Introduction
CVE-2026-99288 is a race condition in the Linux kernel's TCP established-socket lookup path (__inet_lookup_established and __inet6_lookup_established). The flaw allows a remote attacker to trigger a NULL pointer dereference in tcp_v4_rcv(), causing a kernel panic.
The vulnerable lookups run without locks under RCU, and socket memory uses SLAB_TYPESAFE_BY_RCU, which allows memory to be reused while a reader still holds the old address. After a tuple match, the lookup takes a reference but never checks whether the object is still hashed. If the memory was reused for a new socket with the same tuple, the lookup can return an unhashed object whose rsk_listener is NULL, and the subsequent dereference panics the kernel.
This article covers CIQ RLC Pro, RLC Pro LTS, and RLC Pro FIPS variants built on Rocky Linux 8, 9, and 10. It describes what is affected, how to confirm exposure, the available mitigations, and the current patch status.
Problem
The race requires two conditions that are common on production systems:
- SYN cookies are enabled (
net.ipv4.tcp_syncookies = 1or2). This is the default on most Rocky Linux installations. - Multiple RX queues or RPS/RFS are active, allowing the socket lookup to race across CPUs. This is the default on most multi-core systems with modern NICs.
When both conditions are met, an attacker who can send TCP packets to the system can trigger the race, causing the kernel to dereference a NULL pointer and panic. The attack is network-based and does not require authentication on the target system.
Treat the following as affected unless they are running a patched kernel listed in the Patched Kernels section or have the mitigation in place:
- RLC Pro LTS 8.6, 9.2, and 9.6
- RLC Pro 8 (el8_10), RLC Pro 9 (el9_8), and RLC Pro 10 (el10_2)
- RLC Pro FIPS 8.6, 8.10, 9.2, and 9.6
- RLC Pro Hardened variants based on Rocky Linux 8, 9, or 10
CIQ Bridge customers and CentOS 7.9 users are not affected. The vulnerable code path is not present in the el7 kernel.
Status
- Patched kernels are available for most supported variants as of 2026-11-14. See the Patched Kernels table below. CIQ Linux Kernel (CLK) builds are still in progress.
- Recommended action: install the patched kernel. Run
sudo dnf update kernel*and reboot. See Installing the Update below for variant-specific instructions. If you cannot schedule a reboot immediately, apply one of the temporary mitigations in the Mitigation section to reduce exposure until the update can be applied. The mitigations are not required on systems already running a patched kernel. - CVSS 3.1 score: 7.5 (AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H). Rated Important. The primary demonstrated impact is a kernel panic (denial of service). The attack complexity is rated High because the race requires specific timing across CPU queues.
- A researcher proof-of-concept exists that forces address reuse via a SYN cookie request socket. No public exploit is indexed in Metasploit, Exploit-DB, or nuclei templates as of 2026-11-14.
- Two temporary mitigations are available (see Mitigation below). Both carry tradeoffs. The recommended path is to install the patched kernel.
- CIQ Bridge customers and CentOS 7.9 users are unaffected by this vulnerability.
- Architecture: both x86_64 and arm64 are affected. The flaw is in the generic TCP socket lookup path, not architecture-specific code.
- Open a support case if you need help confirming exposure, assessing mitigation impact, or tracking patched kernel availability for a specific CIQ variant.
Patched Kernels
| Variant | Patched Kernel Version | Repository | Released |
|---|---|---|---|
| RLC Pro LTS 8.6 | kernel-4.18.0-372.32.1+41.1.el8_6_ciq |
rlc-8.6-lts.x86_64 |
2026-11-15 |
| RLC Pro FIPS 8.6 | kernel-4.18.0-553.184.1+2.1.el8_6.ciqfips |
fips-8.6-compliant-x86_64 |
2026-11-15 |
| RLC Pro FIPS 8.10 | kernel-4.18.0-553.184.1+2.1.el8_10_ciq |
fips-8.10-compliant-x86_64 |
2026-11-15 |
| RLC Pro 8 | kernel-4.18.0-553.184.1+2.1.el8_10_ciq |
rlc-pro-8.x86_64 |
2026-11-15 |
| CIQ SIG/Cloud Next 8 | kernel-4.18.0-553.184.1+2.1.el8_10_ciq |
ciq-sigcloud-next-8.x86_64 |
2026-11-15 |
| RLC Pro LTS 9.2 | kernel-5.14.0-284.30.1+46.1.el9_2_ciq |
rlc-9.2-lts.x86_64 |
2026-11-15 |
| RLC Pro FIPS 9.2 | kernel-5.14.0-284.30.1+46.1.el9_2_ciq |
fips-9.2-compliant-x86_64 |
2026-11-15 |
| RLC Pro LTS 9.6 | kernel-5.14.0-570.60.1+28.1.el9_6_ciq |
rlc-9.6-lts.x86_64 |
2026-11-15 |
| RLC Pro FIPS 9.6 | kernel-5.14.0-570.60.1+28.1.el9_6_ciq |
fips-9.6-compliant-x86_64 |
2026-11-15 |
| RLC Pro 9 | kernel-5.14.0-687.66.1+2.1.el9_8_ciq |
rlc-pro-9.x86_64 |
2026-11-15 |
| CIQ SIG/Cloud Next 9 | kernel-5.14.0-687.66.1+2.1.el9_8_ciq |
ciq-sigcloud-next-9.x86_64 |
2026-11-15 |
| RLC Pro 10 | kernel-6.12.0-211.74.1+2.1.el10_2_ciq |
ciq-scn-10.x86_64 |
2026-11-15 |
| CIQ SIG/Cloud Next 10 | kernel-6.12.0-211.74.1+2.1.el10_2_ciq |
ciq-scn-10.x86_64 |
2026-11-15 |
| CIQ Linux Kernel LT 6.12 | pending | pending | pending |
Confirm what is running on a given system with:
uname -r
If your system reports one of the patched versions above (or newer), the fix is in place and the mitigation can be reverted. See Resolution.
Installing the Update
RLC Pro LTS and FIPS Variants
For RLC Pro LTS and FIPS variants (including LTS 8.6, LTS 9.2, LTS 9.6, FIPS 8.6, FIPS 8.10, FIPS 9.2, and FIPS 9.6), the patched kernel is available directly from the long-term support or FIPS repository. No additional repository configuration is required:
sudo dnf update kernel*
sudo reboot
RLC Pro 8, RLC Pro 9, RLC Pro 10, and CIQ SIG/Cloud Next
For RLC Pro and CIQ SIG/Cloud Next variants, refresh your DNF metadata and install the update:
sudo dnf refresh
sudo dnf update kernel*
sudo reboot
If the patched kernel does not appear, confirm the appropriate depot channel is enabled for your variant:
sudo depot enable rlc-pro-8 # RLC Pro 8 / CIQ SIG Cloud Next 8
sudo depot enable rlc-pro-9 # RLC Pro 9 / CIQ SIG Cloud Next 9
sudo depot enable rlc-pro-10 # RLC Pro 10 / CIQ SIG Cloud Next 10
Confirming Exposure
Check whether the two race prerequisites are present on the system.
SYN cookies:
sysctl net.ipv4.tcp_syncookies
A value of 1 (the default) or 2 means SYN cookies are enabled and the system can reach the vulnerable code path.
Multiple RX queues:
ls -d /sys/class/net/*/queues/rx-* 2>/dev/null | head -20
If any interface shows more than one rx-* queue, the race can occur across CPUs. Single-queue interfaces are also vulnerable if RPS is configured:
cat /sys/class/net/*/queues/rx-*/rps_cpus 2>/dev/null
A non-zero value on any queue means RPS is distributing packet processing across CPUs, which enables the race even on a single-queue NIC.
Systems where SYN cookies are enabled and multiple RX queues or RPS are active should be treated as exposed.
Mitigation
Apply one of the mitigations below if you cannot immediately install a patched kernel. Once a patched kernel is installed and the system is rebooted, the mitigations are not required and should be reverted.
Option 1: Disable SYN cookies (recommended)
Disabling SYN cookies removes the code path that the known proof-of-concept uses to force socket address reuse. This is the simpler mitigation with the more bounded tradeoff.
sudo sysctl -w net.ipv4.tcp_syncookies=0
To persist across reboots:
echo "net.ipv4.tcp_syncookies=0" | sudo tee /etc/sysctl.d/cve-2026-99288.conf
sudo sysctl --system
⚠️ WARNING Disabling SYN cookies removes the kernel's built-in protection against SYN flood attacks. Systems that are exposed to the public internet or handle high volumes of incoming TCP connections should evaluate the tradeoff before applying this mitigation. If SYN flood protection is critical, consider Option 2 or prioritize installing the patched kernel.
Container and namespace note: net.ipv4.tcp_syncookies is a per-network-namespace sysctl. If the system runs containers with their own network namespaces, the sysctl must be set inside each namespace as well, not only on the host.
Option 2: Restrict to a single RX queue and disable RPS/RFS (last resort)
Use this only if disabling SYN cookies is not acceptable for your environment (for example, public-facing systems under active SYN flood risk). This mitigation forces all packet processing onto a single CPU per interface, which can cause significant performance degradation on production workloads.
The race requires the socket lookup to execute concurrently on different CPUs. Restricting each NIC to a single RX queue and disabling RPS/RFS eliminates that concurrency.
For each network interface (replace eth0 with the actual interface name):
sudo ethtool -L eth0 combined 1
If ethtool -L returns "Operation not supported," the NIC may not support channel changes. Run ethtool -l eth0 to check whether the NIC is already single-queue.
Disable RPS on all queues:
for f in /sys/class/net/eth0/queues/rx-*/rps_cpus; do
echo 0 | sudo tee "$f"
done
Disable RFS:
sudo sysctl -w net.core.rps_sock_flow_entries=0
for f in /sys/class/net/eth0/queues/rx-*/rps_flow_cnt; do
echo 0 | sudo tee "$f"
done
These changes are not persistent across reboots by default. To persist them, add the ethtool -L and sysctl changes to a startup script or use a systemd unit.
Verification
After applying Option 1, confirm SYN cookies are disabled:
sysctl net.ipv4.tcp_syncookies
The output should be net.ipv4.tcp_syncookies = 0.
After applying Option 2, confirm a single RX queue:
ethtool -l eth0 | grep -A4 "Current"
The Combined value should be 1. Confirm RPS is disabled:
cat /sys/class/net/eth0/queues/rx-*/rps_cpus
All values should be 0 or 00000000.
Resolution
Once a patched kernel is available and installed, reboot and verify the running kernel matches a version from the Patched Kernels table:
uname -r
If your system is running a patched kernel, revert any mitigations that were applied.
To revert Option 1 (SYN cookies):
sudo rm /etc/sysctl.d/cve-2026-99288.conf
sudo sysctl -w net.ipv4.tcp_syncookies=1
sudo sysctl --system
To revert Option 2 (RX queue restriction): restore the original queue count and RPS configuration, or reboot. The original values depend on the hardware and previous configuration.
Root Cause
TCP established-socket lookups in both IPv4 (__inet_lookup_established) and IPv6 (__inet6_lookup_established) use RCU for lockless traversal of the ehash table. The kernel socket allocator uses SLAB_TYPESAFE_BY_RCU, which allows slab objects to be freed and reallocated while an RCU reader still holds a pointer to the old object. This is normally safe because the lookup re-validates the tuple after taking a reference.
The bug is that the lookup did not check whether the socket was still hashed after taking the reference. If the original socket was freed and the memory was immediately reused for a new, unhashed socket with a matching tuple, the lookup returned the unhashed object. The known trigger uses a SYN cookie request socket: the request socket is freed while the lookup holds a stale reference, and the replacement has a NULL rsk_listener. When tcp_v4_rcv() dereferences the listener to read sk_state, the kernel panics.
The upstream fix adds a post-reference check: after taking a reference, the lookup verifies that the nulls node is still hashed and that the tuple still matches. If either check fails, the lookup drops the reference and restarts. The fix also changes the SYN cookie, TCP Fast Open, and request migration paths to release request sockets with reqsk_put() instead of freeing them directly, because a stale lookup may hold a transient reference.
Notes
- The CVSS vector rates confidentiality and integrity impact as High (C:H/I:H), but the known proof-of-concept only demonstrates a kernel panic (availability impact). The higher C/I rating reflects the theoretical potential of the use-after-free primitive; in practice, the demonstrated attack is a denial of service.
- The attack complexity is rated High (AC:H) because the race requires precise timing across CPU queues. A single-CPU system with a single RX queue and no RPS is not exploitable by the known PoC, though it may still contain the vulnerable code.
- This CVE does not affect CIQ Bridge or CentOS 7.9. The vulnerable code path was introduced after the el7 kernel branch.
References & related articles
- CVE-2026-99288 (NVD)
- CVE-2026-99288 (MITRE)
- Upstream fix: inet: reject unhashed sockets during established lookups
- Upstream fix: inet6: reject unhashed sockets during established lookups
- Upstream fix: tcp: use
reqsk_putfor SYN cookie release - Upstream fix: tcp: use
reqsk_putfor Fast Open and request migration