On Nov 12, 2007 09:58 -0500, Theodore Tso wrote:
> On Mon, Nov 12, 2007 at 10:52:45AM +0100, Jan Kara wrote:
> > > Did you test this patch before submitting it?
> >
> > Argh, stupid me. I've just tested that I didn't break anything for normal
> > block size and thought that I cannot make mistake in such a simple thing
> > ;).
>
> Could I ask you to perhaps include some 64k blocksize test cases that
> would exercise the new codepaths?
One thing I had done to verify that 64kB rec_len was broken was to always
create lost+found with at least 2 disk blocks, since the second block in
lost+found will not have any dirents in it and will tickle this bug.
Patch attached.
Cheers, Andreas
--
Andreas Dilger
Sr. Software Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
Index: e2fsprogs-cfs/misc/mke2fs.c
===================================================================
--- e2fsprogs-cfs.orig/misc/mke2fs.c
+++ e2fsprogs-cfs/misc/mke2fs.c
@@ -551,7 +551,10 @@ static void create_lost_and_found(ext2_f
}
for (i=1; i < EXT2_NDIR_BLOCKS; i++) {
- if ((lpf_size += fs->blocksize) >= 16*1024)
+ /* Ensure that lost+found is at least 2 blocks, so we always
+ * test large empty blocks for big-block filesystems. */
+ if ((lpf_size += fs->blocksize) >= 16*1024 &&
+ lpf_size >= 2 * fs->blocksize)
break;
retval = ext2fs_expand_dir(fs, ino);
if (retval) {
