[teamd] Porting teamd patch which solves the race condition between RTM_NEWLINK and port add

[yxieca]

- What I did

Porting teamd patch which solves the race condition between RTM_NEWLINK and port add.

Removed previous patches from us: 0006, 0008, 0009.

Signed-off-by: Ying Xie ying.xie@microsoft.com

- How to verify it
continuous warm-reboot.

[yxieca]

@jipanyang here is the RCA you asked for when I checked in patch 0009. I think patch 0006, 0008, 0009 and 0011 incrementally solved the issue and this should be the last piece of the puzzle.

[jipanyang]

I have not gone through code flow thoroughly. Likely I missed some key points.

When adding member, there is processing to fetch the member attribute via netlink synchronously (team_refresh() ? All the member attribute info should be available at kernel at that time based on the upper layer dependency control.

Why need to have dependency on RTM_NEWLINK event? Would the previous patches be unnecessary but just adding some complexity, if team_refresh() change here works?

[yxieca]

I think they are fixing timing issue too. My previous 2 patches are fixing the timing issue with sequence of (1) create port, (2) RMT_NEWLINK but initial update failed. This last change was fixing the time sequence of (1) RTM_NEWLINK but initial update failed, (2) create port.

[pavel-shirshov]

Have you checked this patch?
jpirko/libteam@45912de

[yxieca]

@pavel-shirshov The change you identified has passed 200-ish iterations of tests where I removed patch 0006, 0007, 0008 and the patch in this PR. It looks like this change could be the fix. Thanks for looking into it!