When kern.hz is 100 timer resolution below 10ms is disgarded. You can see clearly in
the RTT measurement, on a path with 15ms delay TCP stats will report an rtt of 20ms.
It also seems to be the case that all sub 10ms delay paths report a minimum delay of
10ms. This does not present if I boot with kern.hz set to 1000.

I get inconsistent results for TXPB, you can see this with the test script here:
https://gist.github.com/adventureloop/fb68d805696aeaf03c227f08cbe87947

updated

• Use named union and struct in header definition

Fix struct so that will build, last rev I seem to have updated the review from
the wrong tree.

Arcanist tricked me and created this instead of updating an existing review

Remove extra white space around union

Remove accidentally added file

• Only reduce staging size on arm

Extend the sequence number field to include a reserved byte in the long header.

Add comment to explain the seq field is sized to hold either the short or the long sequence number

All the tests seem to work on r350568

I have been digging and I think I have all the background for why this was
removed in r87145.

Respond to review comments

Address review comments

We wrote an ID:

I think we can just use zero for the label if enough entropy is not available. A flow label of zero indicates no flow label "treatment" in the network. On my test machine most early packets are already have a flow label of zero. From RFC 6437

Looks good to me

I am okay with Jonathan's suggested plan.

Diff against correct tree

typo in the commit message stopped this being closed automatically

typo in the commit message stopped this being closed automatically

• Do pullup per protcol that requires it

• Use (ip+exthdr len) + length when pulling up mbuf

• Teach netinet/in.h about dccp
• Make sure length bytes are in the first mbuf

• Correctly locate structs in mbuf chain

• Add support for ipv4 in ipv6, gre and dccp fragments
• Unstatic definition
• address language issues in comments