Setting Up Virtual Security Zones in a Linux Cluster

The Distributed Security Infrastructure lets you create disjoint virtual security zones on a physical Linux cluster.

An increasing number of projects use Linux and other open-source software as basic building blocks to create clusters. Examples range from clusters that perform massive computations of visual effects for movies to clusters used as next-generation telecommunication servers.

More and more often, various issues, including economics, management and flexibility, require applications to run on the same physical cluster. An illustration of this situation in the telecom world is carrier-grade clustered servers shared among different operators. The operators share the global infrastructure of the cluster and provide different services to their clients but want to keep their binaries and data private. In such cases, cluster administrators do not have access to the source code for these applications, and security mechanisms cannot be enforced at the source code level. Hence, a security infrastructure is needed to ensure that a given application's resources cannot be tampered with or used by other entities on the cluster.

The Distributed Security Infrastructure (DSI) provides a solution for such a situation. It attempts to build a coherent security framework dedicated to carrier-grade Linux clusters by dividing a cluster into several virtual subclusters, guaranteeing controlled/restricted connections between them. Even though the project is only in its second year of design and development, we believe DSI is a useful tool for cluster administrators. This article presents how to use DSI to set virtual security zones inside a Linux cluster.

DSI Architecture and DSI Tools

In this section, we briefly introduce DSI's architecture. DSI is composed of one security server (SS) and multiple security managers (SMs), one per node (Figure 1). The SS centralizes management of the cluster: it gathers alarms and warnings sent by SMs and propagates a unique security policy over the cluster. On the other hand, SMs are responsible for enforcing security on their own nodes. Furthermore, messages are exchanged between the SS and SMs over encrypted and authenticated channels, using SSL/TLS over CORBA event channels.

Figure 1. The DSI Architecture

Security mechanisms in DSI are implemented at the process level to check the access privileges a process has to a resource. Each process is identified by its security context identifier (ScID) and the node identification on which it is running (SnID).

SnIDs are assigned using the DSI SetNodeID tool. All processes sharing the same security context are assigned the same ScID. ScIDs can be assigned automatically by the system according to DSP rules (see below), or they can be assigned specifically to a given binary using the DSI SetSID tool. This allows grouping of binaries according to their security contexts.

The DSP Configuration File

In DSI, writing a security policy for the cluster consists of granting or denying permissions to a given SnID and ScID pair. These rules are valid for the whole cluster. All rules are centralized in an XML file on the SS to ease management.

DSI provides a way to update and enforce transparently and automatically a unique homogeneous view of the whole cluster's security. Once the administrator modifies existing rules or adds a new rule to the distributed security policy (DSP), the DSP must be loaded on the SS using the dsiUpdatePolicy tool. Then, dsiUpdatePolicy checks the DSP file against our DSP XML schema (syntactical checks). If the DSP is validated, the SS propagates the new rules to all nodes of the cluster using the secure communication channels. Finally, each SM enforces the rules at kernel level calling the distributed security module (DSM, see Figure 2). DSM is based on the LSM kernel patch. Its detailed description is beyond the scope of this article; see the on-line Resources section for links to more information.

Figure 2. DSP Propagation inside the Cluster

Distributed Access Control

Controlling access to local resources is rather simple: the DSM module retrieves the local ScID and SnID of the requesting process and checks corresponding permissions in the security rules. Actually, the originality of DSI lies in distributed access control. Currently, only socket communications are implemented. To illustrate this, we detail the access control mechanisms when a process tries to access a resource located on another node (Figure 3):

  • The access request first is intercepted by the local DSM, which checks that the process has the privilege to call locally the socket-related systems calls.

  • Then, the ScID and SnID of the requesting process are added by DSM to each IP packet sent to the remote node.

  • On the receiving node, the remote DSM uses the ScID and SnID of the requesting process, extracted from the IP packet, to check its permission to communicate with both the target socket and the process to which the target socket belongs.

  • Finally, the remote DSM locally checks that the process to which the target socket belongs may receive information from the requesting process.

Figure 3. Secure Remote Access Control