all: Ensure pid file exists when respawning child process

keepalived/core/main.c

@@ -169,6 +169,7 @@ const char *snmp_socket;				/* Socket to use for SNMP agent */
 #endif
 static const char *syslog_ident;			/* syslog ident if not default */
 bool use_pid_dir;					/* Put pid files in /run/keepalived or @localstatedir@/run/keepalived */
+bool children_started;					/* Set once children have been run first time */
 
 unsigned os_major;					/* Kernel version */
 unsigned os_minor;
@@ -560,6 +561,8 @@ start_keepalived(__attribute__((unused)) thread_ref_t thread)
 		pidfile_rm(&bfd_pidfile);
 #endif
 
+	children_started = true;
+
 #ifndef _ONE_PROCESS_DEBUG_
 	/* Do we have a reload file to monitor */
 	if (global_data->reload_time_file)

keepalived/core/pidfile.c

@@ -151,18 +151,6 @@ close_other_pidfiles(void)
 #endif
 }
 
-/* Create the running daemon pidfile */
-bool
-pidfile_write(const pidfile_t *pidf)
-{
-	if (pidf->fd == -1)
-		return false;
-
-	dprintf(pidf->fd, "%d\n", getpid());
-
-	return true;
-}
-
 /* return the daemon running state */
 static bool
 create_pidfile(pidfile_t *pidf)
@@ -226,7 +214,7 @@ create_pidfile(pidfile_t *pidf)
 			break;
 		}
 
-		/* pid file is now openned, locked and 0 length */
+		/* pid file is now opened, locked and 0 length */
 		return false;
 	}
 
@@ -258,3 +246,54 @@ keepalived_running(unsigned long mode)
 #endif
 	return false;
 }
+
+/* Create the running daemon pidfile */
+bool
+pidfile_write(pidfile_t *pidf)
+{
+	int ret;
+
+	if (pidf->fd == -1)
+		return false;
+
+	if (children_started) {
+		struct stat statb, fstatb;
+
+		/* There could be more error handling, but that will just
+		 * complicate the code for minimal benefit. */
+		if (stat(pidf->path, &statb)) {
+			/* pidfile no longer exists */
+			close(pidf->fd);
+			create_pidfile(pidf);
+		} else {
+			if (fstat(pidf->fd, &fstatb) ||
+			    statb.st_dev != fstatb.st_dev ||
+			    statb.st_ino != fstatb.st_ino) {
+				/* The pidfile has been deleted and recreated. Open the new one. */
+				close(pidf->fd);
+				while ((pidf->fd = open(pidf->path, O_NOFOLLOW | O_WRONLY | O_NONBLOCK)) == -1 && errno == EINTR);
+
+				if (pidf->fd == -1)
+					return false;
+			}
+
+			/* Since we have already written to the pid file,
+			 * we need to reset the file offset and truncate the file. */
+			off_t offs = lseek(pidf->fd, 0, SEEK_CUR);
+			if (offs) {
+				lseek(pidf->fd, 0, SEEK_SET);
+				if (ftruncate(pidf->fd, 0))
+					log_message(LOG_INFO, "ftruncate error %d - %m", errno);
+			}
+		}
+	}
+
+	ret = dprintf(pidf->fd, "%d\n", getpid());
+
+	if (ret < 0)
+		log_message(LOG_INFO, "pidfile_write returned %d, errno %d - %m", ret, errno);
+	else
+		log_message(LOG_INFO, "pidfile_write returned %d", ret);
+
+	return true;
+}

keepalived/include/main.h

@@ -79,6 +79,7 @@ extern bool snmp_option;		/* Enable SNMP support */
 extern const char *snmp_socket;		/* Socket to use for SNMP agent */
 #endif
 extern bool use_pid_dir;		/* pid files in /run/keepalived */
+extern bool children_started;		/* Set once children have been run first time */
 extern unsigned os_major;		/* Kernel version */
 extern unsigned os_minor;
 extern unsigned os_release;

keepalived/include/pidfile.h

@@ -58,7 +58,7 @@ extern void create_pid_dir(void);
 extern void remove_pid_dir(void);
 extern char *make_pidfile_name(const char *, const char *, const char *);
 extern void pidfile_close(pidfile_t *, bool);
-extern bool pidfile_write(const pidfile_t *);
+extern bool pidfile_write(pidfile_t *);
 extern void pidfile_rm(pidfile_t *);
 extern void close_other_pidfiles(void);
 extern bool keepalived_running(unsigned long);