Skip to main content
Vulnerability Database/CVE-2026-98356

CVE-2026-98356: Linux Kernel RDMA Use-After-Free Vulnerability

CVE-2026-98356 is a use-after-free flaw in the Linux kernel RDMA bnxt_re driver that can cause system crashes through NULL pointer dereference. This article covers the technical details, affected kernel versions, and mitigation strategies.

Published:

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

CVE-2026-98356 Overview

CVE-2026-98356 is a null pointer dereference vulnerability in the Linux kernel's RDMA/bnxt_re driver. The flaw resides in the Data Center Bridging (DCB) workqueue initialization path. The function bnxt_re_init_dcb_wq() fails to check the return value of create_singlethread_workqueue(). When the allocation fails, the workqueue pointer remains NULL. The asynchronous DCB handler later invokes queue_work() on this NULL pointer, causing a kernel crash.

The issue affects systems using Broadcom NetXtreme-E RDMA network adapters with DCB enabled. The vulnerability has been resolved in upstream Linux kernel commits.

Critical Impact

A failed workqueue allocation triggers a kernel null pointer dereference, resulting in a denial-of-service condition on systems using the bnxt_re RDMA driver.

Affected Products

  • Linux kernel versions containing the bnxt_re RDMA driver with the DCB workqueue initialization code
  • Systems using Broadcom NetXtreme-E RDMA network adapters
  • Distributions shipping unpatched upstream Linux kernels

Discovery Timeline

  • 2026-10-06 - CVE-2026-98356 published to NVD
  • 2026-10-06 - Last updated in NVD database

Technical Details for CVE-2026-98356

Vulnerability Analysis

The vulnerability resides in the bnxt_re_init_dcb_wq() function of the Broadcom NetXtreme-E RDMA driver (bnxt_re). This function allocates a single-threaded workqueue to process asynchronous Data Center Bridging events. The original code called create_singlethread_workqueue() but did not validate the returned pointer.

When kernel memory allocation fails under pressure, create_singlethread_workqueue() returns NULL. The driver stored this NULL value and proceeded with initialization. Later, when a DCB event arrived, the asynchronous handler called queue_work() on the NULL workqueue pointer. This dereference produces a kernel oops and typically crashes the affected subsystem or the entire host.

The fix adds explicit error handling to the allocation call and propagates failure up the initialization path. Upstream repair commits are tracked at kernel.org commit 6c368f7b, commit 727bbda3, and commit a0fb2007.

Root Cause

The root cause is a missing return value check on a kernel allocation routine. The bnxt_re_init_dcb_wq() function trusts that create_singlethread_workqueue() always succeeds. This assumption violates kernel memory allocation contracts and leaves a NULL pointer in place for later consumers to dereference.

Attack Vector

Triggering the condition requires memory pressure sufficient to cause workqueue allocation failure during driver initialization. Once the NULL pointer is stored, any incoming asynchronous DCB event from the hardware or firmware will invoke queue_work() on the NULL pointer. The attack vector is local and dependent on resource exhaustion. No authenticated remote attacker interaction is required beyond causing memory pressure on an affected host.

No verified exploit code is publicly available. See the upstream kernel commits for technical remediation details.

Detection Methods for CVE-2026-98356

Indicators of Compromise

  • Kernel oops or panic messages referencing bnxt_re_init_dcb_wq, queue_work, or NULL pointer dereference in the RDMA subsystem
  • dmesg entries showing workqueue allocation failures preceding the crash
  • Unexpected loss of RDMA connectivity on Broadcom NetXtreme-E adapters

Detection Strategies

  • Monitor kernel logs for stack traces that include bnxt_re module symbols combined with NULL pointer dereference signatures
  • Correlate memory pressure events with subsequent RDMA driver instability
  • Audit installed kernel versions against the fixed upstream commits to identify unpatched hosts

Monitoring Recommendations

  • Forward dmesg and /var/log/kern.log output to a centralized logging platform for continuous inspection
  • Alert on kernel crash telemetry and workqueue allocation failure messages tied to the bnxt_re module
  • Track system memory pressure metrics on hosts running RDMA workloads to anticipate conditions that can trigger the bug

How to Mitigate CVE-2026-98356

Immediate Actions Required

  • Identify all hosts running the bnxt_re kernel module with Broadcom NetXtreme-E RDMA adapters
  • Apply the upstream kernel patches referenced in the external commits once distribution-level updates are available
  • Reduce memory pressure on affected hosts to lower the probability of workqueue allocation failure prior to patching

Patch Information

The upstream Linux kernel has merged fixes that validate the return of create_singlethread_workqueue() in bnxt_re_init_dcb_wq(). Review the commits at kernel.org commit 6c368f7b, commit 727bbda3, and commit a0fb2007. Deploy vendor-supplied kernel updates once your distribution backports these fixes.

Workarounds

  • Disable Data Center Bridging functionality on affected adapters where operationally feasible
  • Unload the bnxt_re kernel module on hosts that do not require RDMA until a patched kernel is deployed
  • Ensure adequate memory headroom on systems running the driver to reduce the likelihood of allocation failure
bash
# Check whether the bnxt_re module is loaded
lsmod | grep bnxt_re

# Review recent kernel messages for related crashes or warnings
dmesg | grep -iE 'bnxt_re|queue_work|workqueue'

# Temporarily unload the module if RDMA is not required
sudo modprobe -r bnxt_re

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.