Re: How is policy.31 created from modules under /usr/share/selinux

On 12/9/20 11:13 AM, Richard Haines wrote:
> On Wed, 2020-12-09 at 10:07 -0500, Steve Lawrence wrote:
>>
>> On 12/9/20 9:37 AM, Richard Haines wrote:
>>> On Wed, 2020-12-09 at 19:42 +0530, Ashish Mishra wrote:
>>>> Hi Richard ,
>>>>
>>>> Will check with the monolithic policy to check the behavior of
>>>> the
>>>> semodule as you suggested.
>>>>
>>>> Is there any similar approach / workaround for modular one?
>>>
>>> I've only had a quick look at code and I could see two ways to fix:
>>> 1) Modify the Rules.modular part of the make file to move or copy
>>> the
>>> policy and file contexts set of files over to $DESTDIR.
>>> 2) Modify semodule/semanage to handle $DESTDIR. I think this would
>>> be
>>> more difficult to fix as lots go on here.
>>>
>>
>> semodule does accept the -p option to change the root, so we could
>> feed
>> DESTDIR into that. For example, a minimally tested patch:
>>
>> diff --git a/Rules.modular b/Rules.modular
>> index d6224e95..64d953dc 100644
>> --- a/Rules.modular
>> +++ b/Rules.modular
>> @@ -55,8 +55,8 @@ load: $(instpkg) $(appfiles)
>>   # make sure two directories exist since they are not
>>   # created by semanage
>>          @echo "Loading configured modules."
>> -       @$(INSTALL) -d -m 0755 $(policypath) $(dir $(fcpath))
>> -       $(verbose) $(SEMODULE) -s $(NAME) -i $(modpkgdir)/$(notdir
>> $(base_pkg)) $(foreach mod,$(mod_pkgs),-i $(modpkgdir)/$(mod))
>> +       @$(INSTALL) -d -m 0755 $(policypath) $(dir $(fcpath))
>> $(DESTDIR)/var/lib/selinux
>> +       $(verbose) $(SEMODULE) -p $(DESTDIR)/ -s $(NAME) -i
>> $(modpkgdir)/$(notdir $(base_pkg)) $(foreach mod,$(mod_pkgs),-i
>> $(modpkgdir)/$(mod))
>>
>>   ########################################
>>   #
>>
>>
>> Note that we need to create $(DESTDIR)/var/lib/selinux since semanage
>> expects that to already exist.
>>
>> Though, I would suggest that maybe the "install" target should run
>> the
>> above semodule command with the --noreload option to install all
>> files
>> and build the policy binary but not actually load it into the kernel.
>> Then make load just becomes something like
>>
>>    semodule -p $(DESTDIR)/ --reload
>>
>> Makes a clear distinction between installing everything that's needed
>> vs actually loading the policy into the kernel. Happy to create a
>> patch
>> if that approach makes sense.
> 
> Thanks Steve, that worked for me, however I guess Chris needs to
> comment as the $(DESTDIR)/var/lib/selinux needs to be generated and
> maybe a clarification comment in the README. Also need comment
> regarding the use of --reload/--noreload.

To my knowledge, the history is that semodule was only intended to run on the 
target system.  If you wanted to precreate a policy you could 
semodule_link+semodule_expand like what is leveraged in the validate target.

I'd take a patch that changes the Makefile behavior but would like some real 
testing first.

-- 
Chris PeBenito