Re: [net-next PATCH v2 1/5] introduce IFE action

[John Fastabend <john.fastabend@gmail.com>]

On 16-02-25 01:46 PM, Daniel Borkmann wrote:
> On 02/25/2016 01:20 PM, Jamal Hadi Salim wrote:
>> On 16-02-24 12:37 PM, Daniel Borkmann wrote:
>>> On 02/23/2016 01:49 PM, Jamal Hadi Salim wrote:
>>>> From: Jamal Hadi Salim <jhs@mojatatu.com>
>>> [...]
>>>> +static const struct nla_policy ife_policy[TCA_IFE_MAX + 1] = {
>>>> +    [TCA_IFE_PARMS] = {.len = sizeof(struct tc_ife)},
>>>> +    [TCA_IFE_DMAC] = {.type = NLA_BINARY,.len = ETH_ALEN},
>>>> +    [TCA_IFE_SMAC] = {.type = NLA_BINARY,.len = ETH_ALEN},
>>>
>>> This is buggy btw ...
>>
>> I am sure i cutnpasted that from somewhere. Thanks for catching
>> it; I will remove NLA_BINARY ref.
> 
> Yeah, NLA_BINARY seems to be a bit of a misleading name. We should
> probably audit, if there are more such users already in the tree.
> 

At some point in the past (maybe a year ago?) I went through and
fixed a handful of these but yeah it seems to be a common error.

> [...]
>>> Maybe try to make this lockless in the fast path? Otherwise placing
>>> this on ingress / egress (f.e. clsact) doesn't really scale.
>>
>> Let me think about it. Likely it will be subsequent patches - I just
>> want to get this set out first.
> 
> Yes, I mean one of the key motivation was "[...] to horizontally scale
> packet processing at scope of a chasis or rack [...]". So for people
> who don't have that NIC with embedded Cavium processor, they might
> already hit scalability issues for encode/decode right there.
> 
> Thanks again,
> Daniel