Internet Engineering Task Force RATS Architecture Design Team, 6 Oct 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: RATS Architecture Design Team, 2020-10-06

Description

RATS Architecture Design Team, 2020-10-06

A

Hello, um sharing sharing is caring, hello.

A

Is that why we all get along so well, because we're always sharing our screen.

A

Sharing is caring.

A

No you're, not a fan of the care bears. Oh my god,.

B

I didn't don't really know.

A

Don't really know well, the best part is teddy ruxpin teddy ruxpin's, another character that involves stuffed animals and and the best part about it. Is that the the villains that they're opposed to have a union and and when they get to a union, just fine break the the villains they stop like to go and have a coffee.

C

A

The monsters and villains organization mavo.

D

Well, I prefer witty to poo.

A

Well, you know yeah so winnie.

E

A

Doesn't have any actual, you know the only old threats to winnie the pooh is really winnie. The pooh.

D

Well, there's the heflogs and the woozles and, of course, those jugular.

A

Yeah but those are just acid trips.

A

I mean if you look at what how they're drawn you're, like? Oh, my god, like winnie the pooh, is clearly a creation of the 60s yeah yeah yeah. I know I I learned about teddy ruxpin, because I have brothers who are like 15 years younger than me. So there you go.

A

A

I guess I'll give another minute usually have a greater roster.

A

I've been sick since thursday, maybe wednesday kind of hoping that was my turn with kovid, because it wasn't that bad.

D

Well, I would expect hank to be here since it looked like you've been commenting on stuff.

A

Oh yeah, are you joining me he's in chat with me.

B

He was on for a second, oh.

A

He had to restart his browser, firefox told me he wanted to restart, and so I let it and then it just didn't, do it like it, it shut down, but it didn't start again. So I've noticed that sunum is due. Then you ask and they say I want to restart now and you're like okay and then it doesn't hank.

F

Yes, I am here this is my stupid, webex app. I tried it again. It was a mistake, so I'm here.

F

A

What are we gonna do work on here.

G

What can we um these? uh So? I think actually uh some of these are really hard. Okay and some aren't, I think the tester verifier one. I I didn't. I actually made a comment. I don't know how to deal with this one today, because it was so overloaded with different issues of text and a general disagreement of how this pr should be handled in general, and that was very, very difficult for me to address uh right now.

G

So I made a few comments and actually said: maybe we should cut this up, and uh maybe we should have an idea if you want to uh pursue uh this pr, uh which was provided by lawrence. I think.

B

Yeah so you're talking about 149 or 154, I mean.

G

B

I think the 154 is replacing 149 now.

G

Oh okay, that is good. That is excellent. Thank you because I was not really following that line of thought anymore, and it was really really uh okay. Let me go with. I commented this also and weirdly enough. I did not really see that it was exactly the same thing.

G

Yeah reading is solving problems. You say good yeah, then we can shut off with this one. I think.

G

G

Dave, yes, isn't that here.

A

Ned is not here, oh.

B

Yeah, it would be good to have nick for this.

A

G

No, it's fine, uh but.

A

Would you say he was gonna.

G

Typical joints, so he would say if he won't.

A

G

A

Yeah he's usually got something else beforehand and I think that's what he uh five minutes later.

A

A

um What what do we want to do with this? At this point, then,.

B

Well, I think that we need to get clear on what's local and remote attestation.

D

So everything in this document is supposed to be about uh road attestation, although some may apply to.

H

D

H

D

Think your main comment lawrence is that uh the word implicit you find uh confusing there, and I would not disagree with that. um I think the when you look at the term root of trust right, uh usually when people talk about a root and you think about like a certificate chain or the thing that you're, comparing against your trust, anchor list root would be the root of the chain. If you are using say endorsements, then the local hardware is in fact not the root right.

D

The endorser is, and so it might be equally incorrect to refer to, in this case the piece of hardware as a route of trust, because it's not the manufacturer is signing it and therefore it's not the route. The manufacturer is right, and so that is sentence I agree, is a little bit confusing in the remote attestation case.

I

D

A

Referring to the terminal.

D

A

I got a quebel. I think that the thing that you just described is a trust anchor. Yes,.

H

A

And not a route of trust, and- and we have this at the beginning of the rats conversation where, where we I.

D

I'm saying we've never actually defined the term root of trust and to some people, lawrence's statement would uh lawrence's problem with implicit would have the same problem with the term root of trust.

G

Yes, so this is so so this is saying um I I agree with dave that this text is basically still everything about. This is a remote. I made a comment in this section here that is closely not visible, just a little bit down, and- and so so this is the first observation is this is about remote attestation and it's all applicable to even the parts about hardware and level of assurance are applicable here.

G

It's just elaboration on security conservation for the route of trust that is in hardware, then I would strongly advise not to use the term implicit or implied here, and I think maybe we uh settled with vouch for and if you could scroll down a little bit, I made a a a corresponding.

I

Transition, what I'm wondering.

D

Is do we actually need to use the term root of trust, or can we just delete that line, because all the rest of the text is fine? This is just trying to say uh introduce the term root of trust. In some way. Do we actually need to use that elsewhere in the document? If we never use the term word of trust anyplace, we can just delete.

G

800., I think we use it, but I have not checked that literally uh yeah yeah.

B

I think about your voicemail, because there's.

D

B

10 occurrences.

D

Okay, because that whole sentence, I think, was somebody's attempt to define in context the term of trust, um and so, if we need to use it, then just deleting the term implicit um as laurent suggests would be fine. So there's two uses.

B

Of the word, implicit there's, also implicit trust in firmware, and uh that's the one that really is has got me confused or thinking it's local.

D

Go further up this component, that is the component is trusted. I think you'd say you need. We need to say something like this component is often referred to it's a the component. That is trusted, I think, but I need to read it in.

D

Context um or uh about such hardware just to match line 798 where it says, especially if such hardware, how about such hardware is often referred.

D

J

B

Okay, now there's there's this other thing thing that says that uh yeah, it was configured to implicitly trust, firmware or or even software um that I and I'm you know from the conversation with ned or at least comment ned made. I thought he thought that was referring to local attestation and I don't.

A

Think I don't think it's referring to local attestation. I think that that what it's referring to is that um that bootloader, whether it's a hypervisor or uh a one block thing that thing that that uh that measures itself um is effectively the verifier puts some implicit trust that it knows what it's it's it's work. It's doing! That's what I understand.

D

My understanding was that the key material that is used in remote attestation, in the case that's being talked about the weaker security case, that key material is not provisioned into the hardware that key material might be provisioned into the hypervisor and say you might have a hypervisor in dorser and so on, and that's what you're, using in your remote attestation case. That's how I read it saying it's just not baked into the hardware and so it's weak for security, and that's why 797 says it's stronger when you can make it into the hardware.

B

But the verifier is implicitly trusting the firmware. That's the problem. It's not there's no key material, you, the trust, is implicit without the without the crypto.

D

No, what you're the key, what what that means? You're, implicitly trusting that that's the correct key material right, because you're not completely simplistically trusting in this exam in the eg hypervisor you're, not implicitly, trusting the operating system. The hypervisor is the testing environment. The os is the target environment, so the hypervisor generates. The evidence sends it off to the verifier. The verifier verifies it using the key material that was provisioned at the hypervisor.

K

Different interpretations.

D

Tells me that this sentence should be cleaned up, which is what's this point so.

J

The there are, there are architectures where the the firmware boots, the hypervisor and so there's a trust. There's a trusted relationship between the firmware and the hypervisor. There are also te environments that don't depend on on. You know firmware.

D

Right with tu environment is a software te created by the hypervisor.

D

Or perhaps it's just you know, isolation.

I

Between you know, vms or vtls, or whatever, uh okay, the software t was a little bit confusing yeah.

J

Let's just say the dependent, depending on the architecture, the underlying architecture of the hypervisor you may or may not draw into its tcb. You know boot firmware.

D

D

So I don't know who wrote the sentence: that's now 792, but how should we clean it up to make it clear that this is about remote attack station.

D

This is the as lawrence points out. The word implicitly is confusing. Is there some way to rephrase this? That makes it clear about remote attestation and doesn't use the word implicitly.

G

Yeah, the component that cannot create its own evidence, but has to be vouched for, is often referred to as the root of trust.

J

And why do we need the word implicitly at all.

D

We don't need it right right. What's the right way to rephrase it and remind me which section is this? Is this the trust model section yeah, okay, good, just making sure.

J

But if you just remove the word, the sentence seems to say the same thing.

G

uh Yeah, no, it doesn't so you don't know why they are highlighting this is trusted, so that implicitly gives the sentence some gravitas by, because now you understand why this trust relationship is highlighted. If you remove implicitly, you have to add something here. Otherwise the sentence: it's just a weird fact that you highlighted you can just remove the sentence. I think.

D

Michael, can you scroll up just a little bit? I want to read the text that's before it should get the context here.

D

A

J

What's missing is the is the motivation for why we said weaker.

D

Yeah, the more I read this, the more I'm confused by that sentence too. So.

B

I mean I still think it's better to have any discussion of stronger and weaker security and security considerations.

G

It is okay to highlight a specifically from the security consolidation elaborate here and I think it's worth the while still today.

J

The the topic is on trust, and you know why how trust works and it's okay, to talk about how trust is different in in you know different and weaker environments, but it's still not really nailing down what that difference is, and that could.

D

Trust perspective, I can tell you what I'm thinking, which is different from what the text says, and what I can't tell is whether I want to change the text to say what I'm saying or whether I'm missing some other point. The text is trying to get across to me. The trust model is you can trust, so one approach is to say: I'm going to trust the endorser um and it's the endorser's job to to tell me which devices I should trust right. So that's one model, um and so I'm not trusting the device.

D

I'm trusting it in indirectly by saying, if the endorser vouches for it, then I will trust it. Okay. So that's one approach.

B

That's the remote attestation case.

D

It's the typical remote attestation. In this case it is not the only remote attestation case. The other case is where I sort of implicitly trust the manufacturer.

D

I implicitly trust the manufacturer and I establish direct trust in the device, and so I get as the owner of the device that runs a verifier. In an example, I get the key from the device I manually inspect the device I go through whatever security vetting, and then I put that device's public key in my trust, anchor store. It says I trust that specific instance, because I've checked out this particular device and think that it's good independent of what the manufacturer says that one you might claim is weaker. Some people might claim it stronger.

D

I wouldn't get into that. I would just say it's more of a direct trust in a device rather than an indirect trust in a device via the manufacturer that one does have the property that says um just because they device this manufacture manufacturer. I don't trust it. I only trust the ones that are mine, so the three of them that I bought are the ones that I trust not the three that are in the attacker's possession and so that's worth discussing.

B

D

In my opinion, right.

B

I see that makes that's the other case for you, so it.

D

Is a direct and there's an indirect indirect says, whatever the manufacturer says is good for me direct says only the three that I vetted myself, you know inspected to say these are the three keys that are actually in my possession that I have uh the physical protection around. So nobody can do physical tampering on them because I know which ones these are, and then I put those three in there and that's what that's. What I might personally refer to as direct trust and indirect.

B

um But you you, the the manufacturer, is the one that put the key material in right.

D

um Correct well often, yes, I don't know if there's a case where the answer would be no, but my expectation would be yeah absolutely and I'm just using say I'm extracting the public key when I vet that device and I buy it, and I make sure that this is the one that I know, and so I put that say public key into my trust- anchor store in the verifier directly, so that I don't care what the manufacturer says because he's going to vouch for zillions of other devices. He made.

B

Yeah because I.

D

B

D

Against physical tampering, too right as long as I have these three in a physically secure environment, all those other ones are outside my physical security. I don't know if somebody's physically tampered with them.

J

There there are roo there are cases and then, where the root of trust keys are provisioned by the manufacturer are expunged and uh and then.

D

J

Control of the owner, they're recreated.

D

Yeah there might be a case where say on first boot or setup: that's when the key material is internally created. The fuse is burned at that point in time, and now the manufacturer doesn't have the key either right that that is possible.

D

J

Mean a lot, a lot of this is like seems like it's implementation, detail.

E

It's certainly.

J

Relevant to to trust and and my.

D

Point is the differences between.

J

D

And indirect trust.

J

I mean, I guess you know it comes back to what what's the right. What do we want to say in these few lines that are fairly high level.

D

Right, what I don't know is if some other concept besides the one I'm talking about, is trying to get across here. That's actually important for this section.

C

D

I would just rewrite it to talk about direct and indirect, but I don't know if I'm missing something yeah.

J

So I think, there's another concept that potentially this line is trying to address, which is the notion that we, that weaker trust is based on complexity. So, as I kind of go from a very simple route of trust architecture, to a very complex route of trust architecture based on a hypervisor, one could argue that because of the complexity, it's weaker and so there's a lot of implicit trust built around that larger tcb. If you will, but you could still have um you know an implicit trust in the vendor that says I.

J

I trust everything that you did in that complexity and that's my route of trust. I'm gonna, I'm gonna go with it and you know.

A

Not to provide comparative statements about this is better than this, because they have they have different properties and they have different features and they're for different benefits.

J

And I'm not no, I'm not trying to be comparative, but it's just the you could you could have a a constrained root of trust which is not secure at all, because it's completely done wrong.

J

You could have a but you, but if you were just talking about complexity and and in the context of complexity where in theory, everything is done right right just you know it's like whoever did this knew what they were doing. They're the smartest people on the planet. They did it the best way possible, but it's still more complex and so the complexity.

J

People argue that complexity inherently is less secure.

B

I think that's fine, but I don't think it should be in the in the fundamental text about trust. I think that should be somewhere else, because it's a subtle point and and kind of a bit uh uh far afield from what.

J

We're trying to guess at what these lines are trying to say: that's just one possible that could that could have been what was intended here. We don't know.

A

So let so um so I removed the word weaker and implicitly became directly um and then I add a specific device. I don't know if I don't know. If that's what I heard, you guys saying that maybe not tracking what you were intending so I'm gonna.

D

I'm gonna be quiet for about five minutes, while I author some text just for you guys to review, and so I'm not gonna change any of this, I'm going to generate a a a delta that might add two paragraphs into here and then we can decide what to do with these, whether to keep any of mine whether to merge them into here, whether whether you use my replay stuff or whatever, I'm not sure yet. But let me I'm going to write two paragraphs, so I'm just going to be quiet for five minutes.

D

You guys can keep discussing.

A

Okay, so hank suggested this text down here instead of uh the component that is implicitly trusted. He suggested this as well.

G

And why is it lowercase hank because we don't define it? It's not our. It's not a that's a lawrence.

H

G

From somewhere else that he said, this should be lower case. Was that yes, so I just incorporated this here on the fly.

B

Yeah, I felt another issue yesterday that yeah suggests roots of trust should be lowercase.

G

Which is fine with me, so that's why I uh did this here by editing. It might be out of the blue for most.

J

Right are, you is the word endorsement referring to the the new more constrained definition?

J

G

And also should be plural, but I think it did plural with an edit. So if you have five this, it should be appropriate.

A

What how would you feel if we replace components with such hardware.

G

uh Yeah theory, roots of trust, can be in software, so practically they never are. If you call them that, but um yeah, I'm fine with hardware.

B

G

K

J

There's software has to run somewhere. So in your you know, there's a it's always a combination of software plus hardware. You can you can't just have hardware running in in you know thin air.

A

J

Yeah, unless I checked.

A

Creators of the pond great game would disagree.

G

I mean at the the turtles end somewhere and then there's hardware. I think you are right, so that is correct.

G

But that's not the point here. I think uh there can be virtual tpms. That would be software root of trust and yes, they need a hardware one to really functions, for example, but uh in theory they could run on their own.

J

G

J

G

J

You run your software tpm in a hypervisor yep or in you're gonna guest or whatever just run it as an application.

E

J

C

J

Because of the design of the tpm, as in as a quote, root of trust bit of code you're, just saying that your application is the root of trust, it's a weak route of trust, but you know that's yeah,.

B

That's allowed.

J

B

um And not all attestation is tpms, um you can run it in a in a.

H

B

High level os and that that could be a bit of trust. So that's.

J

The point the point is that tpm is designed as as just software yeah right, you can run. You can run that software anywhere right.

B

Right qualcomm shift shift uh the ftpm that ran in the tee.

J

And it seems like somebody has software tpm software just running as a web service for testing. You know it's just.

G

These exist actually yeah, and so I'm fine with component. We could call it a system component to be very, very clear, but I think that's just more words.

A

I'm going to remove this one.

G

um We could look at the individual comment threads instead of skipping them.

G

I think this is the one we just addressed: correct, yeah, okay, so there's another one here with dave. It's just an editorial. Okay,.

B

Yeah, that's the part that dave's writing text about right now.

G

Oh yeah, this one here, okay, I.

B

Think so yeah in fact.

D

In fact, it is relevant to meaning the the answer as to why typically would be necessary does have to do with the text that I'm writing so in that sense, lawrence is correct. It's related okay,.

G

So lawrence I tried.

D

There now I have one short paragraph, but I need to either have a second short paragraph or another half of the same paragraph. So I'm keep I'm going to keep working.

G

uh Okay, that's what we do, because that doesn't be uh half goal here so now, as I try to uh rephrase yours does it still, is it did they break it or is it still? Okay,.

G

Because you were talking about entities and roles in the same sentence, and it was a little bit confusing to me and I think student. That is something also confusing so.

A

That's what you would like to do is just add a sentence at the end here.

G

I think that lawrence just added a sentence at the end. Yes,.

B

Yeah that was that was the suggestion was to I was suggesting to add a sentence.

B

um I I I still prefer my sentence. um The help of the endorser um I mean I mean. The point: is that the trust flows through uh the endorser or the attester manufacturer.

K

J

But that flows there's dissonance in sort of the cross use of roles and entities. So we we need to stick to we. We should try to stick to one or the other, but not both the roles and entities. An endorsement is an entity.

B

But the problem I mean we have this big problem in the the endorser is endorsements, are just an optimal an optional optimization.

J

I don't know what that means.

B

You could do it, you can do out of station with endorsers and endorsements or not, if you, if you don't, you don't need you don't have to have an endorser or or an endorsement. If you look at the all the usage of endorser and endorsement, it's all about an optional optimization.

B

This is kind of I mean I I spin around on this. But if you look at the.

J

I mean I, I don't think, we've closed on the whole conversation around what an endorser is as a role and we've we've. I think we've added an another role, so I.

A

Don't like that, I don't like endorser, um because I think of the reasons that that lawrence is given is that I that it's not necessarily the um it's not necessarily a required role, um but I do like the word endorsement and and where I think lawrence and I differ, is that I think that endorsement doesn't isn't always signed. It sometimes just arrives in a trusted fashion, because it's keyed into the verifier by the verifier's owner. um Oh no, I don't know. I.

B

Agree with you on that. Well,.

A

Well, so so in my mind we should we should. We should gather around the the endorsement and an endorser is a is a scalable way to to communicate endorsements. But it's not the only way to communicate endorsement.

G

So somehow someone has to provide the key material in the end as uh lawrence always says, and that's why only because we don't uh describe the vehicle here. I think that without the endorser this whole breaks down. It's not an optional role, so the communication.

J

Defines here is optional: yes yeah, I don't think it's. I don't think anyone's saying that endorser is optional. David we've made it optional in our framework, because we.

D

uh Endorse yes, you're correct endorsement is completely optional for reasons. I've explained about three times would be happy to do so again. If I wasn't authoring text right now, so, okay.

A

So so so, in my mind, endorsement is, is key and that it's bizarre, maybe to think that you can have an endorsement without an endorser. But what I'm trying to say is that it's in it that relationship is is one where we we do something cryptographically and if it just arrives by fedex on a piece of paper, you can still get an endorsement. That way.

J

But who sent it to.

A

J

You're getting it from someone sending it to you.

L

J

Even if you're sending it to yourself, you become the endorser.

A

Exactly I I I take your point okay, um but I'm trying to say is that so what I really did is I made a measurement at some point of of some characteristics of that specific system, and then I locked them down on the verifier right and that's that's what I did um and uh uh yes, I'm the I'm acting in the role of the endorser there, but exactly but but what I'm trying to say is that I think that's just that's. That's uh that's not neces.

A

I think that it fogs the notion of of the other, uh the other way in which the endorsements more typically arrive, which is through digital signature, on them and and- and I think that by having to say and or a whole bunch of places that are.

H

A

Actually been we're not actually we're not actually adding light uh understanding to the process.

J

But when you said so, digital signature signed by whom now we're.

A

Talking about the digital signature, then it's got to be signed by somebody. Yes, the endorser, I'm.

J

K

A

J

So and so as entities have keys and entities can sign things.

F

J

Endorser is a role that performs a function and that and you know, the role is hosted by or performed by an entity. So when you say an endorsement is signed by an entity.

J

You're basically saying that the endorser role is performed by the entity, that's that has the key that signed the endorsement.

G

J

You're not using an endorsement.

G

Sorry, if you're not using an endorsement here, uh you are still an endorser, because you configure the verifier to act in a specific way. Then verify your owner and dorsa are the same.

A

So so we wrote the course we wrote endorsement was a secure statement that some entity, typically a manufacturer vouches for the integrity of an attester's signing capability.

D

Yeah I mean they interrupt. How would you like me to submit it? Would you like me to update stuff into a comment here? Would you like me.

A

To be better, what's going to be easy just before we before we do that, because I'd like to close this edit before I hit reload, um I I I will I I we. We had two sentences, this one and then this two pieces and the question was which one did we prefer, and this was lawrence's original sentence, and this was hank's attempt to improve it.

D

I think both of them are wrong.

A

Maybe we shouldn't have to have dave's text. First, I'm going to hit update then.

J

J

Sort of broad observation, after having gone back through version, six and just reading kind of the whole thing we seem to, we seem to that. The text seems to be inconsistent about the difference between an entity and a role, and we often use the role name when we mean an entity and we sometimes- and we even say things in the definition. Some of the definitions say that a a role is an entity blah blah blah, which is wrong.

F

J

Others say it's a role that is typically hosted by an entity, blah blah blah, which is correct, but we are inconsistent about.

G

There's a pr fix in that.

J

So in any of these, in any of these sort of proposed texts, if you're careful to read it in terms of, are you describing a role? Are you describing an entity.

A

A

I just committed your hyphen and I sort of just didn't miss the question you had about idev id and, if you're looking for a reference for idev id. Yes, there is, but we don't use it anywhere in the document. At this point surprising to me.

G

So, which line? Did you uh create your uh proposal for uh dave.

D

I've not yet I'm looking for the right place to copy it in, because I've opened my editing. I need to find in the pr what line it was now that you scrolled up and went down.

C

A

We were, we were all arguing here about 793 before.

C

D

I was looking for that line that had the word uh um uh implicit in it that we had all those comments on. I'm not 792.

J

G

This is one of them.

A

The other one is is, is down here at 800.

C

792 has no dips in it. Let me expand.

A

No, that's right it doesn't. It didn't, have a diff in it until we added.

D

A suggestion 792 so right now I have uh 149 plurecast 149 files and 792 is in the middle of the conceptual messages section. So that's not. It.

A

Are you looking at the original file, then.

D

No, I you asked me to put it in as a comment on 149, and so I went to the files view to figure out where to paste it and which one.

A

I don't know why are you doing it? 149. 154., we're on.

D

Well, that's why yeah you're right! That's! Why thank you? No wonder! I'm having a problem finding where to paste it in my screen.

D

What happens when I switch from your screen to my screen, as I put in the wrong one, all right, 790p yeah! That's what I'm looking for all right, I'm going to start at 791 the blank line here. So it's not necessarily on a paragraph but in between so.

A

Hank hank kind of likes pasting the whole thing into kodi imd, and we can all edit it at the same time, and I'm not sure it's just I. I would like something halfway between oh free-for-all, and this is rather regimented right.

A

C

Preserved paragraph breaks, okay and add single comment. All right text posted feel free to bash.

D

Because I know you're well or whether I want you to or not so.

A

Oh okay, so you didn't make it. I made it a uh comment.

D

Well, true, I could that gives a suggestion. Let me just edit just to make it uh doable that way. I can just do an edit and put the yeah things around it. So.

C

That's trivial to do.

C

Okay, updated it is now in the.

D

Suggested text too so that you can do comments on it. Yeah, please refresh.

G

I see you generalized the uh association with evidence and I think it's.

G

J

So in the current definition of endorser and endorsement there isn't the of reference values and evidence requires reference values. So I don't understand.

D

Evidence certain evidence does not require reference values. The appraisal policy requires the reference values.

J

Evidence evidence expects that there are reference values in the verifier somehow.

D

No, the verifier uses that the evidence has no expectation. The evidence is just data right. It doesn't expect anything roles expect things messages, don't expect anything yeah.

G

E

Evidence could be two claims.

G

One of them is larger than the other, and the verified just compares is the first larger than the second, and, if that's yesterday,.

D

G

My point is, I think,.

I

J

Okay, you're missing you're missing my maybe I use the wrong word, but the the point is that evidence of is is it requires reference values endorsements as we define first, there are no reference values in an endorser, doesn't provide reference values and.

H

J

Don't contain reference values, so the.

H

J

That is sort of talking about endorsements and evidence. Independent of anything else, that's unstated is misleading.

D

Which one which end is you talking about.

J

The one that's I can't see because it's on the top of the page.

D

Oh I'm looking at my own screen, then.

J

I was talking about that indirect trust by having an endorser vouch for the attester's ability to securely generate evidence.

D

And so what's the problem with that one.

D

Because what happens here and again, let me tell you what I'm thinking and you tell me how you change the text. If you want to change um in order to do a praise right, you need to have a belief that the tester can securely generate the evidence, and then you have to weigh that evidence in your term again the values into the evidence itself against the reference values. This is only making a statement about the first part.

D

The second part is your tester may still be using a reference value provider to help it appraise the evidence, but first you have to figure out whether you believe that it could actually generate whether you should trust the values that are in the evidence themselves. Is that actually a correct statement of the current state and then you can decide what that current state matches. What's policy.

J

So I agree with what you said and I think the word say that and it may have been, that it was because it was scrolled off the screen. I was.

D

My text here is all in the context of whether you should believe the evidence is a statement of current state. You still have to put that, together with the use of reference values or whatever, to compare it against. You know a desired stage, or you know policy or whatever you want to call it. uh But all of this text is about the former part, which is um before you can even do that. You got to believe that that's actually correct statements of the current state and there's.

J

Three paragraphs.

D

Here in that category, so yeah.

J

I agree with that.

G

So I added a comment to your text here that is visible after f5, okay, cool.

D

Yeah ned, if you see something you want changed, feel free. I just want to explain what I was thinking when I typed it so.

J

Yeah, I'm trying to find it on my own here.

J

D

um I can't tell which line you were commenting on hank.

G

It is very, very updated. This is the quote: it's uh very uh yeah.

L

D

Trying to find that.

G

It's like seven, nine.

D

Five: seven: five, seven, ninety five yeah! Yes, you can we'll. Let you add a comment in the comment. I don't.

C

D

um Yeah yeah, I understand that's the line, but not now I got what was your head come? I think.

G

So my comment is that the the verifier has this key material thanks to an endorser, because putting that key material there is the endorsement. It's not an endorsement.

H

G

It's the conceptual endorsement so to speak, and that is done by the endorser, so the verifier owner probably is doing that and therefore they endorse a role in the verifier or collapse. In this case,.

D

I disagree with, I think, every sentence that you said did you finish reading all of it? First of all, let me put it. Let me ask that as a career.

G

I read the second paragraph uh up for that: okay,.

D

Please read all of it because um in case where you have the tester's key material directly in there, I claim that that is the case where there is no endorser and the the difference has to do with how many, how many chains of you know one key signing. Another key are there in the uh chain that you're using to verify the evidence, and in this case there was one less layer in that level of uh you know, certificate chains and that extra layer is the endorser that is not present.

D

In such case now there may be an endorser, and I talk about this in um uh line um 799, okay, and I want to distinguish it from that case because there actually is an endorser often, but that endorser is used at the time that you're putting the device's key or the tester's key into the trust anchor store.

D

So the endorser may be the manufacturer, you go to the manufacturer, you say: hey: did you create this device? Yes, okay, and this is the device that I bought. You created it, you vouched for it or whatever, and so I know that this is a you know intel inside and it's the one that I bought, and so I put that key into there and at that point evidence uh checking no longer checked for the endorser, which was still there right. I just did that one time and not the time that I'm doing the evidence.

G

You're elaborating on how the key material arrives at the test. Does that correct.

D

I don't know, I don't know it's a very fair power section.

G

Yeah, that's why I'm confused here so uh so. Why is this wrong? You're saying the same things with different words, I think I I don't see the discrepancy. That is my my personal problem. Only baby.

D

uh An endorsement is one key signing another key.

G

That is a technical implementation of endorsement and.

D

I'm saying okay, that does not happen here. That does not happen because the only key that's in there is the one that you would have signed and so there's no yeah.

A

Because I understood dave that when we wrote the text for endorsement, okay, that we specifically didn't want to assume that it always was in the form of a key signing another key. We wrote executive.

H

Pure statement.

A

That an endorser vouches for the integrity of the tester's various capabilities um and we wrote it that way, because we didn't want to say that that key signing another key was the only way of implementing this that the business such as you described. A direct trust, was also a form of endorsement where you basically configure the three devices you trust in.

A

K

A

D

I I find that confusing, but I would prefer language that that uh doesn't get into that point, but uh meaning if we can find a more generic general way to phrase things.

A

Well, but that's why I thought we said when we wrote the definition of endorsement. We wrote that it was a secure statement. We didn't write, it was a digital signature or uh uh something we wrote. There was a secure statement and, and that implied that included. For instance, I loaded a configuration file with the hash of the public key, and that was a secure statement as well right.

D

um So I did not get that letter fact. I I got the fact that it's a secure statement, yes and there's multiple ways to do a secure statement. Emphasis on the word secure and to me secure meant, signed.

A

Okay, so I thought.

D

That we said whether you're assigning a key you're signing something else. It's like a fingerprint of the key. So in that case it's not a key signing a key, it's adding a fingerprint, and so that was my interpretation.

A

Right but but what I'm trying to say is that that I thought that we wrote secure statement that it could also include a uh privileged file that only the you, the owner can edit right. That was the implication that that that that we're saying look, you can figure this somehow in a secure way.

J

A

Just random yobos: do it right! You.

J

Could you could assert that physical possession is a secure way.

A

Yeah, that was the point.

D

So I'm interpreting what you're saying, as the conceptual message used to configure the trust anchor store.

J

D

In the case of 7.99 right, the case of 799 is the case, we're using the particular message you're talking about to configure the trust anchor store.

K

Yeah, so it's online.

D

Store doesn't uh may or may not be viewed as that. That's what I would I would try to avoid the case saying your trust. Anchor store is an endorsement. That's what I'm trying to avoid saying so.

J

I can quote dave on lines 806 to 809, where he makes the case that physical possession is a form of trust. That's not signed.

A

The debate against dave, the debate is whether or not the physical possession is an endorsement. That, I think, is what it amounts to right dave. I think what we're arguing about is.

D

That's not what I'm arguing about that is interesting. That is an interesting philosophical debate, but that's not what I'm arguing about.

A

Well, I'm not sure philosophical. I understood that. That's why we wrote secure statement rather for endorsement, rather than being very specific about that endorsement is a is a signature. Vibration of some kind.

J

I mean we, we said we in our definition of endorsement. We said last time.

D

I will say that nothing.

J

D

The term endorsement but go ahead, endorser! Yes, endorsement- is not used in here, but.

J

But there we've paired them up right on and only an endorser can provide an endorsement apparently and.

A

J

The way we've defined endorsement is that it's it's it's making a statement about capabilities in the testing environment that we said we couldn't use. That word because of some reason.

E

J

Meant that that's what we really meant was the the testing environment, which one could argue the very first one in the platform is the root of trust, and we all understand that the root of trust- you know, can't speak for itself, and so the endorsement is the thing that speaks for the capabilities in the root of trust and that doesn't have the statements that can be made about those capabilities. Don't have to only involve what key you have.

J

It can talk about the environment that protects the key, even if there are no keys or in the environment that produces the keys, even if there are no keys.

D

Let's respond back.

A

It's not in response to that. So, let's, uh let's go around the room here um and uh I haven't heard from peter or uh at all- and I haven't heard from andrew uh at all and we haven't heard from lawrence lately.

A

Maybe you guys can can can inject something that solves this problem as to.

A

Whether an endorsement is digital signature only or not,.

E

I tend to think of it as more than that, but I I I've been very hesitant to wade into this.

B

I mean I proposed at times eliminating the the use of endorsement, because I think it's got preconceptions of what it is and.

G

But that is not the topic that it's a problem that is the endorser involved here or is it not because it says rather.

D

Than the endorsers, that is pretty much, you have to say, be more specific as to what the question is, so is it involved in what.

G

So is, is the uh in the uh having the verifier the testers key. I think it's boiling to back to that again. uh Dave says, as an endorsement is only a key signed by a key know: the value that doors is not involved here and when you say well, it's not only that it can be any kind of endorsement format.

G

The conceptual message is only our standard way to do it, it'll be defined here you can have them collapsed and they don't have to communicate at all, and then the doors are just uh defines the key with the the measure into the uh uh key store at the trust or whatever you want, and therefore it is not necessarily an endorsement used here when the roles collapse and they collapse. I think and and therefore it's still there and dave says no. That is not the case, because an endorsement outputs only key signed by keys.

G

I think that's the point right.

D

Well, I'm all in favor of finding a wording that doesn't uh have to answer this question. So I don't know if everything's spoken up yet and michael you wanted to ask andrew if he had any comments.

B

A

Andrew now, muted yeah muted, so he obviously was unmuted before.

D

um uh I'm fine with changing wording to to to not have to get into that question. I think it's far more interesting to get the main point that I'm trying to make across than it is to try to uh uh try to argue about whether an endorser exists or not or whatever.

A

Let me so let me dave try to to to make sure that I restate what you just said. So we all agree that there are many ways in which, in a test, a verifier can get trust in an attester and you're.

D

Trying to find.

A

And we know that there's two ways: one is that I copy and paste the a tester's key if it has one into the verifier, and another way is that I receive that key through a signed artifact from something like a manufacturer and I'm specifically saying.

D

A

Here, rather than roles, okay.

D

Sort of, although the way that I would break it down, it's almost that you you're not you're, not incorrect. um I would say you're either doing it at the time of uh evidence, generation, you're, verifying that the key chains up to the endorser's key, which is sitting in your trust, anchor store right, that's or um you're, verifying that it changes up to the device's key.

D

The only difference between what I said and what you said is that you that an endorser may be involved a priori at the time of trust, restore configuration, that's the only difference. So that's the.

A

D

A

I would make, but otherwise yes, so yeah, so so- and it comes back to this and I'm still frustrated in general- that we don't have a way of of you're, essentially talking about a pki of order, zero, where you, where you just configure the tester hey, I have order greater than one where there's some some number of of of um of artifacts right. So right.

D

So we agree that those are.

A

The two, the two.

D

Kinds of spaces here and there's a middle ground with the hybrid which is you run your own endorser and then your own endorser only endorses the things that you would have put into there and that well.

A

So what you're doing is you're accomplishing the second process or this. The second paragraph, where I have some number of devices, decided that I need to indirect, add a layer of indirection right, internal layer of interaction right, but fundamentally it's no different than the other one. It's just that I decided to store it differently. Okay,.

D

A

Hang on lawrence hang on lawrence, so so, if, if we so, um if we all agree with those two to those mechanisms- and I think lawrence wants to disagree- but let me just finish: if we all agree with those things, then we're simply looking for an appropriate set of language which maybe involves the word endorsement or endorser to explain those processes now lawrence. You want to add a another case.

B

Well, the the the the key material, the way the key material gets set up. I mean it could be through kdf functions and on several.

A

Others totally, I totally agree with you. I didn't see how I moved the key from the attester to the uh verifier. They could, yes, all be derived from the same uh secret uh kd through a kdf. Yes, I totally agree with that. I'm not saying anything about how that happened.

B

So there's no I mean the number of levels of indirection could be two or seven or zero or right.

A

The point is that the point in the case where it's zero, then I I have no, I have I I have to configure every device. I have to bless each device separately.

D

A

It's non-zero, then there's some level of indirection. That lets me bless many devices at the same time, which.

B

May, or may not be signing.

A

May or may not be signing, so I think we all agree with that as well.

J

Did we agree with that yeah? So what's so? If I look at the text, that's on the screen, this is entirely describing direct case. Our conversation is about the indirect case for which there is no text to review.

D

Which indirect case you're referring to? Because uh it's before.

H

D

It's the text, that's above 794, is the direct.

J

Word so I read and.

B

I further I had further comments on.

A

J

A

What I would like to do is I would like to accept dave's text right now here: okay um and the other suggestions in this bat in this batch and such that we can now comment on them, because that's the only way to do it.

J

I I'm okay with accepting I'm okay with accepting dave's text.

J

The problem I have is that there's the first sentence, which is the paragraph introduction talks about the the indirect case and then it switches to the direct case and doesn't expound on what we mean by indirect.

J

But the last you know little bit of conversation amongst us was about what is what does the indirect case mean and there's we? We don't have any words to ground to ground us on.

E

J

It seems like we're just we need to write something down and then talk about it, and so I think that dave's point was no. I do talk about the indirect case. It's it's that first sentence, I'm just saying: maybe we need more than one sentence.

A

Is a desire for an xxx, add text here? Add more text about indirect case.

J

J

A

J

I added it at the end.

A

J

A

The end here, but but yes, it could.

J

A

J

Is where we transition to talking about the direct case, so it's between 793 and 794 is where it should go.

D

Yeah, I agree with ned. Oh.

A

Okay, yes, all right.

D

Yep yeah, you can insert a paragraph break there, for example, and add the comment in there or something.

G

And I think michael's comment on: uh you have to touch every device by yourself with a zero in direction, and then you can add levels and bless them more than one. At the same time is, I think, a very intuitive way to phrase this. Maybe we can have some introductory text somewhere in the scope that helps us to ease into this section here.

A

I always wish that was in like rc 5280 or something it's not really. um I mean that's where we start with pkis and stuff, and it's not there. Maybe it's somewhere else in some other document.

G

I have never read something that is very concise.

A

Yeah, well, I tried to start a thread on sag or somewhere else about uh how do you number the levels of a pki um and um because there's two possible numberings right, zero and one based um and right? So so when, when you do put that when you do configure that that end device uh key into your trust as a trust anchor, is that a pki of order, zero or one.

A

G

But the concept, I think it was promoting- actually not hope we don't go into the numbering thing, but direct versus indirect means has an implication on how granular trust relationships has to be established. I think that is fine to know, and I think that makes this whole uh other expositional text from dave uh it eases into that. I think.

A

J

A

What, if what if we, what if we actually went further and split this into two subsections, direct trust and verifier and indirect trust and verifier, or by verifier in or buy, I don't know which.

D

It would be the reason.

A

It would be if.

D

You choose to do that.

A

The reason is that I think that it may be valuable to have be able to refer to section numbers on this topic.

D

I don't know whether it's necessary, but that's why I don't have an opinion yet.

J

It's yeah, I mean this is starting to be a wall of text.

C

A

Let's just briefly talk at the top of the hour um whether this this the changes that wound up to this sentence are acceptable.

D

If you notice, uh that's that one sentence there, if this is the sentence that I posted pasted stuff before my intent, was that that sentence would be replaced. But uh that's up to you as to whether you agree with that or not.

A

Your text would replace the sentence completely.

D

That is one straw man. Yes, there might be more changes necessary, but that would be my original intent because there's a bunch of changes we all had with that uh with that yeah.

A

D

Problems we had laurence had some.

A

Problems I was convinced.

D

That I had problems.

A

Before I read it,.

D

A

Let's, let's leave that for the moment, and what about this this! This change here.

J

So I think we are favoring the term direct and indirect and we have no definition for implicit yeah. I agree.

D

There there there is a use of implicit in a completely different context, with a different meaning in the text that talks about how you can do transmission across a secure channel or rather than some evidence, and we use implicit there in a completely different meaning and so yeah.

A

So so this this this, this change also gets rid of this word, implicit because it doesn't appear so we change the the we had a suggestion before hank had a different suggestion. I deleted my suggest the previous suggestion that had such hardware.

D

um Here I'm curious because this is why I was busy: why did you lowercase roots of trust? Is that no longer a term that you want to use out from the document.

A

It's not, although.

G

That is the thing in.

A

G

Yeah, it's the.

H

G

Not in the terminology section defined everything that is defined, there is capitalized with us. So uh lawrence made a comment inside the terminology. This shouldn't and I have the story. It's right.

D

G

Do that we also do.

A

D

Same thing elsewhere in the doc, so that's fine yeah.

A

We also reason we we decline to define root of trust is because it has definitions elsewhere, and the other definitions are too complicated and involved for us to try to repeat get consensus on. I wasn't.

D

Asking that I know I just think so: that's why it's not an art.

A

That's why it's not a determined term that we have defined we've left it defined from elsewhere. Is the point the point so well.

D

A

Happy this sentence is a definition.

D

This sentence is a definition, that's saying in for the purpose. It applies for the purpose of this document when we use the term roots of trust. We mean components that have to be voted for via endorsements, because no evidence can be created about them. That's that's! That's an implicit.

A

Definition here.

D

Yeah yeah. Second, let me read it again:.

D

um Yeah, I'm fine with that wording. I read it three times and everything I could think of that. I would have a problem with this fixes so.

E

I mean it isn't saying that no evidence about them is a little broad. um You really mean that it can't be created during the attestation or it's part of the attestation right.

G

Yeah ever so, the the testing environment cannot create evidence about itself literally. This implies also, and therefore.

E

Just as it's impossible to create uh evidence.

D

I can give.

E

D

Give you an example, because no evidence is created about them. What about that.

J

Yeah, that's better!

J

I I can assert that a tpm that it's possible to create evidence about tpm.

A

J

You can make the change yeah.

A

Yeah, I'm just I just have to freeze.

J

A

The thing so because no evidence is created.

G

Could you, when you're editing editing here, could you exchange, create it uh and remove credit and generate it, because you generate evidence all the time and.

J

D

Another way to figure.

J

They collected, we use the term collected.

G

This is generated is the term we use um generated. It's fine.

D

I think that's.

E

Another one because.

D

This is late in the document we've already talked about the two types of environment. um I think what we mean here is that those components are never a target environment.

D

Yes, I'm not saying we had to have to phrase it that way, but that's how I'm interpreting what we're talking.

A

About you're right about that, it's never a target environment.

J

But but a a testing environment isn't necessarily a root of trust.

D

Correct because in a testing environment can be in a target environment and layered at the station right. So that's why I'm fine with this text here.

D

Yeah, I think it's fine, except.

D

J

Do we want to elevate it to in a definition since we're defining it here.

C

A

There's arguments about it in the in the group.

D

You need to fix either. uh Let's see. Oh, yes, such components just delete the word a this is not plural in the sentence.

A

As that's what you're saying yep.

D

Yeah! That's right because it's components, plural! Yes,.

B

Being endorsements are not.

A

Optional anymore, doesn't it.

D

No, I read it three times to make sure that it was not it's, because it does not assume that you actually have to vouch for it. That's why I'm saying this is nicely weasel-worded.

A

Oh evidence is generated, it does, it doesn't mean it's impossible. It just means that we don't.

D

It says that's the only way to vouch, for it is by doing that. It doesn't say that you have to vouch for it. It's actually nicely worded. So, okay, all right.

A

The people that have taken the power it tells you.

D

What what lawrence, what lawrence expects the people that care about it they said. Well, then, it's actually fine, and so it's actually very nicely worded because it does not break into jail. In that sense, right.

A

So lawrence and hank, it would be great if you could come up with some exception, acceptable merging of this. These sentence here on this recommendation, because you were a disagreement as to how it went. um Okay,.

A

And then the concept was that this sentence would go away, as would this change if, uh based upon what dave said, I think that everyone wants to think about that for a minute unless we're all conclusive already.

D

um I I'm afraid we're gonna forget it. If we don't do something now at least put it in as a comment, if not a suggested text to delete, you could also just do the plus minus and have it be empty.

A

Yes, okay! Yes, I can do that.

D

It's an alternate is an alternate comment. You can decide which one to commit then.

G

Are you trying to uh accept this complete review here.

D

No, he wants to leave it open, I think, and have people think about it right, yeah, okay, so that that's the other suggestion yeah yep.

A

And you can go here to you know, give your reaction, you know rocket means, we don't know what rocket means. There's no idea blow it up. You know all sorts of things. I.

D

Don't know some of those if you hover over them. It has a word, but.

A

Meanwhile, facebook still doesn't have a frown.

A

Okay, uh all right! Well, um we spent most of this today on this pull request. um um I I will close this one once I make sure that it works, because there was an accident.

G

It can be closed 149.. I think it was uh my mistake to even work on that. So we show lawrence. Would you just close your own um pr.

B

G

Thanks because it looked at me, I was like commenting on it again and I don't want to do this again.

A

Okay, so for nine, oh.

G

G

A

um If we've closed all of the, if we, if we close all of these, are we.

D

J

The definitions, the impact of changing the definitions on other other drafts that are interpreting them differently and what has been changed seems like something we haven't talked about.

D

That's the lower case roots of trust throughout the document.

A

Yes, fine, there's.

D

Still 14 open issues.

G

Exactly lawrence is correct. There are issues I commented on the sixth most recent, it's november, 2nd just listen to me.

G

That's october, 2nd was it no. It was remember a second here. It says, cut off.

A

All right, so that's the cut off date is yeah yeah. I I would like us. I would hope to be finished to the point of being able to start a working group last call before that date, um but that's up to working group.

G

Yeah, I will have to do a consistency pass at least one probably two. If I know for seeking for experience when we uh conclude this meeting uh and then uh I would uh hope that other people's also do, but I will certainly commit 100 to do one before we submit a document before I.

A

I don't think it's a problem if that consistency pass happens during a working group last call, um but um I think it is.

J

A

Well, that's great.

J

We made a decision to not disposition the rest of the open issues or.

G

No, I have to go through them one by one. Actually, we had like 15 over the hour. So sorry, but uh these are some of these- are actually reasonable and very very, very, have merit.

A

All right well, so then we need to to have a uh uh walk through them and ideally try to turn some of them into text.

A

B

D

um By the way, I just created a request, I just created a request that all it does is it liberties root of trust throughout the document, and so is trivial.

J

Okay, that's one issue.

D

All I did was a global search and replace.

D

um Oh apparently, it picked up the same one, uh my bad. That was not intentional. I must have started from that match. Let me just go: do a quick rebase.

C

C

And all right, then, please refresh.

D

Apparently it was based on top of the other change and that, on top of uh oh wait, um it may still not have picked it up correctly. All right, you can.

D

If I have to it, may take me another minute to fix that. I think yeah, I didn't it didn't do it right, so I I just need to delete those paragraphs here so hold on.

D

There should be no changes other than lowercasing of characters.

J

H

J

We should go through the issues list to see if any of them can be closed, based on their pull, requests that have addre that have been addressed. There may be a couple that fit in that category. Okay, there may be some that are not.

E

J

To address uh like the appendix a time, considerations regressions, maybe that's not critical,.

J

But what would we say is our process for know coming.

A

J

A

I would say that, first of all, if you believe it's closed by a pull request, then you should just answer that and if, if at least another person agrees, then then close the issue and that we should spend next week going through. All of that, I have another ietf meeting at 10 o'clock on friday. Now so can't do that um and um I don't know what other times are workable between hank and and uh pacific coast times um I'll. Let you guys decide.

D

You can refresh it's fixed now. I believe.

J

G

Hank is typically available until 9, amp, dt or pt, and then uh I can now allow for more. If I can have an.

A

Indication you want to do 9 a.m. Pdt.

H

On friday, I'm okay with that now friday, until friday at 9am, so it can start at.

G

A

You start at eight, um so that's 11! For me. um I.

D

Am not available at 8. 00 uh am pdt this friday. Normally I would be, but this friday I've already got something booked at that time. All.

A

Right so so we might go for the following friday, then, um as a additional day. Yeah, that's okay! Now.

D

The 16th is okay,.

A

Okay, so I'll, I will just drop the calendar entry onto that as well. um I think we'll need the two extra days for sure, okay, so that one's closed all right, so let's come back. Anything we can resolve on in between is great, uh obviously- um and I kind of wonder whether in some cases we should um take some of these issues to the mailing list, um simply to hear what people have to say.

A

Friday, the 16th back to editing, but what time it's going to be at at what would be 11 for me right, 8, 8 8 am pdt. You said.

D

Yes, it would be, 11 am for you.

A

Okay, that's out all right great! Well I'll talk to you guys next week, then, and if not in in and let's try and see what we can close up on the issues or the meantime.

D

Okay sounds great.

C

E

E