Re: [PATCH net-next] net: Fix vti use case with oif in dst lookups for IPv6

[Hajime Tazaki <thehajime@gmail.com>]

At Sun, 11 Oct 2015 07:31:32 -0600,
David Ahern wrote:

> >> The attached patch applied to Linus' tree works for me. Currently the
> >> above change is not in his tree, so I added it to this patch. Once you
> >> confirm that it works for you I'll create the delta-patch for net and
> >> send out.
> >
> > I gave it a try but without any luck unfortunately.
> > I may need to look carefully what mip6 does here.
> >
> > the code path where I'm looking at for MH packet (raw socket
> > with IPPROTO_MH) is:
> >
> > #0  ip6_route_output (net=0x7ffff3a40a40 <rumpns_init_net>, sk=0x7ffff30b5c50, fl6=0x7ffff0ed2fe0) at net/ipv6/route.c:1195
> > #1  0x00007ffff35d155f in ip6_dst_lookup_tail (net=0x7ffff3a40a40 <rumpns_init_net>, sk=0x7ffff30b5c50, dst=0x7ffff0ed2f18, fl6=0x7ffff0ed2fe0)
> >      at net/ipv6/ip6_output.c:929
> > #2  0x00007ffff35d1707 in ip6_dst_lookup_flow (sk=0x7ffff30b5c50, fl6=0x7ffff0ed2fe0, final_dst=0x0) at net/ipv6/ip6_output.c:1024
> > #3  0x00007ffff36199f7 in rawv6_sendmsg (sk=0x7ffff30b5c50, msg=0x7ffff0ed3320, len=32) at net/ipv6/raw.c:872
> > #4  0x00007ffff3526d21 in inet_sendmsg (sock=0x7ffff30b5810, msg=0x7ffff0ed3320, size=32) at net/ipv4/af_inet.c:737
> > #5  0x00007ffff338dc8a in sock_sendmsg_nosec (msg=0x7ffff0ed3320, sock=0x7ffff30b5810) at net/socket.c:610
> > #6  sock_sendmsg (sock=0x7ffff30b5810, msg=0x7ffff0ed3320) at net/socket.c:620
> >
> >
> > which (*dst)->error of ip6_route_output gives -22 (EINVAL).
> >
> > I will be back once I got any findings.
> 
> How does a xfrm change affect this code path?

I've faced this issue since the following patch was applied.

commit 741a11d9e4103a8e1c590ef1280143fe654e4e33
Author: David Ahern <dsa@cumulusnetworks.com>
Date:   Mon Sep 28 10:12:13 2015 -0700

    net: ipv6: Add RT6_LOOKUP_F_IFACE flag if oif is set

I still couldn't spot which part (other than my posted call
graph) is broken and am not sure whether the xfrm change
affects or not (which I need to check the mip6 code again).

> Can you apply this patch, and then run:
> 
> perf record -e fib6:* -a -g
> perf script

I'm using libos environment right now, so the perf trace
can't be used as it is.

I'll keep post here.

-- Hajime