On April 29, 2018 12:45:26 PM GMT+02:00, "Toke Høiland-Jørgensen" <toke@toke.dk> wrote:

On 29 April 2018 03:01:55 CEST, Ryan Mounce <ryan@mounce.com.au> wrote:
On 29 April 2018 at 07:19, Toke Høiland-Jørgensen <toke@toke.dk> wrote:
 Think I finally managed to fix the ACK filter so it works for both
6in4
 and regular v6 (the latter seems to have been broken since the commit
 "e6b72c2 ack_filter: make less aggressive by default" which
accidentally
 reversed the address compare logic). It should also be safe against
 weirdly fragmented packets (ha! famous last words!).

 If someone wants to either eyeball the code and/or test the ACK
filter
 please do. I'm planning to resubmit the patch to netdev sometime
 tomorrow evening :)

Eyeballing now, only a couple of comments.

- Initialisation of iph and ipv6h can be moved before the while()
loop. The compiler probably does this anyway.

Yeah, figured I'd let the compiler deal with that...

- Whether it is worth the complexity of restoring the
skb->encapsulation check within cake_get_tcphdr to handle other cases
of encapsulated v4/v6 traffic. I may find some time to test this later
today.

The reason I removed the check for skb->encapsulation was that in my tests it wasn't actually set for 6in4 traffic.

And looking at the code that touches it, that field seems to mostly be related to offloads? Or did you actually see any effect of using it when you added it initially? :P

Everything else looks good to me, much cleaner.

Awesome, thanks for taking a look!

Thank you so much for your efforts so far in getting cake upstream,
and in particular to rescue ACK filtering. We may have to send you a
real cake!

Haha, you're very welcome. Real cake will be appreciated, but don't feel any obligations ;)

-Toke

Cake mailing list
Cake@lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/cake