• TransportProtocols
• 09May2019

Hangout 09 May 2019

Old Business

• IPv6 Fragment reviews which need testing:

  • Eliminate KAME circular queues: D16847.

  • Drop packets with overlapping fragments: D16850.

• IPv6 modernization project: Wiki Page

  • Julius proposed importing OpenBSD fragmentation regression test cases (see D17376). thj@ is working on it.

    • Stalled; switched to fragment tests; will restart once basic fragment tests (next item) are imported.

  • RFC 7112 D16851.

  • IP6 IDs (and IP IDs, maybe?): may use an insecure algorithm. Ideally, switch to some sort of per-host-pair hash?
    • What do OpenBSD/NetBSD do?
      • tuexen: OpenBSD does skip32. We use a linear congruential algorithm. NetBSD uses the same algorithm we do. jtl@ says he found an academic paper (analyzing ID-generation functions in another context) which says our function is insecure, but he needs to re-find it.
    • Perhaps, a small cache would be sufficient?

      • RFC 7739 provides three different options to consider.

• jtl@ to send patch to Richard for burst mitigation in the default stack. An idea is in D19047.

  • Still need to do.

• D18811: standardize on snd_recover (vs. sack_newdata). Reviewer: tuexen@, lstewart@.

  • @tuexen will review again.

• D18624: RFC 6675 support. Actual functionality in D18985.

  • Helps when HW LRO is enabled.

  • Also, this enables a PRR patch (based on Lars work). D18892. Needs review.

• CUBIC: Only invoking NewReno after_idle, not resetting ssthreash, not resetting epoch time: D18982, D18954, and D19118. Including another issue, this will be discussed at the IETF.

  • lstewart@ is the best person to review this.

• D19000 deal with a bug that the 1 byte occupied in the sequence number space by SYN segments results in a cwnd increase when acked. This bug exists only on the server side. tuexen@ and Richard are working on it.

• Richard is working on burst mitigation. An initial version (WIP) is available as D19047.

• D20028: NUMA aware INPCB

  • Committed.
• Can we lower TIME_WAIT?
  • No action needed. Administrators can modify as appropriate for their environment via sysctl.
• tun/tap interfaces: should we see outgoing packets in tcpdump?
  • Currently inconsistent. In general, yes, these should be visible.
  • There was a re-write of tun/tap interfaces in the meantime. tuexen@ will re-evaluate.

New Things

Nothing!?!?

Next Meeting

Scheduled for 23 May, 14:00 UTC.