From eee26673e0f5da0589d718e986661c878a4747d2 Mon Sep 17 00:00:00 2001
From: wangjufeng <wangjufeng@huawei.com>
Date: Sat, 20 Oct 2018 14:50:28 +0800
Subject: [PATCH] tt


Conflict: Remove previous patch's unused variable,these unused variable
          would cause compile failure
---
 usr/event_poll.c |  17 +++++++-
 usr/iscsid.c     |   2 +-
 usr/log.c        | 122 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-
 usr/log.h        |   6 +++
 4 files changed, 142 insertions(+), 5 deletions(-)

diff --git a/usr/event_poll.c b/usr/event_poll.c
index ffd12a3..e810de9 100644
--- a/usr/event_poll.c
+++ b/usr/event_poll.c
@@ -151,8 +151,11 @@ void event_loop_exit(queue_task_t *qtask)
 
 void event_loop(struct iscsi_ipc *ipc, int control_fd, int mgmt_ipc_fd)
 {
-	struct pollfd poll_array[POLL_MAX];
+	struct pollfd poll_array[POLL_MAX + 1];
 	int res, has_shutdown_children = 0;
+	int log_timerfd = log_get_timerfd();
+	int pool_count = POLL_MAX;
+	int poll_timerfd_index = -1;
 	sigset_t sigset;
 	int sig_fd;
 
@@ -174,6 +177,13 @@ void event_loop(struct iscsi_ipc *ipc, int control_fd, int mgmt_ipc_fd)
 	poll_array[POLL_ALARM].fd = sig_fd;
 	poll_array[POLL_ALARM].events = POLLIN;
 
+	if (log_timerfd >= 0) {
+		poll_array[POLL_MAX].fd = log_timerfd;
+		poll_array[POLL_MAX].events = POLLIN;
+		poll_timerfd_index = POLL_MAX;
+		pool_count += 1;
+	}
+
 	event_loop_stop = 0;
 	while (1) {
 		if (event_loop_stop) {
@@ -188,7 +198,7 @@ void event_loop(struct iscsi_ipc *ipc, int control_fd, int mgmt_ipc_fd)
 		/* Runs actors and may set alarm for future actors */
 		actor_poll();
 
-		res = poll(poll_array, POLL_MAX, reap_count ? REAP_WAKEUP : -1);
+		res = poll(poll_array, pool_count, reap_count ? REAP_WAKEUP : -1);
 
 		if (res > 0) {
 			log_debug(6, "poll result %d", res);
@@ -198,6 +208,9 @@ void event_loop(struct iscsi_ipc *ipc, int control_fd, int mgmt_ipc_fd)
 			if (poll_array[POLL_IPC].revents)
 				mgmt_ipc_handle(mgmt_ipc_fd);
 
+			if(log_timerfd >= 0 && poll_array[poll_timerfd_index].revents)
+				log_watchdog_handle();
+
 			if (poll_array[POLL_ALARM].revents) {
 				struct signalfd_siginfo si;
 
diff --git a/usr/iscsid.c b/usr/iscsid.c
index 478c83d..cd6f562 100644
--- a/usr/iscsid.c
+++ b/usr/iscsid.c
@@ -66,7 +66,6 @@ struct iscsi_daemon_config daemon_config;
 struct iscsi_daemon_config *dconfig = &daemon_config;
 
 static char program_name[] = "iscsid";
-static pid_t log_pid;
 static gid_t gid;
 static bool daemonize = true;
 static int mgmt_ipc_fd;
@@ -311,6 +310,7 @@ static void iscsid_shutdown(void)
 		log_debug(1, "daemon stopping");
 		log_close(log_pid);
 	}
+	log_close_watchdog_timer();
 }
 
 static void catch_signal(int signo)
diff --git a/usr/log.c b/usr/log.c
index 29cf39f..55c6a08 100644
--- a/usr/log.c
+++ b/usr/log.c
@@ -18,6 +18,7 @@
 #include <sys/types.h>
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <sys/timerfd.h>
 
 #include "iscsi_util.h"
 #include "log.h"
@@ -35,6 +36,13 @@ char *log_name;
 int log_level = 0;
 struct logarea *la = NULL;
 
+pid_t log_pid = -1;
+static int default_watchdog_timeout = 4;
+static char program_name[] = "iscsid";
+static int watchdog_timerfd = -1;
+static int last_watchdog_count = 0;
+static int watchdog_noupdate_count = 0;
+
 static int log_stop_daemon = 0;
 static void (*log_func)(int prio, void *priv, const char *fmt, va_list ap);
 static void *log_func_priv;
@@ -142,6 +150,7 @@ static int logarea_init (int size)
 	la->shmid_buff = shmid;
 	la->ops[0].sem_num = 0;
 	la->ops[0].sem_flg = 0;
+	la->watchdog_count = 0;
 
 	return 0;
 
@@ -421,6 +430,103 @@ static void __log_close(void)
 	}
 }
 
+static int log_restart_daemon(void)
+{
+	log_close(log_pid);
+	log_pid = log_init(program_name, DEFAULT_AREA_SIZE, log_do_log_daemon, NULL);
+	if (log_pid < 0)
+		return -1;
+	return 0;
+}
+
+static int log_watchdog_setup(void)
+{
+	int r;
+	struct itimerspec new_value;
+	uint32_t period_inus = 1 * 1000000; /*1 second*/
+	uint32_t seconds;
+	uint32_t nanoseconds;
+	if (log_stop_daemon || watchdog_timerfd >= 0)
+		return 0;
+
+	watchdog_timerfd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK | TFD_CLOEXEC);
+	if (watchdog_timerfd < 0) {
+		syslog(LOG_ERR, "Create watchdog timer failed. errno: %d", errno);
+		return -1;
+	}
+	seconds = period_inus / 1000000;
+	nanoseconds = (period_inus - (seconds * 1000000)) * 1000;
+	new_value.it_interval.tv_sec = seconds;
+	new_value.it_interval.tv_nsec = nanoseconds;
+	new_value.it_value.tv_sec = seconds;
+	new_value.it_value.tv_nsec = nanoseconds;
+
+	r = timerfd_settime(watchdog_timerfd, 0, &new_value, NULL);
+	if (r < 0) {
+		syslog(LOG_ERR, "Set watchdog timer failed. errno: %d", errno);
+		log_close_watchdog_timer();
+		return -1;
+	}
+	return 0;
+}
+
+static int log_read_comm_frpm_pid(pid_t pid, char *buffer, int len)
+{
+	char comm_file[256];
+	FILE *fd = NULL;
+	buffer[0] = '\0';
+	snprintf(comm_file, sizeof(comm_file), "/proc/%i/comm", pid);
+	fd = fopen(comm_file, "r");
+	if (!fd)
+		return -1;
+	if (!fgets(buffer, len, fd)) {
+		log_warning("Can not read file. File:%s.", comm_file);
+               fclose(fd);
+               fd = NULL;
+		return -1;
+	}
+       fclose(fd);
+       fd = NULL;
+	return 0;
+}
+
+int log_get_timerfd(void)
+{
+	return watchdog_timerfd;
+}
+
+void log_close_watchdog_timer(void)
+{
+	if(watchdog_timerfd >= 0) {
+		close(watchdog_timerfd);
+		watchdog_timerfd = -1;
+	}
+}
+
+void log_watchdog_handle(void)
+{
+	uint64_t expir = 0;
+	int len;
+	int new_watchdog_count;
+	len = read(watchdog_timerfd, &expir, sizeof(uint64_t));
+	if(len > 0 && expir > 0) {
+		if (la && !log_stop_daemon) {
+			new_watchdog_count = la->watchdog_count;
+			if (new_watchdog_count == last_watchdog_count) {
+				watchdog_noupdate_count++;
+				if (watchdog_noupdate_count >= default_watchdog_timeout) {
+					watchdog_noupdate_count = 0;
+					syslog(LOG_ERR, "log daemon is not alive, try to restart.");
+					log_restart_daemon();
+				}
+			} else {
+				watchdog_noupdate_count = 0;
+				last_watchdog_count = new_watchdog_count;
+			}
+		}
+	}
+}
+
 int log_init(char *program_name, int size,
 	void (*func)(int prio, void *priv, const char *fmt, va_list ap),
 	void *priv)
@@ -443,6 +549,8 @@ int log_init(char *program_name, int size,
 			return -1;
 		}
 
+		log_watchdog_setup();
+
 		pid = fork();
 		if (pid < 0) {
 			syslog(LOG_ERR, "starting logger failed");
@@ -463,6 +571,8 @@ int log_init(char *program_name, int size,
 		sigaction(SIGTERM, &sa_new, &sa_old );
 
 		while(1) {
+			if (la)
+				la->watchdog_count++;
 			log_flush();
 			sleep(1);
 
@@ -487,7 +597,15 @@ void log_close(pid_t pid)
 	}
 
 	if (pid > 0) {
-		kill(pid, SIGTERM);
-		waitpid(pid, &status, 0);
+		char comm[1024];
+		int r;
+		r = log_read_comm_frpm_pid(pid, comm, sizeof(comm));
+		if (!r && strstr(comm, "iscsid")) {
+			syslog(LOG_WARNING, "Send SIGTERM to iSCSI logger, pid=%i.", pid);
+			kill(pid, SIGTERM);
+			waitpid(pid, &status, 0);
+		} else {
+			syslog(LOG_WARNING, "iSCSI logger with pid=%i already exited.", pid);
+		}
 	}
 }
diff --git a/usr/log.h b/usr/log.h
index c548791..f0c88c7 100644
--- a/usr/log.h
+++ b/usr/log.h
@@ -28,6 +28,7 @@
 
 #include <stdarg.h>
 #include <sys/types.h>
+#include <stdint.h>
 #include "iscsid.h"
 
 union semun {
@@ -42,6 +43,7 @@ union semun {
 #define MAX_MSG_SIZE 256
 
 extern int log_level;
+extern pid_t log_pid;
 
 struct logmsg {
 	short int prio;
@@ -62,10 +64,14 @@ struct logarea {
 	struct sembuf ops[1];
 	int semid;
 	union semun semarg;
+	uint64_t watchdog_count;
 };
 
 extern struct logarea *la;
 
+extern int log_get_timerfd(void);
+extern void log_close_watchdog_timer(void);
+extern void log_watchdog_handle(void);
 extern int log_init(char *program_name, int size,
 	void (*func)(int prio, void *priv, const char *fmt, va_list ap),
 	void *priv);
-- 
1.8.3.1