13 KiB

Raw Blame History

End-to-end encryption related types

Matrix has support for end-to-end encrypted messaging, this module contains types related to end-to-end encryption, describes a bit how E2EE works in the matrix-sdk, and how to set your [Client] up to support E2EE.

Jump to the Client Setup section if you don't care how E2EE works under the hood.

End-to-end encryption

While all messages in Matrix land are transferred to the server in an encrypted manner, rooms can be marked as end-to-end encrypted. If a room is marked as end-to-end encrypted, using a m.room.encrypted state event, all messages that are sent to this room will be encrypted for the individual room members. This means that the server won't be able to read messages that get sent to such a room.

                              ┌──────────────┐
                              │  Homeserver  │
     ┌───────┐                │              │                ┌───────┐
     │ Alice │═══════════════►│  unencrypted │═══════════════►│  Bob  │
     └───────┘   encrypted    │              │   encrypted    └───────┘
                              └──────────────┘

                              ┌──────────────┐
                              │  Homeserver  │
     ┌───────┐                │              │                ┌───────┐
     │ Alice │≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡►│─────────────►│≡≡≡≡≡≡≡≡≡≡≡≡≡≡≡►│  Bob  │
     └───────┘   encrypted    │   encrypted  │   encrypted    └───────┘
                              └──────────────┘

Encrypting for each end

We already mentioned that a message in a end-to-end encrypted world needs to be encrypted for each individual member, though that isn't completely correct. A message needs to be encrypted for each individual end. An end in Matrix land is a client that communicates with the homeserver. The spec calls an end a Device, while other clients might call an end a Session.

The matrix-sdk represents an end as a [Device] object. Each individual message should be encrypted for each individual [Device] of each individual room member.

Since rooms might grow quite big, encrypting each message for every [Device] becomes quickly unsustainable. Because of that room keys have been introduced.

Room keys

Room keys remove the need to encrypt each message for each end. Instead a room key needs to be shared with each end, after that a message can be encrypted in a single, O(1), step.

A room key is backed by a Megolm session, which in turn consists two parts. The first part, the outbound group session is used for encryption, this one never leaves your device. The second part is the inbound group session, which is shared with each end.

            ┌────────────────────────┬───────────────────────┐
            │       Encryption       │      Decryption       │
            ├────────────────────────┼───────────────────────┤
            │ Outbound group session │ Inbound group session │
            └────────────────────────┴───────────────────────┘

Lifetime of a room key

Create a room key
Share the room key with each participant
Encrypt messages using the room key
If needed, rotate the room key and go back to 1

The m.room.encryption state event of the room decides how long a room key should be used. By default this is for 100 messages or for 1 week, whichever comes first.

Decrypting the room history

Since room keys get relatively often rotated, each room key will need to be stored, otherwise we won't be able to decrypt historical messages. The SDK stores all room keys locally in an encrypted manner.

Besides storing them as part of the SDK store, users can export room keys using the [Encryption::export_room_keys] method.

Verification

One important aspect of end-to-end encryption is to check that the end you are communicating with is indeed the person you expect. This checking is done in Matrix via interactive verification. While interactively verifying, we'll need to exchange some critical piece of information over another communication channel, over the phone, or in person are good candidates for such a channel.

Usually each end will need to verify every end it communicates with. An end is represented as a [Device] in the matrix-sdk. This gets rather complicated quickly as is shown bellow, with Alice and Bob each having two devices. Each arrow represents who needs to verify whom for the communication between Alice and Bob to be considered secure.


              ┌───────────────────────────────────────────┐
              ▼                                           │
        ┌───────────┐                                ┌────┴────┐
      ┌►│Alice Phone├───────────────────────────────►│Bob Phone│◄──┐
      │ └─────┬─────┘                                └─────┬───┘   │
      │       ▼                                            ▼       │
      │ ┌────────────┐                               ┌───────────┐ │
      └─┤Alice Laptop├──────────────────────────────►│Bob Desktop├─┘
        └────────────┘                               └─────┬─────┘
              ▲                                            │
              └────────────────────────────────────────────┘

To simplify things and lower the amount of devices a user needs to verify cross signing has been introduced. Cross signing adds a concept of a user identity which is represented in the matrix-sdk using the UserIdentity struct. This way Alice and Bob only need to verify their own devices and each others user identity for the communication to be considered secure.


           ┌─────────────────────────────────────────────────┐
           │   ┌─────────────────────────────────────────┐   │
           ▼   │                                         ▼   │
    ┌──────────┴─────────┐                   ┌───────────────┴──────┐
    │┌──────────────────┐│                   │  ┌────────────────┐  │
    ││Alice UserIdentity││                   │  │Bob UserIdentity│  │
    │└───┬─────────┬────┘│                   │  └─┬───────────┬──┘  │
    │    │         │     │                   │    │           │     │
    │    ▼         ▼     │                   │    ▼           ▼     │
    │┌───────┐ ┌────────┐│                   │┌───────┐  ┌─────────┐│
    ││ Alice │ │ Alice  ││                   ││  Bob  │  │   Bob   ││
    ││ Phone │ │ Laptop ││                   ││ Phone │  │ Desktop ││
    │└───────┘ └────────┘│                   │└───────┘  └─────────┘│
    └────────────────────┘                   └──────────────────────┘

More info about devices and identities can be found in the [identities] module.

To add interactive verification support to your client please see the [verification] module, also check out the documentation for the [Device::is_verified()] method, which explains in more detail what it means for a [Device] to be verified.

Client setup

The matrix-sdk aims to provide encryption support transparently. If encryption is enabled and correctly set up, events that need to be encrypted will be encrypted automatically. Events will also be decrypted automatically.

Please note that, unless a client is specifically set up to ignore unverified devices, verifying devices is not necessary for encryption to work.

Make sure the e2e-encryption feature is enabled.
To persist the encryption keys, you can use ClientBuilder::store_config or one of the other _store methods on ClientBuilder.

Restoring a client

Restoring a Client is relatively easy, still some things need to be kept in mind before doing so.

There are two ways one might wish to restore a [Client]:

Using an access token
Using the password

Initially, logging in creates a device ID and access token on the server, those two are directly connected to each other, more on this relationship can be found in the spec.

After we log in the client will upload the end-to-end encryption related device keys to the server. Those device keys cannot be replaced once they have been uploaded and tied to a device ID.

Using an access token

Log in with the password using [Client::login_username()].
Store the access token, preferably somewhere secure.
Use [Client::restore_session()] the next time the client starts.

Note that the access token is directly connected to a device ID that lives on a server. If you're skipping step one of this method, remember that you can't use an access token that already has some device keys tied to the device ID.

Using a password.

Log in using [Client::login_username()].
Store the device_id that was returned in the login response from the server.
Use [Client::login_username()] the next time the client starts, make sure to set device_id to the stored device_id from the previous step. This will replace the access token from the previous login call, but won't create a new device.

Note that the default store supports only a single device, logging in with a different device ID (either None or a device ID of another client) is not supported using the default store.

Common pitfalls

Failure	Cause	Fix
No messages get encrypted nor decrypted	The `e2e-encryption` feature is disabled	Enable the feature in your `Cargo.toml` file
Messages that were decryptable aren't after a restart	Storage isn't setup to be persistent	Ensure you've activated the persistent storage backend feature, e.g. `sqlite`
Messages are encrypted but can't be decrypted	The access token that the client is using is tied to another device	Clear storage to create a new device, read the Restoring a Client section
Messages don't get encrypted but get decrypted	The `m.room.encryption` event is missing	Make sure encryption is enabled for the room and the event isn't filtered out, otherwise it might be a deserialization bug

13 KiB Raw Blame History