CVE-2026-99715 in Linux
Summary
by MITRE • 11/20/2026
In the Linux kernel, the following vulnerability has been resolved:
nvme-rdma: fix NULL queue->ctrl dereference on reconnect
Since commit 2a8513091d2f ("nvme-rdma: parallelize I/O queue allocation and startup"), a reconnect that gets more I/O queues than before grows the tag set before the new queues are set up. queue->ctrl is only set by nvme_rdma_setup_io_queues(), so I/O submitted in between reaches nvme_rdma_queue_rq() on a queue that was never allocated and dereferences a NULL queue->ctrl:
BUG: kernel NULL pointer dereference, address: 000000000000034c RIP: 0010:nvme_rdma_queue_rq+0x52/0xb90
Set queue->ctrl for every queue when allocating the controller. Such requests are then requeued like on any other queue that is not live.
If you want to get the best quality for vulnerability data then you always have to consider VulDB.
Analysis
by VulDB Data Team • 11/20/2026
The Linux kernel vulnerability identified in the nvme-rdma subsystem represents a critical race condition arising from changes to parallelize I/O queue allocation and startup processes. Specifically, this issue was introduced by commit 2a8513091d2f which aimed to improve performance by allowing concurrent initialization of multiple input-output queues for NVMe over RDMA devices. The fundamental flaw lies in the ordering of operations during a reconnect scenario where the number of I/O queues increases compared to the previous connection state. In this specific sequence, the kernel expands the tag set before fully initializing the new queue structures. Because the control pointer within each queue structure is only populated by the nvme_rdma_setup_io_queues function, any input-output request submitted during this narrow window targets a queue object that has not been properly allocated or initialized.
This architectural oversight leads to a direct NULL pointer dereference when the kernel attempts to process these requests via the nvme_rdma_queue_rq routine. The system crashes with a BUG message indicating a null pointer access at address 0x34c, which corresponds to an offset within the queue control structure that was never populated. From a technical perspective, this is a classic use-after-free or uninitialized memory access scenario where the software attempts to interact with data structures that do not yet exist in a valid state. The operational impact of this vulnerability includes immediate kernel panic and system crash for affected systems utilizing NVMe over RDMA storage controllers during reconnection events involving queue count changes. This effectively results in a denial-of-service condition, disrupting availability for critical workloads relying on high-performance networked storage solutions.
From a classification standpoint, this flaw aligns with CWE-476 which denotes NULL Pointer Dereference vulnerabilities where software fails to check if an object is null before using it. Furthermore, the attack vector relates to timing conditions and race states inherent in concurrent execution paths, mapping closely to ATT&CK techniques involving resource exhaustion or availability impact through system instability. The vulnerability highlights the risks associated with optimizing parallel processing without ensuring strict synchronization of dependent initialization steps across all affected components.
The resolution implemented by the kernel maintainers addresses this issue by enforcing proper initialization order during controller allocation. Specifically, the fix ensures that queue->ctrl is set for every allocated queue at the time of controller setup rather than deferring it to later stages. This change guarantees that any requests arriving while queues are being provisioned will encounter a valid control structure and be handled appropriately. Instead of causing a crash through invalid memory access, these premature requests are now requeued similarly to how other non-live queues handle incoming traffic. This approach maintains system stability by deferring processing until the queue is fully operational, thereby eliminating the race condition window entirely.
Mitigation strategies for organizations running affected kernel versions involve applying the latest security patches that include this specific fix. Administrators should monitor their NVMe over RDMA infrastructure for any signs of unexpected reconnections or latency spikes which might indicate attempted exploitation or accidental triggering of the bug during normal operation. Since this vulnerability requires a reconnect event with an increased queue count, ensuring stable connection parameters can reduce exposure risk until patches are deployed. Long-term remediation includes rigorous code review practices that verify initialization sequences in parallelized subsystems to prevent similar race conditions in future kernel updates.