Re: Fwd: [EXTERNAL] Re: ioctl.c:undefined reference to `__get_user_bad'

[Randy Dunlap <rdunlap@infradead.org>]

On 5/18/21 3:03 AM, Russell King (Oracle) wrote:
> On Mon, May 17, 2021 at 02:06:33PM -0700, Randy Dunlap wrote:
>> [adding back linux-arm-kernel; what happened to it? ]

That was about what I thought was Steve F. dropping LAK on his response email.

> Nothing. I'm not interested in trying to do major disgusting
> contortions to make get_user() work for 8-byte values. If someone
> else wants to put the effort in and come up with an elegant solution
> that doesn't add warnings over the rest of the kernel, that's fine.
> 
> As far as I remember, everything in __get_user_err() relies on
> __gu_val _not_ being 64-bit. If we use the same trick that we do
> in __get_user_check():
> 
> 	__inttype(x) __gu_val = (x);
> 
> then if get_user() is called with a 64-bit integer value and a
> pointer-to-32-bit location to fetch from, we'd end up passing a
> 64-bit integer into the __get_user_asm() which could access the
> wrong 32-bit half of the value in BE mode. Similar issue with
> 64-bit vs pointer-to-16-bit.

Yes, trying to handle get_user() of size 8 bytes is quite messy.
I have a few versions and they are all ugly and cause build warnings.

So we are down to what bugzilla calls WONTFIX. I'm OK with that.

Thanks.

-- 
~Randy