Re: [PATCH] cls_flow: Add tunnel support to the flow classifier

On Mon, 2011-10-17 at 08:40 +0200, Eric Dumazet wrote:

Le dimanche 16 octobre 2011 à 19:06 -0400, Dan Siemon a écrit :

quoted

When used on an interface carrying tunneled traffic the flow classifier
can't look into the tunnels so all of the traffic within the tunnel is
treated as a single flow. This does not allow any type of intelligent
queuing to occur. This patch adds new keys to the flow classifier which
look inside the tunnel. Presently IP-IP, IP-IPv6, IPv6-IPv6 and IPv6-IP
tunnels are supported.

If you are interested I have posted some background and experimental
results at:
http://www.coverfire.com/archives/2011/10/16/making-the-linux-flow-classifier-tunnel-aware/

The related iproute2 patch can be found at the above URL as well.

Signed-off-by: Dan Siemon <redacted>

Hi Dan

You're adding a lot of code (omitting the diffstat :( ) for a specific
usage, yet GRE tunnels are not supported.

Thanks for the review.

Are you arguing this use case isn't worth addressing or that there is a
more efficient way to implement this with less code?

IPv6 part is also a bit limited : It assumes TCP/UDP headers are the
first ones. Maybe its time to use ipv6_skip_exthdr() ?

I noticed this too but the existing src-proto and dst-proto don't handle
this case either. Maybe I can look into fixing those as well.

Note also that if we pull (with pskb_network_may_pull()) too many bytes,
we kill routing performance on paged frags devices, wich are now
becoming very common.

I don't know what paged frag devices means but I trust you are correct :)

The existing keys also use pskb_network_may_pull(). Should they be changed as well?

Adding tunnel support and deep packet inspection might require the use
of skb_header_pointer() wich does the copy of needed data without
requiring expensive reallocation of skb head.

I'll look into this but it may be a while before I have an updated
patch.