CVE-2026-31557: nvmet: move async event work off nvmet-wq
In the Linux kernel, the following vulnerability has been resolved:
nvmet: move async event work off nvmet-wq
For target nvmetctrlfree() flushes ctrl->asynceventwork. If nvmetctrlfree() runs on nvmet-wq, the flush re-enters workqueue completion for the same worker:-
A. Async event work queued on nvmet-wq (prior to disconnect): nvmetexecuteasyncevent() queuework(nvmetwq, &ctrl->asynceventwork)
nvmetaddasyncevent() queuework(nvmetwq, &ctrl->asynceventwork)
B. Full pre-work chain (RDMA CM path): nvmetrdmacmhandler() nvmetrdmaqueuedisconnect() nvmetrdmaqueuedisconnect() queuework(nvmetwq, &queue->releasework) processonework() lock((wqcompletion)nvmet-wq) <--------- 1st nvmetrdmareleasequeuework()
C. Recursive path (same worker): nvmetrdmareleasequeuework() nvmetrdmafreequeue() nvmetsqdestroy() nvmetctrlput() nvmetctrlfree() flushwork(&ctrl->asynceventwork) flushwork() touchwqlockdepmap() lock((wqcompletion)nvmet-wq) <--------- 2nd
Lockdep splat:
============================================ WARNING: possible recursive locking detected 6.19.0-rc3nvme+ #14 Tainted: G N -------------------------------------------- kworker/u192:42/44933 is trying to acquire lock: ffff888118a00948 ((wqcompletion)nvmet-wq){+.+.}-{0:0}, at: touchwqlockdepmap+0x26/0x90
but task is already holding lock: ffff888118a00948 ((wqcompletion)nvmet-wq){+.+.}-{0:0}, at: processonework+0x53e/0x660
3 locks held by kworker/u192:42/44933: #0: ffff888118a00948 ((wqcompletion)nvmet-wq){+.+.}-{0:0}, at: processonework+0x53e/0x660 #1: ffffc9000e6cbe28 ((workcompletion)(&queue->releasework)){+.+.}-{0:0}, at: processonework+0x1c5/0x660 #2: ffffffff82d4db60 (rcureadlock){....}-{1:3}, at: flushwork+0x62/0x530
Workqueue: nvmet-wq nvmetrdmareleasequeuework [nvmetrdma] Call Trace: flushwork+0x268/0x530 nvmetctrlfree+0x140/0x310 [nvmet] nvmetcqput+0x74/0x90 [nvmet] nvmetrdmafreequeue+0x23/0xe0 [nvmetrdma] nvmetrdmareleasequeuework+0x19/0x50 [nvmetrdma] processonework+0x206/0x660 workerthread+0x184/0x320 kthread+0x10c/0x240 retfromfork+0x319/0x390
Move async event work to a dedicated nvmet-aen-wq to avoid reentrant flush on nvmet-wq.
Affected Software
Event History
Frequently Asked Questions
Which deployments are most relevant to this issue?
Systems using the Linux NVMe target (nvmet) subsystem are exposed when asynchronous event work and queue teardown can run through the nvmet workqueue. The described triggering path specifically involves the NVMe-over-RDMA connection-manager disconnect and release-work path.
What conditions are needed to trigger the problem?
No authentication or user interaction is described. Exploitation requires an attacker or other initiating party to cause the relevant NVMe target asynchronous-event and disconnect/queue-release sequence, including the RDMA CM path shown in the report.
What is the operational impact?
The reported effect is recursive workqueue completion locking when controller cleanup flushes async event work from the same nvmet workqueue worker. This produces a lockdep recursive-locking warning and is rated as a high-severity availability issue.
How can I determine whether my kernel includes the fix?
The available data does not identify affected or fixed kernel version numbers. Use the referenced stable kernel commits to determine whether the fix is present in the kernel source or package used by the affected system.