[bug#44549] [PATCH] etc: updates for the guix-daemon SELinux policy

[Daniel Brooks <db48x@db48x.net>]

Marius Bakke <marius@gnu.org> writes:

> Hello Daniel,
>
> Thanks a lot for this.

You're welcome.

>
> Daniel Brooks <db48x@db48x.net> writes:
>
>>>From 7dd9ed6da01c5bf125c95592f4978b579198731a Mon Sep 17 00:00:00 2001
>> From: Daniel Brooks <db48x@db48x.net>
>> Date: Mon, 9 Nov 2020 07:03:42 -0800
>> Subject: [PATCH] etc: updates for the guix-daemon SELinux policy
>>
>> * etc/guix-daemon.cil.in: I can't promise that this is a complete list of
>> everything that guix-daemon needs, but it's probably most of them. It can
>> search for, install, upgrade, and remove packages, create virtual machines,
>> update itself, and so on. I haven't tried creating containers yet, which might
>> reveal more things to add.
>
> This commit message is somewhat unorthodox.  :-)
>
> Perhaps it can be shortened to:
>
> * etc/guix-daemon.cil.in (guix_daemon): Specify more permissions for
> guix-daemon to account for daemon updates and newer SELinux.

I suppose. Personally I dislike the changelog style commit messages, but
when in Rome…

>> +;; In particular, you can run semanage permissive -a guix_daemon.guix_daemon_t
>> +;; to allow guix-daemon to do whatever it wants. SELinux will still check its
>> +;; permissions, and when it doesn't have permission it will still send an
>> +;; audit message to your system logs. This lets you know what permissions it
>> +;; ought to have. Use ausearch --raw to find the permissions violations, then
>> +;; pipe that to audit2allow to generate an updated policy. You'll still need
>> +;; to translate that policy into CIL in order to update this file, but that's
>> +;; fairly straight-forward. Annoying, but easy.
>
> I'm not sure about the second paragraph.  It's mainly a rehash of the
> blog post, no?  And there are many other ways to go about
> troubleshooting SELinux (I did not use ausearch at all).

True. I just wanted a quick summary somewhere in the source so that
future us won't have to rely on a random blog post, even one from Dan
Walsh.

> diff --git a/etc/guix-daemon.cil.in b/etc/guix-daemon.cil.in
> index 666e5677a3..b5909f1b18 100644
> --- a/etc/guix-daemon.cil.in
> +++ b/etc/guix-daemon.cil.in
> @@ -84,6 +84,9 @@
>    (allow init_t
>           guix_daemon_t
>           (process (transition)))
> +  (allow init_t
> +         guix_store_content_t
> +         (lnk_file (read)))

This one is a little unusual; is your service file symlinked or something?

>    (allow init_t
>           guix_store_content_t
>           (file (open read execute)))
> @@ -166,6 +169,9 @@
>    (allow guix_daemon_t
>           root_t
>           (dir (mounton)))
> +  (allow guix_daemon_t
> +         guix_daemon_socket_t
> +         (sock_file (unlink)))

That shouldn't be a problem, though we don't have any other rules for
guix_daemon_socket_t. Possibly that is because my socket file is labeled
guix_daemon_conf_t, for unknown reasons. Perhaps it was not labeled
correctly when created, and hasn't been relabeled since.

>    (allow guix_daemon_t
>           fs_t
>           (filesystem (getattr)))
> @@ -348,7 +354,12 @@
>                                getopt setopt)))
>    (allow guix_daemon_t
>           self
> -         (tcp_socket (accept listen bind connect create setopt getopt getattr ioctl)))
> +         (netlink_route_socket (read write)))
> +  (allow guix_daemon_t
> +         self
> +         (tcp_socket (accept
> +                      listen bind connect create read write
> +                      setopt getopt getattr ioctl)))

These are fine; in fact I discovered these myself this morning and was
going to send a patch.

> Can you test these additional changes on Fedora?

Yes, I'll let you know if there are any problems. Also, I'll investigate
the socket file some more.

>
> With this, I no longer have to go through 'guix pack' and 'podman' to
> run Guix packages on my RHEL workstation!  :-)

Ideal :)

>
> Also, is it OK to add you to the list of contributors at the top of the
> file with this name and address?

Certainly.

db48x