From: Gang Yan <yangang@kylinos.cn>

When we test the mptcp_join.sh, we find a bug. The 'signal invalid
addresses timeout' will be FAIL when run in real test environment,
but PASS in virtme. That's mainly because the status of invalid_addr's
tcp connection is different. It is TCPF_SYNC_SENT in virtme, but
TCPF_CLOSE in another cause the bug of 'add_addr_accepted'.

But in virtme, we can reproduce it through an invalid address
in the same LAN, this also result in an error in TCP level. These
patches will fix the issue.

Gang Yan (3):
  mptcp: fix invalid addr occupy 'add_addr_accepted'
  selftests: mptcp: relax conditions in SYN_TX validation
  selftests: mptcp: add a testcase for invalid addr timeout

 net/mptcp/pm_netlink.c                        | 25 +++++++++++++++++++
 net/mptcp/protocol.c                          |  3 +++
 net/mptcp/protocol.h                          |  2 ++
 .../testing/selftests/net/mptcp/mptcp_join.sh | 23 ++++++++++++++++-
 4 files changed, 52 insertions(+), 1 deletion(-)

-- 
2.25.1

Hi Gang,

Thank you for the new version.

I have some questions here and in the individual patches:

On 09/01/2025 14:31, Gang Yan wrote:
> From: Gang Yan <yangang@kylinos.cn>
> 
> When we test the mptcp_join.sh, we find a bug. The 'signal invalid
> addresses timeout' will be FAIL when run in real test environment,
> but PASS in virtme. That's mainly because the status of invalid_addr's
> tcp connection is different. It is TCPF_SYNC_SENT in virtme, but
> TCPF_CLOSE in another cause the bug of 'add_addr_accepted'.

I still don't understand this. Is it because a test uses some IPs that
do a different behaviour? But the tests will run inside a closed netns.
Or because a kernel option is different? Or different sysctl? Or
something else?

> But in virtme, we can reproduce it through an invalid address
> in the same LAN, this also result in an error in TCP level. These
> patches will fix the issue.
> 
> Gang Yan (3):
>   mptcp: fix invalid addr occupy 'add_addr_accepted'
>   selftests: mptcp: relax conditions in SYN_TX validation
>   selftests: mptcp: add a testcase for invalid addr timeout
> 
>  net/mptcp/pm_netlink.c                        | 25 +++++++++++++++++++
>  net/mptcp/protocol.c                          |  3 +++
>  net/mptcp/protocol.h                          |  2 ++
>  .../testing/selftests/net/mptcp/mptcp_join.sh | 23 ++++++++++++++++-
>  4 files changed, 52 insertions(+), 1 deletion(-)
> 

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.

Hi Gang,

On 09/01/2025 14:31, Gang Yan wrote:
> From: Gang Yan <yangang@kylinos.cn>
> 
> When we test the mptcp_join.sh, we find a bug. The 'signal invalid
> addresses timeout' will be FAIL when run in real test environment,
> but PASS in virtme. That's mainly because the status of invalid_addr's
> tcp connection is different. It is TCPF_SYNC_SENT in virtme, but
> TCPF_CLOSE in another cause the bug of 'add_addr_accepted'.
> 
> But in virtme, we can reproduce it through an invalid address
> in the same LAN, this also result in an error in TCP level. These
> patches will fix the issue.
> 
> Gang Yan (3):
>   mptcp: fix invalid addr occupy 'add_addr_accepted'
>   selftests: mptcp: relax conditions in SYN_TX validation
>   selftests: mptcp: add a testcase for invalid addr timeout

I'm not sure how you used 'git send-email', but the 4 emails from your
series are not "attached" to each others. Did you send the patches one
by one instead of selecting them all?

Please next time make sure all patches references the cover-letter, not
to have different threads per patch, otherwise the CI, and many tools,
cannot apply your patches:

https://patchew.org/MPTCP/tencent._5F3C9419313E4EE06582D82709B4EA32A2D009@qq.com/

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.