sunrpc/svc_udp.c (IP_PKTINFO) broken
H . J . Lu
hjl@lucon.org
Sun Sep 24 00:14:00 GMT 2000
On Sun, Sep 24, 2000 at 07:25:20AM +0200, Thorsten Kukuk wrote:
> On Sat, Sep 23, H . J . Lu wrote:
>
> > On Sat, Sep 23, 2000 at 09:59:34PM +0200, Thorsten Kukuk wrote:
> > >
> > > Hi,
> > >
> > > the following patch:
> > >
> > > 2000-02-17 H.J. Lu <hjl@gnu.org>
> > >
> > > * sunrpc/rpc/svc.h (SVCXPRT): Add a "xp_pad" field at the end.
> > >
> > > * sunrpc/svc_udp.c (svcudp_bufcreate): Turn on IP_PKTINFO if
> > > available.
> > > (svcudp_recv): Set up message with the new "xp_pad" field for
> > > IP_PKTINFO.
> > > (svcudp_reply): Support IP_PKTINFO if available.
> > >
> > > breaks RPC/udp connections on SPARC. Run ypbind on SPARC and
> > > try to run ypwhich to reproduce it.
> > > Can anybody look at it who knows something about IP_PKTINFO ?
> > > Has anybody tested this on other architectures ?
> >
> > We have been using it on ia32 with glibc 2.1. Otherwise, RPC over UDP
> > won't work with IP aliases cross a firwall. I will check it out on
> > ia32 with glibc 2.2 next week.
>
> It works fine on ia32. It does not work on sparc and sparc64. I
> don't know if it is a bigendian problem or if IP_PKTINFO does not
> work on SPARC.
Can you find out why IP_PKTINFO doesn't work on sparc? As far as I
can remember, IP_PKTINFO doesn't do much. My patch only keeps what
it is received and sends it back verbatim. The only thing I can think
of is alignment.
H.J.
More information about the Libc-hacker
mailing list