CSE Roadmap

Core tier · about 3 hours · 6 parts

Part 4 of 6 · 2 min read

Uncatchable signals

In plain words: If a program could catch every signal, a buggy or hostile program could refuse to die. The kernel therefore keeps two signals that no program can catch, block or ignore: SIGKILL (number 9) and SIGSTOP (which stops a process until it receives SIGCONT). The kernel carries them out by itself.

A test tries to ignore both:

#include <stdio.h>
#include <signal.h>
#include <string.h>
#include <errno.h>
int main(void) {
    struct sigaction sa;
    memset(&sa, 0, sizeof sa);
    sa.sa_handler = SIG_IGN;
    sigemptyset(&sa.sa_mask);
    int r1 = sigaction(SIGKILL, &sa, NULL); int e1 = errno;
    int r2 = sigaction(SIGSTOP, &sa, NULL);
    printf("SIGKILL: %d (%s)\nSIGSTOP: %d (%s)\n", r1, strerror(e1), r2, strerror(errno));
    return 0;
}
SIGKILL: -1 (Invalid argument)
SIGSTOP: -1 (Invalid argument)

Both calls failed with the error "Invalid argument".

This explains the habit of system administrators: first send SIGTERM (kill PID), which lets the program clean up and end, and use SIGKILL (kill -9 PID) only if it does not. SIGKILL gives the program no chance to save files or release resources.

A parent can see the difference. The program below forks children that wait, and kills them with SIGTERM and SIGKILL, and also makes one crash with a null write:

SIGTERM: signaled=1 sig=15
null write: sig=11 core=0
SIGKILL: sig=9

The core=0 appears because the test machine does not write core files. On a typical system, signal 11 would also produce one.

Common mistake

Using kill -9 first. The process cannot clean up, so it can leave temporary files, locks and half-written data behind. Send SIGTERM first.

Check yourself: 1. Which two signals cannot be caught? 2. What number did the parent read for a child killed with SIGTERM? 3. Why should SIGTERM be tried first?