Re: Fwd: mm/page_alloc.c:4453 with cfg80211_wiphy_work [cfg80211]

[Vlastimil Babka]

On 7/16/23 13:28, Matthew Wilcox wrote:
> On Sun, Jul 16, 2023 at 06:10:44PM +0700, Bagas Sanjaya wrote:
>> Hi,
>> 
>> I notice a regression report on Bugzilla [1]. Quoting from it:
> 
> Maybe you could try doing some work on this bug before just spamming
> people with it?
> 
>         if (WARN_ON_ONCE_GFP(order > MAX_ORDER, gfp))
>                 return NULL;
> 
> This is the page allocator telling the caller that they've asked for an
> unreasonably large allocation.
> 
> Now, this bug is actually interesting to the MM because the caller
> called kmalloc() with a ridiculous size.  Arguable kmalloc should
> protect callers from themselves (alloc_pages() is more low level
> and can presume its users know what they're doing).
> 
> Vlastimil, what do you think?  Something like ...

Hmm should be more robust to check size > KMALLOC_MAX_SIZE before even doing
get_order(size). Ultimately it checks the same limit.
But I'm unsure about just returning NULL. I think warn_on_once might be
useful even there - in case a bug is introduced/exposed, even a
inexperienced user will be easily able to report sufficient information wich
a WARN and its stacktrace, even if the callsite's alloc check doesn't
provide it in an obvious way?

> +++ b/mm/slab_common.c
> @@ -1119,6 +1119,8 @@ static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
>         void *ptr = NULL;
>         unsigned int order = get_order(size);
> 
> +       if (order > MAX_ORDER)
> +               return NULL;
>         if (unlikely(flags & GFP_SLAB_BUG_MASK))
>                 flags = kmalloc_fix_flags(flags);
> 
> 
>