mikaela.github.io/_posts/2015-02-24-znc160-ssl.md

TL;DR: if you don’t verify SSL certificates, don’t use SSL!

ZNC 1.6.0 was released on 2015-02-12 21:05:48Z. It brings multiple improvements such as taking IP addresses from round-robins randomly instead of always resolving them into same IP and most notably it actually verifies SSL certificates.

• Changelog

ZNC 1.6.0 also doesn’t have option to blindly accept certificates, which would be stupid, but sadly Quakenet is right about most of people just accepting certificates blindly as people are asking how to disable the SSL certificate verification on #znc at freenode a lot.

Some people even wrote a patch and scripts to disable the verification. This isn’t a good idea as patching ZNC can cause all kinds of issues as sometimes seen with zncstrap 1 2 3. See also contributing (reporting bugs) guidelines of ZNC.

I believe same policy should apply to patching ZNC as to config files, patch ZNC or edit config file and you will forfeit all support.

And to the subject

If you don’t verify SSL certificates, you only have a false sense of security as you let anyone between your ZNC and the IRC network. This is called as Man-in the middle (or shortly MITM) attack. There are also people asking for ZNC to trust the certificate for the first time and then be alerted if the certificate changes. What if the MITM is there during your first connection attempt and then you are alerted when the real IRC server gives you wrong certificate?

So what is the correct way?

• Check the website of your IRC network in case the fingerprints are listed on their website.
• Try asking the operators of your IRC network somewhere else if you know them (like another network or email).
• This might not be so recommended, but also check the fingerprints from multiple locations.

But the IRC network has hundreds of servers with different certificates!

In this case do what was recommened before ZNC 1.6.0, check some of the servers that are geographically close to you and use them.

Checking the fingerprint from multiple locations

I have shell function (which you can find later on this page) which I run from multiple places:

• my home, Kotka, Finland
• Kapsi (shell), somewhere in Finland
• my VPS, DigitalOcean, London, the UK

# Get server SSL certificate fingerprint in MD5, SHA1 and SHA256.
# Note that OpenSSL doesn't support IPv6 at time of writing (2015-01-13).
serversslcertfp () {
    SSSLCFFN=$(openssl s_client -showcerts -connect $1 < /dev/null)
    # To see all validity information
    echo "$SSSLCFFN"
    # For getting the fingerprints
    echo "$SSSLCFFN" | openssl x509 -md5 -fingerprint -noout
    echo "$SSSLCFFN" | openssl x509 -sha1 -fingerprint -noout
    echo "$SSSLCFFN" | openssl x509 -sha256 -fingerprint -noout
    echo "$SSSLCFFN" | openssl x509 -sha512 -fingerprint -noout
    unset SSSLCFFN
}

I hope this article has helped you to understand the issues with blindly accepting SSL certificates or at least to understand that if you don’t want to verify SSL certificates, don’t use SSL.

• Updated on 2015-02-26 10:43Z: just use environment variables in the function like suggested by @DarthGandalf on #znc.

I am asked to verify fingerprint for network with valid certificate

Added on 2015-09-03.

There are usually three causes for this. Lets use freenode as example network.

1. You don’t have ca-certificates package installed, so your system trusts no certificate authority. Install it and try again.
2. You are connecting to wrong address. freenode’s certificate is valid for *.freenode.net, but there are CNAMEs pointing there. If you connect to CNAME and the certificate isn’t valid for that CNAME, the certificate is invalid.
   • You should always connect to either irc.freenode.net or chat.freenode.net where it points to.
3. There is MITM which is unlikely, but unlikely is not impossible. Validating the certificates either by trusted certificates or verifying the fingerprints securely manually protect you from this. If MITM is the case, you shouldn’t connect.

---

As I seem to be updating this page more than I originally thought I should probably add this link to changelog here.