Re: [PATCH] Introduce the pkill_on_warn boot parameter

[Alexander Popov]

On 30.09.2021 12:15, Petr Mladek wrote:
> On Wed 2021-09-29 12:49:24, Paul E. McKenney wrote:
>> On Wed, Sep 29, 2021 at 10:01:33PM +0300, Alexander Popov wrote:
>>> On 29.09.2021 21:58, Alexander Popov wrote:
>>>> Currently, the Linux kernel provides two types of reaction to kernel
>>>> warnings:
>>>>  1. Do nothing (by default),
>>>>  2. Call panic() if panic_on_warn is set. That's a very strong reaction,
>>>>     so panic_on_warn is usually disabled on production systems.
> 
> Honestly, I am not sure if panic_on_warn() or the new pkill_on_warn()
> work as expected. I wonder who uses it in practice and what is
> the experience.
> 
> The problem is that many developers do not know about this behavior.
> They use WARN() when they are lazy to write more useful message or when
> they want to see all the provided details: task, registry, backtrace.
> 
> Also it is inconsistent with pr_warn() behavior. Why a single line
> warning would be innocent and full info WARN() cause panic/pkill?
> 
> What about pr_err(), pr_crit(), pr_alert(), pr_emerg()? They inform
> about even more serious problems. Why a warning should cause panic/pkill
> while an alert message is just printed?

That's a good question.

I guess various kernel continuous integration (CI) systems have panic_on_warn
enabled.

[Adding Dmitry Vyukov to this discussion]

If we look at the syzbot dashboard [1] with the results of Linux kernel fuzzing,
we see the issues that appear as various kernel crashes and warnings.
We don't see anything from pr_err(), pr_crit(), pr_alert(), pr_emerg(). Maybe
these situations are not considered as kernel bugs that require fixing.

Anyway, from a security point of view, a kernel warning output is interesting
for attackers as an infoleak. The messages printed by pr_err(), pr_crit(),
pr_alert(), pr_emerg() provide less information.

[1]: https://syzkaller.appspot.com/upstream

> It somehow reminds me the saga with %pK. We were not able to teach
> developers to use it correctly for years and ended with hashed
> pointers.
> 
> Well, this might be different. Developers might learn this the hard
> way from bug reports. But there will be bug reports only when
> anyone really enables this behavior. They will enable it only
> when it works the right way most of the time.
> 
> 
>>>> From a safety point of view, the Linux kernel misses a middle way of
>>>> handling kernel warnings:
>>>>  - The kernel should stop the activity that provokes a warning,
>>>>  - But the kernel should avoid complete denial of service.
>>>>
>>>> From a security point of view, kernel warning messages provide a lot of
>>>> useful information for attackers. Many GNU/Linux distributions allow
>>>> unprivileged users to read the kernel log, so attackers use kernel
>>>> warning infoleak in vulnerability exploits. See the examples:
>>>>   https://a13xp0p0v.github.io/2020/02/15/CVE-2019-18683.html
>>>>   https://a13xp0p0v.github.io/2021/02/09/CVE-2021-26708.html
>>>>
>>>> Let's introduce the pkill_on_warn boot parameter.
>>>> If this parameter is set, the kernel kills all threads in a process
>>>> that provoked a kernel warning. This behavior is reasonable from a safety
>>>> point of view described above. It is also useful for kernel security
>>>> hardening because the system kills an exploit process that hits a
>>>> kernel warning.
>>>>
>>>> Signed-off-by: Alexander Popov <alex.popov@xxxxxxxxx>
>>>
>>> This patch was tested using CONFIG_LKDTM.
>>> The kernel kills a process that performs this:
>>>   echo WARNING > /sys/kernel/debug/provoke-crash/DIRECT
>>>
>>> If you are fine with this approach, I will prepare a patch adding the
>>> pkill_on_warn sysctl.
>>
>> I suspect that you need a list of kthreads for which you are better
>> off just invoking panic().  RCU's various kthreads, for but one set
>> of examples.
> 
> I wonder if kernel could survive killing of any kthread. I have never
> seen a code that would check whether a kthread was killed and
> restart it.

The do_group_exit() function calls do_exit() from kernel/exit.c, which is also
called during a kernel oops. This function cares about a lot of special cases
depending on the current task_struct. Is it fine?

Best regards,
Alexander