linux-kernel - Re: [PATCH] ocfs2: kill osb->system_file

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: <faf70481-09dd-4c7a-bd43-f1e8bec877cb@suse.com>
Date: Tue, 24 Jun 2025 09:33:06 +0800
From: Heming Zhao <heming.zhao@...e.com>
To: Tetsuo Handa <penguin-kernel@...ove.SAKURA.ne.jp>,
 Mark Fasheh <mark@...heh.com>, Joel Becker <jlbec@...lplan.org>,
 Joseph Qi <joseph.qi@...ux.alibaba.com>, jiangyiwen <jiangyiwen@...wei.com>,
 Andrew Morton <akpm@...ux-foundation.org>, ocfs2-devel@...ts.linux.dev
Cc: LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] ocfs2: kill osb->system_file_mutex lock

Hello,

Protecting refcnt with a mutex is the right approach, and commit 43b10a20372d
did it properly.
However, I don't see how your patch fixes the syzbot report [1]. Could you
elaborate on the root cause analysis?

My review comments are inline below.

[1]: https://syzkaller.appspot.com/bug?extid=1fed2de07d8e11a3ec1b

On 6/21/25 23:56, Tetsuo Handa wrote:
> Since calling _ocfs2_get_system_file_inode() twice with the same
> arguments returns the same address, there is no need to serialize
> _ocfs2_get_system_file_inode() using osb->system_file_mutex lock.
> 
> Kill osb->system_file_mutex lock in order to avoid AB-BA deadlock.
> cmpxchg() will be sufficient for avoiding the inode refcount leak
> problem which commit 43b10a20372d ("ocfs2: avoid system inode ref
> confusion by adding mutex lock") tried to address.
> 
> Reported-by: Diogo Jahchan Koike <djahchankoike@...il.com>
'Reported-by' should be: https://syzkaller.appspot.com/bug?extid=1fed2de07d8e11a3ec1b

> Closes: https://lkml.kernel.org/r/000000000000ff2d7a0620381afe@google.com
I don't think we need 'Closes'.

> Fixes: 43b10a20372d ("ocfs2: avoid system inode ref confusion by adding mutex lock")
> Signed-off-by: Tetsuo Handa <penguin-kernel@...ove.SAKURA.ne.jp>
> Cc: jiangyiwen <jiangyiwen@...wei.com>
> Cc: Joseph Qi <joseph.qi@...wei.com>
> Cc: Joel Becker <jlbec@...lplan.org>
> Cc: Mark Fasheh <mfasheh@...e.com>
The 'CC's are also useless.

> ---
>   fs/ocfs2/ocfs2.h   | 2 --
>   fs/ocfs2/super.c   | 2 --
>   fs/ocfs2/sysfile.c | 9 +++------
>   3 files changed, 3 insertions(+), 10 deletions(-)
> 
> diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
> index 6aaa94c554c1..8bdeea60742a 100644
> --- a/fs/ocfs2/ocfs2.h
> +++ b/fs/ocfs2/ocfs2.h
> @@ -494,8 +494,6 @@ struct ocfs2_super
>   	struct rb_root	osb_rf_lock_tree;
>   	struct ocfs2_refcount_tree *osb_ref_tree_lru;
>   
> -	struct mutex system_file_mutex;
> -
>   	/*
>   	 * OCFS2 needs to schedule several different types of work which
>   	 * require cluster locking, disk I/O, recovery waits, etc. Since these
> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
> index 3d2533950bae..4461daf909cf 100644
> --- a/fs/ocfs2/super.c
> +++ b/fs/ocfs2/super.c
> @@ -1997,8 +1997,6 @@ static int ocfs2_initialize_super(struct super_block *sb,
>   	spin_lock_init(&osb->osb_xattr_lock);
>   	ocfs2_init_steal_slots(osb);
>   
> -	mutex_init(&osb->system_file_mutex);
> -
>   	atomic_set(&osb->alloc_stats.moves, 0);
>   	atomic_set(&osb->alloc_stats.local_data, 0);
>   	atomic_set(&osb->alloc_stats.bitmap_data, 0);
> diff --git a/fs/ocfs2/sysfile.c b/fs/ocfs2/sysfile.c
> index 53a945da873b..b63af8d64904 100644
> --- a/fs/ocfs2/sysfile.c
> +++ b/fs/ocfs2/sysfile.c
> @@ -98,11 +98,9 @@ struct inode *ocfs2_get_system_file_inode(struct ocfs2_super *osb,
>   	} else
>   		arr = get_local_system_inode(osb, type, slot);
>   
> -	mutex_lock(&osb->system_file_mutex);
>   	if (arr && ((inode = *arr) != NULL)) {
>   		/* get a ref in addition to the array ref */
>   		inode = igrab(inode);
> -		mutex_unlock(&osb->system_file_mutex);
>   		BUG_ON(!inode);
>   

I agree the above mutex_lock and mutex_unlock is useless. we can remove it
without any problem.

>   		return inode;
> @@ -112,11 +110,10 @@ struct inode *ocfs2_get_system_file_inode(struct ocfs2_super *osb,
>   	inode = _ocfs2_get_system_file_inode(osb, type, slot);

In my view, the key of commit 43b10a20372d is to avoid calling
_ocfs2_get_system_file_inode() twice, which lead refcnt+1 but no place to
do refcnt-1.

>   
>   	/* add one more if putting into array for first time */
> -	if (arr && inode) {
> -		*arr = igrab(inode);
> -		BUG_ON(!*arr);
> +	if (inode && arr && !*arr && !cmpxchg(&(*arr), NULL, inode)) {

Bypassing the refcnt+1 here is not a good idea. We should do refcnt+1
before returning to the caller.

> +		inode = igrab(inode);
> +		BUG_ON(!inode);
>   	}
> -	mutex_unlock(&osb->system_file_mutex);
>   	return inode;
>   }
>