regression: bfs endianness bug

BFS_FILEBLOCKS() expects struct bfs_inode * (on-disk data, with little-
endian fields), not struct bfs_inode_info * (in-core stuff, with host-
endian ones).

It's a macro and fields with the right names are present in
bfs_inode_info, so it compiles, but on big-endian host it gives bogus
results.

Introduced in commit f433dc56344cb72cc3de5ba0819021cec3aef807 ("Fixes to
the BFS filesystem driver").

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/bfs/inode.c

@@ -178,7 +178,8 @@ static void bfs_delete_inode(struct inode *inode)
 	brelse(bh);
 
         if (bi->i_dsk_ino) {
-		info->si_freeb += BFS_FILEBLOCKS(bi);
+		if (bi->i_sblock)
+			info->si_freeb += bi->i_eblock + 1 - bi->i_sblock;
 		info->si_freei++;
 		clear_bit(ino, info->si_imap);
 		dump_imap("delete_inode", s);