Re: [RFC] [PATCH] Replace security fields with hashtable

From: Mimi Zohar (zohar@private)
Date: Thu Oct 28 2004 - 12:50:29 PDT


Chris,

>* Serge E. Hallyn (hallyn@private) wrote:
>> LSM hooks can also be used for performance measurements, to aid an audit
>> subsystem, etc.  And with LSM's like bsdjail and securelevel, stacking
with
>> SELinux is still useful even though all are purely security modules.

>Only valid use is for access control (and I'll agree it can aid audit).

Although we're using the LSM hooks for access control, we are also using
them for
other functions as well, such as attestation, as described in
"Attestation-based
Policy Enforcement for Remote Access" (ACM CCS, October 2004), and signed
executables
verification.  With capabilities, you're now up to 4 different usages.

>> > I don't think arbitary composition of security models is a service
that
>> > the Linux kernel should provide.
>>
>> Here we fundamentally disagree.  Something which can be unsafe for some
if
>> improperly used, but useful for others, should not therefore be
disabled.

>Problem is showing it's useful enough to make a change.  The biggest
>issue I see is that the current scheme is tough to code for both ways.
>The module has to make a choice how to code itself w.r.t. stacking and
>accessing it's security labels.

Correct me if I'm wrong, but the issue isn't the difference in coding the
loading
of the module itself, but in accessing the security data pointers.
Personally,
I think it is really important to be able to share the hooks and the
associated
security data pointers in a logical and coherent fashion.  For this reason,
it makes
sense to have stacker be the hook owner and all LSM security modules
writing to it.
None of the LSM security modules, other than perhaps Stacker, should be
builtin, but
should be loaded in the initrd.

Mimi Zohar



This archive was generated by hypermail 2.1.3 : Thu Oct 28 2004 - 13:32:55 PDT