Due to the broadcast paradigm of mcollective security is a complex topic to discuss.
This discussion will focus on strong SSL and AES+RSA based security plugins, these are not the default or only option but is currently the most secure. Both the SSL security plugin and AES security plugin provide strong caller identification, this is used by the SimpleRPC framework for Authorization and Auditing.
As every organisation has its own needs almost all aspects of the security system is pluggable. This is an overview of the current state of SSL based Authentication, Authorization and Auditing.
The image above is a reference to use in the following pages, it shows a MCollective Setup and indicates the areas of discussion.
The focus here is on ActiveMQ, some of the details and capabilities will differ between middleware systems.
- Security Overview
Client Connections and Credentials
Every STOMP connection has a username and password, this is used to gain basic access to the ActiveMQ system. We have a ActiveMQ Security sample setup documented.
ActiveMQ can use LDAP and other security providers, details of this is out of scope here, you should use their documentation or the recently released book for details of that.
The AES+RSA securith plugin
When using the AES security plugin each user also gets a private and public key, like with SSH you need to ensure that the private keys remain private and not shared between users.
This plugin can be configured to distribute public keys at the cost of some security, you can also manually distribute keys for the most secure setup.
The public / private key pair is used to encrypt using AES and then to encrypt the key used during the AES phase using RSA. This provides encrypted payloads securing the reply strucutres.
The client embeds a caller structure in each request, if RSA decryption pass the rest of the MCollective agents, auditing etc can securely know who initiated a request.
This caller is used later during Authorization and Auditing.
This plugin comes with a significant setup, maintenance and performance overhead if all you need is to securely identify users use the SSL security plugin instead.
The SSL security plugin
When using the SSL security plugin each user also gets a private and public certificate, like with SSH you need to ensure that the private keys remain private and not be shared between users. The public part needs to be distributed to all nodes.
The private key is used to cryptographically sign each request being made by a client, later the public key will be used to validate the signature for authenticity.
The client embeds a caller structure in each request, if SSL signature validation pass the rest of the MCollective agents, auditing etc can securely know who initiated a request.
This caller is used later during Authorization and Auditing.
Connection to Middleware
Enabling TLS throughout will secure your connections from any kind of sniffing and Man in The Middle attacks. Unfortunately the Rubygem we use do not provide options for enforcing a specific CA etc. The authors are willing to extend it to support these based on requests, file support tickets if you need our help in working with them as we already have a good working relationship.
At present there is a bug in the TLS setup with ActiveMQ and RabbitMQ does not support it at all. If you need security of the payloads in your messages use the AES security plugin.
Middleware Authorization and Authentication
As mentioned above ActiveMQ has it’s own users and every node and client authenticates using these.
In addition to this you can on the middleware layer restrict access to topics, you can for example run a development and production collective on the same ActiveMQ infrastructure and allow your developers access to just the development collective using these controls. They are not very fine grained but should be a import step to configure for any real setup.
We have a sample ActiveMQ Security setup documented that has this kind of control.
By combining this topic level restrictions with Subcollectives you can create virtually isolated groups of nodes and give certain users access to only those subcollectives. Effectively partitioning out a subset of machines and giving secure access to just those.
Node connections and credentials
As with the client the node needs a username and password to connect to the middleware and can also use TLS.
It’s not a problem if all the nodes share a username and password for the connection since generally nodes do not make new requests. You can enable registration features that will see your nodes make connections, you should restrict this as outlined in the previous section.
When using the SSL security plugin all the nodes share a same SSL private and public key, all replies are signed using this key. It would not be impossible to add a per node certificate setup but I do not think this will add a significant level of security over what we have today.
When using the AES security plugin nodes can have their own sets of keys and registration data can be secured. Replies are encrypted using the clients key and so only the client who made the request can read the replies.
This applied to the service agent will allow different level of access to actions to different people. The caller id is based directly on the SSL Private Key in use and subject to validation on every node.
As with other aspects of mcollective authorization is tied closely with meta data like facts and classes so you can use these to structure your authorization as can be seen above.
You can provide your own authorization layers to fit your ogranizational needs, they can be specific to an agent or apply to the entire collective.
The RPC layer can keep detailed Auditing records of every request received, the audit log shows the - SSL signature or RSA verified - caller, what agent, action and any arguments that was sent for every request.
The audit layer is a plugin based system, we provide one that logs to a file on every node and there are community plugins that keeps a centralized log both in log files and in MongoDB NoSQL database.
Which to use depends on your usecase, obviously a centralized auditing system for thousands of nodes is very complex and will require a special plugin to be developed the community centralized audit log is ok for roughly 100 nodes or so.