[PATCH v17 08/10] PM: hibernate: disable when there are active secretmem users

[David Hildenbrand]

On 08.02.21 12:14, David Hildenbrand wrote:
> On 08.02.21 12:13, David Hildenbrand wrote:
>> On 08.02.21 11:57, Michal Hocko wrote:
>>> On Mon 08-02-21 11:53:58, David Hildenbrand wrote:
>>>> On 08.02.21 11:51, Michal Hocko wrote:
>>>>> On Mon 08-02-21 11:32:11, David Hildenbrand wrote:
>>>>>> On 08.02.21 11:18, Michal Hocko wrote:
>>>>>>> On Mon 08-02-21 10:49:18, Mike Rapoport wrote:
>>>>>>>> From: Mike Rapoport <rppt at linux.ibm.com>
>>>>>>>>
>>>>>>>> It is unsafe to allow saving of secretmem areas to the hibernation
>>>>>>>> snapshot as they would be visible after the resume and this essentially
>>>>>>>> will defeat the purpose of secret memory mappings.
>>>>>>>>
>>>>>>>> Prevent hibernation whenever there are active secret memory users.
>>>>>>>
>>>>>>> Does this feature need any special handling? As it is effectivelly
>>>>>>> unevictable memory then it should behave the same as other mlock, ramfs
>>>>>>> which should already disable hibernation as those cannot be swapped out,
>>>>>>> no?
>>>>>>>
>>>>>>
>>>>>> Why should unevictable memory not go to swap when hibernating? We're merely
>>>>>> dumping all of our system RAM (including any unmovable allocations) to swap
>>>>>> storage and the system is essentially completely halted.
>>>>>>
>>>>> My understanding is that mlock is never really made visible via swap
>>>>> storage.
>>>>
>>>> "Using swap storage for hibernation" and "swapping at runtime" are two
>>>> different things. I might be wrong, though.
>>>
>>> Well, mlock is certainly used to keep sensitive information, not only to
>>> protect from major/minor faults.
>>>
>>
>> I think you're right in theory, the man page mentions "Cryptographic
>> security software often handles critical bytes like passwords or secret
>> keys as data structures" ...
>>
>> however, I am not aware of any such swap handling and wasn't able to
>> spot it quickly. Let me take a closer look.
> 
> s/swap/hibernate/

My F33 system happily hibernates to disk, even with an application that 
succeeded in din doing an mlockall().

And it somewhat makes sense. Even my freshly-booted, idle F33 has

$ cat /proc/meminfo  | grep lock
Mlocked:            4860 kB

So, stopping to hibernate with mlocked memory would essentially prohibit 
any modern Linux distro to hibernate ever.

-- 
Thanks,

David / dhildenb