Fix pg_upgrade to cope with non-default unix_socket_directory scenarios.

When starting either an old or new postmaster, force it to place its Unix
socket in the current directory.  This makes it even harder for accidental
connections to occur during pg_upgrade, and also works around some
scenarios where the default socket location isn't usable.  (For example,
if the default location is something other than "/tmp", it might not exist
during "make check".)

When checking an already-running old postmaster, find out its actual socket
directory location from postmaster.pid, if possible.  This dodges problems
with an old postmaster having a configured location different from the
default built into pg_upgrade's libpq.  We can't find that out if the old
postmaster is pre-9.1, so also document how to cope with such scenarios
manually.

In support of this, centralize handling of the connection-related command
line options passed to pg_upgrade's subsidiary programs, such as pg_dump.
This should make future changes easier.

Bruce Momjian and Tom Lane

Author: tglsfdc

contrib/pg_upgrade/check.c

@@ -184,8 +184,8 @@ issue_warnings(char *sequence_script_file_name)
 		{
 			prep_status("Adjusting sequences");
 			exec_prog(UTILITY_LOG_FILE, NULL, true,
-					  "\"%s/psql\" " EXEC_PSQL_ARGS " --port %d --username \"%s\" -f \"%s\"",
-					  new_cluster.bindir, new_cluster.port, os_info.user,
+					  "\"%s/psql\" " EXEC_PSQL_ARGS " %s -f \"%s\"",
+					  new_cluster.bindir, cluster_conn_opts(&new_cluster),
 					  sequence_script_file_name);
 			unlink(sequence_script_file_name);
 			check_ok();

contrib/pg_upgrade/dump.c

@@ -24,8 +24,8 @@ generate_old_dump(void)
 	 * restores the frozenid's for databases and relations.
 	 */
 	exec_prog(UTILITY_LOG_FILE, NULL, true,
-			  "\"%s/pg_dumpall\" --port %d --username \"%s\" --schema-only --binary-upgrade %s -f %s",
-			  new_cluster.bindir, old_cluster.port, os_info.user,
+			  "\"%s/pg_dumpall\" %s --schema-only --binary-upgrade %s -f %s",
+			  new_cluster.bindir, cluster_conn_opts(&old_cluster),
 			  log_opts.verbose ? "--verbose" : "",
 			  ALL_DUMP_FILE);
 	check_ok();

contrib/pg_upgrade/option.c

@@ -9,6 +9,8 @@
 
 #include "postgres.h"
 
+#include "miscadmin.h"
+
 #include "pg_upgrade.h"
 
 #include <getopt_long.h>
@@ -376,3 +378,64 @@ adjust_data_dir(ClusterInfo *cluster)
 
 	check_ok();
 }
+
+
+/*
+ * get_sock_dir
+ *
+ * Identify the socket directory to use for this cluster.  If we're doing
+ * a live check (old cluster only), we need to find out where the postmaster
+ * is listening.  Otherwise, we're going to put the socket into the current
+ * directory.
+ */
+void
+get_sock_dir(ClusterInfo *cluster, bool live_check)
+{
+#ifdef HAVE_UNIX_SOCKETS
+	if (!live_check)
+	{
+		/* Use the current directory for the socket */
+		cluster->sockdir = pg_malloc(MAXPGPATH);
+		if (!getcwd(cluster->sockdir, MAXPGPATH))
+			pg_log(PG_FATAL, "cannot find current directory\n");
+	}
+	else
+	{
+		/*
+		 *	If we are doing a live check, we will use the old cluster's Unix
+		 *	domain socket directory so we can connect to the live server.
+		 */
+
+		/* sockdir was added to postmaster.pid in PG 9.1 */
+		if (GET_MAJOR_VERSION(cluster->major_version) >= 901)
+		{
+			char		filename[MAXPGPATH];
+			FILE		*fp;
+			int			i;
+
+			snprintf(filename, sizeof(filename), "%s/postmaster.pid",
+					 cluster->pgdata);
+			if ((fp = fopen(filename, "r")) == NULL)
+				pg_log(PG_FATAL, "Could not get socket directory of the old server\n");
+
+			cluster->sockdir = pg_malloc(MAXPGPATH);
+			for (i = 0; i < LOCK_FILE_LINE_SOCKET_DIR; i++)
+				if (fgets(cluster->sockdir, MAXPGPATH, fp) == NULL)
+					pg_log(PG_FATAL, "Could not get socket directory of the old server\n");
+
+			fclose(fp);
+
+			/* Remove trailing newline */
+			if (strchr(cluster->sockdir, '\n') != NULL)
+				*strchr(cluster->sockdir, '\n') = '\0';
+		}
+		else
+		{
+			/* Can't get live sockdir, so assume the default is OK. */
+			cluster->sockdir = NULL;
+		}
+	}
+#else /* !HAVE_UNIX_SOCKETS */
+	cluster->sockdir = NULL;
+#endif
+}

contrib/pg_upgrade/pg_upgrade.c

@@ -88,6 +88,9 @@ main(int argc, char **argv)
 	check_cluster_versions();
 	check_cluster_compatibility(live_check);
 
+	get_sock_dir(&old_cluster, live_check);
+	get_sock_dir(&new_cluster, false);
+
 	check_old_cluster(live_check, &sequence_script_file_name);
 
 
@@ -211,8 +214,8 @@ prepare_new_cluster(void)
 	 */
 	prep_status("Analyzing all rows in the new cluster");
 	exec_prog(UTILITY_LOG_FILE, NULL, true,
-			  "\"%s/vacuumdb\" --port %d --username \"%s\" --all --analyze %s",
-			  new_cluster.bindir, new_cluster.port, os_info.user,
+			  "\"%s/vacuumdb\" %s --all --analyze %s",
+			  new_cluster.bindir, cluster_conn_opts(&new_cluster),
 			  log_opts.verbose ? "--verbose" : "");
 	check_ok();
 
@@ -224,8 +227,8 @@ prepare_new_cluster(void)
 	 */
 	prep_status("Freezing all rows on the new cluster");
 	exec_prog(UTILITY_LOG_FILE, NULL, true,
-			  "\"%s/vacuumdb\" --port %d --username \"%s\" --all --freeze %s",
-			  new_cluster.bindir, new_cluster.port, os_info.user,
+			  "\"%s/vacuumdb\" %s --all --freeze %s",
+			  new_cluster.bindir, cluster_conn_opts(&new_cluster),
 			  log_opts.verbose ? "--verbose" : "");
 	check_ok();
 
@@ -261,8 +264,8 @@ prepare_new_databases(void)
 	 * the template0 template.
 	 */
 	exec_prog(RESTORE_LOG_FILE, NULL, true,
-			  "\"%s/psql\" " EXEC_PSQL_ARGS " --port %d --username \"%s\" -f \"%s\"",
-			  new_cluster.bindir, new_cluster.port, os_info.user,
+			  "\"%s/psql\" " EXEC_PSQL_ARGS " %s -f \"%s\"",
+			  new_cluster.bindir, cluster_conn_opts(&new_cluster),
 			  GLOBALS_DUMP_FILE);
 	check_ok();
 
@@ -290,8 +293,8 @@ create_new_objects(void)
 
 	prep_status("Restoring database schema to new cluster");
 	exec_prog(RESTORE_LOG_FILE, NULL, true,
-			  "\"%s/psql\" " EXEC_PSQL_ARGS " --port %d --username \"%s\" -f \"%s\"",
-			  new_cluster.bindir, new_cluster.port, os_info.user,
+			  "\"%s/psql\" " EXEC_PSQL_ARGS " %s -f \"%s\"",
+			  new_cluster.bindir, cluster_conn_opts(&new_cluster),
 			  DB_DUMP_FILE);
 	check_ok();
 

contrib/pg_upgrade/pg_upgrade.h

@@ -226,6 +226,7 @@ typedef struct
 	char	   *bindir;			/* pathname for cluster's executable directory */
 	char	   *pgopts;			/* options to pass to the server, like pg_ctl
 								 * -o */
+	char	   *sockdir;		/* directory for Unix Domain socket, if any */
 	unsigned short port;		/* port number where postmaster is waiting */
 	uint32		major_version;	/* PG_VERSION of cluster */
 	char		major_version_str[64];	/* string PG_VERSION of cluster */
@@ -387,6 +388,7 @@ void print_maps(FileNameMap *maps, int n,
 
 void		parseCommandLine(int argc, char *argv[]);
 void		adjust_data_dir(ClusterInfo *cluster);
+void		get_sock_dir(ClusterInfo *cluster, bool live_check);
 
 /* relfilenode.c */
 
@@ -407,6 +409,8 @@ PGresult *
 executeQueryOrDie(PGconn *conn, const char *fmt,...)
 __attribute__((format(PG_PRINTF_ATTRIBUTE, 2, 3)));
 
+char	   *cluster_conn_opts(ClusterInfo *cluster);
+
 void		start_postmaster(ClusterInfo *cluster);
 void		stop_postmaster(bool fast);
 uint32		get_major_server_version(ClusterInfo *cluster);

contrib/pg_upgrade/server.c

@@ -46,21 +46,54 @@ connectToServer(ClusterInfo *cluster, const char *db_name)
 /*
  * get_db_conn()
  *
- * get database connection
+ * get database connection, using named database + standard params for cluster
  */
 static PGconn *
 get_db_conn(ClusterInfo *cluster, const char *db_name)
 {
-	char		conn_opts[MAXPGPATH];
+	char		conn_opts[2 * NAMEDATALEN + MAXPGPATH + 100];
 
-	snprintf(conn_opts, sizeof(conn_opts),
-			 "dbname = '%s' user = '%s' port = %d", db_name, os_info.user,
-			 cluster->port);
+	if (cluster->sockdir)
+		snprintf(conn_opts, sizeof(conn_opts),
+				 "dbname = '%s' user = '%s' host = '%s' port = %d",
+				 db_name, os_info.user, cluster->sockdir, cluster->port);
+	else
+		snprintf(conn_opts, sizeof(conn_opts),
+				 "dbname = '%s' user = '%s' port = %d",
+				 db_name, os_info.user, cluster->port);
 
 	return PQconnectdb(conn_opts);
 }
 
 
+/*
+ * cluster_conn_opts()
+ *
+ * Return standard command-line options for connecting to this cluster when
+ * using psql, pg_dump, etc.  Ideally this would match what get_db_conn()
+ * sets, but the utilities we need aren't very consistent about the treatment
+ * of database name options, so we leave that out.
+ *
+ * Note result is in static storage, so use it right away.
+ */
+char *
+cluster_conn_opts(ClusterInfo *cluster)
+{
+	static char	conn_opts[MAXPGPATH + NAMEDATALEN + 100];
+
+	if (cluster->sockdir)
+		snprintf(conn_opts, sizeof(conn_opts),
+				 "--host \"%s\" --port %d --username \"%s\"",
+				 cluster->sockdir, cluster->port, os_info.user);
+	else
+		snprintf(conn_opts, sizeof(conn_opts),
+				 "--port %d --username \"%s\"",
+				 cluster->port, os_info.user);
+
+	return conn_opts;
+}
+
+
 /*
  * executeQueryOrDie()
  *
@@ -140,17 +173,35 @@ stop_postmaster_atexit(void)
 void
 start_postmaster(ClusterInfo *cluster)
 {
-	char		cmd[MAXPGPATH];
+	char		cmd[MAXPGPATH * 4 + 1000];
 	PGconn	   *conn;
 	bool		exit_hook_registered = false;
 	bool		pg_ctl_return = false;
+	char		socket_string[MAXPGPATH + 200];
 
 	if (!exit_hook_registered)
 	{
 		atexit(stop_postmaster_atexit);
 		exit_hook_registered = true;
 	}
 
+	socket_string[0] = '\0';
+
+#ifdef HAVE_UNIX_SOCKETS
+	/* prevent TCP/IP connections, restrict socket access */
+	strcat(socket_string,
+		   " -c listen_addresses='' -c unix_socket_permissions=0700");
+
+	/* Have a sockdir?  Tell the postmaster. */
+	if (cluster->sockdir)
+		snprintf(socket_string + strlen(socket_string),
+				 sizeof(socket_string) - strlen(socket_string),
+				 " -c %s='%s'",
+				 (GET_MAJOR_VERSION(cluster->major_version) < 903) ?
+				 "unix_socket_directory" : "unix_socket_directories",
+				 cluster->sockdir);
+#endif
+
 	/*
 	 * Using autovacuum=off disables cleanup vacuum and analyze, but freeze
 	 * vacuums can still happen, so we set autovacuum_freeze_max_age to its
@@ -159,12 +210,12 @@ start_postmaster(ClusterInfo *cluster)
 	 * not touch them.
 	 */
 	snprintf(cmd, sizeof(cmd),
-			 "\"%s/pg_ctl\" -w -l \"%s\" -D \"%s\" -o \"-p %d %s %s\" start",
+			 "\"%s/pg_ctl\" -w -l \"%s\" -D \"%s\" -o \"-p %d %s %s%s\" start",
 		  cluster->bindir, SERVER_LOG_FILE, cluster->pgconfig, cluster->port,
 			 (cluster->controldata.cat_ver >=
 			  BINARY_UPGRADE_SERVER_FLAG_CAT_VER) ? "-b" :
 			 "-c autovacuum=off -c autovacuum_freeze_max_age=2000000000",
-			 cluster->pgopts ? cluster->pgopts : "");
+			 cluster->pgopts ? cluster->pgopts : "", socket_string);
 
 	/*
 	 * Don't throw an error right away, let connecting throw the error because

doc/src/sgml/pgupgrade.sgml

@@ -519,6 +519,14 @@ psql --username postgres --file script.sql postgres
    <literal>-d /real-data-directory -o '-D /configuration-directory'</>.
   </para>
 
+  <para>
+   If doing <option>--check</> with a running old server of a pre-9.1 version,
+   and the old server is using a Unix-domain socket directory that is
+   different from the default built into the new <productname>PostgreSQL</>
+   installation, set <envar>PGHOST</> to point to the socket location of the
+   old server.  (This is not relevant on Windows.)
+  </para>
+
   <para>
    A Log-Shipping Standby Server (<xref linkend="warm-standby">) cannot
    be upgraded because the server must allow writes.  The simplest way