On 11/11/19 10:57 AM, Dmitry Vyukov wrote:
> On Fri, Nov 8, 2019 at 11:32 PM Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx> wrote:
>>> diff --git a/mm/kasan/generic_report.c b/mm/kasan/generic_report.c
>>> index 36c645939bc9..52a92c7db697 100644
>>> --- a/mm/kasan/generic_report.c
>>> +++ b/mm/kasan/generic_report.c
>>> @@ -107,6 +107,24 @@ static const char *get_wild_bug_type(struct kasan_access_info *info)
>>>
>>> const char *get_bug_type(struct kasan_access_info *info)
>>> {
>>> + /*
>>> + * If access_size is negative numbers, then it has three reasons
>>> + * to be defined as heap-out-of-bounds bug type.
>>> + * 1) Casting negative numbers to size_t would indeed turn up as
>>> + * a large size_t and its value will be larger than ULONG_MAX/2,
>>> + * so that this can qualify as out-of-bounds.
>>> + * 2) If KASAN has new bug type and user-space passes negative size,
>>> + * then there are duplicate reports. So don't produce new bug type
>>> + * in order to prevent duplicate reports by some systems
>>> + * (e.g. syzbot) to report the same bug twice.
>>> + * 3) When size is negative numbers, it may be passed from user-space.
>>> + * So we always print heap-out-of-bounds in order to prevent that
>>> + * kernel-space and user-space have the same bug but have duplicate
>>> + * reports.
>>> + */
>>
>> Completely fail to understand 2) and 3). 2) talks something about *NOT* producing new bug
>> type, but at the same time you code actually does that.
>> 3) says something about user-space which have nothing to do with kasan.
>
> The idea was to use one of the existing bug titles so that syzbot does
> not produce 2 versions for OOBs where size is user-controlled. We
> don't know if it's overflow from heap, global or stack, but heap is
> the most common bug, so saying heap overflow will reduce chances of
> producing duplicates the most.
> But for all of this to work we do need to use one of the existing bug titles.
The "heap-out-of-bounds" is not one of the existing bug titles.
