Skip to main content
Mallory
MediumPublic exploitVulnerability

CVE-2026-99288: NoisyTCP

Published Nov 10, 2026 · Updated today
Trending this week#7of 428,812 vulnerabilities
Mention activity · 30 days125 mentions · peak 59 in a dayToday
CVSS5.9Medium · v3.1
EPSS0.16%5th percentile
CISA KEVNoNot listed
In the wildNonePublic exploit, no exploitation reported
IdentifiersCVE-2026-99288CWE-476

NoisyTCP is a race condition in the Linux kernel TCP networking subsystem, introduced in Linux 4.4. Lockless established-socket lookups use SLAB_TYPESAFE_BY_RCU caches, allowing socket memory to be recycled while readers retain its address. After acquiring a reference, __inet_lookup_established() and __inet6_lookup_established() revalidate the connection tuple without checking hash membership. A stale lookup can therefore return a replacement socket occupying the same address with the same tuple while it remains unhashed. In the demonstrated SYN-cookie path, the replacement request socket has a NULL rsk_listener, which tcp_v4_rcv() dereferences, causing a kernel panic. Unauthenticated remote attackers can trigger the race through crafted traffic to a reachable TCP listener.

Do you run this?See affected versions in your environment and whether adversaries are exploiting this vulnerability against assets like yours.
Check your exposure
Analyst brief

Impact, mitigation & remediation

What it means and what to do now: patch path, interim mitigations, and the assume-compromise checklist.

Impact

What an attacker gets, and what they’ve been doing with it.

Successful exploitation causes host-wide denial of service through a kernel panic, interrupting all applications and containers sharing the affected kernel. Default-configured web servers, SSH servers, and other TCP listeners can expose the vulnerable kernel path. A separate, harder-to-trigger child-socket race can cause a use-after-free, but general-purpose remote code execution, data theft, and privilege escalation have not been demonstrated. Exploitation timing is environment-dependent: reported laboratory crashes ranged from seconds or minutes on virtual machines to approximately 14 hours on a quiet bare-metal server.

Mitigation

If you can’t patch tonight, do this now.

Temporarily setting net.ipv4.tcp_syncookies=0 blocks the demonstrated SYN-cookie request-reconstruction path, but does not remove the underlying race or cover the separate child-socket use-after-free variant. Apply this setting separately in every relevant network namespace, including container namespaces; newly created namespaces also require attention. Disabling SYN cookies weakens SYN-flood protection. Alternatively, serialize receive processing by configuring one receive queue on externally reachable interfaces and disabling Receive Packet Steering and Receive Flow Steering. This can substantially reduce throughput, may not be supported by every interface, and must remain effective after reboot or interface reconfiguration. Restrict unnecessary TCP exposure while awaiting patch deployment.

Remediation

Patch, then assume compromise.

Install a vendor-supported kernel update explicitly addressing CVE-2026-99288, reboot into it, and verify the running kernel. Upstream releases containing the primary correction include 6.12.117, 6.18.63, 7.3.4, and 7.4-rc2; distribution backports require checking vendor advisories rather than relying solely on upstream version comparisons. Debian 13 includes the correction in linux 6.12.117-1. The correction rejects unhashed sockets after reference acquisition, releases the reference, and restarts the lookup. It also replaces direct request-socket frees with reqsk_put() in affected SYN-cookie, Fast Open, and request-migration failure paths.

Public exploits

Exploits

3 exploits validated by Mallory's analysis. Hidden: 2 rejected by Mallory's analysis as not a working exploit.

View more in app →
noisytcp-pocThe repository contains three files: noisytcp.c, a standalone C network-race trigger; Makefile, which builds it with C11 and pthreads and provides debug/clean targets; and README.md, which describes the alleged vulnerability, prerequisites, usage, and limitations. It is not associated with an exploit framework. The intended attack is a pre-authentication remote denial of service against Linux TCP request-socket handling, attributed by the repository to CVE-2026-99288. Its stated mechanism combines SYN-cookie processing, request-socket allocation reuse, and concurrent fragmented lookup traffic to make an established-socket lookup accept an unhashed request with a NULL listener. The documented practical outcome is a kernel panic, not RCE. Modes include probe, verify, race, and an alternative TIME_WAIT-based setup. Visible main() code parses arguments, initializes networking, starts a packet-sniffer thread, measures SYN-ACK timing, checks local RST leakage, and dispatches the selected mode. Documentation additionally describes backlog calibration, ARP handling, cookie harvesting, and timing controls. This is more than a detection script and supplies an attack packet sequence, but successful exploitation is highly timing-dependent. Both the README and C source are explicitly truncated in the supplied content, so the complete implementation could not be audited or validated. The CVE assignment, fix commit, affected versions, and successful exploitation are repository claims, not independently established facts. No evidence of a fake exploit, exfiltration destination, or unrelated destructive command is visible; Makefile's rm -f only removes the built executable. The original repository URL, git reference, and archive size were not supplied; empty strings and size 0 represent unavailable metadata.re4 · Disclosed Nov 17, 2026 · c · makefile · networkOperationalCVE-2026-99288---NoisyTCP-ExploitThe supplied repository contains one standalone C source file, noisytcp.c, reported as 15,759 bytes. It is not associated with an exploit framework. The source presents itself as a denial-of-service exploit for CVE-2026-99288, but the CVE attribution, affected versions, and successful exploitation are not independently established by the supplied material. A substantial middle portion of the file is omitted, preventing complete implementation review. The documented mechanism is an ABA-style request-socket reuse race in __inet_lookup_established(). A socket lookup allegedly accepts a replacement SYN-cookie request with the same tuple without checking whether it remains hashed. That replacement has no listener, potentially leading tcp_v4_rcv() to dereference NULL. The visible main-loop fragment cycles SYN-backlog entries, creates a victim request, synchronizes two workers, sends a delayed fragmented SYN, and refills the queue. Fragmentation is intended to steer competing packets to different receive-processing CPUs. Worker names indicate plain-packet, fragmented-packet, and monitoring threads, but their implementations are omitted. The program also inserts an iptables OUTPUT rule to drop TCP RST packets matching its selected source port and attempts to remove that rule on normal completion. An optional check branch calls do_check(), whose behavior cannot be assessed from the excerpt. No fixed IP address, domain, URL, callback service, or exfiltration destination is visible; the destination is supplied by the operator. No evidence of a deceptive secondary payload appears in the visible code, although the omitted portion prevents a definitive assessment. Repository URL, revision, and archive size were not provided; empty strings and zero represent unavailable metadata.mustafasaydam · Disclosed Nov 17, 2026 · c · networkOperationalnoisytcp-labThis is a standalone research lab and attempted reproducer for the repository-claimed CVE-2026-99288, not a demonstrated end-to-end exploit. The proposed flaw is an ABA lifetime race in __inet_lookup_established(): a request_sock address may be freed and reused for an unhashed SYN-cookie request with the same TCP four-tuple, allowing lookup to return an object whose listener is NULL. The intended impact is an unauthenticated kernel crash, not RCE. The main entry point, attacker/noisytcp.c, implements raw IPv4/TCP packet construction, fragmentation, SYN-cookie handling, configurable timing and workload generation, and probe, verify, race, tw, and fragtest modes. Fragmented lookup traffic is intended to execute on a different RPS-selected CPU from the RST/free and cookie-ACK/reallocation traffic. The tw mode uses TCP TIME_WAIT transitions to create the request socket needed for the race. No shellcode or post-exploitation payload is supplied. The 28-file repository includes 11 executable source/script files: two C files, six bpftrace programs, two shell scripts, and one Python listener. attacker/*.bt measures request-socket address reuse, lookup state, unhashed objects, fragmentation, and SYN-cookie/request-queue activity. Some probes use BTF structures; det.bt instead uses hardcoded offsets that may be kernel-layout dependent. jit/noisyjit.c and its Makefile provide an optional kernel timing module using pinned hardirq timers to delay selected CPUs; this is instrumentation, not an exploit payload. Dockerfiles, two Compose configurations, victim/nginx.conf, and victim/entrypoint.sh establish container or host-targeting labs. run.sh orchestrates setup, verification, tracing, race execution, counter comparisons, and panic-log searches. victim8081.py supplies a host TCP accept-and-close service. Nine files under results/ preserve measurements and run logs; the remaining top-level files provide documentation, licensing, and ignore rules. The evidence supports substantial SYN-cookie processing, request-socket reuse, and cross-CPU traffic distribution. An older trace records three returns of unhashed TCP_NEW_SYN_RECV objects, but does not establish a NULL listener or resulting panic and uses a probe revision different from the supplied aba.bt. final_race.log explicitly reports no panic and includes intermittent sendto permission errors. The README likewise states that the final crash was not reproduced on bare metal, including with artificial CPU delays. Accordingly, this is POC maturity despite extensive instrumentation, and observed precursor events should not be equated with successful exploitation. The actual target is the shared Linux kernel, not nginx, Python, or SSH themselves: a successful crash from this privileged-container lab could affect its host or Docker VM. No exfiltration, command-and-control, persistence, or remote code execution is evident in the visible material. The README's CVE attribution, fix commit, broad affected-version claim, and claimed fixed releases 7.3.4+, 6.18.63+, and 6.12.117+ are unverified repository assertions. Analysis is static and limited by truncation of the main C source, README, and several logs. Repository URL, git reference, and archive size were not supplied; empty strings and zero represent unavailable metadata.h0rn3t · Disclosed Nov 16, 2026 · c · bpftrace · networkPoC
FAQ

Frequently asked questions

Is CVE-2026-99288 exploited in the wild?
As of Nov 18, 2026, Mallory has no reports of CVE-2026-99288 being exploited in the wild. It is not listed in CISA KEV.
Is there a public exploit for CVE-2026-99288?
Yes. Mallory tracks 3 public exploits or proofs of concept for CVE-2026-99288, the earliest disclosed Nov 16, 2026.
Is there a patch for CVE-2026-99288?
Install a vendor-supported kernel update explicitly addressing CVE-2026-99288, reboot into it, and verify the running kernel. Upstream releases containing the primary correction include 6.12.117, 6.18.63, 7.3.4, and 7.4-rc2; distribution backports require checking vendor advisories rather than relying solely on upstream version comparisons.
How severe is CVE-2026-99288?
CVE-2026-99288 has a CVSS v3.1 score of 5.9 (Medium) and an EPSS score of 0.16% (5th percentile) as of Nov 18, 2026.
Activity feed

Recent activity

127 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.

View more in app →
socprime blogNews · Nov 17, 2026CVE-2026-99288: NoisyTCP Linux Kernel DoSA race condition in Linux kernel lockless TCP established-socket lookups allows reused socket memory to be referenced before the replacement socket is hashed. Specially crafted TCP traffic can trigger a NULL pointer dereference and kernel panic, causing unauthenticated remote denial of service and potentially disrupting all containers sharing the host kernel. Exploitation requires concurrent receive contexts. The issue dates to Linux 4.4 and is rated Important with a CVSS 3.1 score of 7.5. Public research demonstrates denial of service, not arbitrary code execution. Fixed stable versions include 6.12.117, 6.18.63, and 7.3.4; distribution backports may also provide protection. Temporary mitigations include disabling SYN cookies or using one RX queue with RPS and RFS disabled, subject to security and performance trade-offs.Read more ↗uganda certNews · Nov 17, 2026“NoisyTCP” 11-Year-Old Linux Kernel Flaw Crashes Servers Running Default Nginx, Apache, and SSH (CVE-2026-99288) | Uganda National Computer Emergency Response Team-National CertAn unauthenticated remote denial-of-service vulnerability in the Linux kernel’s TCP networking subsystem. Crafted TCP packets exploit a timing gap between SYN cookie handling and socket hashing, triggering a NULL pointer dereference that crashes the entire operating system. The content describes broad exposure across Linux distributions, including servers running default Nginx, Apache, and OpenSSH configurations. Remediation requires applying the distribution’s kernel patch and rebooting into the patched kernel.Read more ↗scworldNews · Nov 17, 2026Security researcher identifies critical NoisyTCP vulnerability in Linux kernel | brief | SC MediaAn unauthenticated remote denial-of-service vulnerability in the Linux kernel's TCP networking code. A race condition during traffic processing causes a use-after-free error and kernel panic on systems processing traffic across multiple receive queues or CPUs. The report states that nearly all Linux distributions are affected, container-triggered crashes can take down the underlying host, and virtual machines crashed in testing in as little as five seconds. Red Hat assigned a CVSS score of 7.5. An upstream fix was committed on November 3, but administrators must verify distribution-specific update availability. Temporary mitigations include disabling TCP SYN cookies or restricting packet processing to a single queue, with potential security or performance trade-offs.Read more ↗almalinuxNews · Nov 16, 2026NoisyTCP (CVE-2026-99288): Patched kernels available in testing | AlmaLinux OSAn unauthenticated remote denial-of-service vulnerability in Linux kernel TCP socket lookup, introduced in Linux 4.4. Incomplete validation during lockless lookup can return a reused connection-request object, causing a kernel panic through a missing listener. Exploitation requires incoming packets to be processed on multiple CPUs; a host crash also stops its containers. A second, harder-to-trigger variant causes a use-after-free, but the cited assessment finds no code execution or data alteration. Patched kernels are available in AlmaLinux testing repositories for releases 8, 9, and 10; no fixed RHEL kernel was available at publication. Temporary mitigations include disabling SYN cookies, which covers only the described crash variant and removes SYN-flood protection, or restricting incoming packet processing to one CPU at a performance cost.Read more ↗
+18 more in the timeline. Sign up to see them
Do you run this?See affected versions in your environment and whether adversaries are exploiting this vulnerability against assets like yours.
Check your exposure
Mallory platform

The version that knows your environment.

This page is what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which adversaries are exploiting it right now, and which detections to deploy tonight.
Exposure mappingQuery your assets running an affected version, and investigate the blast radius.
Threat actor evidenceEvery report linking this CVE to a named adversary.
Associated malwareMalware families linked to this CVE, with evidence and IOCs.
Detection signaturesYARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Vendor-by-vendor mappingCross-references every affected SKU, including bundled OEM variants.
Social activity102Community discussion across Reddit, Mastodon, and other social sources.