GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel IPv6: heap overwrite into skb_shared_info via UDPv6 MSG_MORE with MSG_SPLICE_PAGES

CVE-2026-53362Kernel, userspace & hypervisorKnown exploitedcurated

Impact

On the paged-allocation branch of __ip6_append_data(), alloclen does not account for the fraggap bytes carried over from the previous skb, so the linear area is undersized by exactly that amount and the copy runs past skb->end into the trailing skb_shared_info. An unprivileged local user reaches this from a plain UDPv6 socket using MSG_MORE together with MSG_SPLICE_PAGES, giving controlled corruption of adjacent kernel heap metadata - the usual starting point for local privilege escalation. This is host-kernel state, not per-container state, so any tenant workload that can open a UDPv6 socket sits one bug-chain away from breaking isolation for every GPU pod scheduled on that node. The record carries a KEV flag, which raises the urgency above the usual local-kernel triage.

Who can reach it

Local unprivileged user with the ability to open a UDPv6 socket - including inside a container, since no capability is required and the path is ordinary sendmsg() flag handling. No authentication beyond code execution on the node.

What to do

Update to a stable kernel containing the fix (the linked git.kernel.org commits) and reboot each node; the corrupted-accounting fix is in packet-path code that livepatching generally will not cover. On a GPU fleet this means draining and rebooting nodes, so batch it with the next kernel maintenance window unless the KEV listing forces it sooner. As a stopgap, blocking MSG_SPLICE_PAGES-capable workloads is not practical - restricting untrusted local code execution is the only real mitigation.

References

Related entries

All Kernel, userspace & hypervisor entries

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.