[LU-16393] o2iblnd: connections rejected before lnd startup is complete

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Serguei Smirnov
• Reporter: Serguei Smirnov

Description

Before lnd startup is complete, there's a window of time when o2iblnd can reject connection requests similar to the following:

 Nov 16 08:24:18 ai400x2vm-008 kernel: LNetError: 7758:0:(o2iblnd_cb.c:2480:kiblnd_passive_connect()) Can't accept conn from 172.16.16.12@o2ib on NA (ib0:0:172.16.0.192): bad dst nid 172.16.0.192@o2ib
Nov 16 08:24:19 ai400x2vm-008 kernel: LNetError: 7758:0:(o2iblnd_cb.c:2480:kiblnd_passive_connect()) Can't accept conn from 172.16.16.187@o2ib on NA (ib0:0:172.16.0.192): bad dst nid 172.16.0.192@o2ib
Nov 16 08:24:19 ai400x2vm-008 kernel: LNetError: 7758:0:(o2iblnd_cb.c:2480:kiblnd_passive_connect()) Skipped 54 previous similar messages
Nov 16 08:24:19 ai400x2vm-008 kernel: LNet: Added LNI 172.16.0.192@o2ib [32/5120/0/180]
Nov 16 08:24:19 ai400x2vm-008 kernel: LNet: Using FastReg for registration
Nov 16 08:24:20 ai400x2vm-008 kernel: LNetError: 7758:0:(o2iblnd_cb.c:2480:kiblnd_passive_connect()) Can't accept conn from 172.16.0.58@o2ib on NA (ib0:1:172.16.0.192): bad dst nid 172.16.0.192@o2ib
Nov 16 08:24:20 ai400x2vm-008 kernel: LNetError: 7758:0:(o2iblnd_cb.c:2480:kiblnd_passive_connect()) Skipped 180 previous similar messages
Nov 16 08:24:20 ai400x2vm-008 kernel: LNet: Added LNI 172.16.16.192@o2ib [32/5120/0/180]

Look into getting rid of this race condition.

Issue Links

• is related to: LU-17071 o2iblnd: Oops caused by IBLND_REJECT_EARLY code

Comments

[Peter Jones]

Landed for 2.16

[Gerrit Updater]

"Oleg Drokin <[e-mail]>" merged in patch https://review.whamcloud.com/c/fs/lustre-release/+/51651/
Subject: LU-16393 o2iblnd: add IBLND_REJECT_EARLY reject reason
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 673ff86a84ad5d11cde24aa7411c45385ad1c633

[Gerrit Updater]

"Serguei Smirnov <[e-mail]>" uploaded a new patch: https://review.whamcloud.com/c/fs/lustre-release/+/51651
Subject: LU-16393 o2iblnd: add IBLND_REJECT_EARLY reject reason
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 93fe169eef88e8ab31acd01b8c5b3084f1de93ad

[Serguei Smirnov]

Hi Nathan,

Yes, my understanding is that this is correct.

[Nathan Dauchy]

Is it correct that a client with a rejected connection (during this race window on a server) would report an error message like the following?

LNetError: 353407:0:(o2iblnd_cb.c:2951:kiblnd_rejected()) 192.168.23.45@o2ib rejected: o2iblnd fatal error