Re: [PATCH v3] iomap: follow the alignment requirement for iomap_dio_hole_iter()

在 2026/8/4 22:31, Christoph Hellwig 写道:
> On Sat, Aug 01, 2026 at 07:17:26AM +0930, Qu Wenruo wrote:
>> +++ b/fs/iomap/direct-io.c
>> @@ -409,7 +409,10 @@ static inline unsigned int iomap_dio_alignment(struct inode *inode,
>>   {
>>   	if (dio_flags & IOMAP_DIO_FSBLOCK_ALIGNED)
>>   		return i_blocksize(inode);
>> -	return bdev_logical_block_size(bdev);
>> +	/* @bdev can be NULL for hole cases. */
>> +	if (bdev)
>> +		return bdev_logical_block_size(bdev);
>> +	return SECTOR_SIZE;
> 
> No.  As said for previous versions you should not use
> iomap_dio_alignment.  There is no such things as sub-fsblock alignment
> for holes.  All mappings need to be aligned on fs blocks, you can do
> sub-mapping granularity I/O in it, but not align on it.  So don't
> call iomap_dio_alignment, but instead align down the shortened range
> from the fault to a fs block boundary.

OK, got it now.

> 
> And really sort out the btrfs locking that we can't get rid of this
> premature faulting as it keeps messing up core code way too much.

That will be a long term work, meanwhile I'm only fixing a short-term 
bug, already affecting existing bs > ps support.

I hate the btrfs internal locking mechanism just as you, but it won't be 
an easy thing to do in the short term.

Thanks,
Qu