Re: [RFC V1 1/1] net: extend VXLAN header to support more extensions

[Gavin Li <gavinl@nvidia.com>]

On 2/9/2024 11:32 PM, Thomas Monjalon wrote:
> 09/02/2024 15:58, Ferruh Yigit:
>> On 2/9/2024 1:44 PM, Thomas Monjalon wrote:
>>> 09/02/2024 13:11, Ferruh Yigit:
>>>> On 2/9/2024 10:12 AM, Thomas Monjalon wrote:
>>>>> 09/02/2024 00:54, Ferruh Yigit:
>>>>>> On 1/30/2024 11:25 AM, Gavin Li wrote:
>>>>>>> Currently, DPDK supports VXLAN and VXLAN-GPE with similar header
>>>>>>> structures and we are working on adding support for VXLAN-GBP which is
>>>>>>> another extension to VXLAN. More extension of VXLAN may be added in the
>>>>>>> future.
>>>>>>>
>>>>>>> VXLAN and VXLAN-GBP use the same UDP port(4789) while VXLAN-GPE uses a
>>>>>>> different one, 4790. The three protocols have the same header length and
>>>>>>> overall similar header structure as below.
>>>>>>>      0                   1                   2                   3
>>>>>>>      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>     |R|R|R|R|I|R|R|R|            Reserved                           |
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>     |                VXLAN Network Identifier (VNI) |   Reserved    |
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>
>>>>>>>                             Figure 1: VXLAN Header
>>>>>>>
>>>>>>>      0                   1                   2                   3
>>>>>>>      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>     |R|R|Ver|I|P|B|O|       Reserved                |Next Protocol  |
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>     |                VXLAN Network Identifier (VNI) |   Reserved    |
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>
>>>>>>>                           Figure 2: VXLAN-GPE Header
>>>>>>>
>>>>>>>      0                   1                   2                   3
>>>>>>>      0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>     |G|R|R|R|I|R|R|R|R|D|R|R|A|R|R|R|        Group Policy ID        |
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>     |          VXLAN Network Identifier (VNI)       |   Reserved    |
>>>>>>>     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>>>>>>>
>>>>>>>                            Figure 3: VXLAN-GBP Extension
>>>>>>>
>>>>>>> Both VXLAN-GPE and VXLAN-GBP extended VXLAN by redefining its reserved
>>>>>>> bits, which means the packets can be processed with same pattern and most
>>>>>>> of the code can be reused. Instead of adding more new items by
>>>>>>> copying/pasting code for the VXLAN extensions in the future, it’s better
>>>>>>> to use existing VXLAN infrastructure and add support code in it.
>>>>>>>
>>>>>>
>>>>>> Hi Gavin,
>>>>>>
>>>>>> The motivation is to prevent code duplication, and the code mentioned is
>>>>>>   the driver code, right?
>>>>>
>>>>> The motivation is mainly to provide a unified and more explicit API.
>>>>>
>>>>
>>>>  From user perspective, I think existing approach is more explicit,
>>>> because it sets VXLAN or VXLAN_GPE flow types.
>>>>
>>>> I am trying to understand the benefit, how unifying flow type in the API
>>>> helps to the user?
>>>>
>>>>>> Overall OK to unify "struct rte_vxlan_hdr" although it makes the struct
>>>>>> a little complex, perhaps we can consider extraction some nested structs
>>>>>> as named struct, no strong opinion.
>>>>>>
>>>>>>
>>>>>> But not sure about removing the flow item types for VXLAN-GPE, or not
>>>>>> adding for VXLAN-GBP.
>>>>>>
>>>>>> Think about a case user adding a rule, which has a item type as VXLAN
>>>>>> and in the protocol header some bits are set, lets say first word, last
>>>>>> byte is set, how driver will know if to take it as GPE "next protocol"
>>>>>> or "group policy id".
>>>>>
>>>>> The driver may decide depending on the UDP port and some distinguishing flags.
>>>>> If you want to match on GBP, you should includes the GBP flag in your pattern,
>>>>> no need to use a separate item.
>>>>>
>>>>
>>>> Why not be more explicit?
>>>> It helps to driver to know more about the pattern to be able to create
>>>> proper flow rule, if there is an obvious way for driver to differentiate
>>>> these protocol extensions, and flow item type is redundant, I can
>>>> understand the proposal, but otherwise I think better to keep flow items
>>>> for extensions.
>>>
>>> In any case we need the simple VXLAN item.
>>> If we have GPE and GBP specialized items,
>>> what means a match on the simple VXLAN?
>>> Does it include packets with other extensions or exclude them?
>>> Matching the bits in the protocol make such decision explicit.
>>>
>>>> When a rule is set in HW, HW may not care about the protocol, as long as
>>>> bits in the rule match with the packet, HW can apply the action.
>>>> But for driver to be able to set the rule properly, it needs more
>>>> explicit information.
>>>
>>> Yes information is in the pattern to match.
>>>
>>>> Lets assume driver API get a pattern with 'RTE_FLOW_ITEM_TYPE_VXLAN'
>>>> type and "struct rte_flow_item_vxlan", at this point driver doesn't know
>>>> if it is VXLAN or any of the extensions.
>>>
>>> Yes it knows because of the matched bits in the pattern.
>>> If the rule specify a match on GBP flag = 1, it is GBP only.
>>> If the rule specify a match on GBP flag = 0, it excludes GBP.
>>> If the rule does not mask GBP flag, it includes GBP.
>>>
>>
>>
>> OK, VXLAN-GBP protocol has a GBP flag that gives a way to differentiate
>> the extension, so flow item for it becomes redundant and we can get rid
>> of it.
> 
> Yes I think so.
> 
>> Is it same for the other extensions?
>> If we use VXLAN flow item and by setting specific field in pattern can
>> we differentiate VXLAN and any other extension?
>> Or in some cases other information, like UDP port, needs to be taken
>> into account to differentiate protocol/extension?
> 
> For VXLAN-GPE, differentiation is on UDP port.
> Remember we have an API to fill some UDP ports:
> rte_eth_dev_udp_tunnel_port_add with RTE_ETH_TUNNEL_TYPE_VXLAN_GPE
> 
> The UDP port value/mask may be part of the flow rule pattern.
> 
> 
>> I found a spec for VXLAN-GBP, but it shows as sub-header for VXLAN-GPE,
>> different than what this RFC describes:
>> https://datatracker.ietf.org/doc/html/draft-lemon-vxlan-gpe-gbp
>>
>> Can you please share link for VXLAN-GBP Extension spec?
> 
> I will let Gavin explain here, I'm not an expert.

Sorry for the wrong format of my last mail. Let me try again.

The RFC mentioned, aka draft-lemon-vxlan-gpe-gbp was to define VXLAN-GBP
as a sub header of VXLAN-GPE by assigning a next protocol num of 
VXLAN-GPE for VXLAN-GBP, which has been obsoleted since 
draft-ietf-nvo3-vxlan-gpe-09, the latest is 
draft-ietf-nvo3-vxlan-gpe-13. It was temporarily valid from 
draft-ietf-nvo3-vxlan-gpe-05 to draft-ietf-nvo3-vxlan-gpe-08.

The VXLAN-GBP we are discussing in this thread is an VXLAN extension 
parallel to VXLAN-GPE.

The RFC introduced VXLAN-GBP sub header to VXLAN-GPE is 
https://datatracker.ietf.org/doc/html/draft-ietf-nvo3-vxlan-gpe-05.

The RFC obsoleted the sub header is 
https://datatracker.ietf.org/doc/html/draft-ietf-nvo3-vxlan-gpe-09.

The latest RFC of VXLAN-GPE is 
https://datatracker.ietf.org/doc/html/draft-ietf-nvo3-vxlan-gpe-13.


> 
>