Re: [PATCHSET 0/2] Fix splice from random/urandom

[Jens Axboe <axboe@kernel.dk>]

On 5/19/22 6:48 PM, Jason A. Donenfeld wrote:
> Hi Jens,
> 
> On Thu, May 19, 2022 at 06:02:55PM -0600, Jens Axboe wrote:
>> On 5/19/22 6:00 PM, Jason A. Donenfeld wrote:
>>> Hi Jens,
>>>
>>> On Fri, May 20, 2022 at 1:57 AM Jens Axboe <axboe@kernel.dk> wrote:
>>>>
>>>> On 5/19/22 5:27 PM, Jason A. Donenfeld wrote:
>>>>> Hi Jens,
>>>>>
>>>>> On Fri, May 20, 2022 at 1:25 AM Jens Axboe <axboe@kernel.dk> wrote:
>>>>>> I'll leave that to you :-)
>>>>>
>>>>> Alright, I'll have a look. Which one do I want here? This series adds
>>>>> splice_read/splice_write. The new thing would be sendpage? Or
>>>>> copy_file_range? Or something else?
>>>>
>>>> For copying kernel memory? It's really the same thing, you just
>>>> initialize the iter as an ITER_KVEC instead. The nice thing about the
>>>> iov_iter iterator that it then hides that for you, call the same copy
>>>> in/out helpers for it.
>>>
>>> Err, maybe we're talking about different things? I was thinking about
>>> the plumbing to make splice/sendfile work on non-pipes.
>>
>> Ah I see. sendfile() just uses splice() internally, so that'll work
>> (again) with my changes. splice(), by definition, either moves to and
>> from a pipe. Hence one of the fds must be a pipe. Does that answer the
>> question?
> 
> sendfile() returns -EINVAL even with your patches. Only splicing to pipes
> seems to work.

Huh, that really should work. Are you trying to sendfile() to random? If
so, you need that last write_iter patch too, and add the splice_write as
I mentioned.

-- 
Jens Axboe