ecryptfs-devel team mailing list archive

[Dave Kleikamp <shaggy@xxxxxxxxxxxxxxxxxx>]

On Sat, 2009-01-17 at 15:03 +0900, hooanon05@xxxxxxxxxxx wrote:
> Dave Kleikamp:
> > I think you're hitting on something here.  I never understood  the need
> > for the d_drop()s, but taking them out broke things.  They probably are
> > just papering over bugs where the ecryptfs inode is not being properly
> > updated after changes are made to the lower inode.
> 
> As long as cifs_hardlink() calls d_drop() for the target dentry (as the
> old version of NFS did), ecryptfs may have to call d_drop() too. But I
> believe the d_drop() for the source dentry is unnecessary, as long as
> the inode attributes are maintained correctly.
> Additionally, when the lower filesystem does NOT call d_drop(), ecryptfs
> has no necessary to call it. I'd like to suggest ecryptfs_link() to
> check it by d_unhashed().

Does this function make sense (un-compiled, un-tested)?

void ecryptfs_update_inode_from_lower(struct dentry *dentry)
{
	struct dentry *lower_dentry = ecryptfs_dentry_to_lower(dentry);

	if (d_unhashed(lower_dentry))
		d_drop(dentry);
	else {
		struct inode *inode = dentry->d_inode;
		struct inode *lower_inode = lower_dentry->d_inode;

		inode->i_nlink = lower_inode->i_nlink;
		inode->i_ctime = lower_inode->i_ctime;
		/* Should anything else go here ? */
	}
}
		
-- 
David Kleikamp
IBM Linux Technology Center