Rebootless Patch Status for NoisyTCP CVE-2026-99288
No KernelCare rebootless patch for CVE-2026-99288 has been released yet for any distribution, and there’s no release date to share. Red Hat hasn’t shipped a fixed kernel for RHEL 8, 9 or 10, and Ubuntu is still evaluating the issue. Debian 12, Debian 13, Oracle Linux UEK and CIQ’s commercial Rocky Linux builds already have fixed kernels.
Status as of November 16, 2026: RHEL 6, RHEL 7 and CentOS 7 kernels aren’t affected. LibCare isn’t involved because this is a kernel bug. Until a fix reaches your hosts, the mitigations further down can reduce exposure.
Key Takeaways
- NoisyTCP (CVE-2026-99288) lets an unauthenticated remote attacker panic a vulnerable Linux kernel through an exposed TCP listener.
- The race needs two or more receive queues or RPS, a common setup on servers and cloud VMs.
- Upstream fixed it in 6.12.117, 6.18.63 and 7.3.4, and Debian, Oracle UEK and CIQ ship fixed kernels.
- RHEL 8, 9 and 10 and most rebuilds have no fixed kernel yet, and Ubuntu is still evaluating.
The Short Version
What happened: On November 15, 2026, at 16:00 UTC, researcher Hyunwoo Kim posted the full NoisyTCP advisory to oss-security once the linux-distros embargo ended. The bug, CVE-2026-99288, is a race in the kernel’s lockless TCP connection lookup that ends in a NULL pointer dereference and a kernel panic. Jump to this section →
Who’s affected: Kernels from Linux 4.4 onward that carry the vulnerable code and haven’t picked up the fix, on servers that accept TCP connections and spread incoming packets across multiple receive queues or RPS. Default Nginx, Apache and SSH setups can be hit if an attacker can reach their TCP ports. RHEL 8, 9 and 10 and most rebuilds are affected, and Ubuntu hasn’t finished its evaluation. Jump to this section →
How bad it is: The demonstrated impact is a remote kernel panic, which takes down the whole host, including every container on it. A harder use-after-free path exists, but nobody has demonstrated code execution. Red Hat rates it Important, with a preliminary CVSS 3.1 score of 7.5. Jump to this section →
Where things stand: This isn’t a zero day. The fix was upstream before the CVE went out, no exploit code is public, and no exploitation in the wild had been reported as of November 16. Upstream stable, Debian, Oracle UEK and CIQ’s commercial Rocky Linux builds have fixes. Red Hat and Ubuntu don’t yet, and TuxCare’s rebootless patches aren’t out yet either. Jump to this section →
What to do today: Check whether internet-facing hosts have more than one receive queue or RPS turned on. Swap in your own interface name for eth0:
ethtool -l eth0
grep . /sys/class/net/eth0/queues/rx-*/rps_cpus
A Combined or RX value above 1 under “Current hardware settings,” or any nonzero rps_cpus mask, means the host meets the precondition for the race. Without a fixed kernel, two stopgaps exist, and both cost something. Jump to this section →
The Long Version
Plenty of network-facing kernel bugs need something unusual before they matter. NoisyTCP needs none of that. CVE-2026-99288 sits in the code that looks up established TCP connections, and every incoming TCP segment goes through it. A server only has to call listen() and receive packets from the attacker.
What the attacker gets is a kernel panic, not a shell. That keeps it out of drop-everything territory for most teams. Still, an unauthenticated way to take a production host offline on demand is a serious availability problem.
NoisyTCP traces back to the same 2015 commit as CVE-2026-98432, the TCP use-after-free that made headlines in October. Patching that one doesn’t fix NoisyTCP, so hosts that got October’s TCP fix may still be exposed to this one.
How a Recycled Socket Panics the Kernel
Linux looks up established TCP sockets without taking locks, which keeps the network stack fast. The trade-off is that socket memory can be reused for a brand-new socket while a lookup is still holding the old address.
Once the lookup finds a match, it takes a reference and rechecks the address tuple. It never checked whether that socket was still in the hash table, though.
NoisyTCP takes advantage of exactly that. The attacker gets the kernel to free a pending connection, and that memory is then reused for a new request socket built from a SYN cookie, with the same tuple. The stale lookup grabs the new object, the tuple check passes, and it returns a socket that was never hashed and has no listener. tcp_v4_rcv() then dereferences a NULL listener pointer, and the kernel panics.
Winning that race takes two receive contexts, normally on two CPUs, handling packets for the same connection at the same moment. Kim gets there by mixing IPv4 fragments with normal TCP packets. Fragments can hash to a different CPU than the rest of the flow.
The upstream fix, commit 4e2b3ef19e0e (“inet: reject unhashed sockets during established lookups”), adds the missing check. If the socket is no longer hashed once the reference is taken, the lookup drops the reference and starts over. It also changes how the SYN cookie, TCP Fast Open and request migration paths release references.
The bug dates back to commit 079096f103fa (“tcp/dccp: install syn_recv requests into ehash table”), written in October 2015, which allowed request sockets to be processed without holding the listener lock. That commit shipped in Linux 4.4 in January 2016.
On arm64, POWER and other CPUs with weaker memory ordering, a second upstream commit, 5b5d8f351dd9 (“net: restore refcount ordering for lockless socket access”), restores memory ordering around the lookup’s reference. It wasn’t part of 6.12.117, 6.18.63 or 7.3.4, so check that those kernels carry both commits.
Which Servers Can Actually Be Hit
Exposure comes down to two conditions. First, an unpatched kernel with a TCP listener the attacker can reach, true of a big share of internet-facing servers.
The second is concurrency on the receive side. The race needs two or more receive queues, or Receive Packet Steering (RPS) spreading packets across CPUs. Multi-queue NICs are common on modern servers and on cloud VMs with more than one vCPU, so this condition is often met by default.
Timing varies. In Kim’s tests against QEMU/KVM virtual machines running default Nginx, Apache or SSH, the crash took between 5 seconds and 2 minutes, since vCPU scheduling widens the race window. Against a quiet bare-metal Nginx server with 8 receive queues, a tuned proof of concept took about 14 hours.
Containers don’t help, since one panic in the shared kernel takes down every container on the host.
The quick server check further down shows whether a host meets both conditions. If it does and no fixed kernel is available, apply one of the stopgaps below.
A Crash Today and Code Execution Only in Theory
Kim’s oss-security post describes NoisyTCP as a pre-auth remote corruption bug. That’s fair as a bug class, but it overstates what an attacker gets.
The demonstrated path ends in a NULL pointer dereference and a kernel panic, which is a denial of service. Kim also describes a second flow, a race on fully established child sockets, that can lead to a remote use-after-free, and he says it’s hard to trigger without special techniques. Nobody has demonstrated code execution here.
Kim’s own FAQ answers the remote code execution question with “Not yet. and basically, no.” Red Hat’s statement says the flaw doesn’t allow data alteration or arbitrary code execution. The NoisyTCP repository on GitHub holds the write-up, not exploit code. Kim’s own exploit is limited, lab-specific and unreleased.
So plan around a remote crash that’s practical against exposed, multi-queue hosts.
What the 7.5 Score Actually Measures
Red Hat rates NoisyTCP Important, with a preliminary CVSS 3.1 score of 7.5, and the NVD entry had no assessment of its own as of November 16.
The vector behind that 7.5 is AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H. It says the attacker needs low privileges and can hit confidentiality and integrity, while Red Hat’s own statement on the same page says neither is true.
The mismatch has a practical side. Rescore the vector to match the statement, with no privileges required and availability as the only impact (AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H), and CVSS 3.1 gives 5.9, which is Medium. That’s a hypothetical calculation, not a score Red Hat has published, and the technical risk would be exactly the same.
If the score is ever revised downward, any scanner or patch SLA that ranks by CVSS number alone could push this down the queue. For internet-facing hosts, go by the description instead.
Which Distributions Have a Fix
| Distribution | Status as of November 16, 2026 | Fixed kernel or advisory |
| Upstream stable 6.12, 6.18 and 7.3 | Fixed | 6.12.117, 6.18.63 and 7.3.4 |
| Upstream stable 6.6, 6.1 and 5.x | No fixed release yet | None listed |
| Upstream 7.2 | Branch ended before the fix landed upstream | None (move to 7.3.4) |
| RHEL 8, 9 and 10 | Affected, no release date | None released |
| AlmaLinux, CloudLinux, community Rocky Linux and Oracle Linux RHCK 8, 9 and 10 | Affected | None released |
| CIQ Rocky Linux builds (RLC Pro, RLC Pro LTS, RLC Pro FIPS and SIG/Cloud Next) | Fixed November 15 | Patched kernels listed in CIQ’s advisory |
| RHEL 6, RHEL 7 and CentOS 7 | Not affected | Not needed |
| Oracle Linux UEK (Oracle Linux 7 to 10) | Fixed | ELSA-2026-500630, ELSA-2026-500631 and ELSA-2026-500632 |
| Debian 13 (trixie) | Fixed | linux 6.12.117-1 (DSA-6628-1) |
| Debian 12 (bookworm) | Fixed | linux 6.1.192-1 (DLA-4906-1) |
| Ubuntu 16.04 through 26.10 | Needs evaluation | None released |
CloudLinux 9 and 10 run AlmaLinux kernels, so they follow the AlmaLinux row. Debian 11 doesn’t appear on the Debian tracker. Oracle shipped its UEK errata, starting with ELSA-2026-500630, on November 12 and 13. On the Rocky side, CIQ’s advisory lists patched kernels for its commercial RLC Pro, LTS, FIPS and SIG/Cloud Next builds, with CIQ Linux Kernel LT 6.12 still pending. Teams on Amazon Linux, SUSE or Proxmox VE should check their vendor’s tracker directly.
The Debian 12 row deserves a second look. Upstream 6.1.192 lacks the fix, but Debian backported it into 6.1.192-1. Check the installed kernel package against your distribution’s advisory or changelog rather than trusting the upstream version.
For self-built kernels, 6.12.117, 6.18.63 and 7.3.4 carry the fix. The 7.2 branch ended at 7.2.15 on November 1, two days before the fix was applied upstream, so anything still on 7.2 needs to move to 7.3.4 or rely on a distribution backport.
Buying Time Until a Rebootless Patch Lands
Both mitigations come from Kim’s advisory, and Red Hat lists the same two. TuxCare hasn’t tested either against an exploit. Try each one on a staging host first.
The first option is turning off SYN cookies. The demonstrated crash path needs a request socket rebuilt from a SYN cookie, so disabling them blocks that path. Run this and the commands below as root:
sysctl -w net.ipv4.tcp_syncookies=0
Revert with sysctl -w net.ipv4.tcp_syncookies=1. SYN cookies keep a listener answering during a SYN flood, so turning them off weakens that defense. The setting is also per network namespace, and a new namespace starts with the kernel default, so it has to be set inside every container that has its own network namespace. It also closes only the demonstrated route.
The second option goes after the race directly. Drop the internet-facing interface to a single receive queue and turn off RPS and Receive Flow Steering (RFS). On a NIC that uses combined channels, with eth0 standing in for your interface, that looks like this as root:
ethtool -L eth0 combined 1
for q in /sys/class/net/eth0/queues/rx-*/rps_cpus; do echo 0 > "$q"; done
sysctl -w net.core.rps_sock_flow_entries=0
NICs that expose separate RX channels take rx 1 instead of combined 1. This removes the concurrency the race depends on, not just one path. The cost is throughput, since all receive processing lands on one core. Changing channels can briefly drop the link on some physical NICs, so keep console access handy. The ethtool change doesn’t survive a reboot unless the network configuration reapplies it.
If an internet-facing host has no fixed kernel and can spare the throughput, the single-queue option is the more complete stopgap. If throughput can’t drop, turning off SYN cookies is the narrower choice. If the distribution already ships a fixed kernel, install it and reboot into it, or apply a rebootless patch for the running kernel once one is available. Remove the workarounds only after confirming the fix is active.
Checking a Server in Under a Minute
Six commands answer most of the questions, again with eth0 standing in for your interface:
uname -r
ss -ltn
ethtool -l eth0
grep . /sys/class/net/eth0/queues/rx-*/rps_cpus
sysctl net.ipv4.tcp_syncookies
kcarectl --patch-info | grep 'CVE-2026-99288'
Distribution kernels carry backports the version number won’t reveal, so compare the installed package against your vendor’s advisory and the table above. On Debian, zgrep -c CVE-2026-99288 /usr/share/doc/linux-image-$(uname -r)/changelog*.gz returns a nonzero count when the booted kernel’s changelog lists the fix. The ss line lists TCP listeners, the next two show the race precondition, and the sysctl line shows whether SYN cookies are on. The last line shows whether an applied rebootless patch covers this CVE. For a fleet, run them through Ansible, Salt or a plain SSH loop.
Where KernelCare Stands
TuxCare hasn’t released a rebootless patch for CVE-2026-99288 on any distribution yet, and there’s no date to share. RHEL 6, RHEL 7 and CentOS 7 kernels aren’t affected, so they don’t need one, and LibCare isn’t involved because this is a kernel bug. The TuxCare CVE Tracker shows rebootless patch status for this CVE across supported kernels and will update as releases ship.
| Kernel | Rebootless patch status |
| EL 8, 9 and 10 (RHEL, AlmaLinux, CloudLinux, Oracle Linux RHCK, Rocky Linux) | Not released yet, no date |
| Oracle Linux 7, 8 and 9 UEK | Not released yet, no date |
| Debian 12 and 13 | Not released yet, no date |
| Ubuntu | Not released yet, no date |
| RHEL 6, RHEL 7 and CentOS 7 | Not affected, no rebootless patch needed |
Once a patch ships, hosts with automatic updates pick it up on schedule, or kcarectl --update pulls it right away, no reboot needed. Confirm it landed with kcarectl --patch-info | grep 'CVE-2026-99288'. On RHEL and its rebuilds especially, the gap between a vendor kernel appearing and every exposed host actually booting into it is where a bug like this one causes real trouble.
Get KernelCare Enterprise today →
The Fix Landed Early and Many Distributions Are Still Waiting
Coordinated disclosure worked the way it’s supposed to here, and plenty of production Linux servers are still exposed.
The fix was applied upstream on November 3, a week before the CVE and twelve days before the full write-up. That’s why this isn’t a zero day.
None of that helps an unpatched RHEL 9 web server with port 443 open and four receive queues. An upstream fix still has to become a vendor package, then reach every exposed host. For RHEL and most of its rebuilds, that first step hadn’t happened as of November 16, CIQ’s commercial Rocky Linux builds aside.
So find the internet-facing hosts with multiple receive queues or RPS. Pick a stopgap, test it in staging and go in knowing what it costs. Then work out how fast fixed kernels can reach every exposed host – by rebootless patch or by reboot, whichever the fleet will actually tolerate.
Get KernelCare Enterprise today →
How the Disclosure Unfolded
| Date | Event |
| October 3, 2015 | Commit 079096f103fa introduces the bug, which ships in Linux 4.4 in January 2016 |
| November 3, 2026 | Hyunwoo Kim’s fix is applied upstream |
| November 10, 2026 | The kernel CVE team publishes CVE-2026-99288, and stable 6.12.117, 6.18.63 and 7.3.4 ship with the fix |
| November 11, 2026 | Red Hat lists RHEL 8, 9 and 10 as affected and rates the flaw Important (7.5) |
| November 12 and 13, 2026 | Oracle ships fixed UEK kernels (ELSA-2026-500630, ELSA-2026-500631 and ELSA-2026-500632) |
| By November 15, 2026 | Debian fixes trixie (DSA-6628-1) and bookworm (DLA-4906-1) |
| November 15, 2026 | CIQ releases patched kernels for its RLC Pro, LTS, FIPS and SIG/Cloud Next builds |
| November 15, 2026 | The linux-distros embargo ends, and Kim posts the NoisyTCP advisory to oss-security at 16:00 UTC |
| Pending | Fixed RHEL and community rebuild kernels, CIQ Linux Kernel LT 6.12, Ubuntu’s evaluation and kernels, fixes for upstream 6.6 and older stable branches, and rebootless patches for affected kernels |
Sources
Primary sources for this post are Kim’s oss-security advisory and NoisyTCP write-up, the kernel CVE announcement, the CVE record, the upstream fix and the refcount ordering follow-up, plus vendor pages from CIQ, Red Hat, Debian, Ubuntu and Oracle, and the NVD entry. Rebootless patch status reflects TuxCare’s release records as of November 16, 2026.