Bug in sysfs_break_active_protection()

From: Alan Stern <stern@rowland.harvard.edu>
To: Tejun Heo <tj@kernel.org>, Greg KH <gregkh@linuxfoundation.org>
Cc: Kernel development list <linux-kernel@vger.kernel.org>
Subject: Bug in sysfs_break_active_protection()
Date: Mon, 4 Mar 2024 14:17:27 -0500	[thread overview]
Message-ID: <9c2484f4-df62-4d23-97a2-55a160eba55f@rowland.harvard.edu> (raw)
In-Reply-To: <416a8311-c725-419a-8b22-74c80207347f@rowland.harvard.edu>

Tejun:

It looks like there's a memory leak in sysfs_break_active_protection().  
It and its companion routine do this:

struct kernfs_node *sysfs_break_active_protection(struct kobject *kobj,
						  const struct attribute *attr)
{
	struct kernfs_node *kn;

	kobject_get(kobj);
	kn = kernfs_find_and_get(kobj->sd, attr->name);
	if (kn)
		kernfs_break_active_protection(kn);
	return kn;
}

void sysfs_unbreak_active_protection(struct kernfs_node *kn)
{
	struct kobject *kobj = kn->parent->priv;

	kernfs_unbreak_active_protection(kn);
	kernfs_put(kn);
	kobject_put(kobj);
}

If kn is NULL then the kobject_get(kobj) reference is never dropped.  
It looks like this could happen if two processes want to unregister the 
same kobject at the same time.

Shouldn't sysfs_break_active_protection() do this?

	kobject_get(kobj);
	kn = kernfs_find_and_get(kobj->sd, attr->name);
	if (kn)
		kernfs_break_active_protection(kn);
+	else
+		kobject_put(kobj);
	return kn;

Alan Stern