Re: [PATCH] mm/slab: reject unsupported kmalloc sizes

[David Laight <david.laight.linux@gmail.com>]

On Wed, 26 Aug 2026 11:28:12 +0200
"Vlastimil Babka (SUSE)" <vbabka@kernel.org> wrote:

> On 8/17/26 22:40, Zi Yan wrote:
> > kmalloc is used to allocate physically contiguous memory for kernel
> > allocations. For requests larger than KMALLOC_MAX_CACHE_SIZE, kmalloc uses
> > the page allocator and can only support up to KMALLOC_MAX_SIZE. For request
> > sizes bigger than KMALLOC_MAX_SIZE, the page allocator can emit a WARN
> > because kmalloc allocates an order greater than MAX_PAGE_ORDER. Systems
> > with panic_on_warn=1 crash because of this WARN. Fix it by rejecting any
> > kmalloc size bigger than KMALLOC_MAX_SIZE.
> > 
> > Fixes: aadb4bc4a1f9 ("SLUB: direct pass through of page size or higher kmalloc requests")
> > Reported-by: syzbot+805630f1453e490427fa@syzkaller.appspotmail.com
> > Closes: https://lore.kernel.org/all/6a820ebc.9ebadd4d.20b15e.001b.GAE@google.com/
> > Tested-by: syzbot+805630f1453e490427fa@syzkaller.appspotmail.com
> > Signed-off-by: Zi Yan <ziy@nvidia.com>
> > Cc: stable@vger.kernel.org
> > ---
> > It fixes a page allocator warning (order > MAX_PAGE_ORDER) when gadgetfs
> > requests excessively large memory from kmalloc. Instead of adding
> > __GFP_NOWARN to suppress the warning, as was done for usbfs[1], change
> > kmalloc to return NULL without a warning for this specific issue.
> > 
> > [1] commit 4f2629ea67e72 ("USB: usbfs: Don't WARN about excessively large memory allocations")  
> 
> So I checked and for kvmalloc() we have in __kvmalloc_node_noprof()
> 
>         /* Don't even allow crazy sizes */
>         if (unlikely(size > INT_MAX)) {
>                 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
>                 return NULL;
>         }
> 
> This comes from Linus in commit 7661809d493b4. I'd do the same thing here
> then. Thus there would be a useful warning for e.g. development mistakes
> resulting in the size to be unexpectedly high.
> 
> Callers passing size that comes from userspace or similar untrusted source
> can either pass __GFP_NOWARN or sanitize the size to what they expect to be
> sane (which is context dependent and I assume actually way lower than
> kmalloc limits in practice). Note that passing even sizes within but close
> to/at the limit, trusting blindly some external source, can succeed the
> allocations but effectively DoS the system with heavy reclaim/compaction. So
> caller sanitization should still be preferred IMHO.

Indeed, and sanitising the values early on saves all the size_add() and
size_mult() operations (that are just saturating maths) but still let
through the 'DoS the system' sizes.
Mostly the actual maximum size is actually small. I suspect limits like
64k, 1M or 16M would be appropriate.

David

> Some of the recent
> arguments from Linus [1] would apply to this too, I think.
> 
> [1]
> https://lore.kernel.org/all/CAHk-=wiSmgwwLKCqJwGS-dVHnSLU8W+7q1UQq-G9=TBGGZbuhQ@mail.gmail.com/
> 
> > ---
> >  mm/slub.c | 7 ++++++-
> >  1 file changed, 6 insertions(+), 1 deletion(-)
> > 
> > diff --git a/mm/slub.c b/mm/slub.c
> > index 0337e60db5ace..a3071f4ef1945 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -5263,7 +5263,12 @@ static void *___kmalloc_large_node(size_t size, gfp_t flags, int node)
> >  {
> >  	struct page *page;
> >  	void *ptr = NULL;
> > -	unsigned int order = get_order(size);
> > +	unsigned int order;
> > +
> > +	if (size > KMALLOC_MAX_SIZE)
> > +		return NULL;
> > +
> > +	order = get_order(size);
> >  
> >  	if (unlikely(flags & GFP_SLAB_BUG_MASK))
> >  		flags = kmalloc_fix_flags(flags);
> > 
> > ---
> > base-commit: 8d3ae59288f1e7d58d76558a6ee96d533bc5019f
> > change-id: 20260817-limit_kmalloc_size-3a4a2c73beac
> > 
> > Best regards,
> > --  
> > Yan, Zi
> >   
> 
>