Incorrect signature verification


An incorrect check of return values in the signature validation utilities allows an attacker to get invalid signatures accepted as valid by forcing an error during validation.


The SAML2\Utils class offers a set of methods that allow to verify an XML digital signature against a given key. In particular, the validateSignature() method receives a signature and a key to verify it, and throws an exception in case there is any error, either caused by incorrect input or an invalid signature. This method uses the verify() method from the RobRichards\XMLSecDSig class to verify the signature with the given key, which in turn will end up calling openssl_verify() depending on the signature algorithm used.

The openssl_verify() function returns 1 when the signature was successfully verified, 0 if it failed to verify with the given key, and -1 in case an error occurs. PHP allows translating numerical values to boolean implicitly, with the following correspondences:

This means that an implicit conversion to boolean of the values returned by openssl_verify() will convert an error state, signaled by the value -1, to a successful verification of the signature (represented by the boolean true).

The aforementioned validateSignature() method was performing an implicit conversion to boolean of the values returned by the verify() method, which subsequently will return the same output as openssl_verify() under most circumstances. This means an error during signature verification is treated as a successful verification by the method.

Affected versions

All SimpleSAMLphp versions prior to 1.14.10.

All software using the simplesamlphp/saml2 library with versions prior to 1.9.1, 1.10.3 and 2.3.3.


Upon successful exploitation, an invalid signature would be regarded as valid by an affected version of the software. This allows attackers to modify or manually craft SAML response messages and, by triggering a signature validation error in the affected party, get those messages accepted as valid and coming from a trusted entity. In practice, this means full capabilities to impersonate any individual at a given service provider.

The issue can be exploited to get other invalid messages accepted as valid, though the security implications there are minor.

In order to exploit the issue, an incorrect context must be fed to the signature validation routines, or an exceptional error must be triggered. So far, the following cases have been identified:

SimpleSAMLphp does not support DSA signatures or keys. Therefore, it is not possible for an attacker to feed an incorrect context by sending a signature with an incorrect algorithm. Upon reception of a DSA-SHA1 signature, SimpleSAMLphp will refuse to perform the validation due to the algorithm not being supported. On the other hand, if an attacker manages to trick a service provider operator to change the public key associated to a certain IdP to a DSA key, signatures made with any combination of the RSA algorithm will be accepted, regardless of whether they are valid or not. This means some serious misconfiguration or social engineering is needed in this case for a successful attack.

Regarding memory exhaustion, it is in theory possible to attack a service provider causing the consumption of all available memory while a message with an invalid signature is being validated. However, memory exhaustion must happen only during signature validation and not immediately before or after. This means exploitation of this case is extremely difficult due to the small time window available for the attacker and the precise control that is needed over the service provider.

All in all, the consequences of this issue are critical, so even though we consider it difficult to exploit, and considering that other ways to trigger failures in signature validation could be possible but so far unidentified, we recommend updating the affected software as soon as possible.


Upgrade to the latest version.

For software using the simplesamlphp/saml2 library, update the dependency to the latest version available.


This security issue was discovered and reported on November 29, 2016 by artem-diverta.