On Fri, 16 Apr 2004, Philip Hazel wrote:
> On Thu, 15 Apr 2004, Mike Bethune wrote:
>
> > actually, I think I set it in the wrong place before, I've tried the
> > below and it seems to be working for me now... (cross fingers)
>
> That really does suggest that the signal handler gets lost on fork().
> Your patch is certainly harmless, so if it fixes things for you, it is
> the right solution. Noted.
I made a small test program to see if I could confirm the
hypothesis that signal handlers get lost on fork(), but
I could not. It also doesn't seem to be necessary to restore
the handler after it's been used.
Perhaps this code in daemon.c has something to do with it?
/* Reset signals in the child */
signal(SIGALRM, SIG_DFL);
signal(SIGHUP, SIG_DFL);
signal(SIGCHLD, SIG_DFL);
My test results (and source code):
# ./forkalrm
Parent is 5193 and child 5194.
5194 got SIGALRM.
5193 got SIGALRM.
5194 got SIGALRM.
Process 5194 exiting normally.
5193 got SIGALRM.
Process 5193 exiting normally.
--- begin forkalrm.c ---
#include <unistd.h>
#include <signal.h>
#include <sys/poll.h>
#include <errno.h>
void sigalrm();
int pid;
main()
{
int i;
signal(SIGALRM, sigalrm);
i = fork();
pid = getpid();
if (i)
printf("Parent is %u and child %u.\n", pid, i);
alarm(10);
poll(NULL,0,15000);
if (errno==EINTR) {
alarm(5);
poll(NULL,0,10000);
}
printf("Process %u exiting normally.\n", pid);
}
void sigalrm()
{
printf("%u got SIGALRM.\n", pid);
}
--- end forkalrm.c ---
