Re: [PATCH v4 2/6] io_uring/mock: add cmd using vectored regbufs

[Pavel Begunkov <asml.silence@gmail.com>]

On 5/30/25 15:37, Jens Axboe wrote:
> On 5/30/25 7:40 AM, Pavel Begunkov wrote:
>> On 5/30/25 14:25, Jens Axboe wrote:
>>> On 5/30/25 6:51 AM, Pavel Begunkov wrote:
>>>> +static int io_copy_regbuf(struct iov_iter *reg_iter, void __user *ubuf)
>>>> +{
>>>> +    size_t ret, copied = 0;
>>>> +    size_t buflen = PAGE_SIZE;
>>>> +    void *tmp_buf;
>>>> +
>>>> +    tmp_buf = kzalloc(buflen, GFP_KERNEL);
>>>> +    if (!tmp_buf)
>>>> +        return -ENOMEM;
>>>> +
>>>> +    while (iov_iter_count(reg_iter)) {
>>>> +        size_t len = min(iov_iter_count(reg_iter), buflen);
>>>> +
>>>> +        if (iov_iter_rw(reg_iter) == ITER_SOURCE) {
>>>> +            ret = copy_from_iter(tmp_buf, len, reg_iter);
>>>> +            if (ret <= 0)
>>>> +                break;
>>>> +            if (copy_to_user(ubuf, tmp_buf, ret))
>>>> +                break;
>>>> +        } else {
>>>> +            if (copy_from_user(tmp_buf, ubuf, len))
>>>> +                break;
>>>> +            ret = copy_to_iter(tmp_buf, len, reg_iter);
>>>> +            if (ret <= 0)
>>>> +                break;
>>>> +        }
>>>
>>> Do copy_{to,from}_iter() not follow the same "bytes not copied" return
>>> value that the copy_{to,from}_user() do? From a quick look, looks like
>>> they do.
>>>
>>> Minor thing, no need for a respin just for that.
>>
>> One returns 0 on success the other the number of processed bytes.
> 
> copy_{to,from}_user() returns bytes NOT processed, and I guess the iter

Sure, it doesn't contradict it, they follow different semantics.

> versions return bytes processed. Guess the code is fine, it's more so
> the API that's a bit wonky on the copy/iter side.

Which API? This command or copy helpers? copy_{to,from}_user are used
here in the way they're always used in the kernel, and that's fine,
they're not supposed to fail for valid input. That's unlike iter
helpers, which may return a partial result.

-- 
Pavel Begunkov