[refpolicy] load_policy() with upstart on mint 9 fluxbox

[Justin P. Mattock]

On 10/21/2010 12:08 AM, Shaz wrote:
>
>
> On Thu, Oct 21, 2010 at 11:48 AM, Justin P. Mattock
> <justinmattock at gmail.com <mailto:justinmattock at gmail.com>> wrote:
>  > On 10/20/2010 11:26 PM, Shaz wrote:
>  >>
>  >> On Thu, Oct 21, 2010 at 10:15 AM, Justin P. Mattock
>  >> <justinmattock at gmail.com <mailto:justinmattock at gmail.com>>  wrote:
>  >>>
>  >>> On 10/20/2010 07:44 PM, Serge E. Hallyn wrote:
>  >>>>
>  >>>> Quoting Justin P. Mattock (justinmattock at gmail.com
> <mailto:justinmattock at gmail.com>):
>  >>>>>
>  >>>>> o.k. finally connected the dots that I needed to create a initrd.img
>  >>>>> in order for this to load(im a total newbie!!)
>  >>>>>
>  >>>>> Anyways the policy loads everything went in and am now in full
>  >>>>> enforcement mode.. only real issue is with lxde
>  >>>>> same bug here:
>  >>>>> https://bugzilla.redhat.com/show_bug.cgi?id=552885
>  >>>>>
>  >>>>> seems lxde is in /usr/sbin reason probably for the wrong filelabel..
>  >>>>
>  >>>> Cool, so does following the steps outlined in that bug make it
>  >>>> work for you?
>  >>>>
>  >>>
>  >>> What I normally have is /boot/System.map-* and vmlinuz-* to load the
>  >>> kernel.. Seems sysvinit knows how to take things there and
> load_policy()
>  >>>
>
>
>  >>> for upstart whatever it's doing(like what you said) needs to go through
>  >>> initrd. Yesterday I though thats what I had done with:
>  >>> fakeroot make-kpkg --initrd --append-to-version=-custom kernel_image
>  >>> kernel_headers
>  >>>
>  >>> but missed one last step:
>  >>> mkinitramfs -k -o initrd.img-2.6.36-rc8-custom-00022-g2b666ca
>  >>> then after doing this everything loaded as is..
>  >>>
>  >>> Note: guess this is whats being called to do all of this:
>  >>> /usr/share/initramfs-tools/scripts/init-bottom/_load_selinux_policy
>  >>>
>  >>> As for the file labels in /var/run seems most of the files in there are
>  >>> labeled with initrc_t (keep in mind I chose debian as the distro in
>  >>> build.conf, so maybe this is why)..
>  >>>
>  >>> As for lxde, before using chcon I was getting a login context of
>  >>> name:staff_r:netutils_t:s0 then after relabeling those files:
>  >>>
>  >>> (chcon to this context like the bug report had shown)
>  >>> system_u:object_r:xdm_exec_t:s0 /usr/sbin/lxdm
>  >>> system_u:object_r:xdm_exec_t:s0 /usr/sbin/lxdm-binary
>  >>> system_u:object_r:xdm_var_run_t:s0 lxdm.pid
>  >>>
>  >>> I login with the proper context that I chose:
>  >>> name:staff_r:staff_t:s0
>  >>>
>  >>> Right now I think everything is running o.k. on this operating system..
>  >>> (nice,small, and functional..with a touch of SELinux on top...)
>  >>
>  >> Dear Justin,
>  >>
>  >> "initrd" helps to load selinux and label "init" so that transitions
>  >> can take effect. Be it upstart or sysvinit!
>  >>
>  >> If this is not done then all your processes will be loaded with
>  >> unconfined_t. Rest of the details you are considering should not
>  >> matter and they might be confusing if you tried to load selinux with
>  >> experimentation rather then proper bootstrap through initrd.
>  >>
>  >> Hope this helps.
>  >>
>  >
>  > well my other machines do not use an initrd image file in /boot only
>  > System.map and vmlinuz(I guess doing make, make install for the
> kernel is a
>  > bit outdated these days)..
>  >
>  > Anyways sysvinit always loaded the policy just fine and had the processes
>  > correct, as for upstart seems I needed to do the whole
>  > make-kpkg,dpkg,mkinitramfs thing..(which is fine... but would rather
> keep it
>  > more simpler).
>
> AFAIK without initrd your policy will not work properly as labeling is
> not handled and thus the whole security objective fails. If this is
> possible without an initrd then it would be init handling it by loading
> selinux and restarting itself.

pretty much what I use over here..(make bisecting easier since I dont 
have to deal with the hundreds of vmlinuz tags, just one that gets 
written over)

Now the question(I guess)is, is how/why is upstart not _realizing_ this 
is a system without initrd lets go over here and get the system started?
(id imagine the code might have if then/else or something..)

Justin P. Mattock