Simplex Messaging Protocol is a transport agnostic client-server protocol for asynchronous distributed secure unidirectional message transmission via persistent simplex message queues.
It's designed with the focus on communication security and integrity, under the assumption that any part of the message transmission network can be compromised.
It is designed as a low level protocol for other application protocols to solve the problem of secure and private message transmission, making [MITM attack][1] very difficult at any part of the message transmission system.
The objective of Simplex Messaging Protocol (SMP) is to facilitate the secure and private unidirectional transfer of messages from senders to recipients via persistent simplex queues managed by the message broker (server).
SMP is independent of any particular transmission system and requires only a reliable ordered data stream channel. While this document describes transport over TCP, other transports are also possible.
The protocol describes the set of commands that recipients and senders can exchange with SMP servers to create and to operate unidirectional "queues" (a data abstraction identifying one of many communication channels managed by the server) and to send messages from the sender to the recipient via the SMP server.
More complex communication scenarios can be designed using multiple queues - for example, a duplex communication channel can be made of 2 simplex queues.
The protocol is designed with the focus on privacy and security, to some extent deprioritizing reliability by requiring that SMP servers only store messages until they are acknowledged by the recipients and, in any case, for a limited period of time. For communication scenarios requiring more reliable transmission the users should use several SMP servers to pass each message and implement some additional protocol to ensure that messages are not removed, inserted or changed - this is out of scope of this document.
SMP does not use any form of participants' identities and provides [E2EE][2] without the possibility of [MITM attack][1] relying on two pre-requisites:
- the users can establish a secure encrypted transport connection with the SMP server. [Appendix A](#appendix-a) describes SMP transport protocol of such connection over TCP, but any other transport connection protocol can be used.
- the recipient can pass a single message to the sender via a pre-existing secure and private communication channel (out-of-band message) - the information in this message is used to encrypt messages and to establish connection with SMP server.
The SMP model has three communication participants: the recipient, the message broker (SMP server) that is chosen and, possibly, controlled by the recipient, and the sender.
SMP server manages multiple "simplex queues" - data records on the server that identify communication channels from the senders to the recipients. The same communicating party that is the sender in one queue, can be the recipient in another - without exposing this fact to the server.
The queue record consists of 2 unique random IDs generated by the server, one for the recipient and another for the sender, and 2 keys to authenticate the recipient and the sender respectively, provided by the client. The users of SMP protocol must use a unique key for each queue, to avoid the possibility of aggregating and analyzing their queues in case SMP server is compromised.
Creating and using the queue requires sending commands to the SMP server from the recipient and the sender - they are described in detail in [SMP commands](#smp-commands) section.
The out-of-band message with the queue information is sent via some trusted alternative channel from the recipient to the sender. This message is used to share one or several [queue URIs](#smp-queue-uri) that parties can use to establish the initial connection, the encryption scheme and, it can include the public key(s) for end-to-end encryption.
- Recipient of the queue (who created the queue and sent out-of-band message) will use it to retrieve messages from the server, signing the commands by the recipient key. Recipient decrypts the messages with the key negotiated during the creation of the queue.
This simplex queue can serve as a building block for more complex communication network. For example, two (or more, for redundancy) simplex queues can be used to create a duplex communication channel. Higher level primitives that are only known to system participants in their client applications can be created as well - e.g., contacts, conversations, groups and broadcasts. Simplex messaging servers only have the information about the low-level simplex queues. In this way a high level of privacy and security of the communication is provided. Application level primitives are not in scope of this protocol.
Access to each queue is controlled with unique (not shared with other queues) asymmetric key pairs, separate for the sender and the recipient. The sender and the receiver have private keys, and the server has associated public keys to authenticate participants' commands by verifying cryptographic signatures.
Queue is defined by recipient ID `RID` and sender ID `SID`, unique for the server. Sender key (`SK`) is used by the server to verify sender's commands (identified by `SID`) to send messages. Recipient key (`RK`) is used by the server to verify recipient's commands (identified by `RID`) to retrieve messages.
The protocol uses different IDs for sender and recipient in order to provide an additional privacy by preventing the correlation of senders and recipients commands sent over the network - in case the encrypted transport is compromised, it would still be difficult to correlate senders and recipients without access to the queue records on the server.
The SMP queue URIs MUST include server identity, queue hostname, an optional port, sender queue ID and the public key that the clients must use to verify responses. Server identity is used to establish secure connection protected from MITM attack with SMP server (see [Appendix A](#appendix-a) for SMP transport protocol).
`hostname` can be IP address or domain name, as defined in RFC 1123, section 2.1.
`port` is optional, the default TCP port for SMP protocol is 5223.
`serverIdentity` is a required hash of the server certificate SPKI block (without line breaks, header and footer) used by the client to validate server certificate during transport handshake (see [Appendix A](#appendix-a))
The SMP procedure of creating a simplex queue on SMP server is explained using participants Alice (the recipient) who wants to receive messages from Bob (the sender).
1. Decides which SMP server to use (can be the same or different server that Alice uses for other queues) and opens secure encrypted transport connection to the chosen SMP server (see [Appendix A](#appendix-a)).
3. Generates another new random public/private key pair (recipient key - `RK`) that she did not use before for her to sign commands and to decrypt the transmissions received from the server.
4. Generates one more random key pair (recipient DH key - `RDHK`) to negotiate symmetric key that will be used by the server to encrypt message bodies delivered to Alice (to avoid shared cipher-text inside transport connection).
5. Sends `"NEW"` command to the server to create a simplex queue (see `create` in [Create queue command](#create-queue-command)). This command contains previously generated unique "public" keys `RK` and `RDHK`. `RK` will be used to verify the following commands related to the same queue signed by its private counterpart, for example to subscribe to the messages received to this queue or to update the queue, e.g. by setting the key required to send the messages (initially Alice creates the queue that accepts unsigned messages, so anybody could send the message via this queue if they knew the queue sender's ID and server address).
- Server public DH key (`SDHK`) to negotiate a shared secret for message body encryption, that Alice uses to derive a shared secret with the server `SS`.
2. Alice sends an out-of-band message to Bob via the alternative channel that both Alice and Bob trust (see [protocol abstract](#simplex-messaging-protocol-abstract)). The message must include:
4. Sends the encrypted message to the server with queue ID `SID` (see `send` in [Send message](#send-message)). This initial message to the queue must not be signed - signed messages will be rejected until Alice secures the queue (below).
4. Alice receives Bob's message from the server using recipient queue ID `RID` (possibly, via the same transport connection she already has opened - see `message` in [Deliver queue message](#deliver-queue-message)):
3. Even though anybody could have sent the message to the queue with ID `SID` before it is secured (e.g. if communication is compromised), Alice would ignore all messages until the decryption succeeds (i.e. the result contains the expected message format). Optionally, in the client application, she also may identify Bob using the information provided, but it is out of scope of SMP protocol.
1. She sends the `KEY` command with `RID` signed with "private" key `RK` to update the queue to only accept requests signed by "private" key `SK` provided by Bob. This command contains unique "public" key `SK` previously generated by Bob.
2. From this moment the server will accept only signed commands to `SID`, so only Bob will be able to send messages to the queue `SID` (corresponding to `RID` that Alice has).
3. Once queue is secured, Alice deletes `SID` and `SK` - even if Alice's client is compromised in the future, the attacker would not be able to send messages pretending to be Bob.
3. He sends the command to the server (see `send` in [Send message](#send-message)), that the server will authenticate using the "public" key `SK` (that Alice earlier received from Bob and provided to the server via `"KEY"` command).
1. She signs `"SUB"` command to the server to subscribe to the queue `RID` with the "private" key `RK` (see `subscribe` in [Subscribe to queue](#subscribe-to-queue)).
2. The server, having authenticated Alice's command with the "public" key `RK` that she provided, delivers Bob's message(s) (see `message` in [Deliver queue message](#deliver-queue-message)).
A higher level protocol application protocol should define the semantics that allow to use two simplex queues (or two sets of queues for redundancy) for the bi-directional or any other communication scenarios.
The SMP is intentionally unidirectional - it provides no answer to how Bob will know that the transmission succeeded, and whether Alice received any messages. There may be a scenario when Alice wants to securely receive the messages from Bob, but she does not want Bob to have any proof that she received any messages - this low-level protocol can be used in this scenario, as all Bob knows as a fact is that he was able to send one unsigned message to the server that Alice provided, and now he can only send messages signed with the key `SK` that he sent to the server - it does not prove that any message was received by Alice.
For practical purposes of bi-directional conversation, now that Bob can securely send encrypted messages to Alice, Bob can create the second simplex queue that will allow Alice to send messages to Bob in the same way, sending the second queue details via the first queue. If both Alice and Bob have their respective unique "public" keys (Alice's and Bob's `EK`s of two separate queues), or pass additional keys to sign the messages, the conversation can be both encrypted and signed.
This protocol also can be used for off-the-record messaging, as Alice and Bob can use multiple queues between them and only information they pass to each other allows proving their identity, so if they want to share anything off-the-record they can initiate a new queue without linking it to any other information they exchanged. As a result, this protocol provides better anonymity and better protection from [MITM][1] than [OTR][6] protocol.
- Transport agnostic - the protocol does not define how clients connect to the servers. It can be implemented over any ordered data stream channel: TCP connection, HTTP with long polling, websockets, etc.
- Not semantic - the protocol does not assign any meaning to queues and messages. While on the application level the queues and messages can have different meaning (e.g., for messages: text or image chat message, message acknowledgement, participant profile information, status updates, changing "public" key to encrypt messages, changing servers, etc.), on SMP protocol level all the messages are binary and their meaning can only be interpreted by client applications and not by the servers - this interpretation is out of scope of this protocol.
- Multiple servers, that can be deployed by the system users, can be used to send and retrieve messages.
- Servers do not communicate with each other and do not "know" about other servers.
- Clients only communicate with servers (excluding the initial out-of-band message), so the message passing is asynchronous.
- For each queue, the message recipient defines the server through which the sender should send messages.
- While multiple servers and multiple queues can be used to pass each message, it is in scope of application level protocol(s), and out of scope of this protocol.
- Servers store messages only until they are retrieved by the recipients, and in any case, for a limited time.
- Servers are required to NOT store any message history or delivery log, but even if the server is compromised, it does not allow to decrypt the messages or to determine the list of queues established by any participant - this information is only stored on client devices.
- The only element provided by SMP servers is simplex queues:
- Each queue is created and managed by the queue recipient.
- One unique "public" key is used by the servers to authenticate requests to send the messages into the queue, and another unique "public" key - to retrieve the messages from the queue. "Unique" here means that each "public" key is used only for one queue and is not used for any other context - effectively, this key is not public and does not represent any participant identity.
- Both recipient and sender "public" keys are provided to the server by the queue recipient. "Public" key `RK` is provided when the queue is created, public key `SK` is provided when the queue is secured.
- The "public" keys known to the server and used to authenticate commands from the participants are unrelated to the keys used to encrypt and decrypt the messages - the latter keys are also unique per each queue but they are only known to participants, not to the servers.
- Messaging graph can be asymmetric: Bob's ability to send messages to Alice does not automatically lead to the Alice's ability to send messages to Bob.
To encrypt/decrypt message bodies delivered to the recipients, servers/clients MUST use x25519 or x448 algorithm defined in [RFC8709][15] to derive the shared secret (TODO encryption scheme).
Clients MUST encrypt message bodies sent via SMP servers - the protocol for this end-to-end encryption should be chosen by the clients using SMP protocol.
Simplex messaging servers MUST generate 2 different IDs for each new queue - for the recipient (that created the queue) and for the sender. It is REQUIRED that:
- History of deleted queues, retrieved or acknowledged messages (deleted queues MAY be stored temporarily as part of the queue persistence implementation).
- Snapshots of the database they use to store queues and messages (instead simplex messaging clients must manage redundancy by using more than one simplex messaging server). In-memory persistence is recommended.
Supporting message delivery while the client mobile app is not running requires sending push notifications with the device token. All alternative mechanisms for background message delivery are unreliable, particularly on iOS platform. Obviously, supporting push notification delivery by simply subscribing to messages would reduce meta-data privacy as it allows to see all queues that a given device uses.
To protect the privacy of the recipients, there are several commands in SMP protocol that allow enabling and subscribing to message notifications from SMP queues, using separate set of "notifier keys" and via separate queue IDs - as long as SMP server is not compromised, these notifier queue IDs cannot be correlated with recipient or sender queue IDs.
The clients can optionally instruct a dedicated push notification server to subscribe to notifications and deliver push notifications to the device, which can then retrieve the messages in the background and send local notifications to the user - this is out of scope of SMP protocol. The commands that SMP protocol provides to allow it:
-`enableNotifications` (`"NKEY"`) with `notifierId` (`"NID"`) response - see [Enable notifications command](#enable-notifications-command).
[`SEND` command](#send-message) includes the notification flag to instruct SMP server whether to send the notification - this flag is forwarded to the recepient inside encrypted envelope, together with the timestamp and the message body, so even if TLS is compromised this flag cannot be used for traffic correlation.
Some parts of SMP transmission are padded to a fixed size; this padding is uniformly added as a word16 encoded in network byte order - see `paddedString` syntax.
The server should send `queueIds`, `error` and `ok` responses in the same order within each queue ID as the commands received in the transport connection, so that they can be correlated by the clients. To simplify correlation of commands and responses, the server must use the same `corrId` in the response as in the command sent by the client.
SMP servers must authenticate all transmissions (excluding `ping` and initial `send` commands) by verifying the client signatures. Command signature should be generated by applying the algorithm specified for the queue to the `signed` block of the transmission, using the key associated with the queue ID (recipient's, sender's or notifier's, depending on which queue ID is used).
To keep the transport connection alive and to generate noise traffic the clients should use `ping` command to which the server responds with `ok` response. This command should be sent unsigned and without queue ID.
Sending any of the commands in this section (other than `create`, that is sent without queue ID) is only allowed with recipient's ID (`RID`). If sender's ID is used the server must respond with `"ERR AUTH"` response (see [Error responses](#error-responses)).
If the queue is created successfully, the server must send `queueIds` response with the recipient's and sender's queue IDs and public keys to sign all responses and messages and to encrypt delivered message bodies:
Once the queue is created, the recipient gets automatically subscribed to receive the messages from that queue, until the transport connection is closed. The `subscribe` command is needed only to start receiving the messages from the existing queue when the new transport connection is opened.
`NEW` transmission MUST be signed using the private part of the `recipientSignaturePublicKey`– this verifies that the client has the private key that will be used to sign subsequent commands for this queue.
If subscription is successful the server must respond with the first available message or with `ok` response if no messages are available. The recipient will continue receiving the messages from this queue until the transport connection is closed or until another transport connection subscribes to the same simplex queue - in this case the first subscription should be cancelled and [subscription END notification](#subscription-end-notification) delivered.
The first message will be delivered either immediately or as soon as it is available; to receive the following message the recipient must acknowledge the reception of the message (see [Acknowledge message delivery](#acknowledge-message-delivery)).
This command is sent by the recipient to the server to add notifier's key to the queue, to allow push notifications server to receive notifications when the message arrives, via a separate queue ID, without receiving message content.
The recipient should send the acknowledgement of message delivery once the message was stored in the client, to notify the server that the message should be deleted:
Even if acknowledgement is not sent by the recipient, the server should limit the time of message storage, whether it was delivered to the recipient or not.
Having received the acknowledgement, SMP server should immediately delete the message and then send the next available message or respond with `ok` if there are no more messages available in this simplex queue.
This command can be sent multiple times (in case transport connection was interrupted and the response was not delivered), the server should still respond `ok` even if the queue is already suspended.
Currently SMP defines only one command that can be used by senders - `send` message. This command must be used with sender's ID, if recipient's ID is used the server must respond with `"ERR AUTH"` response (see [Error responses](#error-responses)).
This command is sent to the server by the sender both to confirm the queue after the sender received out-of-band message from the recipient and to send messages after the queue is secured:
The first message is sent to confirm the queue - it should contain sender's server key (see decrypted message syntax below) - this first message must be sent without signature.
Until the queue is secured, the server should accept any number of unsigned messages - it both enables the legitimate sender to resend the confirmation in case of failure and also allows the simplex messaging client to ignore any confirmation messages that may be sent by the attackers (assuming they could have intercepted the queue ID in the server response, but do not have a correct encryption key passed to sender in out-of-band message).
`clientHeader` in the initial unsigned message is used to transmit sender's server key and can be used in the future revisions of SMP protocol for other purposes.
20- | agentPublicHeader (the size is for user messages post handshake, without E2E X3DH keys - it is version and 'M' for the messages - 3 bytes in total)
The push notifications server (notifier) must use this command to start receiving message notifications from the queue:
```abnf
subscribeNotifications = %s"NSUB"
```
If subscription is successful the server must respond with `ok` response if no messages are available. The notifier will be receiving the message notifications from this queue until the transport connection is closed or until another transport connection subscribes to notifications from the same simplex queue - in this case the first subscription should be cancelled and [subscription END notification](#subscription-end-notification) delivered.
The first message notification will be delivered either immediately or as soon as the message is available.
`msgId` - unique message ID generated by the server based on cryptographically strong random bytes. It should be used by the clients to detect messages that were delivered more than once (in case the transport connection was interrupted and the server did not receive the message delivery acknowledgement). Message ID is used as a nonce for server/recipient encryption of message bodies.
`timestamp` - system time when the server received the message from the sender as **a number of seconds** since Unix epoch (1970-01-01) encoded as 64-bit integer in network byte order. If a client system/language does not support 64-bit integers, until 2106 it is safe to simply skip the first 4 zero bytes and decode 32-bit unsigned integer (or as signed integer until 2038).
When server delivers the messages to the recipient, message body should be encrypted with the secret derived from DH exchange using the keys passed during the queue creation and returned with `queueIds` response.
This is done to prevent the possibility of correlation of incoming and outgoing traffic of SMP server inside transport protocol.
#### Notifier queue ID response
Server must respond with this message when queue notifications are enabled.
See its syntax in [Enable notifications command](#enable-notifications-command)
#### Deliver message notification
The server must deliver message notifications to all simplex queues that were subscribed with `subscribeNotifications` command ("NSUB") on the currently open transport connection. The syntax for the message notification delivery is:
When another transport connection is subscribed to the same simplex queue, the server should unsubscribe and to send the notification to the previously subscribed transport connection:
- authentication error (`AUTH`) - incorrect signature, unknown (or suspended) queue, sender's ID is used in place of recipient's and vice versa, and some other cases (see [Send message](#send-message) command).
- message queue quota exceeded error (`QUOTA`) - too many messages were sent to the message queue. Further messages can only be sent after the recipient retrieves the messages.
Server implementations must aim to respond within the same time for each command in all cases when `"ERR AUTH"` response is required to prevent timing attacks (e.g., the server should perform signature verification even when the queue does not exist on the server or the signature of different size is sent, using any RSA key with the same size as the signature size).
Both the recipient and the sender can use TCP or some other, possibly higher level, transport protocol to communicate with the server. The default TCP port for SMP server is 5223.
- servers must send the chain of exactly 2 self-signed certificates in the handshake, with the first (offline) certificate one signing the second (online) certificate. Offline certificate fingerprint is used as a server identity - it is a part of SMP server address.
- The clients must abort the connection in case a different number of certificates is sent.
During TLS handshake the client must validate that the fingerprint of the online server certificate is equal to the `serverIdentity` the client received as part of SMP server address; if the server identity does not match the client must abort the connection.
The first block sent by the server should be `serverHello` and the client should respond with `clientHello` - these blocks are used to agree SMP protocol version:
For TLS transport client should assert that `sessionIdentifier` is equal to `tls-unique` channel binding defined in [RFC 5929][14] (TLS Finished message struct); we pass it in `serverHello` block to allow communication over some other transport protocol (possibly, with another channel binding).