Re: pass-through linux ioctl for mfi(4)

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Mon, Sep 17, 2012 at 11:02:13PM +0700, Robert Elz wrote:
>     Date:        Mon, 17 Sep 2012 17:47:08 +0200
>     From:        Manuel Bouyer <bouyer%antioche.eu.org@localhost>
>     Message-ID:  <20120917154708.GA25378%asim.lip6.fr@localhost>
> 
>   | But this assumes that the mfi driver is compiled in. it doesn't
>   | look right, especially in the context of modules.
> 
> Sure - the point is that you just do whatever the system would do
> in order to call the ioctl() function, other than actually calling it,
> to the point where you believe that it is the mfi driver, or that it
> isn't.
> 
> Of course, it doesn't need to be the d_ioctl that gets tested, you
> could use anything that's in cdevsw[] - and you can test anything
> that you can find (including the file name the module was dynamically
> loaded from).
> 
> What you do now (if it is a module) is that it must have already been
> loaded (because we have a fd that refers to it), so its functions are
> all there to test.

Sorry but I can't see how a kernel with COMPAT_LINUX but without
mfi would compile.

> 
> Personally, I wouldn't test at all, I'[d just assume it is an mfi
> if it is an mfi looking ioctl cmd, call the device's d_ioctl() in the
> normal way (after doing whatever arg/cmd massaging is needed to get
> from the compat interface to the native ioctl), and if it succeeds,
> you're done, if it fails, you just revert to the original everything
> and let whatever happens, happen (it wasn't mfi after all.)

that's what my patch does.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--