[Bug 261291] ESX NFS4.1 client hangs, server never responds to EXCHANGE_ID/CREATE_SESSION

From: <bugzilla-noreply_at_freebsd.org>
Date: Thu, 27 Jan 2022 23:12:51 UTC

--- Comment #9 from Rick Macklem <rmacklem@FreeBSD.org> ---
I took another look at the packet trace and I do not see
any problems w.r.t. the session (sequence ops are using
different slots).

However, I noticed that the request it packet #644 seems
to be created from 4 large TCP segment fragments.
(I'm not sure what Wireshark means by TCP segment fragments,
 since each TCP segment is a "segment". Maybe just its way of
 saying the 4 TCP segments make up the two RPC requests?)

Are you using jumbo frames for a net interface that will
use jumbo mbuf clusters?
- I think you probably already know that jumbo mbuf clusters are
  an accident looking for a place to happen, due to fragmentation
  of the mbuf cluster pool.

If this is the case, try either a network interface that can do
jumbo frames without using jumbo mbuf clusters or turn off the
jumbo frames.

The lack of a reply to the requests in packet #644 is what causes
the mount to mess up and it looks like some sort of network fabric
problem, possibly tickled by having a large Write and a Read lumped
together in the same TCP segments?

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