Re: [PATCH 3/3] io_uring: Don't read userspace data in io_probe

[Jens Axboe <axboe@kernel.dk>]

On 6/19/24 8:55 AM, Gabriel Krisman Bertazi wrote:
> Jens Axboe <axboe@kernel.dk> writes:
> 
>> On 6/18/24 8:06 PM, Gabriel Krisman Bertazi wrote:
>>> We don't need to read the userspace buffer, and the kernel side is
>>> expected to write over it anyway.  Perhaps this was meant to allow
>>> expansion of the interface for future parameters?  If we ever need to do
>>> it, perhaps it should be done as a new io_uring opcode.
>>
>> Right, it's checked so that we could use it for input values in the
>> future. By ensuring that userspace must zero it, then we could add input
>> values and flags in the future.
>>
>> Is there a good reason to make this separate change? If not, I'd say
>> drop it and we can always discuss when there's an actual need to do so.
>> At least we have the option of passing in some information with the
>> current code, in a backwards compatible fashion.
> 
> There is no reason other than it is unused.  I'm fine with dropping it.
> 
> I'll wait for feedback on the other patches and, if we need a new
> iteration, I'll skip this one.

I think the rest look fine. The unsupported part was mostly a thing for
backports, did use it myself internally once for the meta kernel. But I
do think we should just kill it, so fine with doing that.

-- 
Jens Axboe