Re: [PATCH 15/25] autofs: move ioctl32 to autofs{,4}/root.c

[Arnd Bergmann <arnd@arndb.de>]

On Sünndag 06 November 2005 07:22, Ian Kent wrote:
> On Sat, 5 Nov 2005, Arnd Bergmann wrote:
>
> I'm not sure if I like conditional compilation in the code proper but I'll 
> leave it to you to make the final decision since your running with the 
> change. Is there a reason the definitions can't simply be left in place?

I think the compat_ptr() macro is not defined on architectures that don't
have 32 bit compat code, but we could change that.
 
> Its been a while since I trawled through the compat ioctl code (please 
> point me to the right place) but with this change I think that the 
> AUTOFS_IOC_SETTIMEOUT32 is redundant. Consider a conditional define for 
> AUTOFS_IOC_SETTIMEOUT in include/linux/auto_fs.h instead. Both autofs and 
> autofs4 use that definition.

The point here is that the two are different on 64 bit platforms, since
sizeof (int) != sizeof (long). You also can't do

switch (cmd) {
case AUTOFS_IOC_SETTIMEOUT32:
case AUTOFS_IOC_SETTIMEOUT:
	return do_stuff();
}

because then gcc would complain about duplicate case targets on 32 bit
targets.
 
> The lock_kernel()/unlock_kernel() in the autofs4 patch is ineffective as 
> the BKL is not used for syncronisation anywhere else in autofs4. If 
> removing it causes problems I need to know about'em so I can fix'em 
> (hopefully).

I used the BKL here in order to maintain the current semantics, because
ioctl is always called with BKL held, and compat_ioctl is called without
it.

If you are sure you don't need the BKL, then you should also replace
".ioctl = ..." with ".unlocked_ioctl = ...".

	Arnd <><