Koffei

CVE-2026-98432

🟠 Medium

Linux kernel — IPv4 TCP. tcp_v4_syn_recv_sock() hands ireq_opt to the child socket without copying it ; a retransmitted SYN on another CPU builds a SYNACK that reads it under RCU only, while inet_sock_destruct() used a plain kfree(), so ip_options_build() reads freed memory and leaks kernel heap straight to the peer. A second path exists through request-socket migration (tcp_migrate_req=1 or a reuseport BPF program). Go programs are specifically exposed: Go 1.24 and later enable MPTCP by default on Linux listen sockets, and the Go Security Team announced on 29/10/2026 that a server with MPTCP listeners is vulnerable to remote code execution, citing an exploit described by Anthropic. That is the first exploitation-capability statement in the record's own references · Vulnerable from 4.4 ; fixed in 5.10.272 / 5.15.223 / 6.1.190 / 6.6.159 / 6.12.113 / 6.18.56 / 7.2.10 / 7.3 (mainline commit e3f33b0a in 7.3-rc7, with eight backports) ; Debian DLA-4841-1, DLA-4850-1 and DSA-6568-1

CVSS (Koffei note)
9.8
CVSS (v3.1, CVE List)
9.8
EPSS (2026-11-18)
0.777%
56.54th percentile

Exploitation

No — no exploitation found, no public PoC and no actor. The record has no CISA ADP container at all, so there is no SSVC rating in either direction ; not in KEV. Two caveats on reachability and severity: the attack needs the target to accept IPv4 Record Route or Timestamp options, which many edges strip, and the CNA justifies AC:L on the grounds that "multicore scheduling and heap grooming do not constitute conditions outside the attacker's influence", so AC:H is defensible. Honest read: remote information leak and remote denial of service are credible, remote code execution is theoretical Update 30/10/2026: a CVE Program reference container appeared today carrying the Go Security Team announcement described above. It is a reference container only — still no CISA ADP container, so still no SSVC rating in either direction, still no kev block, and still no CWE from any supplier. No in-the-wild exploitation and no public PoC. Corroborated 03/11/2026 by the Daily CyberNews note of 03/11/2026, which describes it as a use-after-free in the IPv4 TCP connection-setup path at 9.8 per the kernel CNA. Correction of 04/11/2026: a public proof of concept does exist. This row previously said there was none. CERT Santé's alert of 03/11/2026 records one as available in open sources, and reports that Anthropic demonstrated root command execution against a server listening over MPTCP. Exploitation is still recorded as none by CERT Santé and it remains outside CISA KEV, so the rating stays 🟠. Existing PoC (not opened), recorded from the proof-of-concept list supplied on 09/11/2026. The CVE record carries no exploit-tagged reference and its SSVC does not record a proof of concept, so the supplied list is the only source for this and it is recorded as the user's own finding rather than a vendor or CISA signal.

Timeline

  1. 15/10/2026CVE record published (CVE List)
  2. 28/10/2026Anthropic says an unreleased internal research model chained flaws into working exploits for Chrome, Safari, Windows and Linux
  3. 30/10/2026Linux kernel TCP use-after-free
  4. 03/11/2026Linux kernel TCP use-after-free (MPTCP remote root) — Red Hat now patched, others still exposed
  5. 04/11/2026Linux kernel IPv4/TCP use-after-free, CVSS 9.8
  6. 17/11/2026Juniper Junos OS Evolved inherits the Linux IPv4 TCP use-after-free, fix still pending

Risk

9.8 Critical (v3.1, Linux CNA via bippy) — AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H, EPSS 0.760% (55.33rd percentile, scored 03/11/2026, up from 0.457%) — this score is unmoderated: there is no CISA ADP container, no NVD score, and Red Hat's bug 2553390 carries no impact rating, no CVSS and no RHEL product list. The record also assigns no CWE at all ; this is a use-after-free, classified CWE-416 here rather than by the CNA Correction of 04/11/2026: a second scorer now exists and it agrees. CERT Santé's alert of 03/11/2026 gives the same 9.8 on the same v3.1 vector, so the score is no longer unmoderated even though there is still no CISA ADP container and no NVD score.

Fix

Yes — the fix replaces kfree() with kfree_rcu(). But the fix is incomplete by the record's own admission: the IPv6 / CALIPSO variant in tcp_v6_reqsk_destructor() has the same bug and "will be addressed in a separate patch", and no CVE ID was found for that follow-up For Go deployments specifically, the Go Security Team's workarounds are GODEBUG=multipathtcp=0 at runtime or in go.mod, or per-socket controls which override GODEBUG — but the team calls per-program disabling fragile, notes the bug may not be specific to MPTCP, declines to change Go's defaults for now, and advises upgrading the kernel instead (issue go.dev/issue/82717). Update of 03/11/2026, and it removes an option rather than adding one: an oss-security thread proposed init_on_free=1 on 31/10/2026 as an interim mitigation for hosts that cannot take a fixed kernel yet. On 02/11/2026 a CIQ kernel engineer answered that it "doesn't do anything meaningful against this CVE", because the attack uses pointers leaked from live slab objects rather than freed contents. Do not rely on init_on_free: there is still no mitigation short of a fixed kernel. Workarounds now exist, from CERT Santé on 03/11/2026, and each has a caveat. Drop inbound IPv4 TCP packets carrying IP options with the nftables expression ip hdrlength > 5 — but container-to-container traffic on the same host may bypass the filter, and do not use it where the network relies on CIPSO labels. Disable MPTCP with net.mptcp.enabled=0 and restart anything already listening over MPTCP — this only removes the condition of the demonstrated exploit, since the flaw itself is in standard IPv4 TCP. For Go 1.24 and later set GODEBUG=multipathtcp=0 or add multipathtcp=0 to go.mod — which the Go team itself considers fragile.

Official description

In the Linux kernel, the following vulnerability has been resolved: ipv4: free inet_opt and ireq_opt after an RCU grace period tcp_v4_syn_recv_sock() transfers ownership of ireq->ireq_opt to the child socket (newinet->inet_opt) without copying it. Another cpu can concurrently process a retransmitted SYN for the same request socket, and send a SYNACK from tcp_check_req(). tcp_v4_send_synack() and inet_csk_route_req() read ireq->ireq_opt under rcu_read_lock() only, and ip_build_and_send_pkt() and ip_options_build() then read opt->optlen twice. Note that the SYNACK timer itself is not an issue: it holds a reference on its request socket, and inet_csk_reqsk_queue_drop() calls timer_delete_sync() before the child can be freed. Since commit 079096f103fa ("tcp/dccp: install syn_recv requests into ehash table"), request sockets are processed without holding the listener lock, so nothing prevents the child socket from being freed while the SYNACK is still being built. TCP child sockets do not have SOCK_RCU_FREE, and inet_sock_destruct() frees inet_opt with a plain kfree(), leading to a use-after-free in ip_options_build(). A similar issue exists with request socket migration (net.ipv4.tcp_migrate_req=1, or a BPF_SK_REUSEPORT_SELECT_OR_MIGRATE program). reqsk_timer_handler() clones the request socket with inet_reqsk_clone(), so that the old request socket and its clone share the same ireq_opt, then reqsk_migrate_reset() clears the pointer in the old one. Another cpu holding a reference on the old request socket can still be using these options (sending a SYNACK, or creating a child in tcp_v4_syn_recv_sock()) when the clone is freed, and tcp_v4_reqsk_destructor() also uses a plain kfree(). Readers already use RCU, and other paths replacing inet_opt (do_ip_setsockopt(), cipso_v4_sock_setattr()...) already use kfree_rcu(). Use kfree_rcu() in inet_sock_destruct() and tcp_v4_reqsk_destructor() as well. IPv6 is not affected by the first issue, because tcp_v6_syn_recv_sock() duplicates the options. tcp_v6_reqsk_destructor() has the same migration issue with ipv6_opt, which is only set by CALIPSO. This will be addressed in a separate patch.

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H