Re: [RFC][PATCH 0/4] slab: Allow for type introspection during allocation

[Przemek Kitszel <przemyslaw.kitszel@intel.com>]

On 7/10/24 01:28, Kees Cook wrote:
> On Tue, Jul 09, 2024 at 11:02:55PM +0200, Marco Elver wrote:
>> On Tue, 9 Jul 2024 at 22:28, Kees Cook <kees@kernel.org> wrote:
>>>
>>> On Tue, Jul 09, 2024 at 10:26:32AM -0700, Christoph Lameter (Ampere) wrote:
>>>> On Mon, 8 Jul 2024, Kees Cook wrote:
>>>>
>>>>>
>>>>>             obj = kmalloc(obj, gfp);
>>>>
>>>> Could we avoid repeating "obj" in this pattern?
>>>>
>>>> F.e.
>>>>
>>>>        KMALLOC(obj, gfp);
>>>
>>> This appears to be the common feedback, which is good! :) And we can
>>> still have it return "obj" as well, so it could still be used in
>>> "return" statements, etc. I will work up a new RFC...
>>
>> More macros like this only obfuscate the code further. The name would
>> become something that makes it really clear there's an assignment.
>>
>>    assign_kmalloc(obj, gfp)
>>
>> There may be better options. Also ALLCAPS could be avoided here, as we
>> have done with other language-like features (vs. pure constants).
> 
> So, in looking a code patterns, it seems what we really want more than
> returning the object that was allocated is actually returning the size
> of the allocation size requested. i.e.:
> 
> 	info->size = struct_size(ptr, flex_member, count);
> 	info->obj = kmalloc(info->size, gfp);
> 
> would become:
> 
> 	info->size = kmalloc(info->obj, flex_member, count, gfp);
> 
> -Kees
> 

that will work out also for the (IMO) most common case of checking if
the allocation succeeded:
	if (!kmalloc(my_foo, flex_part, count, gfp))
		return -ENOMEM;