HIGH 7.8

CVE-2026-52908: Linux Kernel RDMA Memory Re-Registration Access Validation Vulnerability

A vulnerability exists in the Linux kernel's RDMA (Remote Direct Memory Access) subsystem where the re-registration of memory regions does not properly validate access level changes. When a memory region's access permissions are changed from read-only to read-write during re-registration, the underlying memory is not re-evaluated to ensure it meets the stricter pinning requirements for writable access. This can result in memory regions being used with insufficient access controls, potentially allowing privileged local users to read, modify, or crash the system.

Source data · NVD / CISA · public domain

CVSS
3.1 · 7.8 HIGH · CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H
Weaknesses (CWE)
Affected products
8 configuration(s)
Published / Modified
2026-06-19 / 2026-07-08

NVD description (verbatim)

In the Linux kernel, the following vulnerability has been resolved: RDMA: During rereg_mr ensure that REREG_ACCESS is compatible If IB_MR_REREG_ACCESS changes from RO to RW then the umem has to be re-evaluated to ensure it is properly pinned as RW. Since the umem is hidden inside each driver's mr struct add a ib_umem_check_rereg() function that each driver has to call before processing IB_MR_REREG_ACCESS. mlx4 has to retain its duplicate ib_access_writable check because it implements IB_MR_REREG_ACCESS | IB_MR_REREG_TRANS by changing both items in place sequentially while the MR is live, so it will continue to not support this combination.

5 reference(s) · View on NVD →

SEC.co analysis · AI-assisted, reviewed against source

Technical summary

The vulnerability is in the RDMA memory region (MR) re-registration process, specifically when IB_MR_REREG_ACCESS transitions from RO (read-only) to RW (read-write). The kernel's umem (user memory) structures are driver-specific and hidden within each driver's mr struct. The fix introduces an ib_umem_check_rereg() function that drivers must call before processing IB_MR_REREG_ACCESS to re-evaluate whether the underlying umem is properly pinned for the new access level. The mlx4 driver retains special handling because it implements combined IB_MR_REREG_ACCESS and IB_MR_REREG_TRANS operations sequentially on live memory regions, requiring it to continue validating access separately and not support this combination.

Business impact

Exploitation of this vulnerability could allow a local attacker with user-level privileges to escalate capabilities within the RDMA subsystem, potentially accessing, modifying, or corrupting data protected by memory region access controls. For organizations relying on RDMA for high-performance computing, storage, or clustering workloads, successful exploitation could compromise data integrity and confidentiality. The impact depends on workload sensitivity and whether RDMA is used in multi-tenant or security-sensitive environments.

Affected systems

This vulnerability affects the Linux kernel across all distributions and versions that include the RDMA subsystem with vulnerable code paths. The issue is most relevant to systems using InfiniBand or RoCE (RDMA over Converged Ethernet) hardware and software stacks, including those with mlx4 and other RDMA-capable drivers. Systems without RDMA capability or those running updated kernels containing the patch are not affected.

Exploitability

Exploitation requires local access to the system with unprivileged user privileges. The attack surface is limited to applications and processes capable of invoking RDMA memory re-registration operations, typically specialized HPC, database, or storage applications. The vulnerability is not remotely exploitable and requires knowledge of RDMA APIs. No public exploit code or active exploitation in the wild has been reported, and this is not listed on CISA's Known Exploited Vulnerabilities catalog.

Remediation

Apply the Linux kernel patch that implements the ib_umem_check_rereg() validation function and ensures all drivers properly call it before processing IB_MR_REREG_ACCESS operations. This fix validates that memory regions are correctly pinned for their new access level during re-registration. Verify the patch against your kernel version and vendor advisory, as backports to older stable kernels may follow the upstream fix.

Patch guidance

Identify your Linux kernel version and distribution. Check the vendor's security advisory or kernel changelog for availability of patches addressing this RDMA re-registration validation issue. Patches should be applied to all systems using RDMA hardware and drivers (mlx4, mlx5, and others). Test patches in a non-production environment first, as kernel updates require validation of RDMA workload stability. For enterprise Linux distributions, security updates are typically delivered through standard patch management channels.

Detection guidance

Monitor kernel logs for errors or warnings related to RDMA memory region operations, particularly those involving access level changes or re-registration failures. Review audit logs for applications invoking RDMA memory re-registration with changing access permissions. If RDMA workloads are critical, consider testing with tools that stress RDMA memory region operations to validate proper pinning behavior. System administrators should correlate any detected issues with known vulnerable kernel versions.

Why prioritize this

This is a high-severity local privilege escalation vulnerability affecting a core kernel subsystem used in high-performance environments. The CVSS 7.8 score reflects the combination of local-only access requirements with significant impact on confidentiality, integrity, and availability. Prioritization should be elevated for systems using RDMA in security-sensitive workloads or multi-tenant environments, and for systems in high-performance computing clusters where data integrity is critical.

Risk score, explained

The CVSS 3.1 score of 7.8 (HIGH) is calculated as CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H. The attack vector is local (L), requiring on-system access. Attack complexity is low (L), meaning exploitation does not require special conditions. Privileges required are low (L), indicating unprivileged user access suffices. No user interaction (N) is needed. The scope is unchanged (U), and all three impact categories—confidentiality, integrity, and availability—are rated high, reflecting the potential for memory corruption and information disclosure through RDMA operations.

Frequently asked questions

Does this vulnerability affect systems without RDMA hardware or drivers?

No. Systems that do not use InfiniBand, RoCE, or other RDMA technologies are not affected by this vulnerability, as the code path requiring the fix is only invoked when RDMA memory region re-registration is performed.

Can this vulnerability be exploited remotely?

No. Exploitation requires local access to the system with unprivileged user privileges. Remote exploitation is not possible with this vulnerability.

What types of applications are at risk?

Applications that use RDMA APIs for memory registration and re-registration are at risk. This typically includes HPC applications, distributed databases, storage systems, and clustering software that leverage RDMA for performance. Standard applications that do not use RDMA are not affected.

Is there a workaround if I cannot patch immediately?

The most effective mitigation is to restrict local access to systems and limit RDMA usage to trusted applications. However, applying the kernel patch is the proper fix and should be prioritized in your patching schedule.

This analysis is based on available vulnerability data as of the published date. Exploit details, patch availability, and vendor advisories should be verified directly with the Linux kernel security team and your distribution vendor. CVSS scoring reflects standard methodology but should be contextualized against your organization's specific risk posture and asset criticality. No exploit code or proof-of-concept is provided. Test all patches thoroughly in non-production environments before deployment. This information is provided for security professionals and system administrators; it is not a substitute for official vendor guidance or comprehensive security assessment. Source: NVD (public-domain), retrieved 2026-07-27. Analysis generated by SEC.co (claude-haiku-4-5).