Question

I have a process and want to start it again when it is killed. To achieve this, I start child “guardian” process, that uses prctl(PR_SET_PDEATHSIG, SIGHUP); to catch killing of its parent and starts it again.

Here is code of guardian (logging omitted):

void restart (int signal) {
    if (getppid() == 1) {
        if (fork() == 0) {
            execl("./process", 0);
        }
        exit(1);
    }
}

int main() {
    prctl(PR_SET_PDEATHSIG, SIGHUP, NULL, NULL, NULL);

    struct sigaction new_action, old_action;
    new_action.sa_handler = restart;
    sigemptyset (&new_action.sa_mask);
    new_action.sa_flags = 0;

    sigaction (SIGHUP, NULL, &old_action);

    if (old_action.sa_handler != SIG_IGN) {
        sigaction (SIGHUP, &new_action, NULL);
    } 

    while (getppid() != 1) {
        sleep(86400000);
    }
    return 0;
}

And parent:

int main() {
    if (fork() == 0) {
        execl("./guardian", 0);
    } 
    while (1) {
        cout << "I am process\n";
        sleep(1);
    }
    return 0;
}

The problem I have is that it works just one time. Here is ps output when process was started first time:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
1012     13058  0.0  0.3  20244  1932 pts/1    Ss   08:22   0:00 -sh
1012     22084  0.0  0.1  11484  1004 pts/1    S+   11:20   0:00  \_ ./process
1012     22085  0.0  0.1  11484  1000 pts/1    S+   11:20   0:00      \_ [guardian]
1012     12510  0.0  0.3  20784  1712 pts/0    Ss   08:14   0:00 -sh
1012     22088  0.0  0.1  17412  1012 pts/0    R+   11:20   0:00  \_ ps fu

which looks good. Next I kill process with kill -9 22084. And again ps output:

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
1012     13058  0.0  0.3  20244  1932 pts/1    Ss+  08:22   0:00 -sh
1012     12510  0.0  0.3  20784  1712 pts/0    Ss   08:14   0:00 -sh
1012     22091  0.0  0.1  17412  1012 pts/0    R+   11:21   0:00  \_ ps fu
1012     22089  0.0  0.1  11484   996 pts/1    S    11:20   0:00 [process]
1012     22090  0.0  0.1  11484   996 pts/1    S    11:20   0:00  \_ [guardian]

and when I kill process again kill -9 22089 guardian does not seem to get SIGHUP callback (I checked from logs, they are omitted here).

USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
1012     13058  0.0  0.3  20244  1932 pts/1    Ss+  08:22   0:00 -sh
1012     12510  0.0  0.3  20784  1712 pts/0    Rs   08:14   0:00 -sh
1012     22339  0.0  0.1  17412  1008 pts/0    R+   11:27   0:00  \_ ps fu
1012     22090  0.0  0.1  11484   996 pts/1    S    11:20   0:00 [guardian]

My question is – why guardian does not get SIGHUP?

I suspect it might have something to to with background process group – when process is restarted it is in background group (compare S+ and S in ps stat).

Answer 1

It looks like SIGHUP is blocked while you are in the signal handler handling SIGHUP. fork() and exec() inherit the signal mask, hence your second guardian never receives it again.

Unblock SIGHUP while in the signal handler after fork() before exec() parent.