XMPP Council - 2017-09-28

dwd, so are you saying we should drop ALPN from the spec or just change it into a MAY ?

because I'm opposed to the first and fine with the second :)

What does ALPN provide that is required for interop?

when you want to easily multiplex a port

a big corporation has IPs to spare, maybe, and doesn't need it, for personal servers you do

Yes, sure. ALPN is useful for multiplexing ports. Why is it required for interop with direct TLS?

Or to put it another way, why is it specifically useful for direct TLS?

if you remove it from the spec then as a server operator I cannot multiplex easily

What does multiplexing have to do with direct TLS?

because alpn is built into direct TLS and specifically for multiplexing?

Well, put it this way - right now, nobody much seems to be implementing it, and yet it works fine.

Lots of TLS options and extensions might be useful in XMPP. But we don't specify many of them - so what makes ALPN something we want to call people out for non-conformance on?

What is the purpose of direct tls?

conversations implements it which is why it works so well

Is it to get through your company firewall?

I don't think you call them non-conformant, but as a server operator I need to be able to run nginx and prosody on the same port, so I need alpn

Zash, That's an interesting question, yes.

the odd client that doesn't send alpn is ok because of SRV records they just fall back

To me it looks way too much like a premature squeezing of all services into port 443 because ... reasons.

everyone else is doing it :)

I agree it's crappy we have to run everything over TLS on 443 to connect

but also not supporting that won't change anything, people just won't use XMPP

I really doubt that. But even if not, it's not clear that ALPN belngs in *this* spec.

Nice things be unattainable.

These days if you want just about anything to work, you have to (sometimes) run it on 443. Same for TURN servers etc.

So therefore we must embrace it and put everything on 443.

dwd, at least for me the spec is 100% useless without alpn

I'm very much for keeping alpn as may

I have to multiplex on 443, and so do lots of other servers

Otherwise you can only differentiate on domain basis

moparisthebest: which leads back to the question of its purpose

And that's not good enough in some scenarios

it has multiple purposes, my personal purpose is to evade shitty firewalls I encounter

now someone else might only want it for 0-RTT or something, in which case maybe alpn isn't needed

What is the problem statement? Does this require ALPN to solve?

but, if they need multiplexing, it still is

daniel, There's literally no difference between MAY and removing it entirely.

daniel, I mean, there's absolutely nothing in XMPP which rules ou using any TLS extension at all.

If you do direct TLS you need SNI, since you can't use the initial stream 'to' attr anymore to select certificates

dwd, ok optional then

Assuming name based certificate selection is a thing you do

should

The current wording (SHOULD) seems perfectly reasonable to me. If the client sets it and it's not used by the server, no big deal. If the client doesn't set it and the server needs it though, that ends poorly. Just tell clients to set it and be done with it.

daniel, Right, "OPTIONAL" and "MAY" are equivalent terms.

SamWhited, If ALPN's a "SHOULD", I don't think we're in a position to advance it. That's fine if that's what we want.

may would have the benefit that if servers and client end up implementing it the at least use the same value. which would allow us to raise it to a should again if we see that people are using it

I tend to prefer SHOULD, a server operator can feel more comfortable requiring it then

MAY over not specifing it at all i mean

MAY is the default? or like, inherited from all the TLS RFCs?

daniel, I haven't checked but it's faster to just ask, does conversations.im require ALPN ?

moparisthebest, no

dwd: That seems fine to me. Is it just that we need another open source implementation? I think I have one I just never published that should be "production ready", I can probably push it this weekend if that's all that's needed.

SamWhited, I think server-side support is also lacking

Server side support doesn't really matter, does it? I mean, this is just saying "clients SHOULD set it because it doesn't hurt them at all and then it provides information that servers can use if they want to"

how do you define 'implementation' ? this spec has quite a few parts to implement

i think one of my customers requires it. because we are running an http server on the same ip with the same host name

but conversations.im has the http server on a different ip

you've got client support, then server c2s support, then server s2s as client and server s2s as server support

but that's not an option everywhere. ipv4 is getting rare

so I count 4, "parts"

with alpn, I can host an xmpp server with full c2s and s2s support and http or whatever on a single port, with ipv4 getting harder to find that seems useful

I don't see the point of hosting s2s on the same port. but c2s + http is a valid argument

so I could see a future host that sells you a dedicated server, but a shared IP and shared port 443, and you get to add your own alpn/sni rules for it

sounds awful but without ipv6 also seems inevitable

I see a future where we get IPv6

s/ipv6/ip/; s/ipv4/legacy ip/

Zash, thank you

SamWhited, Right now it's saying all implementations, so clients, servers, C2S and S2S.

jonasw, I *hope* for that future but, well, I also hope for a future where everything doesn't need shoved through TLS on 443

I'm in the continue to hope, but also do what you can now camp :)

dwd, I don't see a reason to have different rules for s2s and c2s, just overcomplicates things

right now you can send it or not, I say leave it that way

moparisthebest: I'm not sure that writing a spec for shoving everything on 443 is the best way to make that come true

you don't need to handle alpn on an *incoming* s2s or c2s connection. but that's obvious I guess

daniel, don’t you? what if s2s and c2s live on the same port?

again Zash you really are much more of an optimist than me if you think not supporting that will push people to open other ports rather than to abandon xmpp :)

jonasw, fair enough

moparisthebest, I don’t think that IPv4 depletion is an argument here

moparisthebest: optimist? "being draged kicking and screaming into a future I don't want" is what I was going for

some corporate firewalls may very well be

but I’m not confident that ALPN will help there in the long run

user of public wifi in coffee shop: "hi owner-of-coffee-shop, my xmpp client can't connect from here, can you unblock port 5222?" owner of coffee shop: "what's xmpp and what's a port?"

Zash, you just put how I feel in words better than I could have done it :-)

moparisthebest: owner of coffee shop: "stop that dirty talk! I call the police"

moparisthebest, and that assumes the owner is even around…

moparisthebest, as I said, c2s makes sense, s2s doesn’t too much

and this doesn’t have anything to do with IPv4 depletion.

right, more likely you are talking to the 16 year old behind the counter who's never met the owner

jonasw, it's the same code and everything though, it seems more complicated to only support it for c2s and not s2s meh

what if you are running your xmpp server on your laptop from a coffee shop???? >:)

moparisthebest, the difference is that you don’t need to require support for ALPN from servers in that case. Since they will only ever be listening on a single port which would be a candidate for ALPN (the c2s-port), they don’t need to distinguish protocols.

maybe your server doesn't, but maybe someone else's server does, what's the argument against it other than "I don't think it needs it" ?

moparisthebest, is there an S2S-ALPN-capable XMPP server implementation?

which of the 3 server side parts? haha

sslh covers 2 of 3 parts

dwd's metre could have alpn added rather easily I think

moparisthebest, sslh isn’t an XMPP server, right?

metre isn't either I guess

ssl/ssh multiplexer thingymajibber

the combination implements it though, I don't see a difference really

dwd: that still seems perfectly fine to me. I don't see a problem with just setting it even if it's not used, and if there really is a serious problem (option to try and avoid firewalls that inspect that field or something) then that's why it's a SHOULD, they can ignore it if they *really* need too.

that's basically it, it's super handy to have unless you are trying to evade firewalls, and then you can just not send it, and still be compliant with spec

evade packet inspecting firewalls that don't seem to exactly exist yet, but no doubt will

Does http have an official alpn value?

yes

http2 iirc

h2

Does browsers set it?

or http/1.1

https://www.iana.org/assignments/tls-extensiontype-values/tls-extensiontype-values.xhtml#alpn-protocol-ids

yea h2, and yes browsers set it

not sure if they set http/1.1 at all