Limnoria-doc/use/security.rst

Security in Limnoria

Some security features are disabled by default. We know this is arguable, but enabling them would make it quite hard to start using the bot. This guide is for people who want to enable these features to make their bot as secure as possible.

Note

Limnoria (or Gribble or Supybot) have never been audited by a security professional. We do the best we can to make it secure, but we cannot guarantee it is completely safe.

Trust in network operators

As you may know, by default, it is possible to do anything from IRC, including loading the Unix plugin and using the @call command. The only safeguard is checking the user calling the commands is authenticated as the owner of the bot; and network operators are able to spoof hostmasks and collect your password, thus allowing them to execute commands as the owner.

Although network operators of most well-known IRC networks are not known to do that, you should be aware of that risk.

Starting on commit 4f6a5e7db (version 2017.10.01), there is a new configuration variable, supybot.commands.allowShell, to prevent malicious network operators from getting shell access on your bot's computer. It defaults to True to make it easy for new users to install plugins using PluginDownloader, but it is recommended you set it to False if you do not care about that feature.

Finally, you can remove the owner user account entirely (or remove the owner capability) for that account. This causes every privileged commands to be disabled, so neither you nor server operators can access it. Channel-specific configuration variables can still be configured by users with the #channel,op capability (if any), but global configuration variables can only be modified by accessing the config files.

Network connections / SSL

Background on SSL certification validation

It is often believed using SSL magically makes impossible any attack on your connection (from the bot to the server). It is true that it prevents passive eavesdropping, but other attack methods are still possible.

The main one involves man-in-the-middle, ie. someone acting as a proxy between you (your bot, in that case) and the IRC network. If certificates are not validated, the attacker can allow you to connect to itself using their own SSL certificate, and you would never know about it.

This is why it is important to check the SSL certificate of the server you connect to: an attacker cannot spoof a certificate, or the trust of a Certificate Authority in a network's certificates.

Of course, this assumes there is no bug in your SSL library, the network's, and the protocols involved.

Certificate validation in Limnoria

Until version 2016.02.24, Limnoria did not support certificate validation. Starting from this version, it is possible, but disabled by default, in order to not break existing bots when updating.

Certificate validation can be enabled using this command:

@config supybot.protocols.ssl.verifyCertificates true

Available validation mechanisms are Certification Authorities and fingerprint checking.

Certificate Authorities

By default, Limnoria only checks certificates using CA certificates installed on your system. However, some networks use a CA that is not trusted by your system, such as CACert.

Limnoria allows you to add a CA certificate for a network:

@config networks.NETWORKNAME.ssl.authorityCertificate /path/to/the/certificate.crt

Note that you are responsible for making sure this is the right certificate for the CA, and trust this CA to sign correctly certificates valid for the network's hostname(s).

Fingerprint checking

Alternatively, for networks that do not use a CA, you can give Limnoria the list of fingerprints of certificates used by the network:

@config supybot.networks.NETWORKNAME.ssl.serverFingerprints: <fingerprint1> <fingerprint2> ...

Adding fingerprints will disable CA verifications (useful if you do not want to trust CAs).

Note that you are responsible for giving the correct list of fingerprints.

Supported python versions

Fingerprint checking and CA validation are available in all Python versions supported by Limnoria.

Flooding via command abuse

Limnoria answers at most one message per command, but its message can be rather long (up to about 450 to 500 characters) for even a small command.

If this is undesirable for you, you can take the following measures:

• Limit the size of a single message with supybot.reply.mores.length.
• Limit how many messages the @more command may be called to get a response to a command: supybot.reply.mores.maximum
• Disable large error replies with supybot.reply.error.detailed and supybot.reply.error.noCapability, and/or send them in private with supybot.reply.error.inPrivate.
• And check out the various variables in supybot.abuse.flood.

For old bot configurations, you may also want to set the -scheduler capability to prevent users from using the @scheduler add and @scheduler repeat commands (bot configurations created with Limnoria versions greater than 2020.05.13 already have this by default).

We also recommend you report users abusing your bot to network operators, so they take extra measures against these users if this is against their network's policy.

Hardening

By default, Limnoria exposes much of its configuration. This is by design, to improve discoverability and debugging.

Again, if this is undesirable to you, you can do the following:

• Prevent users from using the Config plugin to read the configuration: defaultcapability add -config (note that sensitive configuration variables are, of course, always hidden from users by default).
• Prevent users from listing available plugins and commands: defaultcapability add -misc.list, defaultcapability add -misc.apropos, and defaultcapability add -plugin
• Hide the version from users: defaultcapability add -misc.version, and also make sure it's not in supybot.user or supybot.plugins.Owner.quitMsg.
• Hide capabilities users are missing to run a command: supybot.reply.error.noCapability
• Replace errors with a generic reply: supybot.reply.error.detailed

Note that, when asking for help involving an error, you should enable verbose errors when providing logs (ie. reset these last values to their default), so it is easier to help you diagnose your problems.