[RFC] btrfs: Don't return extent in fiemap if we meet with a hole.

[RFC] btrfs: Don't return extent in fiemap if we meet with a hole.

[Tao Ma]

Recently, my colleague Jeff tried to add fiemap support to cp(1).
http://www.mail-archive.com/bug-coreutils@gnu.org/msg19987.html

He just meet with a strange issue with following command:
dd if=/dev/null of=/btrfs/sparse bs=1 seek=4096
When we use fiemap to the file, btrfs returns an extent with len '4096'
and flag 'unwritten' while actually there is no data allocated.

I just dived into this and to my surprise, it is done by btrfs
intentionally. I checked other file systems which support fiemap.
Actually with the file created by the script, ocfs2, ext3/4 and
xfs all return zero extent. And according to the documentation file
Documentation/filesystems/fiemap.txt, FIEMAP_EXTENT_UNWRITTEN should
be used when the extent is allocated but it's data has not been
initialized. So I think btrfs should work like other filesystems.

Cc: Jeff Liu <jeff.liu@oracle.com>
Cc: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Tao Ma <tao.ma@oracle.com>
---
 fs/btrfs/extent_io.c |   15 +++++++++------
 1 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index b177ed3..7eb4d77 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -2951,7 +2951,7 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 	u32 flags = 0;
 	u64 disko = 0;
 	struct extent_map *em = NULL;
-	int end = 0;
+	int end = 0, hole = 0;
 	u64 em_start = 0, em_len = 0;
 	unsigned long emflags;
 	ret = 0;
@@ -2978,12 +2978,13 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 
 		disko = 0;
 		flags = 0;
+		hole = 0;
 
 		if (em->block_start == EXTENT_MAP_LAST_BYTE) {
 			end = 1;
 			flags |= FIEMAP_EXTENT_LAST;
 		} else if (em->block_start == EXTENT_MAP_HOLE) {
-			flags |= FIEMAP_EXTENT_UNWRITTEN;
+			hole = 1;
 		} else if (em->block_start == EXTENT_MAP_INLINE) {
 			flags |= (FIEMAP_EXTENT_DATA_INLINE |
 				  FIEMAP_EXTENT_NOT_ALIGNED);
@@ -3015,10 +3016,12 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 			end = 1;
 		}
 
-		ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
-					em_len, flags);
-		if (ret)
-			goto out_free;
+		if (!hole) {
+			ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
+						      em_len, flags);
+			if (ret)
+				goto out_free;
+		}
 	}
 out_free:
 	free_extent_map(em);
-- 
1.6.3.3.334.g916e1.dirty

Re: [RFC] btrfs: Don't return extent in fiemap if we meet with a hole.

[jeff.liu]

Tao Ma wrote:
> Recently, my colleague Jeff tried to add fiemap support to cp(1).
> http://www.mail-archive.com/bug-coreutils@gnu.org/msg19987.html
> 
> He just meet with a strange issue with following command:
> dd if=/dev/null of=/btrfs/sparse bs=1 seek=4096
> When we use fiemap to the file, btrfs returns an extent with len '4096'
> and flag 'unwritten' while actually there is no data allocated.
> 
> I just dived into this and to my surprise, it is done by btrfs
> intentionally. I checked other file systems which support fiemap.
> Actually with the file created by the script, ocfs2, ext3/4 and
> xfs all return zero extent. And according to the documentation file
> Documentation/filesystems/fiemap.txt, FIEMAP_EXTENT_UNWRITTEN should
> be used when the extent is allocated but it's data has not been
> initialized. So I think btrfs should work like other filesystems.
> 
> Cc: Jeff Liu <jeff.liu@oracle.com>
> Cc: Chris Mason <chris.mason@oracle.com>
> Signed-off-by: Tao Ma <tao.ma@oracle.com>
> ---
>  fs/btrfs/extent_io.c |   15 +++++++++------
>  1 files changed, 9 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
> index b177ed3..7eb4d77 100644
> --- a/fs/btrfs/extent_io.c
> +++ b/fs/btrfs/extent_io.c
> @@ -2951,7 +2951,7 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
>  	u32 flags = 0;
>  	u64 disko = 0;
>  	struct extent_map *em = NULL;
> -	int end = 0;
> +	int end = 0, hole = 0;
>  	u64 em_start = 0, em_len = 0;
>  	unsigned long emflags;
>  	ret = 0;
> @@ -2978,12 +2978,13 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
>  
>  		disko = 0;
>  		flags = 0;
> +		hole = 0;
>  
>  		if (em->block_start == EXTENT_MAP_LAST_BYTE) {
>  			end = 1;
>  			flags |= FIEMAP_EXTENT_LAST;
>  		} else if (em->block_start == EXTENT_MAP_HOLE) {
> -			flags |= FIEMAP_EXTENT_UNWRITTEN;
> +			hole = 1;
>  		} else if (em->block_start == EXTENT_MAP_INLINE) {
>  			flags |= (FIEMAP_EXTENT_DATA_INLINE |
>  				  FIEMAP_EXTENT_NOT_ALIGNED);
> @@ -3015,10 +3016,12 @@ int extent_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
>  			end = 1;
>  		}
>  
> -		ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
> -					em_len, flags);
> -		if (ret)
> -			goto out_free;
> +		if (!hole) {
> +			ret = fiemap_fill_next_extent(fieinfo, em_start, disko,
> +						      em_len, flags);
> +			if (ret)
> +				goto out_free;
> +		}
>  	}
>  out_free:
>  	free_extent_map(em);

I have just tried a simple test against this patch, it works for me.


Thanks,
-Jeff