Re: UBSAN: Undefined behaviour in mm/page_alloc.c

[Vlastimil Babka]

On 11/14/18 12:32 AM, Andrew Morton wrote:
> On Wed, 14 Nov 2018 00:23:28 +0100 Vlastimil Babka <vbabka@xxxxxxx> wrote:
> 
>> On 11/14/18 12:15 AM, Andrew Morton wrote:
>>> On Tue, 13 Nov 2018 10:43:05 +0100 Michal Hocko <mhocko@xxxxxxxxxx> wrote:
>>>
>>>> --- a/mm/page_alloc.c
>>>> +++ b/mm/page_alloc.c
>>>> @@ -4364,6 +4353,15 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
>>>>  	gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
>>>>  	struct alloc_context ac = { };
>>>>  
>>>> +	/*
>>>> +	 * There are several places where we assume that the order value is sane
>>>> +	 * so bail out early if the request is out of bound.
>>>> +	 */
>>>> +	if (unlikely(order >= MAX_ORDER)) {
>>>> +		WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
>>>> +		return NULL;
>>>> +	}
>>>> +
>>>
>>> I know "everybody enables CONFIG_DEBUG_VM", but given this is fastpath,
>>> we could help those who choose not to enable it by using
>>>
>>> #ifdef CONFIG_DEBUG_VM
>>> 	if (WARN_ON_ONCE(order >= MAX_ORDER && !(gfp_mask & __GFP_NOWARN)))
>>> 		return NULL;
>>> #endif
>>
>> Hmm, but that would mean there's still potential undefined behavior for
>> !CONFIG_DEBUG_VM, so I would prefer not to do it like that.
>>
> 
> What does "potential undefined behavior" mean here?

I mean that it becomes undefined once a caller with order >= MAX_ORDER
appears. Worse if it's directly due to a userspace action, like in this
case.