Re: Re: PATCH? debugfs_remove_recursive() must not rely on list_empty(d_subdirs)

[Oleg Nesterov <oleg@redhat.com>]

On 07/26, Masami Hiramatsu wrote:
>
> (2013/07/26 5:04), Oleg Nesterov wrote:
> >
> > 	parent = dentry;
> >  down:
> > 	mutex_lock(&parent->d_inode->i_mutex);
> > 	list_for_each_entry_safe(child, next, &parent->d_subdirs, d_u.d_child) {
>
> Perhaps, you can use list_for_each_entry_safe_continue() here, as below.
>
> 	parent = dentry;
> down:
> 	child = list_first_entry_or_null(&parent->d_subdirs,
> 					 typeof(*child), d_u.d_child);
> 	mutex_lock(&parent->d_inode->i_mutex);
>
> restart:
> 	list_for_each_entry_safe_continue(child, next, &parent->d_subdirs, d_u.d_child) {
>
> > 		if (!debugfs_positive(child))
> > 			continue;
> >
> > 		/* XXX: simple_empty(child) instead ? */
> > 		if (!list_empty(&child->d_subdirs)) {
> > 			mutex_unlock(&parent->d_inode->i_mutex);
> > 			parent = child;
> > 			goto down;
> > 		}
> >  up:
> > 		__debugfs_remove(child, parent);
> > 	}
>
> Then, you can avoid jumping into the loop, just restart it from
> parent as below.

Yes, but I'd prefer to jump into the loop. This is subjective, but looks
a bit more understandable to me.

Because "goto down/up" are actually "call/return", and "jump up" looks
like return-after-recursive-debugfs_remove_recursive-call.

However,

> 	if (child != dentry)
> 		goto restart;
>
> > 	mutex_unlock(&parent->d_inode->i_mutex);

Yes, I realized this right after I sent the email ;)

We can factor out the final ->d_parent/mutex_lock if we check
"child != dentry" instead of "parent != dentry".

I'll send the patch in a minute. Thanks.

Oleg.