Internet Engineering Task Force 102, 17 Jul 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF102-RTCWEB-20180717-1430

Description

RTCWEB meeting session at IETF102
2018/07/17 1430

https://datatracker.ietf.org/meeting/102/proceedings/

A

We basically have two three drafts that are not yet completed. One is STP examples which we think we're pretty much ready to go s after this, but we don't really need to talk about. We have IP handling where we have this one issue about hem DNS names for most candidates and security, our architecture document.

A

We have a couple of pyaara pyaara later than that, and so sorry, so most of the PRS got merged, which was good so there's that there are two that are outstanding, which is delicious, because that means we could almost be done, but I guess so we'd like to start off with the IP handling stuff, but before we get there. The whole plan here right is to get it done at this point. We need to pick this thing up and get it across the finish line and we need your help.

A

So I think we're in the process of doing that and we're cautiously optimistic that we can actually do that. All right, I think with that I think we have what Justin is gonna, be up.

B

One thing more about getting finished: I I want to note that one outstanding document before get started is topic in the working group, which is the document, and that's hyacinth reviewed comments. Is that right? It would really be good if people comments or cetera or reads and comments and gets this help getting this done, because that's otherwise it's gonna be the last document out so.

A

A

All right just.

C

And I'm Justin Uberti from Google UN from Apple's gonna, join me up here to talk about this.

C

All right, so I'd be handling, let's start by sort of recapping like what a point of this draft is, and the goal here was to you know, take four obscure web pages. Previously we gave out all ice candidates, and this created some problems. The two most specific problems were that by giving out every single sort I address that the I station and find this and a lot of people who are blind behind VPNs to have their public eyes be address identified. This was not good.

C

In addition, not giving out all the set of addresses allowed more fingerprinting. You know and we'll get to more about this later, but these are the two main things we're trying to address with this draft and in addition to solving these problems, we also learn to maintain the ability to actually establish direct peer-to-peer connections, and why do we care about our trailer pages being able to able to establish peer connections? Well, it turns out that this is actually a very frequently use functionality. There there's actually five times as many page loads in Chrome.

C

You did a channels as use actually RTC web for audio and video, and this traffic has been growing exponentially over 400% you every year is actually starting to make a blip on me compared to over HTTP traffic data channels, I started. Actually the traffic theory is that starting to become actually comparable all on its way to becoming humble so like this actually big deal, we don't want to. Unlike cut this thing off, it's a very promising, emergent technology. We want to find some way of having your cake and eating you all right.

C

So we solved the first problem with the first version of this draft, and this was basically saying that we, what page doesn't have consent it should only be able to access the public IP is that the website would normally see, and so this was done by basically determining what interface will be used to talk to the website and only giving out the IPS associated with the adapter that would be used.

C

You know the same way that we used to touch from the website over HTTP using that for actual web RTC traffic, and so what we ended up doing was we gave out just that one public IP, but we all gave out its associated private IP, and so we considered this an acceptable trade-off. You know, because we wanted to make sure we could allow these direct connections. We thought that the actual IP before information given out was fairly minimal. I wasn't really identifying and overall the you know the cutting off the ability to have direct connections.

C

You know seemed like a pretty big downside, however, like that we knew there are some cases that you know the IP for adjusters might actually contain valuable information. There are some efforts going on to try to reduce the number of fingerprinting surfaces. You know on the web and in doing some enumeration of what people are actually doing with WebRTC and analysis of Chrome stats show that there is a non-trivial number of pages that basically create a peer connection, get addresses and then do nothing else, and even after we implanted this solution for problem 1.

C

So what to do turn it over to un.

D

Ok, so in Safari we went to our team and basically they said, there's one rule you can send anything, but it's already public, but if it's private, please do not send it. So we decided to use what free by default. What do you when get you the media is on and this it's better to ship something than nothing so, and some people were happy, so it was already good still we we broke some existing applications that are running. You know the browsers.

D

We got something back. We had some pretty good feedback actually from things like web torrent, where they do not really care, maybe no kind of okay, but still there are some applications like games where latency is important, so turn is not very good and there are also some apps where that are like sending a massive amount of data between devices and we're turn might be too expensive or very small apps, where turn is also is already too expensive.

D

D

Okay, so we looked at muted it and said: hey: why not choose mdns? So when you gather Eiskant days, you have host candidates with ipv4 addresses, so why not generate a new ID register it for n DNS and then replace in the field. V IP address with this new ID. That's what we did. We implemented it into Safari behind the flag. So it's available, you can activate it and play with it. It's not 100% well implemented, so it could be optimized, but still it's available.

D

You can play with it and then the good thing is that we're able to get some early experiments. The good thing is that it's working very well as expected on unmannered networks, meaning like small offices, networks home like if you play a game, maybe it's at home and not at work. So maybe maybe there it will almost get everything working.

D

It's really it's adding some potential latency you to immunise resolution to connection setup. It's not I, don't think it's a big deal, but still it's something that you should be noted. It does not always work on enterprise networks. That's for sure. So it's not like full solution. Some networks will disable multicast. If you're not on the same network segment, then mdns will probably not work as as well.

D

I should note that only add that expects direct host to host connection working in an enterprise network might face some issues anyway, because you have no guarantee so you need to have anyway, some of a solution. Like turn, servers like something like that that will allow you to actually ensure that you will connect the people you actually want, but still.

E

Behind the flag, if you, if you enabled it to test it right now, when you're doing the ice candidate presentation, are you presenting the mdns and a public IP or just mdns, so.

D

What we're doing is that in case getusermedia news, media access is granted. We do not use Indian, as we submit via private IP addresses right before when we get you access is not granted no consent, we're just sending in DMS candidates. Okay,.

E

That's useful for following this analysis, because okay, it means that in this case there was no public IP and therefore no possibility of using a turn server for the current set of tests you're giving their results from.

D

D

So, what's it to get back to just a me.

C

Okay, so to try to you know, come up with the best possible solution that gets the mineral sort, privacy impact and the maximal my code of connectivity. You know the thing we sort of came up with was to use mdns to hide private addresses I, any a v4 v6 addresses that are public and how I determine that's public is like something we can talk about on a later slide.

C

Give this out, because that's on the website to see anyway, and then the nice sort of thing about this solution would be that, as I p v6 deployment proceeds, the need to rely on MDM DNS goes away, and so you can see from the diagram here kind of what would happen that you have this client and it's multihomed. It's got. You know cell interface, WN. It's got Wi-Fi interface, which has both v4 and v6. The v4 address is your typical mattad before that gets hidden and is covered by the this thing work.

C

Yes, the M DNS name. The v6 address is public, that's here and then the actual public you know server reflexive address doesn't launch from the stun server is also given out. You can see that the local address here is emitted from the are ever so. This is what we propose, you know, is the sort of sweet spot in terms of what we can make work today. Yep.

F

Sometimes, just to clarify the v6 address, you determine that it is actually public by observing it as a reflexive candidate. So.

C

We'll get to exactly how that works later, but that's the I think the way you can actually do that is by actually having stun over ipv6. Now that has his own set of complications compared to later, but, like the seem safe. So it.

G

Just made history clarify first, so so you're exposing you had four dresses you're. Exposing three candidates is the idea that that UID D references to both IP addresses in some way.

C

I think she won't speak to us specifically.

D

Usually, what happens is that if you resolve.

D

And if you have grown from ipv4 and ipv6 family.

D

G

You instead had one new ID, / /, / Canada. You would not even expose now, instead of having it's like like what is that saying for three kiddos? Why don't y'all, for which to run course behind a one who's wrong one course by one I do.

G

G

Their show you have plenty situations with it where you'd have addresses right, um I, guess I, just don't understand why, like I, also read in your draft, you said that you only advertise one MBS for both IP through firm V sex, so like Isis I'd have one candidate per address.

F

F

Yes, Thompson I think the I think this is a good observation. It's very simple to say that I'm gonna buy into this address and then do whatever is necessary for this I think this is just an optimization on their front. I, don't I, don't think it actually changes the the the nature of this at all. To use the the same label all same mdns address to refer to multiple host candidates. I mean that's it's it's just an optimization. If.

C

You can solve em DMS, then you're good I mean the whole reason we give out. The I could be six at all is just in case where I mean us does not work. No.

G

Iii remove that but I guess what I'm saying is that you know these are some priorities faces and so and so and so so gluing. Oh the glue, all these sort of private objects together into one thing.

C

Here, just going to glue the e46 with the Wi-Fi address, but yes I. That is true that they would have different priorities. Somebody.

E

I'd also like to point out that there are will be cases where the ipv6 address has a different scope, especially on wired interfaces. So you may be able to use em DNS and have a different scope. Then you would have the public ipv6 address, so you may be getting a different ipv6 address than this ipv6 address, because it reflects a different ipv6 scope, so I think having a separate UUID. If you want to use a UID to front that separate scope is a good idea. So.

C

Just because it would be a separate new idea for v4 and v6 f per UID. This per address.

F

So I was gonna say: that's, that's probably necessary in this case, because of the way that we do, that the pairing up and the prioritization and all those things I'd missed that point and that's that's a really good point.

H

John thanks yeah the ice spec does say that if you have a name and a candidate, you only take one address from it, preferring v6 if there's both, but you never take more than one address for a candidate by name so yeah. We need to have a different, different, UI, deeper address. You want to right.

C

And so you need to make some changes in the meal too, which.

H

C

Dean asked not mes video.

F

Actually, it's a superset, so I think this is okay. I looked into this and I was a little concerned about the use of these labels, but we kind of get away with that. Mdns supports more labels than names can express, so we'll have to be a little bit careful about the soyoung's.

D

F

D

In terms of privacy, the lifetime of their names could be tied to a page, for instance, beef right nice time. I.

F

Should damn well I'm, sorry, but our origin would be what would be what I would recommend, but I wouldn't I, wouldn't do anything special to retain stability over multiple uses, that's double gate: origin, yeah, yep,.

F

Actually we had, we had this discussion and yeah. It's tricky yeah.

C

We decide on this later, so my hope is kind of get three of the stuff that is mostly not confident and controversial first and that we can get to the tricky stuff.

C

Okay, so I'd like to try to sense of given you know the detergency to get across the finish line, as our cherish reminded us earlier, you know, is this a path that we can agree to move forward with, so.

E

Your choice would like you to amend this consensus goal and that is to change IP handling documents to documents, because we believe that, what's in one of the documents is in fact quite well understood where some of this is still being worked out on the floor and that they don't necessarily need to appear at the same time and having some IP handling document would allow you to move forward with the security architecture and get the the baseline covered and then bring the mdns stuff along will keep the working group open to handle it, but not gate on it.

F

Yes, I'm not in Thompson I, think that's my preferred approach for this. If we can find a way to just adjust the IP handling document so that it doesn't add this, this new stuff then and describe some modes and maybe recommends the safer option of the set, and then we can later on update it with the new things with I.

E

Think that the base thing would be requiring a type II handling is to note that other modes may be published right and make this the first of the new modes right and then, once the new mode is published, we can decide as we're processing it, whether it becomes the preferred mode or not.

E

In the meantime, we'll have ones with which we have a fair amount of experimentation and experience available for the community, and we can cross the finish line while the new modes are being experimented with, and this hybrid mode in particular see some real deployment to get right.

F

Because this is this is a in this is now on the critical path and there's a lot of other things that are very, very, very close to finished. If we put another thing on that critical path right now, I think some people might get stabbed right.

E

Cluster 238 can go if we, if we release IP handling and note in IP handling that more modes will be published and keep working on this as a new mode that makes.

C

Sense but I think we also will probably need to amend the problem statement say we are been trying to solve. This sounds good. Okay, yeah.

E

And we can put in the problem statement that new modes are being developed to solve. That problem sounds.

I

G

I guess I would um III was the.

G

Perfect lovely, informative.

A

G

To this exact raft, that's um say: you know losing wealthy, including this drive here.

C

G

Guess I can imagine and I do.

A

G

But on you know like in terms of practical matter on like blocking these documents were a lot making. This will not make a change in the policy change any faster.

J

Here is Jabbar scribe. Do I understand correctly that the current default mode would remain as of draft version? Oh nine. For now.

C

It sounds like what is being recommended by the chairs is that the current IP handling draft would not change adding additional modes or change its default.

A

But that a new mode could change the default on publication, sure, essentially just adding some text in the problem statement section to say there are other modes that could come along. We can point you this document informatively, which me you ain't, get along faster.

C

Someone said this already, but the intention would be to publish the new mode as something outside of this working group would.

E

Be for one work, but IP handling could go and we could start cluster 2:38 processing without this, since the reference to this would be.

G

Informative normative, but.

E

The new draft would still be on the standards track when published and therefore could update the default mode. Okay,.

C

And we would be open for that last time. That's.

E

A different question: I think it probably would be, but the the critical thing is getting the documents out. If they're working group has to stay open to publish this, then your your chairs will be willing to do that. Okay, you can have.

C

More crows, then,.

E

Yes, I am we're hoping to make them ever smaller cuter as time gets on. This is dJG.

K

So the idea that we're going to try to get this out the door with the default mode being the one that leads to private addresses just by adding text that says we are going to add a new mode later or new roads may be added later, does make me nervous as someone who would like to see this thing published so I am concerned that that what we're doing is we're basically saying here's the way you do it not even mentioning what the privacy implications are and something will come along later and then the we lose steam.

K

C

K

The privacy implications are in a document just just.

C

K

And and say this is.

C

A problem with the current specification I mean I think we would say that this is not yet solving the specification and who knows maybe define I mean guess, is practically speaking like each browser now has made its decision of what it's going to do so I. Don't think this will actually change. You know any actual implementations.

C

Everyone who is involved here is kind of at the mic or in line or standing up here and so my way he will be implanting the new draft and think as soon as you know, we actually have experimental details. Bernard.

L

I'm, open Microsoft I just want to point out one little thing, which is: we have had a it's an issue that probably will be a PR soon for a change in the web IDL to address to make it capable of carrying those in.

G

L

Names and we were willing to take it because people hadn't implemented that thing yet and I'm kind of a little scared that they'll implement the old one and then we'll have an argument.

E

L

Changing it, it could be a little nasty. So it's okay, if you do the new thing, but please do it quickly, because if this stays as it is, it'll probably be a web IDL argument about it. Well,.

G

Not for me yeah and I think I mean well, I would say, is you know.

G

You know we're interested in this mode like understand how that works. I think you put you right to accurately, which is some serious. It works in some cells, many more interestin scene measurements about like how eleven there's two things occur, which I suspect you are as well and that you know, but that like having these documents, tough negotiator queue for a year and a half while you buy eight those questions or whatever it is, is not gonna like that's not gonna make it's not gonna. Make us like more aggressive babbling about.

M

uh You know, I think, there's a lot it's hard to make. This call because there's a lot of unknowns. We don't know about this like we have questions about like does just enabling this allow you to do DDoS attacks on the multicast traffic inside the enterprise, just by creating a lot of these and what are sort of the rates and stuff of that. How well does it actually work in enterprise? I mean the case where this tends the the way you described where it works and doesn't work.

M

I agree, in fact, I think it may be even slightly better than as you described it doesn't work in some large enterprises, but not.

G

M

Lots right, it's uh those are, of course exactly some of the places. This is most valuable in the home residential. In that situation, it didn't matter that you hair, pin through the GNAT anyway, it wasn't like you went across the land, because the address filtering was such that you could hair pin on the map versus having to go to a tremor. So I think we start looking at some of those statistics now. The thing I'm interested in about at this is this is not just strictly a security improvement.

M

It makes a bunch of security much worse and what it makes much worse. Is it now forces a whole bunch of traffic that didn't use to have to go through middle boxes, to go up through middle boxes that are typically run by somebody who's not really fully trusted by the end users, in these cases and anytime, that we start taking traffic that could have remained on a local network and not have to push it through a place. That's an awesome place for law enforcement to intercept it yeah. It makes me a little queasy.

M

I mean I it's. This is not a win-win! A situation there's a lot of things to to think about the the trade-offs on on which way that you end up with best security.

C

You may be very sick, so.

M

E

Return to the Consensus call, which has been up here for for a number of moments- I'm gonna take three hums. The first is: do you think you understand the problem.

E

Okay, I'll move up to hybrid solution, so we're gonna ask: do you think you understand the problem? We're then gonna ask if you can live with this being in a separate document and.

E

If you cannot live with this being in a separate document, because I think that if it's in a separate document, we can progress it after we've gotten some of this experimentation in front of the working group. If it has to be in this document, we'll move forward a different way. But what we really want to know is whether or not you can live with this being in a separate document or not. So anybody have questions on the hums before we take the hummus.

E

I think, if you think we shouldn't do this at all, that is it shouldn't even be in a separate document. You should be at the mic. You want to tell us that, okay, sure.

M

I mean I think that we will need to see a lot more of the details on this before the working group comes to consensus about this is awesome or what it is not to us tonight. So.

E

I think the worst-case scenario.

M

E

Put it into a separate document that goes out with a very limited applicability statement or never goes out, but of course,.

M

I thought that I think that putting it into another document is no guarantee that we're actually what happens right. The working group will come to whatever consensus it comes on on whether it goes in the same document or a different document. There's a lot of work. That needs to be done before we're done with this, and it will need working group consensus.

M

G

It ate I think just before we clean for document exceptions now or more something for separation.

E

The three are I.

G

Heard with the homes work, okay, the home.

E

Yes, we can live with it being in a separate document comes through. Then we will talk about accepting the draft as a working group draft okay, Ronnie yeah, you have to go to the things we have remote participants.

N

Even but you said that the reference for both in the main document, which makes it like so.

E

We can put a reference in the main document if, in that, reference doesn't even have to be an accepted working group draft to say people are working on it. For example, over here.

M

In case people read me is the other way. I will be humming in favor of accepting as a working group document right like I, think that was almost inevitable that it'll get that looks across this nightmare. If we come to it so.

E

Any more hum questions before we take the Hummels hum now, if you think you understand the problem I'm relieved if it was dead. Quite at that point, though I would be like you're gonna have to bring me a live crow at this point. That picture is not cutting it. Okay, hum now, if you can live with us separating this document out from IP handling.

E

How now, if you cannot live with that solution,.

E

Okay, I think that what we're gonna do then is okay. I heard the hums as enough people understood the question to go forward with a working group call for consensus on separating or not separating, and then the strong consensus for separating the document out of IP handling.

M

So I'd like to propose we adopt this as a working group item as soon as we can sort out milestones with our 80s and.

E

Now could you take a hum on this point? All in favor of adopting the draft is a working group draft. How now is there anyone opposed to adopting this is a working group breath um now?

E

Okay, so it is adopted as a working group draft we work with the ADA to add a work item and the schedule of CRO consumption. Thank you very much. What.

A

Is 2.4 the minutes that I guess Justin's gonna update the IP handling document to update the problem statement. I have a forward reference, an informative for reference.

E

Okay, I think you had something beyond consensus to try it about how much time you have I.

C

Think I will sir note there are many issues as things that we can. You know, think about. You won't have them at each our discussion, I'm guess I have two three minutes at most: okay.

E

C

So one of the things I mentioned earlier was what to do about ipv6 addresses that are not public, so, first of all to even determine if a v6 addresses public and you need to hit a stun server to see if you get that reflected back to you.

C

This is problem, and this can occur for nat64 addresses where you have basically an ad which is a v4, and then you get a v6 address. That's only usable behind that net. In order to figure this out, you need a stun server that can basically receive v6 traffic.

C

I am not aware, oh they're, very few of these today, so we would basically be asking digital applications to be deploying v6 down servers and then, even once we have this, then, if you are behind that six four, you won't be able to use that v6 to actually give a direct connection. So we'll do some experiments here will understand. If this is a big issue, you know one of the options could be. We just always give our v6, a--'s and well.

C

One of the most coolers give our v6 is regardless of whether they're considered to be public or not. Just has some. You know some questions for privacy, but since we're not coming out before, is it still a move in the right direction? I want to see what the experiment show. How.

N

Low this gem I'll notice, this problem is not unique to be for physics. It also occurs with before who can't tell from an address whether it's private enough or you can tell from some others- is that they are private, that, if, if you address this, not in those ranges, no.

C

Absolutely I mean yeah fortress none before is for all deployed very I agreed to be safety. You.

N

Need to use stun, it also means that no application that doesn't have turn service will work whatsoever, translates as sensitive stone, service button.

B

N

Even if the Euro correspondent has a public IP right.

F

Thompson I think we're if we, if we're judging, is based on on the principles that we've established, then yeah too bad. If it's not. If this is not the address that the site is seeing as well, probably the site is seeing is a matter of communication with it. Then why are we giving this hell.

C

You know we could take a very sort of hardline position on this or if we see that you know it has like a 75 percent impact on my kitchen, all traffic, then we might. We might take a more compromising position like a English. You mission now, let's get painted in by concerns.

F

F

I'm wondering why the principal decision is not the one to take. You I mean.

C

Do you think we'd say that we would lean toward the principal decision, but like we would probably want to see what the impact of making said principal decision would be so I, don't think what Paul just said is entirely.

G

Correct mainly, if your counterparty knows they have a public IP address ipv6 address, they can advertise, it just want to talk to them even with v6 wait. It won't work so like if there are server first as well will work is two clients right, no clients right.

C

That you don't deploy, Stein silver in your address or public key still can connect.

N

You see no the small this year and I guess change. You know the Nandina sadistic surround the one. Second, when are you ever going to do it? Registration.

D

Does not take that much time, it's so nicely the resolution that might take some time.

N

Registration requires that you have the dye just on the network and and then wait for a while to see if anyone they'll sponsor.

D

We're testing we there's no 1 second delay for a connection setup, that's interesting! The second thing is that you know so. For the first connection, you might have a small delay in solutions. We we had like 200, millisecond delay. It's highly depends if you want Wi-Fi or not. If you are like packet loss, so we might be about to measure to make measurements here to precise. You know, yeah.

F

I, don't think you need to worry about this one. These are extremely high entropy values. The chance of a collision is basically a.

D

N

If she's a packet.

F

Wasn't the resolution is, is, is a concern yes, I agree, but the concern that Harold has with result with respect to advertising your name. Do you have to check that someone else? Has it I wouldn't bother? Checking in this case.

N

So what they're saying is that you can get away with breaking the protocol for NDMS.

M

That will be interesting to see we gain by the other working group, so so.

F

If people are uncomfortable with that, we can go talk today and SSD and and talk about having an exception, an exception carved out for that I think that that's a perfectly reasonable thing to do so again, that's not a big concern.

M

Contest I, it gets a little bit more complicated when the address went out through multiple interfaces and then the SD proxies forwarded them back to the other things this just it gets more located. That's all I'm, saying I'm soon,.

F

M

Like regularly on MDS, when the bugs is your printers show up on multiple networks, when you have a complex, mystical.

D

Technology some different.

M

Addresses moving the same DNS name will resolve two different addresses, depending on which interface you take it on.

G

But no I mean like you're gonna. What's gonna happen is you're gonna pick you're going to advertise the you take the host candidate and you're stuffing them the MS record. That's what you broadcast.

G

What happens is that inappropriate address appears on multiple networks, but it's the same address all those networks, because you're gonna generate a new horse candidate for your new MV s, entry for each choice, candidate, and so what will happen is like it'd, be just like ordinary ice, which is say if you make a pile of addresses, none of which you can use on on, but on that that shouldn't um it's just goofy right, um but it's like I still I see Heron like you feeding this way.

G

um I think that will happen is like maybe you're gonna get a bunch of like weird packets. Like networks praise you like a bunch of packets, like you.

A

O

G

Same address, yeah.

O

News old man, um one clarifying question, so this is, you need to stun server only for nat64 or are there other cases where you would want to require a stun server, because I I'm, not quite I, don't have like all the details for nat64 in my mind, but I think there are like other ways to detect that you are on the net 64 out there.

D

Metaphor is the only case I know, but I cannot grant even I'm correct.

E

You want to go into the next slide. Yes,.

I

You want talk to this one yeah.

D

So when you have a single WebRTC connection between two peers, you can identify whether they're running on the same device by checking the latency of the connection. Some might say it's fine.

D

Some might say that if you have two pages with two iframes that our trackers and you are trying to isolate these two trackers that so that they do not state which is actually what is implement in severy, then allowing these two Tigra tracker iframes to know that they're running in the same device and start to use WebSocket to share stage cookies and everything might be an issue. So we are actually investigating that the the point, where is, should we document this issue somewhere? Is it an issue and if so, what's the path forward? I.

G

Mean this is an issue: there's a lot I mean I, guess I mean I'm, you know, I got I feel like or worst nightmares about, like co-located code have now come true. So, like the the it's a quaint tour in the face of Specter, it's can equate to worry about like determining you're missing the host team at this point in the game.

G

Perhaps if you're chrome, you feel differently, but here people don't so you come so I I'm we're also working on it I'm more seriously, they're just a lot of ways to determine I. Think if you're like on the same post, um you service Lee, if not not poker, solution or v6, but as far as I can tell there's no clear solution at all. um Unless um I mean maybe- or perhaps you can disable.

A

G

Like some context, but imagine like the common thing would be the two parts like these two double key right: you're still gonna have the same mdns problem on.

D

It's up to the browser to decide which you I need to share with, on which contexts no.

G

No, but no I I miss I'm, assuming user different, you IDs, but then basically, what I do is I take the you know: I take the I, I did I, you know I had to dive to two domains that I want to see. They're co-located in the double keyed and I have both of exfiltrate the addresses and then yeah I'm trying to grab pretty call and like it works then like I'm on basically the same machine or I'm that same or in the same local network. So um so I I, don't know.

G

I I tend to think that, like this is a Emmy, it's fine to put this stuff. Doesn't that be handling document is innocent paragraph, but I. Don't think much else is gonna be. Is it is use for what it's really open research question so.

A

um Before we're gonna cut the line, because we do have one other presentation and I guess we should float the idea, maybe even having a virtual in or something that's talked about, something's issues, specifically because I think we can go down a rabbit hole on this one for a while.

C

Since we were trying to progress the document should we just can we agree that we should add the paragraphs of document this issue and then we can resolve this in the to be announced, sequel, absolutely.

M

In a similar way that we identify and w3c documents say, a fingerprinting issue, I mean I. Think it's calling out this issue is perfectly fine, I mean, but there are so.

F

M

You can track this like as you. The which way is up force sensor on iPhones seem to be a high high risk candidate. Let's.

D

Convert over yeah.

M

No one wants to kill the really high runner ones. So what about this one? The question here is I, don't understand what the solution would be for people like I, don't and how we solve this at all for new v6 or before.

D

M

D

Can have you can ice? My friends.

D

So one potential solution is that's for party I. Friends do not have in DNS.

M

If we're gonna allow WebRTC connections like like, where I'm going here is the only Slough, this is to add huge latency to our real-time communication system.

D

That's not really sure, and then you can use probability to actually measure whether it's fake latency and so on. Iii, don't have a good solution there right now, so so.

M

What's your take on, should we try to solve this.

D

It's not a priority because it's true that there are other issues but I think it should be in the one map and maybe we'll find some solutions. Maybe on our side, when we just disabled like when we are spaced in Safari, where we are very restrictive and what we're hoping to do is to open the door a little bit. So we can open video like just a little bit or a little bit more, not on iframes, maybe a my friends. We have some granularity there that we can play with since.

E

We I think that the current proposal is that we will document this issue in the IP handling draft and then consider solutions in the follow-on draft. Please indicate when you're standing up at the mic line, whether you have any problem with us, documenting the issue, because I think we can postpone discussion of the solutions till the discussion of the follow-on draft. If we have agreement that we're documenting the issue.

O

My only point was that sorry Neil told me is that I think the RTT is actually not the biggest problem. I mean if you have like two iframes being able to establish data channel, they can exchange whatever they want, I'm in Europe. Far worse with so.

E

If you would like to document something else, a PR is welcome.

K

This is DJ G I'm for documenting the issue. I just wanted to point out because it seemed like there was some confusion and discussion that this is a web RTC issue and not an MD NS issue. Okay,.

E

K

Making sure that we're all on the same page on that I.

E

Think we are on that page.

C

Twenty one final note, one more in common choice: I'd mentioned that Mike. If you can't stablish a track connection will typically fall back hair, pinning through that and in doing some tests here. I was surprised by how infrequently hair printing actually occurred even in today's internet on some from popular nap products, and you need, did you see the same thing.

D

Yeah working in local conditions personally so but I, my testing is irritating I hope that as soon as we get.

C

There so you can confer from that popular Google and Apple Wi-Fi products may not support her pending super well, but anyway, there are. The tests that will do will probably bring some more concrete numbers on this anyway tax time. So.

E

Thanks both of you for the presentation today and for agreeing to move forward I think it's it's a good thing to move forward with this in in the way that we've described I think the cherish will caucus offline with their Area Director about having an intra meeting as I think dkg pointed out just making this afford reference as a promise and losing steam is probably problematic so trying to keep some of the discussion going on the list and in an interim meeting, seems to us like a valuable way to avoid that and we'll do what we can to do so.

E

I think the next speaker is crystal.

I

Okay, so I did as to be the Directorate review or obvious TP attribute, which is defined in this RT civil security architecture document. Just to clarify this is not my document I'm.

I

Not the author is the reason our period is is because well I have indicated I'm willing to help with STP issues and so on, but but before really starting to write text, there were some some issues that I think we need to turn me on or to make sure we have a common understanding, not how to really write lower things, as also to make clear least, this is not generated general overview of the whole document. Only the STP I attribute part of it.

I

So if you have other issues, you know with the IDP and an identity assertion in general, they. This is not a time and place place for that. So this and also I'm not going the review that I did. For example, Martinez already created a request in that I think the link is on. The mailing list is also I'm going to show it in other presentations.

I

So there are basically a couple of main issues which I'm going to bring up here, which I think is is is what we need to agree on and to make sure: okay, okay, so basically yeah but background I also said about, but the the Assumption here is because this draft is foreign foreign as it's written in this in this working group is very web. Rtc specific, but assumption here is the best EP identity attribute is a generic attribute which can be used to carry identity Association.

I

Also in other environments, not really the bar can see specific, and even if one pair is of a party CN point, the other pair may not not be a, but it could still use the same identity assertion mechanism to do so. That's the assumption to have a generic attribute which can be used by anyone supporting in an iid p assertion mechanism.

I

So this is one of the issues with which I raced in this is that currently in WebRTC, the fingerprint is is used to create this identity assertion.

I

One of the question I raised is that do you have to use the fingerprint always or could you use also something else and the answer I got and now when I say answer here, is this discussion I had and I? Both people have indicated on the list? Is it's not something we have agreed on, but but what those who have provided input instead from a protocol expect that you could use something else than a fingerprint.

I

It's outside the scope of our TC web draft, we're not going to document that or make an instance, but but but basically the way for would be would be to have some takes saying: yeah, you could use something else, but if you use something else, you need to write a draft or some Scott specification describing how you're gonna do it, how hard a JSON structure that you're gonna put into this attribute it looks like and so on, because the the digestion structure defining the document now contains the fingerprint and the identity.

I

So if you use something else, you may need to define adjacent structure which carries this other thing. It's.

E

Already just a clarifying question: you have here in the answer from a protocol perspective, but here what you mean in particular is this usage of SDP is allowed to use other information, not J cept. The protocol will use other information, but this SDP identity may use other information. That's what you mean by protocol. Yes,.

I

I mean I'm not suggesting any changes to chase or RDC.

E

I

A good con about I should have made it clear, I'm, not suggesting any changes to what we have in J said or any RTC VapoRub a bar can see specification. This is pure, pure STP. So to say, this.

F

Is about making sure that the definition of the STP attribute is consistent with other attributes and all those things largely editorial? What Krista has actually done is highlight a few places where there's a little hole or two and that's been quite helpful, but other than that yeah don't change anything. We did change something by the way, but.

I

That issue was, which is interesting, is if you have a subsequent offer a server. Actually, the pair or even yourself, updates the identity and and I have one example where this cause case. We often talk about this third party call, can cross and call transfers done just by updating the STP. So you have your browser, for example, is communicating with another pair, and then you have a call controller which is going to to make a call transfer.

I

The only way it's gonna do is to send you, then this DP of this new peer and obviously this new pair is going to have another identity, so so you're gonna get an updated, STP and updated offer within you within your identity. Again.

I

The the feedback here is that you gain from our pure purko spur sexy. We we shouldn't. We shouldn't disallow this. It's allowed. However, it may not be supported by the receiver and, for example, web RTC. That's not allowed to update the identity of appear, so they suggested way forward. Here is that you say yeah. You can do this, for example, when you send us, you can do it, but also to indicate that this is not something that that anyone must report. If the receiver doesn't support it, it doesn't accept it.

I

It could, for example, not accept the offer also or do something else, I mean we don't need to go exactly into those details, but I think there. The important thing here is that we're not going to mandate endpoint to support it, but but the game it is allowed from our from our protocol perspective.

H

I

That's yeah, that's more or less the same thing. I mean you you're, basing your providing you. You change your modifying the existing identity by moving their existing identity, but it's the.

H

Same thing, I hope Jason already says that if you ever had yet it'd be all subsequent offers, your answers have to contain the same identity.

F

So the web RTC Martin Thompson, sorry, the WebRTC specs- are very clear on this point. Once you have an identity for a Pierre, you can't go backwards from that point. It has to remain the same thereafter. So any any session description you receive from that point onwards has to have the same identity, because we don't know how to do anything else. Nothing else really makes any sense if you want to have a different identity, make a new peer connection and then first and.

I

Actually, what I'm going to put in it you know it takes about. This- is there's going to be a section about modifying a session and there is going to be a must. You must include identity and, of course, if even if it's the same as previously, not not including it means that you're actually removing it. So so so, of course, if that's what you want to do, that's what you want to do, but if you want to keep the same identity as previously, then then you must include it in this subsequent offer.

M

Right, yeah, I think a general policy I'm in favor. That's the general policy here is what you do with identity is effectively a policy decision about what the security and it means in whatever you're using application. In WebRTC, we have very clear things of what that policy is and other things they might have different policies and different rules, and you know even the the call that way of doing transfer is not generally how it's done because of all kinds of problems like this.

M

So it's not like we're losing transfer in WebRTC or anything, but like the way most SPC's do transfer will work fine with Park UCS yeah.

I

I just wanted to bring it up, because, unfortunately, why I wanted a whenever we talk about it. It's very often when people, when we talk about something and someone asked how does it work it's third button, so those were the two actually, the main I consider technical issues or clarifications.

I

So the way forward is, and asaba indicated a I'm willing to help with this. For example, one of the things I didn't bring up here. What would I had in my review was that we should also use this template for for defining your SDP attributes this, if you read I, mean sending a mission offer or sending the answer or modifying a session and so on, and so on and I'm now willing to help to put a pull request regarding that together.

I

But but I want to be spirit, things clarified because those things affect what it takes there would. It would be ya.

K

I

Everlasting was that they'd be here's a link to this request, which, which Martin already put together I, wish I understood, has been known most things. There were some, for example, related to day to day today to the a B and F, and so minor things, yeah I'm good.

G

For loads, um so thank you for doing this review and thank you for if you just reload I, think those are all great all gone yeah there. You go it's linking interview and thank you for catching these um um I guess: I'm reluctant to screw the structure at this point in the game. um I understand your willingness to do it, but it seems like it seems, like that's really kind of like gonna make our lives harder. um Maybe there's pacific points. Do you think?

G

Maybe you called out I'm really happy to add into the document, but I think having a baby talks that we don't have to review. Like seems like kind of mother away, but IIIi.

I

Think I had a look at a document. Currently there did attribute is defined somewhere in a subsection level, 3 or level 4. So my yeah something like that, but my idea would be just to move that into a separate main section and then, of course we have to add this sending an issue of her answer and blah blah blah. But it's going to be very it's not going to impact other parts or the documents.

I

So it's gonna, it's gonna, be a big pull request in a sense that there's a lot of takes that's going elsewhere, but most reduction that most of the existing text, which is not or all of existing text which is not related to this attribute, is going to remain.

G

The same things like going review so I think like unless there's like some, like actual other instructor. This way I look for not yet, but.

E

The terms do not believe that there's a natural law, so we would recommend that you generate the pull request and have it reviewed, but it will be up to the editors discretion to or not except the pull request, depending on the the amount of confusion and work he believed it will take. So the the key thing is that the material be right and the structure is not normative and certain and.

G

Solely if you have any points.

E

G

You need pacific points, you think we're missing. Please do call about specific weeks, I'm happy to put them in well.

I

I think that means we're a mission now is they're. Sending this initial offer and modifies a session I mean it could be written in other ways to and of course, those things that we previously discussed. You're clarifying that that, for example, if you use something else and fingerprint is in name just- could prescribe and also about this updating identity, there there's nothing about that, but I mean I'm.

G

Okay, but again again, I think I, big, there's like a huge PR and a half detective with the same. That's gonna be like unfortunate.

I

And my suggestion, since all these peers here-now emerged my suggestion to make life a little easier would be that we submit a new version now, where we merged or include SP this PR, and then this other PR would be based on that new version, because then it that way we avoid you know, requests.

E

I

Yeah, that's all I had okay.

A

From a timing perspective, if you were to produce such a PR, obviously you're not gonna, get done this week. Do you have a timeframe when you might think you be today be able well.

I

I'm, this is not gonna, be I'm, gonna, be in a way after this week for two weeks so vacation, but but.

A

I

Food, it I mean, let's say in a month.

A

So the only other thing of note is that we are gonna work. A new blast call the STP examples drafts probably and.

E

And we have fun in several outside reviews, for it. I really want to thank Paul and Fleming and the other reviewers who have already done that. But it is a big piece of work to look at. Please do take some time grab a few of the examples go through them. I think anybody who does the whole document is is going to be very tired indeed, by the end of it, because the productions are very, very complicated, but please do sample a few of giving your expertise and give us comments as fast as you can I.

E

Believe that closes the meeting uh thanks. If you haven't signed the blue sheet, please do so before leaving.

A

Where are the blue sheets by the way? Okay in the back Thanks.