Linux's October 2026 TCP Use-After-Free: CVE-2026-98432
CVE-2026-98432 is a CVSS 9.8 remote use-after-free in Linux's IPv4 TCP handshake, fixed upstream Oct 15. RHEL has fixes; Ubuntu's are still in testing.
CVE-2026-98432 is a CVSS 9.8 remote use-after-free in Linux's IPv4 TCP handshake, fixed upstream Oct 15. RHEL has fixes; Ubuntu's are still in testing.
Linux's stable kernels shipped a fix on October 11 and 15, 2026 for CVE-2026-98432 (opens in a new tab), a use-after-free in the IPv4 TCP handshake that a remote, unauthenticated peer can reach. The kernel's own CVE Numbering Authority (CNA) scores it CVSS 3.1 9.8, Critical. The fixed releases are 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10, and 7.3, and the bug goes back to Linux 4.4. The upstream fix is a four-line change, but most servers run a distribution's kernel. Red Hat, which now rates the bug Critical too, shipped fixed kernels for Red Hat Enterprise Linux (RHEL) 8, 9, and 10 on November 1, and AlmaLinux followed on November 2. Ubuntu has shipped a fixed kernel only for 26.10: as of 9:00 AM CST on November 5, its fixes for 22.04, 24.04, and 26.04 are still in testing, and its tracker reads "Vulnerable" for every release, 26.10 included.
Why it's urgent: the bug sits in code every Linux server runs when it accepts an IPv4 TCP connection, before any application sees the connection. Anthropic, whose researcher reported it, says its Claude model wrote a working proof-of-concept exploit that runs commands as root on "Linux servers with a multipath TCP listener," with no credentials. Go programs open those listeners by default, per the Go Security Team. Security researchers report what looks like threat-actor scanning for such servers. See the timeline.
| Field | Detail |
|---|---|
| CVE | CVE-2026-98432, "ipv4: free inet_opt and ireq_opt after an RCU grace period," issued by the Linux kernel CNA |
| Severity | Critical, CVSS 3.1 9.8 (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H), the kernel CNA's score. Red Hat also rates it Critical, CVSS 3.1 9.8, raised from Important (8.1) on October 30. Ubuntu rates it Critical, raised from High on November 3. SUSE rates it important, and Oracle and Amazon rate it Important with CVSS 3.1 8.1. |
| Weakness | Use-after-free of IPv4 option data during the TCP handshake (the record lists no CWE) |
| Impact | Per the CNA's scoring notes: kernel heap contents disclosed to the remote peer, kernel memory corruption, or a crash. Red Hat's bulletin: a remote, unauthenticated attacker "could exploit a race condition during TCP connection request processing to trigger remote code execution." |
| Fixed in | 7.2.10 and 6.18.56 (Oct 11) · 6.12.113, 6.6.159, 6.1.190, 5.15.223, and 5.10.272 (Oct 15) · 7.3 (Oct 18) |
| No fix yet | Ubuntu 16.04 through 26.04, as of 9:00 AM CST, November 5; fixes for 22.04, 24.04, and 26.04 are in testing. Red Hat still lists OpenShift's RHEL CoreOS as affected, and now RHEL AI 3 too. Upstream branches that reached end-of-life before 5.10 get no fix. |
| Exploitation | Anthropic says Claude built a remote root exploit for servers with a multipath TCP listener; it hasn't published the exploit. Security researchers report suspected threat-actor scanning for those servers since November 2. At least ten GitHub repositories say they include a race trigger or a proof of concept. Not in CISA's Known Exploited Vulnerabilities catalog as of 9:00 AM CST, November 5. |
| CVE records | Published on cve.org (opens in a new tab) on October 15. NVD (opens in a new tab) lists it as "Received," with no analysis of its own, as of 9:00 AM CST, November 5. |
| Credit | Reported by Xinyang Ge of Anthropic. Fixed by Eric Dumazet and reviewed by Kuniyuki Iwashima, per the fix commit. |
Any Linux kernel from 4.4 on that lacks the fix, whether it serves the network directly or hosts containers: a container runs on the host's kernel, so a new image doesn't change anything. IPv6 connections aren't affected by the main issue; the commit says "IPv6 is not affected by the first issue."
| Release line | Status on Nov 5, 2026 | Fixed in | Released |
|---|---|---|---|
| 7.3 | Stable | 7.3 (from 7.3-rc7) | Oct 18, 2026 |
| 7.2 | End of life (last release 7.2.15, Nov 1) | 7.2.10 | Oct 11, 2026 |
| 6.18 | Longterm | 6.18.56 | Oct 11, 2026 |
| 6.12 | Longterm | 6.12.113 | Oct 15, 2026 |
| 6.6 | Longterm | 6.6.159 | Oct 15, 2026 |
| 6.1 | Longterm | 6.1.190 | Oct 15, 2026 |
| 5.15 | Longterm | 5.15.223 | Oct 15, 2026 |
| 5.10 | Longterm | 5.10.272 | Oct 15, 2026 |
| 4.4 – 5.4 | End-of-life upstream | distro fixes only | — |
Distribution kernels carry backported fixes under their own version numbers, so check the tracker, not the upstream number. As of 9:00 AM CST, November 5:
When a Linux server accepts a TCP connection, the kernel tracks the half-open handshake in a lightweight "request socket," then creates a full socket once the handshake completes. If the connecting peer's IPv4 header carried options, the kernel keeps them to echo in its replies. The bug: the kernel handed those saved options from the request socket to the new connection without copying them, then freed them immediately when the connection went away, while another CPU could still be reading them to build a handshake reply. Reading freed kernel memory is a use-after-free.
The peer needs no login, and it all happens inside the kernel's handshake handling, before any application code runs. The CNA's scoring notes say the stale data can leak kernel heap memory back to the peer, steer kernel writes out of bounds, or crash the machine. Anthropic says a lost race "only fails the handshake without crashing the whole system, so a determined attacker can retry indefinitely until they succeed."
Eric Dumazet's commit, "ipv4: free inet_opt and ireq_opt after an RCU grace period," changes two plain kfree() calls to kfree_rcu(), which waits until every reader has finished with the memory before freeing it. Other code paths that replace these options already did that. The commit also fixes the same problem in request-socket migration (net.ipv4.tcp_migrate_req=1, or a BPF_SK_REUSEPORT_SELECT_OR_MIGRATE program), and says a related IPv6 case "will be addressed in a separate patch."
torvalds/linux · e3f33b0 · Oct 1, 2026
net/ipv4/af_inet.c
1 lines added, 1 removed
net/ipv4/tcp_ipv4.c
1 lines added, 1 removed
2 files changed2 lines added, 2 removed
go line in the program's go.mod, and applies only where the kernel allows MPTCP (net.mptcp.enabled = 1, as on the Debian 12 kernel we checked; AlmaLinux says it ships MPTCP off). The Go team adds that "the underlying bug is not specific to MPTCP. It is possible that systems with no MPTCP listeners are vulnerable to similar attacks." An internet-wide scan by security researchers on October 30 found about 2.1 million servers reachable over MPTCP on port 443, and by November 2 the same team was seeing what it suspects is threat-actor scanning for such servers (below). The count shows total exposure, not whether those kernels are patched. The check is below.Red Hat now rates it as the CNA does. It raised the bug from Important (8.1) to Critical (9.8) on October 30, citing "the lack of network prerequisites and the potential for severe system instability." Its bulletin says "Systems accepting inbound TCP connections are most directly exposed."
Compare the result with your distribution's status above, not with the upstream version. If the installed kernel is newer than the running one, the fix isn't active until you reboot. Run these on container hosts and Kubernetes nodes, not inside containers: a container reports the host's kernel. We ran them on a Debian 12 host and in a Rocky Linux 9 container.
ss -M lists only MPTCP sockets, so an empty result means no MPTCP listener. net.mptcp.enabled = 1 alone doesn't create one: an application has to open an MPTCP socket. We tested both with iproute2 6.1 in a Debian 12 container, where ss -Mlnp listed a test MPTCP listener and skipped a plain TCP one beside it. On October 31 it also listed a Go HTTP server whose go.mod says go 1.24, and skipped the same server built with go 1.23.
This check tells you which hosts to patch first, not which ones are safe. Red Hat's bulletin says it "is not a test for exposure," and that "a result showing no MPTCP listeners should be treated as informational rather than as reassurance." Every host on an affected kernel needs the fix.
Install your distribution's fixed kernel as soon as it ships, then reboot: a kernel update doesn't take effect until the machine boots into it. Reboot container hosts and Kubernetes nodes too. Appliances and cloud images that bundle a kernel need their vendor's update.
On Debian, RHEL and AlmaLinux 8, 9, and 10, SUSE Linux Enterprise Server 16.0, 15 SP7, and the 15 LTSS releases, openSUSE Leap 16.0 and 15.6, Oracle Linux, Amazon Linux 2 and 2023, and Ubuntu 26.10, the fix is out. On Ubuntu 22.04, 24.04, and 26.04, it's in testing in the -proposed pocket, and Canonical's post (opens in a new tab) explains how to install it now if you run your own kernel acceptance tests. Ubuntu 20.04 and older, OpenShift's RHEL CoreOS, and RHEL AI 3 have no fix yet, as of 9:00 AM CST, November 5. Watch the trackers linked above. Running a self-built kernel? Move to the fixed stable release for its line.
Red Hat's bulletin (opens in a new tab), AlmaLinux (opens in a new tab), and Ubuntu's tracker (opens in a new tab) now publish nftables rules that drop incoming IPv4 TCP packets carrying IP options. Red Hat says its rule "reduces exposure to this vulnerability but cannot fully prevent it," and Ubuntu that its rule "reduces the risk but does not fully prevent exploitation." The following is Atlas's interim recommendation, the same rule as AlmaLinux's, based on the CNA's scoring notes, which say the attacking peer supplies IPv4 Record Route or Timestamp options in its SYN. TCP traffic almost never carries IPv4 options legitimately, so dropping TCP packets that have them should cost nothing:
ip hdrlength gt 5 matches any IPv4 header longer than the 20-byte minimum, which only options produce. The rule runs before connection tracking and doesn't touch IPv6 or non-TCP traffic. We tested it in a Debian 12 container: it dropped a SYN carrying a Record Route option, and ordinary TCP connections still worked. We haven't tested it against an exploit, and it doesn't fix the kernel. It lasts until reboot, which is fine if the reboot is into a fixed kernel; if a host reboots before the update, reapply it, as Red Hat's bulletin now warns for its own rule. Ubuntu's tracker gives steps to load its version of the rule at boot from /etc/nftables.conf. Remove it with nft delete table ip cve_2026_98432. If your edge firewall can drop IPv4 packets with options, doing it there protects every host behind it.
If you manage the firewall with iptables, this rule does the same job from the raw table, which also runs before connection tracking:
The u32 match reads the IPv4 header-length field, and a value of 6 to 15 means the header carries options. We tested it with iptables 1.8.9, on both its nf_tables and legacy backends, in a Debian 12 container: it dropped a Record Route SYN sent to the host and one routed through it, and ordinary TCP connections still worked. Remove it by running the same command with -D in place of -I; like the nft rule, it lasts until reboot. It needs the kernel's xt_u32 module, which the Debian 12, Ubuntu 22.04 and 24.04, and RHEL 8 kernels include. RHEL 9 and 10 kernels, and their rebuilds, leave it out, so use the nft rule there.
Red Hat's and Ubuntu's rules hook input; ours hooks prerouting, so it also drops packets the host routes on to containers. On October 31 we tested it against two pairs of network namespaces. It dropped a Record Route SYN routed between one pair, and never saw one bridged between the other, as AlmaLinux warns: "Traffic between containers on the same host may bypass it." AlmaLinux also says not to use it on networks that rely on CIPSO security labels, which travel in IPv4 options.
Turn off multipath TCP if you don't use it. That removes the setup Anthropic's exploit needs, though not the bug. AlmaLinux's steps:
New MPTCP sockets then fail, and Go programs fall back to plain TCP when they start. The setting is per network namespace, and each new one starts with MPTCP on, so a container keeps it unless you set it there too (Docker: --sysctl net.mptcp.enabled=0). We tested the steps and the Docker flag on October 31. For a single Go program, GODEBUG=multipathtcp=0 in its environment turns the default off, but the Go Security Team calls disabling MPTCP program by program "a fragile mitigation."
Red Hat recommends filtering packets that carry IP options over this step, which it says "only blocks one method of exploitation." RHEL ships with MPTCP disabled, so on a default RHEL system, Red Hat says, setting it to 0 "changes nothing." It adds that the filter takes effect immediately, "for both new and existing connections," without restarting services.
The init_on_free=1 boot parameter, suggested on the oss-security list as another stopgap, isn't one either: it "doesn't do anything meaningful against this CVE," a CIQ kernel engineer replied (opens in a new tab).
Until you can patch, keep MPTCP listeners off the internet, and put the hosts that have one at the top of the patch list.
All times are US Central: CDT (UTC−5) through October 31, CST (UTC−6) from November 1.
Nov 5
Nov 4–5
SUSE fixes openSUSE Leap 15.6
Nov 4
Nov 3
Nov 2–3
Nov 2
Red Hat fixes RHEL for NVIDIA
Nov 1
Oct 31
Oct 30
Oct 29
Oct 28
Oct 27
Oct 26–27
Oct 24
Oct 18
Oct 16
Oct 15
Oct 11
Oct 5
Oct 1
Exploitation status. Anthropic says Claude wrote the exploit "after initial prompting," that it runs "arbitrary commands as root, without any credentials or user interaction" on servers with a multipath TCP listener, and that in its proof of concept "it takes 48 minutes to leak a kernel pointer and another five minutes to reach the shell command execution." Anthropic hasn't published the exploit. Its disclosure policy (opens in a new tab) gives developers 90 days to patch and a further 45 days before Anthropic discusses details beyond a public advisory. Red Hat's bulletin now says the bug could be exploited "to trigger remote code execution."
Security researcher Marcus Hutchins wrote on November 3 that "we've picked up several residential proxies and VPNs scanning the internet on port 443 using MPTCP-enabled connections," and that "we strongly suspect this activity is scanning for CVE-2026-98432 specifically." The post adds that "The use of residential proxies and RU-based VPN servers suggests that this is likely threat actor scanning, rather than security researchers," and that "several denial of service proof-of-concepts" are circulating.
By 9:00 AM CST on November 5, 78 GitHub repositories referenced the CVE. Most describe themselves as audit or mitigation tooling, among them Ansible roles, an XDP program, a checker for the interim packet filter, Wazuh detection rules, and a livepatch for AlmaLinux and Rocky Linux. At least ten say they include a race trigger or a proof of concept for the use-after-free; the first was created at 7:22 PM CDT on October 29, and the latest at 5:51 AM CST on November 4. One, created November 2, calls itself a "lab root-shell kit" that needs a kernel modified to help the exploit, "not a stock remote exploit." The CVE isn't in CISA's Known Exploited Vulnerabilities catalog as of 9:00 AM CST. We haven't tested any public proof of concept, and we aren't linking them.
We'll revise this post and mark it updated when any of these change: Ubuntu's fixed kernels reaching its updates pocket, or a fixed RHEL CoreOS; NVD's analysis; a KEV listing or any report of exploitation; the separate IPv6 fix.
November 5, 9:00 AM CSTLatest
SUSE fixed openSUSE Leap 15.6, and Red Hat now lists RHEL AI 3 as affected. AWS's bulletin, updated November 4, adds Elastic Beanstalk's fixed Amazon Linux 2023 platform versions and Bottlerocket AMIs with the fix. Added Ubuntu's steps for loading its packet filter at boot. 78 GitHub repositories now reference the CVE, still at least ten describing a race trigger or a proof of concept. Unchanged: Ubuntu's fixed kernels are still in its -proposed pocket with no Ubuntu Security Notice, OpenShift's RHEL CoreOS is still listed as affected, AlmaLinux's errata for 9 and 10, Debian, Oracle, Amazon Linux's packages, NVD's status, and KEV haven't moved, and no IPv6 follow-up has landed in mainline or the stable kernels.
November 4, 9:30 AM CST
Added a security researcher's November 3 report of residential proxies and VPNs scanning port 443 over MPTCP, which the researcher's team strongly suspects is scanning for this CVE, and the team's count of about 2.1 million servers reachable over MPTCP on port 443; the key facts' "Proof of concept" row is now "Exploitation." Red Hat published mitigation guidance for self-managed OpenShift Container Platform 4 and fixed its RHEL 8.6 AUS and EUS Long-Life streams, and its CVE page no longer lists RHEL 9's separate real-time kernel package as affected. AWS's bulletin, updated November 3, says Global Accelerator drops packets carrying IP options, so the endpoints behind it aren't affected, and no longer says Network Load Balancer TCP listeners pass IP options to targets. Corrected: we were still listing both among the services whose hosts AWS says to patch. 70 GitHub repositories now reference the CVE, at least ten describing a race trigger or a proof of concept. Unchanged: Ubuntu's fixed kernels are still in its -proposed pocket, OpenShift's RHEL CoreOS is still listed as affected, AlmaLinux's errata for 9 and 10, Debian, SUSE, Oracle, Amazon Linux's packages, NVD's status, and KEV haven't moved, and no IPv6 follow-up has landed in mainline or the stable kernels.
November 3, 11:05 PM CST
Red Hat published mitigation steps for managed OpenShift (ROSA, OSD, and ARO): it mitigates control-plane nodes itself, and customers can apply the nftables rule to worker and infrastructure nodes. Its guidance for self-managed OpenShift is still in preparation. Its bulletin now warns that the packet filter doesn't survive a reboot, so we added that to If you can't patch yet. 68 GitHub repositories now reference the CVE; the new one is mitigation tooling. Unchanged: Ubuntu's fixed kernels are still in its -proposed pocket, OpenShift's RHEL CoreOS and RHEL 9's separate real-time kernel package are still listed as affected, and KEV, NVD, and the other distributions' status haven't moved.
November 3, 3:15 PM CST
SUSE fixed SUSE Linux Enterprise Server 15 SP7, and Red Hat fixed its RHEL 8.4 AUS and EUS Extension streams and the RHEL 9.2 extended-support real-time kernel. Red Hat's bulletin now says checking for MPTCP listeners "is not a test for exposure" and that turning MPTCP off "only blocks one method of exploitation"; we added both to the MPTCP sections, and added the bulletin's OpenShift guidance to Red Hat's entry. The bulletin no longer names RHEL CoreOS in its "potentially impacted" sentence, so we took it out of our quote; OpenShift is still among its directly affected products. 67 GitHub repositories now reference the CVE, still at least nine describing a race trigger or a proof of concept. Ubuntu's fixes for 22.04, 24.04, and 26.04 are still in testing, and AlmaLinux's errata for 9 and 10, Debian, Oracle, Amazon Linux, NVD's status, and KEV are unchanged; no IPv6 follow-up has landed in mainline or the stable kernels.
November 3, 8:30 AM CST
Canonical says Ubuntu's fixed kernels for 22.04, 24.04, and 26.04, in its -proposed pocket since October 26 and 27, are in testing, and lists 26.10's 7.3.0-10.10, in its updates pocket since October 28, as fixed. Ubuntu raised its priority from High to Critical and added an nftables mitigation; its tracker still reads "Vulnerable" for every release. Red Hat fixed its RHEL 8.8, 9.2, 9.4, 9.6, and 10.0 extended-support streams, and lists RHEL for NVIDIA as fixed by RHSA-2026:93775 on November 2, so it's no longer in the affected list. Oracle fixed its Red Hat Compatible Kernel for Oracle Linux 9. Added a CIQ kernel engineer's reply that the init_on_free=1 boot parameter isn't a mitigation. 66 GitHub repositories now reference the CVE, still at least nine describing a race trigger or a proof of concept. AlmaLinux's errata for 9 and 10, Debian, SUSE, Amazon Linux's packages, NVD's status, and KEV are unchanged, and no IPv6 follow-up has landed in mainline or the stable kernels.
November 2, 12:00 PM CST
AlmaLinux moved fixed kernels for AlmaLinux 8, 9, and 10 from testing to its production repositories, the same versions as RHEL's, and Oracle fixed its Red Hat Compatible Kernel for Oracle Linux 8 and 10. Added AWS's bulletin: which of its load balancers pass IP options on to the hosts behind them, the Amazon Linux 2023 release that locked instances need, and its fixed EKS and ECS AMIs. 57 GitHub repositories now reference the CVE, at least nine describing a race trigger or a proof of concept. Ubuntu, Red Hat, Debian, SUSE, Amazon Linux's packages, NVD's status, and KEV are unchanged, and Linux 7.4-rc1, released November 1, doesn't include the IPv6 follow-up.
November 1, 6:15 PM CST
Red Hat shipped fixed kernels for RHEL 8, 9, and 10, including the real-time kernels, between 2:52 and 4:16 PM, so RHEL moved out of the "no fix yet" row; OpenShift's RHEL CoreOS and RHEL for NVIDIA are still listed as affected. Linux 7.2 reached end of life with 7.2.15, and 7.3 is now a stable line. 41 GitHub repositories now reference the CVE, at least seven describing a race trigger or a proof of concept. Ubuntu, AlmaLinux, Debian, SUSE, Oracle, Amazon, NVD's status, and KEV are unchanged, and no IPv6 follow-up has landed in mainline or the stable kernels.
October 31, 5:15 PM CDT
Red Hat raised its rating from Important (8.1) to Critical (9.8) on October 30 and published a mitigation in bulletin RHSB-2026-016, so we replaced its earlier rating and its "no mitigation" statement. Added the Go Security Team's notice that Go programs open MPTCP listeners by default, AlmaLinux's fixed kernels in testing, Amazon Linux's fixes for its remaining kernels, openSUSE Leap 16.0's fix, steps to turn off MPTCP, an iptables version of the interim rule, and the interim rule's limits with bridged container traffic and CIPSO. 36 GitHub repositories now reference the CVE. Ubuntu, Debian, Oracle, NVD's status, and KEV are unchanged, and the stable kernels released October 30 don't include the IPv6 follow-up.
October 29, 8:00 PM CDT
Added Oracle Linux, fixed in its UEK kernels since October 16, and Amazon Linux, fixed in Amazon Linux 2023's 6.12 and 6.18 kernels and pending elsewhere. Eight GitHub repositories now reference the CVE, one describing a race trigger. Ubuntu, Red Hat, Debian, SUSE, NVD, and KEV are unchanged since 11:00 AM, and nothing was corrected.
The bug is reachable through any listening TCP port, so it helps to know which ones the internet can see. Our free security audit checks the domains and public IPs you verify for exposed services, publicly visible software versions with known CVEs, TLS, email security, DNS, and HTTP headers.
Atlas Service
Free Security Audit
Non-destructive checks of your verified domains and public IPs, and a PDF report with every finding ranked and a fix for each, within 3 business days of verification.
It's the second Linux kernel fix this month that only takes effect after a reboot. The other, Fragnesia, is being chained with PHP2Shell to get root.
e3f33b0 (opens in a new tab)github.cominit_on_free, via oss-security (opens in a new tab)openwall.commultipathtcp setting (opens in a new tab)go.devlinux publishing history (opens in a new tab))ubuntu.comTexto de atlasum.net tal como lo publicó, leído de su página y mostrado aquí con el formato de El Autómata. El original, en atlasum.net · Volver a la noticia de El Autómata