxt = NULL;
6454 3919746 : rel->trigdesc = NULL;
6455 3919746 : rel->rd_rsdesc = NULL;
6456 3919746 : rel->rd_partkey = NULL;
6457 3919746 : rel->rd_partkeycxt = NULL;
6458 3919746 : rel->rd_partdesc = NULL;
6459 3919746 : rel->rd_partdesc_nodetached = NULL;
6460 3919746 : rel->rd_partdesc_nodetached_xmin = InvalidTransactionId;
6461 3919746 : rel->rd_pdcxt = NULL;
6462 3919746 : rel->rd_pddcxt = NULL;
6463 3919746 : rel->rd_partcheck = NIL;
6464 3919746 : rel->rd_partcheckvalid = false;
6465 3919746 : rel->rd_partcheckcxt = NULL;
6466 3919746 : rel->rd_indexprs = NIL;
6467 3919746 : rel->rd_indpred = NIL;
6468 3919746 : rel->rd_exclops = NULL;
6469 3919746 : rel->rd_exclprocs = NULL;
6470 3919746 : rel->rd_exclstrats = NULL;
6471 3919746 : rel->rd_fdwroutine = NULL;
6472 :
6473 : /*
6474 : * Reset transient-state fields in the relcache entry
6475 : */
6476 3919746 : rel->rd_smgr = NULL;
6477 3919746 : if (rel->rd_isnailed)
6478 639738 : rel->rd_refcnt = 1;
6479 : else
6480 3280008 : rel->rd_refcnt = 0;
6481 3919746 : rel->rd_indexvalid = false;
6482 3919746 : rel->rd_indexlist = NIL;
6483 3919746 : rel->rd_pkindex = InvalidOid;
6484 3919746 : rel->rd_replidindex = InvalidOid;
6485 3919746 : rel->rd_attrsvalid = false;
6486 3919746 : rel->rd_keyattr = NULL;
6487 3919746 : rel->rd_pkattr = NULL;
6488 3919746 : rel->rd_idattr = NULL;
6489 3919746 : rel->rd_pubdesc = NULL;
6490 3919746 : rel->rd_statvalid = false;
6491 3919746 : rel->rd_statlist = NIL;
6492 3919746 : rel->rd_fkeyvalid = false;
6493 3919746 : rel->rd_fkeylist = NIL;
6494 3919746 : rel->rd_createSubid = InvalidSubTransactionId;
6495 3919746 : rel->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
6496 3919746 : rel->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
6497 3919746 : rel->rd_droppedSubid = InvalidSubTransactionId;
6498 3919746 : rel->rd_amcache = NULL;
6499 3919746 : rel->pgstat_info = NULL;
6500 :
6501 : /*
6502 : * Recompute lock and physical addressing info. This is needed in
6503 : * case the pg_internal.init file was copied from some other database
6504 : * by CREATE DATABASE.
6505 : */
6506 3919746 : RelationInitLockInfo(rel);
6507 3919746 : RelationInitPhysicalAddr(rel);
6508 : }
6509 :
6510 : /*
6511 : * We reached the end of the init file without apparent problem. Did we
6512 : * get the right number of nailed items? This is a useful crosscheck in
6513 : * case the set of critical rels or indexes changes. However, that should
6514 : * not happen in a normally-running system, so let's bleat if it does.
6515 : *
6516 : * For the shared init file, we're called before client authentication is
6517 : * done, which means that elog(WARNING) will go only to the postmaster
6518 : * log, where it's easily missed. To ensure that developers notice bad
6519 : * values of NUM_CRITICAL_SHARED_RELS/NUM_CRITICAL_SHARED_INDEXES, we put
6520 : * an Assert(false) there.
6521 : */
6522 58158 : if (shared)
6523 : {
6524 29768 : if (nailed_rels != NUM_CRITICAL_SHARED_RELS ||
6525 : nailed_indexes != NUM_CRITICAL_SHARED_INDEXES)
6526 : {
6527 0 : elog(WARNING, "found %d nailed shared rels and %d nailed shared indexes in init file, but expected %d and %d respectively",
6528 : nailed_rels, nailed_indexes,
6529 : NUM_CRITICAL_SHARED_RELS, NUM_CRITICAL_SHARED_INDEXES);
6530 : /* Make sure we get developers' attention about this */
6531 : Assert(false);
6532 : /* In production builds, recover by bootstrapping the relcache */
6533 0 : goto read_failed;
6534 : }
6535 : }
6536 : else
6537 : {
6538 28390 : if (nailed_rels != NUM_CRITICAL_LOCAL_RELS ||
6539 : nailed_indexes != NUM_CRITICAL_LOCAL_INDEXES)
6540 : {
6541 0 : elog(WARNING, "found %d nailed rels and %d nailed indexes in init file, but expected %d and %d respectively",
6542 : nailed_rels, nailed_indexes,
6543 : NUM_CRITICAL_LOCAL_RELS, NUM_CRITICAL_LOCAL_INDEXES);
6544 : /* We don't need an Assert() in this case */
6545 0 : goto read_failed;
6546 : }
6547 : }
6548 :
6549 : /*
6550 : * OK, all appears well.
6551 : *
6552 : * Now insert all the new relcache entries into the cache.
6553 : */
6554 3977904 : for (relno = 0; relno < num_rels; relno++)
6555 : {
6556 3919746 : RelationCacheInsert(rels[relno], false);
6557 : }
6558 :
6559 58158 : pfree(rels);
6560 58158 : FreeFile(fp);
6561 :
6562 58158 : if (shared)
6563 29768 : criticalSharedRelcachesBuilt = true;
6564 : else
6565 28390 : criticalRelcachesBuilt = true;
6566 58158 : return true;
6567 :
6568 : /*
6569 : * init file is broken, so do it the hard way. We don't bother trying to
6570 : * free the clutter we just allocated; it's not in the relcache so it
6571 : * won't hurt.
6572 : */
6573 0 : read_failed:
6574 0 : pfree(rels);
6575 0 : FreeFile(fp);
6576 :
6577 0 : return false;
6578 : }
6579 :
6580 : /*
6581 : * Write out a new initialization file with the current contents
6582 : * of the relcache (either shared rels or local rels, as indicated).
6583 : */
6584 : static void
6585 7804 : write_relcache_init_file(bool shared)
6586 : {
6587 : FILE *fp;
6588 : char tempfilename[MAXPGPATH];
6589 : char finalfilename[MAXPGPATH];
6590 : int magic;
6591 : HASH_SEQ_STATUS status;
6592 : RelIdCacheEnt *idhentry;
6593 : int i;
6594 :
6595 : /*
6596 : * If we have already received any relcache inval events, there's no
6597 : * chance of succeeding so we may as well skip the whole thing.
6598 : */
6599 7804 : if (relcacheInvalsReceived != 0L)
6600 226 : return;
6601 :
6602 : /*
6603 : * We must write a temporary file and rename it into place. Otherwise,
6604 : * another backend starting at about the same time might crash trying to
6605 : * read the partially-complete file.
6606 : */
6607 7578 : if (shared)
6608 : {
6609 3790 : snprintf(tempfilename, sizeof(tempfilename), "global/%s.%d",
6610 : RELCACHE_INIT_FILENAME, MyProcPid);
6611 3790 : snprintf(finalfilename, sizeof(finalfilename), "global/%s",
6612 : RELCACHE_INIT_FILENAME);
6613 : }
6614 : else
6615 : {
6616 3788 : snprintf(tempfilename, sizeof(tempfilename), "%s/%s.%d",
6617 : DatabasePath, RELCACHE_INIT_FILENAME, MyProcPid);
6618 3788 : snprintf(finalfilename, sizeof(finalfilename), "%s/%s",
6619 : DatabasePath, RELCACHE_INIT_FILENAME);
6620 : }
6621 :
6622 7578 : unlink(tempfilename); /* in case it exists w/wrong permissions */
6623 :
6624 7578 : fp = AllocateFile(tempfilename, PG_BINARY_W);
6625 7578 : if (fp == NULL)
6626 : {
6627 : /*
6628 : * We used to consider this a fatal error, but we might as well
6629 : * continue with backend startup ...
6630 : */
6631 0 : ereport(WARNING,
6632 : (errcode_for_file_access(),
6633 : errmsg("could not create relation-cache initialization file \"%s\": %m",
6634 : tempfilename),
6635 : errdetail("Continuing anyway, but there's something wrong.")));
6636 0 : return;
6637 : }
6638 :
6639 : /*
6640 : * Write a magic number to serve as a file version identifier. We can
6641 : * change the magic number whenever the relcache layout changes.
6642 : */
6643 7578 : magic = RELCACHE_INIT_FILEMAGIC;
6644 7578 : if (fwrite(&magic, 1, sizeof(magic), fp) != sizeof(magic))
6645 0 : ereport(FATAL,
6646 : errcode_for_file_access(),
6647 : errmsg_internal("could not write init file: %m"));
6648 :
6649 : /*
6650 : * Write all the appropriate reldescs (in no particular order).
6651 : */
6652 7578 : hash_seq_init(&status, RelationIdCache);
6653 :
6654 1045764 : while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
6655 : {
6656 1038186 : Relation rel = idhentry->reldesc;
6657 1038186 : Form_pg_class relform = rel->rd_rel;
6658 :
6659 : /* ignore if not correct group */
6660 1038186 : if (relform->relisshared != shared)
6661 519186 : continue;
6662 :
6663 : /*
6664 : * Ignore if not supposed to be in init file. We can allow any shared
6665 : * relation that's been loaded so far to be in the shared init file,
6666 : * but unshared relations must be ones that should be in the local
6667 : * file per RelationIdIsInInitFile. (Note: if you want to change the
6668 : * criterion for rels to be kept in the init file, see also inval.c.
6669 : * The reason for filtering here is to be sure that we don't put
6670 : * anything into the local init file for which a relcache inval would
6671 : * not cause invalidation of that init file.)
6672 : */
6673 519000 : if (!shared && !RelationIdIsInInitFile(RelationGetRelid(rel)))
6674 : {
6675 : /* Nailed rels had better get stored. */
6676 : Assert(!rel->rd_isnailed);
6677 0 : continue;
6678 : }
6679 :
6680 : /* first write the relcache entry proper */
6681 519000 : write_item(rel, sizeof(RelationData), fp);
6682 :
6683 : /* next write the relation tuple form */
6684 519000 : write_item(relform, CLASS_TUPLE_SIZE, fp);
6685 :
6686 : /* next, do all the attribute tuple form data entries */
6687 3045776 : for (i = 0; i < relform->relnatts; i++)
6688 : {
6689 2526776 : write_item(TupleDescAttr(rel->rd_att, i),
6690 : ATTRIBUTE_FIXED_PART_SIZE, fp);
6691 : }
6692 :
6693 : /* next, do the access method specific field */
6694 519000 : write_item(rel->rd_options,
6695 519000 : (rel->rd_options ? VARSIZE(rel->rd_options) : 0),
6696 : fp);
6697 :
6698 : /*
6699 : * If it's an index, there's more to do. Note we explicitly ignore
6700 : * partitioned indexes here.
6701 : */
6702 519000 : if (rel->rd_rel->relkind == RELKIND_INDEX)
6703 : {
6704 : /* write the pg_index tuple */
6705 : /* we assume this was created by heap_copytuple! */
6706 325796 : write_item(rel->rd_indextuple,
6707 325796 : HEAPTUPLESIZE + rel->rd_indextuple->t_len,
6708 : fp);
6709 :
6710 : /* write the vector of opfamily OIDs */
6711 325796 : write_item(rel->rd_opfamily,
6712 325796 : relform->relnatts * sizeof(Oid),
6713 : fp);
6714 :
6715 : /* write the vector of opcintype OIDs */
6716 325796 : write_item(rel->rd_opcintype,
6717 325796 : relform->relnatts * sizeof(Oid),
6718 : fp);
6719 :
6720 : /* write the vector of support procedure OIDs */
6721 325796 : write_item(rel->rd_support,
6722 325796 : relform->relnatts * (rel->rd_indam->amsupport * sizeof(RegProcedure)),
6723 : fp);
6724 :
6725 : /* write the vector of collation OIDs */
6726 325796 : write_item(rel->rd_indcollation,
6727 325796 : relform->relnatts * sizeof(Oid),
6728 : fp);
6729 :
6730 : /* write the vector of indoption values */
6731 325796 : write_item(rel->rd_indoption,
6732 325796 : relform->relnatts * sizeof(int16),
6733 : fp);
6734 :
6735 : Assert(rel->rd_opcoptions);
6736 :
6737 : /* write the vector of opcoptions values */
6738 859946 : for (i = 0; i < relform->relnatts; i++)
6739 : {
6740 534150 : bytea *opt = rel->rd_opcoptions[i];
6741 :
6742 534150 : write_item(opt, opt ? VARSIZE(opt) : 0, fp);
6743 : }
6744 : }
6745 : }
6746 :
6747 7578 : if (FreeFile(fp))
6748 0 : ereport(FATAL,
6749 : errcode_for_file_access(),
6750 : errmsg_internal("could not write init file: %m"));
6751 :
6752 : /*
6753 : * Now we have to check whether the data we've so painstakingly
6754 : * accumulated is already obsolete due to someone else's just-committed
6755 : * catalog changes. If so, we just delete the temp file and leave it to
6756 : * the next backend to try again. (Our own relcache entries will be
6757 : * updated by SI message processing, but we can't be sure whether what we
6758 : * wrote out was up-to-date.)
6759 : *
6760 : * This mustn't run concurrently with the code that unlinks an init file
6761 : * and sends SI messages, so grab a serialization lock for the duration.
6762 : */
6763 7578 : LWLockAcquire(RelCacheInitLock, LW_EXCLUSIVE);
6764 :
6765 : /* Make sure we have seen all incoming SI messages */
6766 7578 : AcceptInvalidationMessages();
6767 :
6768 : /*
6769 : * If we have received any SI relcache invals since backend start, assume
6770 : * we may have written out-of-date data.
6771 : */
6772 7578 : if (relcacheInvalsReceived == 0L)
6773 : {
6774 : /*
6775 : * OK, rename the temp file to its final name, deleting any
6776 : * previously-existing init file.
6777 : *
6778 : * Note: a failure here is possible under Cygwin, if some other
6779 : * backend is holding open an unlinked-but-not-yet-gone init file. So
6780 : * treat this as a noncritical failure; just remove the useless temp
6781 : * file on failure.
6782 : */
6783 7574 : if (rename(tempfilename, finalfilename) < 0)
6784 0 : unlink(tempfilename);
6785 : }
6786 : else
6787 : {
6788 : /* Delete the already-obsolete temp file */
6789 4 : unlink(tempfilename);
6790 : }
6791 :
6792 7578 : LWLockRelease(RelCacheInitLock);
6793 : }
6794 :
6795 : /* write a chunk of data preceded by its length */
6796 : static void
6797 6572702 : write_item(const void *data, Size len, FILE *fp)
6798 : {
6799 6572702 : if (fwrite(&len, 1, sizeof(len), fp) != sizeof(len))
6800 0 : ereport(FATAL,
6801 : errcode_for_file_access(),
6802 : errmsg_internal("could not write init file: %m"));
6803 6572702 : if (len > 0 && fwrite(data, 1, len, fp) != len)
6804 0 : ereport(FATAL,
6805 : errcode_for_file_access(),
6806 : errmsg_internal("could not write init file: %m"));
6807 6572702 : }
6808 :
6809 : /*
6810 : * Determine whether a given relation (identified by OID) is one of the ones
6811 : * we should store in a relcache init file.
6812 : *
6813 : * We must cache all nailed rels, and for efficiency we should cache every rel
6814 : * that supports a syscache. The former set is almost but not quite a subset
6815 : * of the latter. The special cases are relations where
6816 : * RelationCacheInitializePhase2/3 chooses to nail for efficiency reasons, but
6817 : * which do not support any syscache.
6818 : */
6819 : bool
6820 2714742 : RelationIdIsInInitFile(Oid relationId)
6821 : {
6822 2714742 : if (relationId == SharedSecLabelRelationId ||
6823 2706562 : relationId == TriggerRelidNameIndexId ||
6824 2706260 : relationId == DatabaseNameIndexId ||
6825 : relationId == SharedSecLabelObjectIndexId)
6826 : {
6827 : /*
6828 : * If this Assert fails, we don't need the applicable special case
6829 : * anymore.
6830 : */
6831 : Assert(!RelationSupportsSysCache(relationId));
6832 8802 : return true;
6833 : }
6834 2705940 : return RelationSupportsSysCache(relationId);
6835 : }
6836 :
6837 : /*
6838 : * Invalidate (remove) the init file during commit of a transaction that
6839 : * changed one or more of the relation cache entries that are kept in the
6840 : * local init file.
6841 : *
6842 : * To be safe against concurrent inspection or rewriting of the init file,
6843 : * we must take RelCacheInitLock, then remove the old init file, then send
6844 : * the SI messages that include relcache inval for such relations, and then
6845 : * release RelCacheInitLock. This serializes the whole affair against
6846 : * write_relcache_init_file, so that we can be sure that any other process
6847 : * that's concurrently trying to create a new init file won't move an
6848 : * already-stale version into place after we unlink. Also, because we unlink
6849 : * before sending the SI messages, a backend that's currently starting cannot
6850 : * read the now-obsolete init file and then miss the SI messages that will
6851 : * force it to update its relcache entries. (This works because the backend
6852 : * startup sequence gets into the sinval array before trying to load the init
6853 : * file.)
6854 : *
6855 : * We take the lock and do the unlink in RelationCacheInitFilePreInvalidate,
6856 : * then release the lock in RelationCacheInitFilePostInvalidate. Caller must
6857 : * send any pending SI messages between those calls.
6858 : */
6859 : void
6860 78312 : RelationCacheInitFilePreInvalidate(void)
6861 : {
6862 : char localinitfname[MAXPGPATH];
6863 : char sharedinitfname[MAXPGPATH];
6864 :
6865 78312 : if (DatabasePath)
6866 78312 : snprintf(localinitfname, sizeof(localinitfname), "%s/%s",
6867 : DatabasePath, RELCACHE_INIT_FILENAME);
6868 78312 : snprintf(sharedinitfname, sizeof(sharedinitfname), "global/%s",
6869 : RELCACHE_INIT_FILENAME);
6870 :
6871 78312 : LWLockAcquire(RelCacheInitLock, LW_EXCLUSIVE);
6872 :
6873 : /*
6874 : * The files might not be there if no backend has been started since the
6875 : * last removal. But complain about failures other than ENOENT with
6876 : * ERROR. Fortunately, it's not too late to abort the transaction if we
6877 : * can't get rid of the would-be-obsolete init file.
6878 : */
6879 78312 : if (DatabasePath)
6880 78312 : unlink_initfile(localinitfname, ERROR);
6881 78312 : unlink_initfile(sharedinitfname, ERROR);
6882 78312 : }
6883 :
6884 : void
6885 78312 : RelationCacheInitFilePostInvalidate(void)
6886 : {
6887 78312 : LWLockRelease(RelCacheInitLock);
6888 78312 : }
6889 :
6890 : /*
6891 : * Remove the init files during postmaster startup.
6892 : *
6893 : * We used to keep the init files across restarts, but that is unsafe in PITR
6894 : * scenarios, and even in simple crash-recovery cases there are windows for
6895 : * the init files to become out-of-sync with the database. So now we just
6896 : * remove them during startup and expect the first backend launch to rebuild
6897 : * them. Of course, this has to happen in each database of the cluster.
6898 : */
6899 : void
6900 1856 : RelationCacheInitFileRemove(void)
6901 : {
6902 1856 : const char *tblspcdir = PG_TBLSPC_DIR;
6903 : DIR *dir;
6904 : struct dirent *de;
6905 : char path[MAXPGPATH + sizeof(PG_TBLSPC_DIR) + sizeof(TABLESPACE_VERSION_DIRECTORY)];
6906 :
6907 1856 : snprintf(path, sizeof(path), "global/%s",
6908 : RELCACHE_INIT_FILENAME);
6909 1856 : unlink_initfile(path, LOG);
6910 :
6911 : /* Scan everything in the default tablespace */
6912 1856 : RelationCacheInitFileRemoveInDir("base");
6913 :
6914 : /* Scan the tablespace link directory to find non-default tablespaces */
6915 1856 : dir = AllocateDir(tblspcdir);
6916 :
6917 7520 : while ((de = ReadDirExtended(dir, tblspcdir, LOG)) != NULL)
6918 : {
6919 3808 : if (strspn(de->d_name, "0123456789") == strlen(de->d_name))
6920 : {
6921 : /* Scan the tablespace dir for per-database dirs */
6922 96 : snprintf(path, sizeof(path), "%s/%s/%s",
6923 96 : tblspcdir, de->d_name, TABLESPACE_VERSION_DIRECTORY);
6924 96 : RelationCacheInitFileRemoveInDir(path);
6925 : }
6926 : }
6927 :
6928 1856 : FreeDir(dir);
6929 1856 : }
6930 :
6931 : /* Process one per-tablespace directory for RelationCacheInitFileRemove */
6932 : static void
6933 1952 : RelationCacheInitFileRemoveInDir(const char *tblspcpath)
6934 : {
6935 : DIR *dir;
6936 : struct dirent *de;
6937 : char initfilename[MAXPGPATH * 2];
6938 :
6939 : /* Scan the tablespace directory to find per-database directories */
6940 1952 : dir = AllocateDir(tblspcpath);
6941 :
6942 13764 : while ((de = ReadDirExtended(dir, tblspcpath, LOG)) != NULL)
6943 : {
6944 9860 : if (strspn(de->d_name, "0123456789") == strlen(de->d_name))
6945 : {
6946 : /* Try to remove the init file in each database */
6947 5810 : snprintf(initfilename, sizeof(initfilename), "%s/%s/%s",
6948 5810 : tblspcpath, de->d_name, RELCACHE_INIT_FILENAME);
6949 5810 : unlink_initfile(initfilename, LOG);
6950 : }
6951 : }
6952 :
6953 1952 : FreeDir(dir);
6954 1952 : }
6955 :
6956 : static void
6957 164290 : unlink_initfile(const char *initfilename, int elevel)
6958 : {
6959 164290 : if (unlink(initfilename) < 0)
6960 : {
6961 : /* It might not be there, but log any error other than ENOENT */
6962 161000 : if (errno != ENOENT)
6963 0 : ereport(elevel,
6964 : (errcode_for_file_access(),
6965 : errmsg("could not remove cache file \"%s\": %m",
6966 : initfilename)));
6967 : }
6968 164290 : }
6969 :
6970 : /*
6971 : * ResourceOwner callbacks
6972 : */
6973 : static char *
6974 0 : ResOwnerPrintRelCache(Datum res)
6975 : {
6976 0 : Relation rel = (Relation) DatumGetPointer(res);
6977 :
6978 0 : return psprintf("relation \"%s\"", RelationGetRelationName(rel));
6979 : }
6980 :
6981 : static void
6982 43712 : ResOwnerReleaseRelation(Datum res)
6983 : {
6984 43712 : Relation rel = (Relation) DatumGetPointer(res);
6985 :
6986 : /*
6987 : * This reference has already been removed from the resource owner, so
6988 : * just decrement reference count without calling
6989 : * ResourceOwnerForgetRelationRef.
6990 : */
6991 : Assert(rel->rd_refcnt > 0);
6992 43712 : rel->rd_refcnt -= 1;
6993 :
6994 43712 : RelationCloseCleanup((Relation) res);
6995 43712 : }
|
