RefluXFS Linux Kernel Vulnerability Grants Root Access via XFS Filesystem Flaw
A race condition in the Linux kernel's XFS filesystem, dubbed RefluXFS (CVE-2026-64600), allows local users to overwrite protected files and gain root privileges, bypassing security measures like SELinux.
A newly disclosed vulnerability, nicknamed "RefluXFS" and tracked as CVE-2026-64600, presents a significant threat to Linux systems by allowing unprivileged local users to achieve root privileges. This race condition resides within the XFS filesystem's copy-on-write (CoW) path, enabling attackers to silently overwrite critical system files and bypass security mechanisms such as SELinux, even when in enforcing mode.
The vulnerability is triggered by a timing flaw during concurrent O_DIRECT writes to reflinked files on an XFS volume. Normally, XFS handles writes to shared blocks by allocating a new private block and remapping the file. However, the kernel temporarily releases its inode lock while waiting for transaction log space. This brief window allows a second writer to intervene, remap the file, and decrement its reference count. Consequently, the initial writer mistakenly writes to the original block using a stale reference, leading to data corruption. Since O_DIRECT bypasses the page cache and skips revalidation, this corrupted write is permanently applied to disk.
Exploiting RefluXFS allows an unprivileged local user to overwrite any readable file on a reflink-enabled XFS volume at the block layer. This capability directly translates to gaining host root access. Researchers from Qualys Threat Research Unit (TRU) demonstrated a proof-of-concept on a default RHEL 10.2 deployment, successfully removing root account password protection and enabling passwordless root access within seconds. Crucially, these modifications persist across reboots and leave no discernible artifacts in the kernel logs, making detection exceptionally challenging.
This flaw has been present in mainline and stable Linux kernels since version 4.11, released in 2017, meaning it has existed undetected for years. Qualys estimates that the vulnerability potentially affects over 16.4 million systems globally, based on an analysis conducted using their CyberSecurity Asset Management platform. For a system to be vulnerable, three conditions must be met: a kernel version 4.11 or later without the patch, an XFS filesystem with reflink=1 enabled, and a directory writable by an unprivileged user that is adjacent to a high-value target, such as a SUID-root binary.
RefluXFS operates at a low level within the filesystem allocation layer, making it resistant to conventional kernel hardening techniques. Standard memory protection mechanisms like KASLR, SMEP, and SMAP are ineffective against this attack surface. Furthermore, kernel lockdown features do not restrict O_DIRECT or FICLONE calls, and SELinux failed to block the vulnerable code path in testing. Container isolation methods, including user-namespace restrictions and capability limits, also do not mitigate this threat, leaving patching as the only reliable workaround.
Interestingly, the discovery of RefluXFS was a result of a collaborative research effort between Qualys and Anthropic. By integrating Anthropic's Claude Mythos Preview model into TRU's manual audit workflow under Anthropic's Project Glasswing, researchers tasked the AI with identifying Dirty COW-style race conditions. Through iterative prompting, the model pinpointed the XFS flaw and generated a working proof-of-concept, which Qualys engineers then independently verified before coordinating responsible disclosure with upstream kernel maintainers.
This AI-accelerated vulnerability research mirrors a growing trend in 2026, following other significant Linux privilege escalation disclosures this year. Vendor-fixed kernels are now available and are being backported to major enterprise distributions, including RHEL, Oracle Linux, AlmaLinux, Rocky Linux, and Fedora. Organizations are strongly advised to prioritize patching internet-facing and multi-tenant systems immediately. Applying the relevant distribution security update and performing a full reboot are essential to confirm the fix is active, as no interim mitigation or configuration workaround currently neutralizes the flaw.