"pkill_on_warn" Proposed For Killing Linux Processes That Cause A Kernel Warning

[phoronix]

Phoronix: "pkill_on_warn" Proposed For Killing Linux Processes That Cause A Kernel Warning

A new kernel option was proposed today called "pkill_on_warn" that would kill all threads in a process if that process provoked a kernel warning...

https://www.phoronix.com/scan.php?page=news_item&px=Linux-pkill_on_warn

[tildearrow]

I thought this already happened. When I hit an oops or bug condition, I think I just see "Killed" in the output.

[drjohnnyfever]

I can understand panicking on warning, but isn't shooting down programs that triggered a kernel bug sort of missing the point? The kernel should do the honorable thing and kill itself.

[perpetually high]

I can understand panicking on warning, but isn't shooting down programs that triggered a kernel bug sort of missing the point? The kernel should do the honorable thing and kill itself.

Maybe I’m missing something but I think the #1 job of a kernel is *not* to panic. Don’t care what caused it. Just don’t panic. Warn me and then let me tell the developers and never run it again

[tildearrow]

I can understand panicking on warning, but isn't shooting down programs that triggered a kernel bug sort of missing the point? The kernel should do the honorable thing and kill itself.

It's a kernel option. You are able to configure this.

Personally having some time to save my work and reboot is better than losing my work due to a kernel bug causing a panic.

[billbo]

Err what about the parent of the process? If I have my exploit fork before trying the "dangerous" code, the parent can keep chugging along and try again in a new child.

[drjohnnyfever]

Maybe I’m missing something but I think the #1 job of a kernel is *not* to panic. Don’t care what caused it. Just don’t panic. Warn me and then let me tell the developers and never run it again

Well, if that is the case we don't actually need panic() at all. Just remove it.

[drjohnnyfever]

It's a kernel option. You are able to configure this.

Personally having some time to save my work and reboot is better than losing my work due to a kernel bug causing a panic.

Unless meanwhile that kernel is eating all your data already on the disk instead of panicking. The reason you panic is because you are in *a bad state*

[perpetually high]

Well, if that is the case we don't actually need panic() at all. Just remove it.

The kernel can still panic from bad code or unintended consequences of a code change whatever. Separate from this issue. You would still want the mechanism in place to not panic the kernel when it did nothing wrong

[drjohnnyfever]

The kernel can still panic from bad code or unintended consequences of a code change whatever. Separate from this issue. You would still want the mechanism in place to not panic the kernel when it did nothing wrong

A WARN means an assertion was tripped in the kernel. That is the definition of doing something wrong.