Denis V. Lunev wrote:
> Daniel Lezcano wrote:
>> Denis V. Lunev wrote:
>>> Daniel Lezcano wrote:
>>>> Denis V. Lunev wrote:
>>>>> I could hardly imagine why sombady needs to assign 0.0.0.0 as an
>>>>> interface
>>>>> address or interface destination address. The kernel will behave in a
>>>>> strage
>>>>> way in several places if this is possible, as ifa_local != 0 is
>>>>> considered
>>>>> as initialized/non-initialized state of the ifa.
>>>> AFAICS, we should be able to set at an interface address to 0.0.0.0, in
>>>> order to remove an IP address from an interface and keep this one up.
>>>> I see two trivial cases:
>>>>  * remove the ipv4 on an interface but continue to use it through ipv6
>>>>  * move ipv4 address from the interface to an attached bridge
>>> For this case there is an IOCTL/netlink "remove IP address".
>> And I forgot to mention the general broadcast.
>> This is need for the dhcp protocol. If you are not able to set your
>> interface to 0.0.0.0, you will be not able to send a 255.255.255.255
>> broadcast message to have your IP address.
>>
> 
> OK. Dave, pls disregard this patch. I suspect that others in the set
> should not intersect with this one.
> 
> To summarize the discussion:
> there is the only reason for this assignment: old IOCTL interface does
> not have a way to remove IP address except this, though netlink has a
> method for it that's why I am a little bit confused :)
> 
> This is handled in the __inet_insert_ifa: ifa is just removed there and,
> correctly, ifa with 0.0.0.0 address can't exists in the kernel.

Yes, my last statement is false.