Re: [PATCH net] net: enable interface alias removal via rtnl

[David Ahern <dsahern@gmail.com>]

On 10/9/17 9:25 AM, Nicolas Dichtel wrote:
> Le 09/10/2017 à 16:02, David Ahern a écrit :
>> On 10/9/17 2:23 AM, Nicolas Dichtel wrote:
>>> Le 06/10/2017 à 22:10, Oliver Hartkopp a écrit :
>>>>
>>>>
>>>> On 10/06/2017 08:18 PM, David Ahern wrote:
>>>>> On 10/5/17 4:19 AM, Nicolas Dichtel wrote:
>>>>>> IFLA_IFALIAS is defined as NLA_STRING. It means that the minimal length of
>>>>>> the attribute is 1 ("\0"). However, to remove an alias, the attribute
>>>>>> length must be 0 (see dev_set_alias()).
>>>>>
>>>>> why not add a check in dev_set_alias that if len is 1 and the 1
>>>>> character is '\0' it means remove the alias?
>>> Because it requires an iproute2 patch. iproute2 doesn't send the '\0'. With the
>>> command 'ip link set dummy0 alias ""', the attribute length is 0.
>>
>> iproute2 needs the feature for 0-len strings or perhaps a 'noalias' option.
> iproute2 needs nothing ...
> 
>>
>> You can reset the alias using the sysfs file. Given that there is a
>> workaround for existing kernels and userspace, upstream can get fixed
>> without changing the UAPI.
>>
> I don't get the point with the UAPI. What will be broken?

never mind; I see the error of my ways.

> I don't see why allowing an attribute with no data is a problem.
> 

I remember the problem now. I made a patch back in March 2016 that
adjusted the policy validation to allow 0-length string. I never sent it
and forgot about it until today. You changing ifla_policy to NLA_BINARY
is achieving the same thing.

I think a comment above the policy line is warranted that clarifies
IFLA_IFALIAS is a string but to allow a 0-length string to remove the
alias NLA_BINARY is used for policy validation.

Comparing the validation done for NLA_STRING vs NLA_BINARY it does
change the behavior for 256-character strings.