Re: [PATCH] slab: Introduce kmalloc_obj() and family

[Kees Cook <kees@kernel.org>]

On Fri, Jul 19, 2024 at 08:50:41PM -0700, David Rientjes wrote:
> On Fri, 19 Jul 2024, Kees Cook wrote:
> 
> > diff --git a/include/linux/slab.h b/include/linux/slab.h
> > index 7247e217e21b..3817554f2d51 100644
> > --- a/include/linux/slab.h
> > +++ b/include/linux/slab.h
> > @@ -665,6 +665,44 @@ static __always_inline __alloc_size(1) void *kmalloc_noprof(size_t size, gfp_t f
> >  }
> >  #define kmalloc(...)				alloc_hooks(kmalloc_noprof(__VA_ARGS__))
> >  
> > +#define __alloc_obj3(ALLOC, P, COUNT, FLAGS)			\
> > +({								\
> > +	size_t __obj_size = size_mul(sizeof(*P), COUNT);	\
> > +	void *__obj_ptr;					\
> > +	(P) = __obj_ptr = ALLOC(__obj_size, FLAGS);		\
> > +	if (!__obj_ptr)						\
> > +		__obj_size = 0;					\
> > +	__obj_size;						\
> > +})
> > +
> > +#define __alloc_obj2(ALLOC, P, FLAGS)	__alloc_obj3(ALLOC, P, 1, FLAGS)
> > +
> > +#define __alloc_obj4(ALLOC, P, FAM, COUNT, FLAGS)		\
> > +({								\
> > +	size_t __obj_size = struct_size(P, FAM, COUNT);		\
> > +	void *__obj_ptr;					\
> > +	(P) = __obj_ptr = ALLOC(__obj_size, FLAGS);		\
> > +	if (!__obj_ptr)						\
> > +		__obj_size = 0;					\
> > +	__obj_size;						\
> > +})
> > +
> > +#define kmalloc_obj(...)					\
> > +	CONCATENATE(__alloc_obj,				\
> > +		    COUNT_ARGS(__VA_ARGS__))(kmalloc, __VA_ARGS__)
> > +
> > +#define kzalloc_obj(...)					\
> > +	CONCATENATE(__alloc_obj,				\
> > +		    COUNT_ARGS(__VA_ARGS__))(kzalloc, __VA_ARGS__)
> > +
> > +#define kvmalloc_obj(...)					\
> > +	CONCATENATE(__alloc_obj,				\
> > +		    COUNT_ARGS(__VA_ARGS__))(kvmalloc, __VA_ARGS__)
> > +
> > +#define kvzalloc_obj(...)					\
> > +	CONCATENATE(__alloc_obj,				\
> > +		    COUNT_ARGS(__VA_ARGS__))(kvzalloc, __VA_ARGS__)
> > +
> >  static __always_inline __alloc_size(1) void *kmalloc_node_noprof(size_t size, gfp_t flags, int node)
> >  {
> >  	if (__builtin_constant_p(size) && size) {
> 
> I'm supportive of this especially because it will pave a pathway toward 
> future hardening work.  Request: could we get an addition to 

Thanks!

> Documentation/ that explains how common idioms today can be converted to 
> these new macros for future users?  The above makes sense only when 
> accompanied by your commit description :)

Oh, yes. Very good point! I will figure out a place to add this. I'm not
sure if kerndoc would be best here.

-- 
Kees Cook