Re: [RFC] refining systemd mountpoints

From: Dominick Grift <dac.override@gmail.com>
To: Chris PeBenito <pebenito@ieee.org>,
	refpolicy <selinux-refpolicy@vger.kernel.org>
Subject: Re: [RFC] refining systemd mountpoints
Date: Mon, 13 Jan 2020 10:42:10 +0100	[thread overview]
Message-ID: <20200113094210.GB870816@brutus.lan> (raw)
In-Reply-To: <20200109214240.GA2283901@brutus.lan>

[-- Attachment #1: Type: text/plain, Size: 2929 bytes --]

On Thu, Jan 09, 2020 at 10:42:40PM +0100, Dominick Grift wrote:
> On Thu, Jan 09, 2020 at 04:06:38PM -0500, Chris PeBenito wrote:
> > I'd like to refine how the policy handles systemd's mounton so that it works
> > similar to how we manage mountpoints for mount_t. Since systemd can be made
> > to mount over just about anything, I'm looking at adding a new conditional
> > that would allow init_t to mounton non_security_file_type, and then an
> > interface like files_mountpoint().
> > 
> > The question is for the implementation of the interface; I see two options,
> > either the interface allows mounton for all file-like classes, or the
> > classes are specified as a parameter:
> > 
> > --------
> > init.te:
> > attribute init_mountpoint_type;
> > allow init_t init_mountpoint_type:dir_file_class_set mounton;
> > 
> > init.if:
> > interface(`init_mountpoint',`
> > typeattribute $1 init_mountpoint_type;
> > ')
> > --------
> > 
> > or
> > 
> > --------
> > init.if:
> > interface(`init_mountpoint',`
> > allow init_t $1:$2 mounton;
> > ')
> > --------
> > 
> > I like the first option because it is clearer since you can see the mounton
> > in init.te, but that is excessive access.  The second option could be made
> > to look like the first option, but it would need several attributes and
> > interfaces, e.g. init_dir_mountpoint_type, init_file_mountpoint_type, etc.
> > which isn't so desirable.
> > 
> > Any thoughts on this?
> 
> I implemented the former in my policy. ie the dir_file_class_set equiv..
> 
> 4163               (allow subj bind_path_obj_type_attribute (dirs (create)))
> 4164               (allow subj bind_path_obj_type_attribute list_dir_perms)
> 4165               (allow subj bind_path_obj_type_attribute (dir (mounton)))
> 4166               (allow subj bind_path_obj_type_attribute create_file_perms)
> 4167               (allow subj bind_path_obj_type_attribute (file (mounton)))
> 
> As you can see i even allow systemd to create the mountpoint in case it does not exist. For example if /etc/machine-id does not exist and I have a BindReadOnlyPath=/etc/machine-id then systemd will touch /etc/machine-id and mount it ro

Okay, I think I am wrong. It will not create the bind_path if it does not exist. Not sure how I got to this...

> 
> It also generally buggy. Systemd does not (alway's) use setfscreatecon to create the mountpoints. And sometimes it does use setfscreatecon where it shouldnt.
> 
> https://github.com/systemd/systemd/issues/13762
> 
> > 
> > -- 
> > Chris PeBenito
> 
> -- 
> Key fingerprint = 5F4D 3CDB D3F8 3652 FBD8 02D5 3B6C 5F1D 2C7B 6B02
> https://sks-keyservers.net/pks/lookup?op=get&search=0x3B6C5F1D2C7B6B02
> Dominick Grift

-- 
Key fingerprint = 5F4D 3CDB D3F8 3652 FBD8 02D5 3B6C 5F1D 2C7B 6B02
https://sks-keyservers.net/pks/lookup?op=get&search=0x3B6C5F1D2C7B6B02
Dominick Grift

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]