So what exactly is it?
IT-security has become more and more important over the last decades. While at first security was frequently treated as a necessary evil, nowadays it has matured into a separate area of expertise. There can be a lot of confusion about what exactly encompasses security and identity management; everyone has a different view on it. When discussing these topics first agree upon its scope before you delve into it. In this blog it will be divided into:
- Identity management
- Authentication; including Single Sign-On (SSO)
- Logging and monitoring
- “Hard” security; more technical security including confidentiality and integrity of data, usage of firewalls, IDS/IPS products, and so on.
The first three together encompass Identity and Access Management (IAM). “Soft” security like creating security awareness, training employees, applying physical security to buildings and IT-assets, and availability of IT-assets (together with confidentiality and integrity forming the so-called CIA-triad) are out of scope for this blog.
Security and SOA
When compared to traditional software development the important question is not whether security in an SOA-environment is important, but if it is any different -and should therefore be designed differently? The answer to both questions is yes.
To understand why security should be handled differently we need to understand the characteristics of SOA that are key to the security aspect compared to those of traditional software development:
- Next to human-machine interaction there is more machine-machine interaction. This means there is a greater need for automated security mechanisms for purposes of authentication, authorisation, encryption, and so on.
- A SOA-environment generally contains more intermediary stations such as ESB’s and other middleware components. There are more locations for users and administrators to view -possibly confidential- message contents such as credit card information. In this case transport security alone is not enough.
- How can you manage and control various (external) clients that want to access data and/or services if systems are loosely-coupled? E.g. not every client must be allowed to invoke a banking service.
- SOA results in more Straight-Through-Processing (STP) meaning processes are more frequently executed in an entirely automated fashion without human interference. Good security is key since consequences of possible security breaches could be detected later on. Also, the consequences can be graver due to the possible large amount of process instances.
- Services are invoked by both internal and external consumers. A service’s security level is usually determined by its owner. In case of external services, security will be largely determined and enforced outside an organization’s own span-of-control. The level of security determines the consumers trust: “What happens with my data if a service is not secured?”, “Can I trust a service’s result?”, and so on.
These differences clearly impact the way security should be designed within an SOA-environment. It furthermore warrants the need for an integrated and holistic approach on security in an SOA-environment. Use a layered approach to security as for example promoted by the defense-in-depth strategy.
For a number of reasons it is a good design-principle to externalize identity management and security; even more so in an SOA-environment that frequently consists of heterogeneous infrastructure. Every service having its own IAM and security design and implementation leads to a suboptimal solution, more overhead, and greater chance for security breaches. If security is part of the infrastructure’s components -for instance intertwined in an ESB product- different products will most likely also support different security standards and protocols. E.g. an application server might support SAML version 1.1, the WS-Security Username Token profile, transport security using HTTPS, and LDAPS while the ESB product supports SAML version 2.1, the WS-Security X.509 Token Profile, message security using XML DSIG, and LDAP rather than LDAPS. This is worsened in case external infrastructure supports yet another subset of standards and protocols. This can cause poor interoperability. Use a separate -specialized- component for security instead. This promotes both reuse of better security throughout your SOA-environment and promotes separation of concerns.
The agents and gateway patterns are very well suited to externalize security. Use gateways for appliance of common security policies and agents for more service-specific security policies.
Define a limited set of security classifications; for example based on the CIA-triad (confidentially, integrity, and availability) ranging from e.g. “public” to “highly classified”. Determine a minimum set of security measures per classification level. For each new service determine its classification levels; this is usually the responsibility of the service owner. Make classification levels part of your service repository and governance processes. This results in more understandable security regulations, gives better insight in the current and future security of your environment, better reuse of existing security policies, and prevents reinventing the wheel when establishing security for new services. Most important it results in just the right amount of security to be applied; thereby saving money (strive for the lowest possible classification levels without endangering security) while applying (just) enough security.
Transport versus message security
There are roughly two types of security for message invocation: transport and message security. Transport security secures a message only during transport between service consumer and service provider using the transport layer; e.g. using HTTP over SSL or TLS (HTTPS). That means messages are not protected in intermediary components such as an ESB and not protected directly after being received by the endpoint. Message security secures the message itself, mainly through encryption of the payload using for example public and private keys. Since message security can provide security in the scope you want to -so also in intermediaries and after the message has been received- it is generally preferable over transport security. Both transport and message security can be used for authentication (e.g. signature based on certificates), integrity (e.g. digest), and confidentiality (e.g. encryption).
Maybe trivial but very much important: use standards to promote interoperability. This includes the usage of security standards such as LDAP(S), HTTPS, SAML, XML DSIG, WS-Security (WSS), and other WS-* standards. Using standards results in secured services being reused by (both internal and external) heterogeneous infrastructures. Next to technical standards there are also a number of security reference architectures and principles and guidelines you can leverage.
Before we wrap up some best-practices per area.
Use a centralized identity management repository. This avoids duplicate user management and possible inconsistencies. Divide users into different identity types if needed -such as employees, customers, suppliers, and so on since different rules and administration may apply to each category. Be careful in allowing external IT-assets and organizations direct access to your identity management solution. Consider identity provisioning in such cases as external hosting to minimize security risks.
Usually you want a service provider to authenticate the original service consumer (user identity) and not some intermediary component such as an ESB. Implement identity propagation of tokens, username/password, etc. so the service provider authenticates and authorizes the identity of the original user that invoked the service. That implies that all intermediary components between service consumer and provider need to be able to transport identity tokens -and possibly transform these from one format to another (e.g. from SSO token into username/password).
Especially in case of authenticating and authorizing external organizations consider the trade off between using specific identities (Mr. X or Mrs. Y) versus more general identities (organization Z). Specific identities result in better traceability and can provide for more fine-grained access control while more general identities can result in less administration: the number of different identities to manage and synchronize decreases dramatically.
Avoid generic identities such as “consultant” and “trainee” all together.
Define a limited set of authentication levels and differentiate on information (password), possession (token, physical key, text message to a phone), and attribute (voice, fingerprint) as mechanisms. E.g. “basic-level” authentication requiring information, “middle-level” authentication requiring information and possession, and “high-level” authentication requiring attribute or possession together with a check of ownership.
Most organizations promote SSO to improve user-friendliness and provide for better user-experience. Determine however if you want SSO for your most classified IT-assets. SSO can provide access to a multitude of IT-assets due to a security breach in only one of the IT-assets. A best-practice is to grant access to IT-assets based on authentication level; if you authenticated using basic authentication, SSO will only grant you access to IT-assets requiring the same or a lower authentication level; not to IT-assets requiring “high-level” authentication.
The SSO-provider needs to be verified and trusted before you can hand over authentication to that provider.
Don’t tie rights to IT-assets directly to user identities to avoid high maintenance costs, inflexibility, and lock-in of users. A good design-principle is to use a form of Role-Based Access Control (RBAC) to decouple authorization. Use attributes such as organizational units and/or job titles that do not change frequently over time as intermediary layers in the authorization model. Assign rights in IT-assets to entities in this layer (e.g. organization unit and/or job title) and vice versa assign user identities to these intermediary layer(s). Design the authorization model per identity type (customer, employee, supplier, etc.).
Base authorization on the work/function someone or some organization needs to do; no more, no less. Avoid “super-users”; usually management and/or IT-staff that have gathered much more privileges over time than they’re entitled to. Increase security by assigning more than one role to the various steps in sensitive processes thereby preventing one user to be able to execute the process entirely.
Logging and monitoring
Functionality and processes in an SOA are spread over different loosely-coupled components. Some logging and monitoring needs to be executed on a higher level than on that of an elementary service; but rather on composite service or process level. This gives rise to the need for a central logging and monitoring component that is able to combine and correlate decentral logs and enables monitoring on process-level. The Wire Tap pattern can be used to publish logs, sensors, and other types of messages from services and middleware to the central monitoring component. Notifications can be managed and implemented separately of the logging and notifications can be published by this central monitoring component. Note that this requires synchronization of date and times of the several components that are managed to enable correct correlation. Determine for every service if it is allowed to continue operation in case the central monitoring component fails. Is it e.g. allowed from a security point-of-view to use decentral -localized- logging and monitoring in case the central monitoring component is down?
A best-practice is to divide security in a number of layers. Chart possible vulnerabilities, threats, and corresponding principles and guidelines to counteract them. This approach results in a more effective and efficient security. Examples of such layers are: network security, platform security, application security, integrity & confidentiality, content security, and mobile security. Examples of principles and guidelines are applying compartioning (network security), to have a central list of allowed and non-allowed file extensions for inbound and outbound traffic (content security), and the use of hardening (platform and application security).
In case of Oracle’s SOA product stack (SOA Suite 11g) security is externalized from almost all products and can be applied using policies. These policies can be configured in a management console and reused by processes and services that are packaged and deployed as SCA composites and components. These policies are based on standards such as WS-Security. Oracle Service Bus (OSB) contains security functionality though. As stated in OSB’s SOD: “The ability to attach, detach, author and monitor policies in a central fashion will be extended to the Oracle Service Bus (as it is has been extended to all other components in the SOA Suite 11g).” In any case you can already secure OSB projects using OWSM.