Page History

Scalable testing for insecure SAML signature validation

The SAML 2.0 protocol relies on XML signatures as the foundation of its security. A SAML assertion is signed with XMLDsig and the SP must properly validate this signature. If it does not, basically anyone in the world can trivially provide it with assertions thereby logging in as anyone, which also cannot be easily detected or even seen by the IdP. XMLDsig (and SAML) is notoriously complex and allows for many ways to create one or more signatures for any document. This makes that an implementation can easily fall victim to accepting not properly signed data - and even common implementations in our world like Shibboleth and SimpleSAMLphp have had issues here in the past. Besides these common products, which at least are periodically audited for such problems, a much larger risk is custom implementations that use different or even home grown libraries. Most of the times, the happy path is tested (does login work), but the unhappy path (do invalid assertions fail), not so much.

Given the paramount importance of signature validation, we should have a way to test whether SPs check signatures correctly. Although this can be done manually already, what's lacking is a scalable way that can test e.g. eduGAIN-like size of service providers (repeatedly) and for a large proportion of that set, determine if signatures are processed correctly. This requires to devise tests to fire off at these SPs and heuristics to determine automatically whether the tests passed or failed.

Some ideas of specific scenarios to test, all of which we've seen in real life to fail:

• Signature not checked at all, modified message accepted
• Modified message with signature rejected, but message without any signature accepted
• Multiple signatures on the same message/signature wrapping attacks
• Correctly signing a part of the message but unsigned part with attributes accepted.

Peter Brand (ACOnet)