On Mon, 2020-09-14 at 17:34 +0800, Geliang Tang wrote:
> Hi Paolo,
> 
> Thanks for your patch. I tested it. It fixed the memory leak problem.
> But I got this
> selftest fail output:
> 
> 11 single subflow with syn cookies      syn[ ok ] - synack[ ok ] - ack[ ok ]
> 12 multiple subflows with syn cookies   syn[fail] got 1 JOIN[s] syn expected 2
>  - synack[fail] got 1 JOIN[s] synack expected 2
>  - ack[fail] got 1 JOIN[s] ack expected 2
> Server ns stats
> MPTcpExtMPCapableSYNRX          1                  0.0
> MPTcpExtMPCapableACKRX          1                  0.0
> MPTcpExtMPJoinSynRx             1                  0.0
> MPTcpExtMPJoinAckRx             1                  0.0
> Client ns stats
> MPTcpExtMPJoinSynAckRx          1                  0.0
> 13 subflows limited by server w cookies syn[fail] got 1 JOIN[s] syn expected 2
>  - synack[fail] got 1 JOIN[s] synack expected 2
>  - ack[ ok ]
> Server ns stats
> MPTcpExtMPCapableSYNRX          1                  0.0
> MPTcpExtMPCapableACKRX          1                  0.0
> MPTcpExtMPJoinSynRx             1                  0.0
> MPTcpExtMPJoinAckRx             1                  0.0
> Client ns stats
> MPTcpExtMPJoinSynAckRx          1                  0.0
> 14 signal address with syn cookies      syn[ ok ] - synack[ ok ] - ack[ ok ]
> 15 subflow and signal w cookies         syn[ ok ] - synack[ ok ] - ack[ ok ]
> 16 subflows and signal w. cookies       syn[fail] got 1 JOIN[s] syn expected 3
>  - synack[fail] got 1 JOIN[s] synack expected 3
>  - ack[fail] got 1 JOIN[s] ack expected 3
> Server ns stats
> MPTcpExtMPCapableSYNRX          1                  0.0
> MPTcpExtMPCapableACKRX          1                  0.0
> MPTcpExtMPJoinSynRx             1                  0.0
> MPTcpExtMPJoinAckRx             1                  0.0
> Client ns stats
> MPTcpExtMPJoinSynAckRx          1                  0.0

I sometime see similar errors, but they are quite unfrequent, and looks
tied to the self-test fragility: we use some hard-coded sleep to allow
for MPJ/ADD_ADDR handshake. If the VM running the test is my change
slowed down a bit, self-tests will fail.

Is the above failure reproducible for you?

Thanks,

Paolo