[Bug 292293] Accurate ECN on the loopback interface causes an infinite ACK loop.

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=292293

Richard Scheffenegger <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|[email protected]             |[email protected]

--- Comment #7 from Richard Scheffenegger <[email protected]> ---
Thanks! The binary shows the AE and ECE (ACE == 5) as expected - no congestion
event during the initial 3WHS or increasing this beyond the initial value. 

If you plan to use AccECN, you may want to update the version of tcpdump to
decode these header bits, btw.

AccECN in FBSD itself currently doesn't feature the Change-triggered ACKs, so
this effect may only incidentially be connected with enabling AccECN; (Standard
ECN would issue the sending of an immediate ACK when a CE mark is received).

Which congestion control module, and TCP stack  are you using on the loopback
interfaces / relevant jail?

> sockstat -P tcp -CS

You could try enabling blackbox loggging against the affected session, once it
gets into this state.

The following article has the details of how to selectively enable blackbox
logging on only a specific session, using tcpsso:

https://freebsdfoundation.org/adventures-in-tcp-ip-tcp-black-box-logging/

And it also describes how to extract the log into a file which can be reviewed
(the base stack in 15.0 may not have logging hooks in all the instances, where
ACKs are triggered though).

-- 
You are receiving this mail because:
You are the assignee for the bug.