linux-kernel - Re: [BUG] -next lockdep invalid wait context

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <Zy5EIHUwoXjK1sAJ@slm.duckdns.org>
Date: Fri, 8 Nov 2024 07:02:24 -1000
From: Tejun Heo <tj@...nel.org>
To: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Cc: Boqun Feng <boqun.feng@...il.com>, Hillf Danton <hdanton@...a.com>,
	"Paul E. McKenney" <paulmck@...nel.org>,
	Marco Elver <elver@...gle.com>, linux-kernel@...r.kernel.org,
	gregkh@...uxfoundation.org, tglx@...utronix.de,
	syzbot <syzbot+6ea37e2e6ffccf41a7e6@...kaller.appspotmail.com>
Subject: Re: [BUG] -next lockdep invalid wait context

Hello,

On Fri, Nov 08, 2024 at 11:05:03AM +0100, Sebastian Andrzej Siewior wrote:
...
>  static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen)
>  {
> +	struct kernfs_node *kn_parent;
> +	const char *kn_name;
> +
>  	if (!kn)
>  		return strscpy(buf, "(null)", buflen);
>  
> -	return strscpy(buf, kn->parent ? kn->name : "/", buflen);
> +	kn_parent = rcu_dereference(kn->parent);

kn->parent can never change. This can just be a regular deref.

> +	kn_name = rcu_dereference(kn->name);
> +	return strscpy(buf, kn_parent ? kn_name : "/", buflen);
>  }
>  
>  /* kernfs_node_depth - compute depth from @from to @to */
> @@ -66,7 +70,7 @@ static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to)
>  
>  	while (to->parent && to != from) {
>  		depth++;
> -		to = to->parent;
> +		to = rcu_dereference(to->parent);

Ditto here and other places.

>  int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
>  {
> -	unsigned long flags;
>  	int ret;
>  
> -	read_lock_irqsave(&kernfs_rename_lock, flags);
> +	rcu_read_lock();
>  	ret = kernfs_name_locked(kn, buf, buflen);
> -	read_unlock_irqrestore(&kernfs_rename_lock, flags);
> +	rcu_read_unlock();
>  	return ret;
>  }
>  
> @@ -223,12 +226,11 @@ int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen)
>  int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from,
>  			  char *buf, size_t buflen)
>  {
> -	unsigned long flags;
>  	int ret;
>  
> -	read_lock_irqsave(&kernfs_rename_lock, flags);
> +	rcu_read_lock();
>  	ret = kernfs_path_from_node_locked(to, from, buf, buflen);
> -	read_unlock_irqrestore(&kernfs_rename_lock, flags);
> +	rcu_read_unlock();
>  	return ret;
>  }

The _locked suffix looks awkward afterwards. Given that there's no
restriction on how the function is called anymore, maybe we can just
collapse _locked bodies into the callers?

...
> diff --git a/include/linux/kernfs.h b/include/linux/kernfs.h
> index 87c79d076d6d7..733d89de40542 100644
> --- a/include/linux/kernfs.h
> +++ b/include/linux/kernfs.h
> @@ -199,8 +199,8 @@ struct kernfs_node {
>  	 * never moved to a different parent, it is safe to access the
>  	 * parent directly.
>  	 */
> -	struct kernfs_node	*parent;
> -	const char		*name;
> +	struct kernfs_node	__rcu *parent;
> +	const char		__rcu *name;

->parent doesn't have to be converted to RCU. As long as the node is
accessible, the parent is guaranteed to be pinned and unchanged. We only
need to RCUfy ->name.

Thanks.

-- 
tejun