[LU-5664] assertion in failure handling of LNetNIInit

• Type: Bug
• Resolution: Duplicate
• Priority: Minor
• Severity: 3
• Assignee: WC Triage
• Reporter: Liang Zhen (Inactive)

Description

I hit this in my testing, it seems like failure handling of LNetNIInit is not correct, for example, if we have initialised some NIs before the failure, then we should finalise those initialised NIs before calling lnet_unprepare

LNetError: 2843:0:(api-ni.c:1505:lnet_startup_lndnis()) Can't load LND tcp, module ksocklnd, rc=256
LNetError: 2843:0:(api-ni.c:823:lnet_unprepare()) ASSERTION( list_empty(&the_lnet.ln_nis) ) failed: 
LNetError: 2843:0:(api-ni.c:823:lnet_unprepare()) LBUG
Kernel panic - not syncing: LBUG
Pid: 2843, comm: insmod Tainted: P           ---------------    2.6.32.431.lustre #1
Call Trace:
 [<ffffffff8152528a>] ? panic+0xa7/0x16f
 [<ffffffffa041aeeb>] ? lbug_with_loc+0x9b/0xb0 [libcfs]
 [<ffffffffa04c0d6d>] ? lnet_unprepare+0x2ad/0x320 [lnet]
 [<ffffffffa04c4998>] ? LNetNIInit+0x1f8/0x3f0 [lnet]
 [<ffffffffa052a06e>] ? srpc_startup+0x5e/0x220 [lnet_selftest]
 [<ffffffffa052f585>] ? init_module+0x215/0x500 [lnet_selftest]
 [<ffffffffa052f370>] ? init_module+0x0/0x500 [lnet_selftest]
 [<ffffffff8100204c>] ? do_one_initcall+0x3c/0x1d0
 [<ffffffff810bc511>] ? sys_init_module+0xe1/0x250
 [<ffffffff8100b072>] ? system_call_fastpath+0x16/0x1b

Issue Links

• is duplicated by: LU-5568 kernel crash when when network initialization failed

Comments

[Liang Zhen (Inactive)]

thanks Amir!

[Amir Shehata (Inactive)]

I believe this is a duplicate of LU-5568

There is already a patch to fix this issue:
http://review.whamcloud.com/#/c/11718/