Re: Fix race conditions in timeout(1)

[Pádraig Brady]

On 11/03/2024 16:22, Pádraig Brady wrote:
> On 11/03/2024 15:17, Pádraig Brady wrote:
> > There was a narrow race condition reported at:
> > https://github.com/coreutils/coreutils/issues/82
> >
> > Then I noticed another related race condition
> > in the case of failing fork().
> >
> > Both should be addressed by the attached.
>
> Actually the second one to ensure processes are killed
> is only moving the race around, since the child may
> receive the signal in the same narrow window before assigning
> monitored_pid, and ignore it.
>
> The attached avoids the issue by blocking signal handling
> until monitored_pid is in a deterministic state wrt fork().
>
> I'll think a bit more about this before applying.

The previous patch flip flops the signal mask a couple of times.
We can simplify by expanding the critical section a little
and thus only blocking the cleanup signals once,
and continue to use sigsuspend() to reenable signals in the parent.
I'll push the attached later.

cheers,
Pádraig

0002-timeout-fix-narrow-race-in-failing-to-kill-processes.patch
Description: Text Data