Re: [PATCH 4/7] btrfs: Use iomap_dio_rw() for direct I/O

[Christoph Hellwig <hch@infradead.org>]

On Fri, Nov 15, 2019 at 10:16:57AM -0600, Goldwyn Rodrigues wrote:
> From: Goldwyn Rodrigues <rgoldwyn@suse.com>
> 
> This is the main patch to switch call from
> __blockdev_direct_IO() to iomap_dio_rw(). In this patch:
> 
> Removed buffer_head references
> Removed inode_dio_begin() and inode_dio_end() functions since
> they are called in iomap_dio_rw().
> Renamed btrfs_get_blocks_direct() to direct_iomap_begin() and
> used it as iomap_begin()
> address_space.direct_IO now is a noop since direct_IO is called
> from __btrfs_write_direct().
> 
> Removed flags parameter used for __blockdev_direct_IO(). iomap is
> capable of direct I/O reads from a hole, so we don't need to
> return -ENOENT.

There isn't really any need to describe the low-level changes,
but more what this changes at a high level, and more importantly
the reasons for that.

>  static int btrfs_get_blocks_direct_write(struct extent_map **map,
> -					 struct buffer_head *bh_result,
>  					 struct inode *inode,
>  					 struct btrfs_dio_data *dio_data,
>  					 u64 start, u64 len)

Should this function be renamed as well? btrfs_iomap_begin_write?

> +static int direct_iomap_begin(struct inode *inode, loff_t start,
> +		loff_t length, unsigned flags, struct iomap *iomap,
> +		struct iomap *srcmap)

This needs a btrfs_ prefix.

> +	if ((em->block_start == EXTENT_MAP_HOLE) ||

No need for the inner braces.

> -	dio_end_io(dio_bio);

You removed the only users of dio_end_io and the submit hook in the
old dio code.  Please add a patch to remove those at the end of the
series.

> -				   btrfs_submit_direct, flags);
> +	ret = iomap_dio_rw(iocb, iter, &dio_iomap_ops, NULL, is_sync_kiocb(iocb));

This adds a line > 80 chars.