# EEF-CVE-2026-125275: A device certificate whose CA isn't registered can keep a TLS handshake running forever

- Database: EEF
- Published: 2026-11-17
- Modified: 2026-11-17

## Tracked as

- [CVE-2026-125275](https://cve.now/CVE-2026-125275.md): nerves\_hub\_web unauthenticated denial of service in TLS handshake

## Description

## Summary

Loop with Unreachable Exit Condition ('Infinite Loop') vulnerability in device certificate verification in nerves-hub nerves_hub_web allows an unauthenticated remote attacker to cause a denial of service via a TLS client certificate whose CA is not registered. When a client certificate chains to a root in the device endpoint's trust store but no CA in the chain is registered in NervesHub, `NervesHub.SSL.decide/2` refuses it with the reason `:unknown_ca`. Erlang/OTP `ssl` reads that reason as a missing trust anchor, looks for an alternative root, finds the same root and validates again without end. The handshake process queries the database on each pass and keeps running after the client disconnects, until the node restarts.

The device endpoint trusts the public roots from CAStore unless `DEVICE_SSL_CACERTFILE` is set, so a client certificate from a public CA whose chain allows client authentication is sufficient. A chain that ends at a root outside the trust store is refused once.

This issue affects nerves_hub: from 1.0.0 before 2.5.4.

## Details

**1. Refusal reason.** The device endpoint gives `ssl` the verify function `NervesHub.DeviceLink.PeerVerification.verify_fun/3`, which asks `NervesHub.SSL.decide/2`. For a `:valid_peer` event on a certificate that is not on file, `decide/2` tries to register the certificate. `check_known_ca/1` in `lib/nerves_hub/ssl.ex` looks for the signer among the registered CAs and returns `:unknown_ca` when it finds none. `do_verify/2` passes that to `ssl` as `{:fail, :unknown_ca}`.

**2. Retry in `ssl`.** `unknown_ca` is also the reason Erlang/OTP uses for a path with no trust anchor. In `ssl_handshake:path_validate/11`, a `{bad_cert, unknown_ca}` result makes `ssl` look for a cross-signed alternative root and validate again. Unlike the `root_cert_expired` branch, this branch does not add the root it tried to the list of roots to skip.

**3. Loop.** When the chain ends at a self-signed root in the trust store, `ssl` finds the same root, calls the verify function again, and receives `unknown_ca` again. The recursion is a tail call, so the process does not grow or crash. It does not return to its mailbox, so it sees neither the handshake timeout nor the closed socket. Each pass runs the database queries of the registration checks.

## Proof of concept

1. Make a self-signed root and a client certificate signed by it, directly or through an intermediate that the client sends. Do not register either CA in NervesHub.
2. Run a device endpoint whose trust store holds that root, by adding it to the `DEVICE_SSL_CACERTFILE` bundle. Against the default trust store, a client certificate from a public CA replaces steps 1 and 2.
3. Open one TLS connection to the device endpoint with that certificate, then disconnect.
4. The handshake process on the server keeps running. The `nerves_hub.ssl.fail` telemetry event with reason `unknown_ca` and the database queries repeat until the node restarts.

## Impact

An unauthenticated client can leave a handshake process on a device node running and querying the database after it disconnects. Each further connection adds another such process, and they stop only when the node restarts. Devices and users of the instance lose service while the nodes and the database are saturated.

## Workarounds

Set `DEVICE_SSL_CACERTFILE` to a bundle that holds no public root and no root that signs device certificates, for example a file with one self-signed certificate made for this purpose. A certificate from an unregistered CA is then refused once and the handshake ends. Devices whose certificate or CA is registered in NervesHub continue to connect.

Restarting a node ends the handshakes that are already stuck on it.

## Configurations

The certificate chain must end at a root in the device endpoint's trust store. With `DEVICE_SSL_CACERTFILE` unset, which is the default from 2.0.0, the trust store holds the public roots from CAStore. With `DEVICE_SSL_CACERTFILE` set to a bundle that holds a root which signs device certificates, any certificate from that root whose CA is not registered is sufficient, including the certificate of a deleted device.

The retry in Erlang/OTP `ssl` was confirmed on Erlang/OTP 28.3, 29.0 and 29.1. Earlier releases were not checked.

## References

- https://github.com/nerves-hub/nerves_hub_web/security/advisories/GHSA-8j48-4mg5-7v6v
- https://cna.erlef.org/cves/CVE-2026-125275.html
- https://github.com/nerves-hub/nerves_hub_web/commit/a49fb8875f3aab40062881028ef4e406d985aaa9
- https://github.com/nerves-hub/nerves_hub_web/commit/064e78cc189dd9b87748746f44179632a52ddf58

Page: https://cve.now/EEF-CVE-2026-125275
