Re: [RFC 1/2] io_uring: disallow overlapping ranges for buffer registration

[Jens Axboe <axboe@kernel.dk>]

On 6/12/20 12:22 PM, Bijan Mottahedeh wrote:
> On 6/12/2020 8:16 AM, Jens Axboe wrote:
>> On 6/11/20 8:23 PM, Bijan Mottahedeh wrote:
>>> Buffer registration is expensive in terms of cpu/mem overhead and there
>>> seems no good reason to allow overlapping ranges.
>> There's also no good reason to disallow it imho, so not sure we should
>> be doing that.
>>
> 
> My concern was about a malicious user without CAP_IPC_LOCK abusing its 
> memlock limit and repeatedly register the same buffer as that would tie 
> up cpu/mem resources to pin up to 1TB of memory, but maybe the argument 
> is that the user should have not been granted that large of memlock 
> limit?  Also, without any restrictions, there are a huge number of ways 
> overlapping ranges could be specified, creating a very large validation 
> matrix.  What the use cases for that flexibility are though, I don't know.

Not sure I follow, so I must be missing something. We're accounting each
buffer as it is, so how are we abusing the limit?

My concern on the overlaps is mainly that someone could be
(inadvertently, perhaps) doing it already, and now we're breaking that.
Or maybe that are doing it purposely, for some reason I just don't know
about.

-- 
Jens Axboe