I spoke with the authors and they plan to update the draft with my
suggestion.

With the update I support adoption.

Thanks

Gyan
On Mon, Jul 27, 2026 at 11:33 PM Gyan Mishra <[email protected]> wrote:

> HI Feng & authors
>
> To make the draft more useful with the concept you are trying to do you
> could angle it a different way.
>
> If you change the problem statement to not say asymmetrical routing that
> firewall is dropping the flow as this is not TCP traffic so the problem
> statement is not correct.
>
> You could update the use case and say to avoid DPI as firewalls are not
> SRv6 aware and cannot filter on the payload.  However with this drafts
> solution I can now have 1 flow per CE or per vrf Sid allocation basically
> now you can do per flow level TCP stateful filtering through the firewall
> without requiring SRv6 proxy or firewall supporting SRv6.
>
> If you are able to make the change to the problem statement then I would
> happily support adoption.
>
> Kind Regards
>
> Gyan
> On Fri, Jul 24, 2026 at 2:39 AM Gyan Mishra <[email protected]> wrote:
>
>> Hi Feng
>>
>> Thank you for your patience going through all the relevant questions I
>> have on the draft.
>>
>> There are a lot of things that bother me about  this draft which is why I
>> am speaking up.
>>
>> How was the feedback from Spring WG when you presented the draft ?
>>
>> Were any concerns raised as the ones I am mentioning?
>>
>> I think overall there are two things that bother me from a
>> standardization perspective. Also doing something super corner case that
>> could lead to issues and instability if implemented is what I am worried
>> about.
>> The protocol next hop is the SRv6 tunnel which sits in the global table
>> but now using the source address that is in a VRF and not in the global
>> table to me seems could have some unknown consequences.  I have done lots
>> of R&D testing with SRv6 and I am surprised that this works and would need
>> to analyze further how it’s possible that it works.
>>
>> The second thing  that also really bothers me about this draft is that
>> for each  locator you are now adding more complexity it seems as now you
>> have individual SRv6 tunnel SRv6 encapsulation h.encap.red per VRF source
>> address.  You are now essentially creating per flow state with each tunnel
>> session per locator anchor. That seems to be pretty crazy overhead
>> unnecessary state you are creating where SRv6 is meant to be ultra
>> simplistic and stateless.  You could do multipoint something similar to
>> P2MP RSVP-TE with Leaf AD S2L but way super complex. Imagine you would have
>> a 1-1 mapping per-vrf or per-CE Sid allocation mode to number of SRv6
>> tunnels which is an insane amount of state.
>>
>> See my issue with the problem statement detailed write up as there is a
>> big disconnect as far as TCP sync state and dropping flow versus two
>> unidirectional SRv6 6in6 tunnels in either direction.  Most firewalls don’t
>> support 6in6 DPI but as long payload flow is symmetrical as shown in your
>> diagram nothing gets dropped.
>>
>> I will go over in more detail in my responses in-line.
>>
>> On Thu, Jul 23, 2026 at 11:18 PM <[email protected]> wrote:
>>
>>> Hi Gyan,
>>>
>>>
>>>
>>> We appreciate the time and effort you have put into raising these
>>> concerns. I think this draft does not have the issues that you mentioned.
>>>
>>>
>>>
>>> See my comments in-line.
>>>
>>>
>>>
>>> BR,
>>>
>>> Feng
>>>
>>>
>>>
>>> *发件人:* Gyan Mishra <[email protected]>
>>> *发送时间:* 2026年7月24日 8:54
>>> *收件人:* Robert Raszuk <[email protected]>
>>> *抄送:* [email protected]; spring <[email protected]>;
>>> [email protected]
>>> *主题:* [spring] Re: Call for adoption:
>>> draft-yang-spring-sid-as-source-address
>>>
>>>
>>>
>>> I do not support adoption of this draft for the reasons below:
>>>
>>>
>>>
>>> Major issues with the draft below:
>>>
>>>
>>>
>>> 1st major issue with this subject draft is it proposes using the source
>>> IP of SRv6 tunnel to use  the service sid  endpoint behavior which is part
>>> of IP VPN VRF.  That technically won’t work as just as the loopback IGP
>>> best path selection lowest metric tie breaker next hop must be in the
>>> global table so does the source IP for the tunnel must be in the global
>>> table.
>>>
>>>
>>>
>>> [fyang] We would like to gently point out that the draft does not modify
>>> any control‑plane protocol, and the source address is not used for routing
>>> decisions. In fact, if people try a very simple test that SRv6 tunnel
>>> source address does not in global table, and that works fine. So we believe
>>> there is no requirement for the source address to be in the global table.
>>>
>>>  Gyan>  As I started above I am very surprised this works and would have
>>> to be throughly tested that with this corner case solution it’s not
>>> creating any instability.  Maybe some of what I have mentioned can be added
>>> to a operational considerations section.
>>>
>>> 2nd major issue with this draft is that there is only one source address
>>> you can set for the SRv6 tunnel.
>>>
>>>
>>>
>>> [fyang] Single source address is the current implementation model. This
>>> is unnecessary restriction. We would respectfully note that the why not
>>> move beyond this.
>>>
>>>  Gyan>  For simplicity of a single SRv6 tunnel and not creating x number
>>> of tunnels for every source address per vrf or per CE.
>>>
>>      In section 1 the problem statement is mentioned relating to the
>> issue with asymmetrical flow and the firewall blocking the flow,
>>
>> The SRv6 tunnel is not a TCP or UDP flow,  Most firewalls have TCP state
>> sync enabled for asymmetric flows that if the source packet arrives with
>> TCP SYN then the response packet TCP SYN/AAK should be received in the
>> reverse direction or the flow is dropped.
>> If the TCP SYN/ACK is not received the the flow is dropped by the
>> firewall due to asymmetrical routing.
>>
>> We absolutely do not have asymmetrical routing here as well don’t even
>> have a TCP flow for that matter.
>>
>> SRv6 uses the concept of explicit tunneling vs implicit tunneling as with
>> MPLS using a locator prefix node Sid  that is advertised that draws traffic
>> across the algo 0 or flex also path based on the AG/EAG link affinity
>> defining the path for that plane.  The head SRv6 encap h.encap.red creates
>> the outer header and reads the SRH  based on active Sid or SRv6 compression
>> performs the topological next or replace Sid function.  So the DA address
>> hop by hop is changing for full Sid or SRv6 compression next or replace
>> Sid.  Depending upon where the firewall sits the firewall in the middle of
>> the path would be an ephemeral DA address.
>>
>> So the main point here is the firewall seems encapsulated IPv6 in IPv6
>> packets with epidermal DA address and it’s the same for the unidirectional
>> tunnel in both directions if SR policy has a static or dynamic Sid list
>> steering through the firewall.
>>
>> For both SRv6 full Sid and SRv6 compression the source addres is set to
>> the loopback but what the firewall seems in the middle of the path as shown
>> in figure 1 is an ephemeral DA based on the Sid list for SE policy but if
>> no policy then the DA address would  egress PE Sid + endpoint behavior.
>>
>> However since the tunnels are unidirectional the firewall even with
>> stateful packet inspection would not drop any packets since it’s a one way
>> flow.
>>
>> So you have two one way tunnel flows in opposite directions
>>
>> Forward direction
>>
>> PE1 - FW - PE2  - single undirectional flow
>>
>> Reverse direction
>>
>> PE2 - FW - PE1 - single unidirectional flow
>>
>> Both flows are independent from each other and should never get dropped
>> by the stateful firewall
>>
>> The key here is  stateful firewall is looking for TCP state so the two
>> unidirectional tunnels will pass through not dropped with TCP stateful
>> packet inspection enabled.
>>
>> Most firewalls as I mentioned earlier in our thread do not support DPI
>> for 6 in 6 tunnel however even if DPI was possible and the traffic was
>> routed symmetrically the traffic would not be dropped for inner payload TCP
>> flows.
>>
>> I don’t agree with the problem statement.
>>
>> You are trying to fix something that is not broken and at the same time
>> adding tremendous unnecessary complexity to SRv6.
>>
>>> 3rd major issue is how would that work if you have many endpoint
>>> behaviors per vrf or CE Sid allocation that many Sid addresses in different
>>> vrf which would you even pick.
>>>
>>>
>>>
>>> [fyang] Selection among multiple sid can be found in the subject draft
>>> section 2. And this has been presented in spring WG meeting.
>>>
>>>  Gyan> That is fine however I don’t agree with the problem statement
>>> that anything is broken.
>>>
>>
>> 4th major issue that the SRv6 tunnel carries all the IP VPN  VRFs and not
>>> just the single VRF service Sid endpoint behavior that you decide to set
>>> the source address per this draft.
>>>
>>>
>>>
>>> [fyang] We never propose to use same VRF sid for multipole VRF. While an
>>> SR Policy or tunnel may carry traffic from multiple VRFs, each packet is
>>> encapsulated independently:
>>>
>>>    - Packets belonging to VRF-A use VRF-A's SID as the SA.
>>>    - Packets belonging to VRF-B use VRF-B's SID as the SA.
>>>
>>>
>> Gyan>
>> for each  locator you are now adding more complexity it seems as now you
>> have individual SRv6 tunnel SRv6 encapsulation h.encap.red per VRF source
>> address.  You are now essentially creating per flow state with each tunnel
>> session per locator. That seems to be pretty crazy overhead unnecessary
>> state you are creating where SRv6 is meant to be ultra simplistic and
>> stateless.  You could do multipoint something similar to P2MP RSVP-TE with
>> Leaf AD S2L but way super complex. Imagine you would have a 1-1 mapping
>> per-vrf or per-CE Sid allocation mode to number of SRv6 tunnels which is an
>> insane amount of state.
>>
>>>
>>>
>>
>>> Kind Regards
>>>
>>>
>>>
>>> Gyan
>>>
>>>
>>>
>>> On Thu, Jul 23, 2026 at 4:26 PM Robert Raszuk <[email protected]> wrote:
>>>
>>> Hi,
>>>
>>>
>>>
>>> I have read the draft and support the adoption. It is a well written
>>> and useful document.
>>>
>>>
>>>
>>> It provides a solution to data plane consistency and symmetry for ICMP.
>>>
>>>
>>>
>>> While the draft is not explicitly discussing it - it also actually
>>> solves the problem with BGP NEXT_HOP being accidentally different from
>>> Service SIDs creating all sources of bgp related issues.
>>>
>>>
>>>
>>> The service SID used as source can be in the same time advertised in BGP
>>> protocol NEXT_HOP via proper use of existing update-source knob.
>>>
>>>
>>>
>>> Kind regards,
>>>
>>> Robert
>>>
>>>
>>>
>>>
>>>
>>> On Fri, Jul 10, 2026 at 2:52 PM <[email protected]> wrote:
>>>
>>> Dear WG,
>>>
>>>
>>>
>>> This message starts a 3-week WG adoption call, ending 2026-07-31, for
>>> draft-yang-spring-sid-as-source-address-13 [1]
>>>
>>>
>>> After review of the document, please indicate support (or not) for WG
>>> adoption of the document to the mailing list.
>>> Please also provide comments/reasons for your support (or lack thereof)
>>> as this is a stronger way to indicate your (non) support as this is not a
>>> vote.
>>> If you are willing to work on or review the document, please state this
>>> explicitly. This gives the chairs an indication of the energy level of
>>> people in the working group willing to work on the document.
>>>
>>> Thanks!
>>> Alvaro, Bruno, Joel
>>>
>>>
>>>
>>> [1]
>>> https://datatracker.ietf.org/doc/html/draft-yang-spring-sid-as-source-address-13
>>>
>>>
>>>
>>> ____________________________________________________________________________________________________________
>>>
>>> Ce message et ses pieces jointes peuvent contenir des informations 
>>> confidentielles ou privilegiees et ne doivent donc
>>>
>>> pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu 
>>> ce message par erreur, veuillez le signaler
>>>
>>> a l'expediteur et le detruire ainsi que les pieces jointes. Les messages 
>>> electroniques etant susceptibles d'alteration,
>>>
>>> Orange decline toute responsabilite si ce message a ete altere, deforme ou 
>>> falsifie. Merci.
>>>
>>>
>>>
>>> This message and its attachments may contain confidential or privileged 
>>> information that may be protected by law;
>>>
>>> they should not be distributed, used or copied without authorisation.
>>>
>>> If you have received this email in error, please notify the sender and 
>>> delete this message and its attachments.
>>>
>>> As emails may be altered, Orange is not liable for messages that have been 
>>> modified, changed or falsified.
>>>
>>> Thank you.
>>>
>>> _______________________________________________
>>> spring mailing list -- [email protected]
>>> To unsubscribe send an email to [email protected]
>>>
>>> _______________________________________________
>>> spring mailing list -- [email protected]
>>> To unsubscribe send an email to [email protected]
>>>
>>>
_______________________________________________
spring mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to