CVE-2021-47132

What it actually does

Race condition in MPTCP retransmission path allows concurrent updates to sk_forward_memory without acquiring the msk socket spin lock, leading to memory field corruption.

Why it matters to the business

Unpatched systems may experience networking instability or crashes during MPTCP retransmissions, potentially disrupting services on affected AlmaLinux 8 deployments; low urgency given age and minimal observed exploitation.

How an attack unfolds

An attacker establishes MPTCP connections and forces retransmissions, triggering the unprotected code path in __mptcp_retrans that calls into __mptcp_clean_una_wakeup helpers and corrupts sk_forward_memory.

How to fix it

Upgrade fence-agents-aliyun, fence-agents-aws, fence-agents-azure-arm, and fence-agents-gce to version 4.2.1-53.el8_3.2 or later on AlmaLinux 8; apply any corresponding kernel updates that backport the MPTCP lock fix.

How to confirm the fix worked

Confirm package versions match or exceed the listed fixed releases and verify MPTCP retransmission paths now acquire the msk spin lock before updating sk_forward_memory.

Typical attack path

1. Establish MPTCP session with target
2. Induce packet loss to trigger retransmission
3. Invoke __mptcp_retrans without lock protection
4. Corrupt sk_forward_memory via unlocked updates

How to detect it

Search kernel logs and network stack traces for MPTCP retransmission events coinciding with memory corruption symptoms; no widespread public exploitation artifacts reported.

What your security team should do

• Prioritize patching the listed fence-agents packages to the specified fixed versions
• Monitor EPSS trend and kernel changelogs for any renewed exploitation signals
• Review MPTCP usage in production workloads and restrict where possible until patched