Internet Engineering Task Force 101, 22 Mar 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF101-UTA-20180322-1810

Description

UTA meeting session at IETF101
2018/03/22 1810

https://datatracker.ietf.org/meeting/101/proceedings/

A

All right time to start so welcome to UTA and it's London and.

A

Let's see how far we can take this, so note the note. Well, you know what it says and if you're here, if you.

B

A

And note the fact that there there has been there have been IPR disclosures related to these one of our drafts that we're working on. So this means that we have to be. You have to be extra careful about this, so think about the note. Well, so this is what we're going to do today. We don't have a lot of time, but we're gonna basically try to cover our document status.

A

We're gonna cover some spend some face-to-face time on the MTA outstanding MTA STS issues charges all those we believe, those to be actually quite simple and then we're gonna get an update on the required here last draft and then there's open mic time now proceeding.

A

So this working group has produced quite a lot of output, our core documents from early on our or RFC. Some of them have even got a rota. I, don't know that bit whether that's a good or bad thing, but anyway, our current work is getting done all of the work except one. The the two of the core documents that we're working on right now are scheduled for inclusion on the next, and the near term is di tella Jets.

A

Right new term is t-tell shots, not on the next election, and then we have the SMTP record here as option to work on. So we don't have a lot work left there. There has been a couple of what's called the related drafts that turn up in data tracker. We haven't talked to harness yet, but Hannah's has has has created. This draft called draft show finish to be a TLS one, three profile that you know he called it UTA.

A

He must I think he means something by that, but we haven't talked to him yet and he hasn't made any contact with the chairs or the working you will see, and he, if he turns up before we're done today during open mic. Maybe we can ask him just.

C

A quick comment about the two documents that MTA STS and report so report is on IG agenda for April 19th. So at the moment we have IG has too many documents on the next to tell the charts. So I put the STS one on the my beginning of May, but anyway I'll try to get it on the same April 19th one but I, just I apologized I promised April 19th but I hope that's, okay, right! Fine!

C

A

Right thanks, Alexa all right, so we're gonna spend a bit of time. Come back to our agenda savoring it some of the MTA STS issues resolved. Now we have a we actually created a separate slide deck for that and Alexei. We sort of Dragoon you into talking to this, because it's mainly area director review.

A

And then those a set of issues raised during sick secretary review that we didn't actually put on the draft. But if feel free, to talk to those two.

C

Yeah I think the AIA natio from Segura review is the main question right. So.

C

Yes, the major issue there by major issue: I mean it's not necessary to fix. It's just I think it needs to be fixed. One question is about ignoring all parameters to content, type and I. Think I suggest you don't ignore the church set parameter because it effects objects. You know if somebody puts text file on the web server with wrong character set which is not based on a scheme subset and I I think it's, it seems as potentially problematic.

C

So uh in this case I'm sort of thinking of malicious but malicious or ignorant use- or somebody put you know if this is a file editable or on the web server than somebody can use a very obscure encoding, for example, in one of the Asian character, says it doesn't, which is not utf-8 and then that just becomes a bit problematic or even, if you just use unicode, 16 or 32-bit based encoding, it's not going to parse correctly or utf-8 yeah I, just yeah.

C

A

I think noting a landline.

C

A

Step up to the mic, so it will streaming. We have actually caught a lot of remote.

D

D

Finest required, like I, think now I think we say it has to be ETA there. Us it has to be as ASCII and we don't check the char set. You said you're saying we should actually have a charge that that's that's, I check it and I.

E

Think that's reasonable because otherwise.

F

So we think it if it is boot approach, because polishing, please Urals Urals can be internationalized and you can expressed them so.

D

D

F

C

So, for practical purposes at the moment there is nothing in the policy document which is not a ski but because it allows for extensibility. We want to just be specific about saying up, you know, if you know in the future, it might be GF 8 or something yeah right.

A

I mean this seamless Alexei says it's important to do, but it seems in controvert not controversial. So you, the authors in the working Europe horrifying doing this.

B

Channeling Ned freed from the Java room says, require utf-8 or ASCII done. Ignoring. Labels is a bad idea.

A

This seems like consensus at this point in the workgroup. So let's, let's move on that's your second issue: Alex I'm.

C

Right so this issue is just some sort of minor well I think it's very easy fix, but the reason why we went with colon separated key key value pairs because you can reuse parser for email, header fields, so you can just stick this into your existing parser and use it in them. You know masala, so if you love for trailing whitespace, then.

C

Basically, it interacts with message: blindfolding facility which use spaces on continuation lines right spaces. Sheryl F is going so it will. If you feed it into a parse, then the second line become parts of the first and then it's going to be wrong. Syntax, so I suggest just prohibit threading quite space, and then you don't. You can just use this. You know there are no continuations in in this format, but you will never be tricked into using one by allowing trading whitespace.

G

Victor channeling victor to come up the leading, not trailing.

C

Sorry to repeat drilling white spaces are fine, I, don't care. Leading white spaces are problematic.

A

Yep I'm not I'm, not seeing a lot of people disagreeing so I think this too is an easy fix. Yeah.

C

And I already suggested, the fix is.

A

C

Line, maybe ni fix.

G

Channeling, the dirty or not sorry, Ned, free sorry, is there a reason why they were allowed.

A

That Daniel is trying to stand up.

D

So no reason lost affinity right so.

A

Let's not, let's not dwell on this.

G

Aka brain say: yeah.

A

Let's not dwell on it fix move on alright, so you got a bunch of minor issues. Do you want to talk to each of them, or do you just want to they're, mostly editorial right, missing.

C

Nate's updated keywords: you know just carry on.

A

Tell me when to stop I can't.

C

Stop come back here, just an example: it shows to IMAX records, but it shown them in quotes as if it's it's a valid part of the policy document, but it actually requires here 11 between them, so I think I suggest. You know to point out that there is a zero left between them or I mean in case. You know people will try to cross-check the example against you know the syntax they might just get confused here. That's next.

C

One I mean typically, if I want to be pedantic, you want to, you know, might want to tighten they be enough for this, but I don't care as long. If you want really, you know on leading zeros here, it's fine, but it's, but if you say so, if you don't, you know you can tighten the AV enough.

C

C

Yeah this for the policy document I wasn't sure why certain characters were prohibited. I mean it looks like the war, it's exactly the same syntax as in the DNS record, but I'm just asking is there reason why it has to be the same.

D

So I thought the reason Daniel Margolis again I thought the reason that we were disallowing.

D

It's just basically doesn't matter as much the 30 degree. Disallowing equal sign is because that's significant between the key value pairs. So it makes it easier to just do a split, but you don't escape the ring like that, admittedly say the first.

D

C

Another thing you know you just long space here so if somebody adds a field which is just a three line comment with spaces, this is actually just allowed by your grandma I. Think that would be annoying for people because effectively you are the field and you make the whole policy document invalid I, don't know, yeah, that's sort of why I thought you know it. I suggest you relax this.

G

And this is gonna, go down. Okay,.

D

So I said to the to the white space Dana Margolis again, I think.

H

D

White space is absolutely fine, because it sort of doesn't make party more difficult, I, don't have a strong feeling about equals and I, don't think it sound like Colin there and I think. Maybe because we reuse the ground um equals seems like it requires that you should be a little more careful in the parser to do sort of split up the first equals and then treat the trailing string as a value, but actually.

C

Equals is not relevant for the policy document, because you don't use it as a separate at all. You call :, does ya and semicolon as well, is not actually relevant.

G

So we're done with minor issues: a okay, Jeff I was channeling Victor D on back to minor issues. 6 Victor says that he would list two quoted strings and that one.

C

That is a perfectly fine solution to the CS and say you know two water strings and insert the boat end in between or something like that, all right, just a clarity question. Basically you know to make sure that is. You know either you're clear that it's just two pieces of from the document or it's a you know, fully complete piece. You know or documents of it right.

A

I mean as long as you're everybody's fine with it all right. This one.

C

Yes, so this is basically your mansion yeah. Actually, this is probably slightly less minor than intended to be, because our RFC 50 to 80 is already pickax document is a requirement. It describes how you validate chains, how you check for certificate, expiring and key usage and other things. The text sounds like you describe everything that needs to be checked, but it might be taken as a contradiction with how you will do it. Otherwise,.

D

Daniel Marton, sir, so I think we're fine differencing. The RFC really do any in just sort of eliminating most of this text. I think we weren't sure what to do have to use a jurisdiction so I would service everyone. A mirror sort of HTTPS is okay, not having a certain key usage or anything like that, but I.

C

Don't make sure I'm sorry I, don't actually know the exact answer to this question, but you can have certificates, voice, mom or you can have certificate for websites and you might not. You know one might not be usable for other purpose so, depending whether the client agree well, no I'm not suggesting you usage as suggests that maybe should verify the usage.

I

Jim Pelton: do we allow certificates to be heard after again, rather than by train whistle just a minute.

A

Yeah move all right pick: two Europe.

J

As well because there's a whole bunch of practice- that's not just driving 50 to 80, but it's common around ik use further up the chain and I, don't know which he could use. He was asking about.

A

Yeah, you broke up a little bit. Victor sorry.

J

About that, maybe I should lower the volume on my end so that it was clear. I was asking about ek, use a higher up the chain, or only in the lease certificate which he can use to. Daniel want to ask about.

J

Because I'm, maintaining the validator and openness to sell these days, and so I can speak to at least what that does.

D

So I think we want to do you know the design was to do as close to what you would do when validating any other HTTP resource. If your surrender so I think that's sort of the guiding principle, I had to be honest, actually don't know what browsers do about any video certificate, key user choice and they're restricted by but.

J

So users are generally honored. So if your certificate is not valid for TLS as a server, then it will not validate. If, if that's, what the application tells the verification, library that we're verifying a TLS server will fail if it's an s-line, sir generally yeah.

A

All right Martin, not the Michael.

K

Thompson I think there may be a simplification you can make. You know you can simply say that it must be valid for that hosts in descendants because we're talking about HTTPS and there are certain expectations of a city PS. You could go to the trouble of referencing, 61, 25 and 28 a 18 and all of those things. But frankly, what you're gonna get is pickles HTTP decks hmm and that's the intent. You know all this worry about 8k use and whether or not.

D

K

That's on properly chasing OCSP stapling, the CT, it's just a mess, right so just say: HTTP and you'll get all the baggage that comes along with that.

C

Actually- and this is.

L

Not this is not a girl browser.

H

C

This is automated client so that you know.

H

C

Well, I think the text should be clear what you know whether it was exhaustive list of things you need to check and I suspect it isn't so or you rewarded, saying you know you know maybe bit more hand-wavy but but more specific about what the expectations are. Sorry.

K

I need to be clear about this point: I was being hand-wavy all of those things policy decisions that the client makes, so you could choose to do any one of these things or all of these things or whatever subset of these things you tame appropriate, but right now, all of our documents are very, very careful to sort of fudge their way through this thing and and if you want to try to fudge your way through it here as well, you're likely to step on something or in something I.

C

Think we're sort of venturing into this is might not be quite interoperable, but I think a lot of people are doing the same things.

C

Territory, yes, I know, and maybe we need a document. That sort would be more specific about this project are.

G

You hi Jeff Hodges. Are you referring to 28:18, which one? Oh? That's only about matching common name and stuff. It doesn't talk about all that other stuff, so he already in the draft he already references, 6125 and seven or nine places I just grabbed it in Aris, yeah.

A

They used everybody.

G

Needs to use the mike is kind of some technical debt. We haven't addressed yeah. We we probably need to write something about what you were just postulate. He might exist and I would claim, doesn't really.

K

Yeah I will concede that none of this is written down and that's not important. Yeah, that's kind of a deficit and I don't want to put someone in that pit because that's a I mean do you remember how long it took to get 6125 out and that's just by this tiny little tiny little slice. It was years right.

G

Yeah, it was a lot of work, but that slices one could claim done. 28:18 is actually a lot smaller, but whatever you know, we guess its technical debt that we should probably address, and this speck is begging that question.

D

But we shouldn't address it as possible, so I think like our sort of issue and clearly we sort of get it wrong when you try and specify actual behavior references, but our intent is to reference existing RFC's when possible and I think they're sort of at least for me sort of startlingly confusing anyway. So the the sort of ideal feedback is which we're not referencing. It should be where we're too specific and should not be.

A

Right so Morgan's proposal on the table is to replace all of that. All of this with a one-liner basically saying this should be a valid TCP, a valid HTTPS connection, which.

C

Is which is not specified anywhere right? Yeah.

A

Yeah I'm just understood, but it will, it will lead to interoperability most likely.

C

Okay, I think you know I think well, roughly agreeing where were slightly bike-sharing on specific text. You know another way of addressing this text is saying instead of making sure that it's not a full list of things, you need to check, say, for example, and saying the intent is check that it's a valid HTTP. Yes,.

I

This may sound like I'm, creating a normative reference dependency here, but it's sounding like right. Jennifer, oh I am I've reread our charter, but it sounds like we're identifying an item that has to do with using TLS and applications. It's something that may be another work item for this working room.

K

Tone so I was just about to say exactly what Jim said.

A

Anybody feel free to write up an individual submission for the working group and we'll consider it.

K

To the point that I made before it took Jeff and PSA so long to get 6125 out the door, which is the thing that.

K

Jew-Jew warning to everyone: yeah 28:18 is wouldn't that go down. Well, it's been discussed yeah. It's a massive amount of work. So just make sure this full disclosure here yeah.

L

E

Hobby I, just I, think the for example. Approach is actually a great idea, especially if you use the name as the example of one of the things you should validate, because that's what people always forget to validate every time. They're doing this stuff, hi.

G

Jeff, how does it it may be sufficient in this respect to do to plug in the suggestion that Alexi provides your punt to 50 to 80, and this already address it? It already prompt more or less properly references 6125 in appropriate places.

C

Moving on yeah.

A

But yes, let's Jeff has something Victor.

G

I'm channeling Victor again victor, says: I. Don't expect any interesting deviations from normal HTTP practice. So a brief. Do it like other HTTP clients do is probably okay, +14 mark yeah.

A

All right so Daniel do you guys feel comfortable crafting text based on this yeah.

C

A

Work it out on the list, but I think we have concerns as well on where we're going.

C

Actually, this is something that Chris brought up in his review: the use of a Senayan requirements on its use.

C

This is not a very well form concern I, just feel that certain limitations will.

C

Suffer was trying to satisfy these requirements, especially well on there I.

C

Think server-side I'm less concerned about I'm more concerned about requiring on a client-side.

D

The server doesn't.

F

Have to implement it.

D

Because you could have a server that serves a policy that only has used ever for one domain right, but the client doesn't know if the server needs it or not. So if we're going to allow servers to ever use us and I, we have to require that by the Sunda it rain.

D

hmm I know. Why are you white raised it that way? Yeah.

M

I think the way I put it is I was interested in hearing from clients implementers on the state of S&I implementation in their teeth in their SSL stacks because I, it's not clear to me how widely implemented it is, in essence all stacks. Okay, so.

K

Mom Thompson there are until I stacks out there that don't support SNI anyone using one of those two less taxes own already. Basically, the this is Android 2.2 and what are we up to now? 17 or something yeah, this it's crazy, but that's kind of the the era in which we we're talking about even even Windows XP supports S&I. So I, don't think this is a problem. This is very widely deployed. There's.

K

Actually, if people are interested, there's slides in the mukherjee research group they on on how off how much they see of this, it's still not universally deployed. But if someone's writing new code to do this, you're using a stack, that's got, it doesn't like any way any any rational person does.

G

So I'd be Victor requiring sni on the client works in is necessary. Yes, those stacks are not used by mta's.

H

A

It's probably uncommon for for Android 2 devices to be running empties.

M

Which was say stating them. The server must implement s and I and I kind of like a clarification on that of yoyo, that it's okay for the server to implement the extension and ignore it, because if the server doesn't need to use it, yeah.

D

Daniels that wasn't our intent to require the server to, as I said before it may not be necessary. So.

A

So maybe the right sounds like the right clarification areas to go to say that the clients must support an implement as an I but and be silent, and whether the server does or not.

K

You said, implement I, think we're talking about use. It's.

A

Like a used, sorry absolutely.

K

So maybe maybe you have a victor speak to this and then oh you're.

A

On I think, oh, you went away.

K

There's one other point that I wanted to make: do we connect to servers using IP addresses at all ever never, okay, okay, good! All right, maybe.

J

K

J

K

H

J

Two things, first of all in the deign spec: this is a nice. She is covered in a bit more detail and it explains that servers are free to not implement this anion completely ignore it. If they do implement this and I, they should certainly choose the right certificate if they can, but what they must not do is abort the connection if they don't have a matching identity for the SNI. Clients often support multiple identities, but can only send one, and so the server should not decide that just because it doesn't have a matching name.

J

You know we need to hang up. It should return some default certificate and leave it up to the client to figure out if that's acceptable- and this is important for a person TP, because multiple identities may be supported. We don't know whether the client is an STS client or a Dane, client or whatever, or even even opportunistic, but doing us an eye anyway, but doesn't actually care what's in certificate. So service should not get all picky about us and I. They should do their best, but continue under all circumstances.

J

Now, as for IP addresses, in fact, there are cases when SMTP uses IP addresses only when there are manually specified gateways for destination if an administrative override saying send mail to place X via smart host given by IP. That can happen, and in that case at least Dane just says, if I remember correctly, that that it's not applicable sorry, there's no danger. I P addresses and you can say the same thing about STS, but if you specify a gateway by IP, then all of this stuff is out of scope.

J

You can do any kind of pinning or whatever certificate checks you want by local policy, but both STS and then can't really deal with IP addresses. That's it from me for the moment.

K

Something you can have, you can have an ugly IP address certificate, so those are those a valid not but yeah. If you know, if everything's many okay many way configured at this point, you you're fine, yeah,.

C

Alright moving on is it the last issue from this I.

A

Just want to make sure we agree on what the consensus areas is. I.

C

Feel the consensus is now change the current Exodus possible adequate. Is it.

A

So so I actually heard Victor's make a suggestion to ripoff text from the SMTP specification. It.

C

Doesn't talk about a semi, it's very old spec smtp start TLS yeah.

A

Yeah, that's so.

B

The consensus that I've noted in the notes, which could be wrong and tell me the change it is that we've agreed that clients must implement and use, and we don't care about the server's, whether that's a textual change or not. That's what should end up and serve they eat.

K

B

Change then make a change and if there's not, then do it all right or people. Okay, with that yeah.

C

Just authors to review for consistency.

A

Yes, module all right next I'm 12 yeah. They.

M

Suggest two considerations.

M

Like yeah yeah, this is just you know, noting the you know, noting that this increases the attack surface of an MTA, and so you know that needs to be mitigated, and this discusses that I, don't think you know. Hopefully this is not controversial, it's just you know AB paragraph and we're done I think all.

H

K

Maaan Thompson V from this threat can also be partially mitigated by isolating I cut. That the key point here is that you've got a greater exposure and I think. That's all you really need to say I, don't mind a little bit of extra. You know you might like to do the following, but there are many many many things you might do to avoid these style of attacks. These.

A

Are just more or less text less sticks.

K

To the both of them, starting with this threat, no, it's there's two low starting with this threat, so this.

M

Threat can be partially mitigated.

A

Anyone will ever be otherwise.

M

Good suggest ways to mitigate they don't have to be the only ways.

A

I'm sure this will sort of if the security directors or.

M

A

Like some support for keeping the text, as is some support for not carrying- and you know, I think.

C

I prefer the text but I'm sort of more adult about this, so I don't mind either way. I.

A

Remember this is gonna go through secretary, vo and they're gonna need to pick out security considerations, so you know.

M

We have another chance.

A

M

All right, but it's not a showstopper for me and.

H

J

I just wanted to briefly back to us and I, really think it's important that not only do we not care about the server, but actually we do want the server to be liberal and continue, even if it can't find a specific certificate for the request of this and I. Otherwise, you get into interoperability problems with clients that are indicating SMI for reasons other than STS and may Peter, as I said, opportunistic ordained or whatever.

J

You simply don't know how the client will behave in response to a default certificate and if that decision to the client, not abort the connection I think so.

A

So Victoria seemed like you had text. um Could you like drop the text to the authors as a suggestion sure.

J

It's an RC 76-72 yeah.

A

You sort of just cut and paste or give them the reference or something right. Okay,.

J

There's a whole section on it, yeah sure is.

A

That, okay with everybody all right, I'm seeing nods and interim yeah all right good.

A

Okay, all right and I hear that weakens the consensus on issue number 12 is that we keep the text I think there were some people who expressed support for doing that, and you know again we're gonna revisit this during Sectary review zone, all right number 13.

A

This is DNA dick I think this wasn't Christian yeah.

A

All right, we know that and that's the last one so so. That concludes our walk. Through six year of you secretary review. They are nothing yeah. Let's it's yeah! Let's you.

C

Mentioned the resolution so film home Baker said that we must register the underscore SMTP report or whatever it is in there I on the registry and well. He was suggesting that we should do the work to actually define the registry. If is not established as it turns out, there is a registry in work. We should just use it. So my suggestion is: whichever documents this one or the one creating the registry gets published. First.

C

Doesn't include the text if it makes sense yeah.

A

So so, to explain this that this is a registry being created in DNS or- and it's basically you know the kitchen saying stuff registry of all, all the underscore bits that they found in in various documents. They're just collected and nothing created an eye on a registry for its sort of household painting and I. Guess. Your suggestion Alexi, is that we we sort of do a race condition on publication of that document and if they make it to RFC status and establish that registry.

A

Before we do, then we can add a note to the RFC editor saying we need a registration in in that right.

H

A

If we get there first, we as we don't have to do anything because they we tell them. They have to pick up another code point from our document and we become legacy another bit of legacy for them correct in either way.

C

I will suggest specific text to it. Yeah.

A

So so, but what I think you're talking about here is to create a an eye on a registration entry for the registry which doesn't yet exist and and and an RFC editor's note saying you know, use this. If you, if you can- or you know if the registry doesn't exist, you know we'll handle it in another way. Oh the DNS or thing it's called I can leave something yeah.

C

It sends progress, but I think it's getting there just.

F

A qualification feel review tool is pretty draft, but the issue could apply to both drugs and Dallas days and the rest are pretty so they both use specific DNS prefix. So just the clarification so also are the same.

K

So you've got a normative dependency on that document. That's just well! That's yeah! Yeah.

L

C

We don't have a normative dependency, I, don't want it, but we can fix this by whichever documents published second will have the text. It is registry establishing the published second, and it will have a reference to our C, which is already done. It's the other way. You know other checks anyway, I think it's it.

A

Isn't it's exactly a chicken race with the DNS of working group.

N

When it goes in, are you know, I will check out the documents and it'll drop it because it they.

J

N

Say that there is no drug is to created with this name. Iein actions cannot be finished and if it worked there- and so it that will happen in it before actually I, think they actually do that the same time they do this, some other is I'm sure I can manage this all.

A

Right, the question is Alexei in order to manage this, what text do you need.

C

I will provide it all.

A

Right, in that case, we're done with this document.

C

Just one that the side comment of this was suggestion that the label is blue. There is a separate registry for top-level underscore label or second of underscore label, and most labels are second double underscore labels, so I think Philip, suggested rate change, change to levels to underscore you know SMTP and dot underscore report, or something like that does.

A

That mean we have to specify what the underscore report thing is. Yes,.

C

Well, it's them adjust to two records instead of one I can provide the text date doesn't matter right.

A

But it doesn't means writing a separate orosi to establish an underscore report registry. No.

C

They are both registered in the same. Thank.

A

C

Know all right, let it do people have a strong preference. One word about two levels: verse one I mean it's potentially slightly operation. Eight well, I think the idea is, if you create underscore report, you can start other things underneath it, which is not just this document. What other things I think that that's? Why I think.

A

That was, fails playing train. You know other things.

C

D

We had two levels because we thought it would be harder to abuse domains like.

L

Those random subdomains and Exodus and we had sort of two to nestings that make it hard. And then we.

D

Got rid of it and went back the way it is for reasons I, don't remember, I think, because people have wildcard domain certificates already we want to let them use. It actually was the reason that we think about it. So.

L

There was sort of some.

C

Rationale behind that.

D

A

Think no change is fine. All right well alex is doing. The house has volunteered to do the chicken race with dns opened the oracy editor excellent I. Think that does it for this document. We've got our final presentation for today the rtls good.

C

A

C

Do revision of this yeah.

A

C

A

Do it quickly because well, this is past working applause, call it post everything right! This is the final relation. It goes out.

A

So you're plenty of time, one.

C

Of your cord, it is, is very active, so I like Alex is I think was quite.

C

A

All right, let's try to get it talking.

C

Well, talk to them all.

A

Right all right record hit us Jim, Europe, okay,.

I

So put your hands: how many people will already know what required TLS does how many people don't know and neither review a lot of people are in either category okay, no, no review, then, because that's significant show of hands: let's go the next slide. Okay, so basically what the I've done done. One update since the since the last IETF, which basically does what we discussed at the last IETF. We now have two different mechanisms that are part of this same graph.

I

One is an SMTP option which allows you to require TLS for the transmission of a particular message and allows you to specify some things about that: the characteristics of that TLS connection and then there's now a header field. That's it's! The header.

L

I

Actually required TLS, but it has one possible value which is no, which is used to override policy based requirements like MTA STS, like Dane and so forth. Basically saying this is a message that I want to get through, regardless of TLS security. Next slide.

I

And this is kind of this is kind of saying that much of that well, the issue with a header field, of course, is that you don't have any way of assuring that downstream hops will implement, require TLS. No, so it might be ignored, but it's basically it's basically a best-effort sort.

H

I

As opposed to the required TLS SMTP option, which first, which requires that it be negotiated, requires that the SMTP option be negotiated on each hop, which basically guarantees the or basically asserts a promise from the risk from the SMTP server that it will continue to heed the require TLS requirements next slide and I'm just ahead of myself here. So that's basically what I just said.

I

There are three options: one is to require the MX lookup to be DNS, SEC protected, which I think is probably the security concern that Chris had started to think about in the context of MTA STS you do in order to, you know, really be protected against attackers here. You do need to make sure that you're doing a DNS, SEC ma MX lookup, because of the direction that this mechanism proceeds.

I

You have the ability to restrict how certificate verification can be done. If you don't like certificate chains, because you don't you know- maybe you're in the country you're dealing with a country that owns the CA and you're concerned that they're going to manufacture a CA or something like a certificate for the domain that you're sending to or something like that. You have the ability to say well, I just want this message to go by adding a verified certificate, so these options are somewhat negotiable.

I

We've had some opinion that maybe this is a little bit too much optioning and you know there are other opinions, I think around that you know. Maybe we should even have more options to specify what end ciphers and all those sorts of things we haven't done that far next slide and I'm continue to be ahead of myself here. So.

I

Next slide main issue here is you know if, then, the update and so forth kind of sent how to send out a request for reviews said it was probably during the holiday, so I sent had another request for reviews and still for a little for a little comment. I have gotten indications from some other organizations that that they feel like this capability is important, but these are not the sorts of organizations that necessarily participate in.

I

You know commenting on things at IETF, so I would really appreciate. If people looked at this- and you know- is this valuable- this is not valuable or are the details right all right, I think.

K

That's it. Why would you turn off to us I mean then they're, you know just have an optional security mechanism again and we don't like those anymore use case Oh.

K

What why why do you turn off TLS? Why would you do that.

I

Well suppose suppose you had some sort of a some sort of a.

I

You wanted to send a tornado warning, you know the weather service, you want to send tornado warnings just triggered victim and and and and so the idea is, it's important that the message get there: the security, it's completely public information. So really you don't want the the message to be inhibited by a policy mechanism.

I

J

Yes, I, don't think the required. Tls no case is turning off TLS, its turning off mandatory, authenticated TLS from Dane or STS or similar in recognition of the fact that sometimes there are operational errors and one thing one needs to do is inform the people on the other end that you know they have a problem. So that's a good use case for turning off TLS.

J

Another good use case for turning off TLS is you're, a customer of an ISP that, or you know, service provider that implements Dane or STS or both, and you have an urgent message. You know tornado warning or the like to send. Somebody and the destination is broken, they're messed up their policy, but you have an urgent non sensitive message. You should be able to say you know some damn little, you know damn it's and the torpedoes. You know what. Why can't we let the user decide. The delivery is more important than confidentiality or or security.

J

It's their choice with a browser. They can click OK and you know and go through all the warnings and go to an insecure site with email. It's asynchronous. In the background, these are can't click through all the hops. This just provides them an equivalent mechanism. It's the click. Ok, but you know dumb asynchronously. In the background. It's perfectly fine. Ok,.

K

That sounds far more convincing than the first example. The actually got up to talk about another thing, which was all of the optional knobs and the idea that you might control policy on on subsequent hops.

K

With with this the option that you're talking about I, don't think that makes hollow sense, particularly when you get down to things like controlling so I persuades and and all of these things, yeah there's an infinite number of knobs you can, you can twiddle and what you're asking is for someone to override their policy in a myriad ways for the next hop where you should just be saying: I have an expectation of security. That is, you know, please don't send it in clear text and please authenticate the other end.

K

You you want to establish the baseline here and I. Think that there's reasonable guidance in the existing documents. For for this you want TLS. Then you want to authenticate the other end right. That's basically what we're looking for and to some extent you have to have to trust, go to the next that the next hop is going to respect that in the white in the best way. That I, can you just handed them a message and so I think.

K

That's probably all you need it boolean flag, turn it on make sure it's on or don't deliver it and I. Think that's a reasonable thing to do. How.

I

Do you feel about the the the DNS SEC of required flag, I.

K

Don't see that as particularly meaningful, it's another one of those policy knobs that someone who did write.

I

So it could but it, but without that you have a threat that somebody who can spoof DNS can you know, send your mails completely someplace else. Then.

K

That why don't? Why don't we just require that all ones.

K

Yeah, that's the question right well,.

I

So if because it's moon, the reason, the reason that we.

K

Wanted to be secure or not is the question sure and and.

I

I'm open to whatever the consensus is on this by the way, but I mean I'll. Express the the flip side of this is that the sender of the message has some idea of what their, what their threat profile is, and you know might be concerned about nation-state attackers in some cases or might be concerned about attackers that can spoof DNS. So so.

K

I just sent an email that there will sort of explain my position on this one. Okay,.

A

K

Right, no, it's good I'm relaying! So.

B

A

Before you do what the picture was first in the character.

J

So Victor all right, you can speak I completely agree with Martin I've already in past meetings.

J

Multiple times express the fact that I think all the fine-grained features are a bad idea should be a please deliver securely over some kind of TLS that you know to be secure and any attempt to be much more specific than that is a non-starter I. Think all the people who might know how to specify additional properties and have a reasonable expectation that they're doing the right thing are probably in in the meeting room and- and nobody else will ever you know, be able to correctly specify all the mobs for multiple hops. That's just insane.

J

As for DMS neck, the the only policies that require TLS, you know presumably, is compatible with his either sts ordain. We don't have any other secure delivery mechanism for start TLS, opportunistic, isn't it and both SDS and Dane find various ways of dealing with the insecurities of DNS each with their own mechanisms.

J

Let the realize that each op decide keep it simple.

I

Well, the difference here is that if the MX record is changed, you're actually changing the the common name that you're going to be comparing a certificate off of so really verification of the of the certificate sort of becomes meaningless. At that point,.

B

This is D key to Mike without being a diverse card. Did you.

A

J

In DNS, SEC have mechanisms to avoid forgery of MX records. This is a non-issue.

I

Which is why I'm proposing to have an option to require DNS SEC, no.

J

You don't need you simply do not need it gain. Of course we are see in a sec, but STS works, as you know, as well as it works. You know it's got a first contact issue, but otherwise without the intersect. But you know here's me trumpeting. You know it's TSI, usually you know say bad things about it, but it's you know mostly good enough if get effect isn't available. Let's it's. Let's get mostly good enough in the field. It would raise the security card to buy people turn off security, read RFC.

J

Seventy four thirty five, opportunistic security is a good idea.

J

A

Right, Victor I got a person on the mic. This.

B

Is dkg I'm gonna relay Ned freed from the jabber room? He says the history of TLS and SMTP began with Microsoft shipping, a product which, out of the box, failed whenever search LS was attempted I'm much more concerned about the ability to control the character about the ability to control the characteristics of TLS across multiple hops. The essential characteristic of email is that it gets through when nothing else does compromising. That has to be done carefully.

B

So, to be honest, I'm not not entirely clear from the text whether he is in support of the fine grained flags or opposed to the fine grain facts taking off my jabber relay hat as dkg I am I think opposed to the fine grain Flags because they seem to require that everyone who implements require TLS have a suite of options available to them and I think that there's a sort of narrowing of possibilities that your message will actually get through when you do that, and so I think I agree with Victor.

B

I'm gonna need to go shortly and I'm taking notes right.

A

M minutes left and we've got one minute left so um final thoughts on the mic very quickly.

L

Absolutely as simple as possible and also a future-proof is possible the more we tighten down exactly how it's done, the more the sender's going have to be revised. If anything, changes in the environment worried it's better to say, a secure. Channel is the best practice at the time right.

A

So there seems to be quite storms to put in the room for, for you know, removing as many policy notes as possible from this draft. So and but of course you don't, as you say, you don't have like a lot of review any any of you who have participated indeed in this discussion. We really appreciate this if you could take the time to go and read and review Jim's draft, because this is a working group document and we're gonna get it done with that.

A

We don't even have time for open mic and we'll see you when we see you we'll be in Montreal. Thank you.

O