Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <1588044545-59405-1-git-send-email-wubo40@huawei.com>
In-Reply-To: <1588044545-59405-1-git-send-email-wubo40@huawei.com>
From:   Andreas Gruenbacher <agruenba@redhat.com>
Date:   Tue, 28 Apr 2020 11:56:00 +0200
Message-ID: <CAHc6FU5PHELD9N4tdfn663ao1RqyiByGLb8JOfEJGJdyNUwpQw@mail.gmail.com>
Subject: Re: [PATCH] fs/gfs2:lock a spinlock always before returning from do_xmote()
To:     Wu Bo <wubo40@huawei.com>
Cc:     Bob Peterson <rpeterso@redhat.com>,
        cluster-devel <cluster-devel@redhat.com>,
        LKML <linux-kernel@vger.kernel.org>, liuzhiqiang26@huawei.com,
        linfeilong@huawei.com
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi,

On Tue, Apr 28, 2020 at 5:30 AM Wu Bo <wubo40@huawei.com> wrote:
> The call stack is as follows:
> finish_xmote()
>         ...
>         spin_lock(&gl->gl_lockref.lock);
>         ...
>         --> do_xmote()
>             spin_unlock(&gl->gl_lockref.lock);
>             ...
>             return;
>         ...
>         spin_unlock(&gl->gl_lockref.lock);
>
> do_xmote function needs to be locked before returning,
> Otherwise, there will be a double release lock in finish_xmote() function.
>
> Signed-off-by: Wu Bo <wubo40@huawei.com>
> ---
>  fs/gfs2/glock.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/fs/gfs2/glock.c b/fs/gfs2/glock.c
> index 29f9b66..7129d10 100644
> --- a/fs/gfs2/glock.c
> +++ b/fs/gfs2/glock.c
> @@ -613,6 +613,7 @@ static void do_xmote(struct gfs2_glock *gl, struct gfs2_holder *gh, unsigned int
>                                 fs_err(sdp, "Error %d syncing glock \n", ret);
>                                 gfs2_dump_glock(NULL, gl, true);
>                         }
> +                       spin_lock(&gl->gl_lockref.lock);
>                         return;
>                 }
>         }
> --
> 1.8.3.1

this patch looks correct. We've independently discovered this bug as
well in the meantime, and we'll send the fix upstream.

Thanks,
Andreas