CVE-2025-71117: 
Linux Kernel vulnerability analysis and mitigation

CVE-2025-71117 is a deadlock vulnerability in the Linux kernel's block I/O subsystem caused by improper locking (CWE-667) when freezing request queues from within sysfs store callbacks. It affects Linux kernel versions 6.11 and 6.11.1 through 6.18.2, as well as 6.19-rc1 through 6.19-rc8. The vulnerability was published on January 14, 2026, with NVD initial analysis completed on March 25, 2026. It carries a CVSS v3.1 base score of 5.5 (Medium) (Feedly, kernel.org).

The root cause is improper locking (CWE-667): the blk_mq_freeze_queue() and blk_mq_unfreeze_queue() calls inside sysfs store callbacks for attributes io_poll_delay, io_timeout, nomerges, read_ahead_kb, and rq_affinity can create a circular wait condition. The deadlock occurs when multipathd holds a dm-multipath lock while attempting to call dm_table_set_restrictions(), and simultaneously a udev worker process writes to a sysfs queue attribute (e.g., read_ahead_kb), causing blk_mq_freeze_queue_wait() to block indefinitely. The interaction is specifically triggered when the dm-multipath driver is configured with the queue_if_no_path option. The fix removes unnecessary queue freeze calls from these callbacks and adds the __data_racy annotation to request_queue.rq_timeout to suppress KCSAN data race reports (Feedly, kernel.org).

Successful triggering of this vulnerability results in a system deadlock where the kernel scheduler becomes blocked, causing hung processes and potential system hangs that may require a hard reboot — a high availability impact with no confidentiality or integrity impact. Systems using dm-multipath with queue_if_no_path are at highest risk, particularly during boot when udev workers concurrently write sysfs queue attributes. Additionally, even without a full deadlock, the unnecessary queue freezing operations introduce performance degradation during system boot on systems that set sysfs attributes synchronously (Feedly).

• Logs: Kernel logs (dmesg or /var/log/kern.log) showing hung task warnings involving multipathd or udev-worker processes; stack traces referencing blk_mq_freeze_queue_wait, queue_ra_store, dm_table_set_restrictions, or dm_swap_table.
• Process: Processes in uninterruptible sleep (D state) — observable via ps aux or top — specifically multipathd or udev-worker that do not recover.
• System Behavior: System becoming unresponsive during boot or during dm-multipath reconfiguration events, particularly on storage-heavy systems with multipath configured.

Update the Linux kernel to version 6.18.3 or later, which removes the unnecessary blk_mq_freeze_queue() / blk_mq_unfreeze_queue() calls from the affected sysfs store callbacks. Two upstream commits address this issue: commit 3997b31 and commit 935a20d. As a temporary workaround on unpatched systems, avoid simultaneous dm-multipath reconfiguration operations and sysfs attribute modifications, and prioritize patching on systems using dm-multipath with queue_if_no_path enabled (Feedly, kernel.org).