SCTPhantom: An 18-Year-Old Flaw in Linux's SCTP Stack Leads to Root and Container Escape

Translated by a model; the original is in Russian.

· Security
A use-after-free in the SCTP networking code, present since 2008, lets a local user gain root. Tencent researchers claim they used it to break out of a container onto the host.

cover.svg

Researchers at Tencent Zhuque Lab have disclosed a vulnerability in the Linux kernel's SCTP subsystem, assigned the identifier CVE-2026-64564 and the name SCTPhantom. It is a use-after-free bug that, under the right conditions, lets a local user escalate privileges to root — and, according to the authors, escape a container onto the machine underneath it.

What the bug is

SCTP is a transport protocol capable of carrying a single connection across several network paths at once. An additional dynamic address reconfiguration feature lets the peer add and remove those addresses in the middle of a session. It is precisely at this junction that the bug lives: the kernel validates a removal request against the packet's source address, but acts on a path selected by a different address inside the message itself. A single message can carry an address, the removal of that same address, and a “remove by wildcard” — such a sequence frees the path and then dereferences an already-dead pointer. The connection is left pointing at memory the kernel has already handed back to the system.

The fix rejects a removal that targets the very path in whose context the message is being processed. The defect traces back to kernel 2.6.25 from 2008 and has been present in every release since.

Scope and caveats

The vulnerability is local rather than remote, and it requires SCTP to be reachable on the target, which narrows the set of affected systems. Where those conditions held, the lab reports obtaining root on kernel builds for Debian 13, Ubuntu 24.04, Rocky Linux 9, RHEL 9 and OpenCloudOS.

The container-escape claim rests on Tencent's own tests. According to the lab, the test kept the default seccomp profile and granted neither CAP_NET_ADMIN nor CAP_SYS_ADMIN; of eight attempts, six reached root on the host. An early version of the exploit required the addip sysctls to be enabled, but a path was then found that turns on the needed capabilities at the socket level. Outside the lab, no one has yet reproduced the results, and the name of the runtime used is not disclosed. The openKylin distribution's advisory for the same bug is limited to a kernel panic and denial of service.

The severity rating is unsettled as well: Tencent assigned 8.5 on CVSS v4.0, whereas as of August 7 the NVD had assigned neither a score nor a weakness classification. There was no public exploit code at the time of publication, and there was no entry in the CISA KEV catalog.

What to do

A fix is already out: the stable kernels 7.1.6, 6.18.42, 6.12.101 and 6.6.148, released on August 3, close the problem. Because vendors often backport fixes without bumping the version, you should not rely on the version string alone — check your distribution's tracker. Where SCTP is not needed, blacklisting the module removes the attack surface entirely.

A curious detail: Tencent credits the discovery to Corvus AI, its own multi-agent pipeline for working on the kernel, which makes SCTPhantom yet another long-dormant vulnerability surfaced with the help of machines this year.


Source: The Hacker News

Comments

Sign in, to comment.

  • No comments yet.