Jump to: navigation, search

OSSN/OSSN-0026


Unrestricted write permission to config files can allow code execution

Summary

In numerous places throughout OpenStack projects, variables are read directly from configuration files and used to construct statements which are executed with the privileges of the OpenStack service. Since configuration files are trusted, the input is not checked or sanitized. If a malicious user is able to write to these files, they may be able to execute arbitrary code as the OpenStack service.

Affected Services / Software

Nova / All versions, Trove / Juno, possibly others

Discussion

Some OpenStack services rely on operating system commands to perform certain actions. In some cases these commands are created by appending input from configuration files to a specified command, and passing the complete command directly to the operating system shell to execute. For example:

 command='ls -al ' + config.DIRECTORY
 subprocess.Popen(command, shell=True)

In this case, if config.DIRECTORY is set to something benign like '/opt' the code behaves as expected. If, on the other hand, an attacker is able to set config.DIRECTORY to something malicious such as '/opt ; rm -rf /etc', the shell will execute both 'ls -al /opt' and 'rm -rf /etc'. When called with shell=True, the shell will blindly execute anything passed to it. Code with the potential for shell injection vulnerabilities has been identified in the above mentioned services and versions, but vulnerabilities are possible in other services as well.

Please see the links at the bottom for a couple of examples in Nova and Trove.

Recommended Actions

Ensure permissions for configuration files across all OpenStack services are set so that only the owner user can read/write to them. In cases where other processes or users may have write access to configuration files, ensure that all settings are sanitized and validated.

Additionally the principle of least privilege should always be observed - files should be protected with the most restrictive permissions possible. Other serious security issues, such as the exposure of plaintext credentials, can result from permissions which allow malicious users to view sensitive data (read access).

Contacts / References