TRENDING
A black manual telephone switchboard with rows of orange jacks and red patch cords plugged into the lower row, above a rotary dial and a handset
November 16, 2026
How to Route Alerts in Alertmanager So Pages Reach the On-Call and Tickets Reach the Queue, and Test Grouping, Inhibition, and Silences Before an Outage Does
A brass nail with a threaded screw section lying on graph paper, a fastener that is both nail and screw
November 16, 2026
Ai2’s XYBench Turns the XY Problem Into a Benchmark, but the Model That Drafted Every Rubric Also Judges the Answers, the Human Baseline Is Graded on Rubrics Read Off Its Own Answers, and the Emphasis Judge Disagrees With the Annotators on 20 to 26 Percent of Validation Responses
An amber warning beacon on a white surface, a flashing-light stand-in for the FBI’s FLASH alert
November 16, 2026
The FBI’s ARTEX Warning Says a Supply Chain Compromise “Enabled Access” to 2,250 Tenants Where Microsoft Counted Sign-In Attempts and Cites “Over 1,600” Consoles Where Censys Counted 1,465, and a Microsoft Post It Cites Mistypes One Indicator in Its Hunting Query
Two eggs and a colour-changing egg timer in a saucepan of water on an induction hob
November 16, 2026
How to Set gRPC Deadlines and Retries in Python, Pass Them Down a Call Chain, and Stop Retries From Charging a Card Twice
A whole lemon and a halved lemon side by side on a black background
November 16, 2026
IBM’s LLMON Turns Prompt-Injection Defense Into a Token Layout, but the “Nearly Zero” Chart Is One Benchmark, the Paper’s Own Red Team Lets 15% to 20% of Indirect Attacks Through, and the Shipped Template Lets a Document Type the Delimiters
16 Nov 2026
SXZ.io SXZ.io
  • Home
Search the Site
Popular Searches:
Technology Amazon AI
Recent Posts
A manual paper cutter with a wooden grid board, a clear plastic guard and a long cutting arm, a picture of text being trimmed to fit
How to Stop Ollama From Silently Cutting Your Prompt, Find the Real Context Window, and Set num_ctx Without Running Out of Memory
November 16, 2026
A brown llama and her gray cria walking across a salt flat beside a red lake in Bolivia
llama.cpp’s v1.0.0 Promises Major Bumps for API and ABI Breaks, but Its Rule Table Lists Only the C API, libmtmd Shares the Version While Its Header Warns of Breaking Changes, and the Release Page Carries a 7-Byte Pointer to a Nightly Build Instead of Binaries
November 16, 2026
A stack of chocolate chip cookies on a white plate beside a cup of tea
NoisyTCP Can Crash a Linux Server Over the Network, but the Fix Has Reached Only Three of the Seven Kernel Lines Still Being Released, Red Hat and Ubuntu Have No Update, and the Researcher’s Own Timings Run From 5 Seconds to 14 Hours
November 16, 2026
SXZ.io SXZ.io
  • Home

Categories

Articles 302 Posts
News 307 Posts
Learning Hub 274 Posts
Home/News/NoisyTCP Can Crash a Linux Server Over the Network, but the Fix Has Reached Only Three of the Seven Kernel Lines Still Being Released, Red Hat and Ubuntu Have No Update, and the Researcher’s Own Timings Run From 5 Seconds to 14 Hours
News

NoisyTCP Can Crash a Linux Server Over the Network, but the Fix Has Reached Only Three of the Seven Kernel Lines Still Being Released, Red Hat and Ubuntu Have No Update, and the Researcher’s Own Timings Run From 5 Seconds to 14 Hours

A kernel TCP race that can panic a server over the network has had a mainline fix since November 3, but only three of seven kernel.org lines carry it, Red Hat and Ubuntu have no update, and the...

November 16, 2026 15 Min Read
10

Hyunwoo Kim, whose October disclosure of six kernel bugs we covered as Infinity Core, published an advisory on November 15 for NoisyTCP (CVE-2026-99288), an 11-year-old race in the Linux kernel’s TCP code that he says can be triggered “pre-auth and remotely” and that panics the host. The fix has been in the mainline kernel since November 3. When we checked on November 16, it had reached only three of the seven kernel.org lines that still get releases (6.12.117, 6.18.63 and 7.3.4), Red Hat listed RHEL 8, 9 and 10 as affected with no update, and Ubuntu’s data showed every release from 16.04 to 26.10 as needs-triage.

Table Of Content

  • What the advisory says
  • Where the headline is wider than the advisory
  • The timeline
  • Which kernels have the fix
  • Where each distribution stands
  • Three scores, none for a crash only
  • What to do now
  • Check what you run
  • If you cannot patch yet
  • Look for crashes
  • What we could not confirm
  • What we checked

BleepingComputer’s report of November 16 says NoisyTCP “affects almost all Linux distributions and lets attackers crash systems running Nginx, Apache, and SSH servers with default configurations.” The advisory supports that in outline, but it is narrower in four ways that change how fast to move, and the patch picture is more uneven than the report’s “patch is already available” suggests. It follows our earlier coverage of kernel bugs: Infinity Core, the skb zerocopy bug, CVE-2026-98432 and CVE-2026-98276.

What the advisory says

The kernel looks up established TCP connections without taking a lock, and the memory for the socket objects involved can be reused while a reader still holds an old pointer. The kernel team’s CVE text puts it this way: “TCP established ehash lookups are lockless.” In the flaw, a lookup that had loaded the address of an old SYN cookie request socket can take a reference to a new request socket that reuses the same address and passes the tuple check. The text continues: “The new request socket is unhashed and has a NULL rsk_listener.” The kernel then reads through that missing listener and panics. The fix, commit 4e2b3ef19e0e, checks after taking the reference whether the socket is unhashed and, if it is, drops the reference and restarts the lookup.

  • Two paths. The advisory lists two ways to trigger it. The first, a “request_sock Reuse ABA Race,” ends in the NULL dereference and a remote denial of service. The second, a “Full child socket ABA Race,” “can trigger a remote use-after-free, but since complex race scenarios have to succeed in a nested way, it is generally hard to trigger without using special techniques.”
  • Conditions. The README of the researcher’s GitHub repository has an FAQ that the mailing-list post lacks. It says the target needs no special configuration, but the race “only needs a server with two or more RX queues, or with RPS enabled.” The mailing-list post adds that attacker packets have to be spread across receive contexts, using the difference in receive hash between IPv4 fragments and regular TCP packets. In the README’s words, the flaw “triggers even when the server does nothing more than call” listen().
  • Time to trigger. “Against Nginx, Apache, and SSH servers running with their default configuration on VMs such as QEMU/KVM, where vCPU preemption can occur, it took about 5 seconds to 2 minutes to trigger.” On a quiet bare-metal server it was far slower: “Against Nginx running with its default configuration on a quiet bare-metal server (8 RX queues), it took about 14 hours to trigger, even though a PoC with race timing carefully tuned for the target environment was used.”
  • Containers. “Because this is a Linux kernel TCP vulnerability, if the DoS is triggered against a web server running in a container, it takes down not only the web server application but the entire host.”

Where the headline is wider than the advisory

  • Default configurations. By the README’s own conditions, a host with one receive queue and RPS off is outside the race, and that is the setup the advisory’s second mitigation creates. The kernel’s scaling document says RPS “is disabled when it is zero (the default)” and that it “remains disabled until explicitly configured,” so in practice the condition is a NIC with several receive queues. The advisory does not say how many typical hosts meet it.
  • Crash systems. The range runs from seconds to hours. The 5-second to 2-minute figures are for QEMU/KVM guests, where the hypervisor can preempt a virtual CPU. The 14-hour figure is a quiet bare-metal server with a PoC tuned for it. The README adds that “the race window is narrow, so triggering it against a quiet bare-metal server can take a long time.” The advisory gives no figure for a busy bare-metal host.
  • No data alteration or code execution. BleepingComputer says exploits “don’t allow threat actors to alter data or gain arbitrary code execution.” That matches the statement on Red Hat’s CVE page: “While this flaw causes a kernel panic that disrupts host availability, it does not allow data alteration or arbitrary code execution.” The researcher is less categorical. Asked in the README whether this is a TCP RCE, he answers: “Not yet. and basically, no. So far I have only written a very limited, lab-specific exploit that targets VM servers. I am still researching a general-purpose exploit.” He adds: “But remote DoS can be achieved without much difficulty.” He does not say what the lab exploit achieves, and his second trigger path is described as a remote use-after-free. Red Hat’s statement is categorical about the flaw as a whole, while the researcher’s text calls the second path a use-after-free that is hard to trigger and does not say it cannot be exploited.
  • Almost all distributions. The README says the flaw “affects almost all Linux distributions.” Red Hat’s data marks RHEL 6 and 7 as not affected, which fits the record: the bug arrived with a change committed on October 3, 2015, 079096f103fa (“tcp/dccp: install syn_recv requests into ehash table”), in the kernel that became 4.4. RHEL 8, 9 and 10 are affected.

The timeline

When (UTC) What the record shows Where we read it
2015-10-03, 11:32 Eric Dumazet’s commit 079096f103fa is committed (authored October 2); the record names it as the origin GitHub commit data, CVE record
Sep 25, 10:25 The CVE ID is reserved CVE record
Oct 29, 21:42 Kim’s patch is posted (the date in the commit and in its message link) Commit text
Nov 3, 14:33 Paolo Abeni commits the fix to mainline as 4e2b3ef19e0e, reviewed by Eric Dumazet and Kuniyuki Iwashima GitHub commit data
Nov 6 Stable releases 6.6.162, 6.1.192, 5.15.225 and 5.10.274 ship without the patch kernel.org ChangeLog files
Nov 10, 18:05 to 18:10 Stable releases 6.12.117, 6.18.63 and 7.3.4 ship with the backport kernel.org ChangeLog files
Nov 10, 18:19 The kernel team publishes the CVE record; Red Hat’s public date is the same minute CVE record, Red Hat CVE data
Nov 10, 19:22 and 19:36 NVD creates its entry (status Received); Red Hat opens bug 2570103 NVD, Red Hat Bugzilla
Nov 11 Ubuntu’s CVE page is published with needs-triage and no priority Ubuntu CVE data
Nov 12 and 13 Oracle publishes UEK errata ELSA-2026-500630 and ELSA-2026-500632 (Nov 12) and ELSA-2026-500631 (Nov 13) Oracle Linux CVE page
Nov 15, 16:00 Kim posts the advisory to oss-security; his GitHub repository is created at 16:33 oss-security, GitHub
Nov 16, 09:30 BleepingComputer publishes its report BleepingComputer

By our arithmetic, the mainline commit came 12 days before the advisory and the CVE record nearly 5 days before it. Both describe the SYN cookie path and carry the crash trace. The queue condition, the timings and the mitigations come from the advisory.

Which kernels have the fix

Line Latest release (UTC date) Has the patch
7.4 (mainline) 7.4-rc3 (Nov 15) Yes, since 7.4-rc2
7.3 7.3.4 (Nov 10) Yes
7.2 7.2.15 (Nov 1, end of life) No
6.18 (longterm) 6.18.63 (Nov 10) Yes
6.12 (longterm) 6.12.117 (Nov 10) Yes
6.6 (longterm) 6.6.162 (Nov 6) No
6.1 (longterm) 6.1.192 (Nov 6) No
5.15 (longterm) 5.15.225 (Nov 6) No
5.10 (longterm) 5.10.274 (Nov 6) No

The four older long-term lines have had no release since November 6. Their latest releases are small ones (each ChangeLog is about 3 KB, and the one for 6.1.192 holds Xen fixes), and none contains the patch. The CVE record marks every kernel from 4.4 up to the fixed releases as affected and lists no fixed version for those four lines. Distribution kernels on them carry fixes as backports, which a version number does not show.

Where each distribution stands

Source What its page showed on November 16 (10:05 to 10:15 UTC)
Red Hat Important. RHEL 8, 9 and 10 (kernel, plus kernel-rt on 8 and 9) and the OpenShift 4 and 5 node images are listed as affected; RHEL 6 and 7 are not affected. The CVE data lists no errata, and Bugzilla bug 2570103 (opened Nov 10) was still NEW.
Ubuntu CVE data published Nov 11, with no priority and no kernel notices. The linux package is needs-triage on every release from 16.04 (xenial) to 26.10 (stonking) and for upstream; 14.04 is marked ignored.
Debian Fixed in bookworm security 6.1.192-1, trixie security 6.12.117-1 and sid 7.3.4-1 (DSA-6628-1); forky 7.2.15-1 is vulnerable. Upstream 6.1.192 does not carry the patch, so the bookworm fix must come from Debian’s own patch, which we did not read.
Oracle Important. UEK errata ELSA-2026-500630, ELSA-2026-500631 and ELSA-2026-500632 cover Oracle Linux 7 to 10 (Nov 12 and 13); only UEK kernels are listed.
NVD Status Received, with no score.

We could not load SUSE’s or Amazon Linux’s pages from here, so they are not in the table.

Three scores, none for a crash only

Source Score Vector
Red Hat (draft) 7.5, Important AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H
Oracle (preliminary) 8.1 AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
NVD None yet The CVE record carries no metrics
Our recomputation: a crash only, no privileges, high complexity 5.9 AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H

Both vendor vectors rate confidentiality and integrity impact High, which does not match Red Hat’s own statement that the flaw “does not allow data alteration or arbitrary code execution.” Red Hat’s vector also says low privileges are required, while its statement describes “an unauthenticated remote attacker” acting “without requiring privileges.” Both are labelled draft or preliminary, so we do not read them as final. We recomputed with the CVSS 3.1 equations from FIRST’s specification in a short standard-library script, and its results match Red Hat’s 7.5 and Oracle’s 8.1:

# cvss_check.py: CVSS 3.1 base score from the FIRST specification equations (stdlib only)
import math
W = {"AV": {"N": 0.85, "A": 0.62, "L": 0.55, "P": 0.2}, "AC": {"L": 0.77, "H": 0.44},
     "UI": {"N": 0.85, "R": 0.62}, "CIA": {"H": 0.56, "L": 0.22, "N": 0.0}}
PR = {"U": {"N": 0.85, "L": 0.62, "H": 0.27}, "C": {"N": 0.85, "L": 0.68, "H": 0.5}}

def roundup(x):
    i = round(x * 100000)
    return i / 100000.0 if i % 10000 == 0 else (math.floor(i / 10000) + 1) / 10.0

def score(vec):
    m = dict(p.split(":") for p in vec.split("/")[1:])
    iss = 1 - (1 - W["CIA"][m["C"]]) * (1 - W["CIA"][m["I"]]) * (1 - W["CIA"][m["A"]])
    imp = 6.42 * iss if m["S"] == "U" else 7.52 * (iss - 0.029) - 3.25 * (iss - 0.02) ** 15
    ex = 8.22 * W["AV"][m["AV"]] * W["AC"][m["AC"]] * PR[m["S"]][m["PR"]] * W["UI"][m["UI"]]
    if imp <= 0:
        return 0.0
    return roundup(min(imp + ex, 10)) if m["S"] == "U" else roundup(min(1.08 * (imp + ex), 10))

rows = [("Red Hat draft", "CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H"),
        ("Oracle preliminary", "CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H"),
        ("Crash only, no privileges, high complexity", "CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H"),
        ("Crash only, no privileges, low complexity", "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H")]
for name, v in rows:
    print(f"{score(v):>4}  {v}  {name}")

It prints one line per vector, with the score first:

 7.5  CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H  Red Hat draft
 8.1  CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H  Oracle preliminary
 5.9  CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H  Crash only, no privileges, high complexity
 7.5  CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H  Crash only, no privileges, low complexity

The same vector with low attack complexity scores 7.5, so the 5.9 rests on treating the narrow race as high attack complexity, which is a judgment call. FIRST’s EPSS put the probability of exploitation at 0.16 percent (4.8th percentile) on November 15. The CVE is not in CISA’s Known Exploited Vulnerabilities catalog (version 2026.11.13), and the sources we read report no exploitation in the wild.

What to do now

Check what you run

Install a kernel with the fix: 6.12.117, 6.18.63, 7.3.4 or 7.4-rc2 and later from kernel.org, or your distribution’s own fixed package from the table above. If you run a distribution kernel, the version number cannot tell you, because vendors backport. The script below is ours, read-only and standard-library only. It compares an upstream version string with the fixed releases, answers UNKNOWN for any string that is not a plain upstream version, and reads the settings the advisory ties to the race: the SYN cookie sysctl and the receive-queue and RPS settings. Run it on the host, not in a container, because the sysctl is per network namespace.

# noisycheck.py
"""Read-only look at the three things the NoisyTCP (CVE-2026-99288) advisory says matter on one Linux host.

1. The kernel release against the upstream releases that carry the fix. Only kernel.org kernels can
   be judged by version number. Distribution kernels backport fixes under version strings that do
   not change, so anything that is not a plain x.y.z string is reported as UNKNOWN on purpose.
2. net.ipv4.tcp_syncookies in this network namespace (the advisory's first mitigation sets it to 0).
3. Receive queues and RPS per interface (the advisory says the race needs two or more RX queues,
   or RPS spreading packets across CPUs).

Usage: python3 noisycheck.py [--root DIR] [release ...]
Run it on the host, not inside a container: the sysctl is per network namespace.
"""
import argparse
import glob
import os
import platform
import re

# First upstream release on each line that contains the fix (from the CVE record and the ChangeLog files).
FIRST_FIXED = {(6, 12): 117, (6, 18): 63, (7, 3): 4}
MAINLINE = (7, 4)      # the fix is in 7.4-rc2, so 7.4-rc2 and everything newer has it
MAINLINE_RC = 2

PLAIN = re.compile(r"^(\d+)\.(\d+)(?:\.(\d+))?(?:-rc(\d+))?\+?$")


def check_release(release):
    m = PLAIN.match(release.strip())
    if not m:
        return "UNKNOWN: not a plain upstream version string, ask your vendor"
    major, minor = int(m[1]), int(m[2])
    patch = int(m[3]) if m[3] is not None else 0
    rc = int(m[4]) if m[4] is not None else None
    line = (major, minor)
    if line < (4, 4):
        return "NOT AFFECTED (upstream): older than 4.4, which is where the record starts"
    if line > MAINLINE:
        return "FIXED: newer than 7.4"
    if line == MAINLINE:
        if rc is not None and rc < MAINLINE_RC:
            return f"VULNERABLE: 7.4-rc{rc} is older than 7.4-rc{MAINLINE_RC}"
        return "FIXED: 7.4-rc2 and later have the patch"
    if line in FIRST_FIXED:
        if rc is not None:
            return "UNKNOWN: release candidate of a stable line"
        need = FIRST_FIXED[line]
        if patch >= need:
            return f"FIXED: {major}.{minor}.{need} was the first release with the patch"
        return f"VULNERABLE: update to {major}.{minor}.{need} or later"
    return "NO UPSTREAM FIX: no release on this line has the patch, ask your vendor"


def read(path):
    try:
        with open(path, encoding="utf-8") as f:
            return f.read().strip()
    except OSError:
        return None


def rx_report(root):
    rows = []
    for dev in sorted(glob.glob(os.path.join(root, "sys/class/net/*"))):
        name = os.path.basename(dev)
        if name == "lo":
            continue
        queues = sorted(glob.glob(os.path.join(dev, "queues/rx-*")))
        rps_on = []
        for q in queues:
            mask = read(os.path.join(q, "rps_cpus")) or "0"
            if mask.replace(",", "").strip("0"):
                rps_on.append(os.path.basename(q))
        if len(queues) >= 2 or rps_on:
            verdict = "meets the advisory's condition (2 or more RX queues, or RPS on)"
        else:
            verdict = "one queue and RPS off, which is the advisory's mitigation 2"
        rps = "RPS on for " + ",".join(rps_on) if rps_on else "RPS off"
        rows.append(f"{name:<8} {len(queues)} RX queue(s), {rps}: {verdict}")
    return rows


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--root", default="/")
    ap.add_argument("releases", nargs="*")
    args = ap.parse_args()
    if args.releases:
        for rel in args.releases:
            print(f"{rel:<26} {check_release(rel)}")
        return
    rel = read(os.path.join(args.root, "proc/sys/kernel/osrelease")) or platform.release()
    print(f"{'kernel release':<24} {rel}")
    print(f"{'fix status':<24} {check_release(rel)}")
    sc = read(os.path.join(args.root, "proc/sys/net/ipv4/tcp_syncookies"))
    note = {"0": "off, which is the advisory's mitigation 1",
            "1": "on when a SYN backlog overflows, the kernel default",
            "2": "always on"}.get(sc, "unreadable")
    print(f"{'net.ipv4.tcp_syncookies':<24} {sc} ({note})")
    for row in rx_report(args.root):
        print(row)


if __name__ == "__main__":
    main()

We have no Linux machine here, so we tested it against made-up release strings and made-up /proc and /sys trees, not a live kernel. First, release strings on either side of each fix, the lines with no upstream release that carries it, and three vendor-style strings:

6.12.116                   VULNERABLE: update to 6.12.117 or later
6.12.117                   FIXED: 6.12.117 was the first release with the patch
6.18.62                    VULNERABLE: update to 6.18.63 or later
6.18.63                    FIXED: 6.18.63 was the first release with the patch
7.3.3                      VULNERABLE: update to 7.3.4 or later
7.3.4                      FIXED: 7.3.4 was the first release with the patch
7.4.0-rc1                  VULNERABLE: 7.4-rc1 is older than 7.4-rc2
7.4-rc2                    FIXED: 7.4-rc2 and later have the patch
7.5.1                      FIXED: newer than 7.4
7.2.15                     NO UPSTREAM FIX: no release on this line has the patch, ask your vendor
6.6.162                    NO UPSTREAM FIX: no release on this line has the patch, ask your vendor
6.1.192                    NO UPSTREAM FIX: no release on this line has the patch, ask your vendor
5.15.225                   NO UPSTREAM FIX: no release on this line has the patch, ask your vendor
5.10.274                   NO UPSTREAM FIX: no release on this line has the patch, ask your vendor
3.10.0                     NOT AFFECTED (upstream): older than 4.4, which is where the record starts
6.1.0-27-amd64             UNKNOWN: not a plain upstream version string, ask your vendor
5.14.0-570.el9.x86_64      UNKNOWN: not a plain upstream version string, ask your vendor
6.8.0-100-generic          UNKNOWN: not a plain upstream version string, ask your vendor

Then four made-up hosts, each a folder with a proc/sys/kernel/osrelease file, a proc/sys/net/ipv4/tcp_syncookies file and one sys/class/net/<dev>/queues/rx-<n>/rps_cpus file per receive queue (--root points at the folder):

Host osrelease tcp_syncookies Interfaces and the rps_cpus value of each receive queue
A 6.12.116 1 eth0: four queues, all 0; eth1: one queue, 0
B 6.12.117 0 eth0: one queue, 0
C 6.1.0-27-amd64 1 ens5: one queue, 00000000,000000ff
D 5.15.225 2 eth0: two queues, both 0

The checker prints:

--- host A
kernel release           6.12.116
fix status               VULNERABLE: update to 6.12.117 or later
net.ipv4.tcp_syncookies  1 (on when a SYN backlog overflows, the kernel default)
eth0     4 RX queue(s), RPS off: meets the advisory's condition (2 or more RX queues, or RPS on)
eth1     1 RX queue(s), RPS off: one queue and RPS off, which is the advisory's mitigation 2
--- host B
kernel release           6.12.117
fix status               FIXED: 6.12.117 was the first release with the patch
net.ipv4.tcp_syncookies  0 (off, which is the advisory's mitigation 1)
eth0     1 RX queue(s), RPS off: one queue and RPS off, which is the advisory's mitigation 2
--- host C
kernel release           6.1.0-27-amd64
fix status               UNKNOWN: not a plain upstream version string, ask your vendor
net.ipv4.tcp_syncookies  1 (on when a SYN backlog overflows, the kernel default)
ens5     1 RX queue(s), RPS on for rx-0: meets the advisory's condition (2 or more RX queues, or RPS on)
--- host D
kernel release           5.15.225
fix status               NO UPSTREAM FIX: no release on this line has the patch, ask your vendor
net.ipv4.tcp_syncookies  2 (always on)
eth0     2 RX queue(s), RPS off: meets the advisory's condition (2 or more RX queues, or RPS on)

If you cannot patch yet

The advisory gives two temporary mitigations. The first is sudo sysctl -w net.ipv4.tcp_syncookies=0, which it says “blocks the path where a request socket is reconstructed through a SYN cookie.” The setting is per network namespace, so each container with its own namespace needs it, and the advisory warns that “Disabling it lowers the protection against SYN floods.” The kernel documentation describes the knob as one that will “Send out syncookies when the syn backlog queue of a socket overflows,” with a default of 1, so turning it off trades this exposure for SYN-flood exposure. The advisory does not say whether the second, child-socket path also depends on a SYN cookie.

The second mitigation is to reduce the receive queues of the interface where outside packets arrive to one and to disable RPS and RFS, which “can block the race.” The advisory gives no commands for it. The kernel’s scaling document lists the files: /sys/class/net/<dev>/queues/rx-<n>/rps_cpus, where zero means RPS is off, and /proc/sys/net/core/rps_sock_flow_entries for RFS. The queue count is a NIC and driver setting. We could not test either change here.

Look for crashes

The patch text includes the crash from the reporter’s test. On a host that rebooted unexpectedly, a search of the previous boot’s kernel log (it needs a persistent journal; use -b -2 for earlier boots) can show whether the same trace appeared:

journalctl -k -b -1 | grep -E 'RIP: 0010:tcp_v4_rcv|address: 0000000000000012'

We tested only the pattern, against this six-line file of made-up kernel log lines, two of which carry the trace lines from the patch text:

Nov 16 03:12:44 web1 kernel: eth0: renamed from veth9f2c1d
Nov 16 03:12:45 web1 kernel: BUG: kernel NULL pointer dereference, address: 0000000000000012
Nov 16 03:12:45 web1 kernel: RIP: 0010:tcp_v4_rcv+0x44e/0x1280
Nov 16 03:12:46 web1 kernel: Kernel panic - not syncing: Fatal exception in interrupt
Nov 16 03:15:01 web1 CRON[1881]: (root) CMD (run-parts /etc/cron.hourly)
Nov 16 03:15:09 web1 kernel: TCP: request_sock_TCP: Possible SYN flooding on port 443. Sending cookies.

grep -E printed exactly those two:

Nov 16 03:12:45 web1 kernel: BUG: kernel NULL pointer dereference, address: 0000000000000012
Nov 16 03:12:45 web1 kernel: RIP: 0010:tcp_v4_rcv+0x44e/0x1280

A match is not proof of an attack: the offset in the trace may differ between builds, other bugs can crash in tcp_v4_rcv, and a host that panics outright may never write the trace to disk unless kdump, pstore or a network console is set up. A clean log does not rule out an attempt that did not crash the host.

What we could not confirm

  • Exploit code. When we listed the researcher’s GitHub repository through GitHub’s API at around 10:10 UTC, it held only a README.md of 4,943 bytes, last pushed at 08:03 UTC and with 54 stars. The advisory mentions PoCs and a lab exploit, none of which we saw. We did not run, fetch or link any exploit code.
  • Backports to the older lines. The latest releases on 6.6, 6.1, 5.15 and 5.10 predate the fix. We could not see kernel.org’s stable queue (git.kernel.org sits behind a bot check), so we do not know whether releases for those lines are coming.
  • Impact beyond a crash. Whether the second path is exploitable, and what the researcher’s lab exploit achieves, are not stated in any source we read.
  • How the test traffic reaches the SYN cookie path. The kernel documentation says SYN cookies are sent when a socket’s SYN backlog overflows, and the default is 1. The advisory does not say how its test traffic gets a default host there.

What we checked

We read the advisory on oss-security and the README of the researcher’s repository (we have not linked the repository, because it could later hold exploit code); the CVE.org record, NVD and CISA’s KEV feed; FIRST’s EPSS data; the fix commit and the commit it names as the origin through GitHub’s mirror of the kernel tree; the kernel.org ChangeLog files of the eight releases in the table and the kernel.org release list; the Red Hat, Ubuntu, Debian and Oracle pages below; the kernel documentation; and BleepingComputer’s report. These pages change, so treat the tables as a reading from 10:05 to 10:15 UTC on November 16.

  • Advisory: NoisyTCP: Linux Kernel TCP Pre-auth Remote Corruption (CVE-2026-99288) on oss-security
  • Records: CVE.org, NVD, CISA’s KEV feed and FIRST’s EPSS API
  • Fix: mainline commit 4e2b3ef19e0e, the commit it names as the origin, and the ChangeLog files for 6.12.117, 6.18.63, 7.3.4 and 6.1.192, plus kernel.org’s release list
  • Distribution pages: Red Hat and its bug, Ubuntu, Debian and Oracle
  • Kernel documentation: IP sysctl and scaling in the Linux networking stack
  • Coverage: BleepingComputer
  • Earlier on sxz.io: Infinity Core, the skb zerocopy bug, CVE-2026-98432 and CVE-2026-98276

Tags:

CVE-2026-99288Denial of ServiceLinux KernelNoisyTCPRed HatTCPUbuntu

Share

A black manual telephone switchboard with rows of orange jacks and red patch cords plugged into the lower row, above a rotary dial and a handset
Previous Post

How to Route Alerts in Alertmanager So Pages Reach the On-Call and Tickets Reach the Queue, and Test Grouping, Inhibition, and Silences Before an Outage Does

A brown llama and her gray cria walking across a salt flat beside a red lake in Bolivia
Next Post

llama.cpp’s v1.0.0 Promises Major Bumps for API and ABI Breaks, but Its Rule Table Lists Only the C API, libmtmd Shares the Version While Its Header Warns of Breaking Changes, and the Release Page Carries a 7-Byte Pointer to a Nightly Build Instead of Binaries

No Comment! Be the first one.

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Latest
16 Nov
How to Stop Ollama From Silently Cutting Your Prompt, Find the Real Context Window, and Set num_ctx Without Running Out of Memory
16 Nov
llama.cpp’s v1.0.0 Promises Major Bumps for API and ABI Breaks, but Its Rule Table Lists Only the C API, libmtmd Shares the Version While Its Header Warns of Breaking Changes, and the Release Page Carries a 7-Byte Pointer to a Nightly Build Instead of Binaries
Trending
November 16, 2026
How to Stop Ollama From Silently Cutting Your Prompt, Find the Real Context Window, and Set num_ctx Without Running Out of Memory
November 16, 2026
llama.cpp’s v1.0.0 Promises Major Bumps for API and ABI Breaks, but Its Rule Table Lists Only the C API, libmtmd Shares the Version While Its Header Warns of Breaking Changes, and the Release Page Carries a 7-Byte Pointer to a Nightly Build Instead of Binaries
November 16, 2026
NoisyTCP Can Crash a Linux Server Over the Network, but the Fix Has Reached Only Three of the Seven Kernel Lines Still Being Released, Red Hat and Ubuntu Have No Update, and the Researcher’s Own Timings Run From 5 Seconds to 14 Hours
November 15, 2026
How to Route Alerts in Alertmanager So Pages Reach the On-Call and Tickets Reach the Queue, and Test Grouping, Inhibition, and Silences Before an Outage Does
November 15, 2026
Ai2’s XYBench Turns the XY Problem Into a Benchmark, but the Model That Drafted Every Rubric Also Judges the Answers, the Human Baseline Is Graded on Rubrics Read Off Its Own Answers, and the Emphasis Judge Disagrees With the Annotators on 20 to 26 Percent of Validation Responses
November 15, 2026
The FBI’s ARTEX Warning Says a Supply Chain Compromise “Enabled Access” to 2,250 Tenants Where Microsoft Counted Sign-In Attempts and Cites “Over 1,600” Consoles Where Censys Counted 1,465, and a Microsoft Post It Cites Mistypes One Indicator in Its Hunting Query

Related Posts

Rows of server racks in a data center representing network infrastructure targeted by botnets
News

C0XMO Botnet Shows Why Old Router Firmware Still Matters

June 7, 2026
Close-up of a USB flash drive, representing physical data-theft risk in office security incidents
News

Fake IT Support Is Now Walking Through the Front Door

June 7, 2026
A phone security app on a smartphone resting on a laptop keyboard.
News

Everest Forms Pro Flaw Is Being Exploited to Create Rogue WordPress Admins

June 7, 2026
A phone secured by a padlock, illustrating AI data-leak containment and security controls.
News

OpenAI’s Lockdown Mode Is a Data-Leak Brake, Not a Prompt-Injection Cure

June 8, 2026
SXZ.io SXZ.io
  • [email protected]

Categories

Articles
Learning Hub
News

All Rights Reserved by SXZ.io ©2026