CVE-2026-99660: mm/userfaultfd: write-lock the VMA before waking waiters when unregistering
In the Linux kernel, the following vulnerability has been resolved:
mm/userfaultfd: write-lock the VMA before waking waiters when unregistering
userfaultfdunregister() wakes pending missing faults before clearing the VMA's userfaultfd registration. With per-VMA fault locking, holding mmaplock for write does not prevent another fault from queuing after the wakeup but before userfaultfdclearvma() write-locks the VMA. That waiter can then remain asleep indefinitely.
Call vmastartwrite() before waking waiters. Existing fault handlers queue themselves before dropping the VMA read lock, and new readers are excluded until mmapwriteunlock().
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 7.3.5 - Compensating control
In the Linux kernel, call vma_start_write() before waking waiters in userfaultfd_unregister().
Frequently Asked Questions
What is the practical impact of this flaw?
A faulting task can remain asleep indefinitely during userfaultfd unregistration, causing a hang or denial of service for the affected workload.
What conditions are needed to trigger it?
The issue requires userfaultfd use with pending missing-page faults and a race between unregistering the userfaultfd and another fault queuing on the VMA. The race is associated with kernels using per-VMA fault locking.
Are systems that do not use userfaultfd exposed?
The described race occurs in userfaultfd_unregister() and involves waiters for missing faults. Systems or workloads that do not use userfaultfd missing-fault handling are not described as triggering this condition.
What should be done if patching cannot be performed immediately?
Avoid or limit userfaultfd missing-fault workflows that unregister userfaultfd registrations while faults may still be pending. This reduces exposure to the described unregister-versus-fault race.