NoisyTCP (CVE-2026-99288): relayed as a crash only; Red Hat's preliminary vector says C:H/I:H/A:H
BleepingComputer relayed CVE-2026-99288 as a Linux kernel crash only. The researcher's advisory is titled Pre-auth Remote Corruption, and Red Hat's preliminary vector sets confidentiality, integrity and availability to High (7.5, recomputed by this site).
By Parminder Kumar Sharma · LinkedIn · · 18 min read
Three pages give three answers to what CVE-2026-99288 can do
The saved pages do not agree on what CVE-2026-99288, the Linux kernel TCP flaw its researcher calls NoisyTCP, can do to a server. BleepingComputer's relay of 16 November 2026 says it crashes the host and that "they don't allow threat actors to alter data or gain arbitrary code execution". The researcher's own advisory is titled "NoisyTCP: Linux Kernel TCP Pre-auth Remote Corruption". Red Hat's page rates the impact Important and describes a denial of service, and its statement says the flaw "does not allow data alteration or arbitrary code execution", yet its preliminary CVSS v3.1 vector sets confidentiality, integrity and availability all to High, with attack complexity High and privileges required Low. This site recomputed that vector from the v3.1 formula: it gives a base score of 7.5, the figure Red Hat prints (derived).
What that does not establish. The arithmetic checks the printed score against the printed vector. It does not show the vector is right: Red Hat says its metrics and score are preliminary and subject to review. For comparison, the same vector with only confidentiality and integrity changed to None scores 5.3 (derived; this site's own what-if, not any vendor's score). None of the saved pages reports an attack, counts the servers that meet the preconditions, or gives the score that NVD assigns, because NVD was not read.
Whose words these are. The advisory is the researcher's own, and the relay is a news report of it, so the relay is a secondary source. Red Hat sells support subscriptions for its Linux products, and its page is a statement about its own products. The Debian tracker is a downstream distribution's record, and Debian and Red Hat each ship on their own timelines. No page here sells a fix. The relay page itself carries an advertisement for an email security scan that has nothing to do with this flaw. This site names no individual: the relay and the advisory both name and link the researcher, and this article says only "the researcher".
What each page prints about the impact
What each saved page prints about the impact. Quotations are short and exact. The right column is what the same page does not say.
| Page | What it prints | What it does not say |
|---|---|---|
| Relay, BleepingComputer, 16 Nov | A pre-authenticated remote corruption flaw, a NULL pointer dereference, causes a kernel panic. Says attackers cannot alter data or gain arbitrary code execution. | Anything about the advisory's second trigger flow, or what the researcher's lab exploit does. |
| Researcher's advisory (README) | Titled "Pre-auth Remote Corruption". Flow 1 ends in a kernel panic, a remote DoS. Flow 2 "can trigger a remote use-after-free", but is generally hard to trigger. Asked if it is a TCP RCE: "Not yet. and basically, no.". | What the author's lab exploit achieves. The page says only that it is "very limited, lab-specific exploit that targets VM servers" and that a general purpose exploit is still being researched. |
| Red Hat, description | A race during lockless socket lookups causes a NULL pointer dereference, a kernel crash and a denial of service. | Any other effect. |
| Red Hat, statement | Important, because an unauthenticated remote attacker can trigger an immediate crash "without requiring privileges". The flaw "does not allow data alteration or arbitrary code execution". | Why the vector below scores integrity and confidentiality High and privileges required Low. |
| Red Hat, preliminary vector | Base score 7.5. Network, attack complexity High, privileges Low, no user interaction, scope unchanged, confidentiality High, integrity High, availability High. | A reason for the three High impacts. The page calls the metrics preliminary and subject to review. NVD and cve.org columns read N/A. |
| Debian tracker (kernel commit message) | A reproduced crash: a NULL pointer dereference, then a kernel panic. The fix rejects an unhashed socket found during an established lookup. | Any impact beyond the crash shown. This site did not read the commit itself. |
The difference is in the wording of the impact, not in the crash, which every page prints. The relay's sentence about data and code execution matches Red Hat's statement in substance; this site cannot say which the relay drew on. The advisory is more guarded: it answers the RCE question with "Not yet. and basically, no.", and its second flow is described as a possible remote use-after-free that is hard to trigger. That is the advisory's own basis for the word corruption in its title, and neither the relay nor the Red Hat text mentions that flow.
Red Hat's page also prints the generic text for CWE-476, the weakness class the relay links as well. That text lists a technical impact of executing unauthorised code or commands, reading memory and modifying memory, and says it can happen "in rare circumstances" when address zero is accessible to privileged code. It is a description of the weakness class, not an assessment of this CVE, and it is not evidence of any of those effects here.
The preconditions, as each page states them
The advisory prints its trigger flows. This site does not repeat them: this article stays at the level of the advisory's abstract and the kernel commit message, and prints no packet layout, timing recipe or trigger code. What follows is only the conditions each page names.
The conditions each saved page names. A dash is not used: an empty answer says Silent.
| Condition | What the pages print | Page |
|---|---|---|
| Plain TCP listener, default configuration | It can be triggered against plain TCP listeners, such as Nginx, Apache and SSH servers in their default configuration. The FAQ says no special configuration is needed and that listen() is enough. The relay says default configurations. Red Hat says any system with active network listeners is susceptible. | Advisory, relay, Red Hat statement |
| Two or more RX queues, or RPS | The race needs a server with two or more RX queues, or with RPS spreading packets across CPUs. The relay repeats it in a quotation. Red Hat's statement is Silent. | Advisory, relay |
| SYN cookies | Flow 1 reuses a request socket reconstructed through a SYN cookie. The advisory says the setting defaults to 1 on most distributions, and its first mitigation turns it off. The relay mentions SYN cookie handling. | Advisory, relay, Red Hat mitigation |
| Packets reach the host directly | The advisory says servers that an attacker's TCP packets can reach directly. Red Hat says over the network. | Advisory, Red Hat vector |
| Virtual machine or bare metal | The advisory gives different trigger times for each (next section). The relay gives only the bare metal caution. Red Hat is Silent. | Advisory, relay |
| Containers | If the DoS is triggered against a web server in a container, the entire host goes down. The SYN cookie setting is per network namespace, so each container with its own namespace needs it. | Advisory, Red Hat mitigation |
The narrow race window is a bare metal figure
The relay carries one caution from the researcher: the race window is narrow, so triggering it against a quiet bare metal server "can take a long time". The advisory attaches two measurements to that sentence. Against Nginx, Apache and SSH servers 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. Against Nginx in its default configuration on a quiet bare metal server with 8 RX queues, it took about 14 hours, even though a proof of concept with race timing tuned for that environment was used.
Set side by side, 14 hours is 840 minutes, which is 420 times the 2 minute upper figure, and 50,400 seconds, which is 10,080 times the 5 second lower figure (derived: 14 x 60 / 2 = 420; 14 x 3,600 / 5 = 10,080). So the relay's caution is true of the slow case and says nothing of the fast one. A reader who saw only the relay would not learn that the same advisory prints a figure of seconds for virtual machines.
What that does not establish. These are the researcher's own measurements. The advisory does not say how many runs were made, on which networks, or on which hypervisors beyond the example given. It does not say how long an untuned attacker would need, and no independent measurement was read. It does not say what share of servers are virtual machines, so it gives no way to turn the two figures into an exposure count.
How old it is, and the dates around the fix
The relay calls it an 11 year old flaw. The advisory prints the introducing commit as 079096f103fa, dated 2015-10-03, and the fixing commit as 4e2b3ef19e0e, dated 2026-11-03. The span is 4,049 days, which is 11 years and 1 month (derived: 2015-10-03 to 2026-11-03). That fits the advisory's own "11 years", which it gives as approximate. This site did not read either commit: git.kernel.org's robots.txt lists many AI crawler agents, so the commit pages were not requested. The Debian tracker prints the commit message and no hash, and Red Hat lists three stable branch commit references whose hashes do not begin with the advisory's 12 characters. This site draws no conclusion from that difference.
Dates as the saved pages print them. Spans between them are derived. The Red Hat dates are in the page's embedded data, not in its visible text; the repository date is GitHub's own metadata.
| When (UTC unless stated) | What | Source and derived span |
|---|---|---|
| 2015-10-03 | Introducing commit, prefix 079096f103fa | Advisory. 4,049 days before the fix date (derived) |
| 2026-11-03 | Fixing commit, prefix 4e2b3ef19e0e | Advisory. The relay says merged two weeks ago; 13 days before 16 Nov (derived) |
| 2026-11-10, 18:19 | Red Hat page public date | Red Hat page data. 7 days after the fix date (derived) |
| 2026-11-11, 07:57 | Red Hat page last changed (as read) | Red Hat page data |
| 2026-11-15, 16:33 | The researcher's repository created | GitHub metadata. 5 days after the Red Hat public date (derived) |
| 2026-11-16, 09:30 (04:30 US Eastern) | Relay published | Relay page data. 16 hours 56 minutes after the repository was created (derived); 6 days after the Red Hat public date (derived) |
What the dates do not establish. The advisory says only that the embargo agreed after reporting to the linux-distros list has ended. No page says when it ended, whether the repository existed privately earlier, or whether Red Hat's public date is the first public statement. The relay's phrase about a fix merged two weeks ago agrees with the 13 days between the advisory's fix date and 16 November. Neither the relay nor the Red Hat page says when distribution kernels were first available.
Who is fixed, as each page prints it
Debian security tracker, per release, as printed on 16 November 2026. The tracker links advisories DLA-4906-1 for bookworm and DSA-6628-1 for trixie.
| Release | Version as printed | Status as printed |
|---|---|---|
| bookworm | 6.1.176-1 | vulnerable |
| bookworm (security) | 6.1.192-1 | fixed |
| trixie | 6.12.107-1 | vulnerable |
| trixie (security) | 6.12.117-1 | fixed |
| forky | 7.2.15-1 | vulnerable |
| sid | 7.3.4-1 | fixed |
Red Hat product status as printed in its page data on 16 November 2026. The advisory column is empty on every row and the page data says it has no errata. Counts are derived: 9 Affected plus 3 Not affected is 12 rows.
| Product | Package | State as printed |
|---|---|---|
| Red Hat Enterprise Linux 10 | kernel | Affected |
| Red Hat Enterprise Linux 6 | kernel | Not affected, Vulnerable Code not Present |
| Red Hat Enterprise Linux 7 | kernel | Not affected, Vulnerable Code not Present |
| Red Hat Enterprise Linux 7 | kernel-rt | Not affected, Vulnerable Code not Present |
| Red Hat Enterprise Linux 8 | kernel | Affected |
| Red Hat Enterprise Linux 8 | kernel-rt | Affected |
| Red Hat Enterprise Linux 9 | kernel | Affected |
| Red Hat Enterprise Linux 9 | kernel-rt | Affected |
| Red Hat OpenShift Container Platform 4 | openshift/ose-rhel-coreos-8 | Affected |
| Red Hat OpenShift Container Platform 4 | openshift/ose-rhel-coreos-9 | Affected |
| Red Hat OpenShift Container Platform 5 | openshift/ose-rhel-coreos-10 | Affected |
| Red Hat OpenShift Container Platform 5 | openshift/ose-rhel-coreos-9 | Affected |
Debian's two security rows carry fixed versions, and its main archive rows for bookworm and trixie still list the older versions as vulnerable. Debian's forky row lists no fixed version, while sid has one. Red Hat's own FAQ says Affected means its analysis team has determined the product is affected and that a fix may be released in the near future. Its page lists no advisory against any of the 12 rows, so as read it names no fixed Red Hat kernel. RHEL 6 and 7 are listed Not affected with the justification Vulnerable Code not Present, while the relay and the advisory both say the flaw affects almost all Linux distributions: these are different statements about different scopes, and this site does not resolve them.
What this does not establish. Nothing here says whether Ubuntu, SUSE, Gentoo or any cloud image vendor has shipped a fixed kernel, because those pages were not read. Debian's tracker and Red Hat's table say nothing about appliances, embedded devices or vendor kernels that carry their own patches. The relay says a patch is available and was merged into the mainline kernel, and links the Debian tracker as its evidence for that.
Where the pages differ
These are differences between the saved pages. This article reports them and resolves none.
Differences between the saved pages and how this article handles each.
| Point | What the pages print | Handling |
|---|---|---|
| Impact | Relay: crash only. Advisory: title says corruption, flow 2 is a possible use-after-free, RCE answer is "Not yet. and basically, no.". Red Hat: text says DoS only; vector says C:H/I:H/A:H. | Each printed, none chosen. Red Hat's metrics are preliminary. |
| Default configuration | Relay: servers with default configurations. Advisory: needs two or more RX queues or RPS, yet no special configuration. Red Hat: any system with active listeners. | The conditions are listed side by side in the preconditions table. |
| Privileges | Red Hat's statement says "without requiring privileges"; its vector says privileges required Low. | A difference inside one page. Not resolved. |
| Mitigation | Relay: disable SYN cookies and serialize to one RX queue. Advisory: one of the following two. Red Hat prints both as mitigations in the same words as the advisory. | Reported as the advisory words it: either one. The SYN flood cost is printed in the advisory and in Red Hat's Mitigation section, not in its Statement. |
| Scope | Advisory and relay: almost all Linux distributions. Red Hat: RHEL 6 and 7 Not affected. | Both printed. Neither page counts distributions. |
| The fix commit | Advisory prints 4e2b3ef19e0e. Debian prints the commit message title only. Red Hat lists three other stable commit hashes. | This site did not read the commits and does not match them. |
| The researcher | The relay and the advisory name him and link social media. | Not named here. The article says the researcher. |
On the Mitigation row: Red Hat's Mitigation section tells the reader, in the advisory's words, that disabling SYN cookies lowers the protection against SYN floods. A reader who disables them swaps one exposure for another. The pages give no figure for how much protection is lost, and they do not say whether the second trigger flow, the use-after-free, is also blocked by either mitigation. This site does not know.
What the pages state, and what they do not
A negative column for the saved pages. Left: what is printed. Right: what is not printed anywhere this site read.
| Question | Printed | Not stated |
|---|---|---|
| Is it exploited? | No page reports an attack. Red Hat's page data carries a field named exploit known, set to false. | Any telemetry, honeypot or scan data. That field is not visible text on the page. |
| How many servers are exposed? | The preconditions in words. | Any count of hosts, or any share with two or more RX queues or RPS. |
| Who has a fixed kernel? | Debian: three fixed rows. Red Hat: no advisory on 12 rows. | Ubuntu, SUSE, Gentoo, cloud and appliance vendors. Dates of release. |
| Does it work against cloud guests? | Times on VMs such as QEMU/KVM. | Any cloud provider, managed service or other hypervisor. |
| What score does NVD give? | Red Hat's table shows N/A for NVD and cve.org. | NVD's score. NVD was not read. |
| Can the crash be repeated at a useful rate? | About 5 seconds to 2 minutes on VMs; about 14 hours on one quiet bare metal setup. | Runs, spread, untuned conditions, noisy production networks. |
| Does the patched kernel change SYN cookie behaviour or speed? | The commit message says a lookup that finds an unhashed socket now drops the reference and restarts, and that failed SYN cookie, Fast Open and request migration paths release references with reqsk_put(). | Any performance effect. |
| Can it go beyond a crash? | Advisory: "Not yet. and basically, no."; flow 2 is hard; a lab exploit targets VM servers. Relay and Red Hat text: no code execution. | What the lab exploit does. Whether a general exploit exists. |
What the pages print: identifiers, crash lines and the stopgap setting
Only what the saved pages print, copied exactly. The advisory also prints trigger flows; this site does not repeat them, and prints no packet layout, timing recipe or trigger code. Nothing below is a working exploit.
The identifiers. The CVE id and the title of the kernel commit, as the Debian tracker prints them. A defender uses the title to match the change in a distribution's changelog or kernel source.
CVE-2026-99288
inet: reject unhashed sockets during established lookups
The weakness and Red Hat's rating. Printed on Red Hat's page. The vector is preliminary and the page says it is subject to review. A defender reads it as Red Hat's current scoring of its own products, not as NVD's score.
CWE-476: NULL Pointer Dereference
Impact: Important
Base Score: 7.5
CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H
What a crash looks like. Three lines from the crash the kernel commit message prints for its reproduction, in the order printed, with the lines between them left out. A defender searches kernel logs, the console and any crash dump after an unexplained reboot of an edge server. The frame named virtnet_poll in the full trace is a virtual network driver's polling function, which suggests (inference, not printed) that the reproduction ran in a virtual machine. A panic with these lines does not by itself show this flaw or an attack: other faults can print similar text.
BUG: kernel NULL pointer dereference, address: 0000000000000012
RIP: 0010:tcp_v4_rcv+0x44e/0x1280
Kernel panic - not syncing: Fatal exception in interrupt
The stopgap setting. Printed in the advisory and, in the same words, in Red Hat's Mitigation section. As printed it changes the running value on the host. The pages do not say how to make it survive a reboot. They say the setting is managed per network namespace, so each container with a separate namespace needs it, and that disabling SYN cookies lowers the protection against SYN floods.
sudo sysctl -w net.ipv4.tcp_syncookies=0
The second temporary mitigation has no command in either page. Both say to reduce the RX queues of the interface where external packets arrive to one and to disable RPS and RFS, which can block the race.
What to do, in order
Take this with you
In the order worth doing. The first three are what the pages advise; the rest are this site's suggestions.
- The sources' advice: look up CVE-2026-99288 in your distribution's own tracker and install the fixed kernel when it ships. Debian's tracker prints 6.1.192-1 for bookworm and 6.12.117-1 for trixie in its security archive. Red Hat's page listed no advisory as read, so check it again.
- The sources' advice, where you cannot patch yet: either disable SYN cookies, accepting weaker protection against SYN floods and applying it in each separate container namespace, or reduce the receiving interface to one RX queue and disable RPS and RFS.
- The sources' advice: the advisory says applying the fix is the priority and the mitigations are temporary.
- This site's suggestion: a kernel update needs a reboot, so book the maintenance window now rather than when a patch is already waiting.
- This site's suggestion: inventory internet facing Linux hosts and note which have two or more RX queues or RPS. Start with virtual machines, because the advisory's fastest figures are for them, and keep quiet bare metal on the list, because its slow figure came from one setup.
- This site's suggestion: do not treat the narrow race window as protection. The advisory prints seconds to minutes on VMs.
- This site's suggestion: watch for unexplained kernel panics or reboots on edge servers and keep the console and crash logs. No saved page reports observed attacks, and a panic can have other causes.
- This site's suggestion: if you disable SYN cookies, write down where, and restore the setting once the fixed kernel is running.
The question the pages leave open
Each page answers a different question: what the flaw does, what it needs, whether a fix exists. None counts how many servers are in the position the advisory describes, and none says whether anyone is using it. So the question is about your own estate, not the news: which of your Linux edge servers has two or more receive queues or RPS, and has anyone checked its kernel?
Key facts
Sources
- PrimaryThe advisory (README): abstract, trigger flows named not repeated, trigger characteristics, affected range and dates, mitigations, FAQ. The researcher's own account. Read by this site on 16 November 2026The researcher's GitHub repositoryaccessed 2026-11-16
- PrimaryCVE page: description, statement, mitigation, preliminary CVSS v3.1 vector and 7.5 score, CWE-476, product status rows. Read by this site on 16 November 2026Red Hat Customer Portalaccessed 2026-11-16
- PrimaryCVE page: the kernel commit message (inet: reject unhashed sockets during established lookups) and per release status. Read by this site on 16 November 2026Debian Security Trackeraccessed 2026-11-16
- PrimaryRepository metadata used only for the creation, push and update times. Read by this site on 16 November 2026GitHub (repository metadata)accessed 2026-11-16
- Reported byRelay dated 16 November 2026, 04:30 US Eastern: crash only, default configurations, narrow race window, mitigations. SecondaryBleepingComputeraccessed 2026-11-16


