[sheepdog] [PATCH] sheepdog:show more detail of the crash source

Hitoshi Mitake mitake.hitoshi at lab.ntt.co.jp
Thu Feb 12 08:04:54 CET 2015


At Wed, 11 Feb 2015 21:47:41 -0800,
Wang Zhengyong wrote:
> 
> In the current sheepdog, when sheepdog crashed,
> there is too little information about the signal source.
> 
> This patch use (*handler)(int, siginfo_t *, void *)
> instead of (*handler)(int). In this way, can show more detail of the
> crash problem, especially the pid of singal sender
> 
> Cc: Hitoshi Mitake <mitake.hitoshi at gmail.com>
> Signed-off-by: Wang Zhengyong <wangzhengyong at cmss.chinamobile.com>
> ---
>  dog/dog.c           |    2 +-
>  include/util.h      |    4 ++--
>  lib/logger.c        |    4 ++--
>  lib/util.c          |   11 +++++++----
>  sheep/sheep.c       |    9 +++++----
>  shepherd/shepherd.c |    2 +-
>  6 files changed, 18 insertions(+), 14 deletions(-)
> 
> diff --git a/dog/dog.c b/dog/dog.c
> index 54520dd..77aa27b 100644
> --- a/dog/dog.c
> +++ b/dog/dog.c
> @@ -368,7 +368,7 @@ static const struct sd_option *build_sd_options(const char *opts)
>  	return sd_opts;
>  }
>  
> -static void crash_handler(int signo)
> +static void crash_handler(int signo, siginfo_t *info, void *context)
>  {
>  	sd_err("dog exits unexpectedly (%s).", strsignal(signo));
>  
> diff --git a/include/util.h b/include/util.h
> index 6a513e0..3c34b40 100644
> --- a/include/util.h
> +++ b/include/util.h
> @@ -108,8 +108,8 @@ int rmdir_r(const char *dir_path);
>  int purge_directory(const char *dir_path);
>  bool is_numeric(const char *p);
>  const char *data_to_str(void *data, size_t data_length);
> -int install_sighandler(int signum, void (*handler)(int), bool once);
> -int install_crash_handler(void (*handler)(int));
> +int install_sighandler(int signum, void (*handler)(int, siginfo_t *, void *), bool once);
> +int install_crash_handler(void (*handler)(int, siginfo_t *, void *));
>  void reraise_crash_signal(int signo, int status);
>  pid_t gettid(void);
>  int tkill(int tid, int sig);
> diff --git a/lib/logger.c b/lib/logger.c
> index 02bab00..da0ebac 100644
> --- a/lib/logger.c
> +++ b/lib/logger.c
> @@ -531,7 +531,7 @@ static bool is_sheep_dead(int signo)
>  	return signo == SIGHUP;
>  }
>  
> -static void crash_handler(int signo)
> +static void crash_handler(int signo, siginfo_t *info, void *context)
>  {
>  	if (is_sheep_dead(signo))
>  		sd_err("sheep pid %d exited unexpectedly.", sheep_pid);
> @@ -552,7 +552,7 @@ static void crash_handler(int signo)
>  	reraise_crash_signal(signo, 1);
>  }
>  
> -static void sighup_handler(int signo)
> +static void sighup_handler(int signo, siginfo_t *info, void *context)
>  {
>  	rotate_log();
>  }
> diff --git a/lib/util.c b/lib/util.c
> index 21e0143..089455d 100644
> --- a/lib/util.c
> +++ b/lib/util.c
> @@ -524,25 +524,28 @@ const char *data_to_str(void *data, size_t data_length)
>   * If 'once' is true, the signal will be restored to the default state
>   * after 'handler' is called.
>   */
> -int install_sighandler(int signum, void (*handler)(int), bool once)
> +int install_sighandler(int signum, void (*handler)(int, siginfo_t *, void *), bool once)
>  {
>  	struct sigaction sa = {};
>  
>  	sa.sa_handler = handler;

The handler is now a function which should be called via
sa.sa_sigaction. You need to assign handler to the sa_sigaction.

Other part looks good to me.

Thanks,
Hitoshi

> +	sa.sa_flags = SA_SIGINFO;
> +
>  	if (once)
> -		sa.sa_flags = SA_RESETHAND | SA_NODEFER;
> +		sa.sa_flags = sa.sa_flags | SA_RESETHAND | SA_NODEFER;
>  	sigemptyset(&sa.sa_mask);
>  
>  	return sigaction(signum, &sa, NULL);
>  }
>  
> -int install_crash_handler(void (*handler)(int))
> +int install_crash_handler(void (*handler)(int, siginfo_t *, void *))
>  {
>  	return install_sighandler(SIGSEGV, handler, true) ||
>  		install_sighandler(SIGABRT, handler, true) ||
>  		install_sighandler(SIGBUS, handler, true) ||
>  		install_sighandler(SIGILL, handler, true) ||
> -		install_sighandler(SIGFPE, handler, true);
> +		install_sighandler(SIGFPE, handler, true) ||
> +		install_sighandler(SIGQUIT, handler, true);
>  }
>  
>  /*
> diff --git a/sheep/sheep.c b/sheep/sheep.c
> index e0a034f..6c540ae 100644
> --- a/sheep/sheep.c
> +++ b/sheep/sheep.c
> @@ -239,7 +239,7 @@ static void signal_handler(int listen_fd, int events, void *data)
>  
>  	ret = read(sigfd, &siginfo, sizeof(siginfo));
>  	assert(ret == sizeof(siginfo));
> -	sd_debug("signal %d", siginfo.ssi_signo);
> +	sd_debug("signal %d, ssi pid %d", siginfo.ssi_signo, siginfo.ssi_pid);
>  	switch (siginfo.ssi_signo) {
>  	case SIGTERM:
>  		sys->cinfo.status = SD_STATUS_KILLED;
> @@ -276,9 +276,10 @@ static int init_signal(void)
>  	return 0;
>  }
>  
> -static void crash_handler(int signo)
> +static void crash_handler(int signo, siginfo_t *info, void *context)
>  {
> -	sd_emerg("sheep exits unexpectedly (%s).", strsignal(signo));
> +	sd_emerg("sheep exits unexpectedly (%s), si pid %d, uid %d, errno %d, code %d",
> +		strsignal(signo), info->si_pid, info->si_uid, info->si_errno, info->si_code);
>  
>  	sd_backtrace();
>  	sd_dump_variable(__sys);
> @@ -639,7 +640,7 @@ end:
>  	return status;
>  }
>  
> -static void sighup_handler(int signum)
> +static void sighup_handler(int signo, siginfo_t *info, void *context)
>  {
>  	if (unlikely(logger_pid == -1))
>  		return;
> diff --git a/shepherd/shepherd.c b/shepherd/shepherd.c
> index fd59229..dc6df74 100644
> --- a/shepherd/shepherd.c
> +++ b/shepherd/shepherd.c
> @@ -637,7 +637,7 @@ static int set_listen_fd_cb(int fd, void *data)
>  	return 0;
>  }
>  
> -static void crash_handler(int signo)
> +static void crash_handler(int signo, siginfo_t *info, void *context)
>  {
>  	sd_emerg("shepherd exits unexpectedly (%s).", strsignal(signo));
>  
> -- 
> 1.7.1
> 
> 
> 



More information about the sheepdog mailing list