XSF Discussion - 2020-04-21

pep.> I find the current experimental XEP updating flow a bit weird. Yep, it is not ideal that some big changes to experimental XEPs get merged without asking for feedback first

why?

experimental XEPs shouldn’t be deployed either way :>

doesn't matter, gathering feedback prior merge is still sensible as it potentially improves the XEP while avoiding frequent namespace bumps

versions are cheap if nothing is deployed :)

no they are not

There is a problem to with inbox folder, there are not redirection to official XEPs...

Neustradamus, Thanks! Now make a PR to fix it.

I also question the sentiment that experimental XEPs shouldn't get deployed, at least for certain definitions of deployed. you obviously want to have them deployed in experimental setups

It is not up-to-date but I have listed here: https://github.com/xsf/xmpp.org/issues/421

Neustradamus, that’s an issue, not a PR.

jonas’: I have said, I have listed ;)

and I said we need a PR to fix it

not an issue

issues are worthless if there’s no one working on it.

flow: I wasn't talking about namespace bumps, even though I disagree about jonas’' “nothing gets deployed”, it's really all just a waste of time for editors not to leave a short period of time here for authors to hear/incorporate feedback :(

MattJ: are you interested in the 401 discussion?

https://xmpp.org/extensions/xep-0191.html#example-9 This error you should get when you send a stanza to a JID you yourself have blocked, it should have an error/@by attribute of the account, right?

pep., I know that you didn't talk about namespace bump, I just mentioned it as consequence of this. To be fair, even if editors would keep PRs open for longer period of time to gather feedback, your process does not make it easy to review the proposed changes

Zash, I think there is no reason it couldn't have the 'by' attribute. I haven't read the spec if it is required though ✏

flow: agreed about the reviews

pep., \o/ ;)

marc [11:21]: > MattJ: are you interested in the 401 discussion? For sure!

if that's a public discussion I'll definitely read it :)

can you use chat markers in MUC?

I tried and the clients I'm using don't seem to care about the markers

MattJ: didn't you get the invitation?

Maybe, MUC invitation?

I tried sending you one too

Not sure if yaxim supports them

Or maybe my firewall rules

Ge0rG, fix it

eta: I think it's done in some conditions. I merged MattJ 's PR re 0333 in MUC. waiting for iteam to pull the updates

MattJ: I tried to reinvite you :-/

pep., oh right, apparently you might need to include the 'id' field in the <displayed/> element

that's odd

oh wait, nvm, you always do that

pep., is there a link to that PR?

eta, https://xmpp.org/extensions/xep-0333.html#appendix-revs it's live now :)

https://github.com/xsf/xeps/pull/927 PR

pep., oh, you need unique and stable stanza IDs

Zash: fix what? MattJ's broken default firewall?

I'm pretty sure I reviewed the rules and highlighted that they break MUC invitation an eternity ago

TIL `make preview` in xsf/xeps

do we have unicode symbol for muc topic?

rion: you can sponsor one

👥🗩

Just read latest version of XEP-0333.

sponsor? no idea how.

rion: https://wiki.xmpp.org/web/Adopt-a-Character

I don't think that's what he asked

there should be a way to pay the unicode consortium to add *new* characters.

I'm sure there is

bribery?

lobbying?

Little bit of both

nah, they demand proof of importance or somesuch

Ge0rG, "My brand is very important!"

kinda-sorta

It seems all the mentions about message type are removed. But section 8.1 says that "Only Messages that can be displayed in a chat SHOULD be markable".

Does that mean that I should not set "type" attribute for a chat marker Message and if my client displays Normal message not in a chat but in a separate windows, I should no make Normal messages markable?

That's actually a very sad thing about proof of importance. The wikipedia is into removing the article in russian about fediverse due to the lack of importance

Yagiza, I'm not sure I understand. I made the last update to the XEP, I didn't remove anything, I only added.

MattJ, I know. It seems that part was removed before your last update.

Yagiza, I don't see any such changes

It looks like it has all been typo fixes and state changes since 2013

Yagiza, hmm. Are you maybe confusing it with XEP-0085: Chat State Notifications?

That one discusses types

Quick question to get an overview of what people here think: When writing https://xmpp.org/extensions/inbox/password-storage.html I specified that SCRAM should be used because RFC 6120 mandates it anyways. However, I tend to think it would be better to start phasing it out and just using PLAIN, so maybe it's best if I don't add requirements for SCRAM-SHA-256. What do people think?

I'm all for it but am not sure you'll easily get a consensus on that 😉

SamWhited, I think this should be an RFC

The main reasons I consider PLAIN to be a better option are that it makes the password hash more agile (it's easy to update it to use a new hash, or a higher workfactor on login), it makes it so that server policy can contain password requirements like a minimum length and have a blacklist of common passwords (eg. it can reject "passw0rd", etc.)

not a XEP

Can an Informational XEP even make such demands?

it’s not really specific to XMPP, is it?

jonas’: I disagree, those are too hard for us to update regularly.

Zash: it's just recommendations. I did feel that RFC 2119 language was important to distinguish the importance of various bits of the XEP, but at the end of the day it is informational

> start phasing [SCRAM] out and just using PLAIN Strongly disagree

One downside is that hash recalculation is expensive so more prone to DoS.

I'd rather get rid of PLAIN and use only SCRAM

Zash: why? As far as I can tell PLAIN is much better in all the ways that really matter (eg. in terms of common attacks that can be prevented). The property of not sending the actual password to the server is kind of nice, but also that isn't where passwords are normally attacked

Holger: that's a fair point, but then again web servers always seem to manage to do it okay

So strong is my disagreement that I can't put it into words

SamWhited, you mean we should switch to OAuth?

MattJ, yes, maybe

Sending passwords to the server requires storing them on the client though, and this is a very common stealing vector.

Ge0rG: also a good point, adding that to my pros/cons list

"The web does it" is not an argument that's going to work on me.

Not unless it's "bring back XHTML-IM"

OTOH, you need to store _something_ on the client, and that will be usable to login to xmpp. However, it won't be usable to login into your Gmail

but surely everyone uses separate credentials for all the things!!k

SCRAM lets you store some magic hash on the client and login with that.

It's not an argument for "we should do it because we need to do what the web does", it's a counterpoint that it doesn't seem to lead to DOS' in practice

Without doing the *expensive* PBKDF2 operation again

And without the server doing that either

Mmmhmmm! Magic hash!

Just like 5 hash operations and some XOR magic and you're golden

Even if you set the iteration count to a larger number than I'll bother typing out

So rehashing may be expensive, and that's sad, but it's not a security issue so I'm not sure how much I care (unless there were literally no other common security issues to worry about). Requiring plaintext on the client is one, so I guess the question is: is that more common than not being able to set a password policy on the server or passwords being broken because SCRAm provides no agility for hashes and work factors

My guess is that the latter are much more common. In particular, letting servers define a blacklist (no "passw0rd", "aaaa", a dump downloaded from a recent <common web service> breach, etc.) is probably the biggest single way people break into accounts

This is why we need SASL2, which was supposed to provide credential upgrades and the like.

Yah, if we could at least get that problem solved it would be good. Not being able to set server policy seems much more dangerous than anything else though, and I think that's just not solvable with SCRAM.

I suppose I could mandate that clients all have a policy for disallowing common passwords and what not ("mandate" meaning "if you're following this XEPs advice, do this please", of course)

But that feels wrong and less likely to be widely adopted.

It's not solvable with SCRAM doesn't mean SCRAM needs to go.

It's not like you send SCRAM hashes from the client to register or change password, all that's by sending the plain text already.

Ahh, that's a good point that I was stupidly forgetting.

Okay, so maybe it's just the agility thing and promoting SASL2 is the way to go

The pains with upgrading to SCRAM-SHA-256 is more implementation issue and that you need the plain text password again

Zash: did you just say PLAIN auth? :D

no

I mean Corporate Enterprise Mandatory Password Change Policy

Okay, thanks, that's what I needed to get out of this. I'm convinced for now; leaving it as is. Thanks for sanity checking me (and finding me insane)

I should re-read SASL2. Did an experimental implementation of it when it came out, but I don't recall the details.

I still need convincing that this is a thing concerning XMPP specifically

Does it provide a way to force reset passwords? That seems like the most important thing to me

jonas’: it's not concerning XMPP specifically

why is this a XEP then?

an XMPP Extension Proposal document

You need some signal to say "hey, password change time", if you receive the password encoded in SASL PLAIN doesn't matter

Without letting clients fall victim to MITM

jonas’: because the XSF is in a position to make recommendations for the public jabber network and it's much better if people can get them in one place instead of all over the place and specific bits of it may be specific and common problems when implementing things for the Jabber network, eg. how to handle PLAIN alongside SCRAM-SHA-1 which are both mandated by RFC 6120)

that's not true, PLAIN isn't mandated by 6120 IIRC, but it's common to implement in Jabber alongside SCRAM-SHA-1 for compatibility with the widest range of clients, so having a recommendation for how to do that is good.

Another example is the PBKDF2 iteration count. Right now most people use 4096 which is rather outdated (NIST and OWASP recommend 10000). If you look around on the web you'll find a mix of recommendations for various numbers, but which one applies to the Jabber network? Is it a low security environment or a high security environment? Should I use 4096, 10k, or 100k?

Zash, any clues of how to use Coporate Enterprise Mandatory Password Change sasl policy without sending PLAIN? or was PLAIN assumed

This document clarifies that, even though you're right, technically PBKDF2 iteration count isn't specific to jabber.

I know corporate setups can require password policies, but I'd prefer encouraging SCRAM also fwiw, even during account creation/password change (where it's not used atm)

Ideally I'd use client certs, but for now we all have to deal with passwords..

pep. you can't mandate it during password change or you lose the ability to be agile and then when your password hashing mechanism is broken all your users get their passwords stolen. This is *much* worse than the server getting your password once in a blue moon.

once and for all you mean

Not entirely sure what you mean by not being able to mandate it during password change

You can't mandate using SCRAM or something during password change, like you were suggesting, I mean.

Ah otherwise I can't rehash myself

Hash and work factor agility is the most important thing, in my mind. Right now we don't have that and it's a major problem.

However, if SASL2 provides a way to say "you have to reset your password" or "you need to rehash with this iteration count and HMAC hash" that solves the problem nicely in my mind.

But via requiring another channel I can clear the hash/password altogether and for a reset via that other channel

pep.: didn't understand that bit about another channel, sorry?

email etc.

If SASL2 lets you say "reset your password please" *after* you authed, then it's better than forcing PLAIN auth.

Zash: right, definitely after auth :)

But yeah I would prefer what Zash just said

Then you can use SCRAM mutual auth magic to be sure you're giving the new password to someone who knew your previous password

*once* knew, I think.

Anyways, got what I needed. Ping me via PM or email or something if you have more suggestions for the best practices doc. Thanks again!

SamWhited, questions on IBR2

pep. good timing, I was just about to /part. What do you need?

You specify a way to do the register procedure via <iq>, does that mean I can register an account after bind? Or is it specifically for components?

pep. yah, that all needs to be flushed out a bit more. The idea was to make it compatible with components, but sure, if you're an admin and want to register some accounts for somebody it could be used to make that possible after bind

We'd probably want to add a discovery mechanism for that though so that servers could decide whether to implement it or not. And it should probably be ad-hoc based instead, but I'm not sure if that should all go in this XEP or a separate one.

ok

Bye for real now; thanks again for the discussion on PLAIN v SCRAM!

Is there a way to maybe break the fork relation between linkmauve/memberbot and xsf/memberbot?

So I get suggested by default to PR against xsf/memberbot

And github is sloooooow

"We are investigating degradations to GitHub.com."

yeah

degradations lol

It's dead Jim

It's some evil M$ plot, I bet 😂

https://igniterealtime.org:443/httpfileupload/c463ba66-b75a-4fe0-b500-11202e11be40/dt180322.gif

Why GIF?

because efficient 8-bit palette encoding?

pep., you can ask github to change the primary repo

i've done that in the past

simply write them a mail

you can do that with a simple click in gitlab :/

Also I'm no owner of that repo so maybe they will just tell me off

MattJ, ^ (?)

I don’t seem to wield power on that repo either.

!XSF_Martin [18:51]: > Why GIF? The patent has expired. GIF is free now!

I've sent Github a request

Thanks! :)

These should be easy changes: https://github.com/xsf/memberbot/pull/2 https://github.com/xsf/memberbot/pull/3

That email came out longer than I planned

Awesome, thanks!&

I'm a little surprised that you don't see a difference between responses and actions, maybe I'm still missing something important. I think the two use-cases of memberbot and and mercurial (:D) bot are already an example of why these two are separate, aren't they? One of them you only want for the last message, the other you want for all (or more) messages. One of them you want with backward compatibility using only plaintext, the other one you want with unique ids so that multiple messages are flexibly possible.

You could merge both and then define that responses that define a body may only be available on the most recent message.

I'd imagine you'd end up with a set of rules that depend on the body to be there or not, essentially splitting the merged element based on that. But that might just be wrong?

I don't think I necessarily agree that actions can't have a body

"!merge 245" seems like a good example

"merge !245" :p

!merge !245

!merge!!245!!!

hmm I wasn't even picturing "normal" bot commands using this tbh

In general I'd also prefer bots to use ad-hoc I think

in muc?

dunno, it just rejoins my usual rants about semantics in body :P

Even though with a bot you can generally make that more obvious and slightly restricted

When one of the people proposing forms or ad-hoc commands goes and implements either in one of the mobile clients in a nice way, I'll listen

Shall I dig up my Buttons implementation?

Had a working prototype hacked up in JabberCat back when I wrote the protoxep

> "!merge 245" seems like a good example It is a good example, there is no reason to forbid this. But how to handle this cleanly. Have an "unique" flag on each <response>, if set allow selecting the response even if it's not the most recent message, if not set limit to the most recent message?

Or actually, my argument was that a bot can totally still accept this, just that there is no button for it

The button would say "merge" and send an <action-selected> with the corresponding id, the bot could obviously still accept a manually written "!merge 45"

Sure. But I see no reason to hide the button either

And as noted there is UX weirdness if buttons just disappear

No it wouldn't be hidden, just the button would not generate the "!merge 45" output

Why not?

No hard reason. It probably doesn't have to, because the mercurial bot will send a "45 merged by foo" notification as soon as you click "merge"

MattJ, question about room activity indicators: Is it supposed to only work while the client is online?

Yes

Which is why it uses presence

Mhm

My idea was that the bot knows best when to print stuff and when not to. Thus the action buttons would not print anything, because the bot can just do so with much finer control.

That's kinda my reasoning for giving it the decision about whether there should be a <body> in the user response

MattJ, I'm not really thinking about mobile right now :)

definitely not my main use-case

pep.: non-mobile is not this protocol's main use case... console users are used to typing ;)

But accepting and having this buttons xep being used does bring "concerns"

As dwd said

"concerns"

> So I can see that these are useful, but I'm worried they'll end up baking in a data-as-text-body construction which is less than ideal, and tricky to move away from.

So much text

Sorry, but that's what we have today

Yeah and I don't like that either :/

I know you would like to replace all text messaging with semantic XML

But users don't care, sorry

Users don't have to know, as usual

Take the following scenario: a bot for voting. The votes of some members count double, for whatever reason. Now the bot could decide to only print votes of these double-weighted voters, and hide all single-weighted votes to prevent spam. While a little theoretical, that would only possible if the body isn't fixed in the <response>.

Sure, so it sets unique ids, no body, and echoes whatever it wants whenever it wants

(fwiw, dave's quote here is a borderline argument in favour of XHTML-IM. I note :p)

The bot would decide what goes into the body. So why not just have the bot do the printing itself?

I think in most cases it will flow nicer if the response comes from the user rather than the bot. But I agree it depends on the use case

Feels borderline security-ish too

Seeing Ralph write "+1" in a meeting is more reliable than seeing a not write "Ralph said +1"

Having the bot specify text that appears to be from the user?

Not == bot, my phone refuses to recognise the word

Oh, I guess both directions can appear borderline security-ish :D

I'm sure ralph would be able to correct that itself if it happened

Clicking on "yes" and having "no" appear in the chat because the bot defined it that way is also weird

True, that is definitely a consideration

Agreed, no UI should send a body the user isn't aware of

okay. That's an argument for a) removing labels from (what is now) <response> b) not adding a body to (what is now) <action>

But you have to trust the bot either way

Otherwise it will just echo the "wrong" thing

yes, you'd have to clarify that then followed by kicking the bot from the MUC :P

Could recommend showing it in UI like "Merge (sends "merge #123")" or somesuch.

or something something descriptions

I'd like to avoid mandating such kinds of UI details, though if you want something like that super hard I won't veto

I think the cleanest thing we can do is: - Have buttons for plaintext responses display exactly what they will send - Have buttons that trigger actions display what the action does

Yeah, protocol specifications should stay away from UI design, tho offering suggestions is fine.

There is plenty of precedent for UI requirements in security considerations ("This piece of information MUST be clearly visible to the user")

I'm going to default to "that's silly"

Syndace, makes sense I think

But we still need to figure out visibility I think

e.g. now if I wanted to make an in-MUC voting bot, and I wanted to make quick responses for +1/-1, I can't detect what they are in response to

Those kinds of buttons would need to be hidden if the bot sent a new voting item

I imagined that the (action) buttons would be displayed right with the message they belong to

and then people with clients that don't do actions can't vote?

or they have to manually type +1 now, and we aren't able to offer them quick responses on mobile

they could only vote via +1/-1 for the newest vote then, yes

Unless! <delay/> tag something something

Let's keep the awkward <delay/> tag out of it :)

oh you're talking about clients that support only the quick response subset of the xep?

i.e. not the actions thing

No, I'm working on the assumption that clients would implement the whole thing

But probably not explaining myself well, tired

Okay. Then if actions are used for +1/-1, people with supporting clients could also vote for older items. Which the bot would and should probably prevent. People with clients that don't support would have to manually type +1/-1 and have it count only to the most recent vote item

Yeah, you're right

That's usually not how voting works tho

There is a third option which is not covered currently: a quick response that is not limited to the most recent messrage, but maybe until it is explicitly ended. MattJ> !vote votebot> vote started. *+1 and -1 quick responses appear for everybody* someone> +1 anotherone> -1 MattJ> !done votebot> vote done. *quick responses disappear for everybody*

I worry about adding work for clients

But indeed, that is a possible solution

yeah, that would probably be complicated.

`<in-reply to="xmpp:id"/>`

(removing the merge button after merging a PR would be great)

Going to sleep on this - good night!

I should too, good night :)