Threat reportVulnerabilityTL-2026-6114
NoisyTCP (CVE-2026-99288): 11-Year-Old Linux Kernel TCP Race Condition Enables Remote Unauthenticated Crash
NoisyTCP is a race condition in the Linux kernel's TCP established-socket lookup that lets a remote, unauthenticated attacker panic any host with a plain TCP listener by sending IPv4 fragments and TCP packets that are steered to different CPUs, causing a NULL rsk_listener dereference.
Key facts
| Identifiers | TL-2026-6114, CVE-2026-99288 |
|---|---|
| Also known as | NoisyTCP |
| Severity | HIGH |
| CVSS | 7.5 |
| First disclosed | |
| EPSS | CVE-2026-99288: 0.1% (4.8th percentile) |
| Affected products | Linux Linux kernel (TCP); Debian linux (kernel package); Red Hat Red Hat Enterprise Linux 8, 9, 10 (kernel); Canonical Ubuntu kernel 16.04 to 26.10 |
| Weaknesses (CWE) | CWE-362, CWE-476, CWE-416 |
| MITRE ATT&CK techniques | T1190 Exploit Public-Facing Application, T1499.004 Application or System Exploitation, T1529 System Shutdown/Reboot |
| Motivation | UNKNOWN |
| Target sectors | technology, cloud, telecoms, finance, government administration, health |
| Target regions | Global |
| Detection rules | 9 (1 shown free) |
| Sources | 8 cited sources |
Detection coverage
As of 2026-11-16, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-6114 across Splunk SPL, Microsoft KQL and Sigma, covering 14 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: NoisyTCP (CVE-2026-99288) Kernel Fault in TCP Receive Path (tcp_v4_rcv / rsk_listener) — Exploitation Outcome
Detects the described NoisyTCP crash signature: a kernel panic, oops, KASAN use-after-free or NULL-pointer dereference logged within the same 2-minute bucket as tcp_v4_rcv / tcp_check_req / rsk_listener / inet_lookup_established frames on one host (NULL rsk_listener dereference after the established-lookup ABA race, CVE-2026-99288; the KASAN branch covers the harder 'Full child socket ABA race' use-after-free path, CWE-416, for which no RCE is claimed). Oops text is multi-line, so the search correlates lines per host per bucket. The exact oops text is unpublished and there is no network IOC, so the crash itself is the highest-fidelity signal. Requires kernel logs (syslog/journald/netconsole/kdump) forwarded to the SIEM.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1190, T1499.004, T1529
- Data sources: linux_syslog, kernel_log, journald, kdump
- Severity: high
- Confidence: medium
Splunk SPL
index=os sourcetype=syslog earliest=-24h ("Kernel panic" OR "NULL pointer dereference" OR "Oops:" OR "Fatal exception" OR "KASAN" OR "tcp_v4_rcv" OR "tcp_check_req" OR "rsk_listener" OR "inet_lookup_established")
| eval fault=if(match(_raw,"Kernel panic|NULL pointer dereference|Oops:|Fatal exception|KASAN: [\\w-]*use-after-free"),1,0), tcp_frame=if(match(_raw,"tcp_v4_rcv|tcp_check_req|rsk_listener|inet_lookup_established"),1,0)
| bin _time span=2m
| stats sum(fault) as fault_lines sum(tcp_frame) as tcp_frame_lines count as total_lines values(source) as sources by _time host
| where fault_lines>0 AND tcp_frame_lines>0
| table _time host fault_lines tcp_frame_lines total_lines sourcesMicrosoft KQL
Syslog
| where TimeGenerated > ago(24h)
| extend IsFault = SyslogMessage has_any ("Kernel panic", "NULL pointer dereference", "Oops:", "Fatal exception") or SyslogMessage matches regex @"KASAN: [\w-]*use-after-free"
| extend IsTcpFrame = SyslogMessage has_any ("tcp_v4_rcv", "tcp_check_req", "rsk_listener", "inet_lookup_established")
| where IsFault or IsTcpFrame
| summarize FaultLines = countif(IsFault), TcpFrameLines = countif(IsTcpFrame), TotalLines = count(), Sample = take_any(SyslogMessage) by Computer, HostIP, bin(TimeGenerated, 2m)
| where FaultLines > 0 and TcpFrameLines > 0
| project TimeGenerated, Computer, HostIP, FaultLines, TcpFrameLines, TotalLines, SampleSigma
title: NoisyTCP CVE-2026-99288 Fatal Kernel Exception in Interrupt Context
id: 6b0f3a52-8c1d-4e7a-9a21-5d3c7e1f4a01
status: experimental
description: Detects a fatal kernel exception raised in interrupt (softirq) context, the panic form expected when the NoisyTCP NULL rsk_listener dereference in tcp_v4_rcv fires with panic_on_oops=1. Generic by itself; correlate with TCP-path RIP frames (see SIG-001) and with the host running a TCP listener.
references:
- https://github.com/V4bel/NoisyTCP
- https://access.redhat.com/security/cve/CVE-2026-99288
- https://securityonline.info/noisytcp-vulnerability-cve-2026-99288/
author: Threadlinqs Team
date: 2026/11/16
tags:
- attack.impact
- attack.t1499.004
- attack.t1529
- attack.initial-access
- attack.t1190
- cve.2026.99288
logsource:
product: linux
service: syslog
detection:
keywords_panic:
- 'Kernel panic - not syncing: Fatal exception in interrupt'
- 'Kernel panic - not syncing: Fatal exception'
condition: keywords_panic
falsepositives:
- Hardware faults, driver bugs or other kernel defects that panic in interrupt context
- Deliberate crash tests (sysrq-c, kdump validation)
level: high
Known false positives:
- Unrelated kernel/driver bugs or hardware faults that crash in the network receive path
- Planned crash-dump tests (echo c > /proc/sysrq-trigger) on TCP-listening hosts
Tuning notes: Adjust index/sourcetype to where kernel logs land (e.g. linux_messages_syslog, journald, netconsole). Hosts without kdump/pstore/netconsole lose the oops on panic, so enable remote kernel logging. Suppress hosts in planned crash-test windows. Prioritise internet-facing multi-RX-queue/RPS hosts running nginx/apache/sshd. Kernels containing fix commit 4e2b3ef19e0e should not crash this way.
- NoisyTCP (CVE-2026-99288) Multi-Host Kernel Panic Burst — Fleet-Wide Remote DoS · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) Mitigation / Posture Change And Exposure Checks — tcp_syncookies=0, RX Queue Reduction, RPS/RFS, ethtool -l · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) Exposure — Hosts Booted On Affected Upstream Kernel Versions · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) Kernel Oops Or KASAN Report In TCP Established-Lookup Path · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) kdump Vmcore Saved After Kernel Crash · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) Post-Crash Recovery Artifacts — Unclean Journal Or Filesystem Recovery · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) Unclean Host Reboot After Kernel Crash — Impact · unlock with TL Creds
- NoisyTCP (CVE-2026-99288) Trigger Traffic — IPv4 Fragments Plus TCP SYN Flood From One Source To A Listener · unlock with TL Creds
Indicators of compromise
As of 2026-11-16, Threadlinqs Intelligence tracks 14 indicators of compromise for TL-2026-6114.
Sample indicators (3 of 14)
- URL:
hxxps://github[.]com/V4bel/NoisyTCP - File hash:
079096f103faca2dd87342cca6f23d4b34da8871 - File hash:
4e2b3ef19e0e04adca38442f07dd6bc93c712bf1
Defanged, TLP:CLEAR. Full set of 14 indicators: unlock with TL Creds.
Technical analysis
The flaw dates to commit 079096f103fa (2015-10-03) and was fixed upstream by commit 4e2b3ef19e0e (2026-11-03). Impact is remote denial of service; no in-the-wild exploitation, public PoC, or CISA KEV listing is reported.
NoisyTCP (CVE-2026-99288) was reported and patched by researcher Hyunwoo Kim (@v4bel), who published an advisory at github.com/V4bel/NoisyTCP and posted to oss-security on 2026-11-15 after the distribution embargo ended. The bug is an ABA-style race in the kernel's TCP receive path: a lookup of established connections matches on the connection 4-tuple but does not check whether the socket is still hashed. One CPU finds an existing half-open request socket while another CPU frees that request_sock and reuses the same memory for a new SYN-cookie request with the same connection details. The first CPU takes a reference on the new object, which has no listener attached, and the TCP receive path (tcp_v4_rcv) dereferences the NULL rsk_listener, producing a kernel panic. The upstream fix is titled "inet: reject unhashed sockets during established lookups".
The attacker spreads processing across receive contexts by exploiting the difference in receive hash between IPv4 fragments and ordinary TCP packets, so the racing packets land on different CPUs. Exposure requires two or more NIC receive queues, or Receive Packet Steering (RPS) distributing packets across CPUs, and a plain TCP listener (listen()). No unusual sysctls, firewall rules or special sockets are needed; default Nginx, Apache and SSH servers are reachable. In a container host a panic takes down the whole machine because the kernel is shared. A second, separate path (the 'Full child socket ABA race') could produce a remote use-after-free, but it needs several successive races and is much harder to trigger; the researcher does not claim remote code execution, and sources state it is not a TCP RCE 'at least for now'.
Reported trigger timing in the researcher's lab: about 5 seconds to 2 minutes against default configurations on QEMU/KVM virtual machines (hypervisor vCPU pausing widens the race window), and about 14 hours against a quiet bare-metal Nginx server with 8 RX queues even with a tuned PoC. The researcher holds only a very limited, lab-specific exploit for VM servers that has not been released; the GitHub repository contains the write-up only. No exploitation in the wild is reported and the CVE is not in CISA KEV (securityonline reports a 30-day EPSS of 0.2%). Severity: NVD analysis and CNA CVSS were still awaiting at time of writing; Red Hat rates it 7.5 and Oracle rates it 8.1 Important (ELSA-2026-500630 to 500632). The cvss_score recorded here is the Red Hat figure; no vector is published. CWE identifiers are analyst-assigned from the described mechanics (race condition, NULL dereference, and use-after-free for the secondary path) since no source states a CWE.
Affected: Linux kernel from the 2015 introducing commit (stated as 4.4 and later) up to 6.12.116, 6.18.62 and 7.3.3. Fixed in stable 6.12.117, 6.18.63 and 7.3.4 (released 2026-11-10) and 7.4-rc2. The 7.2 branch ended at 7.2.15 on 2026-11-01 before the fix existed; 5.10, 5.15, 6.1 and 6.6 had no fixed upstream release as of 2026-11-15. Debian patched bookworm (6.1.192-1, DLA-4906-1) and trixie (6.12.117-1, DSA-6628-1); Arch shipped linux 7.3.4.arch1-2 and linux-lts 6.18.63-3; Ubuntu 16.04 to 26.10 were 'needs evaluation'; RHEL 8/9/10 were listed affected without an erratum. Weakly ordered CPU architectures (arm64, POWER, RISC-V) additionally need upstream commit 5b5d8f351dd9. Defender notes: there are no network IOCs; detection is based on exposure checks (RX queue count, RPS masks, kernel version/changelog), unexplained kernel panics on TCP-listening hosts, and unusual IPv4 fragment traffic accompanying TCP to listeners (the last is analyst inference, not a published signature).
MITRE ATT&CK techniques by tactic:
Initial Access
T1190 Exploit Public-Facing Application
Impact
T1499.004 Application or System Exploitation; T1529 System Shutdown/Reboot
Remediation
Patches
- Upstream stable: 6.12.117, 6.18.63, 7.3.4 (and 7.4-rc2)
- Debian 13 (trixie): linux 6.12.117-1 (DSA-6628-1); Debian 12 (bookworm): linux 6.1.192-1 (DLA-4906-1); Debian sid: 7.3.4-1
- Arch: linux 7.3.4.arch1-2, linux-lts 6.18.63-3
- Oracle: ELSA-2026-500630 to ELSA-2026-500632
Immediate actions
- Install the vendor kernel containing upstream fix 4e2b3ef19e0e ('inet: reject unhashed sockets during established lookups') and reboot
- Inventory internet-facing Linux hosts with 2+ RX queues or RPS enabled (ethtool -l <nic>; grep . /sys/class/net/<nic>/queues/rx-*/rps_cpus; cat /proc/sys/net/core/rps_sock_flow_entries)
- Check distro trackers (Debian, Red Hat, Ubuntu, Oracle) for the fixed package; an empty updateinfo result means no advisory yet, not that the host is safe
- Alert on unexpected kernel panics and reboots on hosts that run TCP listeners (nginx, apache, sshd, proxies)
Workarounds
- sysctl -w net.ipv4.tcp_syncookies=0 in every network namespace (including containers); blocks the demonstrated crash path but weakens SYN-flood protection and does not cover the secondary memory-corruption path
- Reduce the external NIC to one RX queue (ethtool -L <nic> combined 1) and disable RPS and RFS (rps_cpus=0, rps_sock_flow_entries=0); not persistent across reboots; test for performance impact first
Longer-term hardening
- Track unpatched LTS branches (5.10, 5.15, 6.1, 6.6) and arm64/POWER/RISC-V hosts that also need commit 5b5d8f351dd9
- Enable kernel crash dumps (kdump) and central logging so panics on edge servers are captured and correlated
- Place critical listeners behind redundant load-balanced nodes so a single-host panic does not cause an outage
Affected and fixed versions
- Linux — Linux kernel (TCP)
Vulnerable versions: 4.4 and later through 6.12.116; 6.18.x before 6.18.63; 7.3.x before 7.3.4; 7.2.x (branch ended at 7.2.15 without fix); 5.10, 5.15, 6.1, 6.6 (no fixed upstream release as of 2026-11-15)
Fixed in: 6.12.117; 6.18.63; 7.3.4; 7.4-rc2 - Debian — linux (kernel package)
Vulnerable versions: forky 7.2.15-1; earlier packages
Fixed in: trixie 6.12.117-1; bookworm 6.1.192-1; sid 7.3.4-1 - Red Hat — Red Hat Enterprise Linux 8, 9, 10 (kernel)
Vulnerable versions: 8; 9; 10 - Canonical — Ubuntu kernel 16.04 to 26.10
Vulnerable versions: 16.04 to 26.10 (needs evaluation as of 2026-11-15)
Timeline
- Commit 079096f103fa introduces the latent TCP established-lookup race into the Linux kernel.
- Upstream 7.2 branch ends at 7.2.15, before a fix existed, leaving it without a fixed release.
- Upstream fix commit 4e2b3ef19e0e 'inet: reject unhashed sockets during established lookups' lands.
- Stable kernels 6.12.117, 6.18.63 and 7.3.4 are released with the fix.
- Debian 12 (bookworm) backport changelog entry for the fix (6.1.192-1, DLA-4906-1).
- Hyunwoo Kim posts the NoisyTCP advisory to oss-security and GitHub after the distribution embargo ends.
- Cyber Security News, iTnews, SecurityOnline, FOSS Linux and others report NoisyTCP; Red Hat rates it 7.5 and Oracle 8.1; no in-the-wild exploitation reported.
Sources cited
- oss-security mailing list (NoisyTCP disclosure, 2026-11-15) openwall.com · Advisory · 2026-11-15
- Red Hat CVE page for CVE-2026-99288 access.redhat.com · Vendor report
- NoisyTCP: Linux Kernel TCP Pre-Auth Remote Corruption (CVE-2026-99288) - researcher advisory GitHub · Code repository · 2026-11-15
- CVE-2026-99288 (NoisyTCP): Linux Kernel TCP Pre-Auth Remote Memory Corruption Fixed in 6.12.117, 6.18.63 and 7.3.4 securityonline.info · Blog · 2026-11-16
- NoisyTCP (CVE-2026-99288): Check, Mitigate, Patch the Kernel computingforgeeks.com · Blog · 2026-11-16
- NoisyTCP (CVE-2026-99288): Kernel Panic Your Patch Missed fosslinux.com · Blog · 2026-11-16
- NoisyTCP: 11-Year-Old Linux Kernel Flaw Lets Unauthenticated Attackers Crash Servers Remotely cybersecuritynews.com · News · 2026-11-16
- Kernel-crashing NoisyTCP bug sat in Linux for 11 years itnews.com.au · News · 2026-11-16
Drafted by the Threadlinqs research pipeline from 8 sources; published ; last updated —.