On Fri, 2010-01-08 at 10:07 -0500, James Carter wrote: > On Fri, 2010-01-08 at 09:30 -0500, Stephen Smalley wrote: > > On Wed, 2009-12-23 at 18:25 -0500, Caleb Case wrote: > > > This patch moves the final files from inside > > > /var/lib/selinux/<store>/[active|previous|tmp] to > > > /var/lib/selinux/tmp/<store>. The move is done to facilitate using > > > source control management on the /var/lib/selinux/<store> directory. If > > > these files remain in /var/lib/selinux/<store> they will pose a size > > > problem if an SCM like git is used as we'd be storing lots of binary > > > diffs. We are suggesting making this change now, rather than later when > > > source policy, SCM, and CIL[1] support are available, to ease the > > > migration burden. > > > > > > These are the files that have been moved: > > > > > > /var/lib/selinux/<store>/active/... /var/lib/selinux/tmp/<store>/... > > > > > > file_contexts contexts/files/file_contexts > > > file_contexts.homedirs contexts/files/file_contexts.homedirs > > > file_contexts.local contexts/files/file_contexts.local > > > netfilter_contexts contexts/netfilter_contexts > > > policy.kern policy/policy.<policyversion> > > > seusers.final seusers > > > > > > The layout of these files in /var/lib/selinux/tmp/<store> is designed to > > > mirror their locations in /etc/selinux/<store>. This should help clarify > > > the relationship between these final files and the files installed in > > > etc. > > > > > > One consequence of this move is that reverting to the previous policy > > > version requires a policy rebuild. Currently you can revert without > > > rebuilding. > > > > That seems a little worrisome to me, as a rebuild might fail, e.g. what > > happens if we abort a transaction due to a lack of disk space and then > > try to revert, requiring a rebuild, only to run out of space during the > > rebuild? > > > If the transaction is aborted then the policy hasn't actually been > changed, so I don't think that this example would be a problem. It is > only after the transaction is complete that everything is written to the > final location. Or am I missing something? > > It would be a problem only if changes were made to the policy, that > policy loaded, there were problems, and then the rebuild of the previous > policy fails. I'm unclear on what state things would be left in. I'm also unclear on the implications of writing these files to a single tmp/ location rather than having separate copies in active/, previous/, and tmp/ - I don't want us to unwittingly clobber files or leave them in intermediate states (as happened with the earlier attempt to hard link files among active/, previous/ and tmp/). I tried running the code but I seem to see these "final files" still under the individual <store> directories in /var/lib/selinux rather than under tmp/. -- Stephen Smalley National Security Agency -- 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.
