Skip to main content
Last updated: 2026-11-12 CVE-2026-98432 is a use-after-free in the Linux kernel’s IPv4 TCP code. A peer that can reach a listening IPv4 TCP socket can trigger it before the application sees the connection. The kernel CNA rates it CVSS 3.1 9.8. A remote root exploit against a server with a Multipath TCP (MPTCP) listener has been demonstrated but not published. The Go team notes that the underlying bug is not specific to MPTCP. We are not aware of reports of exploitation in the wild. The bug is in the host’s Linux kernel, which comes from AWS, Google or an image you choose, not from Aspect Workflows. Container images run on the host kernel and need no changes. Every Linux host in an Aspect Workflows 5.17 through 6.1 deployment on AWS or GCP runs an affected kernel until it is patched, including CI runners. Windows hosts are not affected. Public traffic reaches Workflows through Layer 7 load balancers, so the internet can’t reach a Workflows host’s TCP stack directly. The risk comes from code running inside your network, such as a malicious CI job from a fork pull request or a poisoned dependency.

What we recommend

  • Limit which sources in your network can reach the remote cache, BES and Postgres hosts. CI runners are the main source of untrusted traffic.
  • On AWS ECS, re-apply your deployment now that AWS has published a patched Amazon Linux 2023 ECS-optimized AMI (release 20261031), then replace the remote cache and executor instances yourself. Their Auto Scaling groups don’t refresh instances. To patch sooner, set remote.storage.instance_image_id for remote cache hosts or ec2.instance_image for each executor worker to an AMI you choose.
  • The Ivy BEP hosts are replaced by instance refresh when you re-apply.
  • The Ivy Postgres host is replaced with a patched AMI when you upgrade to 6.0.48 or 6.1.9 and apply. Earlier releases from 6.0.40 pin an AMI that predates the fix, so upgrade or set bes.db.ami_id to a patched AMI. On 6.0.0 through 6.0.39 the host is replaced when you re-apply. Replacing it causes downtime, so schedule it.
  • On GKE, upgrade to 6.0.48 or 6.1.9 and apply. The apply upgrades the control plane in place to 1.35.9-gke.1505000, the fixed 1.35 version in Google’s bulletin GCP-2026-078.
  • From 6.1.10 the cluster is on a GKE release channel, and GKE upgrades the control plane on the channel’s schedule instead of on apply. Confirm it runs a fixed version, or upgrade it yourself.
  • On GKE, the remote cache and executor node pools don’t auto-upgrade. After the apply, upgrade them by hand.
  • Other GKE node pools follow Google’s upgrades. Confirm they run the patched image and replace any nodes that don’t.
  • As a stopgap on hosts you can’t patch yet, Canonical’s CVE page gives an nftables rule that drops incoming IPv4 TCP packets carrying IP options. Canonical says it “reduces the risk but does not fully prevent exploitation”. It breaks connections that depend on IPv4 options, and containers with their own network namespace need the rule applied separately.
  • Turning off MPTCP with net.mptcp.enabled=0 on hosts and GODEBUG=multipathtcp=0 on Go services blocks only the demonstrated exploit. Canonical says it is not an appropriate mitigation because the bug isn’t specific to MPTCP, and the Go team calls the GODEBUG route fragile.

Upgrade GKE node pools by hand

Upgrade the control plane to a fixed version first, then upgrade each pool to match it. Remote cache pools are named remote-cache-0, remote-cache-1 and so on, one per cache shard.
Each upgrade replaces the pool’s nodes and empties their local SSD cache, so do it outside peak CI hours. The Terraform module doesn’t set a node version, so later plans show no change.

CI runner images

  • You choose the CI runner image through the runner image_id setting. We recommend you check your CI runner AMIs and GCP images and rebuild them on a fixed kernel per your company policies.
  • Aspect-provided Amazon Linux 2, Amazon Linux 2023, Debian 12 and Debian 13 starter images have been patched. Images dated 20261102-0 or later carry the fix, on AWS and, for Debian, on GCP. See workflows-images and Starter machine images.
  • Aspect-provided Ubuntu 22.04, 24.04 and 26.04 starter images, including those dated 20261021-0, are still affected. Until we publish patched images, rebuild your own Ubuntu images on a fixed kernel, or use the network restrictions and stopgaps above on Ubuntu runners.
  • On AWS, Canonical’s Ubuntu 24.04 and 26.04 AMIs still boot a vulnerable kernel as of 2026-11-12, so upgrade the kernel and reboot when you build from them. Canonical’s Ubuntu 22.04 AMI and its GCP images for 22.04, 24.04 and 26.04 boot fixed kernels.
  • Debian 12: fixed in linux 6.1.190-1 (DLA-4841-1) and linux-6.12 6.12.113-1~deb12u1 (DLA-4850-1).
  • Debian 13: fixed in linux 6.12.113-1 (DSA-6568-1).
  • Ubuntu 22.04: fixed in linux 5.15.0-203.213, linux-aws 5.15.0-1120.127, linux-gcp 5.15.0-1121.131, linux-aws-6.8 6.8.0-1069.7422.04.1 and linux-gcp-6.8 6.8.0-1073.8122.04.1.
  • Ubuntu 24.04: fixed in linux 6.8.0-149.149, linux-aws 6.8.0-1069.74, linux-gcp 6.8.0-1073.81, linux-aws-7.0 7.0.0-1017.1824.04.1 and linux-gcp-7.0 7.0.0-1017.1924.04.1.
  • Ubuntu 26.04: fixed in linux 7.0.0-43.43, linux-aws 7.0.0-1017.18 and linux-gcp 7.0.0-1017.19.

Vendor fixes

As of 2026-11-12:
  • Amazon Linux 2 and Amazon Linux 2023: fixed. Use Amazon Linux 2023 release 2023.12.20261102 or later, or ECS-optimized AMI release 20261031 or later, and reboot after a kernel update.
  • Ubuntu 22.04, 24.04 and 26.04: fixed in the security notices Canonical published on 2026-11-10 (USN-9018-1 through USN-9027-1, and USN-9029-1). Use the kernel versions listed under CI runner images.
  • GKE: fixed. Bulletin GCP-2026-078 sets 1.35.9-gke.1505000 as the minimum fixed 1.35 version for the control plane and nodes.

References