bug#25624: [PATCH] timeout: Fix signal race in SIGALRM handling

[Tobias Stoeckmann]

On Sun, Feb 05, 2017 at 10:26:35AM -0800, Paul Eggert wrote:
> Pádraig Brady wrote:
> > In general this is a largely theoretical race right?
> > I.E. pids would need to be recycled between the waitpid() and exit()?
> 
> Not that theoretical, in the common case of systems with wraparaound PIDs
> and a small PID space. All you need is a long-running child on a busy
> system.

Yes, normally it is small enough to overflow in less than a minute if an
attacker runs fork() kill() in a loop.

I have updated the patch so it passes the test. As I don't have enough
experience in portable #ifdef's for all supported systems, I hope you can
adjust the patch as needed.


Tobias