On Thu, 8 Feb 2001, Solar Designer wrote:

> 1. You're calling exit(3) from a signal handler, while in a "bad
> state" (possibly somewhere inside libc stdio code).

Yes. It's a SIGCHLD handler which is essentially just "getting the hell
out", after shutting down/closing a few fd's.

I'm now using _exit(), but I'm still curious why exit() segfaults
sometimes. The call stack is something like

read(2) <-- blocking on a unix dgram socket read
SIGCHLD arrives, enters signal handler
exit(3)

What sort of "bad state" do you envisage?

> BTW, the OpenBSD folks are fixing two interesting classes of
> bugs now: fd_set overflows and various signal races. The latter
> are relevant to your problem.

Interesting...

Chris

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]