Critical RefluxFS Linux Kernel Bug Lets Attackers Gain Root Access

Key Takeaways

• A critical vulnerability, dubbed “RefluXFS” (CVE-2026-64600), has been discovered in the Linux kernel’s XFS filesystem.
• This race condition allows an unprivileged local user to achieve root access by silently overwriting system files.
• The flaw impacts Linux kernel versions 4.11 and newer, affecting an estimated 16.4 million systems globally.
• No effective workarounds exist; immediate patching and system reboots are required to mitigate the risk.
• The discovery was aided by AI, specifically Anthropic’s Claude Mythos Preview model, highlighting a growing trend in vulnerability research.

RefluXFS: A Critical Linux Kernel Privilege Escalation

A severe race condition, now named “RefluXFS,” has been identified within the Linux kernel’s XFS filesystem copy-on-write mechanism. This vulnerability, tracked as CVE-2026-64600, enables a local attacker without elevated privileges to covertly modify protected system files, thereby gaining full root access to the host. Notably, this exploit can succeed even on systems configured with SELinux in Enforcing mode.

The Qualys Threat Research Unit (TRU) uncovered this timing-based flaw. It manifests when two simultaneous O_DIRECT write operations target the same reflinked file residing on an XFS volume.

Understanding the Vulnerability

Under normal operation, XFS manages writes to shared data blocks by allocating a new, private block and remapping the file’s pointer to this new location. However, the kernel momentarily releases its inode lock while awaiting available space in the transaction log. This brief window of vulnerability allows a second writer to intervene. During this interval, the second writer can remap the file and decrement its reference count. Consequently, the initial writer, operating with a stale reference, erroneously writes directly to the original block. Since O_DIRECT bypasses the page cache and skips revalidation, this corrupted write is permanently committed to disk.

This flaw permits an unprivileged local user to maliciously overwrite any readable file at the block layer on an XFS volume where reflink functionality is enabled, directly leading to root access on the host.

Qualys’s proof-of-concept demonstrated the severity of RefluXFS on a default RHEL 10.2 installation. Within seconds, the exploit silently removed password protection from the root account, granting immediate, passwordless root access. Crucially, these modifications persist across reboots and leave no discernible artifacts in kernel logs, making detection exceptionally challenging.

The RefluXFS bug has been present in all mainline and stable Linux kernel versions since 4.11, which was released in 2017. Qualys estimates that this vulnerability could potentially affect over 16.4 million systems worldwide, based on analysis performed with its CyberSecurity Asset Management platform.

For a system to be vulnerable, three specific conditions must be met: it must be running a kernel version 4.11 or later without the necessary patch, feature an XFS filesystem with reflink=1 enabled, and contain a directory writable by an unprivileged user, alongside a critical target file such as a SUID-root binary.

RefluXFS operates at the filesystem allocation layer, a level beneath most conventional kernel hardening measures. Memory protection mechanisms like KASLR, SMEP, and SMAP address different attack vectors. Kernel lockdown imposes no restrictions on O_DIRECT or FICLONE calls, and testing has confirmed that SELinux does not block the vulnerable code path. Similarly, container isolation techniques, including user-namespace restrictions and capability limits, do not affect this low-level vulnerability, indicating that there is currently no reliable workaround apart from applying the official patch.

AI-Assisted Discovery

Notably, this discovery resulted from a collaborative research effort between Qualys and Anthropic, which integrated the Claude Mythos Preview model into TRU’s manual audit workflow under Anthropic’s Project Glasswing. Researchers tasked the AI model with identifying Dirty COW-style race conditions. Through iterative prompting, the model successfully pinpointed the XFS flaw and generated a functional proof-of-concept, which Qualys engineers subsequently verified independently before coordinating disclosure with upstream maintainers.

This approach of AI-accelerated, human-validated vulnerability research aligns with a broader trend observed in 2026. This year has seen several other significant Linux privilege escalation vulnerabilities, such as CVE-2026-46333 (a nine-year-old ptrace flaw) and CrackArmor, an AppArmor-based root-access chain impacting over 12 million systems.

The emergence of these four critical vulnerabilities within approximately three months underscores a growing pattern where researchers are discovering novel methods to trick the kernel into writing attacker-controlled data into protected memory or disk blocks that back setuid binaries or configuration files.

What You Should Do

• Apply Patches Immediately: Vendor-fixed kernels are already available and are being backported across various enterprise distributions, including RHEL, Oracle Linux, AlmaLinux, Rocky Linux, and Fedora.
• Prioritize Critical Systems: Organizations must prioritize patching internet-facing and multi-tenant systems without delay.
• Perform Full Reboot: After applying the relevant security update, a full system reboot is essential to ensure the fix is active and fully implemented.
• No Interim Mitigation: Be aware that no interim mitigation or configuration workaround currently exists to neutralize this flaw short of patching.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.