[DRBD-user] WARNING: at net/core/skbuff.c:398 skb_release_head_state+0x64/0xc8

[Massimo Cetra]

Lars Ellenberg ha scritto:
>> [60257.728500] ------------[ cut here ]------------
>> [60257.728500] WARNING: at net/core/skbuff.c:398 skb_release_head_state+0x64/0xc8()
>>     
>
> This is _NOT_ and oops.
> it is a _WARNING_.
>
> different thing.
>
>   
Well sorry, but anyway it was a problem :-D
> specifically, it is this warning:
> static void skb_release_head_state(struct sk_buff *skb)
> {
>         skb_dst_drop(skb);
> #ifdef CONFIG_XFRM
>         secpath_put(skb->sp);
> #endif
>         if (skb->destructor) {
>                 WARN_ON(in_irq());		<=====
>                 skb->destructor(skb);
>         }
>
> so apparently virtio_net does something
> that skb_release_head_state does not expect,
> namely destruct from irq context.
>
> No, I don't think DRBD has anything in particular to do with these
> warnings, we do not touch skbs directly.
>
> If it comes down to pointing fingers,
> from the information at hand, it looks like an issue with virtio_net.
>
>   
Thanks, Lars.
You're right.
It seems to be a problemin virtio_net.
See this post:  http://marc.info/?l=linux-netdev&m=125542939211671
where Eric Dumazet points out the problem and proposes a patch.
I'm testing it right now.

> possibly it is even because you try to use netconsole and virtio_net
> at the same time, and they do interfere?
>
>   
No, i added netconsole because the server was rebooting (drbd was 
loosing sync and was triggering a reboot).
The problem was there before netcosole was used.

> using netconsole to debug network issues may not be the smartest move.
> don't do that, then ;)
>   
It worked this time, fortunately.
That was only one of the ways i was trying to debug the problem.

It's very weird that no one has noticed that Warning before.
I've been using KVM for a long time and i bet that a lot of users use 
DRBD with KVM and virtio drivers like i do.

I hope that that patch has fixed the problem.

Thanks again, Lars.

Max