LCOV - code coverage report
Current view: top level - src/backend/utils/cache - relcache.c (source / functions) Hit Total Coverage
Test: PostgreSQL 17devel Lines: 1833 2003 91.5 %
Date: 2023-12-05 09:10:49 Functions: 80 81 98.8 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /*-------------------------------------------------------------------------
       2             :  *
       3             :  * relcache.c
       4             :  *    POSTGRES relation descriptor cache code
       5             :  *
       6             :  * Portions Copyright (c) 1996-2023, 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/nbtree.h"
      36             : #include "access/parallel.h"
      37             : #include "access/reloptions.h"
      38             : #include "access/sysattr.h"
      39             : #include "access/table.h"
      40             : #include "access/tableam.h"
      41             : #include "access/tupdesc_details.h"
      42             : #include "access/xact.h"
      43             : #include "access/xlog.h"
      44             : #include "catalog/binary_upgrade.h"
      45             : #include "catalog/catalog.h"
      46             : #include "catalog/indexing.h"
      47             : #include "catalog/namespace.h"
      48             : #include "catalog/partition.h"
      49             : #include "catalog/pg_am.h"
      50             : #include "catalog/pg_amproc.h"
      51             : #include "catalog/pg_attrdef.h"
      52             : #include "catalog/pg_auth_members.h"
      53             : #include "catalog/pg_authid.h"
      54             : #include "catalog/pg_constraint.h"
      55             : #include "catalog/pg_database.h"
      56             : #include "catalog/pg_namespace.h"
      57             : #include "catalog/pg_opclass.h"
      58             : #include "catalog/pg_proc.h"
      59             : #include "catalog/pg_publication.h"
      60             : #include "catalog/pg_rewrite.h"
      61             : #include "catalog/pg_shseclabel.h"
      62             : #include "catalog/pg_statistic_ext.h"
      63             : #include "catalog/pg_subscription.h"
      64             : #include "catalog/pg_tablespace.h"
      65             : #include "catalog/pg_trigger.h"
      66             : #include "catalog/pg_type.h"
      67             : #include "catalog/schemapg.h"
      68             : #include "catalog/storage.h"
      69             : #include "commands/policy.h"
      70             : #include "commands/publicationcmds.h"
      71             : #include "commands/trigger.h"
      72             : #include "miscadmin.h"
      73             : #include "nodes/makefuncs.h"
      74             : #include "nodes/nodeFuncs.h"
      75             : #include "optimizer/optimizer.h"
      76             : #include "pgstat.h"
      77             : #include "rewrite/rewriteDefine.h"
      78             : #include "rewrite/rowsecurity.h"
      79             : #include "storage/lmgr.h"
      80             : #include "storage/smgr.h"
      81             : #include "utils/array.h"
      82             : #include "utils/builtins.h"
      83             : #include "utils/catcache.h"
      84             : #include "utils/datum.h"
      85             : #include "utils/fmgroids.h"
      86             : #include "utils/inval.h"
      87             : #include "utils/lsyscache.h"
      88             : #include "utils/memutils.h"
      89             : #include "utils/relmapper.h"
      90             : #include "utils/resowner.h"
      91             : #include "utils/snapmgr.h"
      92             : #include "utils/syscache.h"
      93             : 
      94             : #define RELCACHE_INIT_FILEMAGIC     0x573266    /* version ID value */
      95             : 
      96             : /*
      97             :  * Whether to bother checking if relation cache memory needs to be freed
      98             :  * eagerly.  See also RelationBuildDesc() and pg_config_manual.h.
      99             :  */
     100             : #if defined(RECOVER_RELATION_BUILD_MEMORY) && (RECOVER_RELATION_BUILD_MEMORY != 0)
     101             : #define MAYBE_RECOVER_RELATION_BUILD_MEMORY 1
     102             : #else
     103             : #define RECOVER_RELATION_BUILD_MEMORY 0
     104             : #ifdef DISCARD_CACHES_ENABLED
     105             : #define MAYBE_RECOVER_RELATION_BUILD_MEMORY 1
     106             : #endif
     107             : #endif
     108             : 
     109             : /*
     110             :  *      hardcoded tuple descriptors, contents generated by genbki.pl
     111             :  */
     112             : static const FormData_pg_attribute Desc_pg_class[Natts_pg_class] = {Schema_pg_class};
     113             : static const FormData_pg_attribute Desc_pg_attribute[Natts_pg_attribute] = {Schema_pg_attribute};
     114             : static const FormData_pg_attribute Desc_pg_proc[Natts_pg_proc] = {Schema_pg_proc};
     115             : static const FormData_pg_attribute Desc_pg_type[Natts_pg_type] = {Schema_pg_type};
     116             : static const FormData_pg_attribute Desc_pg_database[Natts_pg_database] = {Schema_pg_database};
     117             : static const FormData_pg_attribute Desc_pg_authid[Natts_pg_authid] = {Schema_pg_authid};
     118             : static const FormData_pg_attribute Desc_pg_auth_members[Natts_pg_auth_members] = {Schema_pg_auth_members};
     119             : static const FormData_pg_attribute Desc_pg_index[Natts_pg_index] = {Schema_pg_index};
     120             : static const FormData_pg_attribute Desc_pg_shseclabel[Natts_pg_shseclabel] = {Schema_pg_shseclabel};
     121             : static const FormData_pg_attribute Desc_pg_subscription[Natts_pg_subscription] = {Schema_pg_subscription};
     122             : 
     123             : /*
     124             :  *      Hash tables that index the relation cache
     125             :  *
     126             :  *      We used to index the cache by both name and OID, but now there
     127             :  *      is only an index by OID.
     128             :  */
     129             : typedef struct relidcacheent
     130             : {
     131             :     Oid         reloid;
     132             :     Relation    reldesc;
     133             : } RelIdCacheEnt;
     134             : 
     135             : static HTAB *RelationIdCache;
     136             : 
     137             : /*
     138             :  * This flag is false until we have prepared the critical relcache entries
     139             :  * that are needed to do indexscans on the tables read by relcache building.
     140             :  */
     141             : bool        criticalRelcachesBuilt = false;
     142             : 
     143             : /*
     144             :  * This flag is false until we have prepared the critical relcache entries
     145             :  * for shared catalogs (which are the tables needed for login).
     146             :  */
     147             : bool        criticalSharedRelcachesBuilt = false;
     148             : 
     149             : /*
     150             :  * This counter counts relcache inval events received since backend startup
     151             :  * (but only for rels that are actually in cache).  Presently, we use it only
     152             :  * to detect whether data about to be written by write_relcache_init_file()
     153             :  * might already be obsolete.
     154             :  */
     155             : static long relcacheInvalsReceived = 0L;
     156             : 
     157             : /*
     158             :  * in_progress_list is a stack of ongoing RelationBuildDesc() calls.  CREATE
     159             :  * INDEX CONCURRENTLY makes catalog changes under ShareUpdateExclusiveLock.
     160             :  * It critically relies on each backend absorbing those changes no later than
     161             :  * next transaction start.  Hence, RelationBuildDesc() loops until it finishes
     162             :  * without accepting a relevant invalidation.  (Most invalidation consumers
     163             :  * don't do this.)
     164             :  */
     165             : typedef struct inprogressent
     166             : {
     167             :     Oid         reloid;         /* OID of relation being built */
     168             :     bool        invalidated;    /* whether an invalidation arrived for it */
     169             : } InProgressEnt;
     170             : 
     171             : static InProgressEnt *in_progress_list;
     172             : static int  in_progress_list_len;
     173             : static int  in_progress_list_maxlen;
     174             : 
     175             : /*
     176             :  * eoxact_list[] stores the OIDs of relations that (might) need AtEOXact
     177             :  * cleanup work.  This list intentionally has limited size; if it overflows,
     178             :  * we fall back to scanning the whole hashtable.  There is no value in a very
     179             :  * large list because (1) at some point, a hash_seq_search scan is faster than
     180             :  * retail lookups, and (2) the value of this is to reduce EOXact work for
     181             :  * short transactions, which can't have dirtied all that many tables anyway.
     182             :  * EOXactListAdd() does not bother to prevent duplicate list entries, so the
     183             :  * cleanup processing must be idempotent.
     184             :  */
     185             : #define MAX_EOXACT_LIST 32
     186             : static Oid  eoxact_list[MAX_EOXACT_LIST];
     187             : static int  eoxact_list_len = 0;
     188             : static bool eoxact_list_overflowed = false;
     189             : 
     190             : #define EOXactListAdd(rel) \
     191             :     do { \
     192             :         if (eoxact_list_len < MAX_EOXACT_LIST) \
     193             :             eoxact_list[eoxact_list_len++] = (rel)->rd_id; \
     194             :         else \
     195             :             eoxact_list_overflowed = true; \
     196             :     } while (0)
     197             : 
     198             : /*
     199             :  * EOXactTupleDescArray stores TupleDescs that (might) need AtEOXact
     200             :  * cleanup work.  The array expands as needed; there is no hashtable because
     201             :  * we don't need to access individual items except at EOXact.
     202             :  */
     203             : static TupleDesc *EOXactTupleDescArray;
     204             : static int  NextEOXactTupleDescNum = 0;
     205             : static int  EOXactTupleDescArrayLen = 0;
     206             : 
     207             : /*
     208             :  *      macros to manipulate the lookup hashtable
     209             :  */
     210             : #define RelationCacheInsert(RELATION, replace_allowed)  \
     211             : do { \
     212             :     RelIdCacheEnt *hentry; bool found; \
     213             :     hentry = (RelIdCacheEnt *) hash_search(RelationIdCache, \
     214             :                                            &((RELATION)->rd_id), \
     215             :                                            HASH_ENTER, &found); \
     216             :     if (found) \
     217             :     { \
     218             :         /* see comments in RelationBuildDesc and RelationBuildLocalRelation */ \
     219             :         Relation _old_rel = hentry->reldesc; \
     220             :         Assert(replace_allowed); \
     221             :         hentry->reldesc = (RELATION); \
     222             :         if (RelationHasReferenceCountZero(_old_rel)) \
     223             :             RelationDestroyRelation(_old_rel, false); \
     224             :         else if (!IsBootstrapProcessingMode()) \
     225             :             elog(WARNING, "leaking still-referenced relcache entry for \"%s\"", \
     226             :                  RelationGetRelationName(_old_rel)); \
     227             :     } \
     228             :     else \
     229             :         hentry->reldesc = (RELATION); \
     230             : } while(0)
     231             : 
     232             : #define RelationIdCacheLookup(ID, RELATION) \
     233             : do { \
     234             :     RelIdCacheEnt *hentry; \
     235             :     hentry = (RelIdCacheEnt *) hash_search(RelationIdCache, \
     236             :                                            &(ID), \
     237             :                                            HASH_FIND, NULL); \
     238             :     if (hentry) \
     239             :         RELATION = hentry->reldesc; \
     240             :     else \
     241             :         RELATION = NULL; \
     242             : } while(0)
     243             : 
     244             : #define RelationCacheDelete(RELATION) \
     245             : do { \
     246             :     RelIdCacheEnt *hentry; \
     247             :     hentry = (RelIdCacheEnt *) hash_search(RelationIdCache, \
     248             :                                            &((RELATION)->rd_id), \
     249             :                                            HASH_REMOVE, NULL); \
     250             :     if (hentry == NULL) \
     251             :         elog(WARNING, "failed to delete relcache entry for OID %u", \
     252             :              (RELATION)->rd_id); \
     253             : } while(0)
     254             : 
     255             : 
     256             : /*
     257             :  * Special cache for opclass-related information
     258             :  *
     259             :  * Note: only default support procs get cached, ie, those with
     260             :  * lefttype = righttype = opcintype.
     261             :  */
     262             : typedef struct opclasscacheent
     263             : {
     264             :     Oid         opclassoid;     /* lookup key: OID of opclass */
     265             :     bool        valid;          /* set true after successful fill-in */
     266             :     StrategyNumber numSupport;  /* max # of support procs (from pg_am) */
     267             :     Oid         opcfamily;      /* OID of opclass's family */
     268             :     Oid         opcintype;      /* OID of opclass's declared input type */
     269             :     RegProcedure *supportProcs; /* OIDs of support procedures */
     270             : } OpClassCacheEnt;
     271             : 
     272             : static HTAB *OpClassCache = NULL;
     273             : 
     274             : 
     275             : /* non-export function prototypes */
     276             : 
     277             : static void RelationCloseCleanup(Relation relation);
     278             : static void RelationDestroyRelation(Relation relation, bool remember_tupdesc);
     279             : static void RelationClearRelation(Relation relation, bool rebuild);
     280             : 
     281             : static void RelationReloadIndexInfo(Relation relation);
     282             : static void RelationReloadNailed(Relation relation);
     283             : static void RelationFlushRelation(Relation relation);
     284             : static void RememberToFreeTupleDescAtEOX(TupleDesc td);
     285             : #ifdef USE_ASSERT_CHECKING
     286             : static void AssertPendingSyncConsistency(Relation relation);
     287             : #endif
     288             : static void AtEOXact_cleanup(Relation relation, bool isCommit);
     289             : static void AtEOSubXact_cleanup(Relation relation, bool isCommit,
     290             :                                 SubTransactionId mySubid, SubTransactionId parentSubid);
     291             : static bool load_relcache_init_file(bool shared);
     292             : static void write_relcache_init_file(bool shared);
     293             : static void write_item(const void *data, Size len, FILE *fp);
     294             : 
     295             : static void formrdesc(const char *relationName, Oid relationReltype,
     296             :                       bool isshared, int natts, const FormData_pg_attribute *attrs);
     297             : 
     298             : static HeapTuple ScanPgRelation(Oid targetRelId, bool indexOK, bool force_non_historic);
     299             : static Relation AllocateRelationDesc(Form_pg_class relp);
     300             : static void RelationParseRelOptions(Relation relation, HeapTuple tuple);
     301             : static void RelationBuildTupleDesc(Relation relation);
     302             : static Relation RelationBuildDesc(Oid targetRelId, bool insertIt);
     303             : static void RelationInitPhysicalAddr(Relation relation);
     304             : static void load_critical_index(Oid indexoid, Oid heapoid);
     305             : static TupleDesc GetPgClassDescriptor(void);
     306             : static TupleDesc GetPgIndexDescriptor(void);
     307             : static void AttrDefaultFetch(Relation relation, int ndef);
     308             : static int  AttrDefaultCmp(const void *a, const void *b);
     309             : static void CheckConstraintFetch(Relation relation);
     310             : static int  CheckConstraintCmp(const void *a, const void *b);
     311             : static void InitIndexAmRoutine(Relation relation);
     312             : static void IndexSupportInitialize(oidvector *indclass,
     313             :                                    RegProcedure *indexSupport,
     314             :                                    Oid *opFamily,
     315             :                                    Oid *opcInType,
     316             :                                    StrategyNumber maxSupportNumber,
     317             :                                    AttrNumber maxAttributeNumber);
     318             : static OpClassCacheEnt *LookupOpclassInfo(Oid operatorClassOid,
     319             :                                           StrategyNumber numSupport);
     320             : static void RelationCacheInitFileRemoveInDir(const char *tblspcpath);
     321             : static void unlink_initfile(const char *initfilename, int elevel);
     322             : 
     323             : 
     324             : /*
     325             :  *      ScanPgRelation
     326             :  *
     327             :  *      This is used by RelationBuildDesc to find a pg_class
     328             :  *      tuple matching targetRelId.  The caller must hold at least
     329             :  *      AccessShareLock on the target relid to prevent concurrent-update
     330             :  *      scenarios; it isn't guaranteed that all scans used to build the
     331             :  *      relcache entry will use the same snapshot.  If, for example,
     332             :  *      an attribute were to be added after scanning pg_class and before
     333             :  *      scanning pg_attribute, relnatts wouldn't match.
     334             :  *
     335             :  *      NB: the returned tuple has been copied into palloc'd storage
     336             :  *      and must eventually be freed with heap_freetuple.
     337             :  */
     338             : static HeapTuple
     339     1032028 : ScanPgRelation(Oid targetRelId, bool indexOK, bool force_non_historic)
     340             : {
     341             :     HeapTuple   pg_class_tuple;
     342             :     Relation    pg_class_desc;
     343             :     SysScanDesc pg_class_scan;
     344             :     ScanKeyData key[1];
     345     1032028 :     Snapshot    snapshot = NULL;
     346             : 
     347             :     /*
     348             :      * If something goes wrong during backend startup, we might find ourselves
     349             :      * trying to read pg_class before we've selected a database.  That ain't
     350             :      * gonna work, so bail out with a useful error message.  If this happens,
     351             :      * it probably means a relcache entry that needs to be nailed isn't.
     352             :      */
     353     1032028 :     if (!OidIsValid(MyDatabaseId))
     354           0 :         elog(FATAL, "cannot read pg_class without having selected a database");
     355             : 
     356             :     /*
     357             :      * form a scan key
     358             :      */
     359     1032028 :     ScanKeyInit(&key[0],
     360             :                 Anum_pg_class_oid,
     361             :                 BTEqualStrategyNumber, F_OIDEQ,
     362             :                 ObjectIdGetDatum(targetRelId));
     363             : 
     364             :     /*
     365             :      * Open pg_class and fetch a tuple.  Force heap scan if we haven't yet
     366             :      * built the critical relcache entries (this includes initdb and startup
     367             :      * without a pg_internal.init file).  The caller can also force a heap
     368             :      * scan by setting indexOK == false.
     369             :      */
     370     1032028 :     pg_class_desc = table_open(RelationRelationId, AccessShareLock);
     371             : 
     372             :     /*
     373             :      * The caller might need a tuple that's newer than the one the historic
     374             :      * snapshot; currently the only case requiring to do so is looking up the
     375             :      * relfilenumber of non mapped system relations during decoding. That
     376             :      * snapshot can't change in the midst of a relcache build, so there's no
     377             :      * need to register the snapshot.
     378             :      */
     379     1032028 :     if (force_non_historic)
     380        2638 :         snapshot = GetNonHistoricCatalogSnapshot(RelationRelationId);
     381             : 
     382     1032028 :     pg_class_scan = systable_beginscan(pg_class_desc, ClassOidIndexId,
     383     1032028 :                                        indexOK && criticalRelcachesBuilt,
     384             :                                        snapshot,
     385             :                                        1, key);
     386             : 
     387     1032022 :     pg_class_tuple = systable_getnext(pg_class_scan);
     388             : 
     389             :     /*
     390             :      * Must copy tuple before releasing buffer.
     391             :      */
     392     1032016 :     if (HeapTupleIsValid(pg_class_tuple))
     393     1032006 :         pg_class_tuple = heap_copytuple(pg_class_tuple);
     394             : 
     395             :     /* all done */
     396     1032016 :     systable_endscan(pg_class_scan);
     397     1032016 :     table_close(pg_class_desc, AccessShareLock);
     398             : 
     399     1032016 :     return pg_class_tuple;
     400             : }
     401             : 
     402             : /*
     403             :  *      AllocateRelationDesc
     404             :  *
     405             :  *      This is used to allocate memory for a new relation descriptor
     406             :  *      and initialize the rd_rel field from the given pg_class tuple.
     407             :  */
     408             : static Relation
     409      949026 : AllocateRelationDesc(Form_pg_class relp)
     410             : {
     411             :     Relation    relation;
     412             :     MemoryContext oldcxt;
     413             :     Form_pg_class relationForm;
     414             : 
     415             :     /* Relcache entries must live in CacheMemoryContext */
     416      949026 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
     417             : 
     418             :     /*
     419             :      * allocate and zero space for new relation descriptor
     420             :      */
     421      949026 :     relation = (Relation) palloc0(sizeof(RelationData));
     422             : 
     423             :     /* make sure relation is marked as having no open file yet */
     424      949026 :     relation->rd_smgr = NULL;
     425             : 
     426             :     /*
     427             :      * Copy the relation tuple form
     428             :      *
     429             :      * We only allocate space for the fixed fields, ie, CLASS_TUPLE_SIZE. The
     430             :      * variable-length fields (relacl, reloptions) are NOT stored in the
     431             :      * relcache --- there'd be little point in it, since we don't copy the
     432             :      * tuple's nulls bitmap and hence wouldn't know if the values are valid.
     433             :      * Bottom line is that relacl *cannot* be retrieved from the relcache. Get
     434             :      * it from the syscache if you need it.  The same goes for the original
     435             :      * form of reloptions (however, we do store the parsed form of reloptions
     436             :      * in rd_options).
     437             :      */
     438      949026 :     relationForm = (Form_pg_class) palloc(CLASS_TUPLE_SIZE);
     439             : 
     440      949026 :     memcpy(relationForm, relp, CLASS_TUPLE_SIZE);
     441             : 
     442             :     /* initialize relation tuple form */
     443      949026 :     relation->rd_rel = relationForm;
     444             : 
     445             :     /* and allocate attribute tuple form storage */
     446      949026 :     relation->rd_att = CreateTemplateTupleDesc(relationForm->relnatts);
     447             :     /* which we mark as a reference-counted tupdesc */
     448      949026 :     relation->rd_att->tdrefcount = 1;
     449             : 
     450      949026 :     MemoryContextSwitchTo(oldcxt);
     451             : 
     452      949026 :     return relation;
     453             : }
     454             : 
     455             : /*
     456             :  * RelationParseRelOptions
     457             :  *      Convert pg_class.reloptions into pre-parsed rd_options
     458             :  *
     459             :  * tuple is the real pg_class tuple (not rd_rel!) for relation
     460             :  *
     461             :  * Note: rd_rel and (if an index) rd_indam must be valid already
     462             :  */
     463             : static void
     464     1024248 : RelationParseRelOptions(Relation relation, HeapTuple tuple)
     465             : {
     466             :     bytea      *options;
     467             :     amoptions_function amoptsfn;
     468             : 
     469     1024248 :     relation->rd_options = NULL;
     470             : 
     471             :     /*
     472             :      * Look up any AM-specific parse function; fall out if relkind should not
     473             :      * have options.
     474             :      */
     475     1024248 :     switch (relation->rd_rel->relkind)
     476             :     {
     477      584330 :         case RELKIND_RELATION:
     478             :         case RELKIND_TOASTVALUE:
     479             :         case RELKIND_VIEW:
     480             :         case RELKIND_MATVIEW:
     481             :         case RELKIND_PARTITIONED_TABLE:
     482      584330 :             amoptsfn = NULL;
     483      584330 :             break;
     484      430742 :         case RELKIND_INDEX:
     485             :         case RELKIND_PARTITIONED_INDEX:
     486      430742 :             amoptsfn = relation->rd_indam->amoptions;
     487      430742 :             break;
     488        9176 :         default:
     489        9176 :             return;
     490             :     }
     491             : 
     492             :     /*
     493             :      * Fetch reloptions from tuple; have to use a hardwired descriptor because
     494             :      * we might not have any other for pg_class yet (consider executing this
     495             :      * code for pg_class itself)
     496             :      */
     497     1015072 :     options = extractRelOptions(tuple, GetPgClassDescriptor(), amoptsfn);
     498             : 
     499             :     /*
     500             :      * Copy parsed data into CacheMemoryContext.  To guard against the
     501             :      * possibility of leaks in the reloptions code, we want to do the actual
     502             :      * parsing in the caller's memory context and copy the results into
     503             :      * CacheMemoryContext after the fact.
     504             :      */
     505     1015072 :     if (options)
     506             :     {
     507       25460 :         relation->rd_options = MemoryContextAlloc(CacheMemoryContext,
     508       12730 :                                                   VARSIZE(options));
     509       12730 :         memcpy(relation->rd_options, options, VARSIZE(options));
     510       12730 :         pfree(options);
     511             :     }
     512             : }
     513             : 
     514             : /*
     515             :  *      RelationBuildTupleDesc
     516             :  *
     517             :  *      Form the relation's tuple descriptor from information in
     518             :  *      the pg_attribute, pg_attrdef & pg_constraint system catalogs.
     519             :  */
     520             : static void
     521      949026 : RelationBuildTupleDesc(Relation relation)
     522             : {
     523             :     HeapTuple   pg_attribute_tuple;
     524             :     Relation    pg_attribute_desc;
     525             :     SysScanDesc pg_attribute_scan;
     526             :     ScanKeyData skey[2];
     527             :     int         need;
     528             :     TupleConstr *constr;
     529      949026 :     AttrMissing *attrmiss = NULL;
     530      949026 :     int         ndef = 0;
     531             : 
     532             :     /* fill rd_att's type ID fields (compare heap.c's AddNewRelationTuple) */
     533      949026 :     relation->rd_att->tdtypeid =
     534      949026 :         relation->rd_rel->reltype ? relation->rd_rel->reltype : RECORDOID;
     535      949026 :     relation->rd_att->tdtypmod = -1;  /* just to be sure */
     536             : 
     537      949026 :     constr = (TupleConstr *) MemoryContextAllocZero(CacheMemoryContext,
     538             :                                                     sizeof(TupleConstr));
     539      949026 :     constr->has_not_null = false;
     540      949026 :     constr->has_generated_stored = false;
     541             : 
     542             :     /*
     543             :      * Form a scan key that selects only user attributes (attnum > 0).
     544             :      * (Eliminating system attribute rows at the index level is lots faster
     545             :      * than fetching them.)
     546             :      */
     547      949026 :     ScanKeyInit(&skey[0],
     548             :                 Anum_pg_attribute_attrelid,
     549             :                 BTEqualStrategyNumber, F_OIDEQ,
     550             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
     551      949026 :     ScanKeyInit(&skey[1],
     552             :                 Anum_pg_attribute_attnum,
     553             :                 BTGreaterStrategyNumber, F_INT2GT,
     554             :                 Int16GetDatum(0));
     555             : 
     556             :     /*
     557             :      * Open pg_attribute and begin a scan.  Force heap scan if we haven't yet
     558             :      * built the critical relcache entries (this includes initdb and startup
     559             :      * without a pg_internal.init file).
     560             :      */
     561      949026 :     pg_attribute_desc = table_open(AttributeRelationId, AccessShareLock);
     562      949026 :     pg_attribute_scan = systable_beginscan(pg_attribute_desc,
     563             :                                            AttributeRelidNumIndexId,
     564             :                                            criticalRelcachesBuilt,
     565             :                                            NULL,
     566             :                                            2, skey);
     567             : 
     568             :     /*
     569             :      * add attribute data to relation->rd_att
     570             :      */
     571      949026 :     need = RelationGetNumberOfAttributes(relation);
     572             : 
     573     3234088 :     while (HeapTupleIsValid(pg_attribute_tuple = systable_getnext(pg_attribute_scan)))
     574             :     {
     575             :         Form_pg_attribute attp;
     576             :         int         attnum;
     577             : 
     578     3233062 :         attp = (Form_pg_attribute) GETSTRUCT(pg_attribute_tuple);
     579             : 
     580     3233062 :         attnum = attp->attnum;
     581     3233062 :         if (attnum <= 0 || attnum > RelationGetNumberOfAttributes(relation))
     582           0 :             elog(ERROR, "invalid attribute number %d for relation \"%s\"",
     583             :                  attp->attnum, RelationGetRelationName(relation));
     584             : 
     585     3233062 :         memcpy(TupleDescAttr(relation->rd_att, attnum - 1),
     586             :                attp,
     587             :                ATTRIBUTE_FIXED_PART_SIZE);
     588             : 
     589             :         /* Update constraint/default info */
     590     3233062 :         if (attp->attnotnull)
     591     1307578 :             constr->has_not_null = true;
     592     3233062 :         if (attp->attgenerated == ATTRIBUTE_GENERATED_STORED)
     593        7040 :             constr->has_generated_stored = true;
     594     3233062 :         if (attp->atthasdef)
     595       35054 :             ndef++;
     596             : 
     597             :         /* If the column has a "missing" value, put it in the attrmiss array */
     598     3233062 :         if (attp->atthasmissing)
     599             :         {
     600             :             Datum       missingval;
     601             :             bool        missingNull;
     602             : 
     603             :             /* Do we have a missing value? */
     604        7056 :             missingval = heap_getattr(pg_attribute_tuple,
     605             :                                       Anum_pg_attribute_attmissingval,
     606             :                                       pg_attribute_desc->rd_att,
     607             :                                       &missingNull);
     608        7056 :             if (!missingNull)
     609             :             {
     610             :                 /* Yes, fetch from the array */
     611             :                 MemoryContext oldcxt;
     612             :                 bool        is_null;
     613        7056 :                 int         one = 1;
     614             :                 Datum       missval;
     615             : 
     616        7056 :                 if (attrmiss == NULL)
     617             :                     attrmiss = (AttrMissing *)
     618        3252 :                         MemoryContextAllocZero(CacheMemoryContext,
     619        3252 :                                                relation->rd_rel->relnatts *
     620             :                                                sizeof(AttrMissing));
     621             : 
     622        7056 :                 missval = array_get_element(missingval,
     623             :                                             1,
     624             :                                             &one,
     625             :                                             -1,
     626        7056 :                                             attp->attlen,
     627        7056 :                                             attp->attbyval,
     628        7056 :                                             attp->attalign,
     629             :                                             &is_null);
     630             :                 Assert(!is_null);
     631        7056 :                 if (attp->attbyval)
     632             :                 {
     633             :                     /* for copy by val just copy the datum direct */
     634        4364 :                     attrmiss[attnum - 1].am_value = missval;
     635             :                 }
     636             :                 else
     637             :                 {
     638             :                     /* otherwise copy in the correct context */
     639        2692 :                     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
     640        5384 :                     attrmiss[attnum - 1].am_value = datumCopy(missval,
     641        2692 :                                                               attp->attbyval,
     642        2692 :                                                               attp->attlen);
     643        2692 :                     MemoryContextSwitchTo(oldcxt);
     644             :                 }
     645        7056 :                 attrmiss[attnum - 1].am_present = true;
     646             :             }
     647             :         }
     648     3233062 :         need--;
     649     3233062 :         if (need == 0)
     650      948000 :             break;
     651             :     }
     652             : 
     653             :     /*
     654             :      * end the scan and close the attribute relation
     655             :      */
     656      949024 :     systable_endscan(pg_attribute_scan);
     657      949024 :     table_close(pg_attribute_desc, AccessShareLock);
     658             : 
     659      949024 :     if (need != 0)
     660           0 :         elog(ERROR, "pg_attribute catalog is missing %d attribute(s) for relation OID %u",
     661             :              need, RelationGetRelid(relation));
     662             : 
     663             :     /*
     664             :      * The attcacheoff values we read from pg_attribute should all be -1
     665             :      * ("unknown").  Verify this if assert checking is on.  They will be
     666             :      * computed when and if needed during tuple access.
     667             :      */
     668             : #ifdef USE_ASSERT_CHECKING
     669             :     {
     670             :         int         i;
     671             : 
     672             :         for (i = 0; i < RelationGetNumberOfAttributes(relation); i++)
     673             :             Assert(TupleDescAttr(relation->rd_att, i)->attcacheoff == -1);
     674             :     }
     675             : #endif
     676             : 
     677             :     /*
     678             :      * However, we can easily set the attcacheoff value for the first
     679             :      * attribute: it must be zero.  This eliminates the need for special cases
     680             :      * for attnum=1 that used to exist in fastgetattr() and index_getattr().
     681             :      */
     682      949024 :     if (RelationGetNumberOfAttributes(relation) > 0)
     683      948000 :         TupleDescAttr(relation->rd_att, 0)->attcacheoff = 0;
     684             : 
     685             :     /*
     686             :      * Set up constraint/default info
     687             :      */
     688      949024 :     if (constr->has_not_null ||
     689      659102 :         constr->has_generated_stored ||
     690      651774 :         ndef > 0 ||
     691      651750 :         attrmiss ||
     692      651750 :         relation->rd_rel->relchecks > 0)
     693             :     {
     694      301748 :         relation->rd_att->constr = constr;
     695             : 
     696      301748 :         if (ndef > 0)            /* DEFAULTs */
     697       25032 :             AttrDefaultFetch(relation, ndef);
     698             :         else
     699      276716 :             constr->num_defval = 0;
     700             : 
     701      301748 :         constr->missing = attrmiss;
     702             : 
     703      301748 :         if (relation->rd_rel->relchecks > 0)   /* CHECKs */
     704       10088 :             CheckConstraintFetch(relation);
     705             :         else
     706      291660 :             constr->num_check = 0;
     707             :     }
     708             :     else
     709             :     {
     710      647276 :         pfree(constr);
     711      647276 :         relation->rd_att->constr = NULL;
     712             :     }
     713      949024 : }
     714             : 
     715             : /*
     716             :  *      RelationBuildRuleLock
     717             :  *
     718             :  *      Form the relation's rewrite rules from information in
     719             :  *      the pg_rewrite system catalog.
     720             :  *
     721             :  * Note: The rule parsetrees are potentially very complex node structures.
     722             :  * To allow these trees to be freed when the relcache entry is flushed,
     723             :  * we make a private memory context to hold the RuleLock information for
     724             :  * each relcache entry that has associated rules.  The context is used
     725             :  * just for rule info, not for any other subsidiary data of the relcache
     726             :  * entry, because that keeps the update logic in RelationClearRelation()
     727             :  * manageable.  The other subsidiary data structures are simple enough
     728             :  * to be easy to free explicitly, anyway.
     729             :  *
     730             :  * Note: The relation's reloptions must have been extracted first.
     731             :  */
     732             : static void
     733       28078 : RelationBuildRuleLock(Relation relation)
     734             : {
     735             :     MemoryContext rulescxt;
     736             :     MemoryContext oldcxt;
     737             :     HeapTuple   rewrite_tuple;
     738             :     Relation    rewrite_desc;
     739             :     TupleDesc   rewrite_tupdesc;
     740             :     SysScanDesc rewrite_scan;
     741             :     ScanKeyData key;
     742             :     RuleLock   *rulelock;
     743             :     int         numlocks;
     744             :     RewriteRule **rules;
     745             :     int         maxlocks;
     746             : 
     747             :     /*
     748             :      * Make the private context.  Assume it'll not contain much data.
     749             :      */
     750       28078 :     rulescxt = AllocSetContextCreate(CacheMemoryContext,
     751             :                                      "relation rules",
     752             :                                      ALLOCSET_SMALL_SIZES);
     753       28078 :     relation->rd_rulescxt = rulescxt;
     754       28078 :     MemoryContextCopyAndSetIdentifier(rulescxt,
     755             :                                       RelationGetRelationName(relation));
     756             : 
     757             :     /*
     758             :      * allocate an array to hold the rewrite rules (the array is extended if
     759             :      * necessary)
     760             :      */
     761       28078 :     maxlocks = 4;
     762             :     rules = (RewriteRule **)
     763       28078 :         MemoryContextAlloc(rulescxt, sizeof(RewriteRule *) * maxlocks);
     764       28078 :     numlocks = 0;
     765             : 
     766             :     /*
     767             :      * form a scan key
     768             :      */
     769       28078 :     ScanKeyInit(&key,
     770             :                 Anum_pg_rewrite_ev_class,
     771             :                 BTEqualStrategyNumber, F_OIDEQ,
     772             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
     773             : 
     774             :     /*
     775             :      * open pg_rewrite and begin a scan
     776             :      *
     777             :      * Note: since we scan the rules using RewriteRelRulenameIndexId, we will
     778             :      * be reading the rules in name order, except possibly during
     779             :      * emergency-recovery operations (ie, IgnoreSystemIndexes). This in turn
     780             :      * ensures that rules will be fired in name order.
     781             :      */
     782       28078 :     rewrite_desc = table_open(RewriteRelationId, AccessShareLock);
     783       28078 :     rewrite_tupdesc = RelationGetDescr(rewrite_desc);
     784       28078 :     rewrite_scan = systable_beginscan(rewrite_desc,
     785             :                                       RewriteRelRulenameIndexId,
     786             :                                       true, NULL,
     787             :                                       1, &key);
     788             : 
     789       54768 :     while (HeapTupleIsValid(rewrite_tuple = systable_getnext(rewrite_scan)))
     790             :     {
     791       26690 :         Form_pg_rewrite rewrite_form = (Form_pg_rewrite) GETSTRUCT(rewrite_tuple);
     792             :         bool        isnull;
     793             :         Datum       rule_datum;
     794             :         char       *rule_str;
     795             :         RewriteRule *rule;
     796             :         Oid         check_as_user;
     797             : 
     798       26690 :         rule = (RewriteRule *) MemoryContextAlloc(rulescxt,
     799             :                                                   sizeof(RewriteRule));
     800             : 
     801       26690 :         rule->ruleId = rewrite_form->oid;
     802             : 
     803       26690 :         rule->event = rewrite_form->ev_type - '0';
     804       26690 :         rule->enabled = rewrite_form->ev_enabled;
     805       26690 :         rule->isInstead = rewrite_form->is_instead;
     806             : 
     807             :         /*
     808             :          * Must use heap_getattr to fetch ev_action and ev_qual.  Also, the
     809             :          * rule strings are often large enough to be toasted.  To avoid
     810             :          * leaking memory in the caller's context, do the detoasting here so
     811             :          * we can free the detoasted version.
     812             :          */
     813       26690 :         rule_datum = heap_getattr(rewrite_tuple,
     814             :                                   Anum_pg_rewrite_ev_action,
     815             :                                   rewrite_tupdesc,
     816             :                                   &isnull);
     817             :         Assert(!isnull);
     818       26690 :         rule_str = TextDatumGetCString(rule_datum);
     819       26690 :         oldcxt = MemoryContextSwitchTo(rulescxt);
     820       26690 :         rule->actions = (List *) stringToNode(rule_str);
     821       26690 :         MemoryContextSwitchTo(oldcxt);
     822       26690 :         pfree(rule_str);
     823             : 
     824       26690 :         rule_datum = heap_getattr(rewrite_tuple,
     825             :                                   Anum_pg_rewrite_ev_qual,
     826             :                                   rewrite_tupdesc,
     827             :                                   &isnull);
     828             :         Assert(!isnull);
     829       26690 :         rule_str = TextDatumGetCString(rule_datum);
     830       26690 :         oldcxt = MemoryContextSwitchTo(rulescxt);
     831       26690 :         rule->qual = (Node *) stringToNode(rule_str);
     832       26690 :         MemoryContextSwitchTo(oldcxt);
     833       26690 :         pfree(rule_str);
     834             : 
     835             :         /*
     836             :          * If this is a SELECT rule defining a view, and the view has
     837             :          * "security_invoker" set, we must perform all permissions checks on
     838             :          * relations referred to by the rule as the invoking user.
     839             :          *
     840             :          * In all other cases (including non-SELECT rules on security invoker
     841             :          * views), perform the permissions checks as the relation owner.
     842             :          */
     843       26690 :         if (rule->event == CMD_SELECT &&
     844       24598 :             relation->rd_rel->relkind == RELKIND_VIEW &&
     845       21430 :             RelationHasSecurityInvoker(relation))
     846         156 :             check_as_user = InvalidOid;
     847             :         else
     848       26534 :             check_as_user = relation->rd_rel->relowner;
     849             : 
     850             :         /*
     851             :          * Scan through the rule's actions and set the checkAsUser field on
     852             :          * all RTEPermissionInfos. We have to look at the qual as well, in
     853             :          * case it contains sublinks.
     854             :          *
     855             :          * The reason for doing this when the rule is loaded, rather than when
     856             :          * it is stored, is that otherwise ALTER TABLE OWNER would have to
     857             :          * grovel through stored rules to update checkAsUser fields. Scanning
     858             :          * the rule tree during load is relatively cheap (compared to
     859             :          * constructing it in the first place), so we do it here.
     860             :          */
     861       26690 :         setRuleCheckAsUser((Node *) rule->actions, check_as_user);
     862       26690 :         setRuleCheckAsUser(rule->qual, check_as_user);
     863             : 
     864       26690 :         if (numlocks >= maxlocks)
     865             :         {
     866          32 :             maxlocks *= 2;
     867             :             rules = (RewriteRule **)
     868          32 :                 repalloc(rules, sizeof(RewriteRule *) * maxlocks);
     869             :         }
     870       26690 :         rules[numlocks++] = rule;
     871             :     }
     872             : 
     873             :     /*
     874             :      * end the scan and close the attribute relation
     875             :      */
     876       28078 :     systable_endscan(rewrite_scan);
     877       28078 :     table_close(rewrite_desc, AccessShareLock);
     878             : 
     879             :     /*
     880             :      * there might not be any rules (if relhasrules is out-of-date)
     881             :      */
     882       28078 :     if (numlocks == 0)
     883             :     {
     884        2644 :         relation->rd_rules = NULL;
     885        2644 :         relation->rd_rulescxt = NULL;
     886        2644 :         MemoryContextDelete(rulescxt);
     887        2644 :         return;
     888             :     }
     889             : 
     890             :     /*
     891             :      * form a RuleLock and insert into relation
     892             :      */
     893       25434 :     rulelock = (RuleLock *) MemoryContextAlloc(rulescxt, sizeof(RuleLock));
     894       25434 :     rulelock->numLocks = numlocks;
     895       25434 :     rulelock->rules = rules;
     896             : 
     897       25434 :     relation->rd_rules = rulelock;
     898             : }
     899             : 
     900             : /*
     901             :  *      equalRuleLocks
     902             :  *
     903             :  *      Determine whether two RuleLocks are equivalent
     904             :  *
     905             :  *      Probably this should be in the rules code someplace...
     906             :  */
     907             : static bool
     908      251556 : equalRuleLocks(RuleLock *rlock1, RuleLock *rlock2)
     909             : {
     910             :     int         i;
     911             : 
     912             :     /*
     913             :      * As of 7.3 we assume the rule ordering is repeatable, because
     914             :      * RelationBuildRuleLock should read 'em in a consistent order.  So just
     915             :      * compare corresponding slots.
     916             :      */
     917      251556 :     if (rlock1 != NULL)
     918             :     {
     919        1802 :         if (rlock2 == NULL)
     920          36 :             return false;
     921        1766 :         if (rlock1->numLocks != rlock2->numLocks)
     922           6 :             return false;
     923        3264 :         for (i = 0; i < rlock1->numLocks; i++)
     924             :         {
     925        1772 :             RewriteRule *rule1 = rlock1->rules[i];
     926        1772 :             RewriteRule *rule2 = rlock2->rules[i];
     927             : 
     928        1772 :             if (rule1->ruleId != rule2->ruleId)
     929           0 :                 return false;
     930        1772 :             if (rule1->event != rule2->event)
     931           0 :                 return false;
     932        1772 :             if (rule1->enabled != rule2->enabled)
     933          22 :                 return false;
     934        1750 :             if (rule1->isInstead != rule2->isInstead)
     935           0 :                 return false;
     936        1750 :             if (!equal(rule1->qual, rule2->qual))
     937           0 :                 return false;
     938        1750 :             if (!equal(rule1->actions, rule2->actions))
     939         246 :                 return false;
     940             :         }
     941             :     }
     942      249754 :     else if (rlock2 != NULL)
     943       11420 :         return false;
     944      239826 :     return true;
     945             : }
     946             : 
     947             : /*
     948             :  *      equalPolicy
     949             :  *
     950             :  *      Determine whether two policies are equivalent
     951             :  */
     952             : static bool
     953         186 : equalPolicy(RowSecurityPolicy *policy1, RowSecurityPolicy *policy2)
     954             : {
     955             :     int         i;
     956             :     Oid        *r1,
     957             :                *r2;
     958             : 
     959         186 :     if (policy1 != NULL)
     960             :     {
     961         186 :         if (policy2 == NULL)
     962           0 :             return false;
     963             : 
     964         186 :         if (policy1->polcmd != policy2->polcmd)
     965           0 :             return false;
     966         186 :         if (policy1->hassublinks != policy2->hassublinks)
     967           0 :             return false;
     968         186 :         if (strcmp(policy1->policy_name, policy2->policy_name) != 0)
     969           0 :             return false;
     970         186 :         if (ARR_DIMS(policy1->roles)[0] != ARR_DIMS(policy2->roles)[0])
     971           0 :             return false;
     972             : 
     973         186 :         r1 = (Oid *) ARR_DATA_PTR(policy1->roles);
     974         186 :         r2 = (Oid *) ARR_DATA_PTR(policy2->roles);
     975             : 
     976         372 :         for (i = 0; i < ARR_DIMS(policy1->roles)[0]; i++)
     977             :         {
     978         186 :             if (r1[i] != r2[i])
     979           0 :                 return false;
     980             :         }
     981             : 
     982         186 :         if (!equal(policy1->qual, policy2->qual))
     983           0 :             return false;
     984         186 :         if (!equal(policy1->with_check_qual, policy2->with_check_qual))
     985           0 :             return false;
     986             :     }
     987           0 :     else if (policy2 != NULL)
     988           0 :         return false;
     989             : 
     990         186 :     return true;
     991             : }
     992             : 
     993             : /*
     994             :  *      equalRSDesc
     995             :  *
     996             :  *      Determine whether two RowSecurityDesc's are equivalent
     997             :  */
     998             : static bool
     999      251556 : equalRSDesc(RowSecurityDesc *rsdesc1, RowSecurityDesc *rsdesc2)
    1000             : {
    1001             :     ListCell   *lc,
    1002             :                *rc;
    1003             : 
    1004      251556 :     if (rsdesc1 == NULL && rsdesc2 == NULL)
    1005      251128 :         return true;
    1006             : 
    1007         428 :     if ((rsdesc1 != NULL && rsdesc2 == NULL) ||
    1008         278 :         (rsdesc1 == NULL && rsdesc2 != NULL))
    1009         288 :         return false;
    1010             : 
    1011         140 :     if (list_length(rsdesc1->policies) != list_length(rsdesc2->policies))
    1012           0 :         return false;
    1013             : 
    1014             :     /* RelationBuildRowSecurity should build policies in order */
    1015         326 :     forboth(lc, rsdesc1->policies, rc, rsdesc2->policies)
    1016             :     {
    1017         186 :         RowSecurityPolicy *l = (RowSecurityPolicy *) lfirst(lc);
    1018         186 :         RowSecurityPolicy *r = (RowSecurityPolicy *) lfirst(rc);
    1019             : 
    1020         186 :         if (!equalPolicy(l, r))
    1021           0 :             return false;
    1022             :     }
    1023             : 
    1024         140 :     return true;
    1025             : }
    1026             : 
    1027             : /*
    1028             :  *      RelationBuildDesc
    1029             :  *
    1030             :  *      Build a relation descriptor.  The caller must hold at least
    1031             :  *      AccessShareLock on the target relid.
    1032             :  *
    1033             :  *      The new descriptor is inserted into the hash table if insertIt is true.
    1034             :  *
    1035             :  *      Returns NULL if no pg_class row could be found for the given relid
    1036             :  *      (suggesting we are trying to access a just-deleted relation).
    1037             :  *      Any other error is reported via elog.
    1038             :  */
    1039             : static Relation
    1040      949030 : RelationBuildDesc(Oid targetRelId, bool insertIt)
    1041             : {
    1042             :     int         in_progress_offset;
    1043             :     Relation    relation;
    1044             :     Oid         relid;
    1045             :     HeapTuple   pg_class_tuple;
    1046             :     Form_pg_class relp;
    1047             : 
    1048             :     /*
    1049             :      * This function and its subroutines can allocate a good deal of transient
    1050             :      * data in CurrentMemoryContext.  Traditionally we've just leaked that
    1051             :      * data, reasoning that the caller's context is at worst of transaction
    1052             :      * scope, and relcache loads shouldn't happen so often that it's essential
    1053             :      * to recover transient data before end of statement/transaction.  However
    1054             :      * that's definitely not true when debug_discard_caches is active, and
    1055             :      * perhaps it's not true in other cases.
    1056             :      *
    1057             :      * When debug_discard_caches is active or when forced to by
    1058             :      * RECOVER_RELATION_BUILD_MEMORY=1, arrange to allocate the junk in a
    1059             :      * temporary context that we'll free before returning.  Make it a child of
    1060             :      * caller's context so that it will get cleaned up appropriately if we
    1061             :      * error out partway through.
    1062             :      */
    1063             : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
    1064             :     MemoryContext tmpcxt = NULL;
    1065             :     MemoryContext oldcxt = NULL;
    1066             : 
    1067             :     if (RECOVER_RELATION_BUILD_MEMORY || debug_discard_caches > 0)
    1068             :     {
    1069             :         tmpcxt = AllocSetContextCreate(CurrentMemoryContext,
    1070             :                                        "RelationBuildDesc workspace",
    1071             :                                        ALLOCSET_DEFAULT_SIZES);
    1072             :         oldcxt = MemoryContextSwitchTo(tmpcxt);
    1073             :     }
    1074             : #endif
    1075             : 
    1076             :     /* Register to catch invalidation messages */
    1077      949030 :     if (in_progress_list_len >= in_progress_list_maxlen)
    1078             :     {
    1079             :         int         allocsize;
    1080             : 
    1081           0 :         allocsize = in_progress_list_maxlen * 2;
    1082           0 :         in_progress_list = repalloc(in_progress_list,
    1083             :                                     allocsize * sizeof(*in_progress_list));
    1084           0 :         in_progress_list_maxlen = allocsize;
    1085             :     }
    1086      949030 :     in_progress_offset = in_progress_list_len++;
    1087      949030 :     in_progress_list[in_progress_offset].reloid = targetRelId;
    1088      949036 : retry:
    1089      949036 :     in_progress_list[in_progress_offset].invalidated = false;
    1090             : 
    1091             :     /*
    1092             :      * find the tuple in pg_class corresponding to the given relation id
    1093             :      */
    1094      949036 :     pg_class_tuple = ScanPgRelation(targetRelId, true, false);
    1095             : 
    1096             :     /*
    1097             :      * if no such tuple exists, return NULL
    1098             :      */
    1099      949036 :     if (!HeapTupleIsValid(pg_class_tuple))
    1100             :     {
    1101             : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
    1102             :         if (tmpcxt)
    1103             :         {
    1104             :             /* Return to caller's context, and blow away the temporary context */
    1105             :             MemoryContextSwitchTo(oldcxt);
    1106             :             MemoryContextDelete(tmpcxt);
    1107             :         }
    1108             : #endif
    1109             :         Assert(in_progress_offset + 1 == in_progress_list_len);
    1110          10 :         in_progress_list_len--;
    1111          10 :         return NULL;
    1112             :     }
    1113             : 
    1114             :     /*
    1115             :      * get information from the pg_class_tuple
    1116             :      */
    1117      949026 :     relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
    1118      949026 :     relid = relp->oid;
    1119             :     Assert(relid == targetRelId);
    1120             : 
    1121             :     /*
    1122             :      * allocate storage for the relation descriptor, and copy pg_class_tuple
    1123             :      * to relation->rd_rel.
    1124             :      */
    1125      949026 :     relation = AllocateRelationDesc(relp);
    1126             : 
    1127             :     /*
    1128             :      * initialize the relation's relation id (relation->rd_id)
    1129             :      */
    1130      949026 :     RelationGetRelid(relation) = relid;
    1131             : 
    1132             :     /*
    1133             :      * Normal relations are not nailed into the cache.  Since we don't flush
    1134             :      * new relations, it won't be new.  It could be temp though.
    1135             :      */
    1136      949026 :     relation->rd_refcnt = 0;
    1137      949026 :     relation->rd_isnailed = false;
    1138      949026 :     relation->rd_createSubid = InvalidSubTransactionId;
    1139      949026 :     relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    1140      949026 :     relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    1141      949026 :     relation->rd_droppedSubid = InvalidSubTransactionId;
    1142      949026 :     switch (relation->rd_rel->relpersistence)
    1143             :     {
    1144      925180 :         case RELPERSISTENCE_UNLOGGED:
    1145             :         case RELPERSISTENCE_PERMANENT:
    1146      925180 :             relation->rd_backend = InvalidBackendId;
    1147      925180 :             relation->rd_islocaltemp = false;
    1148      925180 :             break;
    1149       23846 :         case RELPERSISTENCE_TEMP:
    1150       23846 :             if (isTempOrTempToastNamespace(relation->rd_rel->relnamespace))
    1151             :             {
    1152       23812 :                 relation->rd_backend = BackendIdForTempRelations();
    1153       23812 :                 relation->rd_islocaltemp = true;
    1154             :             }
    1155             :             else
    1156             :             {
    1157             :                 /*
    1158             :                  * If it's a temp table, but not one of ours, we have to use
    1159             :                  * the slow, grotty method to figure out the owning backend.
    1160             :                  *
    1161             :                  * Note: it's possible that rd_backend gets set to MyBackendId
    1162             :                  * here, in case we are looking at a pg_class entry left over
    1163             :                  * from a crashed backend that coincidentally had the same
    1164             :                  * BackendId we're using.  We should *not* consider such a
    1165             :                  * table to be "ours"; this is why we need the separate
    1166             :                  * rd_islocaltemp flag.  The pg_class entry will get flushed
    1167             :                  * if/when we clean out the corresponding temp table namespace
    1168             :                  * in preparation for using it.
    1169             :                  */
    1170          34 :                 relation->rd_backend =
    1171          34 :                     GetTempNamespaceBackendId(relation->rd_rel->relnamespace);
    1172             :                 Assert(relation->rd_backend != InvalidBackendId);
    1173          34 :                 relation->rd_islocaltemp = false;
    1174             :             }
    1175       23846 :             break;
    1176           0 :         default:
    1177           0 :             elog(ERROR, "invalid relpersistence: %c",
    1178             :                  relation->rd_rel->relpersistence);
    1179             :             break;
    1180             :     }
    1181             : 
    1182             :     /*
    1183             :      * initialize the tuple descriptor (relation->rd_att).
    1184             :      */
    1185      949026 :     RelationBuildTupleDesc(relation);
    1186             : 
    1187             :     /* foreign key data is not loaded till asked for */
    1188      949024 :     relation->rd_fkeylist = NIL;
    1189      949024 :     relation->rd_fkeyvalid = false;
    1190             : 
    1191             :     /* partitioning data is not loaded till asked for */
    1192      949024 :     relation->rd_partkey = NULL;
    1193      949024 :     relation->rd_partkeycxt = NULL;
    1194      949024 :     relation->rd_partdesc = NULL;
    1195      949024 :     relation->rd_partdesc_nodetached = NULL;
    1196      949024 :     relation->rd_partdesc_nodetached_xmin = InvalidTransactionId;
    1197      949024 :     relation->rd_pdcxt = NULL;
    1198      949024 :     relation->rd_pddcxt = NULL;
    1199      949024 :     relation->rd_partcheck = NIL;
    1200      949024 :     relation->rd_partcheckvalid = false;
    1201      949024 :     relation->rd_partcheckcxt = NULL;
    1202             : 
    1203             :     /*
    1204             :      * initialize access method information
    1205             :      */
    1206      949024 :     if (relation->rd_rel->relkind == RELKIND_INDEX ||
    1207      584550 :         relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
    1208      370076 :         RelationInitIndexAccessInfo(relation);
    1209      578948 :     else if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) ||
    1210       91818 :              relation->rd_rel->relkind == RELKIND_SEQUENCE)
    1211      492100 :         RelationInitTableAccessMethod(relation);
    1212             :     else
    1213             :         Assert(relation->rd_rel->relam == InvalidOid);
    1214             : 
    1215             :     /* extract reloptions if any */
    1216      949016 :     RelationParseRelOptions(relation, pg_class_tuple);
    1217             : 
    1218             :     /*
    1219             :      * Fetch rules and triggers that affect this relation.
    1220             :      *
    1221             :      * Note that RelationBuildRuleLock() relies on this being done after
    1222             :      * extracting the relation's reloptions.
    1223             :      */
    1224      949016 :     if (relation->rd_rel->relhasrules)
    1225       28078 :         RelationBuildRuleLock(relation);
    1226             :     else
    1227             :     {
    1228      920938 :         relation->rd_rules = NULL;
    1229      920938 :         relation->rd_rulescxt = NULL;
    1230             :     }
    1231             : 
    1232      949016 :     if (relation->rd_rel->relhastriggers)
    1233       50588 :         RelationBuildTriggers(relation);
    1234             :     else
    1235      898428 :         relation->trigdesc = NULL;
    1236             : 
    1237      949016 :     if (relation->rd_rel->relrowsecurity)
    1238        1926 :         RelationBuildRowSecurity(relation);
    1239             :     else
    1240      947090 :         relation->rd_rsdesc = NULL;
    1241             : 
    1242             :     /*
    1243             :      * initialize the relation lock manager information
    1244             :      */
    1245      949016 :     RelationInitLockInfo(relation); /* see lmgr.c */
    1246             : 
    1247             :     /*
    1248             :      * initialize physical addressing information for the relation
    1249             :      */
    1250      949016 :     RelationInitPhysicalAddr(relation);
    1251             : 
    1252             :     /* make sure relation is marked as having no open file yet */
    1253      949016 :     relation->rd_smgr = NULL;
    1254             : 
    1255             :     /*
    1256             :      * now we can free the memory allocated for pg_class_tuple
    1257             :      */
    1258      949016 :     heap_freetuple(pg_class_tuple);
    1259             : 
    1260             :     /*
    1261             :      * If an invalidation arrived mid-build, start over.  Between here and the
    1262             :      * end of this function, don't add code that does or reasonably could read
    1263             :      * system catalogs.  That range must be free from invalidation processing
    1264             :      * for the !insertIt case.  For the insertIt case, RelationCacheInsert()
    1265             :      * will enroll this relation in ordinary relcache invalidation processing,
    1266             :      */
    1267      949016 :     if (in_progress_list[in_progress_offset].invalidated)
    1268             :     {
    1269           6 :         RelationDestroyRelation(relation, false);
    1270           6 :         goto retry;
    1271             :     }
    1272             :     Assert(in_progress_offset + 1 == in_progress_list_len);
    1273      949010 :     in_progress_list_len--;
    1274             : 
    1275             :     /*
    1276             :      * Insert newly created relation into relcache hash table, if requested.
    1277             :      *
    1278             :      * There is one scenario in which we might find a hashtable entry already
    1279             :      * present, even though our caller failed to find it: if the relation is a
    1280             :      * system catalog or index that's used during relcache load, we might have
    1281             :      * recursively created the same relcache entry during the preceding steps.
    1282             :      * So allow RelationCacheInsert to delete any already-present relcache
    1283             :      * entry for the same OID.  The already-present entry should have refcount
    1284             :      * zero (else somebody forgot to close it); in the event that it doesn't,
    1285             :      * we'll elog a WARNING and leak the already-present entry.
    1286             :      */
    1287      949010 :     if (insertIt)
    1288      697454 :         RelationCacheInsert(relation, true);
    1289             : 
    1290             :     /* It's fully valid */
    1291      949010 :     relation->rd_isvalid = true;
    1292             : 
    1293             : #ifdef MAYBE_RECOVER_RELATION_BUILD_MEMORY
    1294             :     if (tmpcxt)
    1295             :     {
    1296             :         /* Return to caller's context, and blow away the temporary context */
    1297             :         MemoryContextSwitchTo(oldcxt);
    1298             :         MemoryContextDelete(tmpcxt);
    1299             :     }
    1300             : #endif
    1301             : 
    1302      949010 :     return relation;
    1303             : }
    1304             : 
    1305             : /*
    1306             :  * Initialize the physical addressing info (RelFileLocator) for a relcache entry
    1307             :  *
    1308             :  * Note: at the physical level, relations in the pg_global tablespace must
    1309             :  * be treated as shared, even if relisshared isn't set.  Hence we do not
    1310             :  * look at relisshared here.
    1311             :  */
    1312             : static void
    1313     4133394 : RelationInitPhysicalAddr(Relation relation)
    1314             : {
    1315     4133394 :     RelFileNumber oldnumber = relation->rd_locator.relNumber;
    1316             : 
    1317             :     /* these relations kinds never have storage */
    1318     4133394 :     if (!RELKIND_HAS_STORAGE(relation->rd_rel->relkind))
    1319      113674 :         return;
    1320             : 
    1321     4019720 :     if (relation->rd_rel->reltablespace)
    1322      684218 :         relation->rd_locator.spcOid = relation->rd_rel->reltablespace;
    1323             :     else
    1324     3335502 :         relation->rd_locator.spcOid = MyDatabaseTableSpace;
    1325     4019720 :     if (relation->rd_locator.spcOid == GLOBALTABLESPACE_OID)
    1326      681350 :         relation->rd_locator.dbOid = InvalidOid;
    1327             :     else
    1328     3338370 :         relation->rd_locator.dbOid = MyDatabaseId;
    1329             : 
    1330     4019720 :     if (relation->rd_rel->relfilenode)
    1331             :     {
    1332             :         /*
    1333             :          * Even if we are using a decoding snapshot that doesn't represent the
    1334             :          * current state of the catalog we need to make sure the filenode
    1335             :          * points to the current file since the older file will be gone (or
    1336             :          * truncated). The new file will still contain older rows so lookups
    1337             :          * in them will work correctly. This wouldn't work correctly if
    1338             :          * rewrites were allowed to change the schema in an incompatible way,
    1339             :          * but those are prevented both on catalog tables and on user tables
    1340             :          * declared as additional catalog tables.
    1341             :          */
    1342     2905784 :         if (HistoricSnapshotActive()
    1343        3924 :             && RelationIsAccessibleInLogicalDecoding(relation)
    1344        2638 :             && IsTransactionState())
    1345             :         {
    1346             :             HeapTuple   phys_tuple;
    1347             :             Form_pg_class physrel;
    1348             : 
    1349        2638 :             phys_tuple = ScanPgRelation(RelationGetRelid(relation),
    1350        2638 :                                         RelationGetRelid(relation) != ClassOidIndexId,
    1351             :                                         true);
    1352        2638 :             if (!HeapTupleIsValid(phys_tuple))
    1353           0 :                 elog(ERROR, "could not find pg_class entry for %u",
    1354             :                      RelationGetRelid(relation));
    1355        2638 :             physrel = (Form_pg_class) GETSTRUCT(phys_tuple);
    1356             : 
    1357        2638 :             relation->rd_rel->reltablespace = physrel->reltablespace;
    1358        2638 :             relation->rd_rel->relfilenode = physrel->relfilenode;
    1359        2638 :             heap_freetuple(phys_tuple);
    1360             :         }
    1361             : 
    1362     2905784 :         relation->rd_locator.relNumber = relation->rd_rel->relfilenode;
    1363             :     }
    1364             :     else
    1365             :     {
    1366             :         /* Consult the relation mapper */
    1367     1113936 :         relation->rd_locator.relNumber =
    1368     1113936 :             RelationMapOidToFilenumber(relation->rd_id,
    1369     1113936 :                                        relation->rd_rel->relisshared);
    1370     1113936 :         if (!RelFileNumberIsValid(relation->rd_locator.relNumber))
    1371           0 :             elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
    1372             :                  RelationGetRelationName(relation), relation->rd_id);
    1373             :     }
    1374             : 
    1375             :     /*
    1376             :      * For RelationNeedsWAL() to answer correctly on parallel workers, restore
    1377             :      * rd_firstRelfilelocatorSubid.  No subtransactions start or end while in
    1378             :      * parallel mode, so the specific SubTransactionId does not matter.
    1379             :      */
    1380     4019720 :     if (IsParallelWorker() && oldnumber != relation->rd_locator.relNumber)
    1381             :     {
    1382       34672 :         if (RelFileLocatorSkippingWAL(relation->rd_locator))
    1383         300 :             relation->rd_firstRelfilelocatorSubid = TopSubTransactionId;
    1384             :         else
    1385       34372 :             relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    1386             :     }
    1387             : }
    1388             : 
    1389             : /*
    1390             :  * Fill in the IndexAmRoutine for an index relation.
    1391             :  *
    1392             :  * relation's rd_amhandler and rd_indexcxt must be valid already.
    1393             :  */
    1394             : static void
    1395     2093886 : InitIndexAmRoutine(Relation relation)
    1396             : {
    1397             :     IndexAmRoutine *cached,
    1398             :                *tmp;
    1399             : 
    1400             :     /*
    1401             :      * Call the amhandler in current, short-lived memory context, just in case
    1402             :      * it leaks anything (it probably won't, but let's be paranoid).
    1403             :      */
    1404     2093886 :     tmp = GetIndexAmRoutine(relation->rd_amhandler);
    1405             : 
    1406             :     /* OK, now transfer the data into relation's rd_indexcxt. */
    1407     2093886 :     cached = (IndexAmRoutine *) MemoryContextAlloc(relation->rd_indexcxt,
    1408             :                                                    sizeof(IndexAmRoutine));
    1409     2093886 :     memcpy(cached, tmp, sizeof(IndexAmRoutine));
    1410     2093886 :     relation->rd_indam = cached;
    1411             : 
    1412     2093886 :     pfree(tmp);
    1413     2093886 : }
    1414             : 
    1415             : /*
    1416             :  * Initialize index-access-method support data for an index relation
    1417             :  */
    1418             : void
    1419      380316 : RelationInitIndexAccessInfo(Relation relation)
    1420             : {
    1421             :     HeapTuple   tuple;
    1422             :     Form_pg_am  aform;
    1423             :     Datum       indcollDatum;
    1424             :     Datum       indclassDatum;
    1425             :     Datum       indoptionDatum;
    1426             :     bool        isnull;
    1427             :     oidvector  *indcoll;
    1428             :     oidvector  *indclass;
    1429             :     int2vector *indoption;
    1430             :     MemoryContext indexcxt;
    1431             :     MemoryContext oldcontext;
    1432             :     int         indnatts;
    1433             :     int         indnkeyatts;
    1434             :     uint16      amsupport;
    1435             : 
    1436             :     /*
    1437             :      * Make a copy of the pg_index entry for the index.  Since pg_index
    1438             :      * contains variable-length and possibly-null fields, we have to do this
    1439             :      * honestly rather than just treating it as a Form_pg_index struct.
    1440             :      */
    1441      380316 :     tuple = SearchSysCache1(INDEXRELID,
    1442             :                             ObjectIdGetDatum(RelationGetRelid(relation)));
    1443      380316 :     if (!HeapTupleIsValid(tuple))
    1444           0 :         elog(ERROR, "cache lookup failed for index %u",
    1445             :              RelationGetRelid(relation));
    1446      380316 :     oldcontext = MemoryContextSwitchTo(CacheMemoryContext);
    1447      380316 :     relation->rd_indextuple = heap_copytuple(tuple);
    1448      380316 :     relation->rd_index = (Form_pg_index) GETSTRUCT(relation->rd_indextuple);
    1449      380316 :     MemoryContextSwitchTo(oldcontext);
    1450      380316 :     ReleaseSysCache(tuple);
    1451             : 
    1452             :     /*
    1453             :      * Look up the index's access method, save the OID of its handler function
    1454             :      */
    1455             :     Assert(relation->rd_rel->relam != InvalidOid);
    1456      380316 :     tuple = SearchSysCache1(AMOID, ObjectIdGetDatum(relation->rd_rel->relam));
    1457      380314 :     if (!HeapTupleIsValid(tuple))
    1458           0 :         elog(ERROR, "cache lookup failed for access method %u",
    1459             :              relation->rd_rel->relam);
    1460      380314 :     aform = (Form_pg_am) GETSTRUCT(tuple);
    1461      380314 :     relation->rd_amhandler = aform->amhandler;
    1462      380314 :     ReleaseSysCache(tuple);
    1463             : 
    1464      380314 :     indnatts = RelationGetNumberOfAttributes(relation);
    1465      380314 :     if (indnatts != IndexRelationGetNumberOfAttributes(relation))
    1466           0 :         elog(ERROR, "relnatts disagrees with indnatts for index %u",
    1467             :              RelationGetRelid(relation));
    1468      380314 :     indnkeyatts = IndexRelationGetNumberOfKeyAttributes(relation);
    1469             : 
    1470             :     /*
    1471             :      * Make the private context to hold index access info.  The reason we need
    1472             :      * a context, and not just a couple of pallocs, is so that we won't leak
    1473             :      * any subsidiary info attached to fmgr lookup records.
    1474             :      */
    1475      380314 :     indexcxt = AllocSetContextCreate(CacheMemoryContext,
    1476             :                                      "index info",
    1477             :                                      ALLOCSET_SMALL_SIZES);
    1478      380314 :     relation->rd_indexcxt = indexcxt;
    1479      380314 :     MemoryContextCopyAndSetIdentifier(indexcxt,
    1480             :                                       RelationGetRelationName(relation));
    1481             : 
    1482             :     /*
    1483             :      * Now we can fetch the index AM's API struct
    1484             :      */
    1485      380314 :     InitIndexAmRoutine(relation);
    1486             : 
    1487             :     /*
    1488             :      * Allocate arrays to hold data. Opclasses are not used for included
    1489             :      * columns, so allocate them for indnkeyatts only.
    1490             :      */
    1491      380314 :     relation->rd_opfamily = (Oid *)
    1492      380314 :         MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
    1493      380314 :     relation->rd_opcintype = (Oid *)
    1494      380314 :         MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
    1495             : 
    1496      380314 :     amsupport = relation->rd_indam->amsupport;
    1497      380314 :     if (amsupport > 0)
    1498             :     {
    1499      380314 :         int         nsupport = indnatts * amsupport;
    1500             : 
    1501      380314 :         relation->rd_support = (RegProcedure *)
    1502      380314 :             MemoryContextAllocZero(indexcxt, nsupport * sizeof(RegProcedure));
    1503      380314 :         relation->rd_supportinfo = (FmgrInfo *)
    1504      380314 :             MemoryContextAllocZero(indexcxt, nsupport * sizeof(FmgrInfo));
    1505             :     }
    1506             :     else
    1507             :     {
    1508           0 :         relation->rd_support = NULL;
    1509           0 :         relation->rd_supportinfo = NULL;
    1510             :     }
    1511             : 
    1512      380314 :     relation->rd_indcollation = (Oid *)
    1513      380314 :         MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(Oid));
    1514             : 
    1515      380314 :     relation->rd_indoption = (int16 *)
    1516      380314 :         MemoryContextAllocZero(indexcxt, indnkeyatts * sizeof(int16));
    1517             : 
    1518             :     /*
    1519             :      * indcollation cannot be referenced directly through the C struct,
    1520             :      * because it comes after the variable-width indkey field.  Must extract
    1521             :      * the datum the hard way...
    1522             :      */
    1523      380314 :     indcollDatum = fastgetattr(relation->rd_indextuple,
    1524             :                                Anum_pg_index_indcollation,
    1525             :                                GetPgIndexDescriptor(),
    1526             :                                &isnull);
    1527             :     Assert(!isnull);
    1528      380314 :     indcoll = (oidvector *) DatumGetPointer(indcollDatum);
    1529      380314 :     memcpy(relation->rd_indcollation, indcoll->values, indnkeyatts * sizeof(Oid));
    1530             : 
    1531             :     /*
    1532             :      * indclass cannot be referenced directly through the C struct, because it
    1533             :      * comes after the variable-width indkey field.  Must extract the datum
    1534             :      * the hard way...
    1535             :      */
    1536      380314 :     indclassDatum = fastgetattr(relation->rd_indextuple,
    1537             :                                 Anum_pg_index_indclass,
    1538             :                                 GetPgIndexDescriptor(),
    1539             :                                 &isnull);
    1540             :     Assert(!isnull);
    1541      380314 :     indclass = (oidvector *) DatumGetPointer(indclassDatum);
    1542             : 
    1543             :     /*
    1544             :      * Fill the support procedure OID array, as well as the info about
    1545             :      * opfamilies and opclass input types.  (aminfo and supportinfo are left
    1546             :      * as zeroes, and are filled on-the-fly when used)
    1547             :      */
    1548      380314 :     IndexSupportInitialize(indclass, relation->rd_support,
    1549             :                            relation->rd_opfamily, relation->rd_opcintype,
    1550             :                            amsupport, indnkeyatts);
    1551             : 
    1552             :     /*
    1553             :      * Similarly extract indoption and copy it to the cache entry
    1554             :      */
    1555      380314 :     indoptionDatum = fastgetattr(relation->rd_indextuple,
    1556             :                                  Anum_pg_index_indoption,
    1557             :                                  GetPgIndexDescriptor(),
    1558             :                                  &isnull);
    1559             :     Assert(!isnull);
    1560      380314 :     indoption = (int2vector *) DatumGetPointer(indoptionDatum);
    1561      380314 :     memcpy(relation->rd_indoption, indoption->values, indnkeyatts * sizeof(int16));
    1562             : 
    1563      380314 :     (void) RelationGetIndexAttOptions(relation, false);
    1564             : 
    1565             :     /*
    1566             :      * expressions, predicate, exclusion caches will be filled later
    1567             :      */
    1568      380308 :     relation->rd_indexprs = NIL;
    1569      380308 :     relation->rd_indpred = NIL;
    1570      380308 :     relation->rd_exclops = NULL;
    1571      380308 :     relation->rd_exclprocs = NULL;
    1572      380308 :     relation->rd_exclstrats = NULL;
    1573      380308 :     relation->rd_amcache = NULL;
    1574      380308 : }
    1575             : 
    1576             : /*
    1577             :  * IndexSupportInitialize
    1578             :  *      Initializes an index's cached opclass information,
    1579             :  *      given the index's pg_index.indclass entry.
    1580             :  *
    1581             :  * Data is returned into *indexSupport, *opFamily, and *opcInType,
    1582             :  * which are arrays allocated by the caller.
    1583             :  *
    1584             :  * The caller also passes maxSupportNumber and maxAttributeNumber, since these
    1585             :  * indicate the size of the arrays it has allocated --- but in practice these
    1586             :  * numbers must always match those obtainable from the system catalog entries
    1587             :  * for the index and access method.
    1588             :  */
    1589             : static void
    1590      380314 : IndexSupportInitialize(oidvector *indclass,
    1591             :                        RegProcedure *indexSupport,
    1592             :                        Oid *opFamily,
    1593             :                        Oid *opcInType,
    1594             :                        StrategyNumber maxSupportNumber,
    1595             :                        AttrNumber maxAttributeNumber)
    1596             : {
    1597             :     int         attIndex;
    1598             : 
    1599     1022480 :     for (attIndex = 0; attIndex < maxAttributeNumber; attIndex++)
    1600             :     {
    1601             :         OpClassCacheEnt *opcentry;
    1602             : 
    1603      642166 :         if (!OidIsValid(indclass->values[attIndex]))
    1604           0 :             elog(ERROR, "bogus pg_index tuple");
    1605             : 
    1606             :         /* look up the info for this opclass, using a cache */
    1607      642166 :         opcentry = LookupOpclassInfo(indclass->values[attIndex],
    1608             :                                      maxSupportNumber);
    1609             : 
    1610             :         /* copy cached data into relcache entry */
    1611      642166 :         opFamily[attIndex] = opcentry->opcfamily;
    1612      642166 :         opcInType[attIndex] = opcentry->opcintype;
    1613      642166 :         if (maxSupportNumber > 0)
    1614      642166 :             memcpy(&indexSupport[attIndex * maxSupportNumber],
    1615      642166 :                    opcentry->supportProcs,
    1616             :                    maxSupportNumber * sizeof(RegProcedure));
    1617             :     }
    1618      380314 : }
    1619             : 
    1620             : /*
    1621             :  * LookupOpclassInfo
    1622             :  *
    1623             :  * This routine maintains a per-opclass cache of the information needed
    1624             :  * by IndexSupportInitialize().  This is more efficient than relying on
    1625             :  * the catalog cache, because we can load all the info about a particular
    1626             :  * opclass in a single indexscan of pg_amproc.
    1627             :  *
    1628             :  * The information from pg_am about expected range of support function
    1629             :  * numbers is passed in, rather than being looked up, mainly because the
    1630             :  * caller will have it already.
    1631             :  *
    1632             :  * Note there is no provision for flushing the cache.  This is OK at the
    1633             :  * moment because there is no way to ALTER any interesting properties of an
    1634             :  * existing opclass --- all you can do is drop it, which will result in
    1635             :  * a useless but harmless dead entry in the cache.  To support altering
    1636             :  * opclass membership (not the same as opfamily membership!), we'd need to
    1637             :  * be able to flush this cache as well as the contents of relcache entries
    1638             :  * for indexes.
    1639             :  */
    1640             : static OpClassCacheEnt *
    1641      642166 : LookupOpclassInfo(Oid operatorClassOid,
    1642             :                   StrategyNumber numSupport)
    1643             : {
    1644             :     OpClassCacheEnt *opcentry;
    1645             :     bool        found;
    1646             :     Relation    rel;
    1647             :     SysScanDesc scan;
    1648             :     ScanKeyData skey[3];
    1649             :     HeapTuple   htup;
    1650             :     bool        indexOK;
    1651             : 
    1652      642166 :     if (OpClassCache == NULL)
    1653             :     {
    1654             :         /* First time through: initialize the opclass cache */
    1655             :         HASHCTL     ctl;
    1656             : 
    1657             :         /* Also make sure CacheMemoryContext exists */
    1658       22162 :         if (!CacheMemoryContext)
    1659           0 :             CreateCacheMemoryContext();
    1660             : 
    1661       22162 :         ctl.keysize = sizeof(Oid);
    1662       22162 :         ctl.entrysize = sizeof(OpClassCacheEnt);
    1663       22162 :         OpClassCache = hash_create("Operator class cache", 64,
    1664             :                                    &ctl, HASH_ELEM | HASH_BLOBS);
    1665             :     }
    1666             : 
    1667      642166 :     opcentry = (OpClassCacheEnt *) hash_search(OpClassCache,
    1668             :                                                &operatorClassOid,
    1669             :                                                HASH_ENTER, &found);
    1670             : 
    1671      642166 :     if (!found)
    1672             :     {
    1673             :         /* Initialize new entry */
    1674       57892 :         opcentry->valid = false; /* until known OK */
    1675       57892 :         opcentry->numSupport = numSupport;
    1676       57892 :         opcentry->supportProcs = NULL;   /* filled below */
    1677             :     }
    1678             :     else
    1679             :     {
    1680             :         Assert(numSupport == opcentry->numSupport);
    1681             :     }
    1682             : 
    1683             :     /*
    1684             :      * When aggressively testing cache-flush hazards, we disable the operator
    1685             :      * class cache and force reloading of the info on each call.  This models
    1686             :      * no real-world behavior, since the cache entries are never invalidated
    1687             :      * otherwise.  However it can be helpful for detecting bugs in the cache
    1688             :      * loading logic itself, such as reliance on a non-nailed index.  Given
    1689             :      * the limited use-case and the fact that this adds a great deal of
    1690             :      * expense, we enable it only for high values of debug_discard_caches.
    1691             :      */
    1692             : #ifdef DISCARD_CACHES_ENABLED
    1693             :     if (debug_discard_caches > 2)
    1694             :         opcentry->valid = false;
    1695             : #endif
    1696             : 
    1697      642166 :     if (opcentry->valid)
    1698      584274 :         return opcentry;
    1699             : 
    1700             :     /*
    1701             :      * Need to fill in new entry.  First allocate space, unless we already did
    1702             :      * so in some previous attempt.
    1703             :      */
    1704       57892 :     if (opcentry->supportProcs == NULL && numSupport > 0)
    1705       57892 :         opcentry->supportProcs = (RegProcedure *)
    1706       57892 :             MemoryContextAllocZero(CacheMemoryContext,
    1707             :                                    numSupport * sizeof(RegProcedure));
    1708             : 
    1709             :     /*
    1710             :      * To avoid infinite recursion during startup, force heap scans if we're
    1711             :      * looking up info for the opclasses used by the indexes we would like to
    1712             :      * reference here.
    1713             :      */
    1714       64108 :     indexOK = criticalRelcachesBuilt ||
    1715        6216 :         (operatorClassOid != OID_BTREE_OPS_OID &&
    1716        4272 :          operatorClassOid != INT2_BTREE_OPS_OID);
    1717             : 
    1718             :     /*
    1719             :      * We have to fetch the pg_opclass row to determine its opfamily and
    1720             :      * opcintype, which are needed to look up related operators and functions.
    1721             :      * It'd be convenient to use the syscache here, but that probably doesn't
    1722             :      * work while bootstrapping.
    1723             :      */
    1724       57892 :     ScanKeyInit(&skey[0],
    1725             :                 Anum_pg_opclass_oid,
    1726             :                 BTEqualStrategyNumber, F_OIDEQ,
    1727             :                 ObjectIdGetDatum(operatorClassOid));
    1728       57892 :     rel = table_open(OperatorClassRelationId, AccessShareLock);
    1729       57892 :     scan = systable_beginscan(rel, OpclassOidIndexId, indexOK,
    1730             :                               NULL, 1, skey);
    1731             : 
    1732       57892 :     if (HeapTupleIsValid(htup = systable_getnext(scan)))
    1733             :     {
    1734       57892 :         Form_pg_opclass opclassform = (Form_pg_opclass) GETSTRUCT(htup);
    1735             : 
    1736       57892 :         opcentry->opcfamily = opclassform->opcfamily;
    1737       57892 :         opcentry->opcintype = opclassform->opcintype;
    1738             :     }
    1739             :     else
    1740           0 :         elog(ERROR, "could not find tuple for opclass %u", operatorClassOid);
    1741             : 
    1742       57892 :     systable_endscan(scan);
    1743       57892 :     table_close(rel, AccessShareLock);
    1744             : 
    1745             :     /*
    1746             :      * Scan pg_amproc to obtain support procs for the opclass.  We only fetch
    1747             :      * the default ones (those with lefttype = righttype = opcintype).
    1748             :      */
    1749       57892 :     if (numSupport > 0)
    1750             :     {
    1751       57892 :         ScanKeyInit(&skey[0],
    1752             :                     Anum_pg_amproc_amprocfamily,
    1753             :                     BTEqualStrategyNumber, F_OIDEQ,
    1754             :                     ObjectIdGetDatum(opcentry->opcfamily));
    1755       57892 :         ScanKeyInit(&skey[1],
    1756             :                     Anum_pg_amproc_amproclefttype,
    1757             :                     BTEqualStrategyNumber, F_OIDEQ,
    1758             :                     ObjectIdGetDatum(opcentry->opcintype));
    1759       57892 :         ScanKeyInit(&skey[2],
    1760             :                     Anum_pg_amproc_amprocrighttype,
    1761             :                     BTEqualStrategyNumber, F_OIDEQ,
    1762             :                     ObjectIdGetDatum(opcentry->opcintype));
    1763       57892 :         rel = table_open(AccessMethodProcedureRelationId, AccessShareLock);
    1764       57892 :         scan = systable_beginscan(rel, AccessMethodProcedureIndexId, indexOK,
    1765             :                                   NULL, 3, skey);
    1766             : 
    1767      241306 :         while (HeapTupleIsValid(htup = systable_getnext(scan)))
    1768             :         {
    1769      183414 :             Form_pg_amproc amprocform = (Form_pg_amproc) GETSTRUCT(htup);
    1770             : 
    1771      183414 :             if (amprocform->amprocnum <= 0 ||
    1772      183414 :                 (StrategyNumber) amprocform->amprocnum > numSupport)
    1773           0 :                 elog(ERROR, "invalid amproc number %d for opclass %u",
    1774             :                      amprocform->amprocnum, operatorClassOid);
    1775             : 
    1776      183414 :             opcentry->supportProcs[amprocform->amprocnum - 1] =
    1777      183414 :                 amprocform->amproc;
    1778             :         }
    1779             : 
    1780       57892 :         systable_endscan(scan);
    1781       57892 :         table_close(rel, AccessShareLock);
    1782             :     }
    1783             : 
    1784       57892 :     opcentry->valid = true;
    1785       57892 :     return opcentry;
    1786             : }
    1787             : 
    1788             : /*
    1789             :  * Fill in the TableAmRoutine for a relation
    1790             :  *
    1791             :  * relation's rd_amhandler must be valid already.
    1792             :  */
    1793             : static void
    1794     1562532 : InitTableAmRoutine(Relation relation)
    1795             : {
    1796     1562532 :     relation->rd_tableam = GetTableAmRoutine(relation->rd_amhandler);
    1797     1562532 : }
    1798             : 
    1799             : /*
    1800             :  * Initialize table access method support for a table like relation
    1801             :  */
    1802             : void
    1803     1562532 : RelationInitTableAccessMethod(Relation relation)
    1804             : {
    1805             :     HeapTuple   tuple;
    1806             :     Form_pg_am  aform;
    1807             : 
    1808     1562532 :     if (relation->rd_rel->relkind == RELKIND_SEQUENCE)
    1809             :     {
    1810             :         /*
    1811             :          * Sequences are currently accessed like heap tables, but it doesn't
    1812             :          * seem prudent to show that in the catalog. So just overwrite it
    1813             :          * here.
    1814             :          */
    1815             :         Assert(relation->rd_rel->relam == InvalidOid);
    1816        6582 :         relation->rd_amhandler = F_HEAP_TABLEAM_HANDLER;
    1817             :     }
    1818     1555950 :     else if (IsCatalogRelation(relation))
    1819             :     {
    1820             :         /*
    1821             :          * Avoid doing a syscache lookup for catalog tables.
    1822             :          */
    1823             :         Assert(relation->rd_rel->relam == HEAP_TABLE_AM_OID);
    1824     1213508 :         relation->rd_amhandler = F_HEAP_TABLEAM_HANDLER;
    1825             :     }
    1826             :     else
    1827             :     {
    1828             :         /*
    1829             :          * Look up the table access method, save the OID of its handler
    1830             :          * function.
    1831             :          */
    1832             :         Assert(relation->rd_rel->relam != InvalidOid);
    1833      342442 :         tuple = SearchSysCache1(AMOID,
    1834      342442 :                                 ObjectIdGetDatum(relation->rd_rel->relam));
    1835      342442 :         if (!HeapTupleIsValid(tuple))
    1836           0 :             elog(ERROR, "cache lookup failed for access method %u",
    1837             :                  relation->rd_rel->relam);
    1838      342442 :         aform = (Form_pg_am) GETSTRUCT(tuple);
    1839      342442 :         relation->rd_amhandler = aform->amhandler;
    1840      342442 :         ReleaseSysCache(tuple);
    1841             :     }
    1842             : 
    1843             :     /*
    1844             :      * Now we can fetch the table AM's API struct
    1845             :      */
    1846     1562532 :     InitTableAmRoutine(relation);
    1847     1562532 : }
    1848             : 
    1849             : /*
    1850             :  *      formrdesc
    1851             :  *
    1852             :  *      This is a special cut-down version of RelationBuildDesc(),
    1853             :  *      used while initializing the relcache.
    1854             :  *      The relation descriptor is built just from the supplied parameters,
    1855             :  *      without actually looking at any system table entries.  We cheat
    1856             :  *      quite a lot since we only need to work for a few basic system
    1857             :  *      catalogs.
    1858             :  *
    1859             :  * The catalogs this is used for can't have constraints (except attnotnull),
    1860             :  * default values, rules, or triggers, since we don't cope with any of that.
    1861             :  * (Well, actually, this only matters for properties that need to be valid
    1862             :  * during bootstrap or before RelationCacheInitializePhase3 runs, and none of
    1863             :  * these properties matter then...)
    1864             :  *
    1865             :  * NOTE: we assume we are already switched into CacheMemoryContext.
    1866             :  */
    1867             : static void
    1868       22316 : formrdesc(const char *relationName, Oid relationReltype,
    1869             :           bool isshared,
    1870             :           int natts, const FormData_pg_attribute *attrs)
    1871             : {
    1872             :     Relation    relation;
    1873             :     int         i;
    1874             :     bool        has_not_null;
    1875             : 
    1876             :     /*
    1877             :      * allocate new relation desc, clear all fields of reldesc
    1878             :      */
    1879       22316 :     relation = (Relation) palloc0(sizeof(RelationData));
    1880             : 
    1881             :     /* make sure relation is marked as having no open file yet */
    1882       22316 :     relation->rd_smgr = NULL;
    1883             : 
    1884             :     /*
    1885             :      * initialize reference count: 1 because it is nailed in cache
    1886             :      */
    1887       22316 :     relation->rd_refcnt = 1;
    1888             : 
    1889             :     /*
    1890             :      * all entries built with this routine are nailed-in-cache; none are for
    1891             :      * new or temp relations.
    1892             :      */
    1893       22316 :     relation->rd_isnailed = true;
    1894       22316 :     relation->rd_createSubid = InvalidSubTransactionId;
    1895       22316 :     relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    1896       22316 :     relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    1897       22316 :     relation->rd_droppedSubid = InvalidSubTransactionId;
    1898       22316 :     relation->rd_backend = InvalidBackendId;
    1899       22316 :     relation->rd_islocaltemp = false;
    1900             : 
    1901             :     /*
    1902             :      * initialize relation tuple form
    1903             :      *
    1904             :      * The data we insert here is pretty incomplete/bogus, but it'll serve to
    1905             :      * get us launched.  RelationCacheInitializePhase3() will read the real
    1906             :      * data from pg_class and replace what we've done here.  Note in
    1907             :      * particular that relowner is left as zero; this cues
    1908             :      * RelationCacheInitializePhase3 that the real data isn't there yet.
    1909             :      */
    1910       22316 :     relation->rd_rel = (Form_pg_class) palloc0(CLASS_TUPLE_SIZE);
    1911             : 
    1912       22316 :     namestrcpy(&relation->rd_rel->relname, relationName);
    1913       22316 :     relation->rd_rel->relnamespace = PG_CATALOG_NAMESPACE;
    1914       22316 :     relation->rd_rel->reltype = relationReltype;
    1915             : 
    1916             :     /*
    1917             :      * It's important to distinguish between shared and non-shared relations,
    1918             :      * even at bootstrap time, to make sure we know where they are stored.
    1919             :      */
    1920       22316 :     relation->rd_rel->relisshared = isshared;
    1921       22316 :     if (isshared)
    1922       14540 :         relation->rd_rel->reltablespace = GLOBALTABLESPACE_OID;
    1923             : 
    1924             :     /* formrdesc is used only for permanent relations */
    1925       22316 :     relation->rd_rel->relpersistence = RELPERSISTENCE_PERMANENT;
    1926             : 
    1927             :     /* ... and they're always populated, too */
    1928       22316 :     relation->rd_rel->relispopulated = true;
    1929             : 
    1930       22316 :     relation->rd_rel->relreplident = REPLICA_IDENTITY_NOTHING;
    1931       22316 :     relation->rd_rel->relpages = 0;
    1932       22316 :     relation->rd_rel->reltuples = -1;
    1933       22316 :     relation->rd_rel->relallvisible = 0;
    1934       22316 :     relation->rd_rel->relkind = RELKIND_RELATION;
    1935       22316 :     relation->rd_rel->relnatts = (int16) natts;
    1936       22316 :     relation->rd_rel->relam = HEAP_TABLE_AM_OID;
    1937             : 
    1938             :     /*
    1939             :      * initialize attribute tuple form
    1940             :      *
    1941             :      * Unlike the case with the relation tuple, this data had better be right
    1942             :      * because it will never be replaced.  The data comes from
    1943             :      * src/include/catalog/ headers via genbki.pl.
    1944             :      */
    1945       22316 :     relation->rd_att = CreateTemplateTupleDesc(natts);
    1946       22316 :     relation->rd_att->tdrefcount = 1; /* mark as refcounted */
    1947             : 
    1948       22316 :     relation->rd_att->tdtypeid = relationReltype;
    1949       22316 :     relation->rd_att->tdtypmod = -1;  /* just to be sure */
    1950             : 
    1951             :     /*
    1952             :      * initialize tuple desc info
    1953             :      */
    1954       22316 :     has_not_null = false;
    1955      426204 :     for (i = 0; i < natts; i++)
    1956             :     {
    1957      403888 :         memcpy(TupleDescAttr(relation->rd_att, i),
    1958      403888 :                &attrs[i],
    1959             :                ATTRIBUTE_FIXED_PART_SIZE);
    1960      403888 :         has_not_null |= attrs[i].attnotnull;
    1961             :         /* make sure attcacheoff is valid */
    1962      403888 :         TupleDescAttr(relation->rd_att, i)->attcacheoff = -1;
    1963             :     }
    1964             : 
    1965             :     /* initialize first attribute's attcacheoff, cf RelationBuildTupleDesc */
    1966       22316 :     TupleDescAttr(relation->rd_att, 0)->attcacheoff = 0;
    1967             : 
    1968             :     /* mark not-null status */
    1969       22316 :     if (has_not_null)
    1970             :     {
    1971       22316 :         TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
    1972             : 
    1973       22316 :         constr->has_not_null = true;
    1974       22316 :         relation->rd_att->constr = constr;
    1975             :     }
    1976             : 
    1977             :     /*
    1978             :      * initialize relation id from info in att array (my, this is ugly)
    1979             :      */
    1980       22316 :     RelationGetRelid(relation) = TupleDescAttr(relation->rd_att, 0)->attrelid;
    1981             : 
    1982             :     /*
    1983             :      * All relations made with formrdesc are mapped.  This is necessarily so
    1984             :      * because there is no other way to know what filenumber they currently
    1985             :      * have.  In bootstrap mode, add them to the initial relation mapper data,
    1986             :      * specifying that the initial filenumber is the same as the OID.
    1987             :      */
    1988       22316 :     relation->rd_rel->relfilenode = InvalidRelFileNumber;
    1989       22316 :     if (IsBootstrapProcessingMode())
    1990         256 :         RelationMapUpdateMap(RelationGetRelid(relation),
    1991             :                              RelationGetRelid(relation),
    1992             :                              isshared, true);
    1993             : 
    1994             :     /*
    1995             :      * initialize the relation lock manager information
    1996             :      */
    1997       22316 :     RelationInitLockInfo(relation); /* see lmgr.c */
    1998             : 
    1999             :     /*
    2000             :      * initialize physical addressing information for the relation
    2001             :      */
    2002       22316 :     RelationInitPhysicalAddr(relation);
    2003             : 
    2004             :     /*
    2005             :      * initialize the table am handler
    2006             :      */
    2007       22316 :     relation->rd_rel->relam = HEAP_TABLE_AM_OID;
    2008       22316 :     relation->rd_tableam = GetHeapamTableAmRoutine();
    2009             : 
    2010             :     /*
    2011             :      * initialize the rel-has-index flag, using hardwired knowledge
    2012             :      */
    2013       22316 :     if (IsBootstrapProcessingMode())
    2014             :     {
    2015             :         /* In bootstrap mode, we have no indexes */
    2016         256 :         relation->rd_rel->relhasindex = false;
    2017             :     }
    2018             :     else
    2019             :     {
    2020             :         /* Otherwise, all the rels formrdesc is used for have indexes */
    2021       22060 :         relation->rd_rel->relhasindex = true;
    2022             :     }
    2023             : 
    2024             :     /*
    2025             :      * add new reldesc to relcache
    2026             :      */
    2027       22316 :     RelationCacheInsert(relation, false);
    2028             : 
    2029             :     /* It's fully valid */
    2030       22316 :     relation->rd_isvalid = true;
    2031       22316 : }
    2032             : 
    2033             : 
    2034             : /* ----------------------------------------------------------------
    2035             :  *               Relation Descriptor Lookup Interface
    2036             :  * ----------------------------------------------------------------
    2037             :  */
    2038             : 
    2039             : /*
    2040             :  *      RelationIdGetRelation
    2041             :  *
    2042             :  *      Lookup a reldesc by OID; make one if not already in cache.
    2043             :  *
    2044             :  *      Returns NULL if no pg_class row could be found for the given relid
    2045             :  *      (suggesting we are trying to access a just-deleted relation).
    2046             :  *      Any other error is reported via elog.
    2047             :  *
    2048             :  *      NB: caller should already have at least AccessShareLock on the
    2049             :  *      relation ID, else there are nasty race conditions.
    2050             :  *
    2051             :  *      NB: relation ref count is incremented, or set to 1 if new entry.
    2052             :  *      Caller should eventually decrement count.  (Usually,
    2053             :  *      that happens by calling RelationClose().)
    2054             :  */
    2055             : Relation
    2056    28995094 : RelationIdGetRelation(Oid relationId)
    2057             : {
    2058             :     Relation    rd;
    2059             : 
    2060             :     /* Make sure we're in an xact, even if this ends up being a cache hit */
    2061             :     Assert(IsTransactionState());
    2062             : 
    2063             :     /*
    2064             :      * first try to find reldesc in the cache
    2065             :      */
    2066    28995094 :     RelationIdCacheLookup(relationId, rd);
    2067             : 
    2068    28995094 :     if (RelationIsValid(rd))
    2069             :     {
    2070             :         /* return NULL for dropped relations */
    2071    28319234 :         if (rd->rd_droppedSubid != InvalidSubTransactionId)
    2072             :         {
    2073             :             Assert(!rd->rd_isvalid);
    2074           4 :             return NULL;
    2075             :         }
    2076             : 
    2077    28319230 :         RelationIncrementReferenceCount(rd);
    2078             :         /* revalidate cache entry if necessary */
    2079    28319230 :         if (!rd->rd_isvalid)
    2080             :         {
    2081             :             /*
    2082             :              * Indexes only have a limited number of possible schema changes,
    2083             :              * and we don't want to use the full-blown procedure because it's
    2084             :              * a headache for indexes that reload itself depends on.
    2085             :              */
    2086      117200 :             if (rd->rd_rel->relkind == RELKIND_INDEX ||
    2087       96582 :                 rd->rd_rel->relkind == RELKIND_PARTITIONED_INDEX)
    2088       20618 :                 RelationReloadIndexInfo(rd);
    2089             :             else
    2090       96582 :                 RelationClearRelation(rd, true);
    2091             : 
    2092             :             /*
    2093             :              * Normally entries need to be valid here, but before the relcache
    2094             :              * has been initialized, not enough infrastructure exists to
    2095             :              * perform pg_class lookups. The structure of such entries doesn't
    2096             :              * change, but we still want to update the rd_rel entry. So
    2097             :              * rd_isvalid = false is left in place for a later lookup.
    2098             :              */
    2099             :             Assert(rd->rd_isvalid ||
    2100             :                    (rd->rd_isnailed && !criticalRelcachesBuilt));
    2101             :         }
    2102    28319218 :         return rd;
    2103             :     }
    2104             : 
    2105             :     /*
    2106             :      * no reldesc in the cache, so have RelationBuildDesc() build one and add
    2107             :      * it.
    2108             :      */
    2109      675860 :     rd = RelationBuildDesc(relationId, true);
    2110      675856 :     if (RelationIsValid(rd))
    2111      675846 :         RelationIncrementReferenceCount(rd);
    2112      675856 :     return rd;
    2113             : }
    2114             : 
    2115             : /* ----------------------------------------------------------------
    2116             :  *              cache invalidation support routines
    2117             :  * ----------------------------------------------------------------
    2118             :  */
    2119             : 
    2120             : /* ResourceOwner callbacks to track relcache references */
    2121             : static void ResOwnerReleaseRelation(Datum res);
    2122             : static char *ResOwnerPrintRelCache(Datum res);
    2123             : 
    2124             : static const ResourceOwnerDesc relref_resowner_desc =
    2125             : {
    2126             :     .name = "relcache reference",
    2127             :     .release_phase = RESOURCE_RELEASE_BEFORE_LOCKS,
    2128             :     .release_priority = RELEASE_PRIO_RELCACHE_REFS,
    2129             :     .ReleaseResource = ResOwnerReleaseRelation,
    2130             :     .DebugPrint = ResOwnerPrintRelCache
    2131             : };
    2132             : 
    2133             : /* Convenience wrappers over ResourceOwnerRemember/Forget */
    2134             : static inline void
    2135    42500684 : ResourceOwnerRememberRelationRef(ResourceOwner owner, Relation rel)
    2136             : {
    2137    42500684 :     ResourceOwnerRemember(owner, PointerGetDatum(rel), &relref_resowner_desc);
    2138    42500684 : }
    2139             : static inline void
    2140    42468354 : ResourceOwnerForgetRelationRef(ResourceOwner owner, Relation rel)
    2141             : {
    2142    42468354 :     ResourceOwnerForget(owner, PointerGetDatum(rel), &relref_resowner_desc);
    2143    42468354 : }
    2144             : 
    2145             : /*
    2146             :  * RelationIncrementReferenceCount
    2147             :  *      Increments relation reference count.
    2148             :  *
    2149             :  * Note: bootstrap mode has its own weird ideas about relation refcount
    2150             :  * behavior; we ought to fix it someday, but for now, just disable
    2151             :  * reference count ownership tracking in bootstrap mode.
    2152             :  */
    2153             : void
    2154    42865420 : RelationIncrementReferenceCount(Relation rel)
    2155             : {
    2156    42865420 :     ResourceOwnerEnlarge(CurrentResourceOwner);
    2157    42865420 :     rel->rd_refcnt += 1;
    2158    42865420 :     if (!IsBootstrapProcessingMode())
    2159    42500684 :         ResourceOwnerRememberRelationRef(CurrentResourceOwner, rel);
    2160    42865420 : }
    2161             : 
    2162             : /*
    2163             :  * RelationDecrementReferenceCount
    2164             :  *      Decrements relation reference count.
    2165             :  */
    2166             : void
    2167    42833090 : RelationDecrementReferenceCount(Relation rel)
    2168             : {
    2169             :     Assert(rel->rd_refcnt > 0);
    2170    42833090 :     rel->rd_refcnt -= 1;
    2171    42833090 :     if (!IsBootstrapProcessingMode())
    2172    42468354 :         ResourceOwnerForgetRelationRef(CurrentResourceOwner, rel);
    2173    42833090 : }
    2174             : 
    2175             : /*
    2176             :  * RelationClose - close an open relation
    2177             :  *
    2178             :  *  Actually, we just decrement the refcount.
    2179             :  *
    2180             :  *  NOTE: if compiled with -DRELCACHE_FORCE_RELEASE then relcache entries
    2181             :  *  will be freed as soon as their refcount goes to zero.  In combination
    2182             :  *  with aset.c's CLOBBER_FREED_MEMORY option, this provides a good test
    2183             :  *  to catch references to already-released relcache entries.  It slows
    2184             :  *  things down quite a bit, however.
    2185             :  */
    2186             : void
    2187    29072766 : RelationClose(Relation relation)
    2188             : {
    2189             :     /* Note: no locking manipulations needed */
    2190    29072766 :     RelationDecrementReferenceCount(relation);
    2191             : 
    2192    29072766 :     RelationCloseCleanup(relation);
    2193    29072766 : }
    2194             : 
    2195             : static void
    2196    29105096 : RelationCloseCleanup(Relation relation)
    2197             : {
    2198             :     /*
    2199             :      * If the relation is no longer open in this session, we can clean up any
    2200             :      * stale partition descriptors it has.  This is unlikely, so check to see
    2201             :      * if there are child contexts before expending a call to mcxt.c.
    2202             :      */
    2203    29105096 :     if (RelationHasReferenceCountZero(relation))
    2204             :     {
    2205    17608162 :         if (relation->rd_pdcxt != NULL &&
    2206       79720 :             relation->rd_pdcxt->firstchild != NULL)
    2207        3510 :             MemoryContextDeleteChildren(relation->rd_pdcxt);
    2208             : 
    2209    17608162 :         if (relation->rd_pddcxt != NULL &&
    2210          94 :             relation->rd_pddcxt->firstchild != NULL)
    2211           0 :             MemoryContextDeleteChildren(relation->rd_pddcxt);
    2212             :     }
    2213             : 
    2214             : #ifdef RELCACHE_FORCE_RELEASE
    2215             :     if (RelationHasReferenceCountZero(relation) &&
    2216             :         relation->rd_createSubid == InvalidSubTransactionId &&
    2217             :         relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId)
    2218             :         RelationClearRelation(relation, false);
    2219             : #endif
    2220    29105096 : }
    2221             : 
    2222             : /*
    2223             :  * RelationReloadIndexInfo - reload minimal information for an open index
    2224             :  *
    2225             :  *  This function is used only for indexes.  A relcache inval on an index
    2226             :  *  can mean that its pg_class or pg_index row changed.  There are only
    2227             :  *  very limited changes that are allowed to an existing index's schema,
    2228             :  *  so we can update the relcache entry without a complete rebuild; which
    2229             :  *  is fortunate because we can't rebuild an index entry that is "nailed"
    2230             :  *  and/or in active use.  We support full replacement of the pg_class row,
    2231             :  *  as well as updates of a few simple fields of the pg_index row.
    2232             :  *
    2233             :  *  We can't necessarily reread the catalog rows right away; we might be
    2234             :  *  in a failed transaction when we receive the SI notification.  If so,
    2235             :  *  RelationClearRelation just marks the entry as invalid by setting
    2236             :  *  rd_isvalid to false.  This routine is called to fix the entry when it
    2237             :  *  is next needed.
    2238             :  *
    2239             :  *  We assume that at the time we are called, we have at least AccessShareLock
    2240             :  *  on the target index.  (Note: in the calls from RelationClearRelation,
    2241             :  *  this is legitimate because we know the rel has positive refcount.)
    2242             :  *
    2243             :  *  If the target index is an index on pg_class or pg_index, we'd better have
    2244             :  *  previously gotten at least AccessShareLock on its underlying catalog,
    2245             :  *  else we are at risk of deadlock against someone trying to exclusive-lock
    2246             :  *  the heap and index in that order.  This is ensured in current usage by
    2247             :  *  only applying this to indexes being opened or having positive refcount.
    2248             :  */
    2249             : static void
    2250       60680 : RelationReloadIndexInfo(Relation relation)
    2251             : {
    2252             :     bool        indexOK;
    2253             :     HeapTuple   pg_class_tuple;
    2254             :     Form_pg_class relp;
    2255             : 
    2256             :     /* Should be called only for invalidated, live indexes */
    2257             :     Assert((relation->rd_rel->relkind == RELKIND_INDEX ||
    2258             :             relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX) &&
    2259             :            !relation->rd_isvalid &&
    2260             :            relation->rd_droppedSubid == InvalidSubTransactionId);
    2261             : 
    2262             :     /* Ensure it's closed at smgr level */
    2263       60680 :     RelationCloseSmgr(relation);
    2264             : 
    2265             :     /* Must free any AM cached data upon relcache flush */
    2266       60680 :     if (relation->rd_amcache)
    2267           0 :         pfree(relation->rd_amcache);
    2268       60680 :     relation->rd_amcache = NULL;
    2269             : 
    2270             :     /*
    2271             :      * If it's a shared index, we might be called before backend startup has
    2272             :      * finished selecting a database, in which case we have no way to read
    2273             :      * pg_class yet.  However, a shared index can never have any significant
    2274             :      * schema updates, so it's okay to ignore the invalidation signal.  Just
    2275             :      * mark it valid and return without doing anything more.
    2276             :      */
    2277       60680 :     if (relation->rd_rel->relisshared && !criticalRelcachesBuilt)
    2278             :     {
    2279           0 :         relation->rd_isvalid = true;
    2280           0 :         return;
    2281             :     }
    2282             : 
    2283             :     /*
    2284             :      * Read the pg_class row
    2285             :      *
    2286             :      * Don't try to use an indexscan of pg_class_oid_index to reload the info
    2287             :      * for pg_class_oid_index ...
    2288             :      */
    2289       60680 :     indexOK = (RelationGetRelid(relation) != ClassOidIndexId);
    2290       60680 :     pg_class_tuple = ScanPgRelation(RelationGetRelid(relation), indexOK, false);
    2291       60674 :     if (!HeapTupleIsValid(pg_class_tuple))
    2292           0 :         elog(ERROR, "could not find pg_class tuple for index %u",
    2293             :              RelationGetRelid(relation));
    2294       60674 :     relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
    2295       60674 :     memcpy(relation->rd_rel, relp, CLASS_TUPLE_SIZE);
    2296             :     /* Reload reloptions in case they changed */
    2297       60674 :     if (relation->rd_options)
    2298         506 :         pfree(relation->rd_options);
    2299       60674 :     RelationParseRelOptions(relation, pg_class_tuple);
    2300             :     /* done with pg_class tuple */
    2301       60674 :     heap_freetuple(pg_class_tuple);
    2302             :     /* We must recalculate physical address in case it changed */
    2303       60674 :     RelationInitPhysicalAddr(relation);
    2304             : 
    2305             :     /*
    2306             :      * For a non-system index, there are fields of the pg_index row that are
    2307             :      * allowed to change, so re-read that row and update the relcache entry.
    2308             :      * Most of the info derived from pg_index (such as support function lookup
    2309             :      * info) cannot change, and indeed the whole point of this routine is to
    2310             :      * update the relcache entry without clobbering that data; so wholesale
    2311             :      * replacement is not appropriate.
    2312             :      */
    2313       60674 :     if (!IsSystemRelation(relation))
    2314             :     {
    2315             :         HeapTuple   tuple;
    2316             :         Form_pg_index index;
    2317             : 
    2318       21206 :         tuple = SearchSysCache1(INDEXRELID,
    2319             :                                 ObjectIdGetDatum(RelationGetRelid(relation)));
    2320       21206 :         if (!HeapTupleIsValid(tuple))
    2321           0 :             elog(ERROR, "cache lookup failed for index %u",
    2322             :                  RelationGetRelid(relation));
    2323       21206 :         index = (Form_pg_index) GETSTRUCT(tuple);
    2324             : 
    2325             :         /*
    2326             :          * Basically, let's just copy all the bool fields.  There are one or
    2327             :          * two of these that can't actually change in the current code, but
    2328             :          * it's not worth it to track exactly which ones they are.  None of
    2329             :          * the array fields are allowed to change, though.
    2330             :          */
    2331       21206 :         relation->rd_index->indisunique = index->indisunique;
    2332       21206 :         relation->rd_index->indnullsnotdistinct = index->indnullsnotdistinct;
    2333       21206 :         relation->rd_index->indisprimary = index->indisprimary;
    2334       21206 :         relation->rd_index->indisexclusion = index->indisexclusion;
    2335       21206 :         relation->rd_index->indimmediate = index->indimmediate;
    2336       21206 :         relation->rd_index->indisclustered = index->indisclustered;
    2337       21206 :         relation->rd_index->indisvalid = index->indisvalid;
    2338       21206 :         relation->rd_index->indcheckxmin = index->indcheckxmin;
    2339       21206 :         relation->rd_index->indisready = index->indisready;
    2340       21206 :         relation->rd_index->indislive = index->indislive;
    2341       21206 :         relation->rd_index->indisreplident = index->indisreplident;
    2342             : 
    2343             :         /* Copy xmin too, as that is needed to make sense of indcheckxmin */
    2344       21206 :         HeapTupleHeaderSetXmin(relation->rd_indextuple->t_data,
    2345             :                                HeapTupleHeaderGetXmin(tuple->t_data));
    2346             : 
    2347       21206 :         ReleaseSysCache(tuple);
    2348             :     }
    2349             : 
    2350             :     /* Okay, now it's valid again */
    2351       60674 :     relation->rd_isvalid = true;
    2352             : }
    2353             : 
    2354             : /*
    2355             :  * RelationReloadNailed - reload minimal information for nailed relations.
    2356             :  *
    2357             :  * The structure of a nailed relation can never change (which is good, because
    2358             :  * we rely on knowing their structure to be able to read catalog content). But
    2359             :  * some parts, e.g. pg_class.relfrozenxid, are still important to have
    2360             :  * accurate content for. Therefore those need to be reloaded after the arrival
    2361             :  * of invalidations.
    2362             :  */
    2363             : static void
    2364      194324 : RelationReloadNailed(Relation relation)
    2365             : {
    2366             :     Assert(relation->rd_isnailed);
    2367             : 
    2368             :     /*
    2369             :      * Redo RelationInitPhysicalAddr in case it is a mapped relation whose
    2370             :      * mapping changed.
    2371             :      */
    2372      194324 :     RelationInitPhysicalAddr(relation);
    2373             : 
    2374             :     /* flag as needing to be revalidated */
    2375      194324 :     relation->rd_isvalid = false;
    2376             : 
    2377             :     /*
    2378             :      * Can only reread catalog contents if in a transaction.  If the relation
    2379             :      * is currently open (not counting the nailed refcount), do so
    2380             :      * immediately. Otherwise we've already marked the entry as possibly
    2381             :      * invalid, and it'll be fixed when next opened.
    2382             :      */
    2383      194324 :     if (!IsTransactionState() || relation->rd_refcnt <= 1)
    2384       94794 :         return;
    2385             : 
    2386       99530 :     if (relation->rd_rel->relkind == RELKIND_INDEX)
    2387             :     {
    2388             :         /*
    2389             :          * If it's a nailed-but-not-mapped index, then we need to re-read the
    2390             :          * pg_class row to see if its relfilenumber changed.
    2391             :          */
    2392         316 :         RelationReloadIndexInfo(relation);
    2393             :     }
    2394             :     else
    2395             :     {
    2396             :         /*
    2397             :          * Reload a non-index entry.  We can't easily do so if relcaches
    2398             :          * aren't yet built, but that's fine because at that stage the
    2399             :          * attributes that need to be current (like relfrozenxid) aren't yet
    2400             :          * accessed.  To ensure the entry will later be revalidated, we leave
    2401             :          * it in invalid state, but allow use (cf. RelationIdGetRelation()).
    2402             :          */
    2403       99214 :         if (criticalRelcachesBuilt)
    2404             :         {
    2405             :             HeapTuple   pg_class_tuple;
    2406             :             Form_pg_class relp;
    2407             : 
    2408             :             /*
    2409             :              * NB: Mark the entry as valid before starting to scan, to avoid
    2410             :              * self-recursion when re-building pg_class.
    2411             :              */
    2412       19674 :             relation->rd_isvalid = true;
    2413             : 
    2414       19674 :             pg_class_tuple = ScanPgRelation(RelationGetRelid(relation),
    2415             :                                             true, false);
    2416       19668 :             relp = (Form_pg_class) GETSTRUCT(pg_class_tuple);
    2417       19668 :             memcpy(relation->rd_rel, relp, CLASS_TUPLE_SIZE);
    2418       19668 :             heap_freetuple(pg_class_tuple);
    2419             : 
    2420             :             /*
    2421             :              * Again mark as valid, to protect against concurrently arriving
    2422             :              * invalidations.
    2423             :              */
    2424       19668 :             relation->rd_isvalid = true;
    2425             :         }
    2426             :     }
    2427             : }
    2428             : 
    2429             : /*
    2430             :  * RelationDestroyRelation
    2431             :  *
    2432             :  *  Physically delete a relation cache entry and all subsidiary data.
    2433             :  *  Caller must already have unhooked the entry from the hash table.
    2434             :  */
    2435             : static void
    2436      901440 : RelationDestroyRelation(Relation relation, bool remember_tupdesc)
    2437             : {
    2438             :     Assert(RelationHasReferenceCountZero(relation));
    2439             : 
    2440             :     /*
    2441             :      * Make sure smgr and lower levels close the relation's files, if they
    2442             :      * weren't closed already.  (This was probably done by caller, but let's
    2443             :      * just be real sure.)
    2444             :      */
    2445      901440 :     RelationCloseSmgr(relation);
    2446             : 
    2447             :     /* break mutual link with stats entry */
    2448      901440 :     pgstat_unlink_relation(relation);
    2449             : 
    2450             :     /*
    2451             :      * Free all the subsidiary data structures of the relcache entry, then the
    2452             :      * entry itself.
    2453             :      */
    2454      901440 :     if (relation->rd_rel)
    2455      901440 :         pfree(relation->rd_rel);
    2456             :     /* can't use DecrTupleDescRefCount here */
    2457             :     Assert(relation->rd_att->tdrefcount > 0);
    2458      901440 :     if (--relation->rd_att->tdrefcount == 0)
    2459             :     {
    2460             :         /*
    2461             :          * If we Rebuilt a relcache entry during a transaction then its
    2462             :          * possible we did that because the TupDesc changed as the result of
    2463             :          * an ALTER TABLE that ran at less than AccessExclusiveLock. It's
    2464             :          * possible someone copied that TupDesc, in which case the copy would
    2465             :          * point to free'd memory. So if we rebuild an entry we keep the
    2466             :          * TupDesc around until end of transaction, to be safe.
    2467             :          */
    2468      898834 :         if (remember_tupdesc)
    2469       16974 :             RememberToFreeTupleDescAtEOX(relation->rd_att);
    2470             :         else
    2471      881860 :             FreeTupleDesc(relation->rd_att);
    2472             :     }
    2473      901440 :     FreeTriggerDesc(relation->trigdesc);
    2474      901440 :     list_free_deep(relation->rd_fkeylist);
    2475      901440 :     list_free(relation->rd_indexlist);
    2476      901440 :     list_free(relation->rd_statlist);
    2477      901440 :     bms_free(relation->rd_keyattr);
    2478      901440 :     bms_free(relation->rd_pkattr);
    2479      901440 :     bms_free(relation->rd_idattr);
    2480      901440 :     bms_free(relation->rd_hotblockingattr);
    2481      901440 :     bms_free(relation->rd_summarizedattr);
    2482      901440 :     if (relation->rd_pubdesc)
    2483        5994 :         pfree(relation->rd_pubdesc);
    2484      901440 :     if (relation->rd_options)
    2485        7986 :         pfree(relation->rd_options);
    2486      901440 :     if (relation->rd_indextuple)
    2487      293598 :         pfree(relation->rd_indextuple);
    2488      901440 :     if (relation->rd_amcache)
    2489           0 :         pfree(relation->rd_amcache);
    2490      901440 :     if (relation->rd_fdwroutine)
    2491         264 :         pfree(relation->rd_fdwroutine);
    2492      901440 :     if (relation->rd_indexcxt)
    2493      293598 :         MemoryContextDelete(relation->rd_indexcxt);
    2494      901440 :     if (relation->rd_rulescxt)
    2495       18284 :         MemoryContextDelete(relation->rd_rulescxt);
    2496      901440 :     if (relation->rd_rsdesc)
    2497        1754 :         MemoryContextDelete(relation->rd_rsdesc->rscxt);
    2498      901440 :     if (relation->rd_partkeycxt)
    2499       14742 :         MemoryContextDelete(relation->rd_partkeycxt);
    2500      901440 :     if (relation->rd_pdcxt)
    2501       14232 :         MemoryContextDelete(relation->rd_pdcxt);
    2502      901440 :     if (relation->rd_pddcxt)
    2503          60 :         MemoryContextDelete(relation->rd_pddcxt);
    2504      901440 :     if (relation->rd_partcheckcxt)
    2505        2592 :         MemoryContextDelete(relation->rd_partcheckcxt);
    2506      901440 :     pfree(relation);
    2507      901440 : }
    2508             : 
    2509             : /*
    2510             :  * RelationClearRelation
    2511             :  *
    2512             :  *   Physically blow away a relation cache entry, or reset it and rebuild
    2513             :  *   it from scratch (that is, from catalog entries).  The latter path is
    2514             :  *   used when we are notified of a change to an open relation (one with
    2515             :  *   refcount > 0).
    2516             :  *
    2517             :  *   NB: when rebuilding, we'd better hold some lock on the relation,
    2518             :  *   else the catalog data we need to read could be changing under us.
    2519             :  *   Also, a rel to be rebuilt had better have refcnt > 0.  This is because
    2520             :  *   a sinval reset could happen while we're accessing the catalogs, and
    2521             :  *   the rel would get blown away underneath us by RelationCacheInvalidate
    2522             :  *   if it has zero refcnt.
    2523             :  *
    2524             :  *   The "rebuild" parameter is redundant in current usage because it has
    2525             :  *   to match the relation's refcnt status, but we keep it as a crosscheck
    2526             :  *   that we're doing what the caller expects.
    2527             :  */
    2528             : static void
    2529     1137218 : RelationClearRelation(Relation relation, bool rebuild)
    2530             : {
    2531             :     /*
    2532             :      * As per notes above, a rel to be rebuilt MUST have refcnt > 0; while of
    2533             :      * course it would be an equally bad idea to blow away one with nonzero
    2534             :      * refcnt, since that would leave someone somewhere with a dangling
    2535             :      * pointer.  All callers are expected to have verified that this holds.
    2536             :      */
    2537             :     Assert(rebuild ?
    2538             :            !RelationHasReferenceCountZero(relation) :
    2539             :            RelationHasReferenceCountZero(relation));
    2540             : 
    2541             :     /*
    2542             :      * Make sure smgr and lower levels close the relation's files, if they
    2543             :      * weren't closed already.  If the relation is not getting deleted, the
    2544             :      * next smgr access should reopen the files automatically.  This ensures
    2545             :      * that the low-level file access state is updated after, say, a vacuum
    2546             :      * truncation.
    2547             :      */
    2548     1137218 :     RelationCloseSmgr(relation);
    2549             : 
    2550             :     /* Free AM cached data, if any */
    2551     1137218 :     if (relation->rd_amcache)
    2552       72210 :         pfree(relation->rd_amcache);
    2553     1137218 :     relation->rd_amcache = NULL;
    2554             : 
    2555             :     /*
    2556             :      * Treat nailed-in system relations separately, they always need to be
    2557             :      * accessible, so we can't blow them away.
    2558             :      */
    2559     1137218 :     if (relation->rd_isnailed)
    2560             :     {
    2561      194324 :         RelationReloadNailed(relation);
    2562      194318 :         return;
    2563             :     }
    2564             : 
    2565             :     /* Mark it invalid until we've finished rebuild */
    2566      942894 :     relation->rd_isvalid = false;
    2567             : 
    2568             :     /* See RelationForgetRelation(). */
    2569      942894 :     if (relation->rd_droppedSubid != InvalidSubTransactionId)
    2570        1636 :         return;
    2571             : 
    2572             :     /*
    2573             :      * Even non-system indexes should not be blown away if they are open and
    2574             :      * have valid index support information.  This avoids problems with active
    2575             :      * use of the index support information.  As with nailed indexes, we
    2576             :      * re-read the pg_class row to handle possible physical relocation of the
    2577             :      * index, and we check for pg_index updates too.
    2578             :      */
    2579      941258 :     if ((relation->rd_rel->relkind == RELKIND_INDEX ||
    2580      589832 :          relation->rd_rel->relkind == RELKIND_PARTITIONED_INDEX) &&
    2581      361080 :         relation->rd_refcnt > 0 &&
    2582       67478 :         relation->rd_indexcxt != NULL)
    2583             :     {
    2584       39746 :         if (IsTransactionState())
    2585       39746 :             RelationReloadIndexInfo(relation);
    2586       39746 :         return;
    2587             :     }
    2588             : 
    2589             :     /*
    2590             :      * If we're really done with the relcache entry, blow it away. But if
    2591             :      * someone is still using it, reconstruct the whole deal without moving
    2592             :      * the physical RelationData record (so that the someone's pointer is
    2593             :      * still valid).
    2594             :      */
    2595      901512 :     if (!rebuild)
    2596             :     {
    2597             :         /* Remove it from the hash table */
    2598      649878 :         RelationCacheDelete(relation);
    2599             : 
    2600             :         /* And release storage */
    2601      649878 :         RelationDestroyRelation(relation, false);
    2602             :     }
    2603      251634 :     else if (!IsTransactionState())
    2604             :     {
    2605             :         /*
    2606             :          * If we're not inside a valid transaction, we can't do any catalog
    2607             :          * access so it's not possible to rebuild yet.  Just exit, leaving
    2608             :          * rd_isvalid = false so that the rebuild will occur when the entry is
    2609             :          * next opened.
    2610             :          *
    2611             :          * Note: it's possible that we come here during subtransaction abort,
    2612             :          * and the reason for wanting to rebuild is that the rel is open in
    2613             :          * the outer transaction.  In that case it might seem unsafe to not
    2614             :          * rebuild immediately, since whatever code has the rel already open
    2615             :          * will keep on using the relcache entry as-is.  However, in such a
    2616             :          * case the outer transaction should be holding a lock that's
    2617             :          * sufficient to prevent any significant change in the rel's schema,
    2618             :          * so the existing entry contents should be good enough for its
    2619             :          * purposes; at worst we might be behind on statistics updates or the
    2620             :          * like.  (See also CheckTableNotInUse() and its callers.)  These same
    2621             :          * remarks also apply to the cases above where we exit without having
    2622             :          * done RelationReloadIndexInfo() yet.
    2623             :          */
    2624          72 :         return;
    2625             :     }
    2626             :     else
    2627             :     {
    2628             :         /*
    2629             :          * Our strategy for rebuilding an open relcache entry is to build a
    2630             :          * new entry from scratch, swap its contents with the old entry, and
    2631             :          * finally delete the new entry (along with any infrastructure swapped
    2632             :          * over from the old entry).  This is to avoid trouble in case an
    2633             :          * error causes us to lose control partway through.  The old entry
    2634             :          * will still be marked !rd_isvalid, so we'll try to rebuild it again
    2635             :          * on next access.  Meanwhile it's not any less valid than it was
    2636             :          * before, so any code that might expect to continue accessing it
    2637             :          * isn't hurt by the rebuild failure.  (Consider for example a
    2638             :          * subtransaction that ALTERs a table and then gets canceled partway
    2639             :          * through the cache entry rebuild.  The outer transaction should
    2640             :          * still see the not-modified cache entry as valid.)  The worst
    2641             :          * consequence of an error is leaking the necessarily-unreferenced new
    2642             :          * entry, and this shouldn't happen often enough for that to be a big
    2643             :          * problem.
    2644             :          *
    2645             :          * When rebuilding an open relcache entry, we must preserve ref count,
    2646             :          * rd_*Subid, and rd_toastoid state.  Also attempt to preserve the
    2647             :          * pg_class entry (rd_rel), tupledesc, rewrite-rule, partition key,
    2648             :          * and partition descriptor substructures in place, because various
    2649             :          * places assume that these structures won't move while they are
    2650             :          * working with an open relcache entry.  (Note:  the refcount
    2651             :          * mechanism for tupledescs might someday allow us to remove this hack
    2652             :          * for the tupledesc.)
    2653             :          *
    2654             :          * Note that this process does not touch CurrentResourceOwner; which
    2655             :          * is good because whatever ref counts the entry may have do not
    2656             :          * necessarily belong to that resource owner.
    2657             :          */
    2658             :         Relation    newrel;
    2659      251562 :         Oid         save_relid = RelationGetRelid(relation);
    2660             :         bool        keep_tupdesc;
    2661             :         bool        keep_rules;
    2662             :         bool        keep_policies;
    2663             :         bool        keep_partkey;
    2664             : 
    2665             :         /* Build temporary entry, but don't link it into hashtable */
    2666      251562 :         newrel = RelationBuildDesc(save_relid, false);
    2667             : 
    2668             :         /*
    2669             :          * Between here and the end of the swap, don't add code that does or
    2670             :          * reasonably could read system catalogs.  That range must be free
    2671             :          * from invalidation processing.  See RelationBuildDesc() manipulation
    2672             :          * of in_progress_list.
    2673             :          */
    2674             : 
    2675      251556 :         if (newrel == NULL)
    2676             :         {
    2677             :             /*
    2678             :              * We can validly get here, if we're using a historic snapshot in
    2679             :              * which a relation, accessed from outside logical decoding, is
    2680             :              * still invisible. In that case it's fine to just mark the
    2681             :              * relation as invalid and return - it'll fully get reloaded by
    2682             :              * the cache reset at the end of logical decoding (or at the next
    2683             :              * access).  During normal processing we don't want to ignore this
    2684             :              * case as it shouldn't happen there, as explained below.
    2685             :              */
    2686           0 :             if (HistoricSnapshotActive())
    2687           0 :                 return;
    2688             : 
    2689             :             /*
    2690             :              * This shouldn't happen as dropping a relation is intended to be
    2691             :              * impossible if still referenced (cf. CheckTableNotInUse()). But
    2692             :              * if we get here anyway, we can't just delete the relcache entry,
    2693             :              * as it possibly could get accessed later (as e.g. the error
    2694             :              * might get trapped and handled via a subtransaction rollback).
    2695             :              */
    2696           0 :             elog(ERROR, "relation %u deleted while still in use", save_relid);
    2697             :         }
    2698             : 
    2699             :         /*
    2700             :          * If we were to, again, have cases of the relkind of a relcache entry
    2701             :          * changing, we would need to ensure that pgstats does not get
    2702             :          * confused.
    2703             :          */
    2704             :         Assert(relation->rd_rel->relkind == newrel->rd_rel->relkind);
    2705             : 
    2706      251556 :         keep_tupdesc = equalTupleDescs(relation->rd_att, newrel->rd_att);
    2707      251556 :         keep_rules = equalRuleLocks(relation->rd_rules, newrel->rd_rules);
    2708      251556 :         keep_policies = equalRSDesc(relation->rd_rsdesc, newrel->rd_rsdesc);
    2709             :         /* partkey is immutable once set up, so we can always keep it */
    2710      251556 :         keep_partkey = (relation->rd_partkey != NULL);
    2711             : 
    2712             :         /*
    2713             :          * Perform swapping of the relcache entry contents.  Within this
    2714             :          * process the old entry is momentarily invalid, so there *must* be no
    2715             :          * possibility of CHECK_FOR_INTERRUPTS within this sequence. Do it in
    2716             :          * all-in-line code for safety.
    2717             :          *
    2718             :          * Since the vast majority of fields should be swapped, our method is
    2719             :          * to swap the whole structures and then re-swap those few fields we
    2720             :          * didn't want swapped.
    2721             :          */
    2722             : #define SWAPFIELD(fldtype, fldname) \
    2723             :         do { \
    2724             :             fldtype _tmp = newrel->fldname; \
    2725             :             newrel->fldname = relation->fldname; \
    2726             :             relation->fldname = _tmp; \
    2727             :         } while (0)
    2728             : 
    2729             :         /* swap all Relation struct fields */
    2730             :         {
    2731             :             RelationData tmpstruct;
    2732             : 
    2733      251556 :             memcpy(&tmpstruct, newrel, sizeof(RelationData));
    2734      251556 :             memcpy(newrel, relation, sizeof(RelationData));
    2735      251556 :             memcpy(relation, &tmpstruct, sizeof(RelationData));
    2736             :         }
    2737             : 
    2738             :         /* rd_smgr must not be swapped, due to back-links from smgr level */
    2739      251556 :         SWAPFIELD(SMgrRelation, rd_smgr);
    2740             :         /* rd_refcnt must be preserved */
    2741      251556 :         SWAPFIELD(int, rd_refcnt);
    2742             :         /* isnailed shouldn't change */
    2743             :         Assert(newrel->rd_isnailed == relation->rd_isnailed);
    2744             :         /* creation sub-XIDs must be preserved */
    2745      251556 :         SWAPFIELD(SubTransactionId, rd_createSubid);
    2746      251556 :         SWAPFIELD(SubTransactionId, rd_newRelfilelocatorSubid);
    2747      251556 :         SWAPFIELD(SubTransactionId, rd_firstRelfilelocatorSubid);
    2748      251556 :         SWAPFIELD(SubTransactionId, rd_droppedSubid);
    2749             :         /* un-swap rd_rel pointers, swap contents instead */
    2750      251556 :         SWAPFIELD(Form_pg_class, rd_rel);
    2751             :         /* ... but actually, we don't have to update newrel->rd_rel */
    2752      251556 :         memcpy(relation->rd_rel, newrel->rd_rel, CLASS_TUPLE_SIZE);
    2753             :         /* preserve old tupledesc, rules, policies if no logical change */
    2754      251556 :         if (keep_tupdesc)
    2755      234378 :             SWAPFIELD(TupleDesc, rd_att);
    2756      251556 :         if (keep_rules)
    2757             :         {
    2758      239826 :             SWAPFIELD(RuleLock *, rd_rules);
    2759      239826 :             SWAPFIELD(MemoryContext, rd_rulescxt);
    2760             :         }
    2761      251556 :         if (keep_policies)
    2762      251268 :             SWAPFIELD(RowSecurityDesc *, rd_rsdesc);
    2763             :         /* toast OID override must be preserved */
    2764      251556 :         SWAPFIELD(Oid, rd_toastoid);
    2765             :         /* pgstat_info / enabled must be preserved */
    2766      251556 :         SWAPFIELD(struct PgStat_TableStatus *, pgstat_info);
    2767      251556 :         SWAPFIELD(bool, pgstat_enabled);
    2768             :         /* preserve old partition key if we have one */
    2769      251556 :         if (keep_partkey)
    2770             :         {
    2771       10608 :             SWAPFIELD(PartitionKey, rd_partkey);
    2772       10608 :             SWAPFIELD(MemoryContext, rd_partkeycxt);
    2773             :         }
    2774      251556 :         if (newrel->rd_pdcxt != NULL || newrel->rd_pddcxt != NULL)
    2775             :         {
    2776             :             /*
    2777             :              * We are rebuilding a partitioned relation with a non-zero
    2778             :              * reference count, so we must keep the old partition descriptor
    2779             :              * around, in case there's a PartitionDirectory with a pointer to
    2780             :              * it.  This means we can't free the old rd_pdcxt yet.  (This is
    2781             :              * necessary because RelationGetPartitionDesc hands out direct
    2782             :              * pointers to the relcache's data structure, unlike our usual
    2783             :              * practice which is to hand out copies.  We'd have the same
    2784             :              * problem with rd_partkey, except that we always preserve that
    2785             :              * once created.)
    2786             :              *
    2787             :              * To ensure that it's not leaked completely, re-attach it to the
    2788             :              * new reldesc, or make it a child of the new reldesc's rd_pdcxt
    2789             :              * in the unlikely event that there is one already.  (Compare hack
    2790             :              * in RelationBuildPartitionDesc.)  RelationClose will clean up
    2791             :              * any such contexts once the reference count reaches zero.
    2792             :              *
    2793             :              * In the case where the reference count is zero, this code is not
    2794             :              * reached, which should be OK because in that case there should
    2795             :              * be no PartitionDirectory with a pointer to the old entry.
    2796             :              *
    2797             :              * Note that newrel and relation have already been swapped, so the
    2798             :              * "old" partition descriptor is actually the one hanging off of
    2799             :              * newrel.
    2800             :              */
    2801        8408 :             relation->rd_partdesc = NULL;    /* ensure rd_partdesc is invalid */
    2802        8408 :             relation->rd_partdesc_nodetached = NULL;
    2803        8408 :             relation->rd_partdesc_nodetached_xmin = InvalidTransactionId;
    2804        8408 :             if (relation->rd_pdcxt != NULL) /* probably never happens */
    2805           0 :                 MemoryContextSetParent(newrel->rd_pdcxt, relation->rd_pdcxt);
    2806             :             else
    2807        8408 :                 relation->rd_pdcxt = newrel->rd_pdcxt;
    2808        8408 :             if (relation->rd_pddcxt != NULL)
    2809           0 :                 MemoryContextSetParent(newrel->rd_pddcxt, relation->rd_pddcxt);
    2810             :             else
    2811        8408 :                 relation->rd_pddcxt = newrel->rd_pddcxt;
    2812             :             /* drop newrel's pointers so we don't destroy it below */
    2813        8408 :             newrel->rd_partdesc = NULL;
    2814        8408 :             newrel->rd_partdesc_nodetached = NULL;
    2815        8408 :             newrel->rd_partdesc_nodetached_xmin = InvalidTransactionId;
    2816        8408 :             newrel->rd_pdcxt = NULL;
    2817        8408 :             newrel->rd_pddcxt = NULL;
    2818             :         }
    2819             : 
    2820             : #undef SWAPFIELD
    2821             : 
    2822             :         /* And now we can throw away the temporary entry */
    2823      251556 :         RelationDestroyRelation(newrel, !keep_tupdesc);
    2824             :     }
    2825             : }
    2826             : 
    2827             : /*
    2828             :  * RelationFlushRelation
    2829             :  *
    2830             :  *   Rebuild the relation if it is open (refcount > 0), else blow it away.
    2831             :  *   This is used when we receive a cache invalidation event for the rel.
    2832             :  */
    2833             : static void
    2834      534292 : RelationFlushRelation(Relation relation)
    2835             : {
    2836      534292 :     if (relation->rd_createSubid != InvalidSubTransactionId ||
    2837      311942 :         relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
    2838             :     {
    2839             :         /*
    2840             :          * New relcache entries are always rebuilt, not flushed; else we'd
    2841             :          * forget the "new" status of the relation.  Ditto for the
    2842             :          * new-relfilenumber status.
    2843             :          *
    2844             :          * The rel could have zero refcnt here, so temporarily increment the
    2845             :          * refcnt to ensure it's safe to rebuild it.  We can assume that the
    2846             :          * current transaction has some lock on the rel already.
    2847             :          */
    2848      232554 :         RelationIncrementReferenceCount(relation);
    2849      232554 :         RelationClearRelation(relation, true);
    2850      232548 :         RelationDecrementReferenceCount(relation);
    2851             :     }
    2852             :     else
    2853             :     {
    2854             :         /*
    2855             :          * Pre-existing rels can be dropped from the relcache if not open.
    2856             :          */
    2857      301738 :         bool        rebuild = !RelationHasReferenceCountZero(relation);
    2858             : 
    2859      301738 :         RelationClearRelation(relation, rebuild);
    2860             :     }
    2861      534286 : }
    2862             : 
    2863             : /*
    2864             :  * RelationForgetRelation - caller reports that it dropped the relation
    2865             :  */
    2866             : void
    2867       59682 : RelationForgetRelation(Oid rid)
    2868             : {
    2869             :     Relation    relation;
    2870             : 
    2871       59682 :     RelationIdCacheLookup(rid, relation);
    2872             : 
    2873       59682 :     if (!PointerIsValid(relation))
    2874           0 :         return;                 /* not in cache, nothing to do */
    2875             : 
    2876       59682 :     if (!RelationHasReferenceCountZero(relation))
    2877           0 :         elog(ERROR, "relation %u is still open", rid);
    2878             : 
    2879             :     Assert(relation->rd_droppedSubid == InvalidSubTransactionId);
    2880       59682 :     if (relation->rd_createSubid != InvalidSubTransactionId ||
    2881       58892 :         relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
    2882             :     {
    2883             :         /*
    2884             :          * In the event of subtransaction rollback, we must not forget
    2885             :          * rd_*Subid.  Mark the entry "dropped" so RelationClearRelation()
    2886             :          * invalidates it in lieu of destroying it.  (If we're in a top
    2887             :          * transaction, we could opt to destroy the entry.)
    2888             :          */
    2889         818 :         relation->rd_droppedSubid = GetCurrentSubTransactionId();
    2890             :     }
    2891             : 
    2892       59682 :     RelationClearRelation(relation, false);
    2893             : }
    2894             : 
    2895             : /*
    2896             :  *      RelationCacheInvalidateEntry
    2897             :  *
    2898             :  *      This routine is invoked for SI cache flush messages.
    2899             :  *
    2900             :  * Any relcache entry matching the relid must be flushed.  (Note: caller has
    2901             :  * already determined that the relid belongs to our database or is a shared
    2902             :  * relation.)
    2903             :  *
    2904             :  * We used to skip local relations, on the grounds that they could
    2905             :  * not be targets of cross-backend SI update messages; but it seems
    2906             :  * safer to process them, so that our *own* SI update messages will
    2907             :  * have the same effects during CommandCounterIncrement for both
    2908             :  * local and nonlocal relations.
    2909             :  */
    2910             : void
    2911     1854720 : RelationCacheInvalidateEntry(Oid relationId)
    2912             : {
    2913             :     Relation    relation;
    2914             : 
    2915     1854720 :     RelationIdCacheLookup(relationId, relation);
    2916             : 
    2917     1854720 :     if (PointerIsValid(relation))
    2918             :     {
    2919      534292 :         relcacheInvalsReceived++;
    2920      534292 :         RelationFlushRelation(relation);
    2921             :     }
    2922             :     else
    2923             :     {
    2924             :         int         i;
    2925             : 
    2926     1337648 :         for (i = 0; i < in_progress_list_len; i++)
    2927       17220 :             if (in_progress_list[i].reloid == relationId)
    2928           4 :                 in_progress_list[i].invalidated = true;
    2929             :     }
    2930     1854714 : }
    2931             : 
    2932             : /*
    2933             :  * RelationCacheInvalidate
    2934             :  *   Blow away cached relation descriptors that have zero reference counts,
    2935             :  *   and rebuild those with positive reference counts.  Also reset the smgr
    2936             :  *   relation cache and re-read relation mapping data.
    2937             :  *
    2938             :  *   Apart from debug_discard_caches, this is currently used only to recover
    2939             :  *   from SI message buffer overflow, so we do not touch relations having
    2940             :  *   new-in-transaction relfilenumbers; they cannot be targets of cross-backend
    2941             :  *   SI updates (and our own updates now go through a separate linked list
    2942             :  *   that isn't limited by the SI message buffer size).
    2943             :  *
    2944             :  *   We do this in two phases: the first pass deletes deletable items, and
    2945             :  *   the second one rebuilds the rebuildable items.  This is essential for
    2946             :  *   safety, because hash_seq_search only copes with concurrent deletion of
    2947             :  *   the element it is currently visiting.  If a second SI overflow were to
    2948             :  *   occur while we are walking the table, resulting in recursive entry to
    2949             :  *   this routine, we could crash because the inner invocation blows away
    2950             :  *   the entry next to be visited by the outer scan.  But this way is OK,
    2951             :  *   because (a) during the first pass we won't process any more SI messages,
    2952             :  *   so hash_seq_search will complete safely; (b) during the second pass we
    2953             :  *   only hold onto pointers to nondeletable entries.
    2954             :  *
    2955             :  *   The two-phase approach also makes it easy to update relfilenumbers for
    2956             :  *   mapped relations before we do anything else, and to ensure that the
    2957             :  *   second pass processes nailed-in-cache items before other nondeletable
    2958             :  *   items.  This should ensure that system catalogs are up to date before
    2959             :  *   we attempt to use them to reload information about other open relations.
    2960             :  *
    2961             :  *   After those two phases of work having immediate effects, we normally
    2962             :  *   signal any RelationBuildDesc() on the stack to start over.  However, we
    2963             :  *   don't do this if called as part of debug_discard_caches.  Otherwise,
    2964             :  *   RelationBuildDesc() would become an infinite loop.
    2965             :  */
    2966             : void
    2967        4086 : RelationCacheInvalidate(bool debug_discard)
    2968             : {
    2969             :     HASH_SEQ_STATUS status;
    2970             :     RelIdCacheEnt *idhentry;
    2971             :     Relation    relation;
    2972        4086 :     List       *rebuildFirstList = NIL;
    2973        4086 :     List       *rebuildList = NIL;
    2974             :     ListCell   *l;
    2975             :     int         i;
    2976             : 
    2977             :     /*
    2978             :      * Reload relation mapping data before starting to reconstruct cache.
    2979             :      */
    2980        4086 :     RelationMapInvalidateAll();
    2981             : 
    2982             :     /* Phase 1 */
    2983        4086 :     hash_seq_init(&status, RelationIdCache);
    2984             : 
    2985      446980 :     while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
    2986             :     {
    2987      442894 :         relation = idhentry->reldesc;
    2988             : 
    2989             :         /* Must close all smgr references to avoid leaving dangling ptrs */
    2990      442894 :         RelationCloseSmgr(relation);
    2991             : 
    2992             :         /*
    2993             :          * Ignore new relations; no other backend will manipulate them before
    2994             :          * we commit.  Likewise, before replacing a relation's relfilelocator,
    2995             :          * we shall have acquired AccessExclusiveLock and drained any
    2996             :          * applicable pending invalidations.
    2997             :          */
    2998      442894 :         if (relation->rd_createSubid != InvalidSubTransactionId ||
    2999      442878 :             relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId)
    3000          18 :             continue;
    3001             : 
    3002      442876 :         relcacheInvalsReceived++;
    3003             : 
    3004      442876 :         if (RelationHasReferenceCountZero(relation))
    3005             :         {
    3006             :             /* Delete this entry immediately */
    3007             :             Assert(!relation->rd_isnailed);
    3008      354124 :             RelationClearRelation(relation, false);
    3009             :         }
    3010             :         else
    3011             :         {
    3012             :             /*
    3013             :              * If it's a mapped relation, immediately update its rd_locator in
    3014             :              * case its relfilenumber changed.  We must do this during phase 1
    3015             :              * in case the relation is consulted during rebuild of other
    3016             :              * relcache entries in phase 2.  It's safe since consulting the
    3017             :              * map doesn't involve any access to relcache entries.
    3018             :              */
    3019       88752 :             if (RelationIsMapped(relation))
    3020       68590 :                 RelationInitPhysicalAddr(relation);
    3021             : 
    3022             :             /*
    3023             :              * Add this entry to list of stuff to rebuild in second pass.
    3024             :              * pg_class goes to the front of rebuildFirstList while
    3025             :              * pg_class_oid_index goes to the back of rebuildFirstList, so
    3026             :              * they are done first and second respectively.  Other nailed
    3027             :              * relations go to the front of rebuildList, so they'll be done
    3028             :              * next in no particular order; and everything else goes to the
    3029             :              * back of rebuildList.
    3030             :              */
    3031       88752 :             if (RelationGetRelid(relation) == RelationRelationId)
    3032        4014 :                 rebuildFirstList = lcons(relation, rebuildFirstList);
    3033       84738 :             else if (RelationGetRelid(relation) == ClassOidIndexId)
    3034        4010 :                 rebuildFirstList = lappend(rebuildFirstList, relation);
    3035       80728 :             else if (relation->rd_isnailed)
    3036       80616 :                 rebuildList = lcons(relation, rebuildList);
    3037             :             else
    3038         112 :                 rebuildList = lappend(rebuildList, relation);
    3039             :         }
    3040             :     }
    3041             : 
    3042             :     /*
    3043             :      * Now zap any remaining smgr cache entries.  This must happen before we
    3044             :      * start to rebuild entries, since that may involve catalog fetches which
    3045             :      * will re-open catalog files.
    3046             :      */
    3047        4086 :     smgrcloseall();
    3048             : 
    3049             :     /* Phase 2: rebuild the items found to need rebuild in phase 1 */
    3050       12110 :     foreach(l, rebuildFirstList)
    3051             :     {
    3052        8024 :         relation = (Relation) lfirst(l);
    3053        8024 :         RelationClearRelation(relation, true);
    3054             :     }
    3055        4086 :     list_free(rebuildFirstList);
    3056       84814 :     foreach(l, rebuildList)
    3057             :     {
    3058       80728 :         relation = (Relation) lfirst(l);
    3059       80728 :         RelationClearRelation(relation, true);
    3060             :     }
    3061        4086 :     list_free(rebuildList);
    3062             : 
    3063        4086 :     if (!debug_discard)
    3064             :         /* Any RelationBuildDesc() on the stack must start over. */
    3065        4092 :         for (i = 0; i < in_progress_list_len; i++)
    3066           6 :             in_progress_list[i].invalidated = true;
    3067        4086 : }
    3068             : 
    3069             : /*
    3070             :  * RelationCloseSmgrByOid - close a relcache entry's smgr link
    3071             :  *
    3072             :  * Needed in some cases where we are changing a relation's physical mapping.
    3073             :  * The link will be automatically reopened on next use.
    3074             :  */
    3075             : void
    3076        3400 : RelationCloseSmgrByOid(Oid relationId)
    3077             : {
    3078             :     Relation    relation;
    3079             : 
    3080        3400 :     RelationIdCacheLookup(relationId, relation);
    3081             : 
    3082        3400 :     if (!PointerIsValid(relation))
    3083           0 :         return;                 /* not in cache, nothing to do */
    3084             : 
    3085        3400 :     RelationCloseSmgr(relation);
    3086             : }
    3087             : 
    3088             : static void
    3089       16974 : RememberToFreeTupleDescAtEOX(TupleDesc td)
    3090             : {
    3091       16974 :     if (EOXactTupleDescArray == NULL)
    3092             :     {
    3093             :         MemoryContext oldcxt;
    3094             : 
    3095       10820 :         oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    3096             : 
    3097       10820 :         EOXactTupleDescArray = (TupleDesc *) palloc(16 * sizeof(TupleDesc));
    3098       10820 :         EOXactTupleDescArrayLen = 16;
    3099       10820 :         NextEOXactTupleDescNum = 0;
    3100       10820 :         MemoryContextSwitchTo(oldcxt);
    3101             :     }
    3102        6154 :     else if (NextEOXactTupleDescNum >= EOXactTupleDescArrayLen)
    3103             :     {
    3104          12 :         int32       newlen = EOXactTupleDescArrayLen * 2;
    3105             : 
    3106             :         Assert(EOXactTupleDescArrayLen > 0);
    3107             : 
    3108          12 :         EOXactTupleDescArray = (TupleDesc *) repalloc(EOXactTupleDescArray,
    3109             :                                                       newlen * sizeof(TupleDesc));
    3110          12 :         EOXactTupleDescArrayLen = newlen;
    3111             :     }
    3112             : 
    3113       16974 :     EOXactTupleDescArray[NextEOXactTupleDescNum++] = td;
    3114       16974 : }
    3115             : 
    3116             : #ifdef USE_ASSERT_CHECKING
    3117             : static void
    3118             : AssertPendingSyncConsistency(Relation relation)
    3119             : {
    3120             :     bool        relcache_verdict =
    3121             :         RelationIsPermanent(relation) &&
    3122             :         ((relation->rd_createSubid != InvalidSubTransactionId &&
    3123             :           RELKIND_HAS_STORAGE(relation->rd_rel->relkind)) ||
    3124             :          relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId);
    3125             : 
    3126             :     Assert(relcache_verdict == RelFileLocatorSkippingWAL(relation->rd_locator));
    3127             : 
    3128             :     if (relation->rd_droppedSubid != InvalidSubTransactionId)
    3129             :         Assert(!relation->rd_isvalid &&
    3130             :                (relation->rd_createSubid != InvalidSubTransactionId ||
    3131             :                 relation->rd_firstRelfilelocatorSubid != InvalidSubTransactionId));
    3132             : }
    3133             : 
    3134             : /*
    3135             :  * AssertPendingSyncs_RelationCache
    3136             :  *
    3137             :  *  Assert that relcache.c and storage.c agree on whether to skip WAL.
    3138             :  */
    3139             : void
    3140             : AssertPendingSyncs_RelationCache(void)
    3141             : {
    3142             :     HASH_SEQ_STATUS status;
    3143             :     LOCALLOCK  *locallock;
    3144             :     Relation   *rels;
    3145             :     int         maxrels;
    3146             :     int         nrels;
    3147             :     RelIdCacheEnt *idhentry;
    3148             :     int         i;
    3149             : 
    3150             :     /*
    3151             :      * Open every relation that this transaction has locked.  If, for some
    3152             :      * relation, storage.c is skipping WAL and relcache.c is not skipping WAL,
    3153             :      * a CommandCounterIncrement() typically yields a local invalidation
    3154             :      * message that destroys the relcache entry.  By recreating such entries
    3155             :      * here, we detect the problem.
    3156             :      */
    3157             :     PushActiveSnapshot(GetTransactionSnapshot());
    3158             :     maxrels = 1;
    3159             :     rels = palloc(maxrels * sizeof(*rels));
    3160             :     nrels = 0;
    3161             :     hash_seq_init(&status, GetLockMethodLocalHash());
    3162             :     while ((locallock = (LOCALLOCK *) hash_seq_search(&status)) != NULL)
    3163             :     {
    3164             :         Oid         relid;
    3165             :         Relation    r;
    3166             : 
    3167             :         if (locallock->nLocks <= 0)
    3168             :             continue;
    3169             :         if ((LockTagType) locallock->tag.lock.locktag_type !=
    3170             :             LOCKTAG_RELATION)
    3171             :             continue;
    3172             :         relid = ObjectIdGetDatum(locallock->tag.lock.locktag_field2);
    3173             :         r = RelationIdGetRelation(relid);
    3174             :         if (!RelationIsValid(r))
    3175             :             continue;
    3176             :         if (nrels >= maxrels)
    3177             :         {
    3178             :             maxrels *= 2;
    3179             :             rels = repalloc(rels, maxrels * sizeof(*rels));
    3180             :         }
    3181             :         rels[nrels++] = r;
    3182             :     }
    3183             : 
    3184             :     hash_seq_init(&status, RelationIdCache);
    3185             :     while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
    3186             :         AssertPendingSyncConsistency(idhentry->reldesc);
    3187             : 
    3188             :     for (i = 0; i < nrels; i++)
    3189             :         RelationClose(rels[i]);
    3190             :     PopActiveSnapshot();
    3191             : }
    3192             : #endif
    3193             : 
    3194             : /*
    3195             :  * AtEOXact_RelationCache
    3196             :  *
    3197             :  *  Clean up the relcache at main-transaction commit or abort.
    3198             :  *
    3199             :  * Note: this must be called *before* processing invalidation messages.
    3200             :  * In the case of abort, we don't want to try to rebuild any invalidated
    3201             :  * cache entries (since we can't safely do database accesses).  Therefore
    3202             :  * we must reset refcnts before handling pending invalidations.
    3203             :  *
    3204             :  * As of PostgreSQL 8.1, relcache refcnts should get released by the
    3205             :  * ResourceOwner mechanism.  This routine just does a debugging
    3206             :  * cross-check that no pins remain.  However, we also need to do special
    3207             :  * cleanup when the current transaction created any relations or made use
    3208             :  * of forced index lists.
    3209             :  */
    3210             : void
    3211      515548 : AtEOXact_RelationCache(bool isCommit)
    3212             : {
    3213             :     HASH_SEQ_STATUS status;
    3214             :     RelIdCacheEnt *idhentry;
    3215             :     int         i;
    3216             : 
    3217             :     /*
    3218             :      * Forget in_progress_list.  This is relevant when we're aborting due to
    3219             :      * an error during RelationBuildDesc().
    3220             :      */
    3221             :     Assert(in_progress_list_len == 0 || !isCommit);
    3222      515548 :     in_progress_list_len = 0;
    3223             : 
    3224             :     /*
    3225             :      * Unless the eoxact_list[] overflowed, we only need to examine the rels
    3226             :      * listed in it.  Otherwise fall back on a hash_seq_search scan.
    3227             :      *
    3228             :      * For simplicity, eoxact_list[] entries are not deleted till end of
    3229             :      * top-level transaction, even though we could remove them at
    3230             :      * subtransaction end in some cases, or remove relations from the list if
    3231             :      * they are cleared for other reasons.  Therefore we should expect the
    3232             :      * case that list entries are not found in the hashtable; if not, there's
    3233             :      * nothing to do for them.
    3234             :      */
    3235      515548 :     if (eoxact_list_overflowed)
    3236             :     {
    3237          70 :         hash_seq_init(&status, RelationIdCache);
    3238       18078 :         while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
    3239             :         {
    3240       18008 :             AtEOXact_cleanup(idhentry->reldesc, isCommit);
    3241             :         }
    3242             :     }
    3243             :     else
    3244             :     {
    3245      614748 :         for (i = 0; i < eoxact_list_len; i++)
    3246             :         {
    3247       99270 :             idhentry = (RelIdCacheEnt *) hash_search(RelationIdCache,
    3248       99270 :                                                      &eoxact_list[i],
    3249             :                                                      HASH_FIND,
    3250             :                                                      NULL);
    3251       99270 :             if (idhentry != NULL)
    3252       97398 :                 AtEOXact_cleanup(idhentry->reldesc, isCommit);
    3253             :         }
    3254             :     }
    3255             : 
    3256      515548 :     if (EOXactTupleDescArrayLen > 0)
    3257             :     {
    3258             :         Assert(EOXactTupleDescArray != NULL);
    3259       27794 :         for (i = 0; i < NextEOXactTupleDescNum; i++)
    3260       16974 :             FreeTupleDesc(EOXactTupleDescArray[i]);
    3261       10820 :         pfree(EOXactTupleDescArray);
    3262       10820 :         EOXactTupleDescArray = NULL;
    3263             :     }
    3264             : 
    3265             :     /* Now we're out of the transaction and can clear the lists */
    3266      515548 :     eoxact_list_len = 0;
    3267      515548 :     eoxact_list_overflowed = false;
    3268      515548 :     NextEOXactTupleDescNum = 0;
    3269      515548 :     EOXactTupleDescArrayLen = 0;
    3270      515548 : }
    3271             : 
    3272             : /*
    3273             :  * AtEOXact_cleanup
    3274             :  *
    3275             :  *  Clean up a single rel at main-transaction commit or abort
    3276             :  *
    3277             :  * NB: this processing must be idempotent, because EOXactListAdd() doesn't
    3278             :  * bother to prevent duplicate entries in eoxact_list[].
    3279             :  */
    3280             : static void
    3281      115406 : AtEOXact_cleanup(Relation relation, bool isCommit)
    3282             : {
    3283      115406 :     bool        clear_relcache = false;
    3284             : 
    3285             :     /*
    3286             :      * The relcache entry's ref count should be back to its normal
    3287             :      * not-in-a-transaction state: 0 unless it's nailed in cache.
    3288             :      *
    3289             :      * In bootstrap mode, this is NOT true, so don't check it --- the
    3290             :      * bootstrap code expects relations to stay open across start/commit
    3291             :      * transaction calls.  (That seems bogus, but it's not worth fixing.)
    3292             :      *
    3293             :      * Note: ideally this check would be applied to every relcache entry, not
    3294             :      * just those that have eoxact work to do.  But it's not worth forcing a
    3295             :      * scan of the whole relcache just for this.  (Moreover, doing so would
    3296             :      * mean that assert-enabled testing never tests the hash_search code path
    3297             :      * above, which seems a bad idea.)
    3298             :      */
    3299             : #ifdef USE_ASSERT_CHECKING
    3300             :     if (!IsBootstrapProcessingMode())
    3301             :     {
    3302             :         int         expected_refcnt;
    3303             : 
    3304             :         expected_refcnt = relation->rd_isnailed ? 1 : 0;
    3305             :         Assert(relation->rd_refcnt == expected_refcnt);
    3306             :     }
    3307             : #endif
    3308             : 
    3309             :     /*
    3310             :      * Is the relation live after this transaction ends?
    3311             :      *
    3312             :      * During commit, clear the relcache entry if it is preserved after
    3313             :      * relation drop, in order not to orphan the entry.  During rollback,
    3314             :      * clear the relcache entry if the relation is created in the current
    3315             :      * transaction since it isn't interesting any longer once we are out of
    3316             :      * the transaction.
    3317             :      */
    3318      115406 :     clear_relcache =
    3319             :         (isCommit ?
    3320      115406 :          relation->rd_droppedSubid != InvalidSubTransactionId :
    3321        3492 :          relation->rd_createSubid != InvalidSubTransactionId);
    3322             : 
    3323             :     /*
    3324             :      * Since we are now out of the transaction, reset the subids to zero. That
    3325             :      * also lets RelationClearRelation() drop the relcache entry.
    3326             :      */
    3327      115406 :     relation->rd_createSubid = InvalidSubTransactionId;
    3328      115406 :     relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    3329      115406 :     relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    3330      115406 :     relation->rd_droppedSubid = InvalidSubTransactionId;
    3331             : 
    3332      115406 :     if (clear_relcache)
    3333             :     {
    3334        3662 :         if (RelationHasReferenceCountZero(relation))
    3335             :         {
    3336        3662 :             RelationClearRelation(relation, false);
    3337        3662 :             return;
    3338             :         }
    3339             :         else
    3340             :         {
    3341             :             /*
    3342             :              * Hmm, somewhere there's a (leaked?) reference to the relation.
    3343             :              * We daren't remove the entry for fear of dereferencing a
    3344             :              * dangling pointer later.  Bleat, and mark it as not belonging to
    3345             :              * the current transaction.  Hopefully it'll get cleaned up
    3346             :              * eventually.  This must be just a WARNING to avoid
    3347             :              * error-during-error-recovery loops.
    3348             :              */
    3349           0 :             elog(WARNING, "cannot remove relcache entry for \"%s\" because it has nonzero refcount",
    3350             :                  RelationGetRelationName(relation));
    3351             :         }
    3352             :     }
    3353             : }
    3354             : 
    3355             : /*
    3356             :  * AtEOSubXact_RelationCache
    3357             :  *
    3358             :  *  Clean up the relcache at sub-transaction commit or abort.
    3359             :  *
    3360             :  * Note: this must be called *before* processing invalidation messages.
    3361             :  */
    3362             : void
    3363       18080 : AtEOSubXact_RelationCache(bool isCommit, SubTransactionId mySubid,
    3364             :                           SubTransactionId parentSubid)
    3365             : {
    3366             :     HASH_SEQ_STATUS status;
    3367             :     RelIdCacheEnt *idhentry;
    3368             :     int         i;
    3369             : 
    3370             :     /*
    3371             :      * Forget in_progress_list.  This is relevant when we're aborting due to
    3372             :      * an error during RelationBuildDesc().  We don't commit subtransactions
    3373             :      * during RelationBuildDesc().
    3374             :      */
    3375             :     Assert(in_progress_list_len == 0 || !isCommit);
    3376       18080 :     in_progress_list_len = 0;
    3377             : 
    3378             :     /*
    3379             :      * Unless the eoxact_list[] overflowed, we only need to examine the rels
    3380             :      * listed in it.  Otherwise fall back on a hash_seq_search scan.  Same
    3381             :      * logic as in AtEOXact_RelationCache.
    3382             :      */
    3383       18080 :     if (eoxact_list_overflowed)
    3384             :     {
    3385           0 :         hash_seq_init(&status, RelationIdCache);
    3386           0 :         while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
    3387             :         {
    3388           0 :             AtEOSubXact_cleanup(idhentry->reldesc, isCommit,
    3389             :                                 mySubid, parentSubid);
    3390             :         }
    3391             :     }
    3392             :     else
    3393             :     {
    3394       27420 :         for (i = 0; i < eoxact_list_len; i++)
    3395             :         {
    3396        9340 :             idhentry = (RelIdCacheEnt *) hash_search(RelationIdCache,
    3397        9340 :                                                      &eoxact_list[i],
    3398             :                                                      HASH_FIND,
    3399             :                                                      NULL);
    3400        9340 :             if (idhentry != NULL)
    3401        8396 :                 AtEOSubXact_cleanup(idhentry->reldesc, isCommit,
    3402             :                                     mySubid, parentSubid);
    3403             :         }
    3404             :     }
    3405             : 
    3406             :     /* Don't reset the list; we still need more cleanup later */
    3407       18080 : }
    3408             : 
    3409             : /*
    3410             :  * AtEOSubXact_cleanup
    3411             :  *
    3412             :  *  Clean up a single rel at subtransaction commit or abort
    3413             :  *
    3414             :  * NB: this processing must be idempotent, because EOXactListAdd() doesn't
    3415             :  * bother to prevent duplicate entries in eoxact_list[].
    3416             :  */
    3417             : static void
    3418        8396 : AtEOSubXact_cleanup(Relation relation, bool isCommit,
    3419             :                     SubTransactionId mySubid, SubTransactionId parentSubid)
    3420             : {
    3421             :     /*
    3422             :      * Is it a relation created in the current subtransaction?
    3423             :      *
    3424             :      * During subcommit, mark it as belonging to the parent, instead, as long
    3425             :      * as it has not been dropped. Otherwise simply delete the relcache entry.
    3426             :      * --- it isn't interesting any longer.
    3427             :      */
    3428        8396 :     if (relation->rd_createSubid == mySubid)
    3429             :     {
    3430             :         /*
    3431             :          * Valid rd_droppedSubid means the corresponding relation is dropped
    3432             :          * but the relcache entry is preserved for at-commit pending sync. We
    3433             :          * need to drop it explicitly here not to make the entry orphan.
    3434             :          */
    3435             :         Assert(relation->rd_droppedSubid == mySubid ||
    3436             :                relation->rd_droppedSubid == InvalidSubTransactionId);
    3437         198 :         if (isCommit && relation->rd_droppedSubid == InvalidSubTransactionId)
    3438          74 :             relation->rd_createSubid = parentSubid;
    3439         124 :         else if (RelationHasReferenceCountZero(relation))
    3440             :         {
    3441             :             /* allow the entry to be removed */
    3442         124 :             relation->rd_createSubid = InvalidSubTransactionId;
    3443         124 :             relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    3444         124 :             relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    3445         124 :             relation->rd_droppedSubid = InvalidSubTransactionId;
    3446         124 :             RelationClearRelation(relation, false);
    3447         124 :             return;
    3448             :         }
    3449             :         else
    3450             :         {
    3451             :             /*
    3452             :              * Hmm, somewhere there's a (leaked?) reference to the relation.
    3453             :              * We daren't remove the entry for fear of dereferencing a
    3454             :              * dangling pointer later.  Bleat, and transfer it to the parent
    3455             :              * subtransaction so we can try again later.  This must be just a
    3456             :              * WARNING to avoid error-during-error-recovery loops.
    3457             :              */
    3458           0 :             relation->rd_createSubid = parentSubid;
    3459           0 :             elog(WARNING, "cannot remove relcache entry for \"%s\" because it has nonzero refcount",
    3460             :                  RelationGetRelationName(relation));
    3461             :         }
    3462             :     }
    3463             : 
    3464             :     /*
    3465             :      * Likewise, update or drop any new-relfilenumber-in-subtransaction record
    3466             :      * or drop record.
    3467             :      */
    3468        8272 :     if (relation->rd_newRelfilelocatorSubid == mySubid)
    3469             :     {
    3470         140 :         if (isCommit)
    3471          72 :             relation->rd_newRelfilelocatorSubid = parentSubid;
    3472             :         else
    3473          68 :             relation->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    3474             :     }
    3475             : 
    3476        8272 :     if (relation->rd_firstRelfilelocatorSubid == mySubid)
    3477             :     {
    3478          96 :         if (isCommit)
    3479          32 :             relation->rd_firstRelfilelocatorSubid = parentSubid;
    3480             :         else
    3481          64 :             relation->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    3482             :     }
    3483             : 
    3484        8272 :     if (relation->rd_droppedSubid == mySubid)
    3485             :     {
    3486          20 :         if (isCommit)
    3487           2 :             relation->rd_droppedSubid = parentSubid;
    3488             :         else
    3489          18 :             relation->rd_droppedSubid = InvalidSubTransactionId;
    3490             :     }
    3491             : }
    3492             : 
    3493             : 
    3494             : /*
    3495             :  *      RelationBuildLocalRelation
    3496             :  *          Build a relcache entry for an about-to-be-created relation,
    3497             :  *          and enter it into the relcache.
    3498             :  */
    3499             : Relation
    3500      105278 : RelationBuildLocalRelation(const char *relname,
    3501             :                            Oid relnamespace,
    3502             :                            TupleDesc tupDesc,
    3503             :                            Oid relid,
    3504             :                            Oid accessmtd,
    3505             :                            RelFileNumber relfilenumber,
    3506             :                            Oid reltablespace,
    3507             :                            bool shared_relation,
    3508             :                            bool mapped_relation,
    3509             :                            char relpersistence,
    3510             :                            char relkind)
    3511             : {
    3512             :     Relation    rel;
    3513             :     MemoryContext oldcxt;
    3514      105278 :     int         natts = tupDesc->natts;
    3515             :     int         i;
    3516             :     bool        has_not_null;
    3517             :     bool        nailit;
    3518             : 
    3519             :     Assert(natts >= 0);
    3520             : 
    3521             :     /*
    3522             :      * check for creation of a rel that must be nailed in cache.
    3523             :      *
    3524             :      * XXX this list had better match the relations specially handled in
    3525             :      * RelationCacheInitializePhase2/3.
    3526             :      */
    3527      105278 :     switch (relid)
    3528             :     {
    3529         448 :         case DatabaseRelationId:
    3530             :         case AuthIdRelationId:
    3531             :         case AuthMemRelationId:
    3532             :         case RelationRelationId:
    3533             :         case AttributeRelationId:
    3534             :         case ProcedureRelationId:
    3535             :         case TypeRelationId:
    3536         448 :             nailit = true;
    3537         448 :             break;
    3538      104830 :         default:
    3539      104830 :             nailit = false;
    3540      104830 :             break;
    3541             :     }
    3542             : 
    3543             :     /*
    3544             :      * check that hardwired list of shared rels matches what's in the
    3545             :      * bootstrap .bki file.  If you get a failure here during initdb, you
    3546             :      * probably need to fix IsSharedRelation() to match whatever you've done
    3547             :      * to the set of shared relations.
    3548             :      */
    3549      105278 :     if (shared_relation != IsSharedRelation(relid))
    3550           0 :         elog(ERROR, "shared_relation flag for \"%s\" does not match IsSharedRelation(%u)",
    3551             :              relname, relid);
    3552             : 
    3553             :     /* Shared relations had better be mapped, too */
    3554             :     Assert(mapped_relation || !shared_relation);
    3555             : 
    3556             :     /*
    3557             :      * switch to the cache context to create the relcache entry.
    3558             :      */
    3559      105278 :     if (!CacheMemoryContext)
    3560           0 :         CreateCacheMemoryContext();
    3561             : 
    3562      105278 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    3563             : 
    3564             :     /*
    3565             :      * allocate a new relation descriptor and fill in basic state fields.
    3566             :      */
    3567      105278 :     rel = (Relation) palloc0(sizeof(RelationData));
    3568             : 
    3569             :     /* make sure relation is marked as having no open file yet */
    3570      105278 :     rel->rd_smgr = NULL;
    3571             : 
    3572             :     /* mark it nailed if appropriate */
    3573      105278 :     rel->rd_isnailed = nailit;
    3574             : 
    3575      105278 :     rel->rd_refcnt = nailit ? 1 : 0;
    3576             : 
    3577             :     /* it's being created in this transaction */
    3578      105278 :     rel->rd_createSubid = GetCurrentSubTransactionId();
    3579      105278 :     rel->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    3580      105278 :     rel->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    3581      105278 :     rel->rd_droppedSubid = InvalidSubTransactionId;
    3582             : 
    3583             :     /*
    3584             :      * create a new tuple descriptor from the one passed in.  We do this
    3585             :      * partly to copy it into the cache context, and partly because the new
    3586             :      * relation can't have any defaults or constraints yet; they have to be
    3587             :      * added in later steps, because they require additions to multiple system
    3588             :      * catalogs.  We can copy attnotnull constraints here, however.
    3589             :      */
    3590      105278 :     rel->rd_att = CreateTupleDescCopy(tupDesc);
    3591      105278 :     rel->rd_att->tdrefcount = 1;  /* mark as refcounted */
    3592      105278 :     has_not_null = false;
    3593      435028 :     for (i = 0; i < natts; i++)
    3594             :     {
    3595      329750 :         Form_pg_attribute satt = TupleDescAttr(tupDesc, i);
    3596      329750 :         Form_pg_attribute datt = TupleDescAttr(rel->rd_att, i);
    3597             : 
    3598      329750 :         datt->attidentity = satt->attidentity;
    3599      329750 :         datt->attgenerated = satt->attgenerated;
    3600      329750 :         datt->attnotnull = satt->attnotnull;
    3601      329750 :         has_not_null |= satt->attnotnull;
    3602             :     }
    3603             : 
    3604      105278 :     if (has_not_null)
    3605             :     {
    3606       15816 :         TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
    3607             : 
    3608       15816 :         constr->has_not_null = true;
    3609       15816 :         rel->rd_att->constr = constr;
    3610             :     }
    3611             : 
    3612             :     /*
    3613             :      * initialize relation tuple form (caller may add/override data later)
    3614             :      */
    3615      105278 :     rel->rd_rel = (Form_pg_class) palloc0(CLASS_TUPLE_SIZE);
    3616             : 
    3617      105278 :     namestrcpy(&rel->rd_rel->relname, relname);
    3618      105278 :     rel->rd_rel->relnamespace = relnamespace;
    3619             : 
    3620      105278 :     rel->rd_rel->relkind = relkind;
    3621      105278 :     rel->rd_rel->relnatts = natts;
    3622      105278 :     rel->rd_rel->reltype = InvalidOid;
    3623             :     /* needed when bootstrapping: */
    3624      105278 :     rel->rd_rel->relowner = BOOTSTRAP_SUPERUSERID;
    3625             : 
    3626             :     /* set up persistence and relcache fields dependent on it */
    3627      105278 :     rel->rd_rel->relpersistence = relpersistence;
    3628      105278 :     switch (relpersistence)
    3629             :     {
    3630       99650 :         case RELPERSISTENCE_UNLOGGED:
    3631             :         case RELPERSISTENCE_PERMANENT:
    3632       99650 :             rel->rd_backend = InvalidBackendId;
    3633       99650 :             rel->rd_islocaltemp = false;
    3634       99650 :             break;
    3635        5628 :         case RELPERSISTENCE_TEMP:
    3636             :             Assert(isTempOrTempToastNamespace(relnamespace));
    3637        5628 :             rel->rd_backend = BackendIdForTempRelations();
    3638        5628 :             rel->rd_islocaltemp = true;
    3639        5628 :             break;
    3640           0 :         default:
    3641           0 :             elog(ERROR, "invalid relpersistence: %c", relpersistence);
    3642             :             break;
    3643             :     }
    3644             : 
    3645             :     /* if it's a materialized view, it's not populated initially */
    3646      105278 :     if (relkind == RELKIND_MATVIEW)
    3647         420 :         rel->rd_rel->relispopulated = false;
    3648             :     else
    3649      104858 :         rel->rd_rel->relispopulated = true;
    3650             : 
    3651             :     /* set replica identity -- system catalogs and non-tables don't have one */
    3652      105278 :     if (!IsCatalogNamespace(relnamespace) &&
    3653       58098 :         (relkind == RELKIND_RELATION ||
    3654       57678 :          relkind == RELKIND_MATVIEW ||
    3655             :          relkind == RELKIND_PARTITIONED_TABLE))
    3656       35160 :         rel->rd_rel->relreplident = REPLICA_IDENTITY_DEFAULT;
    3657             :     else
    3658       70118 :         rel->rd_rel->relreplident = REPLICA_IDENTITY_NOTHING;
    3659             : 
    3660             :     /*
    3661             :      * Insert relation physical and logical identifiers (OIDs) into the right
    3662             :      * places.  For a mapped relation, we set relfilenumber to zero and rely
    3663             :      * on RelationInitPhysicalAddr to consult the map.
    3664             :      */
    3665      105278 :     rel->rd_rel->relisshared = shared_relation;
    3666             : 
    3667      105278 :     RelationGetRelid(rel) = relid;
    3668             : 
    3669      435028 :     for (i = 0; i < natts; i++)
    3670      329750 :         TupleDescAttr(rel->rd_att, i)->attrelid = relid;
    3671             : 
    3672      105278 :     rel->rd_rel->reltablespace = reltablespace;
    3673             : 
    3674      105278 :     if (mapped_relation)
    3675             :     {
    3676        4450 :         rel->rd_rel->relfilenode = InvalidRelFileNumber;
    3677             :         /* Add it to the active mapping information */
    3678        4450 :         RelationMapUpdateMap(relid, relfilenumber, shared_relation, true);
    3679             :     }
    3680             :     else
    3681      100828 :         rel->rd_rel->relfilenode = relfilenumber;
    3682             : 
    3683      105278 :     RelationInitLockInfo(rel);  /* see lmgr.c */
    3684             : 
    3685      105278 :     RelationInitPhysicalAddr(rel);
    3686             : 
    3687      105278 :     rel->rd_rel->relam = accessmtd;
    3688             : 
    3689             :     /*
    3690             :      * RelationInitTableAccessMethod will do syscache lookups, so we mustn't
    3691             :      * run it in CacheMemoryContext.  Fortunately, the remaining steps don't
    3692             :      * require a long-lived current context.
    3693             :      */
    3694      105278 :     MemoryContextSwitchTo(oldcxt);
    3695             : 
    3696      105278 :     if (RELKIND_HAS_TABLE_AM(relkind) || relkind == RELKIND_SEQUENCE)
    3697       50808 :         RelationInitTableAccessMethod(rel);
    3698             : 
    3699             :     /*
    3700             :      * Okay to insert into the relcache hash table.
    3701             :      *
    3702             :      * Ordinarily, there should certainly not be an existing hash entry for
    3703             :      * the same OID; but during bootstrap, when we create a "real" relcache
    3704             :      * entry for one of the bootstrap relations, we'll be overwriting the
    3705             :      * phony one created with formrdesc.  So allow that to happen for nailed
    3706             :      * rels.
    3707             :      */
    3708      105278 :     RelationCacheInsert(rel, nailit);
    3709             : 
    3710             :     /*
    3711             :      * Flag relation as needing eoxact cleanup (to clear rd_createSubid). We
    3712             :      * can't do this before storing relid in it.
    3713             :      */
    3714      105278 :     EOXactListAdd(rel);
    3715             : 
    3716             :     /* It's fully valid */
    3717      105278 :     rel->rd_isvalid = true;
    3718             : 
    3719             :     /*
    3720             :      * Caller expects us to pin the returned entry.
    3721             :      */
    3722      105278 :     RelationIncrementReferenceCount(rel);
    3723             : 
    3724      105278 :     return rel;
    3725             : }
    3726             : 
    3727             : 
    3728             : /*
    3729             :  * RelationSetNewRelfilenumber
    3730             :  *
    3731             :  * Assign a new relfilenumber (physical file name), and possibly a new
    3732             :  * persistence setting, to the relation.
    3733             :  *
    3734             :  * This allows a full rewrite of the relation to be done with transactional
    3735             :  * safety (since the filenumber assignment can be rolled back).  Note however
    3736             :  * that there is no simple way to access the relation's old data for the
    3737             :  * remainder of the current transaction.  This limits the usefulness to cases
    3738             :  * such as TRUNCATE or rebuilding an index from scratch.
    3739             :  *
    3740             :  * Caller must already hold exclusive lock on the relation.
    3741             :  */
    3742             : void
    3743        8982 : RelationSetNewRelfilenumber(Relation relation, char persistence)
    3744             : {
    3745             :     RelFileNumber newrelfilenumber;
    3746             :     Relation    pg_class;
    3747             :     HeapTuple   tuple;
    3748             :     Form_pg_class classform;
    3749        8982 :     MultiXactId minmulti = InvalidMultiXactId;
    3750        8982 :     TransactionId freezeXid = InvalidTransactionId;
    3751             :     RelFileLocator newrlocator;
    3752             : 
    3753        8982 :     if (!IsBinaryUpgrade)
    3754             :     {
    3755             :         /* Allocate a new relfilenumber */
    3756        8950 :         newrelfilenumber = GetNewRelFileNumber(relation->rd_rel->reltablespace,
    3757             :                                                NULL, persistence);
    3758             :     }
    3759          32 :     else if (relation->rd_rel->relkind == RELKIND_INDEX)
    3760             :     {
    3761          16 :         if (!OidIsValid(binary_upgrade_next_index_pg_class_relfilenumber))
    3762           0 :             ereport(ERROR,
    3763             :                     (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
    3764             :                      errmsg("index relfilenumber value not set when in binary upgrade mode")));
    3765             : 
    3766          16 :         newrelfilenumber = binary_upgrade_next_index_pg_class_relfilenumber;
    3767          16 :         binary_upgrade_next_index_pg_class_relfilenumber = InvalidOid;
    3768             :     }
    3769          16 :     else if (relation->rd_rel->relkind == RELKIND_RELATION)
    3770             :     {
    3771          16 :         if (!OidIsValid(binary_upgrade_next_heap_pg_class_relfilenumber))
    3772           0 :             ereport(ERROR,
    3773             :                     (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
    3774             :                      errmsg("heap relfilenumber value not set when in binary upgrade mode")));
    3775             : 
    3776          16 :         newrelfilenumber = binary_upgrade_next_heap_pg_class_relfilenumber;
    3777          16 :         binary_upgrade_next_heap_pg_class_relfilenumber = InvalidOid;
    3778             :     }
    3779             :     else
    3780           0 :         ereport(ERROR,
    3781             :                 (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
    3782             :                  errmsg("unexpected request for new relfilenumber in binary upgrade mode")));
    3783             : 
    3784             :     /*
    3785             :      * Get a writable copy of the pg_class tuple for the given relation.
    3786             :      */
    3787        8982 :     pg_class = table_open(RelationRelationId, RowExclusiveLock);
    3788             : 
    3789        8982 :     tuple = SearchSysCacheCopy1(RELOID,
    3790             :                                 ObjectIdGetDatum(RelationGetRelid(relation)));
    3791        8982 :     if (!HeapTupleIsValid(tuple))
    3792           0 :         elog(ERROR, "could not find tuple for relation %u",
    3793             :              RelationGetRelid(relation));
    3794        8982 :     classform = (Form_pg_class) GETSTRUCT(tuple);
    3795             : 
    3796             :     /*
    3797             :      * Schedule unlinking of the old storage at transaction commit, except
    3798             :      * when performing a binary upgrade, when we must do it immediately.
    3799             :      */
    3800        8982 :     if (IsBinaryUpgrade)
    3801             :     {
    3802             :         SMgrRelation srel;
    3803             : 
    3804             :         /*
    3805             :          * During a binary upgrade, we use this code path to ensure that
    3806             :          * pg_largeobject and its index have the same relfilenumbers as in the
    3807             :          * old cluster. This is necessary because pg_upgrade treats
    3808             :          * pg_largeobject like a user table, not a system table. It is however
    3809             :          * possible that a table or index may need to end up with the same
    3810             :          * relfilenumber in the new cluster as what it had in the old cluster.
    3811             :          * Hence, we can't wait until commit time to remove the old storage.
    3812             :          *
    3813             :          * In general, this function needs to have transactional semantics,
    3814             :          * and removing the old storage before commit time surely isn't.
    3815             :          * However, it doesn't really matter, because if a binary upgrade
    3816             :          * fails at this stage, the new cluster will need to be recreated
    3817             :          * anyway.
    3818             :          */
    3819          32 :         srel = smgropen(relation->rd_locator, relation->rd_backend);
    3820          32 :         smgrdounlinkall(&srel, 1, false);
    3821          32 :         smgrclose(srel);
    3822             :     }
    3823             :     else
    3824             :     {
    3825             :         /* Not a binary upgrade, so just schedule it to happen later. */
    3826        8950 :         RelationDropStorage(relation);
    3827             :     }
    3828             : 
    3829             :     /*
    3830             :      * Create storage for the main fork of the new relfilenumber.  If it's a
    3831             :      * table-like object, call into the table AM to do so, which'll also
    3832             :      * create the table's init fork if needed.
    3833             :      *
    3834             :      * NOTE: If relevant for the AM, any conflict in relfilenumber value will
    3835             :      * be caught here, if GetNewRelFileNumber messes up for any reason.
    3836             :      */
    3837        8982 :     newrlocator = relation->rd_locator;
    3838        8982 :     newrlocator.relNumber = newrelfilenumber;
    3839             : 
    3840        8982 :     if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind))
    3841             :     {
    3842        3784 :         table_relation_set_new_filelocator(relation, &newrlocator,
    3843             :                                            persistence,
    3844             :                                            &freezeXid, &minmulti);
    3845             :     }
    3846        5198 :     else if (RELKIND_HAS_STORAGE(relation->rd_rel->relkind))
    3847        5198 :     {
    3848             :         /* handle these directly, at least for now */
    3849             :         SMgrRelation srel;
    3850             : 
    3851        5198 :         srel = RelationCreateStorage(newrlocator, persistence, true);
    3852        5198 :         smgrclose(srel);
    3853             :     }
    3854             :     else
    3855             :     {
    3856             :         /* we shouldn't be called for anything else */
    3857           0 :         elog(ERROR, "relation \"%s\" does not have storage",
    3858             :              RelationGetRelationName(relation));
    3859             :     }
    3860             : 
    3861             :     /*
    3862             :      * If we're dealing with a mapped index, pg_class.relfilenode doesn't
    3863             :      * change; instead we have to send the update to the relation mapper.
    3864             :      *
    3865             :      * For mapped indexes, we don't actually change the pg_class entry at all;
    3866             :      * this is essential when reindexing pg_class itself.  That leaves us with
    3867             :      * possibly-inaccurate values of relpages etc, but those will be fixed up
    3868             :      * later.
    3869             :      */
    3870        8982 :     if (RelationIsMapped(relation))
    3871             :     {
    3872             :         /* This case is only supported for indexes */
    3873             :         Assert(relation->rd_rel->relkind == RELKIND_INDEX);
    3874             : 
    3875             :         /* Since we're not updating pg_class, these had better not change */
    3876             :         Assert(classform->relfrozenxid == freezeXid);
    3877             :         Assert(classform->relminmxid == minmulti);
    3878             :         Assert(classform->relpersistence == persistence);
    3879             : 
    3880             :         /*
    3881             :          * In some code paths it's possible that the tuple update we'd
    3882             :          * otherwise do here is the only thing that would assign an XID for
    3883             :          * the current transaction.  However, we must have an XID to delete
    3884             :          * files, so make sure one is assigned.
    3885             :          */
    3886         428 :         (void) GetCurrentTransactionId();
    3887             : 
    3888             :         /* Do the deed */
    3889         428 :         RelationMapUpdateMap(RelationGetRelid(relation),
    3890             :                              newrelfilenumber,
    3891         428 :                              relation->rd_rel->relisshared,
    3892             :                              false);
    3893             : 
    3894             :         /* Since we're not updating pg_class, must trigger inval manually */
    3895         428 :         CacheInvalidateRelcache(relation);
    3896             :     }
    3897             :     else
    3898             :     {
    3899             :         /* Normal case, update the pg_class entry */
    3900        8554 :         classform->relfilenode = newrelfilenumber;
    3901             : 
    3902             :         /* relpages etc. never change for sequences */
    3903        8554 :         if (relation->rd_rel->relkind != RELKIND_SEQUENCE)
    3904             :         {
    3905        8326 :             classform->relpages = 0; /* it's empty until further notice */
    3906        8326 :             classform->reltuples = -1;
    3907        8326 :             classform->relallvisible = 0;
    3908             :         }
    3909        8554 :         classform->relfrozenxid = freezeXid;
    3910        8554 :         classform->relminmxid = minmulti;
    3911        8554 :         classform->relpersistence = persistence;
    3912             : 
    3913        8554 :         CatalogTupleUpdate(pg_class, &tuple->t_self, tuple);
    3914             :     }
    3915             : 
    3916        8982 :     heap_freetuple(tuple);
    3917             : 
    3918        8982 :     table_close(pg_class, RowExclusiveLock);
    3919             : 
    3920             :     /*
    3921             :      * Make the pg_class row change or relation map change visible.  This will
    3922             :      * cause the relcache entry to get updated, too.
    3923             :      */
    3924        8982 :     CommandCounterIncrement();
    3925             : 
    3926        8982 :     RelationAssumeNewRelfilelocator(relation);
    3927        8982 : }
    3928             : 
    3929             : /*
    3930             :  * RelationAssumeNewRelfilelocator
    3931             :  *
    3932             :  * Code that modifies pg_class.reltablespace or pg_class.relfilenode must call
    3933             :  * this.  The call shall precede any code that might insert WAL records whose
    3934             :  * replay would modify bytes in the new RelFileLocator, and the call shall follow
    3935             :  * any WAL modifying bytes in the prior RelFileLocator.  See struct RelationData.
    3936             :  * Ideally, call this as near as possible to the CommandCounterIncrement()
    3937             :  * that makes the pg_class change visible (before it or after it); that
    3938             :  * minimizes the chance of future development adding a forbidden WAL insertion
    3939             :  * between RelationAssumeNewRelfilelocator() and CommandCounterIncrement().
    3940             :  */
    3941             : void
    3942       10884 : RelationAssumeNewRelfilelocator(Relation relation)
    3943             : {
    3944       10884 :     relation->rd_newRelfilelocatorSubid = GetCurrentSubTransactionId();
    3945       10884 :     if (relation->rd_firstRelfilelocatorSubid == InvalidSubTransactionId)
    3946       10796 :         relation->rd_firstRelfilelocatorSubid = relation->rd_newRelfilelocatorSubid;
    3947             : 
    3948             :     /* Flag relation as needing eoxact cleanup (to clear these fields) */
    3949       10884 :     EOXactListAdd(relation);
    3950       10884 : }
    3951             : 
    3952             : 
    3953             : /*
    3954             :  *      RelationCacheInitialize
    3955             :  *
    3956             :  *      This initializes the relation descriptor cache.  At the time
    3957             :  *      that this is invoked, we can't do database access yet (mainly
    3958             :  *      because the transaction subsystem is not up); all we are doing
    3959             :  *      is making an empty cache hashtable.  This must be done before
    3960             :  *      starting the initialization transaction, because otherwise
    3961             :  *      AtEOXact_RelationCache would crash if that transaction aborts
    3962             :  *      before we can get the relcache set up.
    3963             :  */
    3964             : 
    3965             : #define INITRELCACHESIZE        400
    3966             : 
    3967             : void
    3968       24270 : RelationCacheInitialize(void)
    3969             : {
    3970             :     HASHCTL     ctl;
    3971             :     int         allocsize;
    3972             : 
    3973             :     /*
    3974             :      * make sure cache memory context exists
    3975             :      */
    3976       24270 :     if (!CacheMemoryContext)
    3977       24270 :         CreateCacheMemoryContext();
    3978             : 
    3979             :     /*
    3980             :      * create hashtable that indexes the relcache
    3981             :      */
    3982       24270 :     ctl.keysize = sizeof(Oid);
    3983       24270 :     ctl.entrysize = sizeof(RelIdCacheEnt);
    3984       24270 :     RelationIdCache = hash_create("Relcache by OID", INITRELCACHESIZE,
    3985             :                                   &ctl, HASH_ELEM | HASH_BLOBS);
    3986             : 
    3987             :     /*
    3988             :      * reserve enough in_progress_list slots for many cases
    3989             :      */
    3990       24270 :     allocsize = 4;
    3991       24270 :     in_progress_list =
    3992       24270 :         MemoryContextAlloc(CacheMemoryContext,
    3993             :                            allocsize * sizeof(*in_progress_list));
    3994       24270 :     in_progress_list_maxlen = allocsize;
    3995             : 
    3996             :     /*
    3997             :      * relation mapper needs to be initialized too
    3998             :      */
    3999       24270 :     RelationMapInitialize();
    4000       24270 : }
    4001             : 
    4002             : /*
    4003             :  *      RelationCacheInitializePhase2
    4004             :  *
    4005             :  *      This is called to prepare for access to shared catalogs during startup.
    4006             :  *      We must at least set up nailed reldescs for pg_database, pg_authid,
    4007             :  *      pg_auth_members, and pg_shseclabel. Ideally we'd like to have reldescs
    4008             :  *      for their indexes, too.  We attempt to load this information from the
    4009             :  *      shared relcache init file.  If that's missing or broken, just make
    4010             :  *      phony entries for the catalogs themselves.
    4011             :  *      RelationCacheInitializePhase3 will clean up as needed.
    4012             :  */
    4013             : void
    4014       24270 : RelationCacheInitializePhase2(void)
    4015             : {
    4016             :     MemoryContext oldcxt;
    4017             : 
    4018             :     /*
    4019             :      * relation mapper needs initialized too
    4020             :      */
    4021       24270 :     RelationMapInitializePhase2();
    4022             : 
    4023             :     /*
    4024             :      * In bootstrap mode, the shared catalogs aren't there yet anyway, so do
    4025             :      * nothing.
    4026             :      */
    4027       24270 :     if (IsBootstrapProcessingMode())
    4028          64 :         return;
    4029             : 
    4030             :     /*
    4031             :      * switch to cache memory context
    4032             :      */
    4033       24206 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    4034             : 
    4035             :     /*
    4036             :      * Try to load the shared relcache cache file.  If unsuccessful, bootstrap
    4037             :      * the cache with pre-made descriptors for the critical shared catalogs.
    4038             :      */
    4039       24206 :     if (!load_relcache_init_file(true))
    4040             :     {
    4041        2908 :         formrdesc("pg_database", DatabaseRelation_Rowtype_Id, true,
    4042             :                   Natts_pg_database, Desc_pg_database);
    4043        2908 :         formrdesc("pg_authid", AuthIdRelation_Rowtype_Id, true,
    4044             :                   Natts_pg_authid, Desc_pg_authid);
    4045        2908 :         formrdesc("pg_auth_members", AuthMemRelation_Rowtype_Id, true,
    4046             :                   Natts_pg_auth_members, Desc_pg_auth_members);
    4047        2908 :         formrdesc("pg_shseclabel", SharedSecLabelRelation_Rowtype_Id, true,
    4048             :                   Natts_pg_shseclabel, Desc_pg_shseclabel);
    4049        2908 :         formrdesc("pg_subscription", SubscriptionRelation_Rowtype_Id, true,
    4050             :                   Natts_pg_subscription, Desc_pg_subscription);
    4051             : 
    4052             : #define NUM_CRITICAL_SHARED_RELS    5   /* fix if you change list above */
    4053             :     }
    4054             : 
    4055       24206 :     MemoryContextSwitchTo(oldcxt);
    4056             : }
    4057             : 
    4058             : /*
    4059             :  *      RelationCacheInitializePhase3
    4060             :  *
    4061             :  *      This is called as soon as the catcache and transaction system
    4062             :  *      are functional and we have determined MyDatabaseId.  At this point
    4063             :  *      we can actually read data from the database's system catalogs.
    4064             :  *      We first try to read pre-computed relcache entries from the local
    4065             :  *      relcache init file.  If that's missing or broken, make phony entries
    4066             :  *      for the minimum set of nailed-in-cache relations.  Then (unless
    4067             :  *      bootstrapping) make sure we have entries for the critical system
    4068             :  *      indexes.  Once we've done all this, we have enough infrastructure to
    4069             :  *      open any system catalog or use any catcache.  The last step is to
    4070             :  *      rewrite the cache files if needed.
    4071             :  */
    4072             : void
    4073       22164 : RelationCacheInitializePhase3(void)
    4074             : {
    4075             :     HASH_SEQ_STATUS status;
    4076             :     RelIdCacheEnt *idhentry;
    4077             :     MemoryContext oldcxt;
    4078       22164 :     bool        needNewCacheFile = !criticalSharedRelcachesBuilt;
    4079             : 
    4080             :     /*
    4081             :      * relation mapper needs initialized too
    4082             :      */
    4083       22164 :     RelationMapInitializePhase3();
    4084             : 
    4085             :     /*
    4086             :      * switch to cache memory context
    4087             :      */
    4088       22164 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    4089             : 
    4090             :     /*
    4091             :      * Try to load the local relcache cache file.  If unsuccessful, bootstrap
    4092             :      * the cache with pre-made descriptors for the critical "nailed-in" system
    4093             :      * catalogs.
    4094             :      */
    4095       22164 :     if (IsBootstrapProcessingMode() ||
    4096       22100 :         !load_relcache_init_file(false))
    4097             :     {
    4098        1944 :         needNewCacheFile = true;
    4099             : 
    4100        1944 :         formrdesc("pg_class", RelationRelation_Rowtype_Id, false,
    4101             :                   Natts_pg_class, Desc_pg_class);
    4102        1944 :         formrdesc("pg_attribute", AttributeRelation_Rowtype_Id, false,
    4103             :                   Natts_pg_attribute, Desc_pg_attribute);
    4104        1944 :         formrdesc("pg_proc", ProcedureRelation_Rowtype_Id, false,
    4105             :                   Natts_pg_proc, Desc_pg_proc);
    4106        1944 :         formrdesc("pg_type", TypeRelation_Rowtype_Id, false,
    4107             :                   Natts_pg_type, Desc_pg_type);
    4108             : 
    4109             : #define NUM_CRITICAL_LOCAL_RELS 4   /* fix if you change list above */
    4110             :     }
    4111             : 
    4112       22164 :     MemoryContextSwitchTo(oldcxt);
    4113             : 
    4114             :     /* In bootstrap mode, the faked-up formrdesc info is all we'll have */
    4115       22164 :     if (IsBootstrapProcessingMode())
    4116          64 :         return;
    4117             : 
    4118             :     /*
    4119             :      * If we didn't get the critical system indexes loaded into relcache, do
    4120             :      * so now.  These are critical because the catcache and/or opclass cache
    4121             :      * depend on them for fetches done during relcache load.  Thus, we have an
    4122             :      * infinite-recursion problem.  We can break the recursion by doing
    4123             :      * heapscans instead of indexscans at certain key spots. To avoid hobbling
    4124             :      * performance, we only want to do that until we have the critical indexes
    4125             :      * loaded into relcache.  Thus, the flag criticalRelcachesBuilt is used to
    4126             :      * decide whether to do heapscan or indexscan at the key spots, and we set
    4127             :      * it true after we've loaded the critical indexes.
    4128             :      *
    4129             :      * The critical indexes are marked as "nailed in cache", partly to make it
    4130             :      * easy for load_relcache_init_file to count them, but mainly because we
    4131             :      * cannot flush and rebuild them once we've set criticalRelcachesBuilt to
    4132             :      * true.  (NOTE: perhaps it would be possible to reload them by
    4133             :      * temporarily setting criticalRelcachesBuilt to false again.  For now,
    4134             :      * though, we just nail 'em in.)
    4135             :      *
    4136             :      * RewriteRelRulenameIndexId and TriggerRelidNameIndexId are not critical
    4137             :      * in the same way as the others, because the critical catalogs don't
    4138             :      * (currently) have any rules or triggers, and so these indexes can be
    4139             :      * rebuilt without inducing recursion.  However they are used during
    4140             :      * relcache load when a rel does have rules or triggers, so we choose to
    4141             :      * nail them for performance reasons.
    4142             :      */
    4143       22100 :     if (!criticalRelcachesBuilt)
    4144             :     {
    4145        1880 :         load_critical_index(ClassOidIndexId,
    4146             :                             RelationRelationId);
    4147        1880 :         load_critical_index(AttributeRelidNumIndexId,
    4148             :                             AttributeRelationId);
    4149        1880 :         load_critical_index(IndexRelidIndexId,
    4150             :                             IndexRelationId);
    4151        1880 :         load_critical_index(OpclassOidIndexId,
    4152             :                             OperatorClassRelationId);
    4153        1880 :         load_critical_index(AccessMethodProcedureIndexId,
    4154             :                             AccessMethodProcedureRelationId);
    4155        1880 :         load_critical_index(RewriteRelRulenameIndexId,
    4156             :                             RewriteRelationId);
    4157        1880 :         load_critical_index(TriggerRelidNameIndexId,
    4158             :                             TriggerRelationId);
    4159             : 
    4160             : #define NUM_CRITICAL_LOCAL_INDEXES  7   /* fix if you change list above */
    4161             : 
    4162        1880 :         criticalRelcachesBuilt = true;
    4163             :     }
    4164             : 
    4165             :     /*
    4166             :      * Process critical shared indexes too.
    4167             :      *
    4168             :      * DatabaseNameIndexId isn't critical for relcache loading, but rather for
    4169             :      * initial lookup of MyDatabaseId, without which we'll never find any
    4170             :      * non-shared catalogs at all.  Autovacuum calls InitPostgres with a
    4171             :      * database OID, so it instead depends on DatabaseOidIndexId.  We also
    4172             :      * need to nail up some indexes on pg_authid and pg_auth_members for use
    4173             :      * during client authentication.  SharedSecLabelObjectIndexId isn't
    4174             :      * critical for the core system, but authentication hooks might be
    4175             :      * interested in it.
    4176             :      */
    4177       22100 :     if (!criticalSharedRelcachesBuilt)
    4178             :     {
    4179        1408 :         load_critical_index(DatabaseNameIndexId,
    4180             :                             DatabaseRelationId);
    4181        1408 :         load_critical_index(DatabaseOidIndexId,
    4182             :                             DatabaseRelationId);
    4183        1408 :         load_critical_index(AuthIdRolnameIndexId,
    4184             :                             AuthIdRelationId);
    4185        1408 :         load_critical_index(AuthIdOidIndexId,
    4186             :                             AuthIdRelationId);
    4187        1408 :         load_critical_index(AuthMemMemRoleIndexId,
    4188             :                             AuthMemRelationId);
    4189        1408 :         load_critical_index(SharedSecLabelObjectIndexId,
    4190             :                             SharedSecLabelRelationId);
    4191             : 
    4192             : #define NUM_CRITICAL_SHARED_INDEXES 6   /* fix if you change list above */
    4193             : 
    4194        1408 :         criticalSharedRelcachesBuilt = true;
    4195             :     }
    4196             : 
    4197             :     /*
    4198             :      * Now, scan all the relcache entries and update anything that might be
    4199             :      * wrong in the results from formrdesc or the relcache cache file. If we
    4200             :      * faked up relcache entries using formrdesc, then read the real pg_class
    4201             :      * rows and replace the fake entries with them. Also, if any of the
    4202             :      * relcache entries have rules, triggers, or security policies, load that
    4203             :      * info the hard way since it isn't recorded in the cache file.
    4204             :      *
    4205             :      * Whenever we access the catalogs to read data, there is a possibility of
    4206             :      * a shared-inval cache flush causing relcache entries to be removed.
    4207             :      * Since hash_seq_search only guarantees to still work after the *current*
    4208             :      * entry is removed, it's unsafe to continue the hashtable scan afterward.
    4209             :      * We handle this by restarting the scan from scratch after each access.
    4210             :      * This is theoretically O(N^2), but the number of entries that actually
    4211             :      * need to be fixed is small enough that it doesn't matter.
    4212             :      */
    4213       22100 :     hash_seq_init(&status, RelationIdCache);
    4214             : 
    4215     3038898 :     while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
    4216             :     {
    4217     3016798 :         Relation    relation = idhentry->reldesc;
    4218     3016798 :         bool        restart = false;
    4219             : 
    4220             :         /*
    4221             :          * Make sure *this* entry doesn't get flushed while we work with it.
    4222             :          */
    4223     3016798 :         RelationIncrementReferenceCount(relation);
    4224             : 
    4225             :         /*
    4226             :          * If it's a faked-up entry, read the real pg_class tuple.
    4227             :          */
    4228     3016798 :         if (relation->rd_rel->relowner == InvalidOid)
    4229             :         {
    4230             :             HeapTuple   htup;
    4231             :             Form_pg_class relp;
    4232             : 
    4233       14558 :             htup = SearchSysCache1(RELOID,
    4234             :                                    ObjectIdGetDatum(RelationGetRelid(relation)));
    4235       14558 :             if (!HeapTupleIsValid(htup))
    4236           0 :                 elog(FATAL, "cache lookup failed for relation %u",
    4237             :                      RelationGetRelid(relation));
    4238       14558 :             relp = (Form_pg_class) GETSTRUCT(htup);
    4239             : 
    4240             :             /*
    4241             :              * Copy tuple to relation->rd_rel. (See notes in
    4242             :              * AllocateRelationDesc())
    4243             :              */
    4244       14558 :             memcpy((char *) relation->rd_rel, (char *) relp, CLASS_TUPLE_SIZE);
    4245             : 
    4246             :             /* Update rd_options while we have the tuple */
    4247       14558 :             if (relation->rd_options)
    4248           0 :                 pfree(relation->rd_options);
    4249       14558 :             RelationParseRelOptions(relation, htup);
    4250             : 
    4251             :             /*
    4252             :              * Check the values in rd_att were set up correctly.  (We cannot
    4253             :              * just copy them over now: formrdesc must have set up the rd_att
    4254             :              * data correctly to start with, because it may already have been
    4255             :              * copied into one or more catcache entries.)
    4256             :              */
    4257             :             Assert(relation->rd_att->tdtypeid == relp->reltype);
    4258             :             Assert(relation->rd_att->tdtypmod == -1);
    4259             : 
    4260       14558 :             ReleaseSysCache(htup);
    4261             : 
    4262             :             /* relowner had better be OK now, else we'll loop forever */
    4263       14558 :             if (relation->rd_rel->relowner == InvalidOid)
    4264           0 :                 elog(ERROR, "invalid relowner in pg_class entry for \"%s\"",
    4265             :                      RelationGetRelationName(relation));
    4266             : 
    4267       14558 :             restart = true;
    4268             :         }
    4269             : 
    4270             :         /*
    4271             :          * Fix data that isn't saved in relcache cache file.
    4272             :          *
    4273             :          * relhasrules or relhastriggers could possibly be wrong or out of
    4274             :          * date.  If we don't actually find any rules or triggers, clear the
    4275             :          * local copy of the flag so that we don't get into an infinite loop
    4276             :          * here.  We don't make any attempt to fix the pg_class entry, though.
    4277             :          */
    4278     3016798 :         if (relation->rd_rel->relhasrules && relation->rd_rules == NULL)
    4279             :         {
    4280           0 :             RelationBuildRuleLock(relation);
    4281           0 :             if (relation->rd_rules == NULL)
    4282           0 :                 relation->rd_rel->relhasrules = false;
    4283           0 :             restart = true;
    4284             :         }
    4285     3016798 :         if (relation->rd_rel->relhastriggers && relation->trigdesc == NULL)
    4286             :         {
    4287           0 :             RelationBuildTriggers(relation);
    4288           0 :             if (relation->trigdesc == NULL)
    4289           0 :                 relation->rd_rel->relhastriggers = false;
    4290           0 :             restart = true;
    4291             :         }
    4292             : 
    4293             :         /*
    4294             :          * Re-load the row security policies if the relation has them, since
    4295             :          * they are not preserved in the cache.  Note that we can never NOT
    4296             :          * have a policy while relrowsecurity is true,
    4297             :          * RelationBuildRowSecurity will create a single default-deny policy
    4298             :          * if there is no policy defined in pg_policy.
    4299             :          */
    4300     3016798 :         if (relation->rd_rel->relrowsecurity && relation->rd_rsdesc == NULL)
    4301             :         {
    4302           0 :             RelationBuildRowSecurity(relation);
    4303             : 
    4304             :             Assert(relation->rd_rsdesc != NULL);
    4305           0 :             restart = true;
    4306             :         }
    4307             : 
    4308             :         /* Reload tableam data if needed */
    4309     3016798 :         if (relation->rd_tableam == NULL &&
    4310     1851652 :             (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) || relation->rd_rel->relkind == RELKIND_SEQUENCE))
    4311             :         {
    4312           0 :             RelationInitTableAccessMethod(relation);
    4313             :             Assert(relation->rd_tableam != NULL);
    4314             : 
    4315           0 :             restart = true;
    4316             :         }
    4317             : 
    4318             :         /* Release hold on the relation */
    4319     3016798 :         RelationDecrementReferenceCount(relation);
    4320             : 
    4321             :         /* Now, restart the hashtable scan if needed */
    4322     3016798 :         if (restart)
    4323             :         {
    4324       14558 :             hash_seq_term(&status);
    4325       14558 :             hash_seq_init(&status, RelationIdCache);
    4326             :         }
    4327             :     }
    4328             : 
    4329             :     /*
    4330             :      * Lastly, write out new relcache cache files if needed.  We don't bother
    4331             :      * to distinguish cases where only one of the two needs an update.
    4332             :      */
    4333       22100 :     if (needNewCacheFile)
    4334             :     {
    4335             :         /*
    4336             :          * Force all the catcaches to finish initializing and thereby open the
    4337             :          * catalogs and indexes they use.  This will preload the relcache with
    4338             :          * entries for all the most important system catalogs and indexes, so
    4339             :          * that the init files will be most useful for future backends.
    4340             :          */
    4341        1918 :         InitCatalogCachePhase2();
    4342             : 
    4343             :         /* now write the files */
    4344        1912 :         write_relcache_init_file(true);
    4345        1912 :         write_relcache_init_file(false);
    4346             :     }
    4347             : }
    4348             : 
    4349             : /*
    4350             :  * Load one critical system index into the relcache
    4351             :  *
    4352             :  * indexoid is the OID of the target index, heapoid is the OID of the catalog
    4353             :  * it belongs to.
    4354             :  */
    4355             : static void
    4356       21608 : load_critical_index(Oid indexoid, Oid heapoid)
    4357             : {
    4358             :     Relation    ird;
    4359             : 
    4360             :     /*
    4361             :      * We must lock the underlying catalog before locking the index to avoid
    4362             :      * deadlock, since RelationBuildDesc might well need to read the catalog,
    4363             :      * and if anyone else is exclusive-locking this catalog and index they'll
    4364             :      * be doing it in that order.
    4365             :      */
    4366       21608 :     LockRelationOid(heapoid, AccessShareLock);
    4367       21608 :     LockRelationOid(indexoid, AccessShareLock);
    4368       21608 :     ird = RelationBuildDesc(indexoid, true);
    4369       21608 :     if (ird == NULL)
    4370           0 :         elog(PANIC, "could not open critical system index %u", indexoid);
    4371       21608 :     ird->rd_isnailed = true;
    4372       21608 :     ird->rd_refcnt = 1;
    4373       21608 :     UnlockRelationOid(indexoid, AccessShareLock);
    4374       21608 :     UnlockRelationOid(heapoid, AccessShareLock);
    4375             : 
    4376       21608 :     (void) RelationGetIndexAttOptions(ird, false);
    4377       21608 : }
    4378             : 
    4379             : /*
    4380             :  * GetPgClassDescriptor -- get a predefined tuple descriptor for pg_class
    4381             :  * GetPgIndexDescriptor -- get a predefined tuple descriptor for pg_index
    4382             :  *
    4383             :  * We need this kluge because we have to be able to access non-fixed-width
    4384             :  * fields of pg_class and pg_index before we have the standard catalog caches
    4385             :  * available.  We use predefined data that's set up in just the same way as
    4386             :  * the bootstrapped reldescs used by formrdesc().  The resulting tupdesc is
    4387             :  * not 100% kosher: it does not have the correct rowtype OID in tdtypeid, nor
    4388             :  * does it have a TupleConstr field.  But it's good enough for the purpose of
    4389             :  * extracting fields.
    4390             :  */
    4391             : static TupleDesc
    4392       44326 : BuildHardcodedDescriptor(int natts, const FormData_pg_attribute *attrs)
    4393             : {
    4394             :     TupleDesc   result;
    4395             :     MemoryContext oldcxt;
    4396             :     int         i;
    4397             : 
    4398       44326 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    4399             : 
    4400       44326 :     result = CreateTemplateTupleDesc(natts);
    4401       44326 :     result->tdtypeid = RECORDOID;    /* not right, but we don't care */
    4402       44326 :     result->tdtypmod = -1;
    4403             : 
    4404     1241140 :     for (i = 0; i < natts; i++)
    4405             :     {
    4406     1196814 :         memcpy(TupleDescAttr(result, i), &attrs[i], ATTRIBUTE_FIXED_PART_SIZE);
    4407             :         /* make sure attcacheoff is valid */
    4408     1196814 :         TupleDescAttr(result, i)->attcacheoff = -1;
    4409             :     }
    4410             : 
    4411             :     /* initialize first attribute's attcacheoff, cf RelationBuildTupleDesc */
    4412       44326 :     TupleDescAttr(result, 0)->attcacheoff = 0;
    4413             : 
    4414             :     /* Note: we don't bother to set up a TupleConstr entry */
    4415             : 
    4416       44326 :     MemoryContextSwitchTo(oldcxt);
    4417             : 
    4418       44326 :     return result;
    4419             : }
    4420             : 
    4421             : static TupleDesc
    4422     1015072 : GetPgClassDescriptor(void)
    4423             : {
    4424             :     static TupleDesc pgclassdesc = NULL;
    4425             : 
    4426             :     /* Already done? */
    4427     1015072 :     if (pgclassdesc == NULL)
    4428       22164 :         pgclassdesc = BuildHardcodedDescriptor(Natts_pg_class,
    4429             :                                                Desc_pg_class);
    4430             : 
    4431     1015072 :     return pgclassdesc;
    4432             : }
    4433             : 
    4434             : static TupleDesc
    4435     1190100 : GetPgIndexDescriptor(void)
    4436             : {
    4437             :     static TupleDesc pgindexdesc = NULL;
    4438             : 
    4439             :     /* Already done? */
    4440     1190100 :     if (pgindexdesc == NULL)
    4441       22162 :         pgindexdesc = BuildHardcodedDescriptor(Natts_pg_index,
    4442             :                                                Desc_pg_index);
    4443             : 
    4444     1190100 :     return pgindexdesc;
    4445             : }
    4446             : 
    4447             : /*
    4448             :  * Load any default attribute value definitions for the relation.
    4449             :  *
    4450             :  * ndef is the number of attributes that were marked atthasdef.
    4451             :  *
    4452             :  * Note: we don't make it a hard error to be missing some pg_attrdef records.
    4453             :  * We can limp along as long as nothing needs to use the default value.  Code
    4454             :  * that fails to find an expected AttrDefault record should throw an error.
    4455             :  */
    4456             : static void
    4457       25032 : AttrDefaultFetch(Relation relation, int ndef)
    4458             : {
    4459             :     AttrDefault *attrdef;
    4460             :     Relation    adrel;
    4461             :     SysScanDesc adscan;
    4462             :     ScanKeyData skey;
    4463             :     HeapTuple   htup;
    4464       25032 :     int         found = 0;
    4465             : 
    4466             :     /* Allocate array with room for as many entries as expected */
    4467             :     attrdef = (AttrDefault *)
    4468       25032 :         MemoryContextAllocZero(CacheMemoryContext,
    4469             :                                ndef * sizeof(AttrDefault));
    4470             : 
    4471             :     /* Search pg_attrdef for relevant entries */
    4472       25032 :     ScanKeyInit(&skey,
    4473             :                 Anum_pg_attrdef_adrelid,
    4474             :                 BTEqualStrategyNumber, F_OIDEQ,
    4475             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
    4476             : 
    4477       25032 :     adrel = table_open(AttrDefaultRelationId, AccessShareLock);
    4478       25032 :     adscan = systable_beginscan(adrel, AttrDefaultIndexId, true,
    4479             :                                 NULL, 1, &skey);
    4480             : 
    4481       60086 :     while (HeapTupleIsValid(htup = systable_getnext(adscan)))
    4482             :     {
    4483       35054 :         Form_pg_attrdef adform = (Form_pg_attrdef) GETSTRUCT(htup);
    4484             :         Datum       val;
    4485             :         bool        isnull;
    4486             : 
    4487             :         /* protect limited size of array */
    4488       35054 :         if (found >= ndef)
    4489             :         {
    4490           0 :             elog(WARNING, "unexpected pg_attrdef record found for attribute %d of relation \"%s\"",
    4491             :                  adform->adnum, RelationGetRelationName(relation));
    4492           0 :             break;
    4493             :         }
    4494             : 
    4495       35054 :         val = fastgetattr(htup,
    4496             :                           Anum_pg_attrdef_adbin,
    4497             :                           adrel->rd_att, &isnull);
    4498       35054 :         if (isnull)
    4499           0 :             elog(WARNING, "null adbin for attribute %d of relation \"%s\"",
    4500             :                  adform->adnum, RelationGetRelationName(relation));
    4501             :         else
    4502             :         {
    4503             :             /* detoast and convert to cstring in caller's context */
    4504       35054 :             char       *s = TextDatumGetCString(val);
    4505             : 
    4506       35054 :             attrdef[found].adnum = adform->adnum;
    4507       35054 :             attrdef[found].adbin = MemoryContextStrdup(CacheMemoryContext, s);
    4508       35054 :             pfree(s);
    4509       35054 :             found++;
    4510             :         }
    4511             :     }
    4512             : 
    4513       25032 :     systable_endscan(adscan);
    4514       25032 :     table_close(adrel, AccessShareLock);
    4515             : 
    4516       25032 :     if (found != ndef)
    4517           0 :         elog(WARNING, "%d pg_attrdef record(s) missing for relation \"%s\"",
    4518             :              ndef - found, RelationGetRelationName(relation));
    4519             : 
    4520             :     /*
    4521             :      * Sort the AttrDefault entries by adnum, for the convenience of
    4522             :      * equalTupleDescs().  (Usually, they already will be in order, but this
    4523             :      * might not be so if systable_getnext isn't using an index.)
    4524             :      */
    4525       25032 :     if (found > 1)
    4526        5014 :         qsort(attrdef, found, sizeof(AttrDefault), AttrDefaultCmp);
    4527             : 
    4528             :     /* Install array only after it's fully valid */
    4529       25032 :     relation->rd_att->constr->defval = attrdef;
    4530       25032 :     relation->rd_att->constr->num_defval = found;
    4531       25032 : }
    4532             : 
    4533             : /*
    4534             :  * qsort comparator to sort AttrDefault entries by adnum
    4535             :  */
    4536             : static int
    4537       10022 : AttrDefaultCmp(const void *a, const void *b)
    4538             : {
    4539       10022 :     const AttrDefault *ada = (const AttrDefault *) a;
    4540       10022 :     const AttrDefault *adb = (const AttrDefault *) b;
    4541             : 
    4542       10022 :     return ada->adnum - adb->adnum;
    4543             : }
    4544             : 
    4545             : /*
    4546             :  * Load any check constraints for the relation.
    4547             :  *
    4548             :  * As with defaults, if we don't find the expected number of them, just warn
    4549             :  * here.  The executor should throw an error if an INSERT/UPDATE is attempted.
    4550             :  */
    4551             : static void
    4552       10088 : CheckConstraintFetch(Relation relation)
    4553             : {
    4554             :     ConstrCheck *check;
    4555       10088 :     int         ncheck = relation->rd_rel->relchecks;
    4556             :     Relation    conrel;
    4557             :     SysScanDesc conscan;
    4558             :     ScanKeyData skey[1];
    4559             :     HeapTuple   htup;
    4560       10088 :     int         found = 0;
    4561             : 
    4562             :     /* Allocate array with room for as many entries as expected */
    4563             :     check = (ConstrCheck *)
    4564       10088 :         MemoryContextAllocZero(CacheMemoryContext,
    4565             :                                ncheck * sizeof(ConstrCheck));
    4566             : 
    4567             :     /* Search pg_constraint for relevant entries */
    4568       10088 :     ScanKeyInit(&skey[0],
    4569             :                 Anum_pg_constraint_conrelid,
    4570             :                 BTEqualStrategyNumber, F_OIDEQ,
    4571             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
    4572             : 
    4573       10088 :     conrel = table_open(ConstraintRelationId, AccessShareLock);
    4574       10088 :     conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
    4575             :                                  NULL, 1, skey);
    4576             : 
    4577       31642 :     while (HeapTupleIsValid(htup = systable_getnext(conscan)))
    4578             :     {
    4579       21554 :         Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(htup);
    4580             :         Datum       val;
    4581             :         bool        isnull;
    4582             : 
    4583             :         /* We want check constraints only */
    4584       21554 :         if (conform->contype != CONSTRAINT_CHECK)
    4585        8622 :             continue;
    4586             : 
    4587             :         /* protect limited size of array */
    4588       12932 :         if (found >= ncheck)
    4589             :         {
    4590           0 :             elog(WARNING, "unexpected pg_constraint record found for relation \"%s\"",
    4591             :                  RelationGetRelationName(relation));
    4592           0 :             break;
    4593             :         }
    4594             : 
    4595       12932 :         check[found].ccvalid = conform->convalidated;
    4596       12932 :         check[found].ccnoinherit = conform->connoinherit;
    4597       25864 :         check[found].ccname = MemoryContextStrdup(CacheMemoryContext,
    4598       12932 :                                                   NameStr(conform->conname));
    4599             : 
    4600             :         /* Grab and test conbin is actually set */
    4601       12932 :         val = fastgetattr(htup,
    4602             :                           Anum_pg_constraint_conbin,
    4603             :                           conrel->rd_att, &isnull);
    4604       12932 :         if (isnull)
    4605           0 :             elog(WARNING, "null conbin for relation \"%s\"",
    4606             :                  RelationGetRelationName(relation));
    4607             :         else
    4608             :         {
    4609             :             /* detoast and convert to cstring in caller's context */
    4610       12932 :             char       *s = TextDatumGetCString(val);
    4611             : 
    4612       12932 :             check[found].ccbin = MemoryContextStrdup(CacheMemoryContext, s);
    4613       12932 :             pfree(s);
    4614       12932 :             found++;
    4615             :         }
    4616             :     }
    4617             : 
    4618       10088 :     systable_endscan(conscan);
    4619       10088 :     table_close(conrel, AccessShareLock);
    4620             : 
    4621       10088 :     if (found != ncheck)
    4622           0 :         elog(WARNING, "%d pg_constraint record(s) missing for relation \"%s\"",
    4623             :              ncheck - found, RelationGetRelationName(relation));
    4624             : 
    4625             :     /*
    4626             :      * Sort the records by name.  This ensures that CHECKs are applied in a
    4627             :      * deterministic order, and it also makes equalTupleDescs() faster.
    4628             :      */
    4629       10088 :     if (found > 1)
    4630        2376 :         qsort(check, found, sizeof(ConstrCheck), CheckConstraintCmp);
    4631             : 
    4632             :     /* Install array only after it's fully valid */
    4633       10088 :     relation->rd_att->constr->check = check;
    4634       10088 :     relation->rd_att->constr->num_check = found;
    4635       10088 : }
    4636             : 
    4637             : /*
    4638             :  * qsort comparator to sort ConstrCheck entries by name
    4639             :  */
    4640             : static int
    4641        2844 : CheckConstraintCmp(const void *a, const void *b)
    4642             : {
    4643        2844 :     const ConstrCheck *ca = (const ConstrCheck *) a;
    4644        2844 :     const ConstrCheck *cb = (const ConstrCheck *) b;
    4645             : 
    4646        2844 :     return strcmp(ca->ccname, cb->ccname);
    4647             : }
    4648             : 
    4649             : /*
    4650             :  * RelationGetFKeyList -- get a list of foreign key info for the relation
    4651             :  *
    4652             :  * Returns a list of ForeignKeyCacheInfo structs, one per FK constraining
    4653             :  * the given relation.  This data is a direct copy of relevant fields from
    4654             :  * pg_constraint.  The list items are in no particular order.
    4655             :  *
    4656             :  * CAUTION: the returned list is part of the relcache's data, and could
    4657             :  * vanish in a relcache entry reset.  Callers must inspect or copy it
    4658             :  * before doing anything that might trigger a cache flush, such as
    4659             :  * system catalog accesses.  copyObject() can be used if desired.
    4660             :  * (We define it this way because current callers want to filter and
    4661             :  * modify the list entries anyway, so copying would be a waste of time.)
    4662             :  */
    4663             : List *
    4664      182766 : RelationGetFKeyList(Relation relation)
    4665             : {
    4666             :     List       *result;
    4667             :     Relation    conrel;
    4668             :     SysScanDesc conscan;
    4669             :     ScanKeyData skey;
    4670             :     HeapTuple   htup;
    4671             :     List       *oldlist;
    4672             :     MemoryContext oldcxt;
    4673             : 
    4674             :     /* Quick exit if we already computed the list. */
    4675      182766 :     if (relation->rd_fkeyvalid)
    4676        1144 :         return relation->rd_fkeylist;
    4677             : 
    4678             :     /* Fast path: non-partitioned tables without triggers can't have FKs */
    4679      181622 :     if (!relation->rd_rel->relhastriggers &&
    4680      178766 :         relation->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
    4681      170536 :         return NIL;
    4682             : 
    4683             :     /*
    4684             :      * We build the list we intend to return (in the caller's context) while
    4685             :      * doing the scan.  After successfully completing the scan, we copy that
    4686             :      * list into the relcache entry.  This avoids cache-context memory leakage
    4687             :      * if we get some sort of error partway through.
    4688             :      */
    4689       11086 :     result = NIL;
    4690             : 
    4691             :     /* Prepare to scan pg_constraint for entries having conrelid = this rel. */
    4692       11086 :     ScanKeyInit(&skey,
    4693             :                 Anum_pg_constraint_conrelid,
    4694             :                 BTEqualStrategyNumber, F_OIDEQ,
    4695             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
    4696             : 
    4697       11086 :     conrel = table_open(ConstraintRelationId, AccessShareLock);
    4698       11086 :     conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
    4699             :                                  NULL, 1, &skey);
    4700             : 
    4701       17232 :     while (HeapTupleIsValid(htup = systable_getnext(conscan)))
    4702             :     {
    4703        6146 :         Form_pg_constraint constraint = (Form_pg_constraint) GETSTRUCT(htup);
    4704             :         ForeignKeyCacheInfo *info;
    4705             : 
    4706             :         /* consider only foreign keys */
    4707        6146 :         if (constraint->contype != CONSTRAINT_FOREIGN)
    4708        3516 :             continue;
    4709             : 
    4710        2630 :         info = makeNode(ForeignKeyCacheInfo);
    4711        2630 :         info->conoid = constraint->oid;
    4712        2630 :         info->conrelid = constraint->conrelid;
    4713        2630 :         info->confrelid = constraint->confrelid;
    4714             : 
    4715        2630 :         DeconstructFkConstraintRow(htup, &info->nkeys,
    4716        2630 :                                    info->conkey,
    4717        2630 :                                    info->confkey,
    4718        2630 :                                    info->conpfeqop,
    4719             :                                    NULL, NULL, NULL, NULL);
    4720             : 
    4721             :         /* Add FK's node to the result list */
    4722        2630 :         result = lappend(result, info);
    4723             :     }
    4724             : 
    4725       11086 :     systable_endscan(conscan);
    4726       11086 :     table_close(conrel, AccessShareLock);
    4727             : 
    4728             :     /* Now save a copy of the completed list in the relcache entry. */
    4729       11086 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    4730       11086 :     oldlist = relation->rd_fkeylist;
    4731       11086 :     relation->rd_fkeylist = copyObject(result);
    4732       11086 :     relation->rd_fkeyvalid = true;
    4733       11086 :     MemoryContextSwitchTo(oldcxt);
    4734             : 
    4735             :     /* Don't leak the old list, if there is one */
    4736       11086 :     list_free_deep(oldlist);
    4737             : 
    4738       11086 :     return result;
    4739             : }
    4740             : 
    4741             : /*
    4742             :  * RelationGetIndexList -- get a list of OIDs of indexes on this relation
    4743             :  *
    4744             :  * The index list is created only if someone requests it.  We scan pg_index
    4745             :  * to find relevant indexes, and add the list to the relcache entry so that
    4746             :  * we won't have to compute it again.  Note that shared cache inval of a
    4747             :  * relcache entry will delete the old list and set rd_indexvalid to false,
    4748             :  * so that we must recompute the index list on next request.  This handles
    4749             :  * creation or deletion of an index.
    4750             :  *
    4751             :  * Indexes that are marked not indislive are omitted from the returned list.
    4752             :  * Such indexes are expected to be dropped momentarily, and should not be
    4753             :  * touched at all by any caller of this function.
    4754             :  *
    4755             :  * The returned list is guaranteed to be sorted in order by OID.  This is
    4756             :  * needed by the executor, since for index types that we obtain exclusive
    4757             :  * locks on when updating the index, all backends must lock the indexes in
    4758             :  * the same order or we will get deadlocks (see ExecOpenIndices()).  Any
    4759             :  * consistent ordering would do, but ordering by OID is easy.
    4760             :  *
    4761             :  * Since shared cache inval causes the relcache's copy of the list to go away,
    4762             :  * we return a copy of the list palloc'd in the caller's context.  The caller
    4763             :  * may list_free() the returned list after scanning it. This is necessary
    4764             :  * since the caller will typically be doing syscache lookups on the relevant
    4765             :  * indexes, and syscache lookup could cause SI messages to be processed!
    4766             :  *
    4767             :  * In exactly the same way, we update rd_pkindex, which is the OID of the
    4768             :  * relation's primary key index if any, else InvalidOid; and rd_replidindex,
    4769             :  * which is the pg_class OID of an index to be used as the relation's
    4770             :  * replication identity index, or InvalidOid if there is no such index.
    4771             :  */
    4772             : List *
    4773     1782166 : RelationGetIndexList(Relation relation)
    4774             : {
    4775             :     Relation    indrel;
    4776             :     SysScanDesc indscan;
    4777             :     ScanKeyData skey;
    4778             :     HeapTuple   htup;
    4779             :     List       *result;
    4780             :     List       *oldlist;
    4781     1782166 :     char        replident = relation->rd_rel->relreplident;
    4782     1782166 :     Oid         pkeyIndex = InvalidOid;
    4783     1782166 :     Oid         candidateIndex = InvalidOid;
    4784             :     MemoryContext oldcxt;
    4785             : 
    4786             :     /* Quick exit if we already computed the list. */
    4787     1782166 :     if (relation->rd_indexvalid)
    4788     1638842 :         return list_copy(relation->rd_indexlist);
    4789             : 
    4790             :     /*
    4791             :      * We build the list we intend to return (in the caller's context) while
    4792             :      * doing the scan.  After successfully completing the scan, we copy that
    4793             :      * list into the relcache entry.  This avoids cache-context memory leakage
    4794             :      * if we get some sort of error partway through.
    4795             :      */
    4796      143324 :     result = NIL;
    4797             : 
    4798             :     /* Prepare to scan pg_index for entries having indrelid = this rel. */
    4799      143324 :     ScanKeyInit(&skey,
    4800             :                 Anum_pg_index_indrelid,
    4801             :                 BTEqualStrategyNumber, F_OIDEQ,
    4802             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
    4803             : 
    4804      143324 :     indrel = table_open(IndexRelationId, AccessShareLock);
    4805      143324 :     indscan = systable_beginscan(indrel, IndexIndrelidIndexId, true,
    4806             :                                  NULL, 1, &skey);
    4807             : 
    4808      348534 :     while (HeapTupleIsValid(htup = systable_getnext(indscan)))
    4809             :     {
    4810      205210 :         Form_pg_index index = (Form_pg_index) GETSTRUCT(htup);
    4811             : 
    4812             :         /*
    4813             :          * Ignore any indexes that are currently being dropped.  This will
    4814             :          * prevent them from being searched, inserted into, or considered in
    4815             :          * HOT-safety decisions.  It's unsafe to touch such an index at all
    4816             :          * since its catalog entries could disappear at any instant.
    4817             :          */
    4818      205210 :         if (!index->indislive)
    4819          58 :             continue;
    4820             : 
    4821             :         /* add index's OID to result list */
    4822      205152 :         result = lappend_oid(result, index->indexrelid);
    4823             : 
    4824             :         /*
    4825             :          * Non-unique, non-immediate or predicate indexes aren't interesting
    4826             :          * for either oid indexes or replication identity indexes, so don't
    4827             :          * check them.
    4828             :          */
    4829      205152 :         if (!index->indisunique ||
    4830      162586 :             !index->indimmediate ||
    4831      162488 :             !heap_attisnull(htup, Anum_pg_index_indpred, NULL))
    4832       42822 :             continue;
    4833             : 
    4834             :         /*
    4835             :          * Remember primary key index, if any.  We do this only if the index
    4836             :          * is valid; but if the table is partitioned, then we do it even if
    4837             :          * it's invalid.
    4838             :          *
    4839             :          * The reason for returning invalid primary keys for foreign tables is
    4840             :          * because of pg_dump of NOT NULL constraints, and the fact that PKs
    4841             :          * remain marked invalid until the partitions' PKs are attached to it.
    4842             :          * If we make rd_pkindex invalid, then the attnotnull flag is reset
    4843             :          * after the PK is created, which causes the ALTER INDEX ATTACH
    4844             :          * PARTITION to fail with 'column ... is not marked NOT NULL'.  With
    4845             :          * this, dropconstraint_internal() will believe that the columns must
    4846             :          * not have attnotnull reset, so the PKs-on-partitions can be attached
    4847             :          * correctly, until finally the PK-on-parent is marked valid.
    4848             :          *
    4849             :          * Also, this doesn't harm anything, because rd_pkindex is not a
    4850             :          * "real" index anyway, but a RELKIND_PARTITIONED_INDEX.
    4851             :          */
    4852      162330 :         if (index->indisprimary &&
    4853      101504 :             (index->indisvalid ||
    4854          30 :              relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE))
    4855      101504 :             pkeyIndex = index->indexrelid;
    4856             : 
    4857      162330 :         if (!index->indisvalid)
    4858          86 :             continue;
    4859             : 
    4860             :         /* remember explicitly chosen replica index */
    4861      162244 :         if (index->indisreplident)
    4862         458 :             candidateIndex = index->indexrelid;
    4863             :     }
    4864             : 
    4865      143324 :     systable_endscan(indscan);
    4866             : 
    4867      143324 :     table_close(indrel, AccessShareLock);
    4868             : 
    4869             :     /* Sort the result list into OID order, per API spec. */
    4870      143324 :     list_sort(result, list_oid_cmp);
    4871             : 
    4872             :     /* Now save a copy of the completed list in the relcache entry. */
    4873      143324 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    4874      143324 :     oldlist = relation->rd_indexlist;
    4875      143324 :     relation->rd_indexlist = list_copy(result);
    4876      143324 :     relation->rd_pkindex = pkeyIndex;
    4877      143324 :     if (replident == REPLICA_IDENTITY_DEFAULT && OidIsValid(pkeyIndex))
    4878       22798 :         relation->rd_replidindex = pkeyIndex;
    4879      120526 :     else if (replident == REPLICA_IDENTITY_INDEX && OidIsValid(candidateIndex))
    4880         458 :         relation->rd_replidindex = candidateIndex;
    4881             :     else
    4882      120068 :         relation->rd_replidindex = InvalidOid;
    4883      143324 :     relation->rd_indexvalid = true;
    4884      143324 :     MemoryContextSwitchTo(oldcxt);
    4885             : 
    4886             :     /* Don't leak the old list, if there is one */
    4887      143324 :     list_free(oldlist);
    4888             : 
    4889      143324 :     return result;
    4890             : }
    4891             : 
    4892             : /*
    4893             :  * RelationGetStatExtList
    4894             :  *      get a list of OIDs of statistics objects on this relation
    4895             :  *
    4896             :  * The statistics list is created only if someone requests it, in a way
    4897             :  * similar to RelationGetIndexList().  We scan pg_statistic_ext to find
    4898             :  * relevant statistics, and add the list to the relcache entry so that we
    4899             :  * won't have to compute it again.  Note that shared cache inval of a
    4900             :  * relcache entry will delete the old list and set rd_statvalid to 0,
    4901             :  * so that we must recompute the statistics list on next request.  This
    4902             :  * handles creation or deletion of a statistics object.
    4903             :  *
    4904             :  * The returned list is guaranteed to be sorted in order by OID, although
    4905             :  * this is not currently needed.
    4906             :  *
    4907             :  * Since shared cache inval causes the relcache's copy of the list to go away,
    4908             :  * we return a copy of the list palloc'd in the caller's context.  The caller
    4909             :  * may list_free() the returned list after scanning it. This is necessary
    4910             :  * since the caller will typically be doing syscache lookups on the relevant
    4911             :  * statistics, and syscache lookup could cause SI messages to be processed!
    4912             :  */
    4913             : List *
    4914      384890 : RelationGetStatExtList(Relation relation)
    4915             : {
    4916             :     Relation    indrel;
    4917             :     SysScanDesc indscan;
    4918             :     ScanKeyData skey;
    4919             :     HeapTuple   htup;
    4920             :     List       *result;
    4921             :     List       *oldlist;
    4922             :     MemoryContext oldcxt;
    4923             : 
    4924             :     /* Quick exit if we already computed the list. */
    4925      384890 :     if (relation->rd_statvalid != 0)
    4926      291518 :         return list_copy(relation->rd_statlist);
    4927             : 
    4928             :     /*
    4929             :      * We build the list we intend to return (in the caller's context) while
    4930             :      * doing the scan.  After successfully completing the scan, we copy that
    4931             :      * list into the relcache entry.  This avoids cache-context memory leakage
    4932             :      * if we get some sort of error partway through.
    4933             :      */
    4934       93372 :     result = NIL;
    4935             : 
    4936             :     /*
    4937             :      * Prepare to scan pg_statistic_ext for entries having stxrelid = this
    4938             :      * rel.
    4939             :      */
    4940       93372 :     ScanKeyInit(&skey,
    4941             :                 Anum_pg_statistic_ext_stxrelid,
    4942             :                 BTEqualStrategyNumber, F_OIDEQ,
    4943             :                 ObjectIdGetDatum(RelationGetRelid(relation)));
    4944             : 
    4945       93372 :     indrel = table_open(StatisticExtRelationId, AccessShareLock);
    4946       93372 :     indscan = systable_beginscan(indrel, StatisticExtRelidIndexId, true,
    4947             :                                  NULL, 1, &skey);
    4948             : 
    4949       93738 :     while (HeapTupleIsValid(htup = systable_getnext(indscan)))
    4950             :     {
    4951         366 :         Oid         oid = ((Form_pg_statistic_ext) GETSTRUCT(htup))->oid;
    4952             : 
    4953         366 :         result = lappend_oid(result, oid);
    4954             :     }
    4955             : 
    4956       93372 :     systable_endscan(indscan);
    4957             : 
    4958       93372 :     table_close(indrel, AccessShareLock);
    4959             : 
    4960             :     /* Sort the result list into OID order, per API spec. */
    4961       93372 :     list_sort(result, list_oid_cmp);
    4962             : 
    4963             :     /* Now save a copy of the completed list in the relcache entry. */
    4964       93372 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    4965       93372 :     oldlist = relation->rd_statlist;
    4966       93372 :     relation->rd_statlist = list_copy(result);
    4967             : 
    4968       93372 :     relation->rd_statvalid = true;
    4969       93372 :     MemoryContextSwitchTo(oldcxt);
    4970             : 
    4971             :     /* Don't leak the old list, if there is one */
    4972       93372 :     list_free(oldlist);
    4973             : 
    4974       93372 :     return result;
    4975             : }
    4976             : 
    4977             : /*
    4978             :  * RelationGetPrimaryKeyIndex -- get OID of the relation's primary key index
    4979             :  *
    4980             :  * Returns InvalidOid if there is no such index.
    4981             :  */
    4982             : Oid
    4983         324 : RelationGetPrimaryKeyIndex(Relation relation)
    4984             : {
    4985             :     List       *ilist;
    4986             : 
    4987         324 :     if (!relation->rd_indexvalid)
    4988             :     {
    4989             :         /* RelationGetIndexList does the heavy lifting. */
    4990           0 :         ilist = RelationGetIndexList(relation);
    4991           0 :         list_free(ilist);
    4992             :         Assert(relation->rd_indexvalid);
    4993             :     }
    4994             : 
    4995         324 :     return relation->rd_pkindex;
    4996             : }
    4997             : 
    4998             : /*
    4999             :  * RelationGetReplicaIndex -- get OID of the relation's replica identity index
    5000             :  *
    5001             :  * Returns InvalidOid if there is no such index.
    5002             :  */
    5003             : Oid
    5004      316668 : RelationGetReplicaIndex(Relation relation)
    5005             : {
    5006             :     List       *ilist;
    5007             : 
    5008      316668 :     if (!relation->rd_indexvalid)
    5009             :     {
    5010             :         /* RelationGetIndexList does the heavy lifting. */
    5011        4620 :         ilist = RelationGetIndexList(relation);
    5012        4620 :         list_free(ilist);
    5013             :         Assert(relation->rd_indexvalid);
    5014             :     }
    5015             : 
    5016      316668 :     return relation->rd_replidindex;
    5017             : }
    5018             : 
    5019             : /*
    5020             :  * RelationGetIndexExpressions -- get the index expressions for an index
    5021             :  *
    5022             :  * We cache the result of transforming pg_index.indexprs into a node tree.
    5023             :  * If the rel is not an index or has no expressional columns, we return NIL.
    5024             :  * Otherwise, the returned tree is copied into the caller's memory context.
    5025             :  * (We don't want to return a pointer to the relcache copy, since it could
    5026             :  * disappear due to relcache invalidation.)
    5027             :  */
    5028             : List *
    5029     3247776 : RelationGetIndexExpressions(Relation relation)
    5030             : {
    5031             :     List       *result;
    5032             :     Datum       exprsDatum;
    5033             :     bool        isnull;
    5034             :     char       *exprsString;
    5035             :     MemoryContext oldcxt;
    5036             : 
    5037             :     /* Quick exit if we already computed the result. */
    5038     3247776 :     if (relation->rd_indexprs)
    5039        3060 :         return copyObject(relation->rd_indexprs);
    5040             : 
    5041             :     /* Quick exit if there is nothing to do. */
    5042     6489432 :     if (relation->rd_indextuple == NULL ||
    5043     3244716 :         heap_attisnull(relation->rd_indextuple, Anum_pg_index_indexprs, NULL))
    5044     3243362 :         return NIL;
    5045             : 
    5046             :     /*
    5047             :      * We build the tree we intend to return in the caller's context. After
    5048             :      * successfully completing the work, we copy it into the relcache entry.
    5049             :      * This avoids problems if we get some sort of error partway through.
    5050             :      */
    5051        1354 :     exprsDatum = heap_getattr(relation->rd_indextuple,
    5052             :                               Anum_pg_index_indexprs,
    5053             :                               GetPgIndexDescriptor(),
    5054             :                               &isnull);
    5055             :     Assert(!isnull);
    5056        1354 :     exprsString = TextDatumGetCString(exprsDatum);
    5057        1354 :     result = (List *) stringToNode(exprsString);
    5058        1354 :     pfree(exprsString);
    5059             : 
    5060             :     /*
    5061             :      * Run the expressions through eval_const_expressions. This is not just an
    5062             :      * optimization, but is necessary, because the planner will be comparing
    5063             :      * them to similarly-processed qual clauses, and may fail to detect valid
    5064             :      * matches without this.  We must not use canonicalize_qual, however,
    5065             :      * since these aren't qual expressions.
    5066             :      */
    5067        1354 :     result = (List *) eval_const_expressions(NULL, (Node *) result);
    5068             : 
    5069             :     /* May as well fix opfuncids too */
    5070        1352 :     fix_opfuncids((Node *) result);
    5071             : 
    5072             :     /* Now save a copy of the completed tree in the relcache entry. */
    5073        1352 :     oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
    5074        1352 :     relation->rd_indexprs = copyObject(result);
    5075        1352 :     MemoryContextSwitchTo(oldcxt);
    5076             : 
    5077        1352 :     return result;
    5078             : }
    5079             : 
    5080             : /*
    5081             :  * RelationGetDummyIndexExpressions -- get dummy expressions for an index
    5082             :  *
    5083             :  * Return a list of dummy expressions (just Const nodes) with the same
    5084             :  * types/typmods/collations as the index's real expressions.  This is
    5085             :  * useful in situations where we don't want to run any user-defined code.
    5086             :  */
    5087             : List *
    5088         244 : RelationGetDummyIndexExpressions(Relation relation)
    5089             : {
    5090             :     List       *result;
    5091             :     Datum       exprsDatum;
    5092             :     bool        isnull;
    5093             :     char       *exprsString;
    5094             :     List       *rawExprs;
    5095             :     ListCell   *lc;
    5096             : 
    5097             :     /* Quick exit if there is nothing to do. */
    5098         488 :     if (relation->rd_indextuple == NULL ||
    5099         244 :         heap_attisnull(relation->rd_indextuple, Anum_pg_index_indexprs, NULL))
    5100         226 :         return NIL;
    5101             : 
    5102             :     /* Extract raw node tree(s) from index tuple. */
    5103          18 :     exprsDatum = heap_getattr(relation->rd_indextuple,
    5104             :                               Anum_pg_index_indexprs,
    5105             :                               GetPgIndexDescriptor(),
    5106             :                               &isnull);
    5107             :     Assert(!isnull);
    5108          18 :     exprsString = TextDatumGetCString(exprsDatum);
    5109          18 :     rawExprs = (List *) stringToNode(exprsString);
    5110          18 :     pfree(exprsString);
    5111             : 
    5112             :     /* Construct null Consts; the typlen and typbyval are arbitrary. */
    5113          18 :     result = NIL;
    5114          36 :     foreach(lc, rawExprs)
    5115             :     {
    5116          18 :         Node       *rawExpr = (Node *) lfirst(lc);
    5117             : 
    5118          18 :         result = lappend(result,
    5119          18 :                          makeConst(exprType(rawExpr),
    5120             :                                    exprTypmod(rawExpr),
    5121             :                                    exprCollation(rawExpr),
    5122             :                                    1,
    5123             :                                    (Datum) 0,
    5124             :                                    true,
    5125             :                                    true));
    5126             :     }
    5127             : 
    5128          18 :     return result;
    5129             : }
    5130             : 
    5131             : /*
    5132             :  * RelationGetIndexPredicate -- get the index predicate for an index
    5133             :  *
    5134             :  * We cache the result of transforming pg_index.indpred into an implicit-AND
    5135             :  * node tree (suitable for use in planning).
    5136             :  * If the rel is not an index or has no predicate, we return NIL.
    5137             :  * Otherwise, the returned tree is copied into the caller's memory context.
    5138             :  * (We don't want to return a pointer to the relcache copy, since it could
    5139             :  * disappear due to relcache invalidation.)
    5140             :  */
    5141             : List *
    5142     3247692 : RelationGetIndexPredicate(Relation relation)
    5143             : {
    5144             :     List       *result;
    5145             :     Datum       predDatum;
    5146             :     bool        isnull;
    5147             :     char       *predString;
    5148             :     MemoryContext oldcxt;
    5149             : 
    5150             :     /* Quick exit if we already computed the result. */
    5151     3247692 :     if (relation->rd_indpred)
    5152        1218 :         return copyObject(relation->rd_indpred);
    5153             : 
    5154             :     /* Quick exit if there is nothing to do. */
    5155     6492948 :     if (relation->rd_indextuple == NULL ||
    5156     3246474 :         heap_attisnull(relation->rd_indextuple, Anum_pg_index_indpred, NULL))
    5157     3245592 :         return NIL;
    5158             : 
    5159             :     /*
    5160             :      * We build the tree we intend to return in the caller's context. After
    5161             :      * successfully completing the work, we copy it into the relcache entry.
    5162             :      * This avoids problems if we get some sort of error partway through.
    5163             :      */
    5164         882 :     predDatum = heap_getattr(relation->rd_indextuple,
    5165             :                              Anum_pg_index_indpred,
    5166             :                              GetPgIndexDescriptor(),
    5167             :                              &isnull);
    5168             :     Assert(!isnull);
    5169         882 :     predString = TextDatumGetCString(predDatum);
    5170         882 :     result = (List *) stringToNode(predString);
    5171         882 :     pfree(predString);
    5172             : 
    5173             :     /*
    5174             :      * Run the expression through const-simplification and canonicalization.
    5175             :      * This is not just an optimization, but is necessary, because the planner
    5176             :      * will be comparing it to similarly-processed qual clauses, and may fail
    5177             :      * to detect valid matches without this.  This must match the processing
    5178             :      * done to qual clauses in preprocess_expression()!  (We can skip the
    5179             :      * stuff involving subqueries, however, since we don't allow any in index
    5180             :      * predicates.)
    5181             :      */
    5182         882 :     result = (List *) eval_const_expressions(NULL, (Node *) result);
    5183             : 
    5184         882 :     result = (List *) canonicalize_qual((Expr *) result, false);
    5185             : 
    5186             :     /* Also convert to implicit-AND format */
    5187         882 :     result = make_ands_implicit((Expr *) result);
    5188             : 
    5189             :     /* May as well fix opfuncids too */
    5190         882 :     fix_opfuncids((Node *) result);
    5191             : 
    5192             :     /* Now save a copy of the completed tree in the relcache entry. */
    5193         882 :     oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
    5194         882 :     relation->rd_indpred = copyObject(result);
    5195         882 :     MemoryContextSwitchTo(oldcxt);
    5196             : 
    5197         882 :     return result;
    5198             : }
    5199             : 
    5200             : /*
    5201             :  * RelationGetIndexAttrBitmap -- get a bitmap of index attribute numbers
    5202             :  *
    5203             :  * The result has a bit set for each attribute used anywhere in the index
    5204             :  * definitions of all the indexes on this relation.  (This includes not only
    5205             :  * simple index keys, but attributes used in expressions and partial-index
    5206             :  * predicates.)
    5207             :  *
    5208             :  * Depending on attrKind, a bitmap covering attnums for certain columns is
    5209             :  * returned:
    5210             :  *  INDEX_ATTR_BITMAP_KEY           Columns in non-partial unique indexes not
    5211             :  *                                  in expressions (i.e., usable for FKs)
    5212             :  *  INDEX_ATTR_BITMAP_PRIMARY_KEY   Columns in the table's primary key
    5213             :  *  INDEX_ATTR_BITMAP_IDENTITY_KEY  Columns in the table's replica identity
    5214             :  *                                  index (empty if FULL)
    5215             :  *  INDEX_ATTR_BITMAP_HOT_BLOCKING  Columns that block updates from being HOT
    5216             :  *  INDEX_ATTR_BITMAP_SUMMARIZED    Columns included in summarizing indexes
    5217             :  *
    5218             :  * Attribute numbers are offset by FirstLowInvalidHeapAttributeNumber so that
    5219             :  * we can include system attributes (e.g., OID) in the bitmap representation.
    5220             :  *
    5221             :  * Caller had better hold at least RowExclusiveLock on the target relation
    5222             :  * to ensure it is safe (deadlock-free) for us to take locks on the relation's
    5223             :  * indexes.  Note that since the introduction of CREATE INDEX CONCURRENTLY,
    5224             :  * that lock level doesn't guarantee a stable set of indexes, so we have to
    5225             :  * be prepared to retry here in case of a change in the set of indexes.
    5226             :  *
    5227             :  * The returned result is palloc'd in the caller's memory context and should
    5228             :  * be bms_free'd when not needed anymore.
    5229             :  */
    5230             : Bitmapset *
    5231     2437094 : RelationGetIndexAttrBitmap(Relation relation, IndexAttrBitmapKind attrKind)
    5232             : {
    5233             :     Bitmapset  *uindexattrs;    /* columns in unique indexes */
    5234             :     Bitmapset  *pkindexattrs;   /* columns in the primary index */
    5235             :     Bitmapset  *idindexattrs;   /* columns in the replica identity */
    5236             :     Bitmapset  *hotblockingattrs;   /* columns with HOT blocking indexes */
    5237             :     Bitmapset  *summarizedattrs;    /* columns with summarizing indexes */
    5238             :     List       *indexoidlist;
    5239             :     List       *newindexoidlist;
    5240             :     Oid         relpkindex;
    5241             :     Oid         relreplindex;
    5242             :     ListCell   *l;
    5243             :     MemoryContext oldcxt;
    5244             : 
    5245             :     /* Quick exit if we already computed the result. */
    5246     2437094 :     if (relation->rd_attrsvalid)
    5247             :     {
    5248     2005020 :         switch (attrKind)
    5249             :         {
    5250      485092 :             case INDEX_ATTR_BITMAP_KEY:
    5251      485092 :                 return bms_copy(relation->rd_keyattr);
    5252         158 :             case INDEX_ATTR_BITMAP_PRIMARY_KEY:
    5253         158 :                 return bms_copy(relation->rd_pkattr);
    5254      572764 :             case INDEX_ATTR_BITMAP_IDENTITY_KEY:
    5255      572764 :                 return bms_copy(relation->rd_idattr);
    5256      468408 :             case INDEX_ATTR_BITMAP_HOT_BLOCKING:
    5257      468408 :                 return bms_copy(relation->rd_hotblockingattr);
    5258      478598 :             case INDEX_ATTR_BITMAP_SUMMARIZED:
    5259      478598 :                 return bms_copy(relation->rd_summarizedattr);
    5260           0 :             default:
    5261           0 :                 elog(ERROR, "unknown attrKind %u", attrKind);
    5262             :         }
    5263             :     }
    5264             : 
    5265             :     /* Fast path if definitely no indexes */
    5266      432074 :     if (!RelationGetForm(relation)->relhasindex)
    5267      417230 :         return NULL;
    5268             : 
    5269             :     /*
    5270             :      * Get cached list of index OIDs. If we have to start over, we do so here.
    5271             :      */
    5272       14844 : restart:
    5273       14844 :     indexoidlist = RelationGetIndexList(relation);
    5274             : 
    5275             :     /* Fall out if no indexes (but relhasindex was set) */
    5276       14844 :     if (indexoidlist == NIL)
    5277        1126 :         return NULL;
    5278             : 
    5279             :     /*
    5280             :      * Copy the rd_pkindex and rd_replidindex values computed by
    5281             :      * RelationGetIndexList before proceeding.  This is needed because a
    5282             :      * relcache flush could occur inside index_open below, resetting the
    5283             :      * fields managed by RelationGetIndexList.  We need to do the work with
    5284             :      * stable values of these fields.
    5285             :      */
    5286       13718 :     relpkindex = relation->rd_pkindex;
    5287       13718 :     relreplindex = relation->rd_replidindex;
    5288             : 
    5289             :     /*
    5290             :      * For each index, add referenced attributes to indexattrs.
    5291             :      *
    5292             :      * Note: we consider all indexes returned by RelationGetIndexList, even if
    5293             :      * they are not indisready or indisvalid.  This is important because an
    5294             :      * index for which CREATE INDEX CONCURRENTLY has just started must be
    5295             :      * included in HOT-safety decisions (see README.HOT).  If a DROP INDEX
    5296             :      * CONCURRENTLY is far enough along that we should ignore the index, it
    5297             :      * won't be returned at all by RelationGetIndexList.
    5298             :      */
    5299       13718 :     uindexattrs = NULL;
    5300       13718 :     pkindexattrs = NULL;
    5301       13718 :     idindexattrs = NULL;
    5302       13718 :     hotblockingattrs = NULL;
    5303       13718 :     summarizedattrs = NULL;
    5304       37170 :     foreach(l, indexoidlist)
    5305             :     {
    5306       23452 :         Oid         indexOid = lfirst_oid(l);
    5307             :         Relation    indexDesc;
    5308             :         Datum       datum;
    5309             :         bool        isnull;
    5310             :         Node       *indexExpressions;
    5311             :         Node       *indexPredicate;
    5312             :         int         i;
    5313             :         bool        isKey;      /* candidate key */
    5314             :         bool        isPK;       /* primary key */
    5315             :         bool        isIDKey;    /* replica identity index */
    5316             :         Bitmapset **attrs;
    5317             : 
    5318       23452 :         indexDesc = index_open(indexOid, AccessShareLock);
    5319             : 
    5320             :         /*
    5321             :          * Extract index expressions and index predicate.  Note: Don't use
    5322             :          * RelationGetIndexExpressions()/RelationGetIndexPredicate(), because
    5323             :          * those might run constant expressions evaluation, which needs a
    5324             :          * snapshot, which we might not have here.  (Also, it's probably more
    5325             :          * sound to collect the bitmaps before any transformations that might
    5326             :          * eliminate columns, but the practical impact of this is limited.)
    5327             :          */
    5328             : 
    5329       23452 :         datum = heap_getattr(indexDesc->rd_indextuple, Anum_pg_index_indexprs,
    5330             :                              GetPgIndexDescriptor(), &isnull);
    5331       23452 :         if (!isnull)
    5332         122 :             indexExpressions = stringToNode(TextDatumGetCString(datum));
    5333             :         else
    5334       23330 :             indexExpressions = NULL;
    5335             : 
    5336       23452 :         datum = heap_getattr(indexDesc->rd_indextuple, Anum_pg_index_indpred,
    5337             :                              GetPgIndexDescriptor(), &isnull);
    5338       23452 :         if (!isnull)
    5339         114 :             indexPredicate = stringToNode(TextDatumGetCString(datum));
    5340             :         else
    5341       23338 :             indexPredicate = NULL;
    5342             : 
    5343             :         /* Can this index be referenced by a foreign key? */
    5344       18684 :         isKey = indexDesc->rd_index->indisunique &&
    5345       42136 :             indexExpressions == NULL &&
    5346             :             indexPredicate == NULL;
    5347             : 
    5348             :         /* Is this a primary key? */
    5349       23452 :         isPK = (indexOid == relpkindex);
    5350             : 
    5351             :         /* Is this index the configured (or default) replica identity? */
    5352       23452 :         isIDKey = (indexOid == relreplindex);
    5353             : 
    5354             :         /*
    5355             :          * If the index is summarizing, it doesn't block HOT updates, but we
    5356             :          * may still need to update it (if the attributes were modified). So
    5357             :          * decide which bitmap we'll update in the following loop.
    5358             :          */
    5359       23452 :         if (indexDesc->rd_indam->amsummarizing)
    5360          56 :             attrs = &summarizedattrs;
    5361             :         else
    5362       23396 :             attrs = &hotblockingattrs;
    5363             : 
    5364             :         /* Collect simple attribute references */
    5365       59436 :         for (i = 0; i < indexDesc->rd_index->indnatts; i++)
    5366             :         {
    5367       35984 :             int         attrnum = indexDesc->rd_index->indkey.values[i];
    5368             : 
    5369             :             /*
    5370             :              * Since we have covering indexes with non-key columns, we must
    5371             :              * handle them accurately here. non-key columns must be added into
    5372             :              * hotblockingattrs or summarizedattrs, since they are in index,
    5373             :              * and update shouldn't miss them.
    5374             :              *
    5375             :              * Summarizing indexes do not block HOT, but do need to be updated
    5376             :              * when the column value changes, thus require a separate
    5377             :              * attribute bitmapset.
    5378             :              *
    5379             :              * Obviously, non-key columns couldn't be referenced by foreign
    5380             :              * key or identity key. Hence we do not include them into
    5381             :              * uindexattrs, pkindexattrs and idindexattrs bitmaps.
    5382             :              */
    5383       35984 :             if (attrnum != 0)
    5384             :             {
    5385       35862 :                 *attrs = bms_add_member(*attrs,
    5386             :                                         attrnum - FirstLowInvalidHeapAttributeNumber);
    5387             : 
    5388       35862 :                 if (isKey && i < indexDesc->rd_index->indnkeyatts)
    5389       27158 :                     uindexattrs = bms_add_member(uindexattrs,
    5390             :                                                  attrnum - FirstLowInvalidHeapAttributeNumber);
    5391             : 
    5392       35862 :                 if (isPK && i < indexDesc->rd_index->indnkeyatts)
    5393       14466 :                     pkindexattrs = bms_add_member(pkindexattrs,
    5394             :                                                   attrnum - FirstLowInvalidHeapAttributeNumber);
    5395             : 
    5396       35862 :                 if (isIDKey && i < indexDesc->rd_index->indnkeyatts)
    5397        5186 :                     idindexattrs = bms_add_member(idindexattrs,
    5398             :                                                   attrnum - FirstLowInvalidHeapAttributeNumber);
    5399             :             }
    5400             :         }
    5401             : 
    5402             :         /* Collect all attributes used in expressions, too */
    5403       23452 :         pull_varattnos(indexExpressions, 1, attrs);
    5404             : 
    5405             :         /* Collect all attributes in the index predicate, too */
    5406       23452 :         pull_varattnos(indexPredicate, 1, attrs);
    5407             : 
    5408       23452 :         index_close(indexDesc, AccessShareLock);
    5409             :     }
    5410             : 
    5411             :     /*
    5412             :      * During one of the index_opens in the above loop, we might have received
    5413             :      * a relcache flush event on this relcache entry, which might have been
    5414             :      * signaling a change in the rel's index list.  If so, we'd better start
    5415             :      * over to ensure we deliver up-to-date attribute bitmaps.
    5416             :      */
    5417       13718 :     newindexoidlist = RelationGetIndexList(relation);
    5418       13718 :     if (equal(indexoidlist, newindexoidlist) &&
    5419       13718 :         relpkindex == relation->rd_pkindex &&
    5420       13718 :         relreplindex == relation->rd_replidindex)
    5421             :     {
    5422             :         /* Still the same index set, so proceed */
    5423       13718 :         list_free(newindexoidlist);
    5424       13718 :         list_free(indexoidlist);
    5425             :     }
    5426             :     else
    5427             :     {
    5428             :         /* Gotta do it over ... might as well not leak memory */
    5429           0 :         list_free(newindexoidlist);
    5430           0 :         list_free(indexoidlist);
    5431           0 :         bms_free(uindexattrs);
    5432           0 :         bms_free(pkindexattrs);
    5433           0 :         bms_free(idindexattrs);
    5434           0 :         bms_free(hotblockingattrs);
    5435           0 :         bms_free(summarizedattrs);
    5436             : 
    5437           0 :         goto restart;
    5438             :     }
    5439             : 
    5440             :     /* Don't leak the old values of these bitmaps, if any */
    5441       13718 :     relation->rd_attrsvalid = false;
    5442       13718 :     bms_free(relation->rd_keyattr);
    5443       13718 :     relation->rd_keyattr = NULL;
    5444       13718 :     bms_free(relation->rd_pkattr);
    5445       13718 :     relation->rd_pkattr = NULL;
    5446       13718 :     bms_free(relation->rd_idattr);
    5447       13718 :     relation->rd_idattr = NULL;
    5448       13718 :     bms_free(relation->rd_hotblockingattr);
    5449       13718 :     relation->rd_hotblockingattr = NULL;
    5450       13718 :     bms_free(relation->rd_summarizedattr);
    5451       13718 :     relation->rd_summarizedattr = NULL;
    5452             : 
    5453             :     /*
    5454             :      * Now save copies of the bitmaps in the relcache entry.  We intentionally
    5455             :      * set rd_attrsvalid last, because that's the one that signals validity of
    5456             :      * the values; if we run out of memory before making that copy, we won't
    5457             :      * leave the relcache entry looking like the other ones are valid but
    5458             :      * empty.
    5459             :      */
    5460       13718 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    5461       13718 :     relation->rd_keyattr = bms_copy(uindexattrs);
    5462       13718 :     relation->rd_pkattr = bms_copy(pkindexattrs);
    5463       13718 :     relation->rd_idattr = bms_copy(idindexattrs);
    5464       13718 :     relation->rd_hotblockingattr = bms_copy(hotblockingattrs);
    5465       13718 :     relation->rd_summarizedattr = bms_copy(summarizedattrs);
    5466       13718 :     relation->rd_attrsvalid = true;
    5467       13718 :     MemoryContextSwitchTo(oldcxt);
    5468             : 
    5469             :     /* We return our original working copy for caller to play with */
    5470       13718 :     switch (attrKind)
    5471             :     {
    5472         846 :         case INDEX_ATTR_BITMAP_KEY:
    5473         846 :             return uindexattrs;
    5474        1706 :         case INDEX_ATTR_BITMAP_PRIMARY_KEY:
    5475        1706 :             return pkindexattrs;
    5476         976 :         case INDEX_ATTR_BITMAP_IDENTITY_KEY:
    5477         976 :             return idindexattrs;
    5478       10190 :         case INDEX_ATTR_BITMAP_HOT_BLOCKING:
    5479       10190 :             return hotblockingattrs;
    5480           0 :         case INDEX_ATTR_BITMAP_SUMMARIZED:
    5481           0 :             return summarizedattrs;
    5482           0 :         default:
    5483           0 :             elog(ERROR, "unknown attrKind %u", attrKind);
    5484             :             return NULL;
    5485             :     }
    5486             : }
    5487             : 
    5488             : /*
    5489             :  * RelationGetIdentityKeyBitmap -- get a bitmap of replica identity attribute
    5490             :  * numbers
    5491             :  *
    5492             :  * A bitmap of index attribute numbers for the configured replica identity
    5493             :  * index is returned.
    5494             :  *
    5495             :  * See also comments of RelationGetIndexAttrBitmap().
    5496             :  *
    5497             :  * This is a special purpose function used during logical replication. Here,
    5498             :  * unlike RelationGetIndexAttrBitmap(), we don't acquire a lock on the required
    5499             :  * index as we build the cache entry using a historic snapshot and all the
    5500             :  * later changes are absorbed while decoding WAL. Due to this reason, we don't
    5501             :  * need to retry here in case of a change in the set of indexes.
    5502             :  */
    5503             : Bitmapset *
    5504         530 : RelationGetIdentityKeyBitmap(Relation relation)
    5505             : {
    5506         530 :     Bitmapset  *idindexattrs = NULL;    /* columns in the replica identity */
    5507             :     Relation    indexDesc;
    5508             :     int         i;
    5509             :     Oid         replidindex;
    5510             :     MemoryContext oldcxt;
    5511             : 
    5512             :     /* Quick exit if we already computed the result */
    5513         530 :     if (relation->rd_idattr != NULL)
    5514          90 :         return bms_copy(relation->rd_idattr);
    5515             : 
    5516             :     /* Fast path if definitely no indexes */
    5517         440 :     if (!RelationGetForm(relation)->relhasindex)
    5518         122 :         return NULL;
    5519             : 
    5520             :     /* Historic snapshot must be set. */
    5521             :     Assert(HistoricSnapshotActive());
    5522             : 
    5523         318 :     replidindex = RelationGetReplicaIndex(relation);
    5524             : 
    5525             :     /* Fall out if there is no replica identity index */
    5526         318 :     if (!OidIsValid(replidindex))
    5527           4 :         return NULL;
    5528             : 
    5529             :     /* Look up the description for the replica identity index */
    5530         314 :     indexDesc = RelationIdGetRelation(replidindex);
    5531             : 
    5532         314 :     if (!RelationIsValid(indexDesc))
    5533           0 :         elog(ERROR, "could not open relation with OID %u",
    5534             :              relation->rd_replidindex);
    5535             : 
    5536             :     /* Add referenced attributes to idindexattrs */
    5537         636 :     for (i = 0; i < indexDesc->rd_index->indnatts; i++)
    5538             :     {
    5539         322 :         int         attrnum = indexDesc->rd_index->indkey.values[i];
    5540             : 
    5541             :         /*
    5542             :          * We don't include non-key columns into idindexattrs bitmaps. See
    5543             :          * RelationGetIndexAttrBitmap.
    5544             :          */
    5545         322 :         if (attrnum != 0)
    5546             :         {
    5547         322 :             if (i < indexDesc->rd_index->indnkeyatts)
    5548         320 :                 idindexattrs = bms_add_member(idindexattrs,
    5549             :                                               attrnum - FirstLowInvalidHeapAttributeNumber);
    5550             :         }
    5551             :     }
    5552             : 
    5553         314 :     RelationClose(indexDesc);
    5554             : 
    5555             :     /* Don't leak the old values of these bitmaps, if any */
    5556         314 :     bms_free(relation->rd_idattr);
    5557         314 :     relation->rd_idattr = NULL;
    5558             : 
    5559             :     /* Now save copy of the bitmap in the relcache entry */
    5560         314 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    5561         314 :     relation->rd_idattr = bms_copy(idindexattrs);
    5562         314 :     MemoryContextSwitchTo(oldcxt);
    5563             : 
    5564             :     /* We return our original working copy for caller to play with */
    5565         314 :     return idindexattrs;
    5566             : }
    5567             : 
    5568             : /*
    5569             :  * RelationGetExclusionInfo -- get info about index's exclusion constraint
    5570             :  *
    5571             :  * This should be called only for an index that is known to have an
    5572             :  * associated exclusion constraint.  It returns arrays (palloc'd in caller's
    5573             :  * context) of the exclusion operator OIDs, their underlying functions'
    5574             :  * OIDs, and their strategy numbers in the index's opclasses.  We cache
    5575             :  * all this information since it requires a fair amount of work to get.
    5576             :  */
    5577             : void
    5578         330 : RelationGetExclusionInfo(Relation indexRelation,
    5579             :                          Oid **operators,
    5580             :                          Oid **procs,
    5581             :                          uint16 **strategies)
    5582             : {
    5583             :     int         indnkeyatts;
    5584             :     Oid        *ops;
    5585             :     Oid        *funcs;
    5586             :     uint16     *strats;
    5587             :     Relation    conrel;
    5588             :     SysScanDesc conscan;
    5589             :     ScanKeyData skey[1];
    5590             :     HeapTuple   htup;
    5591             :     bool        found;
    5592             :     MemoryContext oldcxt;
    5593             :     int         i;
    5594             : 
    5595         330 :     indnkeyatts = IndexRelationGetNumberOfKeyAttributes(indexRelation);
    5596             : 
    5597             :     /* Allocate result space in caller context */
    5598         330 :     *operators = ops = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
    5599         330 :     *procs = funcs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
    5600         330 :     *strategies = strats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
    5601             : 
    5602             :     /* Quick exit if we have the data cached already */
    5603         330 :     if (indexRelation->rd_exclstrats != NULL)
    5604             :     {
    5605         220 :         memcpy(ops, indexRelation->rd_exclops, sizeof(Oid) * indnkeyatts);
    5606         220 :         memcpy(funcs, indexRelation->rd_exclprocs, sizeof(Oid) * indnkeyatts);
    5607         220 :         memcpy(strats, indexRelation->rd_exclstrats, sizeof(uint16) * indnkeyatts);
    5608         220 :         return;
    5609             :     }
    5610             : 
    5611             :     /*
    5612             :      * Search pg_constraint for the constraint associated with the index. To
    5613             :      * make this not too painfully slow, we use the index on conrelid; that
    5614             :      * will hold the parent relation's OID not the index's own OID.
    5615             :      *
    5616             :      * Note: if we wanted to rely on the constraint name matching the index's
    5617             :      * name, we could just do a direct lookup using pg_constraint's unique
    5618             :      * index.  For the moment it doesn't seem worth requiring that.
    5619             :      */
    5620         110 :     ScanKeyInit(&skey[0],
    5621             :                 Anum_pg_constraint_conrelid,
    5622             :                 BTEqualStrategyNumber, F_OIDEQ,
    5623         110 :                 ObjectIdGetDatum(indexRelation->rd_index->indrelid));
    5624             : 
    5625         110 :     conrel = table_open(ConstraintRelationId, AccessShareLock);
    5626         110 :     conscan = systable_beginscan(conrel, ConstraintRelidTypidNameIndexId, true,
    5627             :                                  NULL, 1, skey);
    5628         110 :     found = false;
    5629             : 
    5630         508 :     while (HeapTupleIsValid(htup = systable_getnext(conscan)))
    5631             :     {
    5632         398 :         Form_pg_constraint conform = (Form_pg_constraint) GETSTRUCT(htup);
    5633             :         Datum       val;
    5634             :         bool        isnull;
    5635             :         ArrayType  *arr;
    5636             :         int         nelem;
    5637             : 
    5638             :         /* We want the exclusion constraint owning the index */
    5639         398 :         if (conform->contype != CONSTRAINT_EXCLUSION ||
    5640         230 :             conform->conindid != RelationGetRelid(indexRelation))
    5641         288 :             continue;
    5642             : 
    5643             :         /* There should be only one */
    5644         110 :         if (found)
    5645           0 :             elog(ERROR, "unexpected exclusion constraint record found for rel %s",
    5646             :                  RelationGetRelationName(indexRelation));
    5647         110 :         found = true;
    5648             : 
    5649             :         /* Extract the operator OIDS from conexclop */
    5650         110 :         val = fastgetattr(htup,
    5651             :                           Anum_pg_constraint_conexclop,
    5652             :                           conrel->rd_att, &isnull);
    5653         110 :         if (isnull)
    5654           0 :             elog(ERROR, "null conexclop for rel %s",
    5655             :                  RelationGetRelationName(indexRelation));
    5656             : 
    5657         110 :         arr = DatumGetArrayTypeP(val);  /* ensure not toasted */
    5658         110 :         nelem = ARR_DIMS(arr)[0];
    5659         110 :         if (ARR_NDIM(arr) != 1 ||
    5660         110 :             nelem != indnkeyatts ||
    5661         110 :             ARR_HASNULL(arr) ||
    5662         110 :             ARR_ELEMTYPE(arr) != OIDOID)
    5663           0 :             elog(ERROR, "conexclop is not a 1-D Oid array");
    5664             : 
    5665         110 :         memcpy(ops, ARR_DATA_PTR(arr), sizeof(Oid) * indnkeyatts);
    5666             :     }
    5667             : 
    5668         110 :     systable_endscan(conscan);
    5669         110 :     table_close(conrel, AccessShareLock);
    5670             : 
    5671         110 :     if (!found)
    5672           0 :         elog(ERROR, "exclusion constraint record missing for rel %s",
    5673             :              RelationGetRelationName(indexRelation));
    5674             : 
    5675             :     /* We need the func OIDs and strategy numbers too */
    5676         244 :     for (i = 0; i < indnkeyatts; i++)
    5677             :     {
    5678         134 :         funcs[i] = get_opcode(ops[i]);
    5679         268 :         strats[i] = get_op_opfamily_strategy(ops[i],
    5680         134 :                                              indexRelation->rd_opfamily[i]);
    5681             :         /* shouldn't fail, since it was checked at index creation */
    5682         134 :         if (strats[i] == InvalidStrategy)
    5683           0 :             elog(ERROR, "could not find strategy for operator %u in family %u",
    5684             :                  ops[i], indexRelation->rd_opfamily[i]);
    5685             :     }
    5686             : 
    5687             :     /* Save a copy of the results in the relcache entry. */
    5688         110 :     oldcxt = MemoryContextSwitchTo(indexRelation->rd_indexcxt);
    5689         110 :     indexRelation->rd_exclops = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
    5690         110 :     indexRelation->rd_exclprocs = (Oid *) palloc(sizeof(Oid) * indnkeyatts);
    5691         110 :     indexRelation->rd_exclstrats = (uint16 *) palloc(sizeof(uint16) * indnkeyatts);
    5692         110 :     memcpy(indexRelation->rd_exclops, ops, sizeof(Oid) * indnkeyatts);
    5693         110 :     memcpy(indexRelation->rd_exclprocs, funcs, sizeof(Oid) * indnkeyatts);
    5694         110 :     memcpy(indexRelation->rd_exclstrats, strats, sizeof(uint16) * indnkeyatts);
    5695         110 :     MemoryContextSwitchTo(oldcxt);
    5696             : }
    5697             : 
    5698             : /*
    5699             :  * Get the publication information for the given relation.
    5700             :  *
    5701             :  * Traverse all the publications which the relation is in to get the
    5702             :  * publication actions and validate the row filter expressions for such
    5703             :  * publications if any. We consider the row filter expression as invalid if it
    5704             :  * references any column which is not part of REPLICA IDENTITY.
    5705             :  *
    5706             :  * To avoid fetching the publication information repeatedly, we cache the
    5707             :  * publication actions and row filter validation information.
    5708             :  */
    5709             : void
    5710      170844 : RelationBuildPublicationDesc(Relation relation, PublicationDesc *pubdesc)
    5711             : {
    5712             :     List       *puboids;
    5713             :     ListCell   *lc;
    5714             :     MemoryContext oldcxt;
    5715             :     Oid         schemaid;
    5716      170844 :     List       *ancestors = NIL;
    5717      170844 :     Oid         relid = RelationGetRelid(relation);
    5718             : 
    5719             :     /*
    5720             :      * If not publishable, it publishes no actions.  (pgoutput_change() will
    5721             :      * ignore it.)
    5722             :      */
    5723      170844 :     if (!is_publishable_relation(relation))
    5724             :     {
    5725        5498 :         memset(pubdesc, 0, sizeof(PublicationDesc));
    5726        5498 :         pubdesc->rf_valid_for_update = true;
    5727        5498 :         pubdesc->rf_valid_for_delete = true;
    5728        5498 :         pubdesc->cols_valid_for_update = true;
    5729        5498 :         pubdesc->cols_valid_for_delete = true;
    5730        5498 :         return;
    5731             :     }
    5732             : 
    5733      165346 :     if (relation->rd_pubdesc)
    5734             :     {
    5735      157978 :         memcpy(pubdesc, relation->rd_pubdesc, sizeof(PublicationDesc));
    5736      157978 :         return;
    5737             :     }
    5738             : 
    5739        7368 :     memset(pubdesc, 0, sizeof(PublicationDesc));
    5740        7368 :     pubdesc->rf_valid_for_update = true;
    5741        7368 :     pubdesc->rf_valid_for_delete = true;
    5742        7368 :     pubdesc->cols_valid_for_update = true;
    5743        7368 :     pubdesc->cols_valid_for_delete = true;
    5744             : 
    5745             :     /* Fetch the publication membership info. */
    5746        7368 :     puboids = GetRelationPublications(relid);
    5747        7368 :     schemaid = RelationGetNamespace(relation);
    5748        7368 :     puboids = list_concat_unique_oid(puboids, GetSchemaPublications(schemaid));
    5749             : 
    5750        7368 :     if (relation->rd_rel->relispartition)
    5751             :     {
    5752             :         /* Add publications that the ancestors are in too. */
    5753        1686 :         ancestors = get_partition_ancestors(relid);
    5754             : 
    5755        3956 :         foreach(lc, ancestors)
    5756             :         {
    5757        2270 :             Oid         ancestor = lfirst_oid(lc);
    5758             : 
    5759        2270 :             puboids = list_concat_unique_oid(puboids,
    5760        2270 :                                              GetRelationPublications(ancestor));
    5761        2270 :             schemaid = get_rel_namespace(ancestor);
    5762        2270 :             puboids = list_concat_unique_oid(puboids,
    5763        2270 :                                              GetSchemaPublications(schemaid));
    5764             :         }
    5765             :     }
    5766        7368 :     puboids = list_concat_unique_oid(puboids, GetAllTablesPublications());
    5767             : 
    5768        8000 :     foreach(lc, puboids)
    5769             :     {
    5770         800 :         Oid         pubid = lfirst_oid(lc);
    5771             :         HeapTuple   tup;
    5772             :         Form_pg_publication pubform;
    5773             : 
    5774         800 :         tup = SearchSysCache1(PUBLICATIONOID, ObjectIdGetDatum(pubid));
    5775             : 
    5776         800 :         if (!HeapTupleIsValid(tup))
    5777           0 :             elog(ERROR, "cache lookup failed for publication %u", pubid);
    5778             : 
    5779         800 :         pubform = (Form_pg_publication) GETSTRUCT(tup);
    5780             : 
    5781         800 :         pubdesc->pubactions.pubinsert |= pubform->pubinsert;
    5782         800 :         pubdesc->pubactions.pubupdate |= pubform->pubupdate;
    5783         800 :         pubdesc->pubactions.pubdelete |= pubform->pubdelete;
    5784         800 :         pubdesc->pubactions.pubtruncate |= pubform->pubtruncate;
    5785             : 
    5786             :         /*
    5787             :          * Check if all columns referenced in the filter expression are part
    5788             :          * of the REPLICA IDENTITY index or not.
    5789             :          *
    5790             :          * If the publication is FOR ALL TABLES then it means the table has no
    5791             :          * row filters and we can skip the validation.
    5792             :          */
    5793         800 :         if (!pubform->puballtables &&
    5794        1304 :             (pubform->pubupdate || pubform->pubdelete) &&
    5795         650 :             pub_rf_contains_invalid_column(pubid, relation, ancestors,
    5796         650 :                                            pubform->pubviaroot))
    5797             :         {
    5798          60 :             if (pubform->pubupdate)
    5799          60 :                 pubdesc->rf_valid_for_update = false;
    5800          60 :             if (pubform->pubdelete)
    5801          60 :                 pubdesc->rf_valid_for_delete = false;
    5802             :         }
    5803             : 
    5804             :         /*
    5805             :          * Check if all columns are part of the REPLICA IDENTITY index or not.
    5806             :          *
    5807             :          * If the publication is FOR ALL TABLES then it means the table has no
    5808             :          * column list and we can skip the validation.
    5809             :          */
    5810         800 :         if (!pubform->puballtables &&
    5811        1304 :             (pubform->pubupdate || pubform->pubdelete) &&
    5812         650 :             pub_collist_contains_invalid_column(pubid, relation, ancestors,
    5813         650 :                                                 pubform->pubviaroot))
    5814             :         {
    5815         108 :             if (pubform->pubupdate)
    5816         108 :                 pubdesc->cols_valid_for_update = false;
    5817         108 :             if (pubform->pubdelete)
    5818         108 :                 pubdesc->cols_valid_for_delete = false;
    5819             :         }
    5820             : 
    5821         800 :         ReleaseSysCache(tup);
    5822             : 
    5823             :         /*
    5824             :          * If we know everything is replicated and the row filter is invalid
    5825             :          * for update and delete, there is no point to check for other
    5826             :          * publications.
    5827             :          */
    5828         800 :         if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
    5829         794 :             pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
    5830         782 :             !pubdesc->rf_valid_for_update && !pubdesc->rf_valid_for_delete)
    5831          60 :             break;
    5832             : 
    5833             :         /*
    5834             :          * If we know everything is replicated and the column list is invalid
    5835             :          * for update and delete, there is no point to check for other
    5836             :          * publications.
    5837             :          */
    5838         740 :         if (pubdesc->pubactions.pubinsert && pubdesc->pubactions.pubupdate &&
    5839         734 :             pubdesc->pubactions.pubdelete && pubdesc->pubactions.pubtruncate &&
    5840         722 :             !pubdesc->cols_valid_for_update && !pubdesc->cols_valid_for_delete)
    5841         108 :             break;
    5842             :     }
    5843             : 
    5844        7368 :     if (relation->rd_pubdesc)
    5845             :     {
    5846           0 :         pfree(relation->rd_pubdesc);
    5847           0 :         relation->rd_pubdesc = NULL;
    5848             :     }
    5849             : 
    5850             :     /* Now save copy of the descriptor in the relcache entry. */
    5851        7368 :     oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
    5852        7368 :     relation->rd_pubdesc = palloc(sizeof(PublicationDesc));
    5853        7368 :     memcpy(relation->rd_pubdesc, pubdesc, sizeof(PublicationDesc));
    5854        7368 :     MemoryContextSwitchTo(oldcxt);
    5855             : }
    5856             : 
    5857             : static bytea **
    5858      958648 : CopyIndexAttOptions(bytea **srcopts, int natts)
    5859             : {
    5860      958648 :     bytea     **opts = palloc(sizeof(*opts) * natts);
    5861             : 
    5862     2731628 :     for (int i = 0; i < natts; i++)
    5863             :     {
    5864     1772980 :         bytea      *opt = srcopts[i];
    5865             : 
    5866     1864736 :         opts[i] = !opt ? NULL : (bytea *)
    5867       91756 :             DatumGetPointer(datumCopy(PointerGetDatum(opt), false, -1));
    5868             :     }
    5869             : 
    5870      958648 :     return opts;
    5871             : }
    5872             : 
    5873             : /*
    5874             :  * RelationGetIndexAttOptions
    5875             :  *      get AM/opclass-specific options for an index parsed into a binary form
    5876             :  */
    5877             : bytea     **
    5878     1851922 : RelationGetIndexAttOptions(Relation relation, bool copy)
    5879             : {
    5880             :     MemoryContext oldcxt;
    5881     1851922 :     bytea     **opts = relation->rd_opcoptions;
    5882     1851922 :     Oid         relid = RelationGetRelid(relation);
    5883     1851922 :     int         natts = RelationGetNumberOfAttributes(relation);    /* XXX
    5884             :                                                                      * IndexRelationGetNumberOfKeyAttributes */
    5885             :     int         i;
    5886             : 
    5887             :     /* Try to copy cached options. */
    5888     1851922 :     if (opts)
    5889     1471608 :         return copy ? CopyIndexAttOptions(opts, natts) : opts;
    5890             : 
    5891             :     /* Get and parse opclass options. */
    5892      380314 :     opts = palloc0(sizeof(*opts) * natts);
    5893             : 
    5894     1023968 :     for (i = 0; i < natts; i++)
    5895             :     {
    5896      643660 :         if (criticalRelcachesBuilt && relid != AttributeRelidNumIndexId)
    5897             :         {
    5898      601490 :             Datum       attoptions = get_attoptions(relid, i + 1);
    5899             : 
    5900      601490 :             opts[i] = index_opclass_options(relation, i + 1, attoptions, false);
    5901             : 
    5902      601484 :             if (attoptions != (Datum) 0)
    5903         284 :                 pfree(DatumGetPointer(attoptions));
    5904             :         }
    5905             :     }
    5906             : 
    5907             :     /* Copy parsed options to the cache. */
    5908      380308 :     oldcxt = MemoryContextSwitchTo(relation->rd_indexcxt);
    5909      380308 :     relation->rd_opcoptions = CopyIndexAttOptions(opts, natts);
    5910      380308 :     MemoryContextSwitchTo(oldcxt);
    5911             : 
    5912      380308 :     if (copy)
    5913           0 :         return opts;
    5914             : 
    5915     1023962 :     for (i = 0; i < natts; i++)
    5916             :     {
    5917      643654 :         if (opts[i])
    5918        1768 :             pfree(opts[i]);
    5919             :     }
    5920             : 
    5921      380308 :     pfree(opts);
    5922             : 
    5923      380308 :     return relation->rd_opcoptions;
    5924             : }
    5925             : 
    5926             : /*
    5927             :  * Routines to support ereport() reports of relation-related errors
    5928             :  *
    5929             :  * These could have been put into elog.c, but it seems like a module layering
    5930             :  * violation to have elog.c calling relcache or syscache stuff --- and we
    5931             :  * definitely don't want elog.h including rel.h.  So we put them here.
    5932             :  */
    5933             : 
    5934             : /*
    5935             :  * errtable --- stores schema_name and table_name of a table
    5936             :  * within the current errordata.
    5937             :  */
    5938             : int
    5939        2868 : errtable(Relation rel)
    5940             : {
    5941        2868 :     err_generic_string(PG_DIAG_SCHEMA_NAME,
    5942        2868 :                        get_namespace_name(RelationGetNamespace(rel)));
    5943        2868 :     err_generic_string(PG_DIAG_TABLE_NAME, RelationGetRelationName(rel));
    5944             : 
    5945        2868 :     return 0;                   /* return value does not matter */
    5946             : }
    5947             : 
    5948             : /*
    5949             :  * errtablecol --- stores schema_name, table_name and column_name
    5950             :  * of a table column within the current errordata.
    5951             :  *
    5952             :  * The column is specified by attribute number --- for most callers, this is
    5953             :  * easier and less error-prone than getting the column name for themselves.
    5954             :  */
    5955             : int
    5956         344 : errtablecol(Relation rel, int attnum)
    5957             : {
    5958         344 :     TupleDesc   reldesc = RelationGetDescr(rel);
    5959             :     const char *colname;
    5960             : 
    5961             :     /* Use reldesc if it's a user attribute, else consult the catalogs */
    5962         344 :     if (attnum > 0 && attnum <= reldesc->natts)
    5963         344 :         colname = NameStr(TupleDescAttr(reldesc, attnum - 1)->attname);
    5964             :     else
    5965           0 :         colname = get_attname(RelationGetRelid(rel), attnum, false);
    5966             : 
    5967         344 :     return errtablecolname(rel, colname);
    5968             : }
    5969             : 
    5970             : /*
    5971             :  * errtablecolname --- stores schema_name, table_name and column_name
    5972             :  * of a table column within the current errordata, where the column name is
    5973             :  * given directly rather than extracted from the relation's catalog data.
    5974             :  *
    5975             :  * Don't use this directly unless errtablecol() is inconvenient for some
    5976             :  * reason.  This might possibly be needed during intermediate states in ALTER
    5977             :  * TABLE, for instance.
    5978             :  */
    5979             : int
    5980         344 : errtablecolname(Relation rel, const char *colname)
    5981             : {
    5982         344 :     errtable(rel);
    5983         344 :     err_generic_string(PG_DIAG_COLUMN_NAME, colname);
    5984             : 
    5985         344 :     return 0;                   /* return value does not matter */
    5986             : }
    5987             : 
    5988             : /*
    5989             :  * errtableconstraint --- stores schema_name, table_name and constraint_name
    5990             :  * of a table-related constraint within the current errordata.
    5991             :  */
    5992             : int
    5993        2028 : errtableconstraint(Relation rel, const char *conname)
    5994             : {
    5995        2028 :     errtable(rel);
    5996        2028 :     err_generic_string(PG_DIAG_CONSTRAINT_NAME, conname);
    5997             : 
    5998        2028 :     return 0;                   /* return value does not matter */
    5999             : }
    6000             : 
    6001             : 
    6002             : /*
    6003             :  *  load_relcache_init_file, write_relcache_init_file
    6004             :  *
    6005             :  *      In late 1992, we started regularly having databases with more than
    6006             :  *      a thousand classes in them.  With this number of classes, it became
    6007             :  *      critical to do indexed lookups on the system catalogs.
    6008             :  *
    6009             :  *      Bootstrapping these lookups is very hard.  We want to be able to
    6010             :  *      use an index on pg_attribute, for example, but in order to do so,
    6011             :  *      we must have read pg_attribute for the attributes in the index,
    6012             :  *      which implies that we need to use the index.
    6013             :  *
    6014             :  *      In order to get around the problem, we do the following:
    6015             :  *
    6016             :  *         +  When the database system is initialized (at initdb time), we
    6017             :  *            don't use indexes.  We do sequential scans.
    6018             :  *
    6019             :  *         +  When the backend is started up in normal mode, we load an image
    6020             :  *            of the appropriate relation descriptors, in internal format,
    6021             :  *            from an initialization file in the data/base/... directory.
    6022             :  *
    6023             :  *         +  If the initialization file isn't there, then we create the
    6024             :  *            relation descriptors using sequential scans and write 'em to
    6025             :  *            the initialization file for use by subsequent backends.
    6026             :  *
    6027             :  *      As of Postgres 9.0, there is one local initialization file in each
    6028             :  *      database, plus one shared initialization file for shared catalogs.
    6029             :  *
    6030             :  *      We could dispense with the initialization files and just build the
    6031             :  *      critical reldescs the hard way on every backend startup, but that
    6032             :  *      slows down backend startup noticeably.
    6033             :  *
    6034             :  *      We can in fact go further, and save more relcache entries than
    6035             :  *      just the ones that are absolutely critical; this allows us to speed
    6036             :  *      up backend startup by not having to build such entries the hard way.
    6037             :  *      Presently, all the catalog and index entries that are referred to
    6038             :  *      by catcaches are stored in the initialization files.
    6039             :  *
    6040             :  *      The same mechanism that detects when catcache and relcache entries
    6041             :  *      need to be invalidated (due to catalog updates) also arranges to
    6042             :  *      unlink the initialization files when the contents may be out of date.
    6043             :  *      The files will then be rebuilt during the next backend startup.
    6044             :  */
    6045             : 
    6046             : /*
    6047             :  * load_relcache_init_file -- attempt to load cache from the shared
    6048             :  * or local cache init file
    6049             :  *
    6050             :  * If successful, return true and set criticalRelcachesBuilt or
    6051             :  * criticalSharedRelcachesBuilt to true.
    6052             :  * If not successful, return false.
    6053             :  *
    6054             :  * NOTE: we assume we are already switched into CacheMemoryContext.
    6055             :  */
    6056             : static bool
    6057       46306 : load_relcache_init_file(bool shared)
    6058             : {
    6059             :     FILE       *fp;
    6060             :     char        initfilename[MAXPGPATH];
    6061             :     Relation   *rels;
    6062             :     int         relno,
    6063             :                 num_rels,
    6064             :                 max_rels,
    6065             :                 nailed_rels,
    6066             :                 nailed_indexes,
    6067             :                 magic;
    6068             :     int         i;
    6069             : 
    6070       46306 :     if (shared)
    6071       24206 :         snprintf(initfilename, sizeof(initfilename), "global/%s",
    6072             :                  RELCACHE_INIT_FILENAME);
    6073             :     else
    6074       22100 :         snprintf(initfilename, sizeof(initfilename), "%s/%s",
    6075             :                  DatabasePath, RELCACHE_INIT_FILENAME);
    6076             : 
    6077       46306 :     fp = AllocateFile(initfilename, PG_BINARY_R);
    6078       46306 :     if (fp == NULL)
    6079        4788 :         return false;
    6080             : 
    6081             :     /*
    6082             :      * Read the index relcache entries from the file.  Note we will not enter
    6083             :      * any of them into the cache if the read fails partway through; this
    6084             :      * helps to guard against broken init files.
    6085             :      */
    6086       41518 :     max_rels = 100;
    6087       41518 :     rels = (Relation *) palloc(max_rels * sizeof(Relation));
    6088       41518 :     num_rels = 0;
    6089       41518 :     nailed_rels = nailed_indexes = 0;
    6090             : 
    6091             :     /* check for correct magic number (compatible version) */
    6092       41518 :     if (fread(&magic, 1, sizeof(magic), fp) != sizeof(magic))
    6093           0 :         goto read_failed;
    6094       41518 :     if (magic != RELCACHE_INIT_FILEMAGIC)
    6095           0 :         goto read_failed;
    6096             : 
    6097       41518 :     for (relno = 0;; relno++)
    6098     2733196 :     {
    6099             :         Size        len;
    6100             :         size_t      nread;
    6101             :         Relation    rel;
    6102             :         Form_pg_class relform;
    6103             :         bool        has_not_null;
    6104             : 
    6105             :         /* first read the relation descriptor length */
    6106     2774714 :         nread = fread(&len, 1, sizeof(len), fp);
    6107     2774714 :         if (nread != sizeof(len))
    6108             :         {
    6109       41518 :             if (nread == 0)
    6110       41518 :                 break;          /* end of file */
    6111           0 :             goto read_failed;
    6112             :         }
    6113             : 
    6114             :         /* safety check for incompatible relcache layout */
    6115     2733196 :         if (len != sizeof(RelationData))
    6116           0 :             goto read_failed;
    6117             : 
    6118             :         /* allocate another relcache header */
    6119     2733196 :         if (num_rels >= max_rels)
    6120             :         {
    6121       20220 :             max_rels *= 2;
    6122       20220 :             rels = (Relation *) repalloc(rels, max_rels * sizeof(Relation));
    6123             :         }
    6124             : 
    6125     2733196 :         rel = rels[num_rels++] = (Relation) palloc(len);
    6126             : 
    6127             :         /* then, read the Relation structure */
    6128     2733196 :         if (fread(rel, 1, len, fp) != len)
    6129           0 :             goto read_failed;
    6130             : 
    6131             :         /* next read the relation tuple form */
    6132     2733196 :         if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6133           0 :             goto read_failed;
    6134             : 
    6135     2733196 :         relform = (Form_pg_class) palloc(len);
    6136     2733196 :         if (fread(relform, 1, len, fp) != len)
    6137           0 :             goto read_failed;
    6138             : 
    6139     2733196 :         rel->rd_rel = relform;
    6140             : 
    6141             :         /* initialize attribute tuple forms */
    6142     2733196 :         rel->rd_att = CreateTemplateTupleDesc(relform->relnatts);
    6143     2733196 :         rel->rd_att->tdrefcount = 1;  /* mark as refcounted */
    6144             : 
    6145     2733196 :         rel->rd_att->tdtypeid = relform->reltype ? relform->reltype : RECORDOID;
    6146     2733196 :         rel->rd_att->tdtypmod = -1; /* just to be sure */
    6147             : 
    6148             :         /* next read all the attribute tuple form data entries */
    6149     2733196 :         has_not_null = false;
    6150    16032798 :         for (i = 0; i < relform->relnatts; i++)
    6151             :         {
    6152    13299602 :             Form_pg_attribute attr = TupleDescAttr(rel->rd_att, i);
    6153             : 
    6154    13299602 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6155           0 :                 goto read_failed;
    6156    13299602 :             if (len != ATTRIBUTE_FIXED_PART_SIZE)
    6157           0 :                 goto read_failed;
    6158    13299602 :             if (fread(attr, 1, len, fp) != len)
    6159           0 :                 goto read_failed;
    6160             : 
    6161    13299602 :             has_not_null |= attr->attnotnull;
    6162             :         }
    6163             : 
    6164             :         /* next read the access method specific field */
    6165     2733196 :         if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6166           0 :             goto read_failed;
    6167     2733196 :         if (len > 0)
    6168             :         {
    6169           0 :             rel->rd_options = palloc(len);
    6170           0 :             if (fread(rel->rd_options, 1, len, fp) != len)
    6171           0 :                 goto read_failed;
    6172           0 :             if (len != VARSIZE(rel->rd_options))
    6173           0 :                 goto read_failed;   /* sanity check */
    6174             :         }
    6175             :         else
    6176             :         {
    6177     2733196 :             rel->rd_options = NULL;
    6178             :         }
    6179             : 
    6180             :         /* mark not-null status */
    6181     2733196 :         if (has_not_null)
    6182             :         {
    6183     1019624 :             TupleConstr *constr = (TupleConstr *) palloc0(sizeof(TupleConstr));
    6184             : 
    6185     1019624 :             constr->has_not_null = true;
    6186     1019624 :             rel->rd_att->constr = constr;
    6187             :         }
    6188             : 
    6189             :         /*
    6190             :          * If it's an index, there's more to do.  Note we explicitly ignore
    6191             :          * partitioned indexes here.
    6192             :          */
    6193     2733196 :         if (rel->rd_rel->relkind == RELKIND_INDEX)
    6194             :         {
    6195             :             MemoryContext indexcxt;
    6196             :             Oid        *opfamily;
    6197             :             Oid        *opcintype;
    6198             :             RegProcedure *support;
    6199             :             int         nsupport;
    6200             :             int16      *indoption;
    6201             :             Oid        *indcollation;
    6202             : 
    6203             :             /* Count nailed indexes to ensure we have 'em all */
    6204     1713572 :             if (rel->rd_isnailed)
    6205      269328 :                 nailed_indexes++;
    6206             : 
    6207             :             /* read the pg_index tuple */
    6208     1713572 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6209           0 :                 goto read_failed;
    6210             : 
    6211     1713572 :             rel->rd_indextuple = (HeapTuple) palloc(len);
    6212     1713572 :             if (fread(rel->rd_indextuple, 1, len, fp) != len)
    6213           0 :                 goto read_failed;
    6214             : 
    6215             :             /* Fix up internal pointers in the tuple -- see heap_copytuple */
    6216     1713572 :             rel->rd_indextuple->t_data = (HeapTupleHeader) ((char *) rel->rd_indextuple + HEAPTUPLESIZE);
    6217     1713572 :             rel->rd_index = (Form_pg_index) GETSTRUCT(rel->rd_indextuple);
    6218             : 
    6219             :             /*
    6220             :              * prepare index info context --- parameters should match
    6221             :              * RelationInitIndexAccessInfo
    6222             :              */
    6223     1713572 :             indexcxt = AllocSetContextCreate(CacheMemoryContext,
    6224             :                                              "index info",
    6225             :                                              ALLOCSET_SMALL_SIZES);
    6226     1713572 :             rel->rd_indexcxt = indexcxt;
    6227     1713572 :             MemoryContextCopyAndSetIdentifier(indexcxt,
    6228             :                                               RelationGetRelationName(rel));
    6229             : 
    6230             :             /*
    6231             :              * Now we can fetch the index AM's API struct.  (We can't store
    6232             :              * that in the init file, since it contains function pointers that
    6233             :              * might vary across server executions.  Fortunately, it should be
    6234             :              * safe to call the amhandler even while bootstrapping indexes.)
    6235             :              */
    6236     1713572 :             InitIndexAmRoutine(rel);
    6237             : 
    6238             :             /* read the vector of opfamily OIDs */
    6239     1713572 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6240           0 :                 goto read_failed;
    6241             : 
    6242     1713572 :             opfamily = (Oid *) MemoryContextAlloc(indexcxt, len);
    6243     1713572 :             if (fread(opfamily, 1, len, fp) != len)
    6244           0 :                 goto read_failed;
    6245             : 
    6246     1713572 :             rel->rd_opfamily = opfamily;
    6247             : 
    6248             :             /* read the vector of opcintype OIDs */
    6249     1713572 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6250           0 :                 goto read_failed;
    6251             : 
    6252     1713572 :             opcintype = (Oid *) MemoryContextAlloc(indexcxt, len);
    6253     1713572 :             if (fread(opcintype, 1, len, fp) != len)
    6254           0 :                 goto read_failed;
    6255             : 
    6256     1713572 :             rel->rd_opcintype = opcintype;
    6257             : 
    6258             :             /* read the vector of support procedure OIDs */
    6259     1713572 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6260           0 :                 goto read_failed;
    6261     1713572 :             support = (RegProcedure *) MemoryContextAlloc(indexcxt, len);
    6262     1713572 :             if (fread(support, 1, len, fp) != len)
    6263           0 :                 goto read_failed;
    6264             : 
    6265     1713572 :             rel->rd_support = support;
    6266             : 
    6267             :             /* read the vector of collation OIDs */
    6268     1713572 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6269           0 :                 goto read_failed;
    6270             : 
    6271     1713572 :             indcollation = (Oid *) MemoryContextAlloc(indexcxt, len);
    6272     1713572 :             if (fread(indcollation, 1, len, fp) != len)
    6273           0 :                 goto read_failed;
    6274             : 
    6275     1713572 :             rel->rd_indcollation = indcollation;
    6276             : 
    6277             :             /* read the vector of indoption values */
    6278     1713572 :             if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6279           0 :                 goto read_failed;
    6280             : 
    6281     1713572 :             indoption = (int16 *) MemoryContextAlloc(indexcxt, len);
    6282     1713572 :             if (fread(indoption, 1, len, fp) != len)
    6283           0 :                 goto read_failed;
    6284             : 
    6285     1713572 :             rel->rd_indoption = indoption;
    6286             : 
    6287             :             /* read the vector of opcoptions values */
    6288     1713572 :             rel->rd_opcoptions = (bytea **)
    6289     1713572 :                 MemoryContextAllocZero(indexcxt, sizeof(*rel->rd_opcoptions) * relform->relnatts);
    6290             : 
    6291     4546790 :             for (i = 0; i < relform->relnatts; i++)
    6292             :             {
    6293     2833218 :                 if (fread(&len, 1, sizeof(len), fp) != sizeof(len))
    6294           0 :                     goto read_failed;
    6295             : 
    6296     2833218 :                 if (len > 0)
    6297             :                 {
    6298           0 :                     rel->rd_opcoptions[i] = (bytea *) MemoryContextAlloc(indexcxt, len);
    6299           0 :                     if (fread(rel->rd_opcoptions[i], 1, len, fp) != len)
    6300           0 :                         goto read_failed;
    6301             :                 }
    6302             :             }
    6303             : 
    6304             :             /* set up zeroed fmgr-info vector */
    6305     1713572 :             nsupport = relform->relnatts * rel->rd_indam->amsupport;
    6306     1713572 :             rel->rd_supportinfo = (FmgrInfo *)
    6307     1713572 :                 MemoryContextAllocZero(indexcxt, nsupport * sizeof(FmgrInfo));
    6308             :         }
    6309             :         else
    6310             :         {
    6311             :             /* Count nailed rels to ensure we have 'em all */
    6312     1019624 :             if (rel->rd_isnailed)
    6313      187370 :                 nailed_rels++;
    6314             : 
    6315             :             /* Load table AM data */
    6316     1019624 :             if (RELKIND_HAS_TABLE_AM(rel->rd_rel->relkind) || rel->rd_rel->relkind == RELKIND_SEQUENCE)
    6317     1019624 :                 RelationInitTableAccessMethod(rel);
    6318             : 
    6319             :             Assert(rel->rd_index == NULL);
    6320             :             Assert(rel->rd_indextuple == NULL);
    6321             :             Assert(rel->rd_indexcxt == NULL);
    6322             :             Assert(rel->rd_indam == NULL);
    6323             :             Assert(rel->rd_opfamily == NULL);
    6324             :             Assert(rel->rd_opcintype == NULL);
    6325             :             Assert(rel->rd_support == NULL);
    6326             :             Assert(rel->rd_supportinfo == NULL);
    6327             :             Assert(rel->rd_indoption == NULL);
    6328             :             Assert(rel->rd_indcollation == NULL);
    6329             :             Assert(rel->rd_opcoptions == NULL);
    6330             :         }
    6331             : 
    6332             :         /*
    6333             :          * Rules and triggers are not saved (mainly because the internal
    6334             :          * format is complex and subject to change).  They must be rebuilt if
    6335             :          * needed by RelationCacheInitializePhase3.  This is not expected to
    6336             :          * be a big performance hit since few system catalogs have such. Ditto
    6337             :          * for RLS policy data, partition info, index expressions, predicates,
    6338             :          * exclusion info, and FDW info.
    6339             :          */
    6340     2733196 :         rel->rd_rules = NULL;
    6341     2733196 :         rel->rd_rulescxt = NULL;
    6342     2733196 :         rel->trigdesc = NULL;
    6343     2733196 :         rel->rd_rsdesc = NULL;
    6344     2733196 :         rel->rd_partkey = NULL;
    6345     2733196 :         rel->rd_partkeycxt = NULL;
    6346     2733196 :         rel->rd_partdesc = NULL;
    6347     2733196 :         rel->rd_partdesc_nodetached = NULL;
    6348     2733196 :         rel->rd_partdesc_nodetached_xmin = InvalidTransactionId;
    6349     2733196 :         rel->rd_pdcxt = NULL;
    6350     2733196 :         rel->rd_pddcxt = NULL;
    6351     2733196 :         rel->rd_partcheck = NIL;
    6352     2733196 :         rel->rd_partcheckvalid = false;
    6353     2733196 :         rel->rd_partcheckcxt = NULL;
    6354     2733196 :         rel->rd_indexprs = NIL;
    6355     2733196 :         rel->rd_indpred = NIL;
    6356     2733196 :         rel->rd_exclops = NULL;
    6357     2733196 :         rel->rd_exclprocs = NULL;
    6358     2733196 :         rel->rd_exclstrats = NULL;
    6359     2733196 :         rel->rd_fdwroutine = NULL;
    6360             : 
    6361             :         /*
    6362             :          * Reset transient-state fields in the relcache entry
    6363             :          */
    6364     2733196 :         rel->rd_smgr = NULL;
    6365     2733196 :         if (rel->rd_isnailed)
    6366      456698 :             rel->rd_refcnt = 1;
    6367             :         else
    6368     2276498 :             rel->rd_refcnt = 0;
    6369     2733196 :         rel->rd_indexvalid = false;
    6370     2733196 :         rel->rd_indexlist = NIL;
    6371     2733196 :         rel->rd_pkindex = InvalidOid;
    6372     2733196 :         rel->rd_replidindex = InvalidOid;
    6373     2733196 :         rel->rd_attrsvalid = false;
    6374     2733196 :         rel->rd_keyattr = NULL;
    6375     2733196 :         rel->rd_pkattr = NULL;
    6376     2733196 :         rel->rd_idattr = NULL;
    6377     2733196 :         rel->rd_pubdesc = NULL;
    6378     2733196 :         rel->rd_statvalid = false;
    6379     2733196 :         rel->rd_statlist = NIL;
    6380     2733196 :         rel->rd_fkeyvalid = false;
    6381     2733196 :         rel->rd_fkeylist = NIL;
    6382     2733196 :         rel->rd_createSubid = InvalidSubTransactionId;
    6383     2733196 :         rel->rd_newRelfilelocatorSubid = InvalidSubTransactionId;
    6384     2733196 :         rel->rd_firstRelfilelocatorSubid = InvalidSubTransactionId;
    6385     2733196 :         rel->rd_droppedSubid = InvalidSubTransactionId;
    6386     2733196 :         rel->rd_amcache = NULL;
    6387     2733196 :         rel->pgstat_info = NULL;
    6388             : 
    6389             :         /*
    6390             :          * Recompute lock and physical addressing info.  This is needed in
    6391             :          * case the pg_internal.init file was copied from some other database
    6392             :          * by CREATE DATABASE.
    6393             :          */
    6394     2733196 :         RelationInitLockInfo(rel);
    6395     2733196 :         RelationInitPhysicalAddr(rel);
    6396             :     }
    6397             : 
    6398             :     /*
    6399             :      * We reached the end of the init file without apparent problem.  Did we
    6400             :      * get the right number of nailed items?  This is a useful crosscheck in
    6401             :      * case the set of critical rels or indexes changes.  However, that should
    6402             :      * not happen in a normally-running system, so let's bleat if it does.
    6403             :      *
    6404             :      * For the shared init file, we're called before client authentication is
    6405             :      * done, which means that elog(WARNING) will go only to the postmaster
    6406             :      * log, where it's easily missed.  To ensure that developers notice bad
    6407             :      * values of NUM_CRITICAL_SHARED_RELS/NUM_CRITICAL_SHARED_INDEXES, we put
    6408             :      * an Assert(false) there.
    6409             :      */
    6410       41518 :     if (shared)
    6411             :     {
    6412       21298 :         if (nailed_rels != NUM_CRITICAL_SHARED_RELS ||
    6413             :             nailed_indexes != NUM_CRITICAL_SHARED_INDEXES)
    6414             :         {
    6415           0 :             elog(WARNING, "found %d nailed shared rels and %d nailed shared indexes in init file, but expected %d and %d respectively",
    6416             :                  nailed_rels, nailed_indexes,
    6417             :                  NUM_CRITICAL_SHARED_RELS, NUM_CRITICAL_SHARED_INDEXES);
    6418             :             /* Make sure we get developers' attention about this */
    6419             :             Assert(false);
    6420             :             /* In production builds, recover by bootstrapping the relcache */
    6421           0 :             goto read_failed;
    6422             :         }
    6423             :     }
    6424             :     else
    6425             :     {
    6426       20220 :         if (nailed_rels != NUM_CRITICAL_LOCAL_RELS ||
    6427             :             nailed_indexes != NUM_CRITICAL_LOCAL_INDEXES)
    6428             :         {
    6429           0 :             elog(WARNING, "found %d nailed rels and %d nailed indexes in init file, but expected %d and %d respectively",
    6430             :                  nailed_rels, nailed_indexes,
    6431             :                  NUM_CRITICAL_LOCAL_RELS, NUM_CRITICAL_LOCAL_INDEXES);
    6432             :             /* We don't need an Assert() in this case */
    6433           0 :             goto read_failed;
    6434             :         }
    6435             :     }
    6436             : 
    6437             :     /*
    6438             :      * OK, all appears well.
    6439             :      *
    6440             :      * Now insert all the new relcache entries into the cache.
    6441             :      */
    6442     2774714 :     for (relno = 0; relno < num_rels; relno++)
    6443             :     {
    6444     2733196 :         RelationCacheInsert(rels[relno], false);
    6445             :     }
    6446             : 
    6447       41518 :     pfree(rels);
    6448       41518 :     FreeFile(fp);
    6449             : 
    6450       41518 :     if (shared)
    6451       21298 :         criticalSharedRelcachesBuilt = true;
    6452             :     else
    6453       20220 :         criticalRelcachesBuilt = true;
    6454       41518 :     return true;
    6455             : 
    6456             :     /*
    6457             :      * init file is broken, so do it the hard way.  We don't bother trying to
    6458             :      * free the clutter we just allocated; it's not in the relcache so it
    6459             :      * won't hurt.
    6460             :      */
    6461           0 : read_failed:
    6462           0 :     pfree(rels);
    6463           0 :     FreeFile(fp);
    6464             : 
    6465           0 :     return false;
    6466             : }
    6467             : 
    6468             : /*
    6469             :  * Write out a new initialization file with the current contents
    6470             :  * of the relcache (either shared rels or local rels, as indicated).
    6471             :  */
    6472             : static void
    6473        3824 : write_relcache_init_file(bool shared)
    6474             : {
    6475             :     FILE       *fp;
    6476             :     char        tempfilename[MAXPGPATH];
    6477             :     char        finalfilename[MAXPGPATH];
    6478             :     int         magic;
    6479             :     HASH_SEQ_STATUS status;
    6480             :     RelIdCacheEnt *idhentry;
    6481             :     int         i;
    6482             : 
    6483             :     /*
    6484             :      * If we have already received any relcache inval events, there's no
    6485             :      * chance of succeeding so we may as well skip the whole thing.
    6486             :      */
    6487        3824 :     if (relcacheInvalsReceived != 0L)
    6488          16 :         return;
    6489             : 
    6490             :     /*
    6491             :      * We must write a temporary file and rename it into place. Otherwise,
    6492             :      * another backend starting at about the same time might crash trying to
    6493             :      * read the partially-complete file.
    6494             :      */
    6495        3808 :     if (shared)
    6496             :     {
    6497        1904 :         snprintf(tempfilename, sizeof(tempfilename), "global/%s.%d",
    6498             :                  RELCACHE_INIT_FILENAME, MyProcPid);
    6499        1904 :         snprintf(finalfilename, sizeof(finalfilename), "global/%s",
    6500             :                  RELCACHE_INIT_FILENAME);
    6501             :     }
    6502             :     else
    6503             :     {
    6504        1904 :         snprintf(tempfilename, sizeof(tempfilename), "%s/%s.%d",
    6505             :                  DatabasePath, RELCACHE_INIT_FILENAME, MyProcPid);
    6506        1904 :         snprintf(finalfilename, sizeof(finalfilename), "%s/%s",
    6507             :                  DatabasePath, RELCACHE_INIT_FILENAME);
    6508             :     }
    6509             : 
    6510        3808 :     unlink(tempfilename);       /* in case it exists w/wrong permissions */
    6511             : 
    6512        3808 :     fp = AllocateFile(tempfilename, PG_BINARY_W);
    6513        3808 :     if (fp == NULL)
    6514             :     {
    6515             :         /*
    6516             :          * We used to consider this a fatal error, but we might as well
    6517             :          * continue with backend startup ...
    6518             :          */
    6519           0 :         ereport(WARNING,
    6520             :                 (errcode_for_file_access(),
    6521             :                  errmsg("could not create relation-cache initialization file \"%s\": %m",
    6522             :                         tempfilename),
    6523             :                  errdetail("Continuing anyway, but there's something wrong.")));
    6524           0 :         return;
    6525             :     }
    6526             : 
    6527             :     /*
    6528             :      * Write a magic number to serve as a file version identifier.  We can
    6529             :      * change the magic number whenever the relcache layout changes.
    6530             :      */
    6531        3808 :     magic = RELCACHE_INIT_FILEMAGIC;
    6532        3808 :     if (fwrite(&magic, 1, sizeof(magic), fp) != sizeof(magic))
    6533           0 :         elog(FATAL, "could not write init file");
    6534             : 
    6535             :     /*
    6536             :      * Write all the appropriate reldescs (in no particular order).
    6537             :      */
    6538        3808 :     hash_seq_init(&status, RelationIdCache);
    6539             : 
    6540      514080 :     while ((idhentry = (RelIdCacheEnt *) hash_seq_search(&status)) != NULL)
    6541             :     {
    6542      510272 :         Relation    rel = idhentry->reldesc;
    6543      510272 :         Form_pg_class relform = rel->rd_rel;
    6544             : 
    6545             :         /* ignore if not correct group */
    6546      510272 :         if (relform->relisshared != shared)
    6547      255136 :             continue;
    6548             : 
    6549             :         /*
    6550             :          * Ignore if not supposed to be in init file.  We can allow any shared
    6551             :          * relation that's been loaded so far to be in the shared init file,
    6552             :          * but unshared relations must be ones that should be in the local
    6553             :          * file per RelationIdIsInInitFile.  (Note: if you want to change the
    6554             :          * criterion for rels to be kept in the init file, see also inval.c.
    6555             :          * The reason for filtering here is to be sure that we don't put
    6556             :          * anything into the local init file for which a relcache inval would
    6557             :          * not cause invalidation of that init file.)
    6558             :          */
    6559      255136 :         if (!shared && !RelationIdIsInInitFile(RelationGetRelid(rel)))
    6560             :         {
    6561             :             /* Nailed rels had better get stored. */
    6562             :             Assert(!rel->rd_isnailed);
    6563           0 :             continue;
    6564             :         }
    6565             : 
    6566             :         /* first write the relcache entry proper */
    6567      255136 :         write_item(rel, sizeof(RelationData), fp);
    6568             : 
    6569             :         /* next write the relation tuple form */
    6570      255136 :         write_item(relform, CLASS_TUPLE_SIZE, fp);
    6571             : 
    6572             :         /* next, do all the attribute tuple form data entries */
    6573     1498448 :         for (i = 0; i < relform->relnatts; i++)
    6574             :         {
    6575     1243312 :             write_item(TupleDescAttr(rel->rd_att, i),
    6576             :                        ATTRIBUTE_FIXED_PART_SIZE, fp);
    6577             :         }
    6578             : 
    6579             :         /* next, do the access method specific field */
    6580      255136 :         write_item(rel->rd_options,
    6581      255136 :                    (rel->rd_options ? VARSIZE(rel->rd_options) : 0),
    6582             :                    fp);
    6583             : 
    6584             :         /*
    6585             :          * If it's an index, there's more to do. Note we explicitly ignore
    6586             :          * partitioned indexes here.
    6587             :          */
    6588      255136 :         if (rel->rd_rel->relkind == RELKIND_INDEX)
    6589             :         {
    6590             :             /* write the pg_index tuple */
    6591             :             /* we assume this was created by heap_copytuple! */
    6592      159936 :             write_item(rel->rd_indextuple,
    6593      159936 :                        HEAPTUPLESIZE + rel->rd_indextuple->t_len,
    6594             :                        fp);
    6595             : 
    6596             :             /* write the vector of opfamily OIDs */
    6597      159936 :             write_item(rel->rd_opfamily,
    6598      159936 :                        relform->relnatts * sizeof(Oid),
    6599             :                        fp);
    6600             : 
    6601             :             /* write the vector of opcintype OIDs */
    6602      159936 :             write_item(rel->rd_opcintype,
    6603      159936 :                        relform->relnatts * sizeof(Oid),
    6604             :                        fp);
    6605             : 
    6606             :             /* write the vector of support procedure OIDs */
    6607      159936 :             write_item(rel->rd_support,
    6608      159936 :                        relform->relnatts * (rel->rd_indam->amsupport * sizeof(RegProcedure)),
    6609             :                        fp);
    6610             : 
    6611             :             /* write the vector of collation OIDs */
    6612      159936 :             write_item(rel->rd_indcollation,
    6613      159936 :                        relform->relnatts * sizeof(Oid),
    6614             :                        fp);
    6615             : 
    6616             :             /* write the vector of indoption values */
    6617      159936 :             write_item(rel->rd_indoption,
    6618      159936 :                        relform->relnatts * sizeof(int16),
    6619             :                        fp);
    6620             : 
    6621             :             Assert(rel->rd_opcoptions);
    6622             : 
    6623             :             /* write the vector of opcoptions values */
    6624      424592 :             for (i = 0; i < relform->relnatts; i++)
    6625             :             {
    6626      264656 :                 bytea      *opt = rel->rd_opcoptions[i];
    6627             : 
    6628      264656 :                 write_item(opt, opt ? VARSIZE(opt) : 0, fp);
    6629             :             }
    6630             :         }
    6631             :     }
    6632             : 
    6633        3808 :     if (FreeFile(fp))
    6634           0 :         elog(FATAL, "could not write init file");
    6635             : 
    6636             :     /*
    6637             :      * Now we have to check whether the data we've so painstakingly
    6638             :      * accumulated is already obsolete due to someone else's just-committed
    6639             :      * catalog changes.  If so, we just delete the temp file and leave it to
    6640             :      * the next backend to try again.  (Our own relcache entries will be
    6641             :      * updated by SI message processing, but we can't be sure whether what we
    6642             :      * wrote out was up-to-date.)
    6643             :      *
    6644             :      * This mustn't run concurrently with the code that unlinks an init file
    6645             :      * and sends SI messages, so grab a serialization lock for the duration.
    6646             :      */
    6647        3808 :     LWLockAcquire(RelCacheInitLock, LW_EXCLUSIVE);
    6648             : 
    6649             :     /* Make sure we have seen all incoming SI messages */
    6650        3808 :     AcceptInvalidationMessages();
    6651             : 
    6652             :     /*
    6653             :      * If we have received any SI relcache invals since backend start, assume
    6654             :      * we may have written out-of-date data.
    6655             :      */
    6656        3808 :     if (relcacheInvalsReceived == 0L)
    6657             :     {
    6658             :         /*
    6659             :          * OK, rename the temp file to its final name, deleting any
    6660             :          * previously-existing init file.
    6661             :          *
    6662             :          * Note: a failure here is possible under Cygwin, if some other
    6663             :          * backend is holding open an unlinked-but-not-yet-gone init file. So
    6664             :          * treat this as a noncritical failure; just remove the useless temp
    6665             :          * file on failure.
    6666             :          */
    6667        3808 :         if (rename(tempfilename, finalfilename) < 0)
    6668           0 :             unlink(tempfilename);
    6669             :     }
    6670             :     else
    6671             :     {
    6672             :         /* Delete the already-obsolete temp file */
    6673           0 :         unlink(tempfilename);
    6674             :     }
    6675             : 
    6676        3808 :     LWLockRelease(RelCacheInitLock);
    6677             : }
    6678             : 
    6679             : /* write a chunk of data preceded by its length */
    6680             : static void
    6681     3232992 : write_item(const void *data, Size len, FILE *fp)
    6682             : {
    6683     3232992 :     if (fwrite(&len, 1, sizeof(len), fp) != sizeof(len))
    6684           0 :         elog(FATAL, "could not write init file");
    6685     3232992 :     if (len > 0 && fwrite(data, 1, len, fp) != len)
    6686           0 :         elog(FATAL, "could not write init file");
    6687     3232992 : }
    6688             : 
    6689             : /*
    6690             :  * Determine whether a given relation (identified by OID) is one of the ones
    6691             :  * we should store in a relcache init file.
    6692             :  *
    6693             :  * We must cache all nailed rels, and for efficiency we should cache every rel
    6694             :  * that supports a syscache.  The former set is almost but not quite a subset
    6695             :  * of the latter. The special cases are relations where
    6696             :  * RelationCacheInitializePhase2/3 chooses to nail for efficiency reasons, but
    6697             :  * which do not support any syscache.
    6698             :  */
    6699             : bool
    6700     1674100 : RelationIdIsInInitFile(Oid relationId)
    6701             : {
    6702     1674100 :     if (relationId == SharedSecLabelRelationId ||
    6703     1671786 :         relationId == TriggerRelidNameIndexId ||
    6704     1671724 :         relationId == DatabaseNameIndexId ||
    6705             :         relationId == SharedSecLabelObjectIndexId)
    6706             :     {
    6707             :         /*
    6708             :          * If this Assert fails, we don't need the applicable special case
    6709             :          * anymore.
    6710             :          */
    6711             :         Assert(!RelationSupportsSysCache(relationId));
    6712        2452 :         return true;
    6713             :     }
    6714     1671648 :     return RelationSupportsSysCache(relationId);
    6715             : }
    6716             : 
    6717             : /*
    6718             :  * Invalidate (remove) the init file during commit of a transaction that
    6719             :  * changed one or more of the relation cache entries that are kept in the
    6720             :  * local init file.
    6721             :  *
    6722             :  * To be safe against concurrent inspection or rewriting of the init file,
    6723             :  * we must take RelCacheInitLock, then remove the old init file, then send
    6724             :  * the SI messages that include relcache inval for such relations, and then
    6725             :  * release RelCacheInitLock.  This serializes the whole affair against
    6726             :  * write_relcache_init_file, so that we can be sure that any other process
    6727             :  * that's concurrently trying to create a new init file won't move an
    6728             :  * already-stale version into place after we unlink.  Also, because we unlink
    6729             :  * before sending the SI messages, a backend that's currently starting cannot
    6730             :  * read the now-obsolete init file and then miss the SI messages that will
    6731             :  * force it to update its relcache entries.  (This works because the backend
    6732             :  * startup sequence gets into the sinval array before trying to load the init
    6733             :  * file.)
    6734             :  *
    6735             :  * We take the lock and do the unlink in RelationCacheInitFilePreInvalidate,
    6736             :  * then release the lock in RelationCacheInitFilePostInvalidate.  Caller must
    6737             :  * send any pending SI messages between those calls.
    6738             :  */
    6739             : void
    6740       11984 : RelationCacheInitFilePreInvalidate(void)
    6741             : {
    6742             :     char        localinitfname[MAXPGPATH];
    6743             :     char        sharedinitfname[MAXPGPATH];
    6744             : 
    6745       11984 :     if (DatabasePath)
    6746       11984 :         snprintf(localinitfname, sizeof(localinitfname), "%s/%s",
    6747             :                  DatabasePath, RELCACHE_INIT_FILENAME);
    6748       11984 :     snprintf(sharedinitfname, sizeof(sharedinitfname), "global/%s",
    6749             :              RELCACHE_INIT_FILENAME);
    6750             : 
    6751       11984 :     LWLockAcquire(RelCacheInitLock, LW_EXCLUSIVE);
    6752             : 
    6753             :     /*
    6754             :      * The files might not be there if no backend has been started since the
    6755             :      * last removal.  But complain about failures other than ENOENT with
    6756             :      * ERROR.  Fortunately, it's not too late to abort the transaction if we
    6757             :      * can't get rid of the would-be-obsolete init file.
    6758             :      */
    6759       11984 :     if (DatabasePath)
    6760       11984 :         unlink_initfile(localinitfname, ERROR);
    6761       11984 :     unlink_initfile(sharedinitfname, ERROR);
    6762       11984 : }
    6763             : 
    6764             : void
    6765       11984 : RelationCacheInitFilePostInvalidate(void)
    6766             : {
    6767       11984 :     LWLockRelease(RelCacheInitLock);
    6768       11984 : }
    6769             : 
    6770             : /*
    6771             :  * Remove the init files during postmaster startup.
    6772             :  *
    6773             :  * We used to keep the init files across restarts, but that is unsafe in PITR
    6774             :  * scenarios, and even in simple crash-recovery cases there are windows for
    6775             :  * the init files to become out-of-sync with the database.  So now we just
    6776             :  * remove them during startup and expect the first backend launch to rebuild
    6777             :  * them.  Of course, this has to happen in each database of the cluster.
    6778             :  */
    6779             : void
    6780        1348 : RelationCacheInitFileRemove(void)
    6781             : {
    6782        1348 :     const char *tblspcdir = "pg_tblspc";
    6783             :     DIR        *dir;
    6784             :     struct dirent *de;
    6785             :     char        path[MAXPGPATH + 10 + sizeof(TABLESPACE_VERSION_DIRECTORY)];
    6786             : 
    6787        1348 :     snprintf(path, sizeof(path), "global/%s",
    6788             :              RELCACHE_INIT_FILENAME);
    6789        1348 :     unlink_initfile(path, LOG);
    6790             : 
    6791             :     /* Scan everything in the default tablespace */
    6792        1348 :     RelationCacheInitFileRemoveInDir("base");
    6793             : 
    6794             :     /* Scan the tablespace link directory to find non-default tablespaces */
    6795        1348 :     dir = AllocateDir(tblspcdir);
    6796             : 
    6797        4136 :     while ((de = ReadDirExtended(dir, tblspcdir, LOG)) != NULL)
    6798             :     {
    6799        2788 :         if (strspn(de->d_name, "0123456789") == strlen(de->d_name))
    6800             :         {
    6801             :             /* Scan the tablespace dir for per-database dirs */
    6802          92 :             snprintf(path, sizeof(path), "%s/%s/%s",
    6803          92 :                      tblspcdir, de->d_name, TABLESPACE_VERSION_DIRECTORY);
    6804          92 :             RelationCacheInitFileRemoveInDir(path);
    6805             :         }
    6806             :     }
    6807             : 
    6808        1348 :     FreeDir(dir);
    6809        1348 : }
    6810             : 
    6811             : /* Process one per-tablespace directory for RelationCacheInitFileRemove */
    6812             : static void
    6813        1440 : RelationCacheInitFileRemoveInDir(const char *tblspcpath)
    6814             : {
    6815             :     DIR        *dir;
    6816             :     struct dirent *de;
    6817             :     char        initfilename[MAXPGPATH * 2];
    6818             : 
    6819             :     /* Scan the tablespace directory to find per-database directories */
    6820        1440 :     dir = AllocateDir(tblspcpath);
    6821             : 
    6822        8700 :     while ((de = ReadDirExtended(dir, tblspcpath, LOG)) != NULL)
    6823             :     {
    6824        7260 :         if (strspn(de->d_name, "0123456789") == strlen(de->d_name))
    6825             :         {
    6826             :             /* Try to remove the init file in each database */
    6827        4286 :             snprintf(initfilename, sizeof(initfilename), "%s/%s/%s",
    6828        4286 :                      tblspcpath, de->d_name, RELCACHE_INIT_FILENAME);
    6829        4286 :             unlink_initfile(initfilename, LOG);
    6830             :         }
    6831             :     }
    6832             : 
    6833        1440 :     FreeDir(dir);
    6834        1440 : }
    6835             : 
    6836             : static void
    6837       29602 : unlink_initfile(const char *initfilename, int elevel)
    6838             : {
    6839       29602 :     if (unlink(initfilename) < 0)
    6840             :     {
    6841             :         /* It might not be there, but log any error other than ENOENT */
    6842       28100 :         if (errno != ENOENT)
    6843           0 :             ereport(elevel,
    6844             :                     (errcode_for_file_access(),
    6845             :                      errmsg("could not remove cache file \"%s\": %m",
    6846             :                             initfilename)));
    6847             :     }
    6848       29602 : }
    6849             : 
    6850             : /*
    6851             :  * ResourceOwner callbacks
    6852             :  */
    6853             : static char *
    6854           0 : ResOwnerPrintRelCache(Datum res)
    6855             : {
    6856           0 :     Relation    rel = (Relation) DatumGetPointer(res);
    6857             : 
    6858           0 :     return psprintf("relation \"%s\"", RelationGetRelationName(rel));
    6859             : }
    6860             : 
    6861             : static void
    6862       32330 : ResOwnerReleaseRelation(Datum res)
    6863             : {
    6864       32330 :     Relation    rel = (Relation) DatumGetPointer(res);
    6865             : 
    6866             :     /*
    6867             :      * This reference has already been removed from the resource owner, so
    6868             :      * just decrement reference count without calling
    6869             :      * ResourceOwnerForgetRelationRef.
    6870             :      */
    6871             :     Assert(rel->rd_refcnt > 0);
    6872       32330 :     rel->rd_refcnt -= 1;
    6873             : 
    6874       32330 :     RelationCloseCleanup((Relation) res);
    6875       32330 : }

Generated by: LCOV version 1.14