From 27e4fad9804c0dabe15b00168ad9b65696bedff5 Mon Sep 17 00:00:00 2001 From: Noah Misch Date: Tue, 16 Dec 2025 16:13:54 -0800 Subject: [PATCH] Assert lack of hazardous buffer locks before possible catalog read. Commit 0bada39c83a150079567a6e97b1a25a198f30ea3 fixed a bug of this kind, which existed in all branches for six days before detection. While the probability of reaching the trouble was low, the disruption was extreme. No new backends could start, and service restoration needed an immediate shutdown. Hence, add this to catch the next bug like it. The new check in RelationIdGetRelation() suffices to make autovacuum detect the bug in commit 243e9b40f1b2dd09d6e5bf91ebf6e822a2cd3704 that led to commit 0bada39. This also checks in a number of similar places. It replaces each Assert(IsTransactionState()) that pertained to a conditional catalog read. Back-patch to v14 - v17. This a back-patch of commit f4ece891fc2f3f96f0571720a1ae30db8030681b (from before v18 branched) to all supported branches, to accompany the back-patch of commits 243e9b4 and 0bada39. For catalog indexes, the bttextcmp() behavior that motivated IsCatalogTextUniqueIndexOid() was v18-specific. Hence, this back-patch doesn't need that or its correction from commit 4a4ee0c2c1e53401924101945ac3d517c0a8a559. Reported-by: Alexander Lakhin Discussion: https://postgr.es/m/20250410191830.0e.nmisch@google.com Discussion: https://postgr.es/m/10ec0bc3-5933-1189-6bb8-5dec4114558e@gmail.com Backpatch-through: 14-17 --- src/backend/storage/buffer/bufmgr.c | 67 +++++++++++++++++++++++++++++ src/backend/storage/lmgr/lwlock.c | 15 +++++++ src/backend/utils/adt/pg_locale.c | 3 ++ src/backend/utils/cache/catcache.c | 51 +++++++++++++++------- src/backend/utils/cache/inval.c | 14 +++++- src/backend/utils/cache/relcache.c | 20 ++++++++- src/backend/utils/mb/mbutils.c | 3 +- src/include/storage/bufmgr.h | 3 ++ src/include/storage/lwlock.h | 2 + src/include/utils/relcache.h | 8 ++++ 10 files changed, 166 insertions(+), 20 deletions(-) diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c index e007bd46e93..ef14791e2da 100644 --- a/src/backend/storage/buffer/bufmgr.c +++ b/src/backend/storage/buffer/bufmgr.c @@ -37,6 +37,9 @@ #include "access/xloginsert.h" #include "access/xlogutils.h" #include "catalog/catalog.h" +#ifdef USE_ASSERT_CHECKING +#include "catalog/pg_tablespace_d.h" +#endif #include "catalog/storage.h" #include "catalog/storage_xlog.h" #include "executor/instrument.h" @@ -498,6 +501,10 @@ static void RelationCopyStorageUsingBuffer(RelFileLocator srclocator, ForkNumber forkNum, bool permanent); static void AtProcExit_Buffers(int code, Datum arg); static void CheckForBufferLeaks(void); +#ifdef USE_ASSERT_CHECKING +static void AssertNotCatalogBufferLock(LWLock *lock, LWLockMode mode, + void *unused_context); +#endif static int rlocator_comparator(const void *p1, const void *p2); static inline int buffertag_comparator(const BufferTag *ba, const BufferTag *bb); static inline int ckpt_buforder_comparator(const CkptSortItem *a, const CkptSortItem *b); @@ -3225,6 +3232,66 @@ CheckForBufferLeaks(void) #endif } +#ifdef USE_ASSERT_CHECKING +/* + * Check for exclusive-locked catalog buffers. This is the core of + * AssertCouldGetRelation(). + * + * A backend would self-deadlock on LWLocks if the catalog scan read the + * exclusive-locked buffer. The main threat is exclusive-locked buffers of + * catalogs used in relcache, because a catcache search on any catalog may + * build that catalog's relcache entry. We don't have an inventory of + * catalogs relcache uses, so just check buffers of most catalogs. + * + * It's better to minimize waits while holding an exclusive buffer lock, so it + * would be nice to broaden this check not to be catalog-specific. However, + * bttextcmp() accesses pg_collation, and non-core opclasses might similarly + * read tables. That is deadlock-free as long as there's no loop in the + * dependency graph: modifying table A may cause an opclass to read table B, + * but it must not cause a read of table A. + */ +void +AssertBufferLocksPermitCatalogRead(void) +{ + ForEachLWLockHeldByMe(AssertNotCatalogBufferLock, NULL); +} + +static void +AssertNotCatalogBufferLock(LWLock *lock, LWLockMode mode, + void *unused_context) +{ + BufferDesc *bufHdr; + BufferTag tag; + Oid relid; + + if (mode != LW_EXCLUSIVE) + return; + + if (!((BufferDescPadded *) lock > BufferDescriptors && + (BufferDescPadded *) lock < BufferDescriptors + NBuffers)) + return; /* not a buffer lock */ + + bufHdr = (BufferDesc *) + ((char *) lock - offsetof(BufferDesc, content_lock)); + tag = bufHdr->tag; + + /* + * This relNumber==relid assumption holds until a catalog experiences + * VACUUM FULL or similar. After a command like that, relNumber will be + * in the normal (non-catalog) range, and we lose the ability to detect + * hazardous access to that catalog. Calling RelidByRelfilenumber() would + * close that gap, but RelidByRelfilenumber() might then deadlock with a + * held lock. + */ + relid = tag.relNumber; + + Assert(!IsCatalogRelationOid(relid)); + /* Shared rels are always catalogs: detect even after VACUUM FULL. */ + Assert(tag.spcOid != GLOBALTABLESPACE_OID); +} +#endif + + /* * Helper routine to issue warnings when a buffer is unexpectedly pinned */ diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c index 15c2f8a2ca9..3fae2a157ab 100644 --- a/src/backend/storage/lmgr/lwlock.c +++ b/src/backend/storage/lmgr/lwlock.c @@ -1910,6 +1910,21 @@ LWLockReleaseAll(void) } +/* + * ForEachLWLockHeldByMe - run a callback for each held lock + * + * This is meant as debug support only. + */ +void +ForEachLWLockHeldByMe(void (*callback) (LWLock *, LWLockMode, void *), + void *context) +{ + int i; + + for (i = 0; i < num_held_lwlocks; i++) + callback(held_lwlocks[i].lock, held_lwlocks[i].mode, context); +} + /* * LWLockHeldByMe - test whether my process holds a lock in any mode * diff --git a/src/backend/utils/adt/pg_locale.c b/src/backend/utils/adt/pg_locale.c index 111a7888779..df0c84cf371 100644 --- a/src/backend/utils/adt/pg_locale.c +++ b/src/backend/utils/adt/pg_locale.c @@ -67,6 +67,7 @@ #include "utils/lsyscache.h" #include "utils/memutils.h" #include "utils/pg_locale.h" +#include "utils/relcache.h" #include "utils/syscache.h" #ifdef USE_ICU @@ -1222,6 +1223,8 @@ lookup_collation_cache(Oid collation, bool set_flags) Assert(OidIsValid(collation)); Assert(collation != DEFAULT_COLLATION_OID); + AssertCouldGetRelation(); + if (collation_cache == NULL) { /* First time through, initialize the hash table */ diff --git a/src/backend/utils/cache/catcache.c b/src/backend/utils/cache/catcache.c index 6550f966df7..5690a36755c 100644 --- a/src/backend/utils/cache/catcache.c +++ b/src/backend/utils/cache/catcache.c @@ -999,12 +999,41 @@ RehashCatCacheLists(CatCache *cp) cp->cc_lbucket = newbucket; } +/* + * ConditionalCatalogCacheInitializeCache + * + * Call CatalogCacheInitializeCache() if not yet done. + */ +pg_attribute_always_inline +static void +ConditionalCatalogCacheInitializeCache(CatCache *cache) +{ +#ifdef USE_ASSERT_CHECKING + /* + * TypeCacheRelCallback() runs outside transactions and relies on TYPEOID + * for hashing. This isn't ideal. Since lookup_type_cache() both + * registers the callback and searches TYPEOID, reaching trouble likely + * requires OOM at an unlucky moment. + * + * InvalidateAttoptCacheCallback() runs outside transactions and likewise + * relies on ATTNUM. InitPostgres() initializes ATTNUM, so it's reliable. + */ + if (!(cache->id == TYPEOID || cache->id == ATTNUM) || + IsTransactionState()) + AssertCouldGetRelation(); + else + Assert(cache->cc_tupdesc != NULL); +#endif + + if (unlikely(cache->cc_tupdesc == NULL)) + CatalogCacheInitializeCache(cache); +} + /* * CatalogCacheInitializeCache * * This function does final initialization of a catcache: obtain the tuple - * descriptor and set up the hash and equality function links. We assume - * that the relcache entry can be opened at this point! + * descriptor and set up the hash and equality function links. */ #ifdef CACHEDEBUG #define CatalogCacheInitializeCache_DEBUG1 \ @@ -1139,8 +1168,7 @@ CatalogCacheInitializeCache(CatCache *cache) void InitCatCachePhase2(CatCache *cache, bool touch_index) { - if (cache->cc_tupdesc == NULL) - CatalogCacheInitializeCache(cache); + ConditionalCatalogCacheInitializeCache(cache); if (touch_index && cache->id != AMOID && @@ -1319,16 +1347,12 @@ SearchCatCacheInternal(CatCache *cache, dlist_head *bucket; CatCTup *ct; - /* Make sure we're in an xact, even if this ends up being a cache hit */ - Assert(IsTransactionState()); - Assert(cache->cc_nkeys == nkeys); /* * one-time startup overhead for each cache */ - if (unlikely(cache->cc_tupdesc == NULL)) - CatalogCacheInitializeCache(cache); + ConditionalCatalogCacheInitializeCache(cache); #ifdef CATCACHE_STATS cache->cc_searches++; @@ -1607,8 +1631,7 @@ GetCatCacheHashValue(CatCache *cache, /* * one-time startup overhead for each cache */ - if (cache->cc_tupdesc == NULL) - CatalogCacheInitializeCache(cache); + ConditionalCatalogCacheInitializeCache(cache); /* * calculate the hash value @@ -1659,8 +1682,7 @@ SearchCatCacheList(CatCache *cache, /* * one-time startup overhead for each cache */ - if (unlikely(cache->cc_tupdesc == NULL)) - CatalogCacheInitializeCache(cache); + ConditionalCatalogCacheInitializeCache(cache); Assert(nkeys > 0 && nkeys < cache->cc_nkeys); @@ -2314,8 +2336,7 @@ PrepareToInvalidateCacheTuple(Relation relation, continue; /* Just in case cache hasn't finished initialization yet... */ - if (ccp->cc_tupdesc == NULL) - CatalogCacheInitializeCache(ccp); + ConditionalCatalogCacheInitializeCache(ccp); hashvalue = CatalogCacheComputeTupleHashValue(ccp, ccp->cc_nkeys, tuple); dbid = ccp->cc_relisshared ? (Oid) 0 : MyDatabaseId; diff --git a/src/backend/utils/cache/inval.c b/src/backend/utils/cache/inval.c index 4d3dd6cd15a..140848c2d73 100644 --- a/src/backend/utils/cache/inval.c +++ b/src/backend/utils/cache/inval.c @@ -631,7 +631,8 @@ PrepareInvalidationState(void) { TransInvalidationInfo *myInfo; - Assert(IsTransactionState()); + /* PrepareToInvalidateCacheTuple() needs relcache */ + AssertCouldGetRelation(); /* Can't queue transactional message while collecting inplace messages. */ Assert(inplaceInvalInfo == NULL); @@ -700,7 +701,7 @@ PrepareInplaceInvalidationState(void) { InvalidationInfo *myInfo; - Assert(IsTransactionState()); + AssertCouldGetRelation(); /* limit of one inplace update under assembly */ Assert(inplaceInvalInfo == NULL); @@ -863,6 +864,12 @@ InvalidateSystemCaches(void) void AcceptInvalidationMessages(void) { +#ifdef USE_ASSERT_CHECKING + /* message handlers shall access catalogs only during transactions */ + if (IsTransactionState()) + AssertCouldGetRelation(); +#endif + ReceiveSharedInvalidMessages(LocalExecuteInvalidationMessage, InvalidateSystemCaches); @@ -1326,6 +1333,9 @@ CacheInvalidateHeapTupleCommon(Relation relation, Oid databaseId; Oid relationId; + /* PrepareToInvalidateCacheTuple() needs relcache */ + AssertCouldGetRelation(); + /* Do nothing during bootstrap */ if (IsBootstrapProcessingMode()) return; diff --git a/src/backend/utils/cache/relcache.c b/src/backend/utils/cache/relcache.c index 34c3cfa8e8c..3140f00ea44 100644 --- a/src/backend/utils/cache/relcache.c +++ b/src/backend/utils/cache/relcache.c @@ -2028,6 +2028,23 @@ formrdesc(const char *relationName, Oid relationReltype, relation->rd_isvalid = true; } +#ifdef USE_ASSERT_CHECKING +/* + * AssertCouldGetRelation + * + * Check safety of calling RelationIdGetRelation(). + * + * In code that reads catalogs in the event of a cache miss, call this + * before checking the cache. + */ +void +AssertCouldGetRelation(void) +{ + Assert(IsTransactionState()); + AssertBufferLocksPermitCatalogRead(); +} +#endif + /* ---------------------------------------------------------------- * Relation Descriptor Lookup Interface @@ -2055,8 +2072,7 @@ RelationIdGetRelation(Oid relationId) { Relation rd; - /* Make sure we're in an xact, even if this ends up being a cache hit */ - Assert(IsTransactionState()); + AssertCouldGetRelation(); /* * first try to find reldesc in the cache diff --git a/src/backend/utils/mb/mbutils.c b/src/backend/utils/mb/mbutils.c index 1ccdf9da5b0..33ad7ae7774 100644 --- a/src/backend/utils/mb/mbutils.c +++ b/src/backend/utils/mb/mbutils.c @@ -39,6 +39,7 @@ #include "mb/pg_wchar.h" #include "utils/builtins.h" #include "utils/memutils.h" +#include "utils/relcache.h" #include "utils/syscache.h" #include "varatt.h" @@ -311,7 +312,7 @@ InitializeClientEncoding(void) { Oid utf8_to_server_proc; - Assert(IsTransactionState()); + AssertCouldGetRelation(); utf8_to_server_proc = FindDefaultConversionProc(PG_UTF8, current_server_encoding); diff --git a/src/include/storage/bufmgr.h b/src/include/storage/bufmgr.h index b379c76e273..166524c50e0 100644 --- a/src/include/storage/bufmgr.h +++ b/src/include/storage/bufmgr.h @@ -205,6 +205,9 @@ extern Buffer ExtendBufferedRelTo(BufferManagerRelation bmr, extern void InitBufferPoolAccess(void); extern void AtEOXact_Buffers(bool isCommit); +#ifdef USE_ASSERT_CHECKING +extern void AssertBufferLocksPermitCatalogRead(void); +#endif extern void PrintBufferLeakWarning(Buffer buffer); extern void CheckPointBuffers(int flags); extern BlockNumber BufferGetBlockNumber(Buffer buffer); diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h index 34169e5889e..cab38447d48 100644 --- a/src/include/storage/lwlock.h +++ b/src/include/storage/lwlock.h @@ -131,6 +131,8 @@ extern bool LWLockAcquireOrWait(LWLock *lock, LWLockMode mode); extern void LWLockRelease(LWLock *lock); extern void LWLockReleaseClearVar(LWLock *lock, uint64 *valptr, uint64 val); extern void LWLockReleaseAll(void); +extern void ForEachLWLockHeldByMe(void (*callback) (LWLock *, LWLockMode, void *), + void *context); extern bool LWLockHeldByMe(LWLock *lock); extern bool LWLockAnyHeldByMe(LWLock *lock, int nlocks, size_t stride); extern bool LWLockHeldByMeInMode(LWLock *lock, LWLockMode mode); diff --git a/src/include/utils/relcache.h b/src/include/utils/relcache.h index 38524641f47..10a91c986d1 100644 --- a/src/include/utils/relcache.h +++ b/src/include/utils/relcache.h @@ -37,6 +37,14 @@ typedef Relation *RelationPtr; /* * Routines to open (lookup) and close a relcache entry */ +#ifdef USE_ASSERT_CHECKING +extern void AssertCouldGetRelation(void); +#else +static inline void +AssertCouldGetRelation(void) +{ +} +#endif extern Relation RelationIdGetRelation(Oid relationId); extern void RelationClose(Relation relation); -- 2.39.5