Re: [PATCH] Add timeout utility

[Mike Frysinger]

On Wednesday 02 April 2008, Pádraig Brady wrote:
> Mike Frysinger wrote:
> > On Tuesday 01 April 2008, Pádraig Brady wrote:
> >> +  /* Setup handlers before fork() so that we
> >> +   * handle any signals caused by child, without races.  */
> >> +  signal (SIGALRM, cleanup);    /* our timeout.  */
> >> +  signal (SIGINT, cleanup);     /* Ctrl-C at terminal for example.  */
> >> +  signal (SIGQUIT, cleanup);    /* Ctrl-\ at terminal for example.  */
> >> +  signal (SIGTERM, cleanup);    /* if we're killed, stop monitored
> >> process.  */ +  signal (SIGHUP, cleanup);     /* terminal closed for
> >> example.  */ +  signal (SIGTTIN, SIG_IGN);    /* don't sTop if
> >> background child needs tty.  */ +  signal (SIGTTOU, SIG_IGN);    /*
> >> don't sTop if background child needs tty.  */
> >
> > if you're using signal(), you have race problems.  why not use
> > sigaction() and friends instead ?
>
> I originally wrote this for util-linux-ng.
> BSD and linux at least handle signal() sanely
> so I wrote it for those.
>
> Since coreutils needs to be more portable,
> I'll change to the more complicated sigaction instead.
>
> To be sure, are you referring to races where a signal
> can be received while in the signal handler on some systems?
>
> Also there is the issue of restarting system calls
> after the signal handler has run.
>
> Also there is the issue of resetting the signal
> handler after it has run.
>
> Are there systems that coreutils supports currently
> that doesn't follow BSD/linux semantics wrt the above 3 points?

afaik, Linux behaves as POSIX dictates.  that means it does not behave as you 
describe above.  newer versions of glibc may block the same signal from 
re-occuring, but it doesnt block other signals.
-mike

signature.asc
Description: This is a digitally signed message part.