MDEV-5479: Prevent mysqld from stealing Unix socket of another active instance

mysql-test/main/socket_conflict.result

@@ -0,0 +1,19 @@
+#
+# MDEV-5479: Prevent mysqld from unlinking a Unix socket
+#            file actively used by another process
+#
+#
+# Test 1: Server must refuse to start when a listener is
+#         already present on the socket path
+#
+FOUND 1 /\[ERROR\] Another process is already listening on the socket file/ in socket_conflict.err
+#
+# Test 2: Stale socket file must be cleaned up at startup
+#
+# restart
+SELECT 1;
+1
+1
+#
+# End of 13.0 tests
+#

mysql-test/main/socket_conflict.test

@@ -0,0 +1,118 @@
+--source include/not_windows.inc
+--source include/not_embedded.inc
+
+--echo #
+--echo # MDEV-5479: Prevent mysqld from unlinking a Unix socket
+--echo #            file actively used by another process
+--echo #
+
+# Shut down the server once; both tests run while it is down.
+--source include/shutdown_mysqld.inc
+
+--echo #
+--echo # Test 1: Server must refuse to start when a listener is
+--echo #         already present on the socket path
+--echo #
+
+# Create a fake listener on the default socket path.
+# The child process creates and holds the listening socket;
+# the parent must not touch the socket object at all, because
+# Perl's IO::Socket->close() calls shutdown(SHUT_RDWR) which
+# would kill the listener in the child too.
+perl;
+  use IO::Socket::UNIX;
+  my $path= $ENV{MASTER_MYSOCK};
+  unlink $path if -e $path;
+  my $pid= fork();
+  die "fork: $!" unless defined $pid;
+  if ($pid == 0)
+  {
+    # Detach from parent's process group and close inherited
+    # pipe fds. mysqltest uses popen() for perl blocks and
+    # reads stdout to completion -- the child must close its
+    # copy of the pipe so mysqltest can proceed.
+    setpgrp(0, 0);
+    open(STDIN, '<', '/dev/null');
+    open(STDOUT, '>', '/dev/null');
+    open(STDERR, '>', '/dev/null');
+    my $srv= IO::Socket::UNIX->new(
+      Type   => SOCK_STREAM,
+      Local  => $path,
+      Listen => 1,
+    ) or exit 1;
+    while (1) { sleep 60; }
+  }
+  # Parent: wait for child to create the socket (up to 60s)
+  for (1..600)
+  {
+    last if -S $path;
+    select(undef, undef, undef, 0.1);
+  }
+  die "Fake listener not created at $path\n" unless -S $path;
+  my $pidfile= "$ENV{MYSQLTEST_VARDIR}/tmp/fake_server.pid";
+  open(my $fh, '>', $pidfile) or die "Cannot write $pidfile: $!\n";
+  print $fh $pid;
+  close $fh;
+EOF
+
+--let errorlog=$MYSQL_TMP_DIR/socket_conflict.err
+--let SEARCH_FILE=$errorlog
+
+# Use --loose-skip-innodb to skip InnoDB initialization, which
+# runs before network_init() and would otherwise be slow on CI.
+--error 1
+--exec $MYSQLD --defaults-group-suffix=.1 --defaults-file=$MYSQLTEST_VARDIR/my.cnf --loose-skip-innodb --log-error=$errorlog
+
+--let SEARCH_PATTERN=\[ERROR\] Another process is already listening on the socket file
+--source include/search_pattern_in_file.inc
+
+--remove_file $SEARCH_FILE
+
+# Kill the fake listener and clean up its socket file
+perl;
+  my $pidfile= "$ENV{MYSQLTEST_VARDIR}/tmp/fake_server.pid";
+  open(my $fh, '<', $pidfile) or die "Cannot read $pidfile: $!\n";
+  my $pid= <$fh>;
+  chomp $pid;
+  close $fh;
+  kill 'TERM', $pid;
+  # The child runs in its own process group (setpgrp) and was
+  # reparented to init, so waitpid won't work here. Poll until
+  # the process is gone.
+  for (1..600)
+  {
+    last unless kill(0, $pid);
+    select(undef, undef, undef, 0.1);
+  }
+  unlink $pidfile;
+  unlink $ENV{MASTER_MYSOCK};
+EOF
+
+--echo #
+--echo # Test 2: Stale socket file must be cleaned up at startup
+--echo #
+
+# Create a stale Unix socket at the default socket path.
+# The socket is bound then immediately closed, leaving an
+# orphaned file with no listener behind it.
+perl;
+  use IO::Socket::UNIX;
+  my $path= $ENV{MASTER_MYSOCK};
+  my $srv= IO::Socket::UNIX->new(
+    Type   => SOCK_STREAM,
+    Local  => $path,
+    Listen => 1,
+  ) or die "Cannot create socket at $path: $!\n";
+  $srv->close();
+EOF
+
+# Start the server normally -- it should detect the stale
+# socket, remove it, and bind successfully.
+--source include/start_mysqld.inc
+
+# Verify the server is operational
+SELECT 1;
+
+--echo #
+--echo # End of 13.0 tests
+--echo #

sql/mysqld.cc

@@ -2702,6 +2702,60 @@ static void use_systemd_activated_sockets()
 }
 
 
+#ifdef HAVE_SYS_UN_H
+/*
+  Check if an existing Unix socket file is actively in use.
+  If another process is listening on the socket, abort startup.
+  If the socket is stale (no listener), remove it so we can
+  re-bind.
+*/
+static void handle_stale_unix_socket(const char *path)
+{
+  MY_STAT stat_buf;
+
+  if (!my_stat(path, &stat_buf, MYF(0)))
+    return;  /* File does not exist */
+
+  if (S_ISSOCK(stat_buf.st_mode))
+  {
+    MYSQL_SOCKET test_sock;
+    test_sock= mysql_socket_socket(key_socket_unix,
+                                   AF_UNIX, SOCK_STREAM, 0);
+    if (mysql_socket_getfd(test_sock) >= 0)
+    {
+      struct sockaddr_un test_addr;
+      bzero((char*) &test_addr, sizeof(test_addr));
+      test_addr.sun_family= AF_UNIX;
+      strmov(test_addr.sun_path, path);
+      if (mysql_socket_connect(test_sock,
+            (struct sockaddr *) &test_addr,
+            sizeof(test_addr)) == 0)
+      {
+        /* Socket is active - another process is listening */
+        mysql_socket_close(test_sock);
+        sql_print_error("Another process is already listening "
+                        "on the socket file '%s'. "
+                        "Aborting.", path);
+        unireg_abort(1);
+      }
+      if (socket_errno == EACCES)
+      {
+        mysql_socket_close(test_sock);
+        sql_print_error("Socket file '%s' exists and is "
+                        "owned by another user. Cannot "
+                        "verify or replace it. Aborting.",
+                        path);
+        unireg_abort(1);
+      }
+      mysql_socket_close(test_sock);
+    }
+  }
+  /* File is stale or not a socket - safe to remove */
+  (void) unlink(path);
+}
+#endif /* HAVE_SYS_UN_H */
+
+
 static void network_init(void)
 {
 #ifdef HAVE_SYS_UN_H
@@ -2779,7 +2833,7 @@ static void network_init(void)
     else
 #endif
     {
-      (void) unlink(mysqld_unix_port);
+      handle_stale_unix_socket(mysqld_unix_port);
       port_len= sizeof(UNIXaddr);
     }
     arg= 1;