linux-kernel - Re: [PATCH] watchdog: only print critical log when the watchdog fails to be stopped

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] [day] [month] [year] [list]

Message-ID: <d184241d-beca-4246-ba5a-f7a9ffe2f140@roeck-us.net>
Date: Tue, 4 Feb 2025 18:35:15 -0800
From: Guenter Roeck <linux@...ck-us.net>
To: "liuchao (CR)" <liuchao173@...wei.com>
Cc: caihe <caihe@...wei.com>,
 "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
 "linux-watchdog@...r.kernel.org" <linux-watchdog@...r.kernel.org>,
 lixiaokeng <lixiaokeng@...wei.com>,
 "wim@...ux-watchdog.org" <wim@...ux-watchdog.org>
Subject: Re: [PATCH] watchdog: only print critical log when the watchdog fails
 to be stopped

On 2/4/25 18:02, liuchao (CR) wrote:
> On 1/27/2025 22:30, Guenter Roeck wrote:
>> On 1/27/25 01:35, liuchao (CR) wrote:
>>> On 1/26/25 21:10, Guenter Roeck wrote:
>>>> On 1/26/25 00:38, Liu Chao wrote:
>>>>> Every time the user echoes 0 > /dev/watchdog0, meaningless critical
>>>>> log is printed.
>>>>>
>>>>
>>>> It is not meaningless, and it will still be displayed after this
>>>> change, making the change pointless.
>>>
>>> The change is not pointless. For example, the softdog driver does not
>>> invoke watchdog_stop or print logs in the watchdog_release.
>>>
>>
>> It seems to me that is a problem in the softdog driver.
>>
>> The change is actually worse than I initially thought.
>> The message is _supposed_ to be displayed if watchdog_stop() is not called while
>> the watchdog is running (i.e., if err == -EBUSY).
>> Otherwise it would not be displayed for real hardware watchdogs which are not
>> stopped because they were running and watchdog_stop() is not called because
>> WDIOF_MAGICCLOSE is set in the driver and the magic release byte was not
>> written.
>>
>> Specifically, the softdog driver has WDIOF_MAGICCLOSE set. It is not supposed
>> to be unloadable (or unloaded) while the watchdog is running.
> 
> When echo to /dev/watchdog0, The watchdog_open, watchdog_write, and
> watchdog_release functions are invoked in sequence. Do you mean the softdog
> driver should not call watchdog_release?
> 
> After the user opens /dev/watchdog0, the user feeds the watchdog through ioctl
> WDIOC_KEEPALIVE and never closes. Is this the correct usage?
> 

I tried softdog. It works as advertised. Yes, "echo 0 > watchdog0" triggers the message.
"sudo modprobe -r softdog" then fails with

	modprobe: FATAL: Module softdog is in use.

and, as expected, one minute later (or whatever the timeout is set to) the system reboots.

There is nothing wrong with the message. The softdog _does_ refuse to be unloaded
while running, and it _does_ reboot the system after the timeout expired. This is all
perfectly as expected. The log is not meaningless. Instead, it tells the user that
the system will reboot after the watchdog expired. Which it does.

Guenter