lkml.org 
[lkml]   [2012]   [May]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] staging: comedi: remove __user annotation inside of struct's
On 2012-05-09 11:31, Dan Carpenter wrote:
> On Wed, May 09, 2012 at 11:20:07AM +0100, Ian Abbott wrote:
>> On 2012-05-09 00:55, H Hartley Sweeten wrote:
>>> On Tuesday, May 08, 2012 4:41 PM, H Hartley Sweeten wrote:
>>>>
>>>> The structs' comedi_insn, coomedi_insnlist, comedi_cmd,
>>>> comedi_chaninfo, and comedi_rangeinfo are all passed to
>>>> the kernel from user space using ioctl commands. They
>>>> are then copied to kernel space using copy_from_user()
>>>> before the data is passed to the drivers.
>>>>
>>>> The __user annotation should not be used with variables
>>>> inside the struct. This produces a lot of sparse warnings
>>>> like:
>>>>
>>>> warning: dereference of noderef expression
>>>>
>>>> Note: This patch exposes some new warnings about different
>>>> address space. These will be addressed.
>>>
>>> Please ignore this patch.
>>>
>>> It appears the annotations in the struct definitions are correct.
>>
>> Personally, I think you were on the mark with the patch. It's
>> better to avoid using __user in comedi.h so it can be used as-is in
>> user-space.
>
> Sparse is useful so we shouldn't break it. I always run sparse over
> my patches before submission and look at the warnings. Except if
> they scroll off the page. In that case, I just figure that the
> author deserves the bugs.
>
> We could just do some ifdeferry to fix it for userspace.

That doesn't help in cases such as 'struct comedi_insn' where the 'data'
pointer is a user-space pointer in the user-space copy of the object and
a kernel-space pointer in the kernel-space copy of the object. The only
fix for that is to have separate "k" versions of the struct or to do a
load of casting, which is slightly error-prone and makes the code less
readable.

Are there any handy macros for casting pointers to __user pointers,
something like

#define _user(p) ((typeof(*(p)) __user *)(p))

but preferably without the repeated expansion of 'p' in case of
side-effects?

--
-=( Ian Abbott @ MEV Ltd. E-mail: <abbotti@mev.co.uk> )=-
-=( Tel: +44 (0)161 477 1898 FAX: +44 (0)161 718 3587 )=-


\
 
 \ /
  Last update: 2012-05-09 13:41    [W:0.106 / U:0.472 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site