[LU-16633] obd_get_mod_rpc_slot() is vulnerable to races

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.16.0, Lustre 2.15.5
• Affects Version/s: Lustre 2.16.0
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Alex Zhuravlev

Description

wait_woken() called from obd_get_mod_rpc_slot() resets WQ_FLAG_WOKEN flag, but leaves the corresponding wq_entry on the list.
__wake_up_locked_key() (called from obd_put_mod_rpc() in another thread) finds that wq_entry on the list w/o WQ_FLAG_WOKEN set, consumes one more slot (cl_mod_rpcs_in_flight++).
this way we leak slots.
this can be catched with sanity/182b.

Issue Links

• is related to: LU-16404 sanity test_230n: lfs mirror: cannot get UNLOCK lease, Mirroring failed
• is related to: LU-17938 Client evicted during stress test
• is related to: LU-15947 Spinlock contention during wake_up_all() in obd_put_mod_rpc_slot()