Re: [gentoo-dev] Re: Re: turning off a default compile option

[Greg KH]

On Tue, Jul 06, 2004 at 06:18:12AM -0700, Duncan wrote:
> Greg KH posted <20040705200509.GB18900@×××××.com>, excerpted below,  on
> Mon, 05 Jul 2004 13:05:09 -0700:
>
> > On Mon, Jul 05, 2004 at 09:33:34AM -0700, Duncan wrote:
> >> That's not a sufficient test.  Mainline 2.6 kernels have standard Linux
> >> security capabilities broken out into their own module, now.  Thus, it
> >> can be compiled in as traditional, compiled as a module (my
> >> configuration, with it specifically loaded at boot, since the kernel
> >> isn't smart enough to auto-load it on its own, yet)
> >
> > And how would you expect such a "auto-load" system to work for the
> > capabilities module?
>
> I really don't know.  I really don't know how the kernel knows to load a
> LOT of the modules, but it does.

It knows how to "autoload" modules 2 different ways:
	- A hardware device is added to the system, which generates a
	  hotplug event and that causes the driver that can control the
	  device to be loaded.  See the linux-hotplug.sf.net web site
	  for some (pretty old) information about this, or a paper about
	  how it all works (a bit old, but it still happens the same
	  way) is at:
	  http://www.kroah.com/linux/talks/ols_2001_hotplug_paper/hotplug.ps
	  with some slides from my presentation about this at:
	  http://www.kroah.com/linux/talks/ols_2001_hotplug_talk/html
	  
	- a device node is opened, and there is no driver that is
	  associated with it.  This causes the kernel try try to load
	  any module that says it will support that device.  Now, this
	  method is going out of style, as most major/minor numbers
	  associated with a device node are not directly mapped to a
	  driver (like for all pci devices).  If you use udev, you can
	  not do this kind of loading, as the device node is not present
	  for the kernel to determine it in the first place.

Now before you think that autoloading the capabilities module the second
way mentioned above, grep the kernel source tree for the capable()
function.  Yeah it's called _everywhere_ whenever you do pretty much
_anything_.

So sorry, we can't autoload the capabilities module right now, and I
don't see it happening in the future. :(

> BTW, thanks for the reply.  It's been interesting at times, watching
> people respond to your posts, not having /any/ idea that you ARE one of
> the kernel developers, indeed, UDEV wouldn't be the same without you.  <g>
> You don't wear it on your sleeve, or in a sig, like some would, and your
> name doesn't quite stand out like "Linus" would. =:^)

Shhh, don't blow my cover :)

Hope this helps,

greg k-h

--
gentoo-dev@g.o mailing list