XMPP Council - 2017-02-22

moparisthebest, didn't you plan an update on https://xmpp.org/extensions/xep-0368.html security considerations after last week's discussions with SamWhited?

Tobias, I updated it

so it's just the editor that hasn't processed it yet, ok

he wanted the description updated, and I changed something else too, I think it's good to go

I think he did? I'll check

yea that's the latest version

ah..ok

I thought I merged… refresh?

Hi~

yeah..seems it's the latest version

ah..seems it's about time

1) Roll call

here

Here.

Here too.

great

2) Minute taker

seems jc isn't here, so we need a volunteer

i can do it

great

3) Outstanding votes

https://trello.com/b/ww7zWMlI/xmpp-council-agenda there are 2 ProtoXEPs and 2 other voting issues on the trello, please see recent minute notes and reply on list to vote

I've a few there.

4) Clarify CSI and Carbons state after SM resumption https://github.com/xsf/xeps/pull/402

Sam added this earlier this month, Flow closed his PR and opened indepenent PRs instead. Please take a look and reply feedback in the PRs.

5) The Last Call for XEP-0333: Chat Markers ended, daniel is taking maintainership and will probably send in an update sometime so we can vote on advancing it

If daniel plans on making more changes, I will extend the LC by another week.

more changes? haven't seen changes since the LC was issued

s/more changes/changes/

ah..k

6) Vote on advancing XEP-0368: SRV records for XMPP over TLS, specifically Version 0.1.2, to Draft

I'll vote on list

+1

I still don't like the direct TLS definition text, but that's an editorial change that can be made during draft

so +1

re extending the last call for xep333. please do. i will make changes in the next wee

Dave Cridland, Link Mauve ?

Wasn’t there a pending PR to fix the direct TLS definition?

Apparently not.

Anyway, will vote on list.

I think '368 is ready to go Draft now.

that's probably supposed to be at least a +0

?

+1

great

FTR, I'm sure there's further editorial fixes, but the technical side seems solid.

7) Date of next

I'm probably traveling next Wednesday around this time

should we do tuesday?

same time?

I’m ok with Tuesday.

WFM

that could work better for me, yeah

does that work for you, Dave Cridland?

I think so.

great..that would be 2017-02-28 16:00:00 UTC then

8) AOB

Yes

yeah?

XEP-0233, 0280, 0334, 0352 had their LCs end today. One of them has a pending PR (0280), otherwise do we want to do anything about them?

probably see what the feedback was on them, if there was positive feedback indicating that they are fine and no changes are required, we can vote on them next week

if the feedback indicates changes are required, they should stay proposed until we receive an update on the XEP

334 and 352 are probably good to go

but we can vote on that next time

i think with these widely deployed xeps no feedback is good feedback

334 I'm a bit torn about, but I agree with 352

anyways, next week. In the mean time should I extend the LC or move them back to experimental?

i'd request to extend the LC of XEP-0233 and i'll bug someone who implemented it to respond with feedback on the list

FWIW, I don't think that's true, 'no feedback' is generally a sign of no interest in XEPs, which implies they've probably not seen sufficient review.

(Not saying that's the case with 334/352 necessarily, but as a general point)

Kev, i was specifically talking about those day. as they seem to be used and implemented

s/day/two

yup...and we should try bugging people having implemented those...even if they have little time they could simply reply with a mail answering the questions with a short Yes/No.

Tobias, that sounds sensible.

ok

any other AOB?

Sounds reasonable; in that case I'll extend the LCs again.

SamWhited, sounds sensible

doesn't look like it

thanks everyone

Thank you!

thank you daniel for writing up meeting minutes

note with 368 I made a somewhat major change with the last version

SamWhited, pointed out 3.1 had no proper SHOULD/MUST etc language

in my mind that was always a MUST (mixing priorities of records), but he and ralphm convinced me it should be a SHOULD

so maybe it was only a major change in my mind :)

will give the XEP a total reread this evening then :) luckily it's not that big

yea and it works the same both ways, the server op can't tell if you mixed priorities or some ports were blocked

so in the grand scheme of things, no big deal

So it might be that there exist circumstances to not mix priorities properly, and it'll still interop? Feels like a SHOULD is OK here.

moparisthebest, What were the cases ralphm and SamWhited were unable to mix priorities?

Dave Cridland, ralphm pointed to the python twisted lib that connected to SRV records and how it'd be hard to mix them, but easy to try xmpps-client first then xmpp-client second

which is fair, and like you said doesn't affect interop, and moreover is invisible to the server

moparisthebest: I also believe that'd be easier to implement.

right

Keeping track of 'this srv entry is with tls/with starttls' seems like a recipe for messing up

Zash, I found it easy enough, and RFC 6168 is the same, so it's not uncommon.

it was easy in conversations

you already have to keep an object with 'target' and 'port', throwing 'directTls' in there is trivial?

That's basically what I did in Metre.

It treats a _xmpps-server._tcp record as if it were a _xmpp-server._tcp record with an extra field of "tls=true".

I may have done somesuch for DANE

conversations code starts about here: https://github.com/siacs/Conversations/blob/master/src/main/java/eu/siacs/conversations/utils/DNSHelper.java#L184

and yea I think you'd do DANE the same way, store the key hash on the same object

I think, implementing dnssec+dane is fairly high on my priorities as a next patch for conversations

moparisthebest, See the minidns work that the XSF had s a GSoC project a couple of years back - should work fine. Smack uses it already.

moparisthebest, But just do DNSSEC+RFC 6125 names to beging with.

yea Dave Cridland conversations already uses minidns, so I think it's basically just writing a bit of code

so that brings up an interesting point, DANE can pin certs, public keys, and anywhere in the chain, but the RFC recommends implementations for a specific protocol pick one way and stick with it

like for SMTP they just use public key pinning on the end device and that's all that is supported

sorry had to run off, but is there a standard like that for DANE with xmpp?

or can/should we make one?

DANE-SRV is as close as we get. But I'm happy supporting all forms - I've seen use-cases for all of them.

I despise certificate pinning personally hehe

pinning the end devices public key is the only way to go in my opinion

Dave Cridland: I *can* mix priorities, I just don't want too. I'm going to try connecting and starting TLS and connecting and starting a plain stream at the same time probably; if TLS works, it wins, otherwise start using the plain stream that was already being negotiated.

attempting simultaneous connections sounds , not good for some reason

do you do that now with existing SRV records? attempt them all at once and see which one wins?

moparisthebest: I'm not doing it for each record, just for each type of record (xmpp vs. xmpps); weights and priorities are still used within those records.

However, yes, if you have a few servers with identical weights and priorities, I would like to connect to all of them and use whichever one wins.

(and then if that fails, fall back to the next set of weights/priorities)

what's the difference? the server op mixed priorities when they defined them

I mean none of this really matters, you can do it however you like and interop and everything will work perfectly

If the server op made plain a higher priority than TLS, I'm ignoring it. That's basically the only difference.

just seems strange, I assumed people that wouldn't mix them would just try xmpps- first, then xmpp-

That's what I'm doing; doing it at the same time is just an optimization (if we can't do xmpps, we just fallback to the connection that's already started)

yea but in doing both at the same time it's slower than just doing the first would be

if bandwidth is constrained or anything

maybe, but I really doubt even the slowest connections are *that* slow (unless you're operating of HFR or something)

And no-one would do that, right? :)

Heh, I'm aware that it exists, I'm just not writing a library that aims to target every single use case (although, this will probably be an option when you configure the session, so it will probably support not doing it this way too, haven't figured the API out yet though)

I was wondering about happy-eyeballs across multiple SRV records.

I think I've suggested before that happy-eyeballs across multiple same-priority records is sensible.

how many domains have enough servers where it'd make a difference?

that's a genuine question I really do not have any idea :)

I'm not sure; probably not many, but it's also not hard to do or anything, so I'm not sure that it matters. ¯\_(ツ)_/¯

it just kind of smells of premature optimization, multi-threaded code where connections are racing and one wins doesn't sound like the simplest thing to implement, especially if 99% of cases have 1 server anyway

Nah, I've already done it, it was trivial.

probably depends a lot on the language too :)

what are you using? go still?

Yah, Go (which indeed is the reason why it was trivial)

is that code open source anywhere? I'd like to have a look

I just need to add it to my XMPP library and also add the XMPP/XMPPS version (although that's not a race, that's waiting on XMPPS and just pre-negotiating XMPP in case XMPPS fails)

yea I was thinking about how to implement it in java, but I'm not sure I'd call it trivial

Nothing is trivial in Java… everything is harder than you expect it to be.

Or maybe that's just me.

I wouldn't go that far :) but yea