Re: [PATCH v14 29/42] btrfs: introduce dedicated data write path for ZONED mode

[David Sterba <dsterba@suse.cz>]

On Tue, Jan 26, 2021 at 11:25:07AM +0900, Naohiro Aota wrote:
> If more than one IO is issued for one file extent, these IO can be written
> to separate regions on a device. Since we cannot map one file extent to
> such a separate area, we need to follow the "one IO == one ordered extent"
> rule.
> 
> The Normal buffered, uncompressed, not pre-allocated write path (used by
> cow_file_range()) sometimes does not follow this rule. It can write a part
> of an ordered extent when specified a region to write e.g., when its
> called from fdatasync().
> 
> Introduces a dedicated (uncompressed buffered) data write path for ZONED
> mode. This write path will CoW the region and write it at once.
> 
> Reviewed-by: Josef Bacik <josef@toxicpanda.com>
> Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
> ---
>  fs/btrfs/inode.c | 34 ++++++++++++++++++++++++++++++++--
>  1 file changed, 32 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> index a9bf78eaed42..6d43aaa1f537 100644
> --- a/fs/btrfs/inode.c
> +++ b/fs/btrfs/inode.c
> @@ -1400,6 +1400,29 @@ static int cow_file_range_async(struct btrfs_inode *inode,
>  	return 0;
>  }
>  
> +static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
> +				       struct page *locked_page, u64 start,
> +				       u64 end, int *page_started,
> +				       unsigned long *nr_written)
> +{
> +	int ret;
> +
> +	ret = cow_file_range(inode, locked_page, start, end,
> +			     page_started, nr_written, 0);
> +	if (ret)
> +		return ret;
> +
> +	if (*page_started)
> +		return 0;
> +
> +	__set_page_dirty_nobuffers(locked_page);
> +	account_page_redirty(locked_page);
> +	extent_write_locked_range(&inode->vfs_inode, start, end, WB_SYNC_ALL);
> +	*page_started = 1;
> +
> +	return 0;
> +}
> +
>  static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
>  					u64 bytenr, u64 num_bytes)
>  {
> @@ -1879,17 +1902,24 @@ int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page
>  {
>  	int ret;
>  	int force_cow = need_force_cow(inode, start, end);
> +	const bool do_compress = inode_can_compress(inode) &&
> +		inode_need_compress(inode, start, end);

This would make sense to cache the values, but inode_need_compress is
quite heavy as it runs the compression heuristic. This would affect all
cases and drop some perf.

> +	const bool zoned = btrfs_is_zoned(inode->root->fs_info);
>  
>  	if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
> +		ASSERT(!zoned);
>  		ret = run_delalloc_nocow(inode, locked_page, start, end,
>  					 page_started, 1, nr_written);
>  	} else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
> +		ASSERT(!zoned);
>  		ret = run_delalloc_nocow(inode, locked_page, start, end,
>  					 page_started, 0, nr_written);
> -	} else if (!inode_can_compress(inode) ||
> -		   !inode_need_compress(inode, start, end)) {
> +	} else if (!do_compress && !zoned) {
>  		ret = cow_file_range(inode, locked_page, start, end,
>  				     page_started, nr_written, 1);
> +	} else if (!do_compress && zoned) {
> +		ret = run_delalloc_zoned(inode, locked_page, start, end,
> +					 page_started, nr_written);

The part of the condition is shared so it should be structured lik

	} else if (!<the compression checks>) {
		if (zoned)
			run_delalloc_zoned
		else
			cow_file_range
	} ...

>  	} else {
>  		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
>  		ret = cow_file_range_async(inode, wbc, locked_page, start, end,
> -- 
> 2.27.0