Re: [kernel-hardening] [PATCH 2/2] security: Yama LSM

[Vasiliy Kulikov <segoon@openwall.com>]

Sorry, dropped James from cc.

On Tue, Dec 20, 2011 at 09:35 +0400, Vasiliy Kulikov wrote:
> Hi,
> 
> On Tue, Dec 20, 2011 at 07:52 +1100, James Morris wrote:
> > On Mon, 19 Dec 2011, Kees Cook wrote:
> > > On Mon, Dec 19, 2011 at 11:33:10AM +1100, James Morris wrote:
> > > > On Thu, 15 Dec 2011, Kees Cook wrote:
> > > > > +#ifdef CONFIG_SECURITY_YAMA
> > > > > +	ns->ptrace_scope = parent_pid_ns->ptrace_scope;
> > > > > +#endif
> > > > > +
> > > > 
> > > > I'd like to see this implemented as an LSM hook, something like 
> > > > security_ptrace_set_scope().
> > > 
> > > I must be dense, but I fail to understand the purpose of this. The "ptrace
> > > scope" implemented by Yama is a sysctl, not an system interface. I don't
> > > understand why (or where) other LSMs would want to catch changing this.
> > > Can you explain what you're looking for in more detail?
> > > 
> > 
> > We should not see YAMA-specific code in the core kernel.  However you do 
> > it, the above should happen in LSM.
> 
> Probably this should be security_pid_namespace_create() instead of
> security_ptrace_set_scope()?  (Or even use create an analog of
> register_pernet_subsys() for pid_ns.)
> 
> Then have ->ptrace_scope and similar things as per-LSM private variables
> like in task_struct->cred->security.  ns->security should be dynamically
> allocated.
> 
> Thanks,
> 
> -- 
> Vasiliy