On 2011-09-14, at 1:31 PM, Darrick J. Wong wrote:

On Wed, Sep 14, 2011 at 01:02:59PM -0400, Ted Ts'o wrote:

quoted

On Wed, Aug 31, 2011 at 05:36:22PM -0700, Darrick J. Wong wrote:

quoted

diff --git a/lib/ext2fs/ext2_fs.h b/lib/ext2fs/ext2_fs.h
index 1f08673..367bfdf 100644
--- a/lib/ext2fs/ext2_fs.h
+++ b/lib/ext2fs/ext2_fs.h

@@ -169,7 +169,8 @@ struct ext4_group_desc

	__u16	bg_free_inodes_count_hi;/* Free inodes count MSB */
	__u16	bg_used_dirs_count_hi;	/* Directories count MSB */
	__u16	bg_itable_unused_hi;	/* Unused inodes count MSB */
-	__u32	bg_reserved2[3];
+	__u32	bg_inode_bitmap_csum;	/* crc32c(uuid+group+ibitmap) */
+	__u32	bg_reserved2[2];
};

One of the reasons why I like to coalesce the patches to the data
structures into their own separate commit, is it's hard when I'm
reviewing individual patches in a mail reader what's going on from a
big picture spective.  (Heck, even just *finding* the patches that
modify the on-disk format is hard....)

But as near as I can tell, your patch series only uses one of the
32-bit fields in bg_reserved.  Is there a good reason why
bg_inode_bitmap_csum can't also used one of the two fields in
bg_reserved?  That way we get two 32-bit checksums for both struct
ext2_group_desc and struct ext4_group_desc.  Is there a third 32-bit
per-block group checksum I'm forgetting about? 

No, but I was under the impression that Amir was using one of bg_reserved for
snapshots.  If that's not true, I'll move both bitmap checksums to bg_reserved.

He was, AND the first bg_reserved2 field as well, which the above use
would conflict with.  See my other patch.

Cheers, Andreas