Re: gzip zombie / spawned from init

[Richard B. Johnson]

On Tue, 14 Jun 2005, [iso-8859-1] Måns Rullgård wrote:

> Bart Hartgers <bart@xxxxxxxxxxxxxxxxxx> writes:
>
> > Nico Schottelius wrote:
> > > Hello!
> > > I wrote an init replacement (cinit), which is now in the beta-phase.
> > > The only problem I do have currently is that when calling
> > > 'loadkeys dvorak' directly from init (without a shell or anything)
> > > it will leave behind a gzip zombie (which was forked by loadkeys).
> > > Now my question is: Is that a problem of loadkeys or from my init
> > > and what could be the reasons that it's still there?
> >
> > Not really a kernel issue but:
> >
> > Yes and no. If a parent exits before its child, the child is
> > reparented to init. loadkeys probably doesn't wait properly for gzip
> > to finish.
> >
> > > cinit forks() loadkeys and does waitpid() for it. There is no
> > > loadkeys zombie, only gzip.
> >
> > Use waitpid(-1,...) or wait(...) to wait on all childeren in your
> > init. gzip will become a child of cinit.
>
> In fact, init must reap any zombies that are reparented to it.
> Otherwise, the system will sooner or later run out of PIDs.  There are
> a lot of misbehaving programs out there, and even if they were all
> well-behaved, they could be killed before having waited for their
> children, leaving zombies behind.
>
> --
> Måns Rullgård
> mru@xxxxxxxxxxxxx


Normally an `init` program will provide a SIGCHLD handler like:

void reaper(int sig)
{
    while(wait3(&sig, WNOHANG, NULL) > 0)
        ;
}

This will 'reap' and throw away the status of all dead children,
preventing zombies.


Cheers,
Dick Johnson
Penguin : Linux version 2.6.11.9 on an i686 machine (5537.79 BogoMips).
 Notice : All mail here is now cached for review by Dictator Bush.
                 98.36% of all statistics are fiction.