[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 5 Feb 2018 12:27:14 -0500
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: [PATCH] re-fix child reaping in wordexp
On Mon, Feb 05, 2018 at 08:07:44PM +0300, Alexander Monakov wrote:
> On Mon, 5 Feb 2018, Rich Felker wrote:
> > > In general retrying waitpid on EINTR is not robust in case pid reuse is
> > > possible, but fixing that requires changing waitpid call sites to only
> > > do that with signals blocked (where that's not already the case).
> >
> > I don't follow this. Unless there's a bug in the kernel, this should
> > not be functionally different from SA_RESTART. A return with EINTR
> > means the child was not reaped.
>
> The problem I had in mind is that you don't know if a signal handler or
> another thread had (yes, incorrectly) already reaped that child when you
> are about to retry waitpid.
>
> With signals blocked, you issue just one waitpid, and you need very rapid
> pid reuse to happen, after someone successfully reaps your child
> even before you enter waitpid.
>
> Of course this is a bit moot since the other thread/sighandler shouldn't
> be issuing wildcard waits in the first place, and if rapid pid reuse
> does not happen you safely leave the retry loop with ECHILD. But this is
> why I said "not robust" rather than "incorrect".
OK, that makes sense -- it's a matter of tiny window vs
unboundedly-large window. And in this case EINTR is not relevant; the
same unboundedly-large window can happen if you have a long-running
signal handler with SA_RESTART.
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
