Re: [nf-next] netfilter: Add support for inner IPv6 packet match

On Jan 25, 2018 at 08:13:00PM, Pabel wrote:

On Thu, Jan 18, 2018 at 04:13:25PM +0100, Ahmed Abdelsalam wrote:
[...]

quoted

diff --git a/include/uapi/linux/netfilter_ipv6/ip6t_inner6.h b/include/uapi/linux/netfilter_ipv6/ip6t_inner6.h
new file mode 100644
index 0000000..7017fa4
--- /dev/null
+++ b/include/uapi/linux/netfilter_ipv6/ip6t_inner6.h

@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+#ifndef _IP6T_INNER6_H
+#define _IP6T_INNER6_H
+
+#include <linux/types.h>
+#include <linux/netfilter.h>
+
+/* Values for "invflags" field in struct ip6t_inner6 */
+#define IP6T_INNER6_INV_SRC 0x01
+#define IP6T_INNER6_INV_DST 0x02
+#define IP6T_INNER6_INV_MASK 0x03
+
+struct ip6t_inner6 {
+ /* Source and destination addr of inner IPv6 packet */
+ struct in6_addr inner_src, inner_dst;
+ /* Mask for src and dest addr of inner IPv6 packet */
+ struct in6_addr inner_smsk, inner_dmsk;
+ __u8 invflags;
+};

Matching at inner headers is a very useful, no doubt. Problem is that
this approach is rather limited since it only allows for matching
source and destination address at the inner header. I suspect someone
else will follow up later on to add more fields to this, and we will
end up having a new version of ip6tables... inside ip6t_inner6 :-).

nf_tables is a much more flexible framework, we can store the offset
of this inner header in nft_pktinfo on demand, add new base to
nft_payload and have access to all matching capabilities from any
arbitrary offset. I really think this new feature belongs there.

Ahmed, have you had a chance to follow-up on the nftables approach
suggested here?

We're looking to move some of our backbone network to SRv6, but
require nftables support for inner packets.

I can potentially help with implementation, but don't want to
duplicate any more recent work of yours.

Pablo, what syntax were you envisioning for nft match expressions?
Something like:

   <outer packet match expressions> inner <inner packet match expressions>

Nested inner packets could be handled by allowing repeated "inner
<expr>" clauses:

<outer expr> inner <middle expr> inner <inner expr>

etc.

Curious for your thoughts, since this support could be useful with a
variety of tunnel protocols, not just IPv6 segment routing.

Thanks,
David