Impact

There is a potential vulnerability in Traefik managing the ACME HTTP challenge.

When Traefik is configured to use the HTTPChallenge to generate and renew the Let's Encrypt TLS certificates, the delay authorized to solve the challenge (50 seconds) can be exploited by attackers (slowloris attack). Patches

- https://github.com/traefik/traefik/releases/tag/v2.10.6 - https://github.com/traefik/traefik/releases/tag/v3.0.0-beta5

Workarounds

Replace the HTTPChallenge with the TLSChallenge or the DNSChallenge.

For more information

If you have any questions or comments about this advisory, please open an issue.

Summary

The traefik docker container uses 100% CPU when it serves as its own backend, which is an automatically generated route resulting from the Docker integration in the default configuration.

Details

While attempting to set up Traefik to handle traffic for Docker containers, I observed in the webUI a rule with the following information:

Host(traefik-service) | webwebsecure | traefik-service@docker | traefik-service

I assumed that this is something internal; however, I wondered why it would have a host rule on the web entrypoint configured.

So I have send a request with that hostname with curl -v --resolve "traefik-service:80:xxx.xxx.xxx.xxx" http://traefik-service. That made my whole server unresponsive.

I assume the name comes from a docker container with that name, traefik itself: localhost ~ # docker ps CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES d1414e74aec7 traefik:v2.10 "/entrypoint.sh trae…" 4 minutes ago Up 4 minutes 0.0.0.0:80->80/tcp, :::80->80/tcp, 0.0.0.0:443->443/tcp, :::443->443/tcp, 127.0.0.1:8080->8080/tcp traefik.service

PoC

1. Start traefik with docker run --rm -v /var/run/docker.sock:/var/run/docker.sock -p 80:80 --name foo -p 8080:8080 traefik:v2.10 --api.insecure=true --providers.docker

2. curl -v --resolve "foo:80:127.0.0.1" http://foo

looks like this creates an endless loop of request.

Knowing the name of the docker container seems to be enough to trigger this, if the docker backend is used.

Impact

Server is unreachable and uses 100% CPU

Summary When a request is sent to Traefik with a URL fragment, Traefik automatically URL encodes and forwards the fragment to the backend server. This violates the RFC because in the origin-form the URL should only contain the absolute path and the query.

When this is combined with another frontend proxy like Nginx, it can be used to bypass frontend proxy URI-based access control restrictions.

Details For example, we have this Nginx configuration:

location /admin { deny all; return 403; } This can be bypassed when the attacker is requesting to /#/../admin

This won’t be vulnerable if the backend server follows the RFC and ignores any characters after the fragment.

However, if Nginx is chained with another reverse proxy which automatically URL encode the character # (Traefik) the URL will become

/%23/../admin

And allow the attacker to completely bypass the Access Restriction from the Nginx Front-End proxy.

Here is a diagram to summarize the attack:

!image

PoC !image (1)

This is the POC docker I've set up. It contains Nginx, Traefik proxies and a backend server running PHP.

https://drive.google.com/file/d/1vLnA0g7N7ZKhLNmHmuJ4JJjVJ2akNMt/view?usp=sharing

Impact This allows the attacker to completely bypass the Access Restriction from Front-End proxy.