Re: EXT4_FEATURE_RO_COMPAT_DIR_NLINK and the directory size.

[Andreas Dilger <adilger@sun.com>]

On Feb 12, 2008  20:35 +0530, Aneesh Kumar K.V wrote:
> On Tue, Feb 12, 2008 at 07:23:42PM +0530, Aneesh Kumar K.V wrote:
> > On Tue, Feb 12, 2008 at 06:06:32PM +0530, Aneesh Kumar K.V wrote:
> > > Also after running fsstress fsck fails with zero size directory on
> > > powerpc. I am finding directories with 
> > > 
> > > i_links_count == 1
> > > i_blocks == 0

Is this after some failure to create a subdirectory (e.g. ENOSPC)?

> > (gdb) p *inode
> > $1 = {i_mode = 16895, i_uid = 0, i_size = 4096, i_atime = 1202822535, i_ctime = 1202822535, i_mtime = 1202822535, i_dtime = 0,
> > i_gid = 0, i_links_count = 1, i_blocks = 0, i_flags = 524288, osd1 = {linux1 = {l_i_version = 1}, hurd1 = {h_i_translator = 1},
> > masix1 = {m_i_reserved1 = 1}}, i_block = {4077518848, 262144, 0 <repeats 13 times>}, i_generation = 2950864834, i_file_acl = 0,
> > i_dir_acl = 0, i_faddr = 0, osd2 = {linux2 = {l_i_blocks_hi = 0, i_pad1 = 0, l_i_uid_high = 0, l_i_gid_high = 0, l_i_reserved2 = 0},
> > hurd2 = {h_i_frag = 0 '\0', h_i_fsize = 0 '\0', h_i_mode_high = 0, h_i_uid_high = 0, h_i_gid_high = 0, h_i_author = 0}, masix2 = {
> > m_i_frag = 0 '\0', m_i_fsize = 0 '\0', m_pad1 = 0, m_i_reserved2 = {0, 0}}}}

This is the subdirectory inode?  The i_block values are the extent magic,
and i_flags = EXTENT_FL it appears (which is fine).

> Right now waiting for the test to finish with this change.
> 
> diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
> index a9347fb..fd3b031 100644
> --- a/fs/ext4/namei.c
> +++ b/fs/ext4/namei.c
> @@ -1805,7 +1805,7 @@ retry:
>  	inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
>  	dir_block = ext4_bread (handle, inode, 0, 1, &err);
>  	if (!dir_block) {
> -		ext4_dec_count(handle, inode); /* is this nlink == 0? */
> +		drop_nlink(inode);
>  		ext4_mark_inode_dirty(handle, inode);
>  		iput (inode);
>  		goto out_stop;

It's entirely possible this is the right fix.  The code was changed for
all directory inodes replacing drop_nlink() (or its predecessor) with
ext4_dec_count().  In most cases this is correct, but it seems here it
isn't very clear (even with the useless comment) that the intention is
to drop the nlinks to 0 for the new directory inode.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.