Internet Engineering Task Force Interim Meetings, 6 Oct 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF-CBOR-20211006-1400

Description

CBOR meeting session at IETF
2021/10/06 1400

https://datatracker.ietf.org/meeting//proceedings/

A

Well, it's at the top of the hour, but it looks like we only have dave robin and michael richardson on at this point.

A

So we'll wait a little longer.

A

And we have francesca, I didn't mean that you didn't count francesca.

B

A

Michael, you want to go ahead and get your slides up. You can do the share, preloaded, slides button and I'll approve it.

C

I should have put karsten's name on this too.

A

We figure carsten's name. Is there automatically right.

A

Good morning, carsten or good afternoon.

D

Good morning, michael, is pretty quiet, so if you want to adjust your microphone.

C

I did nothing at all other than talk louder.

A

Good now you sound the same as carsten to me. So same volume.

C

Okay gain is at a hundred.

C

Are we still waiting.

A

Well, let's see we have dave ira, marco yeah ricard, so sean okay, I think we're ready to go. We have enough people to uh to move ahead so um welcome to seaboard what uh who's willing to help take notes. Can we get a couple people to to do that.

D

I can help renew the habit.

A

A

All right, well, it looks like you've got it so go ahead and stick your name in so we uh we can give you credit and uh let's go ahead. Michael.

C

All right, so I got some isg reviews on this document that had passed working group last call a little while ago um and just to recall um this is what the tags look like these three black ones and maybe we're going to talk about the green one today. um This is an ipv6 address, um it's never truncated.

C

This is a subnet, it can be truncated um and um I can't remember if you can see my pointer or not, I think not right um and then we interface definition, which um is a 128-bit uh address, with a prefix length, for how long uh is actually uh on link if it's an interface um and then there's z4 tags, and they don't matter because v4 doesn't matter.

C

um These are the review comments. How do we support link local scope or if we do, um what kind of hex should we use and should we avoid any kind of uh dead memory? Sanitized memory codes uh there what to do with invalid lengths and what to do with ethernet addresses.

C

um So I think I just put them in the order of lowest of uh low hanging fruit first, uh so you can see in this example, we have upper and lower case. We have, you know, attempts to spell cute things in the words uh in the hex uh content and I think the workings needs to say whether we think we should put them in upper or lower case and whether or not we should avoid any of these kind of wordy things uh there. I think we should. We should have a consistent.

C

Probably we should have a consistent case, but I'm agnostic as to whether it's upper lower case.

C

Any comments, any preferences.

A

That sean says, let's not shout and keep it lower. I think I have a minor minor preference for lowercase as well, but I agree that the main thing is that we be consistent.

C

Okay, so I'm going to do that, I will make it all lowercase, and uh that was the easiest one. How are we gonna? Do ipv6 link local addresses, so we had uh so what are they? I guess probably most people here know this, but I'll say this anyway. They typically start with fe 80, a bunch of zeros and then a 64-bit iid, which is sometimes derived from a layer two address um and they're specific to one interface.

C

Those interfaces in often api sometimes are named um with different types of names, but internally it is the if index that is used, and essentially we could support it.

C

If we could find a place to put one extra integer um so a way that I can think is, we could recognize that this is most cases is probably relating to um a bit to an interface address, um even if it we're just specifying um something, and so the idea was that we would add an additional uh integer to the end of an interface address and, if necessary, you. You know not really talking about an interface address, we could talk, we would set the 64 to 128.

C

um and then an additional option since that takes two bytes for 128. Is we could declare that zero is the same as 128, um and that would save us one byte uh there um we could do other things.

C

uh The the difficulty with the bear address is that it's a bare byte string and we have used arrays to indicate other other kind of criteria, and so we can't just come back to the bare byte screen string. We need to do something else.

C

Looking for comments on this uh objections, this is stupid and ugly.

C

We're going to have to write some text clarifying that the if index is system local, um this is not suitable for transfer in you know, I can't put this in my seabor based ftp protocol and uh and inde, and and have the the index mean anything across the network.

C

On the other hand, if what I'm shipping around is network configuration in some kind of uh yang or other type thing, then the indexes make sense locally and it it does make sense that you would talk about interface, link, local addresses and you would be able to specify the I've index of them.

A

D

I think uh we we need to do this a little carefully, because uh this is something that applies to addresses but not to prefixes.

D

um So I think the the cases one and three um actually need to be addressed here and I'd I'd rather have a clear difference between an address that has a zone identifier on it and an interface definition so setting something to 128 or zero um doesn't really suit me, setting it to false or not that that might work.

D

Okay, having a three entry array where the the first entry is, is the the actual uh byte string. The second entry is either the net mask or it's null or false or whatever. um If no net mask is, is intended and the third one would be uh the the zone identifier. That would be the most general thing to do where zone identifier can be an integer or a text string.

D

Okay, um so that's the solution. If we agree that we have a pro, we have that problem.

D

So there is already a way to to send all this information in yang, obviously, um so this is a different way to to send the same information uh which is fine and yes, I've, seen systems where I needed to use the zone id. It's still kind of a special case, but I think using this this triple with with nulls or false, we can avoid uh really becoming a ton of different cases.

E

C

I like false, if there's no net mask um and um um and do we just we're just extending that array by one item, it's either there or it's not. um What does it mean if it's not a link local address and there's a scope id? I think the answer is, I don't know it. You know it's actually valid to do that in a socket bind call, uh but sometimes it's meaningless.

C

um So I don't think we should. We should rule that.

D

What about uh mollycast is, I think, multicast.

C

Does need a a scope, id yep.

D

Where the word scope unfortunately means something different.

C

um It needs a zone yeah, it needs an anything bound to an interface, and then it has a scope, a multicast scope, yes, which is yeah more confusing you're right, yeah.

D

So I think we we don't need to limit this to fe-80 to the fv-80 case, um but we can just provide for that first third array element and then then the subnet sticks out like a sore thumb, but that's okay.

C

All right, uh so um is there any other comments on that proposal?.

C

I will write it up this afternoon. um I know it's on the telechat tomorrow. I don't know if that'll work but uh we'll see uh the other issues. What do we do with invalid lengths?

C

um This is a question from ben caduck.

C

C

I think that's really it we invoke whatever whatever it is. The application thinks to do for an invalid length. um Then I I I don't know what else we would do. Maybe there's some other comments or thoughts about that.

D

Yeah, I think this should be ended like any invalid tag. Okay, so we probably should have a paragraph about tag validity that says what what are the things that a tag validator should check.

C

um So that's not up to the application. It's up to some library to validate this.

D

Well, 8949 leaves this open with a tag: validation, yeah,.

C

D

Exactly application of the library, but you it still makes sense to to fully define tech validation. So you don't have two different libraries that to do different things for validation,.

C

Okay, so we need to specify more clearly what is invalid. Then it's invalid, if the, if the in, for instance, on a um uh for instance in an uh in a an ip addresses, always has to be 128 or 32, um and we've said that trailing zeros are okay for the subnet, so those va lengths could be different but uh and we've said, the interface definition is always 128., so the only cases are what, if it's, not what not 128 bits, then it's invalid or if it's not 0 to.

E

C

In the in the length.

D

I hope most of this is in the ctdl already uh so so I think the certification section could probably just point to the cdl and maybe add one or two sentences. I don't know what what else needs to be said and that everything, of course can always be explained in the city, all right.

C

And um ethernet addresses people have asked okay, so I have changed the text that says that 260 only is deprecated for ip addresses.

C

So if you want to use a tag 260 for ethernet addresses, uh this document is perfectly happy to to support that.

C

I have no particular need to encode them, but maybe others do.

C

I think that we should just we should just remain mute about whether about other things, and there are, of course, also two byte ethernet addresses and six byte ones and eight byte ones. Depending on what transport you have.

C

Maybe they're not ethernet addresses, maybe they're 15.4 mac addresses yeah.

D

So I I like donald eastlake's suggestion to do this in what was it 7201 this?

D

um There is this document, if you haven't seen it, which contains all kinds of ancient and and arcane knowledge about ieee defined identifiers used in the idf context, it's a little bit like like a book. You would check out from the hogwarts library I mean it's really all very ancient and and stuff you, you don't normally uh come into contact with. um So I think defining a sibo tag for for mac addresses in in that document uh would be the right way because we have all the context.

D

So we we know what what kinds of mac addresses we talked about and so on. The only little problem might be that this document is informational, and so the tech definition would be informational and any standards document that that does a normative reference to that tag would need a downright, but I think that that's a manageable.

C

7201 bis. No, uh let me check this. 7201 is not ethernet arcane knowledge.

C

So our recommendation is that this document deal with it.

D

C

C

Okay, well, that's it from me.

C

That's donald's document anyway, there you go, he can go, take it and run with it.

A

Okay, so you think you can get a revision posted for tomorrow's telechat, so perhaps.

C

Yeah I'll get it done by the end of end of today.

A

C

So that may not be ideal for for uh north american uh iesg members, but um um I I hope I'll reply. I will certainly reply to all the emails and tell them what we're doing um you know a little bit after this call. Okay,.

B

um Michael, I have I've seen you and carsten have replied to most emails, but a roman hasn't gotten a reply. I don't know if you've seen it sometimes you know.

C

B

Yes and it's nothing.

C

B

Objection comment and yeah.

C

Okay, I'll make sure I reply to him. I I uh I did combine his comments in with the other, uh some other because he repeated there were repeated comments, so I did reply.

B

Okay, but that's good that that's good enough with just answering him like we're taking.

C

Care of this tell him that we've done that yeah great okay. Thank you.

B

And um I haven't followed the thread with um um eric and donald and who else was in in that thread. But.

C

I haven't either because I haven't read any email yet today, but um I'll catch up on that.

B

Okay, yeah, the conversation seemed to have start going in kind of a side track. It's just sort of.

B

Not very actionable to this document, but yeah it's good to be prepared. So.

C

B

A

F

um I wanted to point out that in the email, if you haven't seen it today, is a comment last night uh from brian carpenter in the thread about link local and scope with eric and others that says uh about the interface index that unfortunately, it's actually possible to change the interface id in linux and it's actually true in other unix-like systems, so mac, os and elsewhere, which breaks the mapping between if index and, if name- and I don't know if that matters. But I suggest you may want to look at that. Come and respond.

C

Okay, I'll call respond.

C

I think that the point is, though, is that anyone who's using this kind of thing, um for, um I would say, configuration long that persists for longer than a second um is going to have to deal with that question anyway, um right where, whereas uh some of the things that drove me originally to define this was that I wanted to be able to pass configuration from well a demon that had discovered uh appear on a link local address to another demon that needed to interact with it and um if that get, if the, if there's a rename that happens in the middle well, then bad things happen and which we try right.

C

I don't care that branch.

F

C

So I'll reply to it, but I don't think we should change anything in the way we're doing it. People use if index, I think at their peril.

F

C

And, and we can put a beast, we can put a a a seabor string in there as well much less compact and that may be the for some people. The correct answer right right: okay, thank you.

A

And francesca posted a link to the email thread she was talking about in the chat.

B

Yeah that contains the one mail that you were just talking about. Michael era,.

A

Okay, do we have anything else we need to talk about today.

D

Well, I wrote a new draft which you had, I think about three hours a chance to uh look at. So maybe we can have a quick look at that.

A

Go ahead and share your screen. If you like, or just talk about yeah there, we go grant screen.

D

Okay, so this this is um essentially the content uh that we should be discussing. We just don't know where that content will go. I just wrote a single draft so for for a while, we can can discuss it in one piece, but of course this. This has dependencies on on things. That really would probably suggest moving some of this to the place where the dependency is.

D

um So basically, we talked about having application-oriented literals um in diagnostic notation, and so um I essentially wrote a few words that said, say the notation for byte strings for encoded byte strings uh can be uh hijacked or extended a little bit, uh because we essentially have a namespace here uh with the the uh tokens h b, 32 h, 32 and b 64 taken, but we might want to use all the other names, um and this explains how how to do this and and also defines a registry uh where new names can be put up.

D

So the the four names that are that are taken are uh described as reserved here, because they are really just by strings and the the other two are the ones that this document uh defines mostly to have an example um and yeah. That would be an iana registry. um I seem to remember. Informational documents can have inner registries, but I I never remember the details of that and it's not written up um in a place where I would find it.

D

Yes, thank you.

D

um So 81 26 is mute about that which kind of sounds like if it's not forbidden, it should be allowed, but it's nicer if you actually know it's allowed, um so um we need to define the policy, which is probably some specification required thing, or maybe not, let's discuss this later and the the actual detail template. But I think this is pretty straightforward.

D

So let's look at the two examples. uh The one we really need is the ci um extension, and essentially we we would be able to write something like this in the diagnostic notation and the diagnostic notation processor would send this through a uri to ci converter and essentially come up with this.

D

Yeah, I think that's pretty much what we already discussed.

D

There is one ugly thing here or maybe interesting thing depending on what your perspective is. The the href document defines a conversion from cri to ui, but not from uri to ci. So theoretically that is ambiguous. There might be more than one ci that can be converted to the same ui yeah. I don't have an example for that, but there are so. There is some cautionary text here. That implementations should favor the simplest variant available. I mean you could you could instead of -4, you could write a string containing https here.

D

So this is trivially true, but it's also trivial. What the right answer is, but there may be more complicated uh cases where we don't have a 100 percent, unique uh mapping and that's new because so far diagnostic notation always gave you exactly um not at the encoding level, but at the data model it will give you exactly what you uh wrote.

D

So that's the cra case and just to have an another case here. um I took the dt syntax from the coral document and converted this into an application uh extension identifier, and so you can now write something like gtd, 1969 or 721 and get an epoch based time which, in this particular example, is a negative.

D

A negative number.

D

So that's it and I probably need to polish this some more, but before I do that, I'd like to have a little bit of feedback, whether that's the direction we want to take or whether we we want to do this at all, uh to support the current thing or whether we decide there should be no application, specific things in the diagnostic notation um because yeah.

D

That, of course uh opens the question whether your diagnostic notation processor is able to do the things that my diagnostic notation processor can do, um but we kind of opened that when we extended diagnostic notation in 8610, so it's it's not necessarily given that all diagnostic notation processes can do the same thing.

D

You will also note the absence of a b and f, which is very unusual for me. If, if I have any text-based information I usually have abnf, but in 8949 we took the position that this is not an interchange format, so we shouldn't try to define it using the same mechanisms we use for interchange forms um yeah. So if this is your comment, where's the abn f, here's your answer. That doesn't mean that you need to agree with that. Of course we can provide abn for this.

D

Okay, any comments.

A

Everything welcome. Let's do it, let's not do it. Let's do it differently. Yay carsten, thanks anything.

A

Everyone is really excited.

F

Yes, uh let's do it? um Yes, I like what you've done. uh I'm all I'm always nervous about not having at least exemplary abn f, but if it you know, it would have to be labeled, as this is not normative. This is you know, for descriptive purposes,.

F

But we obviously need the cri extension and the dt extension. I believe.

D

So we have one more comment in the chat.

D

So I try to have a more polished version done by the internet draft deadline.

A

A

Anything else we need to discuss.

A

Sorry, how many, what.

B

A

Oh um yeah, let me check.

B

A

Think gotta get my calendar up here. Okay, we have yes in two weeks we so on the 20th of october. We have one.

A

And that is the last one that's scheduled, because then we are the wednesday before ietf, so yeah. I think we're good on that. We can always schedule one for for november 3rd. If we need the extra time I'm guessing we don't, we will have a meeting during ietf 112, so we should be good and then at that point we'll schedule them for post 112..

A

B

No, that's me: it's going to be on november 3rd.

A

I'm sorry, I didn't quite get that what about november 3rd.

B

It the plenary, is going to be november 3rd.

A

That's right! Yes! Yes, it is so I I don't need something yeah. I don't think we're going to need one on november 3rd. I think we're good. Does anyone think we might.

A

Now, we'll we'll double check on the list, whether uh uh after the next interim, we'll we'll double check on the list, whether there's any material that needs to be discussed between then and the 112 meeting. I think not.

A

Okay, then, um you all have a half an hour back 27 minutes actually by my clock. uh Thanks for coming and um thanks for the discussion see you in two weeks.

B

Thank you, barry bye,.

A

F