Enterprises often frustrate developers. Why do Enterprises always seem so behind when it comes to the very latest technology? In particular, a trend we are seeing is the continued struggle to marry Enterprise authentication with the burgeoning world of REST APIs. Developers want to use REST, but Enterprises need enterprise grade API security.
We think this problem will only worsen as Enterprises continue their rapid adoption of APIs. It seems clear that SOAP, while capable of Enterprise grade authentication through X.509 and SAML, will be left behind as the “Skinny jeans Facebook generation” puts the final nail in SOAP’s coffin.
Among our own customers and the stories we’ve heard, Enterprises are left with a dilemma with four horns concerning the protection of REST APIs:
(a) Use mutual authentication with client-side SSL
(b) Use HTTP authentication (password or digest) with server-side SSL
(c) Use OAuth, either 3-legged or 2-legged
(d) Use a de-facto standard or “roll your own”
Each option has challenges and benefits. In particular, the challenges balance time to market and time to value (which I term developer friction), security, and manageability. The success of an API is directly related to it’s perceived ease of access and the amount of friction involved in using it. An Enterprise will get more value out of an API that is actually used versus one that lies dormant. At the same time, however, as APIs become a new tunnel into the Enterprise, security and manageability cannot be sacrificed for the sake of adoption.
Current Options and Challenges
Let’s consider the options and “score” each of them. In option (a), most developers would rather shoot themselves than deal with client-side X.509 certificates, especially after experiencing the apparent ease of use of OAuth as evidenced by SaaS providers such as Salesforce.com and social platforms such as Facebook and Twitter. To compound this, the issue of key rotation and certificate management is a weighty burden to deal with from an Enterprise perspective. Echoes are heard from the grave, “But this year, will be the year of PKI!” Now, the “Year of PKI” conjures up not a picture of a secure Enterprise, but an enterprise fraught with wasteland scenes of Armageddon.
SSL with Mutual Auth Score:
- Security: High
- Friction: Extremely High
- Management Burden: Extremely High
- Developer Attitude: I hate my life
- Result: A secure API that nobody uses has lowered value to the Enterprise
In option (b), HTTP Authentication with server-side SSL, we’ve overloaded an interactive web-authentication mechanism developed in 1996 and tried to marry it to the API economy of 2012.
Well, you say…. At least it is a standard – and we console ourselves by the fact that it is a protected channel. The issue here of course is the proliferation and management of passwords.
Security is reduced because we’ve coded in a username and password and we’re probably using the same username and password for multiple applications and not telling management about it. Security is low when the password is in the clear and somewhere around medium-low when digests are used, depending on your perspective on rainbow tables. Friction is high because eventually I’ll need to (or should) rotate that password if I have even a rudimentary password policy
HTTP Basic Auth with SSL Score:
- Security: Medium-Low
- Friction: High
- Management Burden: High
- Developer Attitude: Really? We’re doing this?
- Result: An API with a “hackish” authentication method and suspect security.
In option (c), OAuth, developers will cheer but problems remain. First and foremost, traditional 3-legged OAuth is an authorization protocol that puts permission control in the hands of a user, who ostensibly owns an asset they are expressing authorization for.
In the traditional Enterprise context, however, control should be given to administrators, not users. Most users don’t own Enterprise assets, the Enterprise owns their own assets. With X.509 certificate authentication models, administrators can revoke permissions easily by revoking a certification while OAuth delegates permissions to users. This is why OAuth works well for social websites – content is owned by users.
One solution to this is to use 2-legged OAuth or some organic variation such as xAuth, which is notionally similar to SAML in that it allows the exchange of a username/password for an access token. Note: If xAuth is still OAuth why is it called something different? Answer: It is different (?)
2-legged OAuth has advantages over the previous option in that it stops the proliferation of passwords, which is a good thing, but the failure of the OAuth 2.0 specification to carefully define what 2-legged OAuth entails is a bit worrisome for both security and interoperability.
Third, depending on the specific OAuth data flow, such as the authorization code flow, implicit grant flow, or client credential flow, these all have varying levels of security and friction associated with them. The most secure OAuth data flows involving a confidential client demand strong client authentication….. …with a public/private key pair (ssshhh!!).
Finally, as history has taught us, never place all your bets in the security of a protocol. Many of us were shocked at the late 2009 news of a critical man-in-the-middle attack on SSL. If we count the Netscape days to the time of the vulnerability, the protocol was battle tested in industry for 14 years before the vulnerability was found. Some would argue that OAuth just hasn’t been around the block long enough.
- Security: Medium (or Low, depending on your view of history)
- Friction: Medium-Low
- Management Burden: Medium
- Developer Attitude: I love my job, it’s like Facebook
- Result: An API that uses a rapidly emerging protocol and avoids the use of passwords with low developer friction
In option (d), roll your own, we’ve seen a number of solutions that involve API keys or shared secrets along with authenticators like HMAC-SHA1. Amazon web services (AWS) is a famous example of this approach, and many of our own customers have copied this approach as a best practice.
These solutions have the advantage over HTTP Basic Authentication in that they don’t shoe-horn themselves into an outdated standard designed for another purpose. The flip-side is that these aren’t official standards. The idea is to use an HTTP header to store the credentials in a bespoke way defined by the Enterprise. The security model here is essentially the same as username and password, except for the fact that we’re calling it an API key to make it sound like it’s not a password. Management burden is increased here because interoperability outside the organization is reduced.
“Roll Your Own” Score:
- Security: Medium-Low
- Friction: Medium
- Management Burden: Medium-High
- Developer Attitude: Eh, It’s not OAuth, but at least I don’t have to deal with X.509 certificates
- Result: A reasonable, albeit non-standard solution to API authentication. Friction is reduced but password or shared secret proliferation remains a problem
Appease your Developers
One approach to solving the API access mechanism dilemma is to reject the dilemma and have it all. This can be done with a front-end or security intermediary that, much like a diligent mediator can make both sides happy. In this model, a front-end proxy is used on top of existing APIs, essentially retrofitting them for OAuth without having to get into the ‘weeds’ of actually modifying enterprise APIs. Everything is done on the wire at network layer 4 and above.
To take one example, if you are an Enterprise with REST APIs that use mutual SSL authentication, you don’ t have to enforce this fact on your developers. Instead, have a gateway handle it for you. You can give your developers a choice – they can fall on the sword of X.509 if they’d like, or use OAuth.
Conceptually, it looks like this:
In the previous diagram the gateway handles the OAuth dance for incoming clients and maps the identity to a username from Active Directory. Then, the gateway initiates a mutual SSL connection to the Enterprise REST API and sends the username of the original requestor over this secure channel in an HTTP header. Simple.
Developers are happy because the Enterprise is now “living in the 21st century” and the enterprise application developers are happy because they don’t need to retrofit something that already works, and the security architects are happy because they can manage it all on a security gateway with a few clicks and drop downs. Party down!
This gateway or proxy model is a superb answer for solving some of these real challenges. After all, Enterprises can’t wait for the standards to catch up with the new generation of developers.