Re: [PATCH v3 1/4] mm/mlock: return EINVAL if len overflows for mlock/munlock

From: mawupeng
Date: Sun Feb 05 2023 - 19:48:28 EST




On 2023/2/4 1:14, David Hildenbrand wrote:
> On 28.01.23 07:32, Wupeng Ma wrote:
>> From: Ma Wupeng <mawupeng1@xxxxxxxxxx>
>>
>> While testing mlock, we have a problem if the len of mlock is ULONG_MAX.
>> The return value of mlock is zero. But nothing will be locked since the
>> len in do_mlock overflows to zero due to the following code in mlock:
>>
>>    len = PAGE_ALIGN(len + (offset_in_page(start)));
>>
>> The same problem happens in munlock.
>>
>> Add new check and return -EINVAL to fix this overflowing scenarios since
>> they are absolutely wrong.
>>
>> Return 0 early to avoid burn a bunch of cpu cycles if len == 0.
>>
>> Signed-off-by: Ma Wupeng <mawupeng1@xxxxxxxxxx>
>> ---
>>   mm/mlock.c | 14 ++++++++++++--
>>   1 file changed, 12 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/mlock.c b/mm/mlock.c
>> index 7032f6dd0ce1..eb09968ba27f 100644
>> --- a/mm/mlock.c
>> +++ b/mm/mlock.c
>> @@ -478,8 +478,6 @@ static int apply_vma_lock_flags(unsigned long start, size_t len,
>>       end = start + len;
>>       if (end < start)
>>           return -EINVAL;
>> -    if (end == start)
>> -        return 0;
>>       vma = mas_walk(&mas);
>>       if (!vma)
>>           return -ENOMEM;
>> @@ -575,7 +573,13 @@ static __must_check int do_mlock(unsigned long start, size_t len, vm_flags_t fla
>>       if (!can_do_mlock())
>>           return -EPERM;
>>   +    if (!len)
>> +        return 0;
>> +
>>       len = PAGE_ALIGN(len + (offset_in_page(start)));
>> +    if (!len)
>> +        return -EINVAL;
>> +
>>       start &= PAGE_MASK;
>
> The "ordinary" overflows are detected in apply_vma_lock_flags(), correct?

Overflow is not checked anywhere however the ordinary return early if len == 0 is detected in apply_vma_lock_flags().

do_mlock
apply_vma_lock_flags
end = start + len;
if (end == start)
return 0;

Move the checking to the begin is easier to detect overflows and make the logic clearer
and avoid burn a bunch of cpu cycles.

>