Re: [PATCH 1/1] mm/slab_common: reject zero object_size before calculate_alignment

[Longlong Xia]

在 2026/8/25 20:59, Hao Li 写道:
> On Mon, Aug 24, 2026 at 05:24:54PM +0800, Longlong Xia wrote:
> > From: Longlong Xia <xialonglong@xxxxxxxxxx>
> >
> > calculate_alignment() with SLAB_HWCACHE_ALIGN halves ralign in a
> > while (size <= ralign / 2) loop. When size is 0, ralign eventually
> > reaches 0 and the condition stays true indefinitely, hanging the
> > kernel.
> >
> > kmem_cache_sanity_check() rejected size > KMALLOC_MAX_SIZE but not
> > size == 0, and the sanity check is compiled out without
> > CONFIG_DEBUG_VM. Add a !object_size check in
> > __kmem_cache_create_args(), which is always compiled, and add !size
> > to the DEBUG_VM sanity check for diagnostics.
> >
> > Assisted-by: Codex:gpt-5.6-sol
> > Signed-off-by: Longlong Xia <xialonglong@xxxxxxxxxx>
> > ---
> >   mm/slab_common.c | 4 ++--
> >   1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/mm/slab_common.c b/mm/slab_common.c
> > index 657fd75776ea..209fe838fe71 100644
> > --- a/mm/slab_common.c
> > +++ b/mm/slab_common.c
> > @@ -102,7 +102,7 @@ static bool kmem_cache_is_duplicate_name(const char *name)
> >   
> >   static int kmem_cache_sanity_check(const char *name, unsigned int size)
> >   {
> > -	if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
> > +	if (!name || in_interrupt() || !size || size > KMALLOC_MAX_SIZE) {
> >   		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
> >   		return -EINVAL;
> >   	}
> > @@ -354,7 +354,7 @@ struct kmem_cache *__kmem_cache_create_args(const char *name,
> >   		goto out_unlock;
> >   	}
> >   
> > -	if (flags & ~SLAB_FLAGS_PERMITTED) {
> > +	if (!object_size || flags & ~SLAB_FLAGS_PERMITTED) {
> under !CONFIG_DEBUG_VM, I think we ignore the sanity_check deliberately, so it
> seems we don't need to add !object_size check here. right?

Thanks for the review.

Right, agreed. The !object_size check duplicates what 
kmem_cache_sanity_check()
is for, and that one is intentionally debug-only. I'll drop the
__kmem_cache_create_args() hunk in v2 and keep only the !size addition in
kmem_cache_sanity_check().


Thanks,

Longlong

> >   		err = -EINVAL;
> >   		goto out_unlock;
> >   	}
> > --
> > 2.43.0