CVE-2023-54000

What it actually does

Race condition between external loopback and reset paths allows napi_disable to be invoked twice without an intervening napi_enable, causing the second caller to block indefinitely on rtnl_lock.

Why it matters to the business

Network interface resets or diagnostics can hang the system for extended periods, disrupting connectivity and requiring manual intervention or reboot; low exploitability reduces immediate risk but still affects reliability of affected network hardware.

How an attack unfolds

An attacker with local access triggers concurrent external_lb and reset operations on the hns3 device; the resulting double napi_disable call deadlocks rtnl_lock, blocking the service task and any dependent network operations.

How to fix it

Apply the upstream fix that guards napi_disable/enable calls with the HNS3_NIC_STATE_DOWN state flag, equivalent to the protection already used in ndo_stop/ndo_start; update to a kernel containing commit resolving CVE-2023-54000.

How to confirm the fix worked

Confirm the HNS3_NIC_STATE_DOWN guard is present around external_lb napi calls; test concurrent external_lb and reset sequences to verify absence of deadlock and hung-task warnings.

Typical attack path

1. Obtain local access or trigger diagnostic/reset paths on hns3 hardware
2. Simultaneously invoke external loopback and device reset operations
3. Trigger the unprotected napi_disable sequence leading to mutex acquisition failure
4. Cause prolonged task blocking and service degradation

How to detect it

Search kernel logs for hung-task timeouts exceeding 120 seconds involving kworker threads in hclge and stack traces showing mutex_lock on rtnl_lock during hclge_reset_prepare; look for repeated external_lb operations followed by reset events.

What your security team should do

• Prioritize kernel updates containing the state-guard fix on all systems using hns3 hardware
• Monitor for hung-task reports referencing hclge_service_task or rtnl_lock in logs
• Restrict unprivileged access to network diagnostic and reset interfaces until patched