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;
Sender: Guenter Roeck <groeck7@gmail.com>
Date:   Thu, 5 Nov 2020 06:26:21 -0800
From:   Guenter Roeck <linux@roeck-us.net>
To:     Wang Wensheng <wangwensheng4@huawei.com>
Cc:     wim@linux-watchdog.org, linux-watchdog@vger.kernel.org,
        linux-kernel@vger.kernel.org, rui.xiang@huawei.com,
        guohanjun@huawei.com
Subject: Re: [PATCH -next v3 1/2] watchdog: Fix potential dereferencing of
 null pointer
Message-ID: <20201105142621.GB1389@roeck-us.net>
References: <20201105123848.93735-1-wangwensheng4@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201105123848.93735-1-wangwensheng4@huawei.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On Thu, Nov 05, 2020 at 12:38:47PM +0000, Wang Wensheng wrote:
> A reboot notifier, which stops the WDT by calling the stop hook without
> any check, would be registered when we set WDOG_STOP_ON_REBOOT flag.
> 
> Howerer we allow the WDT driver to omit the stop hook since commit
> "d0684c8a93549" ("watchdog: Make stop function optional") and provide
> a module parameter for user that controls the WDOG_STOP_ON_REBOOT flag
> in commit 9232c80659e94 ("watchdog: Add stop_on_reboot parameter to
> control reboot policy"). Together that commits make user potential to
> insert a watchdog driver that don't provide a stop hook but with the
> stop_on_reboot parameter set, then dereferencing of null pointer occurs
> on system reboot.
> 
> Check the stop hook before registering the reboot notifier to fix the
> issue.
> 
> Fixes: d0684c8a9354 ("watchdog: Make stop function optional")
> Signed-off-by: Wang Wensheng <wangwensheng4@huawei.com>
> ---
>  drivers/watchdog/watchdog_core.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/watchdog/watchdog_core.c b/drivers/watchdog/watchdog_core.c
> index 423844757812..945ab38b14b8 100644
> --- a/drivers/watchdog/watchdog_core.c
> +++ b/drivers/watchdog/watchdog_core.c
> @@ -267,8 +267,15 @@ static int __watchdog_register_device(struct watchdog_device *wdd)
>  	}
>  
>  	if (test_bit(WDOG_STOP_ON_REBOOT, &wdd->status)) {
> -		wdd->reboot_nb.notifier_call = watchdog_reboot_notifier;
> +		if (!wdd->ops->stop) {
> +			pr_err("watchdog%d: Cannot support stop_on_reboot\n",
> +				wdd->id);
> +			watchdog_dev_unregister(wdd);
> +			ida_simple_remove(&watchdog_ida, id);
> +			return -EINVAL;
> +		}

The problem with this is that setting the "stop_on_reboot" module parameter
would now prevent the watchdog from being loaded, which isn't really
desirable and might go unnoticed. I think the initial check should be
above, with the "Mandatory operations" check, and
	if (stop_on_reboot != -1) {
should be extended to
	if (stop_on_reboot != -1 && wdd->ops->stop) {

or possibly more fancy:

	if (stop_on_reboot != -1) {
		if (stop_on_reboot) {
			if (!wdd->ops->stop)
				pr_warn("watchdog%d: stop_on_reboot not supported\n", wdd->id);
			else
				set_bit(WDOG_STOP_ON_REBOOT, &wdd->status);
		} else {
			clear_bit(WDOG_STOP_ON_REBOOT, &wdd->status);
		}
	}

Thanks,
Guenter

>  
> +		wdd->reboot_nb.notifier_call = watchdog_reboot_notifier;
>  		ret = register_reboot_notifier(&wdd->reboot_nb);
>  		if (ret) {
>  			pr_err("watchdog%d: Cannot register reboot notifier (%d)\n",
> -- 
> 2.25.0
>