OpenSSH 5.1: call for testing

[Damien Miller]

On Tue, 8 Jul 2008, Gert Doering wrote:

> Hi,
> 
> On Tue, Jul 08, 2008 at 08:37:55AM +1000, Damien Miller wrote:
> > > NetBSD 3.1 on Sparc64, CVS snapshot from "earlier today", has some 
> > > funnies in the test suite. 
> [..]
> > 
> > OpenBSD suffered from a similar problem a while back, it turned out to be
> > a kernel bug related to CMSG alignment:
> > 
> > http://www.openbsd.org/cgi-bin/cvsweb/src/sys/kern/uipc_usrreq.c.diff?r1=1.37&r2=1.38
> > http://www.openbsd.org/cgi-bin/cvsweb/src/sys/kern/uipc_usrreq.c.diff?r1=1.40&r2=1.41
> > 
> > Short story: OpenSSH's use of the CMSG_* macros used to incorrect, but
> > fixing it uncovered this kernel bug.
> 
> Hmmm. OpenBSD's change log also mentions Sparc64 - this seems to be a
> good platform for uncovering bugs...

Big endian and strict alignment :)

> Is the discussion preceding this OpenBSD kernel change online
> somewhere? I'd like to forward this to the NetBSD kernel list, but
> without a bit more background, I can't really explain what's wrong.

I think most of the discussion happened in person, or at least not by
email. The comments added in the first diff explain it a bit, but I
think the bulk of the problems occur when using the CMSG macros to pack
multiple messages under one msghdr - our previous use was causing the
alignment to be slightly wrong which made robust recovery of subsequent
messages impossible.

-d