Re: [PATCH 3/3] Compare namespaces when comparing addresses in auth_unix cache.

From: Rob Landley <rlandley-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>
To: Alexey Dobriyan <adobriyan-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Cc: linux-nfs-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	Kir Kolyshkin <kir-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>,
	Tim Spriggs <tims-DAzCR1AAk2Rg9hUCZPvPmw@public.gmane.org>,
	Pavel Emelyanov <xemul-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>,
	containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org,
	Trond Myklebust
	<Trond.Myklebust-HgOvQuBEEgTQT0dZR+AlfA@public.gmane.org>,
	linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
Subject: Re: [PATCH 3/3] Compare namespaces when comparing addresses in auth_unix cache.
Date: Mon, 11 Apr 2011 12:45:02 -0500	[thread overview]
Message-ID: <4DA33E1E.701__24038.5155463175$1302544055$gmane$org@parallels.com> (raw)
In-Reply-To: <BANLkTi=sh1CB+e1EhU-c-Bs3BCszWEBTMA-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>

On 04/11/2011 08:57 AM, Alexey Dobriyan wrote:
> On Mon, Apr 11, 2011 at 4:36 PM, Serge E. Hallyn <serge-A9i7LUbDfNHQT0dZR+AlfA@public.gmane.org> wrote:
> 
>> Besides use-after-free, the other concern is an invalid net_eq()
>> result due to the * being re-used for a new netns.
> 
> Exactly.
> 
> "struct net" dies last, no exceptions, anything different is a disaster.

Actually the patch turns out to be irrelevant because Pavel made the
whole thing into a pernet subsystem, so it was already fixed in a
different way.  (Commit 2f72c9b7.  My bad, I initially started working
against an NFS tree with an older kernel version, this fix was to a
different source file so my patch still applied, and I just regression
tested that it worked, not that it still failed without it.  Just caught
it now.)

I actually did talk about addressing potential reuse of the net * in my
email with Stanislav (point of the patch was to allow one cache to
maintain two different contexts at the same time; a situation that
changed the meaning of an existing cache entry by requiring the old
context to go away didn't introduce any new problems that a single
context didn't already have due to DNAT, servers moving, administrators
changing credentials on the server).  But sort of a moot point now.

I believe the third patch can be dropped entirely.

Rob