Skip to content

Advisories checked against the appliances

A published vulnerability in other people's code that runs on these appliances is checked here: which part runs the affected code, whether anything can reach it, and what was tried on the lab appliances. A vulnerability in this project's own code is reported privately, as SECURITY.md says.

advisory in short checked result
CVE-2026-98432 Linux, TCP over IPv4: a SYN carrying IPv4 options, and a retransmitted SYN, use the options after they are freed 2026-11-19 nothing reaches the affected code from any network; no change made

CVE-2026-98432

Published 2026-10-15 by the Linux kernel's own CNA, scored 9.8 critical (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H): its record, and the fix, ipv4: free inet_opt and ireq_opt after an RCU grace period. A peer with no credentials sends a TCP SYN carrying an IPv4 option - Record Route or Timestamp - to a listening port, and races a retransmitted SYN against the end of that connection; the kernel frees the options while it is still building a SYN-ACK from them, and can read or write kernel memory past them. Every Linux kernel from 4.4 is affected until the fixed stable releases: 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56 and 7.2.10. The record's attack goes over IPv4.

Which part runs Linux

part kernel affected code reachable
Firewall, OPNsense FreeBSD no - the fault is in Linux -
Coprocessor, XGS 3300 and XGS 136 Linux 4.14.207-10.22.03 on both, as the appliance ships it yes, and 4.14 has no fixed release no - below

Why nothing reaches the coprocessor's TCP

Read on 2026-11-19 on the second XGS 3300 and the XGS 136, through npu-run.sh:

  • One TCP listener: SSH on port 22, nothing else.
  • One address a peer could use: its Linux has IPv4 addresses on loopback and on the internal link to the host only - 169.254.254.2/30, with the host at 169.254.254.1 (the management link). No front port has an address in it, and it does not forward (ip_forward 0).
  • Link-local is never forwarded: 169.254.0.0/16 is link-local, and FreeBSD drops any packet to or from it that it would otherwise forward (RFC 3927, section 2.7; sys/netinet/ip_input.c). Nothing past the firewall can send to that address through it.
  • pf drops the attack's packet anyway: pf drops an IPv4 packet carrying options unless the rule that passes it says allow-opts. On both appliances three rules say so, and all three pass traffic out.

So the only machine that can open a TCP connection to the coprocessor is the firewall itself, and code running there as root already controls the coprocessor over that link, without needing this fault.

What was tried

On 2026-11-19, on the second XGS 3300 (OPNsense 26.7, this project's drivers loaded), from a lab machine behind one of its ports, with a route to 169.254.254.2 through the firewall:

connection from the lab machine without IP options with an IPv4 Record Route option
Coprocessor's SSH, 169.254.254.2:22, through the firewall timed out timed out
Firewall's own web GUI, on the lab machine's own network connected timed out

The option was set on the socket (IP_OPTIONS, one Record Route slot). The first row is the link-local rule; the second row's right-hand cell is pf dropping a packet with options, the packet this attack needs. On the XGS 136 the same reading was made - the same address, the same single listener, the same three outbound allow-opts rules - and the connection test was not run there: the route it needs on the lab machine would have taken the path the lab uses to control that appliance.

Not done

No change was made. Dropping IPv4 options inside the coprocessor as well - a rule in its own firewall, set again at every boot - was weighed and left out: the only sender it could stop is the host itself. This page is read again if a change ever gives the coprocessor an address a network can reach.

Discussion

The discussion of this page is held in the project's GitHub Discussions. Showing it here loads giscus from giscus.app and GitHub, and writing in it needs a GitHub account - what that means.