service and system-config-services restart daemons in incorrect type under mls policy

[Daniel J Walsh]

Stephen Smalley wrote:
> On Thu, 2007-04-26 at 20:00 -0500, Joe Nall wrote:
>   
>> On Apr 26, 2007, at 3:18 PM, Stephen Smalley wrote:
>>
>>     
>>> On Thu, 2007-04-26 at 14:55 -0500, Joe Nall wrote:
>>>       
>>>> I'm running an mls/permissive policy on FC6 and service and system-
>>>> config-services start daemons in the user's selinux context rather
>>>> than those in /etc/selinux/mls/contexts/initrc_context. Since our
>>>> policies use init_daemon_domain to establish domain transitions, they
>>>> are not transitioning into the correct domain on user initiated (re)
>>>> starts.
>>>>
>>>> "run_init service <service> restart" - works, but leaves us in a
>>>> situation where documentation doesn't match experience. What is the
>>>> right approach to getting the transitions to work properly? Patch
>>>> service and friends? Write a more generic transition?
>>>>         
>>> That should be governed by the DIRECT_INITRC= setting in the refpolicy
>>> build.conf (or as overridden on the make command line in the .spec  
>>> file
>>> for building the policy).  DIRECT_INITRC=y (as in -targeted) turns on
>>> direct role transitions and domain transitions from  
>>> sysadm_r:sysadm_t to
>>> system_r:initrc_t and/or system_r:<daemon domain>, although we  
>>> can't yet
>>> automatically transition the user identity field.
>>>
>>> If you want the DIRECT_INITRC=n situation, then yes, you need to
>>> integrate run_init or similar functionality into the init script  
>>> and/or
>>> service script infrastructure, as they have done in Hardened Gentoo.
>>>       
>> Why does run_init prompt for a root password rather than perform a  
>> role check?
>>     
>
> The role authorization is handled transparently by policy - if you
> weren't in an authorized role/domain, then you couldn't use run_init to
> transition to system_r:initrc_t anyway.  Same as with newrole.  The
> re-authentication stage is purely a (weak) countermeasure against
> invocation by malicious code without user consent - if we had a trusted
> path mechanism in Linux, we'd use that instead.
>
>   
Most people are adding pam_rootok to /etc/pam.d/run_init so that it will 
work for sysadm_t.