Linux kernel io_uring: io_link_skb compares a notification against itself, skipping the context validation check

CVSS 7.8CVE-2025-39963Kernel, userspace & hypervisorcurated

Impact

io_link_skb() derived prev_notif from the current notification instead of from prev_nd, so the check that two linked zero-copy send notifications belong to the same context compared an object with itself and always passed. The guard that was supposed to stop notifications from different contexts being chained therefore did nothing. On a GPU node this lives in the io_uring zero-copy transmit path that high-throughput serving and data-loader processes use; the record describes an incorrect reference and a bypassed validation, and does not demonstrate a working local-root exploit, so treat it as kernel state corruption of unclear reachability rather than a confirmed escalation. Any local process able to use io_uring zero-copy send touches this code without privilege.

Who can reach it

Local, unprivileged: a process using io_uring zero-copy send (SEND_ZC) notifications. No authentication beyond local code execution, and tenant containers qualify unless seccomp blocks io_uring.

What to do

Update to a stable kernel using prev_nd for prev_notif (three backports linked) and reboot - drain and reboot per node. Blocking io_uring syscalls in the container seccomp profile keeps tenants out of the path until the reboot window arrives.

References

https://nvd.nist.gov/vuln/detail/CVE-2025-39963 https://git.kernel.org/stable/c/2c139a47eff8de24e3350dadb4c9d5e3426db826 https://git.kernel.org/stable/c/50a98ce1ea694f1ff8e87bc2f8f84096d1736f6a https://git.kernel.org/stable/c/a89c34babc2e5834aa0905278f26f4dbe4b26b76

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.