CVE-2026-39045 - SSRF in Nym Exit Nodes

• CVE-2026-39045
• CVSS 3.1: AV:N/AC:L/PR:N/UI:N/S:C/C:L/I:N/A:N 6.1 / Medium
• CWE-918
• <=v1.30.0

The bug

Nym exit gateways apply an exit policy (Tor format) to inbound SOCKS5 connections. The default policy, fetched from https://nymtech.net/.wellknown/network-requester/exit-policy.txt, has port-based accept rules (port 80, 22, 443, etc.) but no rules rejecting private, loopback, or link-local ranges — no equivalent to Tor's reject private:*. The network requester adds no compensating check in code: service-providers/network-requester/src/socks5/tcp.rs:33 calls TcpStream::connect(&address) immediately after the policy filter passes.

Any mixnet user can therefore have an exit gateway connect to its own 127.0.0.1, 169.254.169.254, etc., on any accept-listed port.

PoC

./nym-socks5-client init --id test --provider <gateway-provider-address>
./nym-socks5-client run --id test
curl --socks5-hostname 127.0.0.1:1080 http://169.254.169.254/

The exit gateway queries its own cloud-metadata endpoint on behalf of an anonymous mixnet user.

Scope

Every exit gateway using the default policy is vulnerable to the SSRF. To confirm it's exploitable in practice I tested 28 cloud-hosted gateways identified via whois and got cloud metadata back from 12 of them across Linode, DigitalOcean, AWS, and Vultr. Exposed data: instance IDs, internal network config, authorized SSH public keys, and account identifiers. IAM-role-bound AWS nodes would also leak short-lived STS credentials (none in my sample). Port 22 is accept-listed too, so localhost sshd is reachable from the mixnet.

Fix

Commit ad56645f ("Block non-public IPR/NR checks", #6670) added an is_global_ip() check in front of the policy filter in both the network requester and the IP packet router, with a default-off allow_local_ips flag. Shipped in nym-binaries-v2026.9-venaco. The published policy file itself still contains no private-range rejects; the fix is purely code-side, so operators must upgrade for it to apply.

Timeline