fixed severe bug about removing the pidfile in the wrong place (not propagating the state of the pidfile object from parent to daemon!)

2 files changed, 31 insertions, 20 deletions

diff --git a/src/daemon/daemon.c b/src/daemon/daemon.c
index a6fe63a..abcf47c 100644
--- a/src/daemon/daemon.c
+++ b/src/daemon/daemon.c
@@ -370,7 +370,7 @@ wolf_error_t wolf_daemon_start( wolf_daemon_p d ) {
 			 */
 			WOLF_LOG( WOLF_LOG_DEBUG, "Parent after first fork: child is %d", pid );
 			d->state = TERMINATE_EXIT_CODE;
-			return WOLF_ERR_PROGRAMMING; /* intensional error code here, so we get into wolf_daemon_exit! */
+			return( d->error = WOLF_ERR_PROGRAMMING ); /* intensional error code here, so we get into wolf_daemon_exit! */
 	}
 
 	/* second pipe communicates from the daemon back to its parent
@@ -428,7 +428,7 @@ wolf_error_t wolf_daemon_start( wolf_daemon_p d ) {
 			(void)install_signal_handlers_parent( );
 			WOLF_LOG( WOLF_LOG_DEBUG, "Parent after second fork: child (and daemon) is %d", pid );
 			d->state = TERMINATE_PARENT;
-			return WOLF_ERR_PROGRAMMING; /* intensional error code here, so we get into wolf_daemon_exit! */
+			return( d->error = WOLF_ERR_PROGRAMMING ); /* intensional error code here, so we get into wolf_daemon_exit! */
 	}
 
 	/* Now install signal handlers */
@@ -579,7 +579,6 @@ wolf_error_t wolf_daemon_start( wolf_daemon_p d ) {
 
 NORETURN void wolf_daemon_exit( wolf_daemon_p d ) {
 	int exit_code;
-	int pid_file_fd;
 
 	WOLF_LOG( WOLF_LOG_DEBUG, "daemon_exit called int state %d (error %d)",
 		d->state, d->error );
@@ -612,18 +611,18 @@ NORETURN void wolf_daemon_exit( wolf_daemon_p d ) {
 			/* wait for termination signal */
 			WOLF_LOG( WOLF_LOG_DEBUG, "Waiting on parent pipe for termination signal" );
 
-			atomar_read( daemon_parent_pipe[0], &pid_file_fd, sizeof( int ) );
-			WOLF_LOG( WOLF_LOG_DEBUG, "Parent got termination (pidfile fd: %d).. cleaning up now (PID: %lu)",
-				pid_file_fd, getpid( ) );
+			atomar_read( daemon_parent_pipe[0], &d->pidfile.locked, sizeof( bool ) );
+			atomar_read( daemon_parent_pipe[0], &d->pidfile.running, sizeof( bool ) );
+			atomar_read( daemon_parent_pipe[0], &d->pidfile.fd, sizeof( int ) );
+			WOLF_LOG( WOLF_LOG_DEBUG, "Parent got termination (pidfile lock: %d, run: %d, fd: %d).. cleaning up now (PID: %lu)",
+				d->pidfile.locked, d->pidfile.running, d->pidfile.fd, getpid( ) );
 
 			/* we need root permissions for that! */
-			d->pidfile.fd = pid_file_fd;
 			(void)pidfile_remove( &d->pidfile );
 
 			wolf_daemon_signals_terminate( );
 
-			WOLF_LOG( WOLF_LOG_DEBUG, "Terminating parent of daemon pid file fd %d (PID %lu)",
-				pid_file_fd, getpid( ) );
+			WOLF_LOG( WOLF_LOG_DEBUG, "Terminating parent of daemon (PID %lu)", getpid( ) );
 
 			/* exit code is irrelevant here.. */
 			exit( EXIT_SUCCESS );
@@ -635,6 +634,8 @@ NORETURN void wolf_daemon_exit( wolf_daemon_p d ) {
 			/* This is the daemon terminating, signal the
 			 * parent that we are terminating
 			 */
+			atomar_write( daemon_parent_pipe[1], &d->pidfile.locked, sizeof( bool ) );
+			atomar_write( daemon_parent_pipe[1], &d->pidfile.running, sizeof( bool ) );
 			atomar_write( daemon_parent_pipe[1], &d->pidfile.fd, sizeof( int ) );
 
 			wolf_daemon_signals_terminate( );
@@ -659,6 +660,8 @@ NORETURN void wolf_daemon_exit( wolf_daemon_p d ) {
 			atomar_write( exit_code_pipe[1], &exit_code, sizeof( int ) );
 
 			/* also terminate the parent of the daemon */
+			atomar_write( daemon_parent_pipe[1], &d->pidfile.locked, sizeof( bool ) );
+			atomar_write( daemon_parent_pipe[1], &d->pidfile.running, sizeof( bool ) );
 			atomar_write( daemon_parent_pipe[1], &d->pidfile.fd, sizeof( int ) );
 
 			wolf_daemon_signals_terminate( );
diff --git a/src/daemon/pidfile.c b/src/daemon/pidfile.c
index aa4e3b0..839e2cf 100644
--- a/src/daemon/pidfile.c
+++ b/src/daemon/pidfile.c
@@ -57,11 +57,11 @@ bool is_daemon_running( struct pidfile_t *pidfile, pid_t *pid, wolf_error_t *err
 			WOLF_LOG( WOLF_LOG_DEBUG, "No pidfile '%s' found, daemon is not running", pidfile->filename );
 			(void)close( pidfile->fd );
 			*error = WOLF_OK;
-			return pidfile->running;
+			return( pidfile->running = false );
 		} else {
 			WOLF_LOG( WOLF_LOG_EMERG, "Unable to open pidfile '%s' for reading: %s", pidfile->filename, strerror( errno ) );
 			*error = WOLF_ERR_INTERNAL;
-			return pidfile->running;
+			return( pidfile->running = false );
 		}
 	}
 	
@@ -78,11 +78,13 @@ bool is_daemon_running( struct pidfile_t *pidfile, pid_t *pid, wolf_error_t *err
 			WOLF_LOG( WOLF_LOG_EMERG, "Unable to lock pidfile '%s': %s", pidfile->filename, strerror( errno ) );
 			(void)close( pidfile->fd );
 			*error = WOLF_ERR_INTERNAL;
-			pidfile->running = false;
-			return pidfile->running;
+			return( pidfile->running = false );
 		}
 	} else {
-		pidfile->locked = false;
+		/* we can lock the file but it could be a relict, so we are not sure
+		 * yet if the daemon is already running or not
+		 */
+		pidfile->locked = true;
 	}
 	
 	/* try to read the pid from the file */
@@ -91,7 +93,7 @@ bool is_daemon_running( struct pidfile_t *pidfile, pid_t *pid, wolf_error_t *err
 		WOLF_LOG( WOLF_LOG_EMERG, "Unable to read pid from pidfile '%s': %s", pidfile->filename, strerror( errno ) );
 		(void)close( pidfile->fd );
 		*error = WOLF_ERR_INTERNAL;
-		return pidfile->running;
+		return( pidfile->running = false );
 	}
 	
 	/* parse the file and see if we have a valid pid on the first line */
@@ -107,7 +109,7 @@ bool is_daemon_running( struct pidfile_t *pidfile, pid_t *pid, wolf_error_t *err
 		WOLF_LOG( WOLF_LOG_EMERG, "pidfile '%s' contains invalid data, can't read PID from it!", pidfile->filename );
 		(void)close( pidfile->fd );
 		*error = WOLF_ERR_INTERNAL;
-		return pidfile->running;
+		return( pidfile->running = false );
 	}
 	WOLF_LOG( WOLF_LOG_DEBUG, "Found PID '%lu' in pidfile", *pid );
 	
@@ -121,12 +123,12 @@ bool is_daemon_running( struct pidfile_t *pidfile, pid_t *pid, wolf_error_t *err
 			WOLF_LOG( WOLF_LOG_EMERG, "Found PID '%lu' in pidfile '%s', but no such process is running. Check and manually delete the pidfile!", *pid, pidfile->filename );
 			(void)close( pidfile->fd );
 			*error = WOLF_ERR_INTERNAL;
-			return pidfile->running;
+			return( pidfile->running = false );
 		} else {
-			WOLF_LOG( WOLF_LOG_EMERG, "Can't check if processor with PID '%lu' is alive: %s", *pid, strerror( errno ) );
+			WOLF_LOG( WOLF_LOG_EMERG, "Can't check if proces with PID '%lu' is alive: %s", *pid, strerror( errno ) );
 			(void)close( pidfile->fd );
 			*error = WOLF_ERR_INTERNAL;
-			return pidfile->running;
+			return( pidfile->running = false );
 		}
 	}
 	WOLF_LOG( WOLF_LOG_DEBUG, "A process with PID '%lu' is already running", *pid );
@@ -136,7 +138,7 @@ bool is_daemon_running( struct pidfile_t *pidfile, pid_t *pid, wolf_error_t *err
 	 */
 	(void)close( pidfile->fd );
 	*error = WOLF_OK;
-	return pidfile->running;
+	return( pidfile->running = true );
 }
 
 wolf_error_t pidfile_create( struct pidfile_t *pidfile ) {
@@ -233,6 +235,12 @@ wolf_error_t pidfile_remove( struct pidfile_t *pidfile ) {
 	WOLF_LOG( WOLF_LOG_DEBUG, "Removing pidfile '%s' (fd: %d, locked: %d, running: %d)",
 		pidfile->filename, pidfile->fd, pidfile->locked, pidfile->running );
 
+	/* if we know that another process is running, then we also know that we are
+	 * not allowed to remove the pidfile
+	 */
+	if( pidfile->running ) {
+		return WOLF_OK;
+	}
 
 	/* unconditionally remove the pidfile */
 	if( unlink( pidfile->filename ) < 0 ) {