Re: [lisp] Review of RFC6833bis

[Dino Farinacci <[e-mail]>]

> On Nov 29, 2017, at 11:09 AM, Victor Moreno (vimoreno) <[e-mail]> wrote:
> 
> In addition to the comments provided by Alberto and Dino, I have some comments on the text in section 5 of the RFC. 
> 
> 1. The text in section 5 currently prescribes that in response to a map-request for a non-registered or non-existent EID, an ITR should expect an NMR from the Map-Server (or Resolver) with action code “Natively-Forward” (this is stated in section 5.1 and re-stated in 5.3 and 5.4). Prescribing that this is to always be an NMR with a Native-Forward action is limiting. 

I can go along with this Victor. I would like to hear what the working group says about this. See technical details below.

I have also asked myself the question many times if we need or should distinguish between not-registered or not-assigned. We did this in DDT with Map-Referrals but not in negative Map-Replies.

The problem with “not-assigned” is that a big EID-prefix block may be configured in a map-server with “accept-more-specifics” semantics. Which means you don’t know if the more specifics are assigned or not. So what would always be returned is a “not-registered” action.

So I think we could introduce this, I think in practice it could never be used. And as you say a negative prefix that is built for a hole that needs to point to a PETR, is that a EID-prefix or not? So it would be hard to distinguish between not-assigned here versus not-registered unless the EID-record had a flag that said what is being registered.

> After trying to address a few use cases, it would be very useful if the specification allowed:
> 
> a. As an alternative to responding with an NMR, an implementation may respond to a Map-Request for a non-registered or non-existent EID with a “positive" map-reply inclusive of an RLOC-set that can be configured in the Mapping-System. The immediate use case this would address is that of not having to configure a PETR statically at every ITR, but have the map-replies include the PETR's RLOC information. This Map-Reply should have short TTLs and include covering/hole EID prefixes just as specified for the current NMRs, hence this differs from a map-reply for registered 0/0 EIDs.

Why configure a map-server with something it may not have control over. An mapping system provider may not be associaed with data-plane provider, so it unilaterally (or even if agreed upon) should register *to itself*. There is no reason why the data-plane provider register any length EID-prefix with any RLOC-set it determines. Then a LISP site just encapsulates. 

The architecture already supports this.

> b. The use of all action codes possible with the ACT bits (as defined in section 4.4) when Map-Replying with an NMR to a Map-Request for a non-registered or non-existent EID. As I read the text in section 5, it isn’t clear that other action codes are allowed, the text is focused solely on Natively-Forward, there are strong requirements to leverage the Drop action and particularly Drop/Policy-Denied.

As well as Auth-Failure. I can make this more clear in the section describing the ACT bits.

> 2. There are multiple references to LISP-ALT in the text. Do we want to keep those references moving forward? Or can these be removed? The implementations I work with do not use ALT, but maybe there are others that do?

Since LISP-ALT and LISP-DDT are the only RFC’ed mapping database transport systems, we should include both IMO. What do you think? We have done this for other areas as well.

> 3. Section 5.3. Paragraph 2 is describing the behavior for the MS when there is a match on an EID “hole”. Section 5.3. suggests a 15 minute TTL, meanwhile section 5.1. suggests a 1 minute TTL for the same EID. I think it should be 1 minute in both cases. 

Will fix.

> I can contribute with text suggestions once we discuss the above on the list. 

Thanks for that. If you want to take a crack at (b) above that would be great.

Dino

> 
> -v
> 
> 
>> On Nov 29, 2017, at 10:04 AM, Alberto Rodriguez-Natal <[e-mail]> wrote:
>> 
>> Thanks for the responses Dino. I'll get back to you later this week.
>> 
>> Alberto
>> 
>> On Tue, Nov 28, 2017 at 5:12 PM, Dino Farinacci <[e-mail]> wrote:
>>>> Hi all,
>>>> 
>>>> Wanted to send this before the meeting on Friday. I just completed a
>>>> review of 6833bis, you can find my comments below. Like last time,
>>>> extracts from the draft are copied and followed by my comments.
>>>> 
>>>> Thanks,
>>>> Alberto
>>> 
>>> Thanks again Alberto for your comments. See my responses inline and a -07 diff file.
>>> 
>>>> Map-Resolver:  A network infrastructure component that accepts LISP
>>>> Encapsulated Map-Requests,
>>>> 
>>>>  • [AR] We could remove "Encapsulated" and just use "Map-Requests".
>>>> A Map-Resolver may accept non-encapsulated Map-Requests as well.
>>> 
>>> No, they are “control-plane” encapsulated. I’ll make that more clear.
>>> 
>>>> Map-Register message:   A LISP message sent by an ETR to a Map-Server
>>>> to register its associated EID-Prefixes.  In addition to the set of
>>>> EID-Prefixes to register, the message includes one or more RLOCs to be
>>>> used by the Map-Server when forwarding Map-Requests (re-formatted as
>>>> Encapsulated Map-Requests) received through the database mapping
>>>> system.
>>>> 
>>>>  • [AR] This may give the impression that the RLOCs on the
>>>> Map-Register are only to forward Map-Requests, which is not the case
>>>> in proxy-reply mode. I would suggest we rephrase this text as follows:
>>>> "In addition to the set of EID-Prefixes to register, the message
>>>> includes one or more RLOCs to be used to reach the ETR. The Map-Server
>>>> uses these RLOCs when it has to forward Map-Requests (potentially
>>>> re-formatted as Encapsulated Map-Requests) received through the
>>>> database mapping system.”
>>> 
>>> I reworded.
>>> 
>>>> Map-Notify message:   A LISP message sent by a Map-Server to an ETR
>>>> 
>>>>  • [AR] I would replace "ETR" with "xTR" so we cover the PubSub
>>>> behavior as well.
>>> 
>>> Can’t do that because an ITR does not get Map-Notifies as a response to a Map-Register. But I will add that ITRs get Map-Notifies to inform them of RLOC-set changes (for pubsub and signal-free-multicast use-cases).
>>> 
>>>> 
>>>> For definitions of other terms -- notably Map-Request, Map-Reply,
>>>> Ingress Tunnel Router (ITR), and Egress Tunnel Router (ETR) -- please
>>>> consult the LISP specification [I-D.ietf-lisp-rfc6830bis].
>>>> 
>>>>  • [AR] I think that the definitions for Map-Request and Map-Reply
>>>> should be moved here, and probably we should include the definition
>>>> for Map-Notify-Ack as well. 6830bis should reference 6833bis for
>>>> control-plane messages, not the other way around.
>>> 
>>> They did. But the text you identified above was not changed. Changed now.
>>> 
>>>> A Map-Register message contains a list of EID-Prefixes plus a set of
>>>> RLOCs that can be used to reach the ETR when a Map-Server needs to
>>>> forward a Map-Request to it.
>>>> 
>>>>  • [AR] Since proxy-reply is a common case, I'd not constrain the
>>>> meaning of the RLOCs in the Map-Register. I'd remove the last part of
>>>> the sentence that says "when a Map-Server needs to forward a
>>>> Map-Request to it.”
>>> 
>>> Agree.
>>> 
>>>> A Map-Resolver receives Encapsulated Map-Requests from its client ITRs
>>>> and uses a mapping database system to find the appropriate ETR to
>>>> answer those requests.
>>>> 
>>>>  • [AR] A MR can receive Map-Requests that don't come from ITR
>>>> and/or that are not encapsulated. If we don't want to change the text,
>>>> at least I'd add at the beginning: "In a common scenario, a
>>>> Map-Resolver…"
>>> 
>>> I don’t think we should change this. The lig document indicates that a lig client can send Map-Requests. This document is for support of a data-plane.
>>> 
>>>> Note that while it is conceivable that a non-LISP-DDT Map-Resolver
>>>> could cache responses to improve performance,
>>>> 
>>>>  • [AR] The discussion on caching or not at the Map-Resolver goes
>>>> beyond DDT. We could rephrase this removing the mention to
>>>> "non-LISP-DDT”.
>>> 
>>> Agree. Changed.
>>> 
>>>> The LISP UDP-based messages are the Map-Request and Map-Reply
>>>> messages.  When a UDP Map-Request is sent, the UDP source port is
>>>> chosen by the sender and the destination UDP port number is set to
>>>> 4342.  When a UDP Map-Reply is sent, the source UDP port number is set
>>>> to 4342 and the destination UDP port number is copied from the source
>>>> port of either the Map-Request or the invoking data packet.
>>>> 
>>>>  • [AR] I would remove the first sentence and re-phrase this
>>>> paragraph as follows: "When a UDP LISP control message is sent and is
>>>> not a direct reply to a previous control message, the UDP source port
>>>> is chosen by the sender and the destination UDP port number is set to
>>>> 4342.  When a UDP LISP control message is sent as a direct reply to a
>>>> previous message, the source UDP port number is set to 4342 and the
>>>> destination UDP port number is either set to 4342 or copied from the
>>>> source port of the invoking LISP control message. See the following
>>>> subsections for details.”
>>> 
>>> This is not really true. There are too many cases with respect to NAT-traversal that make things more complicated. I think we need to be specific about each message and not generalize it. Plus, Map-Notify messages are response and unsolicited so they vary on whether the destination or source port is 4342.
>>> 
>>> I have added text to identify each message that follow the Map-Request and Map-Reply port rules.
>>> 
>>>> The UDP checksum is computed and set to non-zero for Map-Request,
>>>> Map-Reply, Map-Register, and Encapsulated Control Message (ECM)
>>>> control messages.
>>>> 
>>>>  • [AR] Shouldn't it be computed for all LISP control messages?
>>> 
>>> Yes. Added text.
>>> 
>>>> EID-Prefix:  This prefix is 4 octets for an IPv4 address family and 16
>>>> octets for an IPv6 address family.  When a Map-Request is sent by an
>>>> ITR because a data packet is received for a destination where there is
>>>> no mapping entry, the EID-Prefix is set to the destination IP address
>>>> of the data packet, and the 'EID mask-len' is set to 32 or 128 for
>>>> IPv4 or IPv6, respectively.
>>>> 
>>>>  • [AR] We should probably rephrase this to don't limit it to IPv4/6
>>> 
>>> Good point. Generalized the text a bit.
>>> 
>>>> For the latter two cases, the destination IP address used for the
>>>> Map-Request is one of the RLOC addresses from the Locator-Set of the
>>>> Map-Cache entry.
>>>> 
>>>>  • [AR] To refresh map-caches before TTL expiration, the
>>>> destination IP of the Map-Request can be the address of the Map-Server
>>>> if in proxy-reply. This should be considered here.
>>> 
>>> This is not true. The Map-Request has an IP header but it is encapsulated in an ECM and that IP header destination address is the *Map-Resolvers*.
>>> 
>>>> If the ITR erroneously provides no ITR-RLOC addresses, the Map-Replier
>>>> MUST drop the Map-Request.
>>>> 
>>>>  • [AR] This could probably be a SHOULD since we use AFI = 0 as
>>>> ITR-RLOC to unsubscribe in PubSub.
>>> 
>>> An AFI=0 Is an RLOC address.
>>> 
>>>> EID-Prefix:  This prefix is 4 octets for an IPv4 address family and 16
>>>> octets for an IPv6 address family.
>>>> 
>>>>  • [AR] We should mention the possibility for address families
>>>> other than IPv4/6.
>>> 
>>> Have that covered.
>>> 
>>>> The RLOCs in the Map-Reply are globally routable IP addresses of all
>>>> ETRs for the LISP site.
>>>> 
>>>>  • [AR] We should remove "globally" here. Maybe also add a
>>>> "Generally" at the beginning since we might have LCAFs with AFI = 0
>>>> (LISP-VPN encoding of Home-IID for instance).
>>> 
>>> Removed “globally”. I don’t understand your other “Generally” comment.
>>> 
>>>> For example, a requester with two cached EID-Prefixes that are covered
>>>> by a Map-Reply containing one less-specific prefix replaces the entry
>>>> with the less-specific EID-Prefix.
>>>> 
>>>>  • [AR] Not sure if I follow here. Does this mean that a
>>>> less-specific received in a Map-Reply will erase from the map-cache
>>>> previously cached more-specifics that are covered by the
>>>> less-specific?
>>> 
>>> Yes, because if the Map-Reply returned a more-specific with the less-specific, then it would be replaced so the less-specific replaces the more-specifics that ARE NOT in the Map-Reply.
>>> 
>>>> When more than one EID-Prefix is returned, all SHOULD use the same
>>>> Time to Live value so they can all time out at the same time.  When a
>>>> more-specific EID-Prefix is received later, its Time to Live value in
>>>> the Map-Reply record can be stored even when other less-specific
>>>> entries exist.
>>>> 
>>>>  • [AR] We should explain in which cases a more-specific can be
>>>> received later.
>>> 
>>> I don’t follow. Each EID-record will contain a TTL for the length prefix that is encoded. So the new Map-Reply TTL will be used for any length entry (and in this case the more-specific).
>>> 
>>>> The Locator-Set MUST be sorted in order of ascending IP address where
>>>> an IPv4 locator address is considered numerically 'less than' an IPv6
>>>> locator address.
>>>> 
>>>>  • [AR] LCAF addresses (maybe with AFI=0) should be discussed here as well.
>>> 
>>> It is discussed in the LCAF draft. Don’t want to repeat it.
>>> 
>>>> Nonce:  This 8-octet 'Nonce' field is set to 0 in Map-Register messages.
>>>> 
>>>>  • [AR] Since there may be future cases that benefit from having a
>>>> non-zero nonce on the Map-Register, I would suggest to rephrase this
>>>> sentence to add a CAN.
>>> 
>>> Not until we identify and describe the use-case. No point to speculate. But note, if a Map-Notify is returned, the nonce should be non-zero. I’ll fix that.
>>> 
>>>> E:  This is the to-ETR bit.  When set to 1, the Map-Server's intention
>>>> is to forward the ECM to an authoritative ETR.
>>>> 
>>>>  • [AR] Can M and E be set at the same time?
>>> 
>>> It can. M bit is set to indicate to avoid DDT procedures. The E bit tells the Map-Server what to do with the Map-Request.
>>> 
>>>> LCM:  The format is one of the control message formats described in
>>>> this section.  At this time, only Map-Request messages are allowed to
>>>> be encapsulated.
>>>> 
>>>>  • [AR] Shall we mention the NAT traversal draft?
>>> 
>>> No, this is an ECM, not a data-encapsulated control-message.
>>> 
>>>> A Map-Server's configuration must also include a list of the
>>>> EID-Prefixes for which each ETR is authoritative.
>>>> 
>>>>  • [AR] There may be certain cases where this does not need to be
>>>> pre-configured. I suggest we replace the "must" with a "should". Note
>>>> that this requirement is already a "should" in section 6.
>>> 
>>> Agree.
>>> 
>>>> See [I-D.ietf-lisp-rfc6830bis] for details on how the Map-Server sets
>>>> certain flags (such as those indicating whether the message is
>>>> authoritative and how returned Locators should be treated) when
>>>> sending a Map-Reply on behalf of an ETR.
>>>> 
>>>>  • [AR] It may be useful to discuss at least some of those details here.
>>> 
>>> It does following this.
>>> 
>>> I also made the document RFC2119 compliant where the terms for not capitalized. There were a few places.
>>> 
>>> Thanks again,
>>> Dino
>>> 
>>> 
>>> 
>>> 
>>> 
>>> 
>> 
>> _______________________________________________
>> lisp mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/lisp
>