amcheck: Move common routines into a separate module
authorTomas Vondra <tomas.vondra@postgresql.org>
Sat, 29 Mar 2025 14:14:47 +0000 (15:14 +0100)
committerTomas Vondra <tomas.vondra@postgresql.org>
Sat, 29 Mar 2025 14:14:49 +0000 (15:14 +0100)
Before performing checks on an index, we need to take some safety
measures that apply to all index AMs. This includes:

* verifying that the index can be checked - Only selected AMs are
supported by amcheck (right now only B-Tree). The index has to be
valid and not a temporary index from another session.

* changing (and then restoring) user's security context

* obtaining proper locks on the index (and table, if needed)

* discarding GUC changes from the index functions

Until now this was implemented in the B-Tree amcheck module, but it's
something every AM will have to do. So relocate the code into a new
module verify_common for reuse.

The shared steps are implemented by amcheck_lock_relation_and_check(),
receiving the AM-specific verification as a callback. Custom parameters
may be supplied using a pointer.

Author: Andrey Borodin <amborodin@acm.org>
Reviewed-By: José Villanova <jose.arthur@gmail.com>
Reviewed-By: Aleksander Alekseev <aleksander@timescale.com>
Reviewed-By: Nikolay Samokhvalov <samokhvalov@gmail.com>
Reviewed-By: Andres Freund <andres@anarazel.de>
Reviewed-By: Tomas Vondra <tomas@vondra.me>
Reviewed-By: Mark Dilger <mark.dilger@enterprisedb.com>
Reviewed-By: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Kirill Reshke <reshkekirill@gmail.com>
Discussion: https://postgr.es/m/45AC9B0A-2B45-40EE-B08F-BDCF5739D1E1%40yandex-team.ru

contrib/amcheck/Makefile
contrib/amcheck/expected/check_btree.out
contrib/amcheck/meson.build
contrib/amcheck/verify_common.c [new file with mode: 0644]
contrib/amcheck/verify_common.h [new file with mode: 0644]
contrib/amcheck/verify_nbtree.c
src/tools/pgindent/typedefs.list

index 5e9002d250181005fde6a17dac72b6f4cb6d940c..c3d70f3369cb463a63cda9c32f75bdc1337df81e 100644 (file)
@@ -3,6 +3,7 @@
 MODULE_big = amcheck
 OBJS = \
    $(WIN32RES) \
+   verify_common.o \
    verify_heapam.o \
    verify_nbtree.o
 
index e7fb5f551574a344379a20e79dfc5b055c3a0deb..c6f4b16c55615b1fc1a3cdc0c94bca55e7c19505 100644 (file)
@@ -57,8 +57,8 @@ ERROR:  could not open relation with OID 17
 BEGIN;
 CREATE INDEX bttest_a_brin_idx ON bttest_a USING brin(id);
 SELECT bt_index_parent_check('bttest_a_brin_idx');
-ERROR:  only B-Tree indexes are supported as targets for verification
-DETAIL:  Relation "bttest_a_brin_idx" is not a B-Tree index.
+ERROR:  expected "btree" index as targets for verification
+DETAIL:  Relation "bttest_a_brin_idx" is a brin index.
 ROLLBACK;
 -- normal check outside of xact
 SELECT bt_index_check('bttest_a_idx');
index 61d7eaf2305d7e3359a83934367ebf867baf541f..67a4ac8518d156264a4475258e4797ffbc6b3aac 100644 (file)
@@ -1,6 +1,7 @@
 # Copyright (c) 2022-2025, PostgreSQL Global Development Group
 
 amcheck_sources = files(
+  'verify_common.c',
   'verify_heapam.c',
   'verify_nbtree.c',
 )
diff --git a/contrib/amcheck/verify_common.c b/contrib/amcheck/verify_common.c
new file mode 100644 (file)
index 0000000..d095e62
--- /dev/null
@@ -0,0 +1,191 @@
+/*-------------------------------------------------------------------------
+ *
+ * verify_common.c
+ *     Utility functions common to all access methods.
+ *
+ * Copyright (c) 2016-2025, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ *   contrib/amcheck/verify_common.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "access/genam.h"
+#include "access/table.h"
+#include "access/tableam.h"
+#include "verify_common.h"
+#include "catalog/index.h"
+#include "catalog/pg_am.h"
+#include "commands/tablecmds.h"
+#include "utils/guc.h"
+#include "utils/syscache.h"
+
+static bool amcheck_index_mainfork_expected(Relation rel);
+
+
+/*
+ * Check if index relation should have a file for its main relation fork.
+ * Verification uses this to skip unlogged indexes when in hot standby mode,
+ * where there is simply nothing to verify.
+ *
+ * NB: Caller should call index_checkable() before calling here.
+ */
+static bool
+amcheck_index_mainfork_expected(Relation rel)
+{
+   if (rel->rd_rel->relpersistence != RELPERSISTENCE_UNLOGGED ||
+       !RecoveryInProgress())
+       return true;
+
+   ereport(NOTICE,
+           (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
+            errmsg("cannot verify unlogged index \"%s\" during recovery, skipping",
+                   RelationGetRelationName(rel))));
+
+   return false;
+}
+
+/*
+* Amcheck main workhorse.
+* Given index relation OID, lock relation.
+* Next, take a number of standard actions:
+* 1) Make sure the index can be checked
+* 2) change the context of the user,
+* 3) keep track of GUCs modified via index functions
+* 4) execute callback function to verify integrity.
+*/
+void
+amcheck_lock_relation_and_check(Oid indrelid,
+                               Oid am_id,
+                               IndexDoCheckCallback check,
+                               LOCKMODE lockmode,
+                               void *state)
+{
+   Oid         heapid;
+   Relation    indrel;
+   Relation    heaprel;
+   Oid         save_userid;
+   int         save_sec_context;
+   int         save_nestlevel;
+
+   /*
+    * We must lock table before index to avoid deadlocks.  However, if the
+    * passed indrelid isn't an index then IndexGetRelation() will fail.
+    * Rather than emitting a not-very-helpful error message, postpone
+    * complaining, expecting that the is-it-an-index test below will fail.
+    *
+    * In hot standby mode this will raise an error when parentcheck is true.
+    */
+   heapid = IndexGetRelation(indrelid, true);
+   if (OidIsValid(heapid))
+   {
+       heaprel = table_open(heapid, lockmode);
+
+       /*
+        * Switch to the table owner's userid, so that any index functions are
+        * run as that user.  Also lock down security-restricted operations
+        * and arrange to make GUC variable changes local to this command.
+        */
+       GetUserIdAndSecContext(&save_userid, &save_sec_context);
+       SetUserIdAndSecContext(heaprel->rd_rel->relowner,
+                              save_sec_context | SECURITY_RESTRICTED_OPERATION);
+       save_nestlevel = NewGUCNestLevel();
+   }
+   else
+   {
+       heaprel = NULL;
+       /* Set these just to suppress "uninitialized variable" warnings */
+       save_userid = InvalidOid;
+       save_sec_context = -1;
+       save_nestlevel = -1;
+   }
+
+   /*
+    * Open the target index relations separately (like relation_openrv(), but
+    * with heap relation locked first to prevent deadlocking).  In hot
+    * standby mode this will raise an error when parentcheck is true.
+    *
+    * There is no need for the usual indcheckxmin usability horizon test
+    * here, even in the heapallindexed case, because index undergoing
+    * verification only needs to have entries for a new transaction snapshot.
+    * (If this is a parentcheck verification, there is no question about
+    * committed or recently dead heap tuples lacking index entries due to
+    * concurrent activity.)
+    */
+   indrel = index_open(indrelid, lockmode);
+
+   /*
+    * Since we did the IndexGetRelation call above without any lock, it's
+    * barely possible that a race against an index drop/recreation could have
+    * netted us the wrong table.
+    */
+   if (heaprel == NULL || heapid != IndexGetRelation(indrelid, false))
+       ereport(ERROR,
+               (errcode(ERRCODE_UNDEFINED_TABLE),
+                errmsg("could not open parent table of index \"%s\"",
+                       RelationGetRelationName(indrel))));
+
+   /* Check that relation suitable for checking */
+   if (index_checkable(indrel, am_id))
+       check(indrel, heaprel, state, lockmode == ShareLock);
+
+   /* Roll back any GUC changes executed by index functions */
+   AtEOXact_GUC(false, save_nestlevel);
+
+   /* Restore userid and security context */
+   SetUserIdAndSecContext(save_userid, save_sec_context);
+
+   /*
+    * Release locks early. That's ok here because nothing in the called
+    * routines will trigger shared cache invalidations to be sent, so we can
+    * relax the usual pattern of only releasing locks after commit.
+    */
+   index_close(indrel, lockmode);
+   if (heaprel)
+       table_close(heaprel, lockmode);
+}
+
+/*
+ * Basic checks about the suitability of a relation for checking as an index.
+ *
+ *
+ * NB: Intentionally not checking permissions, the function is normally not
+ * callable by non-superusers. If granted, it's useful to be able to check a
+ * whole cluster.
+ */
+bool
+index_checkable(Relation rel, Oid am_id)
+{
+   if (rel->rd_rel->relkind != RELKIND_INDEX ||
+       rel->rd_rel->relam != am_id)
+   {
+       HeapTuple   amtup;
+       HeapTuple   amtuprel;
+
+       amtup = SearchSysCache1(AMOID, ObjectIdGetDatum(am_id));
+       amtuprel = SearchSysCache1(AMOID, ObjectIdGetDatum(rel->rd_rel->relam));
+       ereport(ERROR,
+               (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+                errmsg("expected \"%s\" index as targets for verification", NameStr(((Form_pg_am) GETSTRUCT(amtup))->amname)),
+                errdetail("Relation \"%s\" is a %s index.",
+                          RelationGetRelationName(rel), NameStr(((Form_pg_am) GETSTRUCT(amtuprel))->amname))));
+   }
+
+   if (RELATION_IS_OTHER_TEMP(rel))
+       ereport(ERROR,
+               (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+                errmsg("cannot access temporary tables of other sessions"),
+                errdetail("Index \"%s\" is associated with temporary relation.",
+                          RelationGetRelationName(rel))));
+
+   if (!rel->rd_index->indisvalid)
+       ereport(ERROR,
+               (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+                errmsg("cannot check index \"%s\"",
+                       RelationGetRelationName(rel)),
+                errdetail("Index is not valid.")));
+
+   return amcheck_index_mainfork_expected(rel);
+}
diff --git a/contrib/amcheck/verify_common.h b/contrib/amcheck/verify_common.h
new file mode 100644 (file)
index 0000000..b2565bf
--- /dev/null
@@ -0,0 +1,31 @@
+/*-------------------------------------------------------------------------
+ *
+ * amcheck.h
+ *     Shared routines for amcheck verifications.
+ *
+ * Copyright (c) 2016-2025, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ *   contrib/amcheck/amcheck.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "storage/bufpage.h"
+#include "storage/lmgr.h"
+#include "storage/lockdefs.h"
+#include "utils/relcache.h"
+#include "miscadmin.h"
+
+/* Typedefs for callback functions for amcheck_lock_relation */
+typedef void (*IndexCheckableCallback) (Relation index);
+typedef void (*IndexDoCheckCallback) (Relation rel,
+                                     Relation heaprel,
+                                     void *state,
+                                     bool readonly);
+
+extern void amcheck_lock_relation_and_check(Oid indrelid,
+                                           Oid am_id,
+                                           IndexDoCheckCallback check,
+                                           LOCKMODE lockmode, void *state);
+
+extern bool index_checkable(Relation rel, Oid am_id);
index d56eb7637d39d5cae28ae6e2d9e5f219225230f6..f11c43a0ed7974085a3007a3ecc696b5b078354b 100644 (file)
@@ -30,6 +30,7 @@
 #include "access/tableam.h"
 #include "access/transam.h"
 #include "access/xact.h"
+#include "verify_common.h"
 #include "catalog/index.h"
 #include "catalog/pg_am.h"
 #include "catalog/pg_opfamily_d.h"
@@ -159,14 +160,22 @@ typedef struct BtreeLastVisibleEntry
    ItemPointer tid;            /* Heap tid */
 } BtreeLastVisibleEntry;
 
+/*
+ * arguments for the bt_index_check_callback callback
+ */
+typedef struct BTCallbackState
+{
+   bool        parentcheck;
+   bool        heapallindexed;
+   bool        rootdescend;
+   bool        checkunique;
+} BTCallbackState;
+
 PG_FUNCTION_INFO_V1(bt_index_check);
 PG_FUNCTION_INFO_V1(bt_index_parent_check);
 
-static void bt_index_check_internal(Oid indrelid, bool parentcheck,
-                                   bool heapallindexed, bool rootdescend,
-                                   bool checkunique);
-static inline void btree_index_checkable(Relation rel);
-static inline bool btree_index_mainfork_expected(Relation rel);
+static void bt_index_check_callback(Relation indrel, Relation heaprel,
+                                   void *state, bool readonly);
 static void bt_check_every_level(Relation rel, Relation heaprel,
                                 bool heapkeyspace, bool readonly, bool heapallindexed,
                                 bool rootdescend, bool checkunique);
@@ -241,15 +250,21 @@ Datum
 bt_index_check(PG_FUNCTION_ARGS)
 {
    Oid         indrelid = PG_GETARG_OID(0);
-   bool        heapallindexed = false;
-   bool        checkunique = false;
+   BTCallbackState args;
+
+   args.heapallindexed = false;
+   args.rootdescend = false;
+   args.parentcheck = false;
+   args.checkunique = false;
 
    if (PG_NARGS() >= 2)
-       heapallindexed = PG_GETARG_BOOL(1);
-   if (PG_NARGS() == 3)
-       checkunique = PG_GETARG_BOOL(2);
+       args.heapallindexed = PG_GETARG_BOOL(1);
+   if (PG_NARGS() >= 3)
+       args.checkunique = PG_GETARG_BOOL(2);
 
-   bt_index_check_internal(indrelid, false, heapallindexed, false, checkunique);
+   amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
+                                   bt_index_check_callback,
+                                   AccessShareLock, &args);
 
    PG_RETURN_VOID();
 }
@@ -267,18 +282,23 @@ Datum
 bt_index_parent_check(PG_FUNCTION_ARGS)
 {
    Oid         indrelid = PG_GETARG_OID(0);
-   bool        heapallindexed = false;
-   bool        rootdescend = false;
-   bool        checkunique = false;
+   BTCallbackState args;
+
+   args.heapallindexed = false;
+   args.rootdescend = false;
+   args.parentcheck = true;
+   args.checkunique = false;
 
    if (PG_NARGS() >= 2)
-       heapallindexed = PG_GETARG_BOOL(1);
+       args.heapallindexed = PG_GETARG_BOOL(1);
    if (PG_NARGS() >= 3)
-       rootdescend = PG_GETARG_BOOL(2);
-   if (PG_NARGS() == 4)
-       checkunique = PG_GETARG_BOOL(3);
+       args.rootdescend = PG_GETARG_BOOL(2);
+   if (PG_NARGS() >= 4)
+       args.checkunique = PG_GETARG_BOOL(3);
 
-   bt_index_check_internal(indrelid, true, heapallindexed, rootdescend, checkunique);
+   amcheck_lock_relation_and_check(indrelid, BTREE_AM_OID,
+                                   bt_index_check_callback,
+                                   ShareLock, &args);
 
    PG_RETURN_VOID();
 }
@@ -287,193 +307,46 @@ bt_index_parent_check(PG_FUNCTION_ARGS)
  * Helper for bt_index_[parent_]check, coordinating the bulk of the work.
  */
 static void
-bt_index_check_internal(Oid indrelid, bool parentcheck, bool heapallindexed,
-                       bool rootdescend, bool checkunique)
+bt_index_check_callback(Relation indrel, Relation heaprel, void *state, bool readonly)
 {
-   Oid         heapid;
-   Relation    indrel;
-   Relation    heaprel;
-   LOCKMODE    lockmode;
-   Oid         save_userid;
-   int         save_sec_context;
-   int         save_nestlevel;
-
-   if (parentcheck)
-       lockmode = ShareLock;
-   else
-       lockmode = AccessShareLock;
-
-   /*
-    * We must lock table before index to avoid deadlocks.  However, if the
-    * passed indrelid isn't an index then IndexGetRelation() will fail.
-    * Rather than emitting a not-very-helpful error message, postpone
-    * complaining, expecting that the is-it-an-index test below will fail.
-    *
-    * In hot standby mode this will raise an error when parentcheck is true.
-    */
-   heapid = IndexGetRelation(indrelid, true);
-   if (OidIsValid(heapid))
-   {
-       heaprel = table_open(heapid, lockmode);
-
-       /*
-        * Switch to the table owner's userid, so that any index functions are
-        * run as that user.  Also lock down security-restricted operations
-        * and arrange to make GUC variable changes local to this command.
-        */
-       GetUserIdAndSecContext(&save_userid, &save_sec_context);
-       SetUserIdAndSecContext(heaprel->rd_rel->relowner,
-                              save_sec_context | SECURITY_RESTRICTED_OPERATION);
-       save_nestlevel = NewGUCNestLevel();
-       RestrictSearchPath();
-   }
-   else
-   {
-       heaprel = NULL;
-       /* Set these just to suppress "uninitialized variable" warnings */
-       save_userid = InvalidOid;
-       save_sec_context = -1;
-       save_nestlevel = -1;
-   }
+   BTCallbackState *args = (BTCallbackState *) state;
+   bool        heapkeyspace,
+               allequalimage;
 
-   /*
-    * Open the target index relations separately (like relation_openrv(), but
-    * with heap relation locked first to prevent deadlocking).  In hot
-    * standby mode this will raise an error when parentcheck is true.
-    *
-    * There is no need for the usual indcheckxmin usability horizon test
-    * here, even in the heapallindexed case, because index undergoing
-    * verification only needs to have entries for a new transaction snapshot.
-    * (If this is a parentcheck verification, there is no question about
-    * committed or recently dead heap tuples lacking index entries due to
-    * concurrent activity.)
-    */
-   indrel = index_open(indrelid, lockmode);
-
-   /*
-    * Since we did the IndexGetRelation call above without any lock, it's
-    * barely possible that a race against an index drop/recreation could have
-    * netted us the wrong table.
-    */
-   if (heaprel == NULL || heapid != IndexGetRelation(indrelid, false))
+   if (!smgrexists(RelationGetSmgr(indrel), MAIN_FORKNUM))
        ereport(ERROR,
-               (errcode(ERRCODE_UNDEFINED_TABLE),
-                errmsg("could not open parent table of index \"%s\"",
+               (errcode(ERRCODE_INDEX_CORRUPTED),
+                errmsg("index \"%s\" lacks a main relation fork",
                        RelationGetRelationName(indrel))));
 
-   /* Relation suitable for checking as B-Tree? */
-   btree_index_checkable(indrel);
-
-   if (btree_index_mainfork_expected(indrel))
+   /* Extract metadata from metapage, and sanitize it in passing */
+   _bt_metaversion(indrel, &heapkeyspace, &allequalimage);
+   if (allequalimage && !heapkeyspace)
+       ereport(ERROR,
+               (errcode(ERRCODE_INDEX_CORRUPTED),
+                errmsg("index \"%s\" metapage has equalimage field set on unsupported nbtree version",
+                       RelationGetRelationName(indrel))));
+   if (allequalimage && !_bt_allequalimage(indrel, false))
    {
-       bool        heapkeyspace,
-                   allequalimage;
+       bool        has_interval_ops = false;
 
-       if (!smgrexists(RelationGetSmgr(indrel), MAIN_FORKNUM))
-           ereport(ERROR,
-                   (errcode(ERRCODE_INDEX_CORRUPTED),
-                    errmsg("index \"%s\" lacks a main relation fork",
-                           RelationGetRelationName(indrel))));
-
-       /* Extract metadata from metapage, and sanitize it in passing */
-       _bt_metaversion(indrel, &heapkeyspace, &allequalimage);
-       if (allequalimage && !heapkeyspace)
-           ereport(ERROR,
-                   (errcode(ERRCODE_INDEX_CORRUPTED),
-                    errmsg("index \"%s\" metapage has equalimage field set on unsupported nbtree version",
-                           RelationGetRelationName(indrel))));
-       if (allequalimage && !_bt_allequalimage(indrel, false))
-       {
-           bool        has_interval_ops = false;
-
-           for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(indrel); i++)
-               if (indrel->rd_opfamily[i] == INTERVAL_BTREE_FAM_OID)
-                   has_interval_ops = true;
-           ereport(ERROR,
-                   (errcode(ERRCODE_INDEX_CORRUPTED),
-                    errmsg("index \"%s\" metapage incorrectly indicates that deduplication is safe",
-                           RelationGetRelationName(indrel)),
-                    has_interval_ops
-                    ? errhint("This is known of \"interval\" indexes last built on a version predating 2023-11.")
-                    : 0));
-       }
-
-       /* Check index, possibly against table it is an index on */
-       bt_check_every_level(indrel, heaprel, heapkeyspace, parentcheck,
-                            heapallindexed, rootdescend, checkunique);
+       for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(indrel); i++)
+           if (indrel->rd_opfamily[i] == INTERVAL_BTREE_FAM_OID)
+           {
+               has_interval_ops = true;
+               ereport(ERROR,
+                       (errcode(ERRCODE_INDEX_CORRUPTED),
+                        errmsg("index \"%s\" metapage incorrectly indicates that deduplication is safe",
+                               RelationGetRelationName(indrel)),
+                        has_interval_ops
+                        ? errhint("This is known of \"interval\" indexes last built on a version predating 2023-11.")
+                        : 0));
+           }
    }
 
-   /* Roll back any GUC changes executed by index functions */
-   AtEOXact_GUC(false, save_nestlevel);
-
-   /* Restore userid and security context */
-   SetUserIdAndSecContext(save_userid, save_sec_context);
-
-   /*
-    * Release locks early. That's ok here because nothing in the called
-    * routines will trigger shared cache invalidations to be sent, so we can
-    * relax the usual pattern of only releasing locks after commit.
-    */
-   index_close(indrel, lockmode);
-   if (heaprel)
-       table_close(heaprel, lockmode);
-}
-
-/*
- * Basic checks about the suitability of a relation for checking as a B-Tree
- * index.
- *
- * NB: Intentionally not checking permissions, the function is normally not
- * callable by non-superusers. If granted, it's useful to be able to check a
- * whole cluster.
- */
-static inline void
-btree_index_checkable(Relation rel)
-{
-   if (rel->rd_rel->relkind != RELKIND_INDEX ||
-       rel->rd_rel->relam != BTREE_AM_OID)
-       ereport(ERROR,
-               (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-                errmsg("only B-Tree indexes are supported as targets for verification"),
-                errdetail("Relation \"%s\" is not a B-Tree index.",
-                          RelationGetRelationName(rel))));
-
-   if (RELATION_IS_OTHER_TEMP(rel))
-       ereport(ERROR,
-               (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-                errmsg("cannot access temporary tables of other sessions"),
-                errdetail("Index \"%s\" is associated with temporary relation.",
-                          RelationGetRelationName(rel))));
-
-   if (!rel->rd_index->indisvalid)
-       ereport(ERROR,
-               (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-                errmsg("cannot check index \"%s\"",
-                       RelationGetRelationName(rel)),
-                errdetail("Index is not valid.")));
-}
-
-/*
- * Check if B-Tree index relation should have a file for its main relation
- * fork.  Verification uses this to skip unlogged indexes when in hot standby
- * mode, where there is simply nothing to verify.  We behave as if the
- * relation is empty.
- *
- * NB: Caller should call btree_index_checkable() before calling here.
- */
-static inline bool
-btree_index_mainfork_expected(Relation rel)
-{
-   if (rel->rd_rel->relpersistence != RELPERSISTENCE_UNLOGGED ||
-       !RecoveryInProgress())
-       return true;
-
-   ereport(DEBUG1,
-           (errcode(ERRCODE_READ_ONLY_SQL_TRANSACTION),
-            errmsg("cannot verify unlogged index \"%s\" during recovery, skipping",
-                   RelationGetRelationName(rel))));
-
-   return false;
+   /* Check index, possibly against table it is an index on */
+   bt_check_every_level(indrel, heaprel, heapkeyspace, readonly,
+                        args->heapallindexed, args->rootdescend, args->checkunique);
 }
 
 /*
index 1279b69422a59e402b160777928915ed5e1fb181..b66affbea56eb8056502cc9185abb641890d0662 100644 (file)
@@ -194,6 +194,7 @@ BOOLEAN
 BOX
 BTArrayKeyInfo
 BTBuildState
+BTCallbackState
 BTCycleId
 BTDedupInterval
 BTDedupState