linux-kernel - Re: [PATCH V4 18/20] rv/monitor: Add safe watchdog monitor

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]

Date:   Fri, 1 Jul 2022 07:45:50 -0700
From:   Guenter Roeck <linux@...ck-us.net>
To:     Steven Rostedt <rostedt@...dmis.org>
Cc:     Gabriele Paoloni <gpaoloni@...hat.com>,
        Daniel Bristot de Oliveira <bristot@...nel.org>,
        Wim Van Sebroeck <wim@...ux-watchdog.org>,
        Jonathan Corbet <corbet@....net>,
        Ingo Molnar <mingo@...hat.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Peter Zijlstra <peterz@...radead.org>,
        Will Deacon <will@...nel.org>,
        Catalin Marinas <catalin.marinas@....com>,
        Marco Elver <elver@...gle.com>,
        Dmitry Vyukov <dvyukov@...gle.com>,
        "Paul E. McKenney" <paulmck@...nel.org>,
        Shuah Khan <skhan@...uxfoundation.org>,
        Juri Lelli <juri.lelli@...hat.com>,
        Clark Williams <williams@...hat.com>,
        "open list:DOCUMENTATION" <linux-doc@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-trace-devel@...r.kernel.org
Subject: Re: [PATCH V4 18/20] rv/monitor: Add safe watchdog monitor

On 6/28/22 12:32, Steven Rostedt wrote:
[ ... ]
>> I am open to suggestions from others, but at this point I have serious doubts
>> that this code is maintainable in the kernel.
>>
> 
> Let me give some background on this. Various safety critical users
> (automotive, power plants, etc) are looking at using Linux in the field.
> Obviously, Linux itself is too big to verify properly. In fact, even small
> OSs that say they are verifiable seldom are. Part of the solution is to add
> a way to check that Linux is running as expected, and be able to detect
> when it is not. When that happens, you fall into a "fail-safe" mode (for
> example, if you have an autonomous driving vehicle, it could sound an alarm
> and inform the driver to go into manual mode, or it is going to simply
> "pull over". Note, this is just a hypothetical example, not something I've
> seen in reality).
> 
> But to implement this, you need to verify the verifier. In this case, the
> watchdog timer is what verifies that Linux is running properly. It is
> possible to verify parts of the kernel when you limit the inputs of the
> kernel. This means we are not trying to verify all use cases of the
> watchdog timer. That would be pretty much impossible, or take decades to
> complete. We only need to verify the use cases used in safety critical
> systems.
> 

I understand the context. However, ...

> Do not confuse this with static analyzers or other general purpose tooling
> to find bugs. This is not the purpose of the run time verify. It is just to
> prove that certain use cases will perform as expected, given a limited
> input.
> 

... this is, unfortunately, not explained in the patch. I would have much less
of a problem with the series if those details were included.

Not that I would mind such a verifier, if it was possible to define one,
but it would have to be tested with a large number of watchdog drivers
to ensure that it addresses all use cases, or at least with a substantial
percentage of use cases. It would also require that the state machine is
readable to give people a chance to fix it if turns out to be broken.
It would also have to be robust, meaning it would have to reject invalid
(unsupported) settings from the start and not only during runtime.

Thanks,
Guenter