On Mon, 31/01/2011 at 14.38 -0500, Christopher J. PeBenito wrote:
> On 1/24/2011 4:01 PM, Dominick Grift wrote:
> > On 01/24/2011 09:54 PM, Guido Trentalancia wrote:
> >> On Mon, 24/01/2011 at 14.57 +0100, Dominick Grift wrote:
> >>> On 01/24/2011 01:44 AM, Guido Trentalancia wrote:
> >>>> diff -pruN -x .git -x booleans.conf -x corenetwork.if -x corenetwork.te -x modules.conf refpolicy-git-18012011/policy/modules/system/selinuxutil.te refpolicy-git-18012011-new/policy/modules/system/selinuxutil.te
> >>>> --- refpolicy-git-18012011/policy/modules/system/selinuxutil.te 2011-01-17 19:36:10.814131755 +0100
> >>>> +++ refpolicy-git-18012011-new/policy/modules/system/selinuxutil.te 2011-01-23 04:14:02.662963912 +0100
> >>>> @@ -444,6 +444,7 @@ files_read_etc_files(semanage_t)
> >>>> files_read_etc_runtime_files(semanage_t)
> >>>> files_read_usr_files(semanage_t)
> >>>> files_list_pids(semanage_t)
> >>>> +files_search_default(semanage_t)
> >>>
> >>> There should not be any default_t directories. Thus this shouldnt be
> >>> allowed.
> >>
> >> This stems from the fact that at some point I came to a state where
> >> while working from the terminal (as opposed to working from a graphical
> >> terminal), semanage had trouble dealing with some temporary local
> >> modules that I was working with for testing purposes (they were labelled
> >> default_t)...
> >>
> >> Of course it can be removed. So, in general default_t should never
> >> appear anywhere in the policy ? Just for curiosity, what is the reason
> >> behind that ? If it is allowed to carry out operations on usr and
> >> etc_runtime files, why shouldn't it allowed to carry out operations on
> >> default_t ?
> >
> > locations unknown to selinux are labelled default_t. So for example if
> > you create a dir named /test in the root of the filesystem. There is no
> > file context specification for it and thus selinux labels it default_t.
> >
> > Basically it signals some incompatibility in that sense because files
> > always need a label, and this should not happen.
> >
> > Why should refpolicy support a scenario that should never happen in the
> > first place? default_t is not like other types like usr_t or
> > etc_runtime_t. types like file_t, default_t, unlabeled_t all signal some
> > other issues. I think this is discussed in "Fedora SELinux user guide"
> > (you can find it with google, its a free document.
>
> This is the right idea, but I want to clarify the labeling.
>
> default_t: this means no entry in file_contexts matches this file, so it
> falls back to default_t.
>
> unlabeled_t: this means the object has an invalid context. Typically
> this happens if a type is removed from the policy while the system is
> running.
>
> file_t: this means a filesystem that supports extended attributes is
> mounted for the first time on a SELinux system. The default for the
> files in this case is file_t.
>
> In all of these cases, the security attributes of the object are unknown.
Yes. I knew already about unlabeled_t and default_t, but I didn't know
about the existence and meaning of file_t, as I thought it was going to
default to unlabeled_t in the case mentioned above.
So good to know (we never stop learning) and thanks for pointing that
out, Christopher !
Basically, what happened is that I had a local problem with semanage
configuration, that led to a mislabeled /root directory. Everything
sorted out now.
Regards,
Guido