Re: [PATCHv3] ipv6: Enable enough of the code to handle GSO when disabled.

On 10/17/2012 12:13 PM, Eric Dumazet wrote:

On Wed, 2012-10-17 at 11:46 -0400, Vlad Yasevich wrote:

quoted

This patch attempts to solve this by enabling just enough code so GSO
is correctly processed.  However, I should point out that if IPv6 is
simply blacklisted or not built for the kernel, this problem will
still persist.

So I guess this should be done in a different way ?

We currently use a single structure (struct packet_type) to hold
pointers to different methods. (The .func() field, and the gso/gro
stuff)

We probably need to split it in two parts, and make one part linked into
kernel, even if CONFIG_IPV6=n, so that GRO/GSO is fully IPv4/IPv6
functional.

The thing about this approach is that if there are any other protocols that could have to provide their own segmentation functionality, such functionality would always have to be part of the kernel.  I wasn't sure how much I liked that.

By the way, do we really need a hash table for this ?
It seems we only have IPv4 (ETH_P_IP) and IPv6 (ETH_P_IPV6) to take care
of ?

Which hash are you talking about?  I didn't add any hashes.

This would remove some tests we currently have in GRO stack, and some
RCU stuff as well.
(GRO is slower if we have many af_packet sockets)

list_for_each_entry_rcu(ptype, head, list) {
     if (ptype->type != type || ptype->dev || !ptype->gro_receive)
         continue;
...

Adding a hook/test in ipv6_rcv() is ugly.

There already is one for device.  I original had a stub ipv6_rcv that just dropped the packet.  I could resurrect that if that's better...

-vlad

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html