Linux IPv4 TCP use-after-free
CVE-2026-98432
Red Hat says a race in the Linux kernel's IPv4 TCP code lets an unauthenticated remote attacker corrupt kernel memory. Researchers report remote root against servers with an MPTCP listener, but the exploit is not public and no exploitation is reported.
This tier is for “Internet-facing server with an MPTCP listener”. Other placements are higher or lower; see below.
Verdict by placement
Your tier depends on where it runs. Find the line that matches your setup.
- Tier 2
Internet-facing server with an MPTCP listener
Handle ahead of your normal cycle. Patch the kernel and reboot.
Why Tier 2
- Reachable from the internet.
- No login needed.
- Remote root reported by researchers against MPTCP listeners.
- Exploit not released. No exploitation reported.
- Tier 3
Internet-facing server without an MPTCP listener
Schedule the kernel update sooner than your normal cycle. Move up if an exploit that does not need MPTCP appears.
Why Tier 3
- Reachable from the internet.
- Red Hat says the flaw can lead to code execution, and the bug is not specific to MPTCP.
- No exploit has been shown or released for servers without an MPTCP listener.
- Tier 4
Internal-only server
Patch in your normal cycle.
Why Tier 4
- Not reachable from the internet.
Does this apply to you?
Do you run Linux servers that accept IPv4 TCP connections from the internet, and are their kernels unpatched?
How to checkCheck each server's kernel with uname -r against your distribution's security tracker for CVE-2026-98432. Upstream fixes are in 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3, but vendor kernels carry their own version numbers, so use the distribution's tracker. To see whether MPTCP listeners exist (the route used in the researchers' exploit), check sysctl net.mptcp.enabled and the listening services.
If you cannot tell, answer Yes.
Patch and mitigate
There are 2 ways in. Each option below says which it closes and which it leaves open.
- AIncoming IPv4 TCP connection carrying IP options (the base flaw)
- BRequest-socket migration, if enabled (a similar flaw)
Install a fixed kernel and reboot
Vendor-statedRed Hat calls kernel updates the only complete solution. Upstream fixes are in 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3. Check your distribution's tracker for its fixed kernel.
- Effort
- A kernel update and a reboot of each server. Live patching, where your distribution offers it for this flaw, can avoid the reboot.
- Might break
- A new kernel can change drivers and behaviour. Test first on a representative host.
Check it worked, undo, guides
- Check it worked
- Run
uname -rafter the reboot and confirm it matches your distribution's fixed kernel for CVE-2026-98432. - Undo
- Boot the previous kernel. This reopens the flaw.
Drop IPv4 TCP packets that carry IP options
Vendor-statedRed Hat's temporary nftables rule drops incoming IPv4 TCP packets with IP options (
ip protocol tcp ip hdrlength > 5 counter drop, in a table named cve_2026_98432). It takes effect at once for new and existing connections and needs no service restart. Red Hat says it reduces exposure but cannot fully prevent the flaw, and is not a substitute for patching. It is not kept across a reboot.- Effort
- Loading one nftables table on each server.
- Might break
- Red Hat warns it may disrupt workloads that need IP options, such as CIPSO labelling. It does not cover pod-to-pod traffic in OpenShift.
Check it worked, undo, guides
- Check it worked
- List the table with
nft list table inet cve_2026_98432and watch its drop counter. - Undo
- Delete the table, as the bulletin describes, once a fixed kernel is running.
Turn off MPTCP and request-socket migration
Reported, not yet confirmedAkamai suggests setting
net.mptcp.enabled=0andnet.ipv4.tcp_migrate_req=0. Akamai and Red Hat both say this does not stop exploitation, because the bug is not specific to MPTCP. It only blocks the route used in the researchers' exploit.- Effort
- Two sysctl changes, which can be applied at runtime.
- Might break
- Services that rely on MPTCP stop using it.
Check it worked, undo, guides
- Check it worked
- Read the two sysctl values back and confirm they are 0.
- Undo
- Set the values back and reload sysctl.
The detail behind this call
Evidence
4 items, as of 2026-11-09
- Vendor bulletin · Vendor-stated
- Red Hat rates it Critical. An unauthenticated remote attacker can exploit a race with no privileges or user interaction, and exploitation needs precise timing. MPTCP is not a precondition, and default RHEL systems, where MPTCP is off, are still affected. RHEL 8, 9 and 10 and OpenShift Container Platform 4 are listed.
- Exploit evidence · Reported, not yet confirmed
- Akamai says Anthropic reported a root exploit on 2026-10-27 and has not released it, and Akamai knew of no public working exploit when it wrote on 2026-11-02.
- Exploitation status · Reported, not yet confirmed
- BleepingComputer reported on 2026-10-30 that there were no confirmed reports of exploitation in the wild. Red Hat and Akamai do not state otherwise.
- CVE record · Reported, not yet confirmed
- Red Hat's CVE page gives a public date of 2026-10-15 and a CVSS v3 base score of 9.8; the same page shows cve.org also at 9.8 and NVD with no score. Akamai says the Linux kernel CNA published it that day. The score is shown for reference only.
Any CVSS (Common Vulnerability Scoring System, a general severity score) figure quoted is for reference. It did not set a tier.
Analysis
Some write-ups describe this as an MPTCP-only problem. Red Hat and Akamai both say the bug itself is not specific to MPTCP, while the researchers' exploit uses an MPTCP listener. The record therefore has separate rows for servers with and without one.
As of 2026-11-09 it has not been confirmed whether the flaw is in the CISA KEV catalogue.