[LU-6088] racer test_1: dir_create.sh mutex deadlock in sys_open->do_lookup

• 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: Di Wang
• Reporter: Maloo

Description

This issue was created by maloo for Andreas Dilger <[e-mail]>

This issue relates to the following test suite run: https://testing.hpdd.intel.com/test_sets/9ec8122a-9608-11e4-af28-5254006e85c2.

The sub-test test_1 failed with the following logs on the client console:

INFO: task dir_create.sh:5686 was blocked for more than 120s
Call trace:
mutex_lock+0x2b/0x50
do_lookup+0x11b/0x230
__link_path_walk+0x200/0x1000
path_walk+0x6a/0xe0
do_filp_open+0x1fa/0xd20
do_sys_open+0x69/0x140
sys_open+0x20/0x30

It looks like this is only being hit with both master client and master server (pre-2.7.0) so is very likely related to DNE striped directories and is a regression on master (possibly due to the addition of a new racer test for striped directories?). Combinations of 2.4/2.5/2.6/master client or server do not hit this problem.

It would be nice to get the LU-4712 patch http://review.whamcloud.com/9689 landed to clean up the DNE striped directory console messages, but this case doesn't have the client oops, just stuck threads.

Info required for matching: racer 1

Issue Links

• is duplicated by: LU-5936 lmv_merge_attr() and callees ignore i_blocks
• is related to: LU-6085 racer stuck on mutex_lock in ll_setattr_raw()
• is related to: LU-4712 racer test_1: oops at __d_lookup+0x8c
• is related to: LU-5285 mdt_reconstruct_setattr() calls mdt_attr_get_complex() without checking that object exists