Handle DROP DATABASE getting interrupted

Until now, when DROP DATABASE got interrupted in the wrong moment, the removal
of the pg_database row would also roll back, even though some irreversible
steps have already been taken. E.g. DropDatabaseBuffers() might have thrown
out dirty buffers, or files could have been unlinked. But we continued to
allow connections to such a corrupted database.

To fix this, mark databases invalid with an in-place update, just before
starting to perform irreversible steps. As we can't add a new column in the
back branches, we use pg_database.datconnlimit = -2 for this purpose.

An invalid database cannot be connected to anymore, but can still be
dropped.

Unfortunately we can't easily add output to psql's \l to indicate that some
database is invalid, it doesn't fit in any of the existing columns.

Add tests verifying that a interrupted DROP DATABASE is handled correctly in
the backend and in various tools.

Reported-by: Evgeny Morozov <postgresql3@realityexists.net>
Author: Andres Freund <andres@anarazel.de>
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Discussion: https://postgr.es/m/20230509004637.cgvmfwrbht7xm7p6@awork3.anarazel.de
Discussion: https://postgr.es/m/20230314174521.74jl6ffqsee5mtug@awork3.anarazel.de
Backpatch: 11-, bug present in all supported versions

Author: anarazel

doc/src/sgml/catalogs.sgml

@@ -3042,7 +3042,8 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</replaceable>:<replaceable>&l
       </para>
       <para>
        Sets maximum number of concurrent connections that can be made
-       to this database.  -1 means no limit.
+       to this database.  -1 means no limit, -2 indicates the database is
+       invalid.
       </para></entry>
      </row>
 

src/backend/commands/dbcommands.c

@@ -719,7 +719,7 @@ createdb(ParseState *pstate, const CreatedbStmt *stmt)
 	int			encoding = -1;
 	bool		dbistemplate = false;
 	bool		dballowconnections = true;
-	int			dbconnlimit = -1;
+	int			dbconnlimit = DATCONNLIMIT_UNLIMITED;
 	char	   *dbcollversion = NULL;
 	int			notherbackends;
 	int			npreparedxacts;
@@ -926,7 +926,7 @@ createdb(ParseState *pstate, const CreatedbStmt *stmt)
 	if (dconnlimit && dconnlimit->arg)
 	{
 		dbconnlimit = defGetInt32(dconnlimit);
-		if (dbconnlimit < -1)
+		if (dbconnlimit < DATCONNLIMIT_UNLIMITED)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
 					 errmsg("invalid connection limit: %d", dbconnlimit)));
@@ -977,6 +977,16 @@ createdb(ParseState *pstate, const CreatedbStmt *stmt)
 				 errmsg("template database \"%s\" does not exist",
 						dbtemplate)));
 
+	/*
+	 * If the source database was in the process of being dropped, we can't
+	 * use it as a template.
+	 */
+	if (database_is_invalid_oid(src_dboid))
+		ereport(ERROR,
+				errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+				errmsg("cannot use invalid database \"%s\" as template", dbtemplate),
+				errhint("Use DROP DATABASE to drop invalid databases."));
+
 	/*
 	 * Permission check: to copy a DB that's not marked datistemplate, you
 	 * must be superuser or the owner thereof.
@@ -1576,6 +1586,7 @@ dropdb(const char *dbname, bool missing_ok, bool force)
 	bool		db_istemplate;
 	Relation	pgdbrel;
 	HeapTuple	tup;
+	Form_pg_database datform;
 	int			notherbackends;
 	int			npreparedxacts;
 	int			nslots,
@@ -1691,17 +1702,6 @@ dropdb(const char *dbname, bool missing_ok, bool force)
 						dbname),
 				 errdetail_busy_db(notherbackends, npreparedxacts)));
 
-	/*
-	 * Remove the database's tuple from pg_database.
-	 */
-	tup = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(db_id));
-	if (!HeapTupleIsValid(tup))
-		elog(ERROR, "cache lookup failed for database %u", db_id);
-
-	CatalogTupleDelete(pgdbrel, &tup->t_self);
-
-	ReleaseSysCache(tup);
-
 	/*
 	 * Delete any comments or security labels associated with the database.
 	 */
@@ -1718,6 +1718,37 @@ dropdb(const char *dbname, bool missing_ok, bool force)
 	 */
 	dropDatabaseDependencies(db_id);
 
+	/*
+	 * Tell the cumulative stats system to forget it immediately, too.
+	 */
+	pgstat_drop_database(db_id);
+
+	tup = SearchSysCacheCopy1(DATABASEOID, ObjectIdGetDatum(db_id));
+	if (!HeapTupleIsValid(tup))
+		elog(ERROR, "cache lookup failed for database %u", db_id);
+	datform = (Form_pg_database) GETSTRUCT(tup);
+
+	/*
+	 * Except for the deletion of the catalog row, subsequent actions are not
+	 * transactional (consider DropDatabaseBuffers() discarding modified
+	 * buffers). But we might crash or get interrupted below. To prevent
+	 * accesses to a database with invalid contents, mark the database as
+	 * invalid using an in-place update.
+	 *
+	 * We need to flush the WAL before continuing, to guarantee the
+	 * modification is durable before performing irreversible filesystem
+	 * operations.
+	 */
+	datform->datconnlimit = DATCONNLIMIT_INVALID_DB;
+	heap_inplace_update(pgdbrel, tup);
+	XLogFlush(XactLastRecEnd);
+
+	/*
+	 * Also delete the tuple - transactionally. If this transaction commits,
+	 * the row will be gone, but if we fail, dropdb() can be invoked again.
+	 */
+	CatalogTupleDelete(pgdbrel, &tup->t_self);
+
 	/*
 	 * Drop db-specific replication slots.
 	 */
@@ -1730,11 +1761,6 @@ dropdb(const char *dbname, bool missing_ok, bool force)
 	 */
 	DropDatabaseBuffers(db_id);
 
-	/*
-	 * Tell the cumulative stats system to forget it immediately, too.
-	 */
-	pgstat_drop_database(db_id);
-
 	/*
 	 * Tell checkpointer to forget any pending fsync and unlink requests for
 	 * files in the database; else the fsyncs will fail at next checkpoint, or
@@ -2248,7 +2274,7 @@ AlterDatabase(ParseState *pstate, AlterDatabaseStmt *stmt, bool isTopLevel)
 	ListCell   *option;
 	bool		dbistemplate = false;
 	bool		dballowconnections = true;
-	int			dbconnlimit = -1;
+	int			dbconnlimit = DATCONNLIMIT_UNLIMITED;
 	DefElem    *distemplate = NULL;
 	DefElem    *dallowconnections = NULL;
 	DefElem    *dconnlimit = NULL;
@@ -2319,7 +2345,7 @@ AlterDatabase(ParseState *pstate, AlterDatabaseStmt *stmt, bool isTopLevel)
 	if (dconnlimit && dconnlimit->arg)
 	{
 		dbconnlimit = defGetInt32(dconnlimit);
-		if (dbconnlimit < -1)
+		if (dbconnlimit < DATCONNLIMIT_UNLIMITED)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
 					 errmsg("invalid connection limit: %d", dbconnlimit)));
@@ -2346,6 +2372,14 @@ AlterDatabase(ParseState *pstate, AlterDatabaseStmt *stmt, bool isTopLevel)
 	datform = (Form_pg_database) GETSTRUCT(tuple);
 	dboid = datform->oid;
 
+	if (database_is_invalid_form(datform))
+	{
+		ereport(FATAL,
+				errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+				errmsg("cannot alter invalid database \"%s\"", stmt->dbname),
+				errhint("Use DROP DATABASE to drop invalid databases."));
+	}
+
 	if (!object_ownercheck(DatabaseRelationId, dboid, GetUserId()))
 		aclcheck_error(ACLCHECK_NOT_OWNER, OBJECT_DATABASE,
 					   stmt->dbname);
@@ -3064,6 +3098,42 @@ get_database_name(Oid dbid)
 	return result;
 }
 
+
+/*
+ * While dropping a database the pg_database row is marked invalid, but the
+ * catalog contents still exist. Connections to such a database are not
+ * allowed.
+ */
+bool
+database_is_invalid_form(Form_pg_database datform)
+{
+	return datform->datconnlimit == DATCONNLIMIT_INVALID_DB;
+}
+
+
+/*
+ * Convenience wrapper around database_is_invalid_form()
+ */
+bool
+database_is_invalid_oid(Oid dboid)
+{
+	HeapTuple	dbtup;
+	Form_pg_database dbform;
+	bool		invalid;
+
+	dbtup = SearchSysCache1(DATABASEOID, ObjectIdGetDatum(dboid));
+	if (!HeapTupleIsValid(dbtup))
+		elog(ERROR, "cache lookup failed for database %u", dboid);
+	dbform = (Form_pg_database) GETSTRUCT(dbtup);
+
+	invalid = database_is_invalid_form(dbform);
+
+	ReleaseSysCache(dbtup);
+
+	return invalid;
+}
+
+
 /*
  * recovery_create_dbdir()
  *

src/backend/commands/vacuum.c

@@ -1850,6 +1850,20 @@ vac_truncate_clog(TransactionId frozenXID,
 		Assert(TransactionIdIsNormal(datfrozenxid));
 		Assert(MultiXactIdIsValid(datminmxid));
 
+		/*
+		 * If database is in the process of getting dropped, or has been
+		 * interrupted while doing so, no connections to it are possible
+		 * anymore. Therefore we don't need to take it into account here.
+		 * Which is good, because it can't be processed by autovacuum either.
+		 */
+		if (database_is_invalid_form((Form_pg_database) dbform))
+		{
+			elog(DEBUG2,
+				 "skipping invalid database \"%s\" while computing relfrozenxid",
+				 NameStr(dbform->datname));
+			continue;
+		}
+
 		/*
 		 * If things are working properly, no database should have a
 		 * datfrozenxid or datminmxid that is "in the future".  However, such

src/backend/postmaster/autovacuum.c

@@ -1972,6 +1972,18 @@ get_database_list(void)
 		avw_dbase  *avdb;
 		MemoryContext oldcxt;
 
+		/*
+		 * If database has partially been dropped, we can't, nor need to,
+		 * vacuum it.
+		 */
+		if (database_is_invalid_form(pgdatabase))
+		{
+			elog(DEBUG2,
+				 "autovacuum: skipping invalid database \"%s\"",
+				 NameStr(pgdatabase->datname));
+			continue;
+		}
+
 		/*
 		 * Allocate our results in the caller's context, not the
 		 * transaction's. We do this inside the loop, and restore the original

src/backend/utils/init/postinit.c

@@ -1116,6 +1116,7 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	if (!bootstrap)
 	{
 		HeapTuple	tuple;
+		Form_pg_database datform;
 
 		tuple = GetDatabaseTuple(dbname);
 		if (!HeapTupleIsValid(tuple) ||
@@ -1125,6 +1126,15 @@ InitPostgres(const char *in_dbname, Oid dboid,
 					(errcode(ERRCODE_UNDEFINED_DATABASE),
 					 errmsg("database \"%s\" does not exist", dbname),
 					 errdetail("It seems to have just been dropped or renamed.")));
+
+		datform = (Form_pg_database) GETSTRUCT(tuple);
+		if (database_is_invalid_form(datform))
+		{
+			ereport(FATAL,
+					errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+					errmsg("cannot connect to invalid database \"%s\"", dbname),
+					errhint("Use DROP DATABASE to drop invalid databases."));
+		}
 	}
 
 	/*

src/bin/pg_amcheck/pg_amcheck.c

@@ -1590,7 +1590,7 @@ compile_database_list(PGconn *conn, SimplePtrList *databases,
 						 "FROM pg_catalog.pg_database d "
 						 "LEFT OUTER JOIN exclude_raw e "
 						 "ON d.datname ~ e.rgx "
-						 "\nWHERE d.datallowconn "
+						 "\nWHERE d.datallowconn AND datconnlimit != -2 "
 						 "AND e.pattern_id IS NULL"
 						 "),"
 

src/bin/pg_amcheck/t/002_nonesuch.pl

@@ -291,6 +291,40 @@
 	'many unmatched patterns and one matched pattern under --no-strict-names'
 );
 
+
+#########################################
+# Test that an invalid / partially dropped database won't be targeted
+
+$node->safe_psql(
+	'postgres', q(
+	CREATE DATABASE regression_invalid;
+	UPDATE pg_database SET datconnlimit = -2 WHERE datname = 'regression_invalid';
+));
+
+$node->command_checks_all(
+	[
+		'pg_amcheck', '-d', 'regression_invalid'
+	],
+	1,
+	[qr/^$/],
+	[
+		qr/pg_amcheck: error: no connectable databases to check matching "regression_invalid"/,
+	],
+	'checking handling of invalid database');
+
+$node->command_checks_all(
+	[
+	    'pg_amcheck', '-d', 'postgres',
+		'-t', 'regression_invalid.public.foo',
+	],
+	1,
+	[qr/^$/],
+	[
+		qr/pg_amcheck: error: no connectable databases to check matching "regression_invalid.public.foo"/,
+	],
+	'checking handling of object in invalid database');
+
+
 #########################################
 # Test checking otherwise existent objects but in databases where they do not exist
 

src/bin/pg_dump/pg_dumpall.c

@@ -1345,7 +1345,7 @@ dropDBs(PGconn *conn)
 	res = executeQuery(conn,
 					   "SELECT datname "
 					   "FROM pg_database d "
-					   "WHERE datallowconn "
+					   "WHERE datallowconn AND datconnlimit != -2 "
 					   "ORDER BY datname");
 
 	if (PQntuples(res) > 0)
@@ -1488,7 +1488,7 @@ dumpDatabases(PGconn *conn)
 	res = executeQuery(conn,
 					   "SELECT datname "
 					   "FROM pg_database d "
-					   "WHERE datallowconn "
+					   "WHERE datallowconn AND datconnlimit != -2 "
 					   "ORDER BY (datname <> 'template1'), datname");
 
 	if (PQntuples(res) > 0)

src/bin/pg_dump/t/002_pg_dump.pl

@@ -1907,6 +1907,17 @@
 		},
 	},
 
+	'CREATE DATABASE regression_invalid...' => {
+		create_order => 1,
+		create_sql => q(
+		    CREATE DATABASE regression_invalid;
+			UPDATE pg_database SET datconnlimit = -2 WHERE datname = 'regression_invalid'),
+		regexp => qr/^CREATE DATABASE regression_invalid/m,
+		not_like => {
+			pg_dumpall_dbprivs => 1,
+		},
+	},
+
 	'CREATE ACCESS METHOD gist2' => {
 		create_order => 52,
 		create_sql =>
@@ -4690,6 +4701,14 @@
 	qr/pg_dump: error: connection to server .* failed: FATAL:  database "qqq" does not exist/,
 	'connecting to a non-existent database');
 
+#########################################
+# Test connecting to an invalid database
+
+$node->command_fails_like(
+	[ 'pg_dump', '-d', 'regression_invalid' ],
+	qr/pg_dump: error: connection to server .* failed: FATAL:  cannot connect to invalid database "regression_invalid"/,
+	'connecting to an invalid database');
+
 #########################################
 # Test connecting with an unprivileged user