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;
Date:   Sun, 26 Jul 2020 09:06:50 +0200
From:   Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To:     Xin Xiong <xiongx18@fudan.edu.cn>
Cc:     Jiri Slaby <jslaby@suse.com>, linux-kernel@vger.kernel.org,
        Xiyu Yang <xiyuyang19@fudan.edu.cn>,
        Xin Tan <tanxin.ctf@gmail.com>, yuanxzhang@fudan.edu.cn
Subject: Re: [PATCH] tty: fix pid refcount leak in tty_signal_session_leader
Message-ID: <20200726070650.GA440555@kroah.com>
References: <20200726052804.GA51199@xin-virtual-machine>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200726052804.GA51199@xin-virtual-machine>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, Jul 26, 2020 at 01:28:04PM +0800, Xin Xiong wrote:
> In the loop, every time when p->signal->leader is true, the function
> tty_signal_session_leader() will invoke get_pid() and return a
> reference of tty->pgrp with increased refcount to the local variable
> tty_pgrp or return NULL if it fails. After finishing the loop, the
> function invokes put_pid() for only once, decreasing the refcount that
> tty_pgrp keeps.
> 
> Refcount leaks may occur when the scenario that p->signal->leader is
> true happens more than once. In this assumption, if the above scenario
> happens n times in the loop, the function forgets to decrease the
> refcount for n-1 times, which causes refcount leaks.
> 
> Fix the issue by decreasing the current refcount of the local variable
> tty_pgrp before assigning new objects to it.
> 
> Signed-off-by: Xiyu Yang <xiyuyang19@fudan.edu.cn>
> Signed-off-by: Xin Tan <tanxin.ctf@gmail.com>
> Signed-off-by: Xin Xiong <xiongx18@fudan.edu.cn>
> ---
>  drivers/tty/tty_jobctrl.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/drivers/tty/tty_jobctrl.c b/drivers/tty/tty_jobctrl.c
> index f8ed50a16848..9e6bf693ade1 100644
> --- a/drivers/tty/tty_jobctrl.c
> +++ b/drivers/tty/tty_jobctrl.c
> @@ -212,6 +212,8 @@ int tty_signal_session_leader(struct tty_struct *tty, int exit_session)
>  			__group_send_sig_info(SIGCONT, SEND_SIG_PRIV, p);
>  			put_pid(p->signal->tty_old_pgrp);  /* A noop */
>  			spin_lock(&tty->ctrl_lock);
> +			if (tty_pgrp)
> +				put_pid(tty_pgrp);

No need to check this before calling it.

But, the real question is why is this needed now?  Nothing has changed
in this area of the kernel for a very long time, so how did things get
broken here?

How are you triggering this and what is the result when we have that
additional reference?

thanks,

greg k-h