Some file_contexts regular expressions in refpolicy-strict are causing
genhomedircon to die; refpolicy is failing to build for me entirely.
The regular expressions seem redundant to me, perhaps I am missing
something, but the following patch fixes the problems for me. Please
review and apply
Specifically:
- polmatch need no longer make an exception for unlabeled_t
since a flow will now always match SPD rules with no contexts (per
the IPSec leak fix patch upstreamed a few weeks back), as
opposed to needing polmatch access to unlabeled_t.
Signed-off-by: Venkat Yekkirala <vyekkirala@TrustedCS.com>
- Allow semanage to read from /root on strict non-MLS for
local policy modules.
- Gentoo init script fixes for udev.
- Allow udev to read kernel modules.inputmap.
- Dnsmasq fixes from testing.
- Allow kernel NFS server to getattr filesystems so df can work
on clients.
> We could add another 'or' on the above constraint:
>
> or ( (t2 == mlsfilewrite_in_range) and (l1 dom l2) and (h1 domby h2) )
>
> I believe that would be the constraint you were looking for. I don't
> like the name of that attribute, but I couldn't come up with a better
> one off the top of my head. :)
>
Attached is a patch which I've tested against selinux-policy-2.4.2-1
that implements this additional constraint. The name is still a bit
forced, but it works.
-matt <mra at hp dot com>
Allow unconfined processes to see unlabeled processes in ps.
Removed a redundant rule in samba.te
Removed support for the pre-Fedora Red Hat code to create sym-links in /boot.
Removed support for devpts_t files in /tmp (there is no way that would ever
work).
Allowed postgrey to create socket files.
Made the specs for the /lib and /lib64 directories better support stem
compression.
This policy is much tighter than the GConf policy from the old example
policy. It only allows gconfd to access configuration data stored by
GConf. Users can modify configuration data using gconftool-2 or
gconf-editor, both of which use gconfd. GConf manages multiple
configuration sources, so gconfd should be used to make any changes
anyway. Normal users who aren't trying to directly edit the
configuration data of GConf won't notice anything different.
There is also a difference between this policy and the old example
policy in handling directories in /tmp. The old example policy
labeled /tmp/gconfd-USER with ROLE_gconfd_tmp_t, but, since there was no
use of the file_type_auto_trans macro, if that directory was deleted
gconfd would create one labeled as tmp_t. This policy uses the
files_tmp-filetrans macro to cause a directory in /tmp created by gconfd
to be labeled as $1_tmp_t. It is not labeled with $1_gconf_tmp_t,
because if /tmp/orbit-USER is deleted, gconfd will create it (through
use of ORBit) and it would get the $1_gconf_tmp_t label. By having
gconfd create $1_tmp_t directories in /tmp and $1_gconf_tmp_t files and
directories in directories labeled with $1_tmp_t, it can control its
data without requiring any future bonobo or Gnome policies to have
access to $1_gconf_tmp_t.
This patch is related to work that I am doing in making gconfd an
userspace object manager. If any user program can modify the
configuration data that GConf stores, than making gconfd an userspace
object manager would be useless.
Signed-off-by: James Carter <jwcart2@tycho.nsa.gov>
Chris PeBenito