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-2026-98559 Overview
CVE-2026-98559 is a use-after-free (UAF) vulnerability in the Linux kernel's Shared Memory Communications over RDMA (SMC-R) subsystem. The flaw resides in smcr_port_err(), which traverses the smc_lgr_list.list linked list without holding smc_lgr_list.lock. A concurrent call to smc_lgr_terminate_sched() can free a link group (lgr) while it is still being dereferenced, resulting in memory corruption. The upstream fix holds smc_lgr_list.lock across the traversal and updates smc_ib_gid_check() to call smcr_port_err() only after releasing the lock.
Critical Impact
Concurrent execution paths in the SMC-R stack can dereference a freed lgr structure, enabling kernel memory corruption, denial of service, or potential privilege escalation on affected Linux hosts.
Affected Products
- Linux kernel (upstream net/smc subsystem)
- Linux distributions shipping vulnerable stable kernels prior to the referenced fix commits
- Systems with SMC-R / RDMA networking enabled
Discovery Timeline
- 2026-10-18 - CVE-2026-98559 published to NVD
- 2026-10-19 - Last updated in NVD database
Technical Details for CVE-2026-98559
Vulnerability Analysis
The vulnerability affects the SMC-R implementation in net/smc, which provides accelerated TCP sockets over RDMA fabrics. The function smcr_port_err() is invoked when an InfiniBand or RoCE port transitions to an error state. It walks smc_lgr_list.list to notify every active link group tied to the affected port.
Because the traversal runs without acquiring smc_lgr_list.lock, another CPU executing smc_lgr_terminate_sched() can concurrently remove and free an lgr entry. The iterator in smcr_port_err() then dereferences freed memory, producing a classic use-after-free condition in kernel context.
Exploitation requires triggering concurrent port error events and link group termination. On systems using SMC-R for application traffic, port flaps or link resets can race with socket teardown to reach the vulnerable window.
Root Cause
The root cause is missing synchronization around a shared list traversal. smc_lgr_list is protected by smc_lgr_list.lock, but smcr_port_err() iterated the list without taking the lock. Any writer path, including smc_lgr_terminate_sched(), could remove and free an element mid-iteration. This is a locking defect classified under the use-after-free class of memory corruption vulnerabilities.
Attack Vector
Triggering the race requires generating a port error event, typically through RDMA device state changes, while link groups are being torn down. On hosts that expose SMC-R to multi-tenant workloads or process attacker-influenced network traffic over RDMA, a local attacker able to open SMC sockets and influence link state could race the two paths. The resulting UAF can lead to kernel panic, memory disclosure, or control-flow corruption depending on the allocator state at the time of the free.
No public proof-of-concept is referenced in the advisory data. See the upstream patches for the precise code paths and locking semantics addressed by the fix.
Detection Methods for CVE-2026-98559
Indicators of Compromise
- Unexpected kernel oops or panic entries referencing smcr_port_err, smc_lgr_terminate_sched, or smc_ib_gid_check in dmesg or /var/log/messages.
- KASAN (Kernel Address Sanitizer) splats indicating use-after-free in the net/smc module on kernels built with sanitizer support.
- Repeated RDMA port error events correlated with SMC socket activity shortly before kernel instability.
Detection Strategies
- Inventory Linux hosts with the smc kernel module loaded (lsmod | grep smc) and determine whether SMC-R is in active use.
- Compare running kernel versions against the fixed stable commits (00a59600, 0edcc428, 34bcf11c, 3d35f069, 3d65a2eb, 61cb282f, a9b968e4) referenced in the kernel.org advisories.
- Enable KASAN on test or canary systems to surface the UAF under synthetic port-flap workloads.
Monitoring Recommendations
- Forward kernel ring buffer and crash dump telemetry to a centralized log and SIEM platform for correlation across hosts.
- Alert on repeated kernel oops conditions or node reboots on hosts running RDMA workloads.
- Track RDMA port state transitions and correlate with SMC socket churn to identify race-prone workloads.
How to Mitigate CVE-2026-98559
Immediate Actions Required
- Identify all Linux systems using SMC-R and prioritize them for patching with a kernel containing the referenced fix commits.
- If patching is not immediately possible, unload the smc module (modprobe -r smc) on hosts that do not require SMC-R connectivity.
- Restrict local user ability to open SMC sockets by applying seccomp or AppArmor profiles to untrusted workloads.
Patch Information
The fix adds smc_lgr_list.lock around the traversal in smcr_port_err() and reorders smc_ib_gid_check() so that smcr_port_err() runs after the lock is released. Apply the stable kernel updates referenced in the advisory:
- Kernel Patch 00a59600
- Kernel Patch 0edcc428
- Kernel Patch 34bcf11c
- Kernel Patch 3d35f069
- Kernel Patch 3d65a2eb
- Kernel Patch 61cb282f
- Kernel Patch a9b968e4
Workarounds
- Disable SMC-R by blacklisting the smc and smc_diag modules on hosts that do not require it.
- Avoid exposing RDMA port administration to unprivileged users or containers.
- Where feasible, fall back to standard TCP sockets until the kernel update is deployed.
# Configuration example: disable SMC modules until the kernel is patched
echo 'blacklist smc' | sudo tee /etc/modprobe.d/disable-smc.conf
echo 'blacklist smc_diag' | sudo tee -a /etc/modprobe.d/disable-smc.conf
sudo modprobe -r smc_diag smc
sudo update-initramfs -u
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

