Cached keystone tokens may be accepted after revocation
Keystone auth_token middleware token and revocation list caching is used to reduce the load on the keystone service. The default token cache time is set to 300 seconds and the default token revocation list cache time is set to 10 seconds. This creates a misleading expectation that revoked tokens will not be accepted more than 10 seconds after revocation, however the maximum validity of a cached token must be assumed to be the cache duration. System owners should make a risk based decision to balance token lifespan with performance requirements and if the use of revoked tokens is an unacceptable risk then caching should be disabled.
Affected Services / Software
OpenStack Services that use Keystone middleware: Juno, Kilo, Liberty
There are multiple options for configuring token caching in the keystone auth_token middleware. These options include token_cache_time, revocation_cache_time and check_revocations_for_cached, with each option affecting the different stages of token caching and revocation. Depending on the configuration the previously mentioned options, an attacker could use a compromised token for up to token_cache_time seconds before the token becomes disabled. To mitigate this vulnerability, a change was issued in Juno where the default Token Revocation List (TRL) cache time was reduced to 10 seconds and the check_revocations_for_cached option was added. The addition of a token to a TRL does not guarantee that cached tokens will be rejected considering the operational nature of token caching. For instance, if the check_revocations_for_cached is disabled then tokens are valid after caching token_cache_time or the designated expiration given to the token. Otherwise (if check_revocations_for_cached is enabled) then tokens are rejected after the revocation_cache_time.
System owners should weigh the risk of an attacker using a revoked token versus the performance implications of reducing the token cache time.
Review the implications of the default 300 second token cache time and any risks associated with the use of revoked tokens for up to that cache time. If this is unacceptable, reduce the cache time to reduce the attack window or disable token caching entirely.
Contacts / References
- Author: Shellee Arnold, HPE
- This OSSN : https://wiki.openstack.org/wiki/OSSN/OSSN-0056
- Original LaunchPad Bug : https://bugs.launchpad.net/python-keystoneclient/+bug/1287301
- OpenStack Security ML : email@example.com
- OpenStack Security Group : https://launchpad.net/~openstack-ossg