Re: [gentoo-hardened] SELinux documentation draft

[Chris PeBenito]

On 1/6/2011 5:32 PM, Sven Vermeulen wrote:
> I've been working on bringing the SELinux handbook as currently available on
> http://www.gentoo.org/proj/en/hardened/selinux/selinux-handbook.xml more
> up2date. It's somewhat of a rewrite, but with all elements of the original
> SELinux handbook still inside it (apart from the troubleshooting as I guess
> those are quite outdated, being from 2006 and older).

The troubleshooting is not outdated, though there could be a few additions.

> The draft is currently available in the hardened-docs.git repository. In
> http://git.overlays.gentoo.org/gitweb/?p=proj/hardened-docs.git;a=tree;f=html/selinux;hb=HEAD
> you should be able to select individual chapters (HTML format) in the "raw"
> tree to view them somewhat like they would on the Gentoo site, but for your
> convenience there's also a PDF available at
> http://git.overlays.gentoo.org/gitweb/?p=proj/hardened-docs.git;a=tree;f=pdf;hb=HEAD

I looked through section 1 and 2 of the pdf version, and here are my 
notes so far:

1.2.1 The best way to look at security is by defining your security 
goals.  An example would be protecting critical system files, such as 
the kernel image, shadow passwords, and SELinux policy binary from being 
written except by trusted processes.

1.2.2 I don't understand the point of this section

1.2.3 I'd say this is not appropriate for this document.

1.2.4 This has problems; authentication is related to access control, 
but not part of it.  Also, "discretionary" in the sense of access 
control means that users can change the policy.  Contrast that with 
mandatory access control, where the security admin sets the policy.

1.3.1 SELinux has been included in the kernel for years, we can drop the 
"kernel patch" part.

2.2.1 This should be users, roles, and types.  "Domain" used in SELinux 
refers to a type that labels a process.  The amount of access to 
unlabeled processes is defined by the policy, not by the SELinux 
mechanisms.  I also think that the newrole example is out of scope for 
this section.

2.2.2 "Contexts for permission rules" doesn't make sense.  "Rules" or 
"Access control policy" or something similar makes more sense.  You also 
jump into rules with no discussion of object classes.  The part where 
you say that if access is denied by DAC, then SELinux isn't checked 
should be earlier in the intro.

2.3.1 same thing about domain vs. type.  A type is not a state, it is a 
security attribute.

2.3.2 I think discussion of the specific rules required for a successful 
domain transition are not relevant for this part of the doc.

2.4.1 You say what can be technically done by using a role, but you 
don't discuss what the concept of a role is.

2.4.2 Technically, the only default role is object_r.  The remainder are 
just common ones that you see if you use refpolicy.

2.4.3 The key reason for having a SELinux user id that is separate from 
the Linux user id is that the seuser doesn't change during a login 
session, but the Linux uid can be changed, eg by setuid or su.

2.5.1 The purpose of MLS is to provide a hierarchical confidentiality 
policy.  "This allows administrators to ensure that
information flow is more closely governed" is incorrect.  MLS can be 
implemented in TE, but it causes an explosion of types, so its easier to 
have an additional mechanism.

2.6.1 The discussion on why an access is allowed or denied always starts 
with TE.  In this case, user_t can't alter the clock, nor can it reach 
the hwclock_t domain since there is no domain transition.

-- 
Chris PeBenito
<pebenito@g.o>
Developer,
Hardened Gentoo Linux