Re: [patch 5/12] lsm stacking v0.2: actual stacker module

[serge@hallyn.com]

Quoting Tony Jones (tonyj@suse.de):
> Hey Serge,
> 
> I don't think your symbol_get() is doing what you think it is ;-)

Hmm, I wonder whether something changed.  It shouldn't be possible to
rmmod module b if module a has done a symbol_get on it...  This may mean
more stringent locking will be required after all to support unloading.
That, or a rmmod lsm hook.

> > + * Add the stacked module (as specified by name and ops).
> > + * If the module is not compiled in, the symbol_get at the end will
> > + * prevent the the module from being unloaded.
> > +*/
> > +static int stacker_register (const char *name, struct security_operations *ops)
> > +{
>  ...
> > +	symbol_get(ops);
> > +
> > +out:
> > +	spin_unlock(&stacker_lock);
> > +	return ret;
> > +}
> 
> 
> Seemed useful to be able to view which modules had been unloaded.
> Easier to maintain them on their own list than to compute the difference
> of <stacked_modules> and <all_modules>.  Patch attached, not sure if you
> are cool with reusing the 'unload' file.

No, that's good, thanks.  Though I guess "unloading" of this type won't
be needed if true module deletion has to be supported.

> Apart from this, looks good.  I ran it against our regression tests using
> AppArmor (SubDomain) composed with Capability and everything was functionally
> as expected.   I still need to run it through our SMP stress tests.

Excellent :)

thanks,
-serge