NoisyTCP: from exposed TCP to the kernel behind the service
A TCP flaw can interrupt every container on a host. NoisyTCP calls for identifying where packets arrive and which kernel fix is actually running there.
Portekizce okuIn this article
By Pentesty
An availability flaw with distinct publication dates
CVE-2026-99288 was published on November 10, 2026. Hyunwoo Kim disclosed the research on November 15 at 16:00 UTC; BleepingComputer covered it on November 16. This timeline separates the original vulnerability record, the technical disclosure and the report that prompted this analysis.
The flaw involves a race during TCP socket reuse in the kernel. Its confirmed consequence is a kernel panic that interrupts the system. Red Hat describes an availability impact and distinguishes it from data alteration or arbitrary code execution; the reporting does not establish a data leak caused by this vulnerability.
Locate the host that actually receives the packets
The research requires attacker packets to reach a TCP listener directly and concurrent receive processing: at least two RX queues, or RPS spreading packets across CPUs. Nginx, Apache and SSH services with default configurations can meet these conditions. This does not establish that every Linux server has the same exposure.
Map the public address, any proxy and the host or VM terminating each connection. For a container service, identify its shared kernel and the other workloads on that host. The researcher explains that the fault can crash the entire host. Replicas concentrated there remain within the same failure domain. Include the receive-processing configuration in that record. An application label or an orchestrator service name alone does not identify which underlying kernel receives external traffic.
Read fix status by branch and distribution
The Linux record identifies fixes in the 6.12 branch from 6.12.117, the 6.18 branch from 6.18.63 and the 7.3 branch from 7.3.4, alongside the mainline milestone 7.4-rc2. The corrective commit was integrated on November 3. These milestones do not replace checking the backports and packages maintained by each vendor.
When checked on November 16, Red Hat classified RHEL 8, 9 and 10 as affected and RHEL 6 and 7 as unaffected; the consulted evidence does not confirm a fixed RPM. Debian records fixes in bookworm 6.1.192-1, trixie 6.12.117-1 and unstable 7.3.4-1. Treat these results as dated snapshots of their respective channels.
Accept the running kernel and workload continuity
Match each machine to its vendor-approved package, update provenance and the kernel loaded after maintenance. Installing a package or recreating a container does not establish that the host is running the fix. Our analysis of Linux kernel update acceptance explains this verification for a different CVE.
In an isolated replica, preserve the relevant network topology and run legitimate transactions before and after maintenance. Record the service owner, shared dependencies and the result of an authorized failover. Acceptance should establish both fix provenance and recovery of the business workflow; exploiting the flaw is unnecessary for this exercise.
Set boundaries for any temporary mitigation
The researcher and Red Hat describe temporary alternatives: disable SYN cookies in each relevant network namespace, reducing SYN-flood protection; or limit reception to one RX queue and disable RPS and RFS. Assign an owner, a maintenance window, a capacity assessment and a documented rollback to any decision. Applying the kernel fix remains the priority.
A blocked-traffic metric needs to connect to a service that remains usable. Our review of DDoS telemetry and availability helps define that outcome. For NoisyTCP, bring together the TCP path, failure domain, vendor decision and evidence of operation after maintenance in a record for each service.
Frequently asked questions
Does NoisyTCP allow remote code execution?
The confirmed impact in this analysis is host interruption through a kernel panic. The verified sources do not establish arbitrary code execution or a data leak resulting from this flaw.
Does a container prevent other applications from going down?
The container shares its host kernel. According to the researcher, triggering the flaw through a service inside it can crash the host and interrupt other workloads depending on that kernel.
Is comparing the kernel number with a published version enough?
Check the branch, package and backport status in the vendor channel. Then verify the kernel actually loaded and the result of maintenance for the service.
Is disabling SYN cookies a permanent fix?
It is a temporary alternative that reduces SYN-flood protection and applies per network namespace. Assess exposure, document rollback and plan remediation before making that decision.
Sources and references
- BleepingComputer: NoisyTCP report, November 16, 2026
- Hyunwoo Kim: original NoisyTCP disclosure on oss-security
- Linux CNA: CVE-2026-99288 record and fixes by branch
- Red Hat: CVE-2026-99288 impact, affected products and mitigations
- Debian Security Tracker: vulnerable and fixed packages
- Official Linux repository: corrective commit and integration date
Original Pentesty analysis, published November 16, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.