Re: [spring] spring WG Adoption Call for draft-cheng-dhc-distribute-srv6-locator-by-dhcp

[Weiqiang Cheng <[e-mail]>]

Hi Joel & Robert,
Thank you for your valuable comments.
We will add some explicit text in the new version to state that this draft aligns with the underlying SRv6 RFCs when the DHCP is used.

B.R.
Weiqiang Cheng
 
From: Joel Halpern
Date: 2023-07-15 02:25
To: Robert Raszuk
CC: Weiqiang Cheng; wangaijun; spring; [e-mail]
Subject: Re: [spring] spring WG Adoption Call for draft-cheng-dhc-distribute-srv6-locator-by-dhcp
My assumption in suggesting the solution I did was that the DHCP server would be configured properly and would comply with what we say the requirements are.  I am not asking the protocol to enforce the domain.  I am asking that the document be clear about when this DHCP usage is consistent with the underlying RFCs.  And to tell folks not to violate those RFCs.
If the authors or the WG want to take it even further, that is their right / privilege. 
Yours,
Joel
On 7/14/2023 2:19 PM, Robert Raszuk wrote:

Ok last email on this ... let's talk IETF and standardization as you prefer.  

You asked: 

Speaking as an individual participant, it seems to me that the description of using CPE as SRv6 endpoints needs to state explicitly and clearly that this assumes that the CPE are managed by the access operator, and that all of the relevant endpoints are part of a single operator / domain.

But subject draft is defining two new DHCPv6 options OPTION_IA_SRV6_LOCATOR & OPTION_IALOCATOR. No more no less. There is no OPTION_LIMITED_DOMAIN_MEMBER which when received the former two can be processed.

So any client will be able to get those options. 

If you (or others) would be really trying to make this work standards wise you would define an extension to protocol(s) to indicate that the querying node is part of what you called a "limited domain" or not. Today to the best of my understanding such boundary is protocol wise undefined. 

Otherwise irrespective if authors add proposed text or not the behaviour will be identical in practice. 

Kind regards,
Robert


On Fri, Jul 14, 2023 at 6:27 PM Joel Halpern <[e-mail]> wrote:
You may not subscribe to the notion of limited domains.  But the RFCs do.  Hence, the WG is required to do so.  An operator can of course do anything they want, and they choose their risks.  But IETF standardization work is subject to the restrictions we as a community agree to.
Yours,
Joel
On 7/14/2023 12:25 PM, Robert Raszuk wrote:
> then anyone, almost anywhere in the Internet, can inject such packets.

Sure can. And just like without SRv6 I am protecting my hosts from such injections. 

See SRv6 has a very broad range of use cases. But it does not promise a secure and encrypted pipe (at least last time I checked). There are better protocols focusing just to do that. 

So imagine I want to enhance my packets between v6 speaking hosts with demux value so when the packet arrives it goes to its own dedicated namespace. And as before I am protecting such namespaces from attacks the same way I protect my default namespace (as example I do have on secure machines a whitelisted set of rules who can talk to me directly). 

That is why I never subscribed to this notion of limited domain in respect to building SRv6 walls :)  

Cheers,
Robert


On Fri, Jul 14, 2023 at 6:15 PM Joel Halpern <[e-mail]> wrote:
If a deployment of SRv6 uses unsecured tunnels as a means to deliver SRv6 packets between pieces of the domain, then anyone, almost anywhere in the Internet, can inject such packets.  That is not a limited domain.  It is a wide open domain.
Yours,
Joel
On 7/14/2023 12:01 PM, Robert Raszuk wrote:

Why would you expect this to always unconditionally use "secure tunnels" ? For lot's of applications basic encapsulation is sufficient. Note that I mentioned the case where my sites are connected over ISP not the Internet. 

Besides if there is concern about data protection or data integrity lot's of applications use app level encryption so asking for transport tunnel security would be redundant at best.  

Thx,
R.

On Fri, Jul 14, 2023 at 5:42 PM Joel Halpern <[e-mail]> wrote:
If they want to specify in the draft that the CPE are operator managed by a single operator, and that they use secure tunnels (not just arbitrary overlay) between the BRAS then we can as a WG see if that suffices.  But the draft doesn't currently say that.
Yours,
Joel
On 7/14/2023 11:40 AM, Robert Raszuk wrote:
Joel,  

Limited domain (if you will) can be also build in the overlay - unless you point me to any standards track RFC which says otherwise. If I connect over ISP with v6 I can run my own limited domain over such ISP between hosts in my sites. 

Of course ISP may recognize that I am actually sending v6 packets with extension headers and drop those ... but this is not a concern for this thread nor for IETF. In such a case I will would switch an ISP :) 

Cheers,
R.





On Fri, Jul 14, 2023 at 5:33 PM Joel Halpern <[e-mail]> wrote:
Nope.
The host case for SRv6 is for hosts run by the operator within their domain.  8986 explicitly requires that SRv6 be used only within a limited domain.
It is clear from other comments that you do not like this restriction.  That is your priviledge as an individual.  But IETF work has to respect the restriction.
Yours,
Joel
On 7/14/2023 11:28 AM, Robert Raszuk wrote:
Joel, 

Point 1: 

Reading RFC8986 I see in section 4.16.2 clear definition of SRv6 running on user hosts. 

One of the applications of the USP flavor is when a packet with an SRH is destined to an 
application on hosts with smartNICs ("Smart Network Interface Cards") implementing 
SRv6. The USP flavor is used to remove the consumed SRH from the extension header 
chain before sending the packet to the host.

Point 2: 

If two hosts want for whatever reason to use SRv6 over the provider's IPv6 underlay (and they may possibly benefit from DHCP allocation of locators) they can happily do so and I see nowhere in any SRv6 spec any mandate which would state a word against such deployment model. 

Thx,
Robert

PS, This entire notion of "limited domain" is as you very well know the history is not a very technical thing ... 


On Fri, Jul 14, 2023 at 5:08 PM Joel Halpern <[e-mail]> wrote:
Speaking as an individual participant, it seems to me that the description of using CPE as SRv6 endpoints needs to state explicitly and clearly that this assumes that the CPE are managed by the access operator, and that all of the relevant endpoints are part of a single operator / domain.  Otherwise, this usage violates the base rules for SRv6.
Personally, I would like to see this fixed before adoption completes.
Yours,
Joel
On 7/14/2023 9:44 AM, Joel Halpern wrote:
I probably need to re-read the draft.  Does the draft assume the CPE is part of the operator domain?  Remember that SRv6 MUST be used ONLY within a limited domain?
Yours,
Joel
On 7/14/2023 2:22 AM, Weiqiang Cheng wrote:
Hi Aijun,
Thank you very much for your comments. 
We will add some text to describe that the CPE utilizes locators obtained through DHCP to provide differentiated services.

Best Regards
Weiqiang Cheng
 
发件人： Aijun Wang
发送时间： 2023-07-14 11:19
收件人： 'Alvaro Retana'; [e-mail]
抄送： [e-mail]; [e-mail]
主题： 答复: [spring] spring WG Adoption Call for draft-cheng-dhc-distribute-srv6-locator-by-dhcp
Support its adoption.
 
It will be more convincible to add some descriptions in the document that the CPE itself needs to obtain different IPv6 address that can differentiate the services that it can provide.
 
 
Best Regards
 
Aijun Wang
China Telecom
 
-----邮件原件-----
发件人: [e-mail] [mailto:[e-mail]] 代表 Alvaro Retana
发送时间: 2023年7月5日 20:00
收件人: [e-mail]
抄送: [e-mail]; [e-mail]
主题: [spring] spring WG Adoption Call for draft-cheng-dhc-distribute-srv6-locator-by-dhcp
 
Dear WG:
 
This message starts a two-week adoption call for draft-cheng-dhc-distribute-srv6-locator-by-dhcp, ending on July/19.
It "describes the method of assigning locators to SRv6 Endpoints through DHCPv6".
 
   https://datatracker.ietf.org/doc/draft-cheng-spring-distribute-srv6-locator-by-dhcp/
 
 
Please review the draft and consider whether you support its adoption by the WG.  Please share any thoughts with the list to indicate support or opposition -- this is not a vote.
 
If you are willing to provide a more in-depth review, please state it explicitly to give the chairs an indication of the energy level in the working group willing to work on the document.
 
WG adoption is the start of the process.  The fundamental question is whether you agree the proposal is worth the WG's time to work on and whether this draft represents a good starting point.  The chairs are particularly interested in hearing the opinion of people who are not authors of the document.
 
Thanks!
 
Alvaro (for the Chairs)
 
_______________________________________________
spring mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/spring
 

_______________________________________________
spring mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/spring

_______________________________________________
spring mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/spring

_______________________________________________
spring mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/spring