Re: [patch net-next v2] ipv6: log autoconfiguration failures

[Vlad Yasevich <vyasevich@gmail.com>]

On 12/12/2013 12:22 PM, David Miller wrote:
> From: Denys Vlasenko <dvlasenk@redhat.com>
> Date: Thu, 12 Dec 2013 12:17:42 +0100
> 
>> I can easily imagine their frustration. Kernel _knows_ why
>> it didn't work, and it's not expected to normally pappen,
>> why didn't it tell anything about it?
> 
> Packets are dropped silently, ARP fails and entries go stale silently,
> none of this is logged with kernel messages, why is ipv6 autoconf so
> unique and important to justify different behavior?

But most of the changes from V2 report an actual error condition in
the kernel: addr_len is not set correctly for the device type.
These changes are worth keeping and not just as pr_debug.  Having
a counter of these failures will not provide much useful data if you
happen to have multiple such device and where one happens to work and
the other doesn't (very unlikely, I know).

> 
> Give it statistics just like we have for every other kind of similar
> event.

The fact that link-local address could not be generated for a specific
device type does not lend itself well to counting.  There is no way
from to tell from the counter which device does not support IPv6 autoconfig.

-vlad