[LU-14069] OBD_FAIL_LDLM_CANCEL_BL_CB_RACE is buggy in ldlm_handle_cp_callback

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.14.0, Lustre 2.12.6
• Severity: 3
• Assignee: Oleg Drokin
• Reporter: Oleg Drokin

Description

There's this code in ldlm_handle_cp_callback:

        if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_BL_CB_RACE)) {
                long to = cfs_time_seconds(1);

                ldlm_callback_reply(req, 0);

                while (to > 0) {
                        schedule_timeout_interruptible(to);
                        if (ldlm_is_granted(lock) ||
                            ldlm_is_destroyed(lock))
                                break;
                }
        }

This looks like it was supposed to be a time-bound wait and indeed looking at when it was introduced (commit 022b1022, bz 11300) it has the to assigned from schedule_timeout.

This got broken by commit adde80ff which is some squashed head commit and lost the to assignment.

Not this seems to be breaking LU-13692 patch

Issue Links

• is related to: LU-13692 MDS slow/hung threads at mdt_object_local_lock
• is related to: LU-11300 LNet: Router Aliveness and Health