[LU-218] deadlock (pagefault vs. blocking ast) in clio

• Status: Closed
• Project: Lustre
• Affects Version/s: Lustre 2.1.0
• Fix Version/s: Lustre 2.1.0
• Type: Bug
• Priority: Major
• Reporter: Niu Yawei (Inactive)
• Assignee: Robert Read (Inactive)
• Resolution: Duplicate

Environment

2.6.18-194.17.1.el5

• Severity: 3
• Epic: client

Description

The following deadlock was hit when I was running mmap tests:

mmap test thread: pagefault -> lock page -> release dlm lock -> cancel dlm lock -> cl_lock_mutex_get;
bl_ast handler thread: cancel dlm lock -> cl_lock_mutex_get -> flush pages -> lock page;

And because of this deadlock, ll_imp_inval thread is blocked on cl_lock_mutex_get, so the client eviction can never be finished. I think it's the root cause of LU-180, but I'm not 100 percent sure because they didn't provide the stack trace to prove it yet.

The stack trace is attached.

Comments

[Niu Yawei (Inactive)]

Jay, could you take a look to see if it's a known bug or not? Thanks.

[Oleg Drokin]

I wonder if this is the same thing as LU-122?

[Niu Yawei (Inactive)]

Right, I think it's same as LU-122, and LU-180 is also probably caused by this deadlock. Will mark it as duplicated bug.

[Niu Yawei (Inactive)]

Duplicated with LU-122.