CVE-2026-89507

In the Linux kernel, the following vulnerability has been resolved:

RDMA/ucma: Lock the handler in ucma write cm event()

ctx->file may only be changed under the handler lock and the xa lock, which is what stops uevents being queued for a ctx while ucma migrate id() moves it to another file. The CM core takes that lock before invoking ucma event handler(), but the write() paths that queue uevents themselves do not.

ucma write cm event() re-reads ctx->file for each of its four dereferences, so ucma migrate id() can swap it mid-sequence:

mutex lock(&ctx->file->mut);			/* file A */
list add tail(&uevent->list, &ctx->file->event list);	/* file B */
mutex unlock(&ctx->file->mut);			/* file B */
wake up interruptible(&ctx->file->poll wait);	/* file B */

The window is the mutex lock() itself: the writer sleeps in it while the migration reassigns ctx->file. The list add tail() then runs on file B's event list holding only file A's mutex:

list add corruption. prev->next should be next (ffff888101320f30), but was ffff88814a08c418. (prev=ffff88814a075c18). kernel BUG at lib/list debug.c:32! Call Trace: ucma write cm event+0x36e/0x5e0

and file A's mut is left held forever, wedging its next writer in D state. The uevent is also stranded on a list ucma cleanup ctx events() will not walk, so it outlives its context. /dev/infiniband/rdma cm is 0666 and no RDMA device is involved, so an unprivileged user reaches all of this.

Take the handler lock, as ucma cleanup mc events() does; ctx->cm id is pinned by the ucma get ctx() reference.