On Sun, Apr 29, 2012 at 12:36 AM, Andrew Morton <akpm@linux-foundation.org> wrote:
On Sun, 29 Apr 2012 00:32:26 +0530 Sasikanth babu <sasikanth.v19@gmail.com> wrote:

> > Fact is, debugfs_create_dir() and debugfs_create_file() are stupid
> > interfaces which don't provide the caller (and hence the user) with any
> > information about why they failed.  Perhaps memblock_init_debugfs()
> > should return -EWESUCK.
> >
>
>    I'm working on a patch which address this issue. debugfs_create_XXX
> calls
>    will return proper error codes, and fixing the existing code not each
> and every part  but the code
>    which handles the values returned by debufs_create_XXX otherwise it will
> break the existing
>    functionality .

Excellent!

> (any suggestions or opinions ?)

Well, don't modify the existing interfaces: create new ones and we can
migrate gradually.  But you're probably already doing that.

  Not going to change the existing interface.  Modified debugfs_create_XXX
  to return ERR_PTR(error) instead of NULL.
 
[sasikantha@localhost linux-2.6]$ git diff
diff --git a/fs/debugfs/inode.c b/fs/debugfs/inode.c
index b80bc84..f5a5783 100644
--- a/fs/debugfs/inode.c
+++ b/fs/debugfs/inode.c
@@ -378,7 +378,7 @@ struct dentry *debugfs_create_file(const char *name, umode_t mode,
        error = debugfs_create_by_name(name, mode, parent, &dentry,
                                       data, fops);
        if (error) {
-               dentry = NULL;
+               dentry = ERR_PTR(error);
                simple_release_fs(&debugfs_mount, &debugfs_mount_count);
                goto exit;
        }

  And from the caller side modifying the code as shown below (Currently started doing
  modification for each subsystem)
 
        dir = debugfs_create_dir("test", NULL);

        if (IS_ERR(dir)) {
            return PTR_ERR(dir);
       }

   I think as you had mentioned creating new interface and migrating gradually is the right of
   way of doing it.
   
   Thanks
   Sasi