NoisyTCP: An 11-Year-Old Linux Kernel Race Lets Anyone Crash a Server Through Nginx, Apache or SSH. Debian Has the Fix. Ubuntu and RHEL Do Not Yet
If you run Linux servers that accept TCP connections from the internet, put CVE-2026-99288, which its finder calls NoisyTCP, on this week's patch list. On Debian 12 or 13, install the kernel from the security archive now and reboot. On Ubuntu or RHEL there is no fixed kernel yet, so decide today whether your internet-facing hosts can afford one of the two stopgaps below, and watch for the vendor kernel. The flaw lets someone with no login crash the whole machine by sending traffic to any listening TCP port, including default Nginx, Apache and SSH. Nobody has reported attacks, and the demonstrated impact is a crash, not a takeover.
This is a different bug from the two Linux TCP stories we covered this month. CVE-2026-98432 (our posts of November 2 and November 4) is a use-after-free reached through IPv4 Record Route and Timestamp options. NoisyTCP is a race in how the kernel looks up established connections, has its own CVE, and has its own fix.
What is confirmed
The CVE record, published by the Linux kernel's CVE team on Tuesday, November 10, describes the flaw. When a TCP packet arrives, the kernel looks up which connection it belongs to without taking a lock. The memory behind those connection objects can be reused while a lookup still holds an old address. The lookup rechecks the connection's addresses and ports after taking a reference, but it never checks that the object is still in the table. If the memory has just been reused for a new SYN-cookie request with the same addresses and ports, the lookup returns that half-built object, and the kernel reads through a NULL pointer and panics. The record includes the crash trace from the proof of concept.
The researcher, Hyunwoo Kim, published the full write-up on the oss-security mailing list on November 15 (UTC), after the embargo with the Linux distribution maintainers ended. He is the same researcher behind Dirty Frag and Januscape earlier this year. Per his post:
Who is exposed. Any server with a plain TCP listener, with no special firewall rules or socket options, as long as incoming packets are spread across more than one CPU: two or more receive queues on the network card, or RPS turned on. Most modern servers and cloud instances have multiple queues. He says the bug was latent for about 11 years and affects "almost all Linux distributions". That is his estimate, not a count anyone has measured.
What it does. The demonstrated result is a kernel panic. On a container host, the whole host goes down, not only the container that was hit. He describes a second path that could lead to memory corruption, but calls it generally hard to trigger without special techniques. No source we read shows code execution.
How long it takes. The race is narrow. He reports 5 seconds to 2 minutes against QEMU/KVM virtual machines, where busy hosts and vCPU preemption widen the window, and about 14 hours against a quiet bare-metal Nginx server with 8 receive queues, even with a tuned exploit. Virtual machines are the easier target.
The bug entered the kernel in a commit dated October 3, 2015 (Linux 4.4). The fix, "inet: reject unhashed sockets during established lookups", is dated November 3, 2026, and is in mainline 7.4-rc2 and the stable releases 6.12.117, 6.18.63 and 7.3.4.
Fix status by distribution
As read from each tracker on Monday, November 16:
Debian 12 (bookworm): fixed in the security archive, kernel 6.1.192. The base release is still at a vulnerable 6.1.176, so you need the security update.
Debian 13 (trixie): fixed in the security archive, kernel 6.12.117. Debian unstable is fixed at 7.3.4; testing (forky) was still vulnerable.
Ubuntu 16.04 through 26.10: every supported release, including the AWS, Azure and GCP kernels, is marked "Needs evaluation". No fixed kernel. Ubuntu rates it Medium priority.
Red Hat Enterprise Linux 8, 9 and 10, and OpenShift 4 and 5: marked Affected, with no fixed kernel released. RHEL 6 and 7 are marked Not affected. Red Hat rates it Important, with a draft CVSS score of 7.5. Its vector says low privileges are required, which does not match its own description of an unauthenticated attacker; we go by the description and the researcher.
The bug is not in CISA's Known Exploited Vulnerabilities catalog as of its November 13 release.
What is not known
Whether anyone is using it. No source we read reports attacks.
Whether a usable exploit is public. The researcher's GitHub page describes "a very limited, lab-specific exploit" aimed at virtual machines and says he is still working on a general one. We found no exploit code there, and our exploit harvester holds no proof-of-concept repository for this CVE as of today.
How many servers are exposed. Nobody has published a count, and there cannot be a simple one: the condition is "has a listening TCP port and more than one receive queue", which describes most internet-facing Linux.
When Ubuntu and Red Hat will ship fixed kernels.
What to do
Patch where you can, and reboot. Debian 12 and 13 have fixed kernels in the security archive now. A kernel update does nothing until the machine boots into it; check the running version with uname -r afterward. If you build your own kernels, take 6.12.117, 6.18.63, 7.3.4 or later.
On Ubuntu and RHEL, watch for the kernel and patch the exposed hosts first. Start with internet-facing virtual machines and container hosts, where the race is fastest and a crash takes the most down with it.
Decide on a stopgap for anything that cannot wait. The researcher and Red Hat give the same two options, and both have a cost:
Know what an attack would look like. The visible result is an unexplained kernel panic or reboot on a host that takes internet traffic, with a NULL pointer dereference in tcp_v4_rcv in the crash log. If you collect kernel crash logs, alert on that.
How we know
We read the CVE record for CVE-2026-99288 from the CVE Services API, the researcher's oss-security post of November 15, his GitHub write-up, the Debian security tracker, the Ubuntu CVE tracker, Red Hat's security data API and CVE page, and the CISA KEV JSON feed (catalog release of November 13), all on November 16. BleepingComputer's report of November 16 was the lead; every technical fact here comes from the researcher's post, the CVE record or the distribution trackers.
Our edge instruments watch HTTP requests to our own web properties. This bug is triggered by the timing of raw TCP packets, not by any request path or user agent, so our ttp-check tool has nothing to match and cannot observe it. That is a blind spot, not a clean result. We hold no indicators for it.
In our feed
Nothing: none published in the sources we read. A kernel race triggered by ordinary connections has no attacker infrastructure to list until someone uses it.
Sources
Hyunwoo Kim on oss-security: NoisyTCP: Linux Kernel TCP Pre-auth Remote Corruption (CVE-2026-99288)
Hyunwoo Kim: NoisyTCP on GitHub
CVE Program: CVE-2026-99288
Linux kernel: fix commit 4e2b3ef19e0e
Debian: security tracker, CVE-2026-99288
Ubuntu: CVE-2026-99288
Red Hat: CVE-2026-99288
BleepingComputer: New NoisyTCP Linux kernel flaw lets attackers crash servers
DugganUSA, November 2: Linux TCP CVE-2026-98432 remote root
DugganUSA, November 4: Two Linux root bugs in one week
If this was useful, or if it wasn't, the rating widget below takes ten seconds and is the only way we know.
Every indicator in this post is in the feed. Free.
1.58M+ IOCs, STIX 2.1 / TAXII, 88% novel vs ThreatFox, exploited-CVE leads ahead of CISA. No credit card — a free API key in 30 seconds, and you can audit every claim above against the live endpoints.



Comments