Well, if the security considerations that applied to traditional application architectures applied to SOAs, then the aforementioned controls would be sufficient. But unfortunately service-oriented solutions are different. Unlike the interoperability challenges that stunted the growth of Common Object Request Broker Architecture (CORBA), today’s SOAs are interoperable, relying often on Web services based on standards such as XML for encapsulation and HTTP/HTTPS for transport. It is in fact this interoperability and interdependence between entities that has influenced a new frontier of security threats.
Unlike conventional application architecture, service-oriented solutions that expose Web services to external users and business partners are more susceptible to unauthorized access. Traditionally application architectures allowed limited external access with adequate security controls in place to do so. Message-level security was unnecessary because all processing was carried out internally, within the application. Instead the emphasis of security was on how the message was transported. The contents of the messages were assumed to be valid since they were coming from one trusted source within the organization.
In a service-oriented infrastructure, a solution invites all HTTP/HTTPS traffic through its firewalls, and does little to differentiate between a malicious or valid message. A typical organization’s border and internal firewalls are packet filtering and “allow” all HTTP/HTTPS traffic back to the application server from an uncontrolled zone to a DMZ to a secure zone without inspection. Thus, the service-oriented solution has introduced a hole in the firewall.
Understanding the Risks of Exposure
To offer a rudimentary understanding of the new threats, we’ll consider them in the context of a sample scenario with “Big Bank”. Suppose Big Bank is developing an application to issue checks and has implemented a Web service called “cut_check”. The cut_check service may be accessed by other Big Bank organizations and business partners on disparate networks across the world. Following best practices, the application will have its Web server in the DMZ and application server (containing business logic) in a separate security zone, away from the DMZ, protected by packet filtering.
Conventional application security architecture would not allow direct access to the cut_check service running on the application server, but instead would only accept requests through a proxy, such as an HTTP Web server in a DMZ. The cut_check service would process each transaction at the application server and would then present the results upon completion of each transaction to the Web server to display back to the user.
As illustrated in the top of portion of Figure 1, the user makes a request to the application, the application makes a request to the service, and the response is provided back to the user. However, as shown on the bottom portion of Figure 1, a service-oriented solution introduces new methods of invocation. The service may be directly exposed to the user, invoked by another Web service, or it may invoke a service directly. The service-to-service invocation can result in a chaining or composition of services to perform a specific operation.
For example, at Big Bank the clerk may issue cut_check which may rely on another service such as calc_taxes (that determines the amount of tax to be drawn prior to issuance). Consequently, the service-to-service use cases also require security controls similar to user-to-service scenarios.
Figure 1: The shifting architectural paradigms introduced by SOA.
Furthermore, SOA-based applications are incubating a new breed of security threats to Web based applications. There are numerous attack scenarios around Web services already being exploited and the list keeps growing. The types of attacks include, but are not limited to: denial of service, replay, message alteration, breach of confidentiality, man in the middle, and spoofing attacks.
In particular, XML attacks tend to be the most pervasive through well-formed SOAP messages embedded in HTTP/HTTPS protocols. Overall, there are four broad classifications of threats associated with XML, as summarized in Table 1.
Table 1 - Common types of XML attacks.
The astute architect might suggest that transport layer security mechanisms such as Secure Sockets Layer (SSL) or Transport Layer Security (TLS) addresses the XML threat. However, security at the transport layer of the OSI model is just not enough. It doesn’t address the context of SOA scenarios because it doesn’t protect against malicious content included in the message. It simply encrypts the malicious code embedded in the XML message written by a, perhaps, authorized user with access the service. Once the packet is opened by the application it is too late.
Thus network and host system controls, identity and access management controls, and data encryption controls during storage and transit must be supplemented with additional controls to protect the message payload. This additional control is required given that the application is now open to a broader audience and is accessed with standard protocols instead of proprietary extensions.
SOA Security Design Considerations
A layered security architecture provides the best defense for a service-oriented architecture. The most effective architectures certainly use strong network and host system controls, identity and access management controls, and data encryption during transit and storage, but also rely on Web service security. Two fundamental design considerations used to secure Web services focus on packet and message-level security.
Since SOAP messages containing XML payloads traverse across runtime service activities using the HTTP/HTTPS protocols, packet filtering firewalls and routers do not provide adequate security controls. Packet filtering devices will not be able to differentiate authorized well-formed XML messages with malicious messages. However, XML firewalls can bridge this gap.
First, XML firewalls provide packet level inspection of all inbound and outbound traffic to the back-end web application server. For inbound traffic, the device tears open packets to scan its contents against a catalog of threats before it forwards it on to the application for processing. For outbound traffic, the device may apply response filters to inspect packets for sensitive information inadvertently included in the response. Response filters can serve as a critical component to prevent confidential information from leaving an organization through Web services.
Next, the XML firewall must be strategically placed on the network. Depending on the application requirements, it is most typically placed in a DMZ security zone in front of the web application server, as illustrated in Figure 2. Coupled with the traditional packet filtering firewalls, placement in the DMZ offers the ideal position for the XML firewall to protect the business logic harvested in the application server. As the services become exposed to external networks beyond the host organization’s control, additional XML firewalls may be deployed to add more depth of security across the network.
Figure 2: A typical XML firewall architecture.
Furthermore, many of the major application servers, such as IBM WebSphere and BEA WebLogic, offer embedded packet inspection services. These value-add features may be appropriate depending on the amount of messages the server processes and the performance requirements of the application. Considering the processing overhead of packet inspection, the use of these embedded features may introduce significant overhead as the application scales. The industry best practice is to offload this security processing into a hardware-based XML firewall, such as IBM Data Power, Intel Sarvega, or Cisco Reactivity. Using a hardware device also adds an additional layer of protection into the overall security architecture.
In addition to packet inspection from an XML firewall, Web services may also be protected using message-level security. The use of standards offers a consistent framework for developers to implement security at the message level, especially since the information security community has helped mature Web services security standards over the past few years. Figure 3 describes the security mechanisms available in reference to layers of the Open Systems Interconnection (OSI) model.
Figure 3: Security protocols categorized by OSI layer.
A key feature of the WS-Security standard is that it makes use of existing technologies instead of inventing new ones. Specifically, WS-Security calls upon the following proven security techniques:
|•||XML Digital Signatures - Used to ensure that a message is not altered while in transit.|
|•||XML Encryption - Used to encrypt messages to make them unreadable except by those possessing the proper keys.|
|•||Authentication Tokens - A mechanism to restrict access to Web services to only those users that are authorized with supported authentication types: usernames and passwords, X.509 digital certificates, Security Assertion Markup Language (SAML).|
For instance, the WS-Policy specification offers security governance for encryption but it does not specify all of the key exchange properties necessary to make this work in the real world. The exchanging parties must still determine and agree upon on a key size, cipher, and algorithm. Also, Web security standards provide little support beyond SOAP and HTTP/HTTPS. Some solutions may still require proprietary specifications and technologies to fully protect their perimeters.
Next, if the technology infrastructure is not sized correctly, the advantages provided by message-level security may result in detrimental performance. Despite these inconsistencies, an architecture based upon standards is still far superior than relying on proprietary methods.
The proliferation of Web services has pushed the business case for SOA, for they have enabled unparalleled interoperability among a heterogeneous set of enterprise applications. Along with the increased integration, there has been a paradigm shift on how to secure an application given the new frontier of attacks on interdependent enterprise solutions. The new generation of attacks categorically include threats based upon XML denial of service, unauthorized access, data integrity, data confidentiality, and system compromise.
As the proverbial saying goes, “Security is only as good as its weakest link.” The external exposure of Web services has warranted inspection of the transaction payload and the use of message-level security. Thus conventional security controls must be amended to include Web service security mechanisms for service-oriented solution designs. The inclusion of XML firewalls for packet inspection and message-level security are therefore critical for the protection of SOA.
Industry standards are available and continue to evolve, but do not provide a comprehensive solution. The onus is on the security architect to use the standards as a baseline and devise appropriate security controls applicable to the specific implementation of the service-oriented solution.