[Cerowrt-devel] trivial 6in4 fix(?)

[Dave Taht <dave.taht@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1488 bytes --]

The 3.8.13-6 dev non-release has a trivial patch in it that I hoped would
improve 6in4 and 6to4 tunneling performance. I'm not in a position to test
at the moment.

I am curious: Is there a current user of cero using simplest.qos, tunneling
ipv6 via hurricane or 6to4, that can hammer it with multiple ipv6 streams
(as in the rrul test)? (If you have rrul and ipv6, you can force it to do
pure ipv6 testing with the -6 option.)

During that test, the output, on 3.8.13-5 and earlier, of

tc -s class show dev ge00

I *think*, will show only a few fq_codel classes in use, rather than one
per stream. (you may need to run the command a few times to identify all
the streams in use)

It looked to me that the following patch would de-encapsulate the 6in4
stuff for purposes of the fq_codel flow hash calculation, and fix it... (I
put this patch into -6 but haven't tested)

---
 net/core/flow_dissector.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/core/flow_dissector.c b/net/core/flow_dissector.c
index 00ee068..25ebce7 100644
--- a/net/core/flow_dissector.c
+++ b/net/core/flow_dissector.c
@@ -138,6 +138,8 @@ ipv6:
                }
                break;
        }
+       case IPPROTO_IPV6:
+               proto = __constant_htons(ETH_P_IPV6);
        case IPPROTO_IPIP:
                goto again;
        default:
--
1.8.1.2

-- 
Dave Täht

Fixing bufferbloat with cerowrt:
http://www.teklibre.com/cerowrt/subscribe.html

[-- Attachment #2: Type: text/html, Size: 1660 bytes --]