Deep Security 11.1 has reached end of support. Use the version selector (above) to see more recent versions of the Help Center.

SSL implementation and credential provisioning

The Deep Security Agent may initiate communication to Deep Security Manager or it may be contacted by the manager if the computer object is set to operate in bi-directional mode. Deep Security Manager treats all connections to agents in a similar way. If the agent has not been activated, a limited set of interactions are possible. If the agent has been activated (either by an administrator or via the agent-initiated activation feature), the full set of interactions are enabled. The Deep Security Manager acts as an HTTP client in all cases, regardless of whether it was the client when forming the TCP connection. Agents cannot ask for data or initiate operations themselves. The manager requests information such as events and status, invokes operations, or pushes configuration to the agent. This security domain is highly controlled to ensure that agents have no access to Deep Security Manager or its host.

Both agent and manager use two different security contexts to establish the secure channel for HTTP requests:

Before activation, the agent accepts the bootstrap certificate to form the SSL or TLS channel.
After authentication, mutual authentication is required to initiate the connection. For mutual authentication, the manager's certificate is sent to the agent and the agent's certificate is sent to the manager. The agent validates that the certificates come from the same certificate authority (which is the Deep Security Manager) before privileged access is granted.

Once the secure channel is established, the agent acts as the server for the HTTP communication. It has limited access to the manager and can only respond to requests. The secure channel provides authentication, confidentiality through encryption, and integrity. The use of mutual authentication protects against man-in-the-middle (MiTM) attacks where the SSL communication channel is proxied through a malicious third party. Within the stream, the inner content uses GZIP and the configuration is further encrypted using PKCS #7.