This is a plugin that provides fine grained action level authorization for agents. Any MCollective agent plugins based on SimpleRPC can be restricted with authorization plugins like this one.

Installation

Follow the plugin deployment guide, using the libdir copy method and placing the actionpolicy.rb and actionpolicy.ddl files in the util directory.

Note that it is not currently possible to use the 'mco plugin package' command to package this plugin.

Configuration

There are three settings available for the actionpolicy plugin:

allow_unconfigured -- whether to allow requests to agents that do not have policy files configured. Boolean, with allowed values of 0, 1, y, n; values of true or false are not allowed. Defaults to 0.
enable_default -- whether to use a default policy file. Boolean, with allowed values of 0, 1, y, n; values of true or false are not allowed. Defaults to 0.
default_name -- the name of the default policy file, if enable_default is set to 1 or y.

General authentication configuration options can be set in the server config file.

# Enables system wide rpc authorization
rpcauthorization = 1
# Sets the authorization provider to use the actionpolicy plugin
rpcauthprovider = action_policy
# Allow requests to agents without policies
plugin.actionpolicy.allow_unconfigured = 1

Default Policy Files

You can optionally have a default policy file that applies in the absence of an agent-specific policy file.

plugin.actionpolicy.enable_default = 1
plugin.actionpolicy.default_name = default

This allows you to create a policy file called default.policy which will be used unless a specific policy file exists. Note that if both allow_unconfigured and enable_default are configured, all requests will go through the default policy, as enable_default takes precedence over allow_unconfigured.

Usage

Policies are defined in files like <configdir>/policies/<agent>.policy

Example: Puppet agent policy file

# /etc/mcollective/policies/puppet.policy
policy default deny
allow   cert=admin          *                       *                *
allow   cert=acme-devs      *                       customer=acme    acme::devserver
allow   cert=acme-devs      enable disable status   customer=acme    *

# /etc/mcollective/policies/service.policy
policy default deny
allow   cert=puppet-admins  restart                 (puppet().enabled=false and environment=production) or environment=development

The above policy can be described as:

Allow the admin user to invoke all Puppet actions on all servers.
Allow the acme-devs user to invoke all Puppet actions on machines with the fact customer=acme and the config class acme::devserver
Allow the acme-devs user to invoke the enable, disable and status actions on all other machines with fact customer=acme
Allow the puppet-admins user to restart services at any time in development but in production only when Puppet has been disabled
All other commands get denied

Policy File Format

Policy files must have the following format:

Any lines starting with # are comments.
A single policy default deny or policy default allow line is permitted; it can go anywhere in the file. This default policy will apply to any commands that don't match a specific rule. If you don't specify a default policy, the value of the plugin.actionpolicy.allow_unconfigured setting will be used as the default.
Any number of policy lines are permitted. These must be tab delimited lines with either four or five fields (the final field is optional) in the following order:
1. allow or deny
2. Caller ID --- must be either * (always matches) or one caller ID string (see below)
3. Actions --- must be either * (always matches) or a space-separated list of actions
4. Facts --- may be either * (always matches), a space-separated list of fact=value pairs (matches if every listed fact matches), or any valid compound filter string
5. Classes --- may be completely absent (always matches), * (always matches), a space-separated list of class names (matches if every listed class is present), or any valid compound filter string

Notes

Like firewall rules, policy lines are processed in order --- ActionPolicy will allow or deny each request using the first rule that matches it. A policy line matches a request if every field after the allow/deny field matches.
Policy lines must use hard tabs; editor features that convert tabs to spaces (like Vim's expandtab) will result in non-functional policy lines.
Compound filter strings may match on facts, classes, and data plugins (MCollective 2.2.x or later). When using data plugins in action policies, you should avoid using slow ones, as this will impact the response times of agents, the client waiting time, etc.

Caller ID

Caller ID strings are always of the form <kind>=<value>, but both the kind and the value of the ID will depend on your security plugin. See your security plugin's documentation or code for details.

The recommended SSL security plugin sets caller IDs of cert=<NAME>, where <NAME> is the filename of the client's public key file (minus the .pem extension). So a request validated with the puppet-admins.pem public key file would be given a caller ID of cert=puppet-admins. This kind of caller ID is cryptographically authenticated.
The PSK security plugin defaults to caller IDs of uid=<UID>, where <UID> is the local UID of the client process. There are several other options available, which can be configured with the plugin.psk.callertype setting. None of PSK's caller IDs are authenticated, and you should generally not be relying on authorization at all if you are using the PSK security plugin.

Hardcoding ActionPolicy Into a Specific Agent

Instead of using the site-wide authorization settings (as described above), you can also hardcode authorization plugins in your agents:

module MCollective::Agent
    class Service<RPC::Agent
        authorized_by :action_policy

        # ...
    end
end