[LU-7096] Unprotected critical section in nrs_policy_stop

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.8.0
• Affects Version/s: Lustre 2.8.0
• Labels: cea patch
• Severity: 3
• Assignee: Emoly Liu
• Reporter: Henri Doreau (Inactive)

Description

The execution flow to stopping a NRS policy is not safe, and concurrent execution of a policy' stop function can happen, leading to instant crashes.

For instance:
lctl set_param ost.OSS.ost_io.nrs_policies=orr
lctl set_param ost.OSS.ost_io.nrs_policies=fifo & sleep 0.1 && \
lctl set_param ost.OSS.ost_io.nrs_orr_quantum=32

leads to a crash in:
[...]
#6 [ffff880098c99cc0] kmem_cache_destroy at ffffffff811772bc
#7 [ffff880098c99ce0] nrs_orr_stop at ffffffffa139d8a8 [ptlrpc]
#8 [ffff880098c99d00] nrs_policy_stop0 at ffffffffa13933cf [ptlrpc]
#9 [ffff880098c99d20] nrs_policy_stop_primary at ffffffffa1393938 [ptlrpc]
#10 [ffff880098c99d40] nrs_policy_start_locked at ffffffffa13940e0 [ptlrpc]
#11 [ffff880098c99d90] nrs_policy_ctl at ffffffffa139452c [ptlrpc]
#12 [ffff880098c99de0] ptlrpc_nrs_policy_control at ffffffffa1396092 [ptlrpc]
[...]

Issue Links

• is related to: LU-7072 sanityn test_78: Expected set_param to return 0 or EAGAIN