On 19/01/2021 05:22, Geliang Tang wrote:
Paolo Abeni <pabeni(a)redhat.com> 于2021年1月19日周二 上午7:07写道：
> (or it's just me lacking coffee?!?)
> I just tried plain net-next at
> commit 213b97b125808aa220ba460a7af09d8f3772f669
> Merge: 20efd2c79afb 1fef8544bf41
> Author: Jakub Kicinski <kuba(a)kernel.org>
> Date: Sat Jan 16 19:06:01 2021 -0800
> Merge branch 'net-fix-the-features-flag-in-sctp_gso_segment'
> selftests mptcp_connect.sh randomly fail with data corruption: the
> amount of transferred data is correct, the fail contains some chunk of
> random data.
> Can anybody else replicate the above?
I just tested it, and got the mptcp_connect.sh failure too. I'll try to
find out which commit breaks it today, and give you the feedback.
It is failing for me too.
In fact, my CI also reported the issue but it was "hidden" by another
one we have (issue #138) :)
It looks like it was introduced recently. Build with 9ab7e76aefc9 ("GTP:
add support for flow based tunneling API") didn't have this issue but
213b97b12580 ("Merge branch
'net-fix-the-features-flag-in-sctp_gso_segment'") had that.
A quick "git bisect" launched during meetings pointed me to:
dbd50f238dec ("net: move the hsize check to the else block in skb_segment")
If I revert it on top of export, I no longer have the corruption.
Do we need to adapt MPTCP code or is there an issue with this commit?
Tessares | Belgium | Hybrid Access Solutions