Re: [PATCH v1] mm/gup: fix FOLL_FORCE COW security issue and remove FOLL_COW

[David Hildenbrand]

On 09.08.22 22:14, Linus Torvalds wrote:
> On Tue, Aug 9, 2022 at 1:07 PM David Hildenbrand <david@xxxxxxxxxx> wrote:
>>
>>>         /* But FOLL_FORCE has no effect on shared mappings */
>>>         if (vma->vm_flags & MAP_SHARED)
>>>                 return false;
>>
>> I'd actually rather check for MAP_MAYSHARE here, which is even stronger.
>> Thoughts?
> 
> Hmm. Adding the test for both is basically free (all those vm_flags
> checks end up being a bit mask and compare), so no objections.
> 
> For some reason I though VM_SHARED and VM_MAYSHARE end up always being
> the same bits, and it was a mistake to make them two bits in the first
> place (unlike the read/write/exec bits that can are about mprotect),
> 
> But as there are two bits, I'm sure somebody ends up touching one and
> not the other.
> 
> So yeah, I don't see any downside to just checking both bits.
> 
> [ That said, is_cow_mapping() only checks VM_SHARED, so if they are
> ever different, that's a potential source of confusion ]

IIUC VM_MAYSHARE is always set in a MAP_SHARED mapping, but for file
mappings we only set VM_SHARED if the file allows for writes (and we can
set VM_MAYWRITE or even VM_WRITE). [don't ask me why, it's a steady
source of confusion]

That's why is_cow_mapping() works even due to the weird MAP_SHARED vs.
VM_MAYSHARE logic.


I'd actually only check for VM_MAYSHARE here, which implies MAP_SHARED.
If someone would ever mess that up, I guess we'd be in bigger trouble.

But whatever you prefer.

-- 
Thanks,

David / dhildenb