[LU-6010] DLC: LNetFini() assert is hit if lustre_rmmod without bringing down NI

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.7.0
• Affects Version/s: Lustre 2.7.0
• Labels: HB
• Severity: 3
• Assignee: Amir Shehata (Inactive)
• Reporter: Amir Shehata (Inactive)

Description

if lnet is not brought down properly (IE: lctl network down or lnetctl lnet unconfigure), then LNetFini():LASSERT(the_lnet.ln_refcount == 0); is hit. We need to modify this assert to account for this scenario. This wasn't a problem before because a network was always loaded irregardless. and LNetNIFini() was always called to set ln_refcoutn to 0

Comments

[Peter Jones]

Landed for 2.7

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch http://review.whamcloud.com/13110/
Subject: LU-6010 lnet: prevent assert on LNet module unload
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: f1a2e6107c124d010d89973cfd716fbd17b689f0

[Isaac Huang (Inactive)]

Isn't the "Use Case 3/6" illegal uses of DLC? The user should NOT be allowed to remove the lnet module if DLC configure has been called successfully but DLC unconfigure has not been called yet.

I think the correct action for "Use Case 3/6" is to prevent the lnet module to be unloaded - i.e. to enforce the correct DLC usage in "Use Case 5". In other words, it should be handled the same way as in "Use Case 8", fail with "lnet in use error".

[Gerrit Updater]

Amir Shehata ([e-mail]) uploaded a new patch: http://review.whamcloud.com/13110
Subject: LU-6010 lnet: prevent assert on LNet module unload
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: 1262d8f1188c3b610354908a0049a8effe0af314

[Amir Shehata (Inactive)]

Here are the use cases for loading/unloading LNet, and which ones correspond to this issue:

Use Case 1
Configure LNet with no mod NI -> initializes NI
Dynamically add NIs
Unconfigure LNet -> uninitializes NI
lustre_rmmod

Use Case 2
Configure LNet with no mod NI
Unconfigure LNet
lustre_rmmod

Use Case 3
Configure LNet with no mod NI
lustre_rmmod
--> LU-6010

Use Case 4
Configure LNet with mod NI
Dynamically add NIs
Unconfigure LNet
lustre_rmmod

Use Case 5
Configure LNet with mod NI
Dynamically del all NIs
Unconfigure LNet
lustre_rmmod

Use Case 6
Configure LNet with mod NI
Dynamically del all NIs
lustre_rmmod
--> LU-6010

MODULE and LNet interaction
-> can not dynamically add NIs from a different module has to be from IOCTL

Use Case 7
Module X: Configure LNet with mod NI
Module Y: Configure LNet with mod NI
MODULE X: Unconfigure LNet
MODULE Y: Unconfigure LNet
lustre_rmmod

Use Case 8
Module X: Configure LNet with mod NI
Module Y: Configure LNet with mod NI
MODULE Y: Unconfigure LNet
lustre_rmmod
 -> lnet in use error

Use Case 9
Module X: Configure LNet with mod NI
Module Y: Configure LNet with mod NI
lustre_rmmod
 -> lnet in use error