On 05/23/12 13:46, Daniel J Walsh wrote: > On 05/23/2012 01:32 PM, Christopher J. PeBenito wrote: >> On 05/23/12 11:46, Daniel J Walsh wrote: >>> On 05/21/2012 04:58 PM, Sven Vermeulen wrote: >>>> Hi guys, >>> >>>> It looks like the current stable sepolgen release has requirements >>>> towards an unofficial (well, fedora/rhel only) patch on setools. With >>>> the current stable setools, it gives the following error when trying to >>>> use audit2allow on a denial that contains write & open: >>> >>>> Traceback (most recent call last): File "/usr/bin/audit2allow-2.7", >>>> line 354, in <module> app.main() File "/usr/bin/audit2allow-2.7", line >>>> 345, in main self.__output() File "/usr/bin/audit2allow-2.7", line 315, >>>> in __output g.add_access(self.__avs) File >>>> "/usr/lib64/python2.7/site-packages/sepolgen/policygen.py", line 211, >>>> in add_access self.__add_allow_rules(raw_allow) File >>>> "/usr/lib64/python2.7/site-packages/sepolgen/policygen.py", line 179, >>>> in __add_allow_rules self.domains = seinfo(ATTRIBUTE, >>>> name="domain")[0]["types"] NameError: global name 'seinfo' is not >>>> defined >>> >>>> The patch that RedHat (and Fedora) provides fixes this in Python 2 >>>> systems, but doesn't work in Python 3 (because Python 3 has a >>>> different setup for Extension-based modules). I have a locally-tested >>>> patch on that, but I'm not sure this is a good way to go forward. >>> >>>> Perhaps it would be wise to remove the dependency towards the setools >>>> binding and instead include the necessary code in the userspace >>>> libraries themselves? policygen.py doesn't require the entire set of >>>> querying that seinfo provides... >>> >>>> The patch that is suggested by RedHat/Fedora doesn't follow the same >>>> structure as the other bindings do (like libqpol/libapol) in setools >>>> too. >>> >>>> Wkr, Sven Vermeulen >>> >>> Well I am not sure if anyone has ever used the setools python binaries >>> other then the setools/sesearch and seinfo bindings. >>> >>> I would suggest we drop the general python bindings or deemphasize them >>> and work on improving the seinfo/sesearch bindings. > >> I don't have a problem with a simpler api, e.g. a single function for rule >> searching, rather than the multiple calls to set up a query, but the >> current implementation in Fedora isn't acceptable to upstream setools. >> Perhaps what should be done is to add a basic query api to the SELinux >> userspace upstream, so that you can create all of these tools. Libqpol in >> setools tries to do this, but its implementation wouldn't be acceptable >> upstream. Then the extra dependency of sepolgen on setools could be >> broken. > > > What is not acceptable, the entire idea of setools.sesearch and setools.seinfo > or something specific about the design? Or do you want setools to just be low > level calls and want us to build a new package that actualy does something > useful with python? > > IE Do you want us to just create a new tool called seanalyze or something that > includes the python bindings that we find useful. Well its up to SELinux upstream to say if they'd accept a libsequery or libseanalyze or augment libsepol. In the absence of that, the issue with what you have is the implementation. The preference would be to add the new single function query API to libapol either in the C library sources or in pure python using the existing swig wrappers. It would also have to be a complete API (i.e. apol could be converted to use it), which is more than sesearch and seinfo support. -- Chris PeBenito Tresys Technology, LLC www.tresys.com | oss.tresys.com -- This message was distributed to subscribers of the selinux mailing list. If you no longer wish to subscribe, send mail to majordomo@xxxxxxxxxxxxx with the words "unsubscribe selinux" without quotes as the message.
