Re: [PATCH] power: disable hibernation if module loading is disabled

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Friday, March 04, 2011, Vasiliy Kulikov wrote:
> On Fri, Mar 04, 2011 at 22:21 +0100, Rafael J. Wysocki wrote:
> > On Friday, March 04, 2011, Vasiliy Kulikov wrote:
> > > If modules_disabled is set to 1, then nobody, even full root may not write
> > > to the kernel, right?  So, if something permits to indirectly pass
> > > modules_disabled restriction, this is a bug.  Otherwise,
> > > modules_disabled is confusing as it gives false sense of security.
> > > 
> > > -OR-
> > > 
> > > modules_disabled's documentation should be changed to note that it
> > > doesn't prevent rootkit uploading, but only forbids modprob'ing modules
> > > via the "official" init_module(2) gate, disallowing e.g. module autoloading.
> > 
> > Why not to change that documentation, then?
> 
> Because it's better to fix something (if it is possible, of course) than
> simply documenting the bug.

modules_disabled surely is not the right interface to disable hibernation
and I don't really think there's a bug because it doesn't work as you'd like
it to.  In fact, there would be a bug if it did work that way.

> > Also, please note that in order to "write" into memory using the hibernation
> > interface you need to have write access to swap,
> 
> No, you may just "write the kernel" via write() /dev/snapshot, this is
> the way uswsusp works.  I didn't check whether it really needs
> temporary file to change the kernel memory or it may be done entirely
> without disk iteraction.  This is irrelevant to modules_disabled policy
> violation, though.

Sorry, but who defined the "modules_disabled policy" and when did that happen
and how come that I'm not aware of it?

> > which you can use to corrupt
> > memory regardless of the modules_disabled setting AFAICS.
> 
> Please correct me if I'm wrong, but kernel memory is not swappable at
> all and only userspace memory is written to the swap.  Root with
> CAP_SYS_ADMIN already may do everything with all processes, so this is
> not a threat.

OK

> If one may change kernel memory via swap then it is another problem with
> modules_disabled.

If you want an interface to disable _any_ kind of writes into the kernel
memory by any means, then please add it and don't call it modules_disabled,
because it's a hell of a confusing name and no amount of documentation
can help that.

Thanks,
Rafael