Re: [PATCH net v2] ipv6: Handle race in addrconf_dad_work

[Eric Dumazet]

On 10/2/19 5:10 PM, David Ahern wrote:
> On 10/2/19 4:36 PM, Eric Dumazet wrote:
>> This might be related to a use of a bonding device, with a mlx4 slave.
>>
> 
> does it only happen with bonds?

All my hosts have bonds, some are just fine with your patch, but others are not.

> 
> bond shows IF_READY even though the underlying device is carrier down
> which seems wrong; if a lower device is not carrier up then DAD does not
> really happen.
> 
> A quick test with a VM (and setting carrier off on the tap) shows that
> with the current patch in place the bond address stays in tentative
> state. Reverting this patch and the tentative flag is removed. DAD never
> happens since carrier is off on the one link in the bond, so the
> tentative flag getting cleared is wrong.
> 
> I need to step out; come back to this later.
> 

Note that in my setup nodad flag is requested.