selftests: `mptcp_connect.sh` sometimes fails when validating SYN cookies

[matttbe]

It is not that often but reported at least one or two times per week by the public CI.

Here is an example:

# ns1 MPTCP -> ns2 (10.0.1.2:10006      ) MPTCP	(duration   730ms) [ OK ]
# ns1 MPTCP -> ns2 (dead:beef:1::2:10007) MPTCP	write: Connection reset by peer
# copyfd_io_poll: poll timed out (events: POLLIN 1, POLLOUT 4)
# (duration 30675ms) [ FAIL ] client exit code 111, server 0
# 
# netns ns2-6386f927-RqxpSD socket stat for 10007:
# Netid State Recv-Q Send-Q Local Address:Port Peer Address:PortProcess
# TcpPassiveOpens                 1                  0.0
# TcpEstabResets                  1                  0.0
# TcpInSegs                       7                  0.0
# TcpOutSegs                      8                  0.0
# TcpOutRsts                      1                  0.0
# TcpExtSyncookiesSent            1                  0.0
# TcpExtSyncookiesRecv            2                  0.0
# TcpExtSyncookiesFailed          1                  0.0
# TcpExtTCPPureAcks               1                  0.0
# TcpExtTCPReqQFullDoCookies      1                  0.0
# TcpExtTCPOrigDataSent           6                  0.0
# MPTcpExtMPCapableSYNRX          1                  0.0
# MPTcpExtMPCapableACKRX          1                  0.0
# 
# netns ns1-6386f927-RqxpSD socket stat for 10007:
# Netid State Recv-Q Send-Q Local Address:Port Peer Address:PortProcess
# TcpActiveOpens                  1                  0.0
# TcpEstabResets                  1                  0.0
# TcpInSegs                       5                  0.0
# TcpOutSegs                      11                 0.0
# TcpOutRsts                      3                  0.0
# TcpExtTCPOrigDataSent           6                  0.0
# TcpExtTCPDelivered              1                  0.0
# MPTcpExtMPCapableSYNTX          1                  0.0
# MPTcpExtMPCapableSYNACKRX       1                  0.0
# 
# ns1 MPTCP -> ns2 (10.0.2.1:10008      ) MPTCP	(duration   691ms) [ OK ]
# ns1 MPTCP -> ns2 (dead:beef:2::1:10009) MPTCP	(duration   706ms) [ OK ]

From:

• https://api.cirrus-ci.com/v1/artifact/task/4941721188958208/summary/summary.txt
• https://cirrus-ci.com/task/4941721188958208

We briefly discussed about that at the meeting last week and it could be eventually fixed by this series: https://lore.kernel.org/all/20221130094448.4119946-1-matthieu.baerts@tessares.net/ → We just got it in IPv4 with the patches applied so it is not linked.
Still, it is good to keep track of this issue.

[matttbe]

I just got an issue but with IPv4 this time:

# ns1 TCP   -> ns1 (dead:beef:1::1:10005) MPTCP	(duration   665ms) [ OK ]
# ns1 MPTCP -> ns2 (10.0.1.2:10006      ) MPTCP	write: Connection reset by peer
# copyfd_io_poll: poll timed out (events: POLLIN 1, POLLOUT 4)
# (duration 30647ms) [ FAIL ] client exit code 111, server 0
# 
# netns ns2-638f4061-faG9Gz socket stat for 10006:
# Netid State Recv-Q Send-Q Local Address:Port Peer Address:PortProcess
# TcpPassiveOpens                 1                  0.0
# TcpEstabResets                  1                  0.0
# TcpInSegs                       7                  0.0
# TcpOutSegs                      7                  0.0
# TcpOutRsts                      1                  0.0
# TcpExtSyncookiesSent            1                  0.0
# TcpExtSyncookiesRecv            2                  0.0
# TcpExtSyncookiesFailed          1                  0.0
# TcpExtTCPPureAcks               1                  0.0
# TcpExtTCPReqQFullDoCookies      1                  0.0
# TcpExtTCPOrigDataSent           5                  0.0
# MPTcpExtMPCapableSYNRX          1                  0.0
# MPTcpExtMPCapableACKRX          1                  0.0
# 
# netns ns1-638f4061-faG9Gz socket stat for 10006:
# Netid State Recv-Q Send-Q Local Address:Port Peer Address:PortProcess
# TcpActiveOpens                  1                  0.0
# TcpEstabResets                  1                  0.0
# TcpInSegs                       5                  0.0
# TcpOutSegs                      11                 0.0
# TcpOutRsts                      3                  0.0
# TcpExtTCPOrigDataSent           6                  0.0
# TcpExtTCPDelivered              1                  0.0
# MPTcpExtMPCapableSYNTX          1                  0.0
# MPTcpExtMPCapableSYNACKRX       1                  0.0
# 
# ns1 MPTCP -> ns2 (dead:beef:1::2:10007) MPTCP	(duration  1345ms) [ OK ]

With a debug kernel (on top of export-net):

• Task - Summary

[matttbe]

Notes from the last meeting:

• still appearing in the public CI (mostly IPv6)
• seems difficult to reproduce
• might help to stress the host and lower prio of the VM
• maybe a sysctl to reduce the timeout → didn't find any
• • add losses? → maybe

[matttbe]

We didn't see this issue for a while. I suggest to close this ticket. We can re-open it later if needed.