Thib

151 posts tagged with "Thib" (See all Author)

Better authentication, session management and permissions in Matrix

21.09.2023 14:00 — General Thib

Before we explain all about our new authentication system - don't panic! This change is part of Matrix 2.0, the next step for Matrix that will be introduced in a blog post later today.

🔗Nothing breaks

Matrix 2.0 is a reference to Matthew’s FOSDEM talk Matrix 2.0, itself a reference to the name he coined in the Matrix Summer Special 2022: evolutions of the spec that are making the Matrix experience so much better it almost feels like a different protocol. One of the major changes in this bundle is the introduction of OpenID Connect-native Matrix and its implementation matrix-authentication-service (MAS).

Don’t panic: your current setup is not at risk, you don’t urgently need to deploy the matrix-authentication-service to keep using Matrix.

Let’s walk together through what OIDC-native Matrix, what changes as compared to the Matrix we know and love, and what’s the impact for your deployment.

🔗From OIDC-ish to OIDC

One of the most popular Matrix server implementations, Synapse, supports authentication via OIDC. This is the only mainstream server implementation to support authentication via an OpenID Connect Provider. OpenID Connect, SAML, and more generally Single Sign On (SSO) are a requirement for most companies past a certain size, as well as a security best practice.

As it is today, Synapse is more OIDC-compatible than OIDC-native. This means that it’s possible to use SSO (typically OpenID Connect) to connect to Synapse, but that’s about where the OpenID Connect adventure stops. Once you’re authenticated, Synapse generates a Matrix Access Token for your client, but that is not an OAuth2/OIDC Access Token. From your client point of view, it is doing the “Matrix SSO dance” defined by the spec, but the fact that Synapse does a login via OIDC, SAML or CAS is irrelevant to the client. Whenever your client asks Synapse for anything that requires being logged in, it sends that Matrix Access Token in the Authorization header of the http requests.

While this approach has served Matrix well, the Matrix Access Token method is reimplementing some concepts of OpenID Connect without all the benefits of thousands of developers battle testing and fixing every edge case they meet.

MSC3861, which proposes the adoption of OIDC in Matrix, is all about embracing the best of OIDC to make Matrix even better. By adopting the standard OIDC flows we allow Matrix to stand on the shoulders of another battle-tested industry standard. Not only does it improve security overall, it also unlocks new use cases for Matrix.

🔗Unlocking use cases

The Matrix Access Tokens Matrix currently relies on are secure, but not very flexible. A unique token is granted to each and every client during user login. This Matrix Access Token is only revoked when the user manages their sessions and logs a device out. A Matrix Access Token gives full access to a Matrix Account. Session management based on Matrix Access Tokens is quite nuclear.

Using OpenID Connect for Matrix comes with three major areas of improvements:

  • Standard authentication process on every client, including on devices without a keyboard
  • Improved security with token rotation
  • Better granularity of permissions (which improves security too!)

🔗Authenticate like you want

When relying on OIDC, the client delegates authentication to the OpenID Provider. This means that the client redirects the user to a web page on the OpenID Provider, where the user needs to authenticate. This authentication can happen through a username and password, WebAuthn (passwordless login), additional steps if you want to add MFA… and much more.

Since the authentication happens on the OpenID Provider, the client doesn’t need to support any authentication method other than “redirect the user to the OpenID Provider, and handle the result”. Once the user has authenticated against the OpenID Provider, it is going to be redirected back to its client, to allow it to retrieve an Access Token on their behalf.

During the whole process, no password was given to the client, and the client doesn’t even know how the user authenticated. All that matters to the client is that it now has an Access Token it can use to perform the regular Matrix API calls.

From the user perspective, the authentication is a very familiar process very well integrated in their password manager regardless of the client they use. They can use any client that supports OIDC, without having to worry about whether it supports every particular step of their authentication process. It’s also worth noting that matrix-authentication-service has a compatibility layer to support the m.login.password flow. This means compatibility with older clients will not break!

From the administrator perspective, it’s possible to force authentication flows (e.g. MFA), making sure the login is secure and matches their organisation’s policies. It also makes it possible to have a central management of all users' devices, including the ability to enforce policies upon them as provided by the OpenID Provider. For example, it becomes possible to the re-authentication of the user if they leave a trusted network.

From the developer perspective, neither clients nor servers have to add support for every new authentication method, existing or to come.

Element has been spearheading the effort to implement MSC3861 and matrix-authentication-service, and Element X is the first client to support them. Here is how the authentication flow and account management look like as seen from Element X iOS.

🔗Token rotation

While current Matrix Access Token can expire thanks to MSC2918, this best practice didn’t make it to most of the clients of the ecosystem. If the access token of a user was leaked for one reason or another, they wouldn’t notice since this wouldn’t create a new session, and the leaked token would be valid forever. Please note that thanks to E2EE, attackers who gain access to a Matrix Access Token would not be able to read the encrypted content in encrypted rooms.

OpenID Connect relies on two tokens:

  • an Access Token, that is used to perform the authenticated API calls, and that expires regularly,
  • and a Refresh Token that is used in conjunction with the Access Token to rotate both.

This ensures that if the Access Token was leaked during an API call, its validity would be very limited in time. Leaking a Refresh Token will also have limited impact, since the server would notice that two different clients are trying to consume the same Refresh Token and would log the user out from devices using this token.

🔗Principle of least privilege

Matrix is not just about apps and servers. It’s also a vibrant ecosystem of widgets, bots and various CLI to toy with. One of the strengths of Matrix has been its hackability: log into a client, dump the matrix access token, and you can hand it to a CLI or a bot for it to do things on your behalf.

While this approach is particularly convenient for development, it comes with its own lot of issues. Most of the time you don’t want the CLI or bot to be able to do everything on your behalf.

Widgets are another case where OIDC can help. While widgets are not yet part of the spec, in practice they are already commonly used. In essence, a widget is a tiny app that has access to an (unspecified) widgets API that allows it to do a limited amount of things on behalf of the user, such as sending a message in the rooms it’s in. While this has served several clients well, this can be improved. Indeed, it’s up to the client to enforce the restrictions mandated by the widgets API. This means you can’t be assured that all the clients will consistently enforce them.

With OpenID Connect, widgets can eventually become tiny embedded Matrix clients on their own, and they can get their own Access Token with explicit and granular permissions of their own. Those restrictions will be enforced by the server, who will not even be aware it’s a widget calling the API, making the experience consistent across clients. This behaviour still needs to be added to the specification, but a more OIDC-native Matrix paves the way for it.

🔗Upgrading to use matrix-authentication-service

MSC3861 (and friends) define how the OIDC flows and concepts (e.g. token exchange, dynamic client registration) should be used in Matrix, as well as the requirements for things which need to be coordinated with the homeserver (e.g. session lifecycle management).

The goals of matrix-authentication-service (MAS) are two-fold: firstly, it provides an implementation of the MSCs to support the Matrix Specification Change (MSC) process; secondly, it provides capabilities to facilitate the transition of the Matrix ecosystem to OIDC.

Because MSC3861 is changing the technical mechanism by which your client gets access to your account it is not something that can be done by pushing a single, big button.Instead we need a transition period where both OIDC and the existing legacy Matrix authentication methods coexist so that different clients and homeservers can migrate at their own pace.

One way in which MAS facilitates this transition is by allowing deployments without an external Identity Provider (IdP) to use OIDC out-of-the-box. This means that MAS will support some of the current flows that Synapse supports, primarily in the form of password-based login. Please note those flows are not supported yet: MAS requires an external IdP for now.

Another way is that MAS allows for existing sessions and access tokens to be imported (just from Synapse so far) so that there is no need for clients to re-authenticate when a homeserver is migrated.

Because of these migration capabilities we expect MAS to be widely used, however, in keeping with the open philosophy of Matrix, we expect other implementations to exist either standalone or integrated with other OIDC providers like Keycloak or Dex.

MAS is not intended to be a fully fledged IdP solution: we suggest using a dedicated software for this if you want to do more than what is included.

Currently MAS is deployable as a sidecar service alongside the homeserver. While larger setups shouldn’t see this as a problem, it might be seen as cumbersome in smaller self-hosted setups. For the sake of convenience, the Synapse and MAS team are investigating how to bundle MAS along with Synapse natively using PyO3. We anticipate such a bundle to exist by the end of Q2 2024.

You can browse the matrix-authentication-service docs to get all the technical details and installation steps.

This Week in Matrix 2023-09-15

15.09.2023 00:00 — This Week in Matrix Thib

🔗Matrix Live

🔗Dept of Spec 📜

TravisR says

Here's your weekly spec update! The heart of Matrix is the specification - and this is modified by Matrix Spec Change (MSC) proposals. Learn more about how the process works at https://spec.matrix.org/proposals.

🔗MSC Status

New MSCs:

MSCs proposed for Final Comment Period:

MSCs in Final Comment Period:

  • No MSCs are in FCP.

Accepted MSCs:

  • No MSCs were accepted this week.

Closed MSCs:

  • No MSCs were closed/rejected this week.

🔗New MSCs in detail

In this new segment, we aim to give a bit more context as to why an MSC was opened, beyond what is available in the MSC's introduction.

MSC4056 stems from a conversation held back at IETF 117, where members of the Spec Core Team (SCT) were attempting to make RBAC work in Matrix. Thankfully, there was prior art in the form of MSC2812, but a problem with decentralization (and specifically state resolution) was discovered. Thoughts were had about how to fix it, and MSC4056 is the result of those thoughts. Implementation work is eventually planned for this MSC, but in the meantime it should see forwards movement with the SCT's involvement in the MIMI working group at the IETF.

If you have thoughts or suggestions about the very Discord-centric approach, please leave them on the MSC :)

Continue reading…

This Week in Matrix 2023-09-01

01.09.2023 00:00 — This Week in Matrix Thib

🔗Matrix Live

🔗Dept of Status of Matrix 🌡️

Thib says

We’d like to thank everyone for their patience as we continue to work toward restoring the Libera.Chat bridge, and apologize for the continued inconvenience. We’ve heard from many people and communities who are impacted, who have confirmed that operating this bridge is an important service and we remain committed to getting it back online.

It’s been a month since our last update and folks have been reaching out, so we wanted to take this opportunity to provide a brief update.

The bridge team at Element is still actively working on the issues that led to the bridge being disabled in the first place. You can see some of the work that’s in flight through GitHub PRs: #1757, #1766, #1764, #1734.

We’re also looking into a way to transition responsibility for the bridge from Element to being directly run by The Matrix.org Foundation over the coming months - more details as we have them.

Unfortunately, we do not yet have a clear timeline for bringing the bridge back online. We’ll continue providing regular updates and will share more information as soon as we can. Thank you again for your patience! Please do not hesitate to reach out at #libera-matrix:libera.chat if you have any questions or concerns.

Continue reading…

This Week in Matrix 2023-07-28

28.07.2023 18:00 — This Week in Matrix Thib

🔗Matrix Live

🔗Dept of Spec 📜

Andrew Morgan (anoa) says

Here's your weekly spec update! The heart of Matrix is the specification - and this is modified by Matrix Spec Change (MSC) proposals. Learn more about how the process works at https://spec.matrix.org/proposals.

🔗MSC Status

New MSCs:

MSCs in Final Comment Period:

Accepted MSCs:

  • No MSCs were accepted this week.

Closed MSCs:

  • No MSCs were closed/rejected this week.

🔗Spec Updates

We've been quite busy at IETF 117 this last week discussing MLS and MIMI in several contexts, meetings, and sessions. Overall things have moved pretty fast in the last week, but the short summary is we're working with MIMI to get (Linearized) Matrix used as the new-found "signalling layer". This layer delegates membership of the room to the crypto layer when the crypto layer (namely MLS) supports being used as such, and is responsible for enforcing all policies. Policies in the context of MIMI are things like join rules, history visibility, and power levels, but with an added twist: we're looking at supporting Role-Based Access Control (RBAC) in combination with power levels in MIMI, which should also bring RBAC to Matrix in the form of a currently-unwritten MSC.

All told, we've got several new documents to write and MSCs to draft, but we'll get there in time. The MIMI working group is expecting solutions in place by about September, so watch this space for more news as we progress. An architecture draft is also in progress on the MIMI side to further explain what all of these new layers mean. In the meantime, if you have questions then please visit the matrix-spec room on Matrix!

We're also looking for more Matrix 1.9 candidates. Currently we have just custom emoji and anything to do with MIMI on the agenda - if you'd like to add more, let us know in the Office of the Matrix Spec Core Team room on Matrix.

🔗Random MSC of the Week

The random MSC of the week is... MSC3062: Bot verification!

This MSC describes a method for verifying (cross-signing) the devices of a bot user, and how verification of that sort could be done. Obviously it wouldn't make much sense to verify emoji with a bot. Instead, this MSC suggests that the bot provide a URL to present to the user. If the URL appears trustworthy (those who would control this URL should also be in charge of this bot), then the user can choose to continue the verification.

The user's Matrix client would then make a request to the URL with details of the verification. If the server responds successfully, some cryptographic magic happens, and your client will consider the bot verified!

This is essentially tying a bot's verification with control of a domain's DNS, which I think is a smart way to do things. But you do need to watch out for those pesky UTF-8 control characters when asking the user to verify the URL!

Continue reading…

Postponing the Libera.Chat deportalling

28.07.2023 14:00 — Bridges Thib

We have recently announced that we will be honouring Libera Chat’s request to turn off portalled rooms on the Libera.Chat bridge maintained by the Matrix.org Foundation. The changes were originally scheduled to be effective on 31st July. In the meantime, we posted instructions for people to turn their portalled rooms into plumbed ones so the bridge keeps working for them.

Some stability issues on the bridge have prevented people from turning their portalled rooms into plumbed ones. We have been actively working on resolving those issues since the first reports and the situation is gradually improving. However, at this point, we do not believe the plumbed mode can be considered sufficiently stable yet.

Continue reading…

This Week in Matrix 2023-07-07

07.07.2023 19:00 — This Week in Matrix Thib

🔗Matrix Live

🔗Dept of Spec 📜

Andrew Morgan (anoa) announces

Here's your weekly spec update! The heart of Matrix is the specification - and this is modified by Matrix Spec Change (MSC) proposals. Learn more about how the process works at https://spec.matrix.org/proposals.

🔗MSC Status

New MSCs:

MSCs in Final Comment Period:

  • No MSCs are in FCP.

Accepted MSCs:

Closed MSCs:

🔗Spec Updates

Work to use Matrix as the standard for interoperable messaging at the IETF is continuing in full stride. At IETF 117 (July 22nd - 28th, 2023) we'll be talking about the precise requirements of an interoperable protocol, and encouraging Matrix be that protocol. Linearized Matrix is our proposal for the room model, with more updates expected in the coming days ahead of the submission deadline, meanwhile yours truly is working on using MSC1767 Extensible Events for a content format. Watch this space for updates leading up to IETF 117 🙂

We're also well on track to test interoperability of different Linearized Matrix implementations at the Hackathon - get in touch with us via the #sct-office:matrix.org if you're working on such an implementation so we can coordinate details. It's not too late to get started either; Linearized Matrix itself is relatively simple to implement compared to the full capability of Matrix, by design.

🔗Random MSC of the Week

The random MSC of the week is... MSC3903: X25519 Elliptic-curve Diffie-Hellman ephemeral for establishing secure channel between two Matrix clients!

This MSC provides a means of establishing a trusted, secure communications channel across a potentially untrusted network. Subsequent MSCs could then use this channel to transfer details such as login tokens or key backup credentials in the context of setting up a new Matrix device. MSC3906 is one proposal that takes advantage of this.

This is just one piece of work building on the tree of MSCs supporting the shift of authentication in Matrix from home-brewed to OIDC. See https://areweoidcyet.com/ for more details on that effort.

Continue reading…

Making Sure The Libera.Chat Bridge Keeps Working

07.07.2023 18:45 — Bridges Thib

Libera Chat recently announced their decision to opt-out of portalled rooms from the Libera.Chat bridge instance hosted by the Matrix.org Foundation (a decision we regret but respect). This means that for the bridge to keep working, all of your portalled rooms need to be turned into plumbed rooms before July 31st. All of this might be a bit obscure, so let’s walk together through these concepts and give you the tools to make sure the bridge keeps working for you.

Continue reading…

What happened with archive.matrix.org

04.07.2023 14:24 — General Matthew Hodgson

We launched the Matrix Public Archive publicly on June 2nd, 2023. We decided to take it down on Sunday, June 25th out of precaution after a member of OFTC staff warned us that the archive made the content of two OFTC IRC channels bridged to Matrix available on the Internet.

After investigating the issue, we determined that the Matrix Public Archive's behaviour was expected for these channels, given an IRC chanop had explicitly configured the Matrix side of the rooms to be world-readable.

Let's talk about how room visibility works in vanilla Matrix, how it works with bridges, and what are the next steps.

Continue reading…

This Week in Matrix 2023-06-30

30.06.2023 00:00 — This Week in Matrix Thib

🔗Matrix Live

🔗Dept of Status of Matrix 🌡️

Thib announces

You might have seen the news already, but the Matrix.org Foundation is pleased to welcome the first public sector organisation as part of its membership: Gematik joined us as a Silver member!

If you're an organisation of the public sector it might not be obvious why spending money on an open source product is important: you will probably be interested in why it makes sense for gematik to contribute, and how it benefits the public sector in general.

Whether you're an individual, a business, a non-profit, a public sector organisation, you can join the Matrix.org Foundation as a member to support us in our mission and help us steer Matrix in the right direction!

Continue reading…