[CRIU] [PATCH] restore: block SIGCHILD for sub-processes

[Tycho Andersen]

On Mon, Oct 05, 2015 at 02:56:06PM +0300, Andrey Vagin wrote:
> From: Andrew Vagin <avagin at openvz.org>
> 
> Otherwise sigchld_handler() will call waitpid()
> and worries about unknown process.
> 
> (00.333012)      1: kernel/hostname nr 15
> (00.333120)      1: kernel/domainname nr 6
> (00.335243)      1: Error (cr-restore.c:1225): BUG at cr-restore.c:1225
> 
> Cc: Tycho Andersen <tycho.andersen at canonical.com>

Whoops, somehow I didn't hit this when I ran the tests, thanks:

Acked-by: Tycho Andersen <tycho.andersen at canonical.com>

> Reported-by: Mr Jenkins
> Signed-off-by: Andrew Vagin <avagin at openvz.org>
> ---
>  sysctl.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)
> 
> diff --git a/sysctl.c b/sysctl.c
> index e9fdf40..9a4bd1a 100644
> --- a/sysctl.c
> +++ b/sysctl.c
> @@ -196,6 +196,7 @@ static int __userns_sysctl_op(void *arg, int unused, pid_t pid)
>  	struct sysctl_userns_req *userns_req = arg;
>  	int op = userns_req->op;
>  	struct sysctl_req *req, **reqs = NULL;
> +	sigset_t blockmask, oldmask;
>  	pid_t worker;
>  
>  	// fix up the pointer
> @@ -266,6 +267,16 @@ static int __userns_sysctl_op(void *arg, int unused, pid_t pid)
>  		req = (struct sysctl_req *) (((char *) req) + total_len);
>  	}
>  
> +	/*
> +	 * Don't let the sigchld_handler() mess with us
> +	 * calling waitpid() on the exited worker. The
> +	 * same is done in cr_system().
> +	 */
> +
> +	sigemptyset(&blockmask);
> +	sigaddset(&blockmask, SIGCHLD);
> +	sigprocmask(SIG_BLOCK, &blockmask, &oldmask);
> +
>  	worker = fork();
>  	if (worker < 0)
>  		goto out;
> @@ -310,6 +321,8 @@ static int __userns_sysctl_op(void *arg, int unused, pid_t pid)
>  	ret = 0;
>  
>  out:
> +	sigprocmask(SIG_BLOCK, &oldmask, NULL);
> +
>  	if (fds) {
>  		for (i = 0; i < userns_req->nr_req; i++) {
>  			if (fds[i] < 0)
> -- 
> 2.4.3
>