Implementing End-To-End Encryption in Matrix Clients | Hacker News Reader

Implementing End-To-End Encryption in Matrix Clients

matrix.org

122 pointsSami_Lehtinen22 comments

warning: the e2e implementor’s guide hasn’t been updated with any of the incremental key backup or interactive key verification stuff which just landed in Riot 1.0, nor the cross-signing epic which is on the horizon.

The details for these for those interested can be found at:

https://github.com/matrix-org/matrix-doc/pull/1703

https://github.com/matrix-org/matrix-doc/issues/1267

https://github.com/matrix-org/matrix-doc/pull/1544

and

https://github.com/matrix-org/matrix-doc/pull/1756

It's a good sign (I mean this sincerely!) when the first comment is the project founder with a caveat, setting realistic expectations.

Keep up the good work! (…but do remember to get some bloody sleep every once in a while)

thanks :)

Is there any plan to expose the Megolm session rotation parameters in the room configuration (I only just found out about them by reading the spec)? As an aside, I just tested setting the rotation parameters in an empty room, with key rotation set to every 5 messages, and the session_id was the same for all of my messages -- is Megolm rotation implemented in Riot?

Hm, it should work, at least on matrix-js-sdk, based on a quick look at https://github.com/matrix-org/matrix-js-sdk/blob/907cf19f054.... However, you're right that the UI isn't exposed in Riot currently, and the code may have bitrotted; looks like there isn't an explicit test for it.

I figured it out -- the problem is that you cannot change the encryption parameters for a room, as any new m.room.encryption events are explicitly ignored[1]. I set up a new room and sent a custom state event and it worked.

There is obviously a reason why you can't disable encryption once enabled, but it does suck that fields within m.room.encryption which are entirely advisory (since as far as I know the current implementation doesn't block messages that use old sessions) can't be changed.

It's good to see that the session is being rolled somewhat frequently though. This does explain some of my questions about why the old-school key backups kept getting larger even though I haven't joined any new chats.

[1]: https://github.com/matrix-org/matrix-js-sdk/blob/03a54353be9...

I love Matrix and I am planning on running a home server of my own and integrate it with all the different chat services I use.

Last time I checked there was the Matrix reference implementation and Riot. Today, there's a bunch of gorgeous Matrix clients that I can see people actually use. Alternative clients like Fractal and Nheko can make or break the ecosystem and I'm glad to see continuous improvement in that area.

It's especially nice to see the cross-signing proposal approved, as trying out just a few Matrix clients over time has added 9 separate keys I need to approve to my account already. Stuff like this will make it easy to switch phones/desktops/apps and will be wonderful for public uptake.

I want to integate all chat services as well. Could you share how you plan to do it? Libpurple/pidgin perhaps?

Pidgin integration would be amazing. It's my favorite client and one of the most usable currently, but nothing uses it. Everything nowadays is custom web or electron apps. I wish someone made a signald plugin for Pidgin...

Quinn Norton had some pretty damning things to say about libpurple and its lack of security in 2014: https://medium.com/message/everything-is-broken-81e5f33a24e1

It's possible libpurple's security has improved since then, but I wouldn't count on it.

Thanks for the link. I actually checked the CVEs so far and read their 'security' page. I have only skimmed the article, however, if it's just reverse engineering the protocols, and you are still using somebody else's server, what's the biggest risk? Maybe a vuln in the client that won't be fixed as fast as the official client?

https://matrix.org/blog/2017/03/11/how-do-i-bridge-thee-let-... is a good overviewer. Only one bridge (matrix-bifrost) uses libpurple as a fallback; everthing else speaks the target protocol natively.

There's a whole bunch of bridges available for Matrix. I plan on trying the whatsapp and telegram bridges, and maybe discord as well.

There's a list here: https://matrix.org/docs/projects/bridges

They should focus on non-Electron client, and server in Go [1] rather than Python. Both are too slow and consume so much memory just for an IM.

[1] https://github.com/matrix-org/dendrite

I love Matrix and Riot is the perfect client! Personally, I found the current way to manage end-to-end encryption in a room a little bit annoying to configure (when using it in a team is even more painful). I am glad to hear that it is getting better.

I can't stop suggesting Matrix to people!

Is there a milestone after which Matrix will only recommend clients which have E2E encryption for all messages by default?

As it is you list stuff like Nheko which claims to E2E encrypt messages but not attachments.

sure. the idea is to provide a proxy such that non-e2e clients get encrypted - see https://github.com/vector-im/riot-web/issues/6779.

nheko lost its maintainer for a few months hence stuff like the missing e2e attachments, but they are trivial to add; hopefully someone will contribute them soon.

I'm new to using all the new chat programs and LOVE the idea of matrix, but was somewhat surprised to learn there isn't a bridge for Signal, since it's open source. Is there a reason and/or any possibility of it happening. The linux Signal desktop client is too resource intensive and it would be amazing to be able to use WeeChat.

There is a signal bridge [0], but it is a little hacky. I would imagine moxie won't condone an 'official' bridge to matrix, given his stance on federated chat. Also: any realistic (short-term) bridges will have to basically be a man-in-the-middle, breaking the end to end crypto. In the (very?) long term MLS [1] may allow for e2ee through a bridge, but that would probably mean that signal has to explicitly support it. Given the track-record of OWS I personally doubt that would happen, but I'm just a random guy from the internet, so what do I know. I do know that the matrix people are looking at MLS and are willing to implement it in matrix if it finalizes, though they'll currently focus on their own megolm implementation, which is very close to being perfect, from my perspective.

[0] https://github.com/matrix-hacks/matrix-puppet-signal [1] https://datatracker.ietf.org/wg/mls/about/

Maybe slightly out of context but: It should be highlighted in <h1> that Riot does not use end to end encryption by default. At least that's how I understand it.

i think the assumption here is that if you are reading an implementor’s guide to e2e encryption, you’ll be doing so because it’s not yet turned on by default (due in part to not all clients implementing it yet).

https://github.com/vector-im/riot-web/issues/6779 is the issue tracking turning it on by default.

apps like riot make it pretty clear when you are in an unencrypted convo (typically the composer says “Send a message (unencrypted)”.

Reply on news.ycombinator.com