Linux kernel io_uring zero-copy send: unchecked chained notification context on completion

CVSS 7.8CVE-2025-68317Kernel, userspace & hypervisorcurated

Impact

io_uring zero-copy send is supposed to link a ubuf_info only to requests originating from the same ring context; the fix adds the missing check after ambiguous syzkaller reports of notification completions crossing contexts. The upstream commit message does not establish a demonstrated exploit, and the 7.8 score is the kernel CNA's standard local-escalation rating rather than a researched impact. It matters on a GPU fleet because io_uring is unprivileged by default and is exactly the kind of interface a tenant workload or sidecar can reach from inside a container, so a memory-safety error here is reachable without any device access.

Who can reach it

Local, unauthenticated beyond holding a shell or running code on the node - any process able to create an io_uring instance and issue zero-copy sends, including a tenant container unless io_uring is blocked by seccomp or the io_uring_disabled sysctl.

What to do

Update to a stable kernel carrying the linked commits through your distribution. Mitigation without a reboot: set kernel.io_uring_disabled=2 or block io_uring_setup in the container seccomp profile, which costs throughput for workloads that use it but needs no downtime. Taking the fix itself means draining the node and rebooting.

References

https://nvd.nist.gov/vuln/detail/CVE-2025-68317 https://git.kernel.org/stable/c/aaafd17d3f4be2c15539359a5b4bfa00237f687f https://git.kernel.org/stable/c/d664a3ce3a604231a0b144c152a3755d03b18b60

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.