Prevent concurrent SimpleLruTruncate() for any given SLRU.

The SimpleLruTruncate() header comment states the new coding rule.  To
achieve this, add locktype "frozenid" and two LWLocks.  This closes a
rare opportunity for data loss, which manifested as "apparent
wraparound" or "could not access status of transaction" errors.  Data
loss is more likely in pg_multixact, due to released branches' thin
margin between multiStopLimit and multiWrapLimit.  If a user's physical
replication primary logged ":  apparent wraparound" messages, the user
should rebuild standbys of that primary regardless of symptoms.  At less
risk is a cluster having emitted "not accepting commands" errors or
"must be vacuumed" warnings at some point.  One can test a cluster for
this data loss by running VACUUM FREEZE in every database.  Back-patch
to 9.5 (all supported versions).

Discussion: https://postgr.es/m/[email protected]

11 files changed, 117 insertions, 13 deletions

diff --git a/doc/src/sgml/catalogs.sgml b/doc/src/sgml/catalogs.sgml
index 0709b35c04..1466606c34 100644
--- a/doc/src/sgml/catalogs.sgml
+++ b/doc/src/sgml/catalogs.sgml
@@ -10215,7 +10215,8 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</replaceable>:<replaceable>&l
    and general database objects (identified by class OID and object OID,
    in the same way as in <structname>pg_description</structname> or
    <structname>pg_depend</structname>).  Also, the right to extend a
-   relation is represented as a separate lockable object.
+   relation is represented as a separate lockable object, as is the right to
+   update <structname>pg_database</structname>.<structfield>datfrozenxid</structfield>.
    Also, <quote>advisory</quote> locks can be taken on numbers that have
    user-defined meanings.
   </para>
@@ -10243,6 +10244,7 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</replaceable>:<replaceable>&l
        Type of the lockable object:
        <literal>relation</literal>,
        <literal>extend</literal>,
+       <literal>frozenid</literal>,
        <literal>page</literal>,
        <literal>tuple</literal>,
        <literal>transactionid</literal>,
diff --git a/doc/src/sgml/monitoring.sgml b/doc/src/sgml/monitoring.sgml
index 408ecbcae0..5243f5f346 100644
--- a/doc/src/sgml/monitoring.sgml
+++ b/doc/src/sgml/monitoring.sgml
@@ -1739,6 +1739,12 @@ postgres   27093  0.0  0.0  30096  2752 ?        Ss   11:34   0:00 postgres: ser
       <entry>Waiting to extend a relation.</entry>
      </row>
      <row>
+      <entry><literal>frozenid</literal></entry>
+      <entry>Waiting to
+       update <structname>pg_database</structname>.<structfield>datfrozenxid</structfield>
+       and <structname>pg_database</structname>.<structfield>datminmxid</structfield>.</entry>
+     </row>
+     <row>
       <entry><literal>object</literal></entry>
       <entry>Waiting to acquire a lock on a non-relation database object.</entry>
      </row>
@@ -1907,6 +1913,11 @@ postgres   27093  0.0  0.0  30096  2752 ?        Ss   11:34   0:00 postgres: ser
       <entry>Waiting to read or update <command>NOTIFY</command> messages.</entry>
      </row>
      <row>
+      <entry><literal>NotifyQueueTail</literal></entry>
+      <entry>Waiting to update limit on <command>NOTIFY</command> message
+       storage.</entry>
+     </row>
+     <row>
       <entry><literal>NotifySLRU</literal></entry>
       <entry>Waiting to access the <command>NOTIFY</command> message SLRU
        cache.</entry>
@@ -2083,6 +2094,11 @@ postgres   27093  0.0  0.0  30096  2752 ?        Ss   11:34   0:00 postgres: ser
       <entry>Waiting for WAL buffers to be written to disk.</entry>
      </row>
      <row>
+      <entry><literal>WrapLimitsVacuum</literal></entry>
+      <entry>Waiting to update limits on transaction id and multixact
+       consumption.</entry>
+     </row>
+     <row>
       <entry><literal>XactBuffer</literal></entry>
       <entry>Waiting for I/O on a transaction status SLRU buffer.</entry>
      </row>
diff --git a/src/backend/access/transam/slru.c b/src/backend/access/transam/slru.c
index 61249f4a12..275058036a 100644
--- a/src/backend/access/transam/slru.c
+++ b/src/backend/access/transam/slru.c
@@ -1208,6 +1208,14 @@ SimpleLruFlush(SlruCtl ctl, bool allow_redirtied)
 
 /*
  * Remove all segments before the one holding the passed page number
+ *
+ * All SLRUs prevent concurrent calls to this function, either with an LWLock
+ * or by calling it only as part of a checkpoint.  Mutual exclusion must begin
+ * before computing cutoffPage.  Mutual exclusion must end after any limit
+ * update that would permit other backends to write fresh data into the
+ * segment immediately preceding the one containing cutoffPage.  Otherwise,
+ * when the SLRU is quite full, SimpleLruTruncate() might delete that segment
+ * after it has accrued freshly-written data.
  */
 void
 SimpleLruTruncate(SlruCtl ctl, int cutoffPage)
diff --git a/src/backend/access/transam/subtrans.c b/src/backend/access/transam/subtrans.c
index f33ae407a6..d9a31d3c8d 100644
--- a/src/backend/access/transam/subtrans.c
+++ b/src/backend/access/transam/subtrans.c
@@ -349,8 +349,8 @@ ExtendSUBTRANS(TransactionId newestXact)
 /*
  * Remove all SUBTRANS segments before the one holding the passed transaction ID
  *
- * This is normally called during checkpoint, with oldestXact being the
- * oldest TransactionXmin of any running transaction.
+ * oldestXact is the oldest TransactionXmin of any running transaction.  This
+ * is called only during checkpoint.
  */
 void
 TruncateSUBTRANS(TransactionId oldestXact)
diff --git a/src/backend/commands/async.c b/src/backend/commands/async.c
index 71b7577afc..4c1286eb98 100644
--- a/src/backend/commands/async.c
+++ b/src/backend/commands/async.c
@@ -244,19 +244,22 @@ typedef struct QueueBackendStatus
 /*
  * Shared memory state for LISTEN/NOTIFY (excluding its SLRU stuff)
  *
- * The AsyncQueueControl structure is protected by the NotifyQueueLock.
+ * The AsyncQueueControl structure is protected by the NotifyQueueLock and
+ * NotifyQueueTailLock.
  *
- * When holding the lock in SHARED mode, backends may only inspect their own
- * entries as well as the head and tail pointers. Consequently we can allow a
- * backend to update its own record while holding only SHARED lock (since no
- * other backend will inspect it).
+ * When holding NotifyQueueLock in SHARED mode, backends may only inspect
+ * their own entries as well as the head and tail pointers. Consequently we
+ * can allow a backend to update its own record while holding only SHARED lock
+ * (since no other backend will inspect it).
  *
- * When holding the lock in EXCLUSIVE mode, backends can inspect the entries
- * of other backends and also change the head and tail pointers.
+ * When holding NotifyQueueLock in EXCLUSIVE mode, backends can inspect the
+ * entries of other backends and also change the head pointer. When holding
+ * both NotifyQueueLock and NotifyQueueTailLock in EXCLUSIVE mode, backends
+ * can change the tail pointer.
  *
  * NotifySLRULock is used as the control lock for the pg_notify SLRU buffers.
- * In order to avoid deadlocks, whenever we need both locks, we always first
- * get NotifyQueueLock and then NotifySLRULock.
+ * In order to avoid deadlocks, whenever we need multiple locks, we first get
+ * NotifyQueueTailLock, then NotifyQueueLock, and lastly NotifySLRULock.
  *
  * Each backend uses the backend[] array entry with index equal to its
  * BackendId (which can range from 1 to MaxBackends).  We rely on this to make
@@ -2177,6 +2180,10 @@ asyncQueueAdvanceTail(void)
 	int			newtailpage;
 	int			boundary;
 
+	/* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
+	LWLockAcquire(NotifyQueueTailLock, LW_EXCLUSIVE);
+
+	/* Compute the new tail. */
 	LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
 	min = QUEUE_HEAD;
 	for (BackendId i = QUEUE_FIRST_LISTENER; i > 0; i = QUEUE_NEXT_LISTENER(i))
@@ -2185,7 +2192,6 @@ asyncQueueAdvanceTail(void)
 		min = QUEUE_POS_MIN(min, QUEUE_BACKEND_POS(i));
 	}
 	oldtailpage = QUEUE_POS_PAGE(QUEUE_TAIL);
-	QUEUE_TAIL = min;
 	LWLockRelease(NotifyQueueLock);
 
 	/*
@@ -2205,6 +2211,17 @@ asyncQueueAdvanceTail(void)
 		 */
 		SimpleLruTruncate(NotifyCtl, newtailpage);
 	}
+
+	/*
+	 * Advertise the new tail.  This changes asyncQueueIsFull()'s verdict for
+	 * the segment immediately prior to the new tail, allowing fresh data into
+	 * that segment.
+	 */
+	LWLockAcquire(NotifyQueueLock, LW_EXCLUSIVE);
+	QUEUE_TAIL = min;
+	LWLockRelease(NotifyQueueLock);
+
+	LWLockRelease(NotifyQueueTailLock);
 }
 
 /*
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index d32de23e62..92c4eb6a87 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -1345,6 +1345,14 @@ vac_update_datfrozenxid(void)
 	bool		dirty = false;
 
 	/*
+	 * Restrict this task to one backend per database.  This avoids race
+	 * conditions that would move datfrozenxid or datminmxid backward.  It
+	 * avoids calling vac_truncate_clog() with a datfrozenxid preceding a
+	 * datfrozenxid passed to an earlier vac_truncate_clog() call.
+	 */
+	LockDatabaseFrozenIds(ExclusiveLock);
+
+	/*
 	 * Initialize the "min" calculation with GetOldestXmin, which is a
 	 * reasonable approximation to the minimum relfrozenxid for not-yet-
 	 * committed pg_class entries for new tables; see AddNewRelationTuple().
@@ -1533,6 +1541,9 @@ vac_truncate_clog(TransactionId frozenXID,
 	bool		bogus = false;
 	bool		frozenAlreadyWrapped = false;
 
+	/* Restrict task to one backend per cluster; see SimpleLruTruncate(). */
+	LWLockAcquire(WrapLimitsVacuumLock, LW_EXCLUSIVE);
+
 	/* init oldest datoids to sync with my frozenXID/minMulti values */
 	oldestxid_datoid = MyDatabaseId;
 	minmulti_datoid = MyDatabaseId;
@@ -1642,6 +1653,8 @@ vac_truncate_clog(TransactionId frozenXID,
 	 */
 	SetTransactionIdLimit(frozenXID, oldestxid_datoid);
 	SetMultiXactIdLimit(minMulti, minmulti_datoid, false);
+
+	LWLockRelease(WrapLimitsVacuumLock);
 }
 
 
diff --git a/src/backend/storage/lmgr/lmgr.c b/src/backend/storage/lmgr/lmgr.c
index 2010320095..7409de9405 100644
--- a/src/backend/storage/lmgr/lmgr.c
+++ b/src/backend/storage/lmgr/lmgr.c
@@ -461,6 +461,21 @@ UnlockRelationForExtension(Relation relation, LOCKMODE lockmode)
 }
 
 /*
+ *		LockDatabaseFrozenIds
+ *
+ * This allows one backend per database to execute vac_update_datfrozenxid().
+ */
+void
+LockDatabaseFrozenIds(LOCKMODE lockmode)
+{
+	LOCKTAG		tag;
+
+	SET_LOCKTAG_DATABASE_FROZEN_IDS(tag, MyDatabaseId);
+
+	(void) LockAcquire(&tag, lockmode, false, false);
+}
+
+/*
  *		LockPage
  *
  * Obtain a page-level lock.  This is currently used by some index access
@@ -1098,6 +1113,11 @@ DescribeLockTag(StringInfo buf, const LOCKTAG *tag)
 							 tag->locktag_field2,
 							 tag->locktag_field1);
 			break;
+		case LOCKTAG_DATABASE_FROZEN_IDS:
+			appendStringInfo(buf,
+							 _("pg_database.datfrozenxid of database %u"),
+							 tag->locktag_field1);
+			break;
 		case LOCKTAG_PAGE:
 			appendStringInfo(buf,
 							 _("page %u of relation %u of database %u"),
diff --git a/src/backend/storage/lmgr/lwlocknames.txt b/src/backend/storage/lmgr/lwlocknames.txt
index e6985e8eed..774292fd94 100644
--- a/src/backend/storage/lmgr/lwlocknames.txt
+++ b/src/backend/storage/lmgr/lwlocknames.txt
@@ -50,3 +50,6 @@ MultiXactTruncationLock				41
 OldSnapshotTimeMapLock				42
 LogicalRepWorkerLock				43
 XactTruncationLock					44
+# 45 was XactTruncationLock until removal of BackendRandomLock
+WrapLimitsVacuumLock				46
+NotifyQueueTailLock					47
diff --git a/src/backend/utils/adt/lockfuncs.c b/src/backend/utils/adt/lockfuncs.c
index e992d1bbfc..f592292d06 100644
--- a/src/backend/utils/adt/lockfuncs.c
+++ b/src/backend/utils/adt/lockfuncs.c
@@ -29,6 +29,7 @@
 const char *const LockTagTypeNames[] = {
 	"relation",
 	"extend",
+	"frozenid",
 	"page",
 	"tuple",
 	"transactionid",
@@ -254,6 +255,17 @@ pg_lock_status(PG_FUNCTION_ARGS)
 				nulls[8] = true;
 				nulls[9] = true;
 				break;
+			case LOCKTAG_DATABASE_FROZEN_IDS:
+				values[1] = ObjectIdGetDatum(instance->locktag.locktag_field1);
+				nulls[2] = true;
+				nulls[3] = true;
+				nulls[4] = true;
+				nulls[5] = true;
+				nulls[6] = true;
+				nulls[7] = true;
+				nulls[8] = true;
+				nulls[9] = true;
+				break;
 			case LOCKTAG_PAGE:
 				values[1] = ObjectIdGetDatum(instance->locktag.locktag_field1);
 				values[2] = ObjectIdGetDatum(instance->locktag.locktag_field2);
diff --git a/src/include/storage/lmgr.h b/src/include/storage/lmgr.h
index 3acc11aa5a..f7cabcbbf5 100644
--- a/src/include/storage/lmgr.h
+++ b/src/include/storage/lmgr.h
@@ -59,6 +59,9 @@ extern bool ConditionalLockRelationForExtension(Relation relation,
 												LOCKMODE lockmode);
 extern int	RelationExtensionLockWaiterCount(Relation relation);
 
+/* Lock to recompute pg_database.datfrozenxid in the current database */
+extern void LockDatabaseFrozenIds(LOCKMODE lockmode);
+
 /* Lock a page (currently only used within indexes) */
 extern void LockPage(Relation relation, BlockNumber blkno, LOCKMODE lockmode);
 extern bool ConditionalLockPage(Relation relation, BlockNumber blkno, LOCKMODE lockmode);
diff --git a/src/include/storage/lock.h b/src/include/storage/lock.h
index fdabf42721..1c3e9c1999 100644
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -138,6 +138,7 @@ typedef enum LockTagType
 {
 	LOCKTAG_RELATION,			/* whole relation */
 	LOCKTAG_RELATION_EXTEND,	/* the right to extend a relation */
+	LOCKTAG_DATABASE_FROZEN_IDS,	/* pg_database.datfrozenxid */
 	LOCKTAG_PAGE,				/* one page of a relation */
 	LOCKTAG_TUPLE,				/* one physical tuple */
 	LOCKTAG_TRANSACTION,		/* transaction (for waiting for xact done) */
@@ -194,6 +195,15 @@ typedef struct LOCKTAG
 	 (locktag).locktag_type = LOCKTAG_RELATION_EXTEND, \
 	 (locktag).locktag_lockmethodid = DEFAULT_LOCKMETHOD)
 
+/* ID info for frozen IDs is DB OID */
+#define SET_LOCKTAG_DATABASE_FROZEN_IDS(locktag,dboid) \
+	((locktag).locktag_field1 = (dboid), \
+	 (locktag).locktag_field2 = 0, \
+	 (locktag).locktag_field3 = 0, \
+	 (locktag).locktag_field4 = 0, \
+	 (locktag).locktag_type = LOCKTAG_DATABASE_FROZEN_IDS, \
+	 (locktag).locktag_lockmethodid = DEFAULT_LOCKMETHOD)
+
 /* ID info for a page is RELATION info + BlockNumber */
 #define SET_LOCKTAG_PAGE(locktag,dboid,reloid,blocknum) \
 	((locktag).locktag_field1 = (dboid), \