Re: [PATCH] slab: deal with NULL pointers passed to kmem_cache_free

[Matt Mackall <mpm@selenic.com>]

On Mon, Mar 19, 2007 at 02:16:01PM -0700, Christoph Lameter wrote:
> On Mon, 19 Mar 2007, Matt Mackall wrote:
> 
> > I think this sort of thing should work:
> > 
> > a = kmalloc(...)
> > b = kmem_cache_alloc(..)
> > c = allocate_some_id(...)
> > if (!a || !b || !c) {
> >    free_some_id(c)
> >    kmem_cache_free(c)
> 
> ^^^^ this requires the specification of a kmem_cache structure and the 
> object must be allocated by that cache.

Yes, omitted for brevity.

> >    kfree(a);
> 
> Here we dynamically determine the slab cache and do not verify even which 
> slab it came from.

That's an implementation detail that you shouldn't rely on and that
SLOB in fact breaks. kfree(kmem_cache_alloc(...)) is bad style to the
point of being a bug.

In my opinion:

xxxfree(xxxalloc(...)); /* should always work, even if allocation fails */
yyyfree(xxxalloc(...)); /* should never be expected to work */

-- 
Mathematics is the supreme nostalgia of our time.