Skip to content

MSC2366: Key verification flow additions: m.key.verification.ready and m.key.verification.done #2366

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 9 commits into from
Jan 31, 2021
Merged
80 changes: 80 additions & 0 deletions proposals/2366-key-verification-accept.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,80 @@
# Key verification flow additions: `m.key.verification.ready` and `m.key.verification.done`

The current key verification framework is asymmetrical in that the user who
requests the verification is unable to select the key verification method.
This makes it harder for more experienced users who wish to guide less
experienced users through the verification process, especially if they are not
verifying in-person, but are using a trusted but remote channel of verification
(such as telephone or video conference).

As an example, let us say that Alice is an experienced Matrix user and is
introducing Bob to the wonders of federated communications. Alice wants to
verify keys with Bob, so she clicks on the "Verify" button in her client on
Bob's profile (which sends a `m.key.verification.request` message to Bob).
Bob's device receives the verification request and prompts Bob to accept the
verification request. At this point, under the current framework, Bob is
responsible for choosing the verification method to use. However, with this
proposal, Bob would be able to just accept the verification request without
choosing a method, and allow Alice to choose the verification method.

In addition, the current key verification framework does not have a method for
clients to signal to the other side that a key verification was successful.
Some clients may wish to wait until the other side has either confirmed a
successful verification or indicated an error before displaying the result of
the verification, in order to give the two users a consistent view of the
verification as a whole.

## Proposal

Two new event types are added to the [key verification
framework](https://matrix.org/docs/spec/client_server/r0.6.1#key-verification-framework)
when verifying in to-device messages. The new event
types are already described in [MSC2241 (Key verification in
DMs)](https://github.com/matrix-org/matrix-doc/pull/2241). This proposal adds
them to verifications in to-device messages.

The first event type is `m.key.verification.ready`, which must be sent by the
target of the `m.key.verification.request` message, upon receipt of the
`m.key.verification.request` event. It has the fields:

- `from_device`: the ID of the device that sent the `m.key.verification.ready`
message
- `methods`: an array of verification methods that the device supports

It also has the usual `transaction_id` or `m.relates_to` fields for key
verification events, depending on whether it is sent as a to-device event
or an in-room event.

After the `m.key.verification.ready` event is sent, either party can send an
`m.key.verification.start` event to begin the verification. If both parties
send an `m.key.verification.start` event, and they both specify the same
verification method, then the event sent by the user whose user ID is the
lexicographically smallest is used, and the other `m.key.verification.start` event is ignored.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's not immediately obvious that this algorithm solves the glare proble. I think it is robust, but when it lands in the specit would be good to clairify exactly what "used" means here, and to add some walk throughs and/or sequence diagrams.

In the case of a single user verifying two of their devices, the device ID is
compared instead. If both parties send an `m.key.verification.start` event,
but they specify different verification methods, the verification should be
cancelled with a `code` of `m.unexpected_message`.

With to-device messages, previously the sender of the
`m.key.verification.request` message would send an `m.key.verification.cancel`
message to the recipient's other devices when it received an
`m.key.verification.start` event. With this new event, the sender of the
`m.key.verification.request` message should send an `m.key.verification.cancel`
message when it receives an `m.key.verification.ready` or
`m.key.verification.start` message, whichever comes first.

The `m.key.verification.ready` event is required for verifications in both DMs
and in to-device messages to accept verifications requested using an
`m.key.verification.request` event.

The second event type is `m.key.verification.done`, which has no fields other
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

How long should the other client wait until it receives a done, as old clients won't send that? The normal timeout of 10min sounds pretty long.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have the same question. Riot doesn't send this, when you are not verifying in DMs it seems. So we kinda need a behaviour for the transition phase, when a client follows the current spec.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It definitely should be sent in DMs.
image

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

when you are not verifying in DMs

It seems to be only sent, when verifying via DMs. So it seems like it is not sent in to_device messaging. This proposal suggests, that clients would be sending them already, but they don't. Unless I am missing something.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, sorry. I read your sentence several times, and managed to miss the "not" every time. :-/

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I just tried in Element Web in a non-DM self-verification, and it does send an m.key.verification.ready in response to m.key.verification.request. Trying to directly verify another user's device seems to send an m.key.verification.start directly, so it is not affected by this case.

I haven't tried Element Android or Element iOS yet.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, that matches my tests as well. It can deal with request, but if you start it from Element, it uses start (for to_device verifications).

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I just tested Element Web, and it sent an m.key.verification.done after verifying using to-device. I haven't tested Android or iOS. In response to the original question in this thread:

How long should the other client wait until it receives a done, as old clients won't send that?

I think that we're at the point where we don't need to worry about "old" clients (it looks like Element is doing this, and just waiting until the normal timeout), but if you do want to worry about old clients, you could 1) wait for the .done, but allow the user to exit the verification UI, or 2) exit the verification UI once it's completed on your side, without waiting for the .done to come in from the other user.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does element-web also send a .done if you don't initiate verification via popup but via the device menu on the right? The device menu on the right also skips the .request and starts directly with a .start, so yeah

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes. I tested initialting verification both via the popup and via the device menu. In both situations, it sent the .done in my tests. (I also noticed that initiating from the device menu skipped the .request step.)

than the usual `transaction_id` or `m.relates_to` field. This indicates that
the device has successfully completed its side of the verification.

## Potential issues

Clients that follow the Client-Server 0.6.0 spec may not expect an
`m.key.verification.ready` message in response to `m.key.verification.request`.
However to our knowledge, no clients implement `m.key.verification.request` in
this way yet -- to our knowledge, all clients that implement verification
implement this proposal.