Re: [PATCH 2/2] iomap: remove the atomic write length check in iomap_dio_bio_iter_one()

[Tal Zussman]

On 11/9/26 1:44 PM, John Garry wrote:
>> 
>>> This looks ok. But I am just thinking that a CoW-based atomic write does
>>> not have the single bio restriction, so we don't need to fail the atomic
>>> in this case. However, we should have consistent rules for iovec's
>>> whether it's a block device or regular file for ext4 or xfs, so this
>>> makes sense.
>>>
>>> thanks
>>>
>>> BTW, you could be a real hero and add an xfstests case for this also.
>>>
>> 
>> What specifically would you want to see tested? I can do an atomic write with
>> a partially unmapped buffer, with and without stable writes, and check that it
>> fails and that nothing is written.
> 
> Just something similar to the blktests case which you added.
> 

Sounds good, I'll add it to the queue.

> BTW, have you tested this partially-unmapped case for CoW-based atomics
> on xfs? You can just do that by setting atomic_wr=0
> 

Yeah, I just tested it. It behaves as expected: returns -EFAULT with the file
unchanged, both with and without bouncing.

The single-bio check depends on REQ_ATOMIC, which only the HW atomic path sets,
so it doesn't apply to CoW writes, both before and after this series.

In terms of consistency, the unmapped buffer returns -EINVAL on the non-bounce
HW path (from bio_iov_iter_get_pages()) but -EFAULT for all other paths. That
should probably be made consistent across HW and SW and bounce vs. non-bounce,
but that's a separate change.

Thanks,
Tal


> Thanks,
> John
> 
>> 
>>>> ---
>>>>    fs/iomap/direct-io.c | 14 --------------
>>>>    1 file changed, 14 deletions(-)
>>>>
>>>> diff --git a/fs/iomap/direct-io.c b/fs/iomap/direct-io.c
>>>> index ad041ac3252a..aca7cbaea4af 100644
>>>> --- a/fs/iomap/direct-io.c
>>>> +++ b/fs/iomap/direct-io.c
>>>> @@ -370,15 +370,6 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
>>>>    		goto out_put_bio;
>>>>    	ret = bio->bi_iter.bi_size;
>>>>    
>>>> -	/*
>>>> -	 * An atomic write bio must cover the complete length.  If it doesn't,
>>>> -	 * error out.
>>>> -	 */
>>>> -	if ((op & REQ_ATOMIC) && WARN_ON_ONCE(ret != iomap_length(iter))) {
>>>> -		ret = -EINVAL;
>>>> -		goto out_bio_release_pages;
>>>> -	}
>>>> -
>>>>    	if (dio->flags & IOMAP_DIO_WRITE)
>>>>    		task_io_account_write(ret);
>>>>    	else if ((dio->flags & IOMAP_DIO_USER_BACKED) &&
>>>> @@ -393,11 +384,6 @@ static ssize_t iomap_dio_bio_iter_one(struct iomap_iter *iter,
>>>>    	iomap_dio_submit_bio(iter, dio, bio, pos);
>>>>    	return ret;
>>>>    
>>>> -out_bio_release_pages:
>>>> -	if (dio->flags & IOMAP_DIO_BOUNCE)
>>>> -		bio_free_folios(bio);
>>>> -	else
>>>> -		bio_release_pages(bio, false);
>>>>    out_put_bio:
>>>>    	bio_put(bio);
>>>>    	return ret;
>>>>
>>>
>> 
>