2   Unpadded Base64

Unpadded Base64 refers to 'standard' Base64 encoding as defined in RFC 4648, without "=" padding. Specifically, where RFC 4648 requires that encoded data be padded to a multiple of four characters using = characters, unpadded Base64 omits this padding.

For reference, RFC 4648 uses the following alphabet for Base 64:

Value Encoding  Value Encoding  Value Encoding  Value Encoding
    0 A            17 R            34 i            51 z
    1 B            18 S            35 j            52 0
    2 C            19 T            36 k            53 1
    3 D            20 U            37 l            54 2
    4 E            21 V            38 m            55 3
    5 F            22 W            39 n            56 4
    6 G            23 X            40 o            57 5
    7 H            24 Y            41 p            58 6
    8 I            25 Z            42 q            59 7
    9 J            26 a            43 r            60 8
   10 K            27 b            44 s            61 9
   11 L            28 c            45 t            62 +
   12 M            29 d            46 u            63 /
   13 N            30 e            47 v
   14 O            31 f            48 w
   15 P            32 g            49 x
   16 Q            33 h            50 y

Examples of strings encoded using unpadded Base64:

UNPADDED_BASE64("") = ""
UNPADDED_BASE64("f") = "Zg"
UNPADDED_BASE64("fo") = "Zm8"
UNPADDED_BASE64("foo") = "Zm9v"
UNPADDED_BASE64("foob") = "Zm9vYg"
UNPADDED_BASE64("fooba") = "Zm9vYmE"
UNPADDED_BASE64("foobar") = "Zm9vYmFy"

When decoding Base64, implementations SHOULD accept input with or without padding characters whereever possible, to ensure maximum interoperability.

3   Signing JSON

Various points in the Matrix specification require JSON objects to be cryptographically signed. This requires us to encode the JSON as a binary string. Unfortunately the same JSON can be encoded in different ways by changing how much white space is used or by changing the order of keys within objects.

Signing an object therefore requires it to be encoded as a sequence of bytes using Canonical JSON, computing the signature for that sequence and then adding the signature to the original JSON object.

3.1   Canonical JSON

We define the canonical JSON encoding for a value to be the shortest UTF-8 JSON encoding with dictionary keys lexicographically sorted by unicode codepoint. Numbers in the JSON must be integers in the range [-(2**53)+1, (2**53)-1].

We pick UTF-8 as the encoding as it should be available to all platforms and JSON received from the network is likely to be already encoded using UTF-8. We sort the keys to give a consistent ordering. We force integers to be in the range where they can be accurately represented using IEEE double precision floating point numbers since a number of JSON libraries represent all numbers using this representation.

import json

def canonical_json(value):
    return json.dumps(
        value,
        # Encode code-points outside of ASCII as UTF-8 rather than \u escapes
        ensure_ascii=False,
        # Remove unnecessary white space.
        separators=(',',':'),
        # Sort the keys of dictionaries.
        sort_keys=True,
        # Encode the resulting unicode as UTF-8 bytes.
    ).encode("UTF-8")

3.1.1   Grammar

Adapted from the grammar in http://tools.ietf.org/html/rfc7159 removing insignificant whitespace, fractions, exponents and redundant character escapes

value     = false / null / true / object / array / number / string
false     = %x66.61.6c.73.65
null      = %x6e.75.6c.6c
true      = %x74.72.75.65
object    = %x7B [ member *( %x2C member ) ] %7D
member    = string %x3A value
array     = %x5B [ value *( %x2C value ) ] %5B
number    = [ %x2D ] int
int       = %x30 / ( %x31-39 *digit )
digit     = %x30-39
string    = %x22 *char %x22
char      = unescaped / %x5C escaped
unescaped = %x20-21 / %x23-5B / %x5D-10FFFF
escaped   = %x22 ; "    quotation mark  U+0022
          / %x5C ; \    reverse solidus U+005C
          / %x62 ; b    backspace       U+0008
          / %x66 ; f    form feed       U+000C
          / %x6E ; n    line feed       U+000A
          / %x72 ; r    carriage return U+000D
          / %x74 ; t    tab             U+0009
          / %x75.30.30.30 (%x30-37 / %x62 / %x65-66) ; u000X
          / %x75.30.30.31 (%x30-39 / %x61-66)        ; u001X

3.1.2   Examples

To assist in the development of compatible implementations, the following test values may be useful for verifying the canonical transformation code.

Given the following JSON object:

{}

The following canonical JSON should be produced:

{}

Given the following JSON object:

{
    "one": 1,
    "two": "Two"
}

The following canonical JSON should be produced:

{"one":1,"two":"Two"}

Given the following JSON object:

{
    "b": "2",
    "a": "1"
}

The following canonical JSON should be produced:

{"a":"1","b":"2"}

Given the following JSON object:

{"b":"2","a":"1"}

The following canonical JSON should be produced:

{"a":"1","b":"2"}

Given the following JSON object:

{
    "auth": {
        "success": true,
        "mxid": "@john.doe:example.com",
        "profile": {
            "display_name": "John Doe",
            "three_pids": [
                {
                    "medium": "email",
                    "address": "john.doe@example.org"
                },
                {
                    "medium": "msisdn",
                    "address": "123456789"
                }
            ]
        }
    }
}

The following canonical JSON should be produced:

{"auth":{"mxid":"@john.doe:example.com","profile":{"display_name":"John Doe","three_pids":[{"address":"john.doe@example.org","medium":"email"},{"address":"123456789","medium":"msisdn"}]},"success":true}}

Given the following JSON object:

{
    "a": "日本語"
}

The following canonical JSON should be produced:

{"a":"日本語"}

Given the following JSON object:

{
    "本": 2,
    "日": 1
}

The following canonical JSON should be produced:

{"日":1,"本":2}

Given the following JSON object:

{
    "a": "\u65E5"
}

The following canonical JSON should be produced:

{"a":"日"}

Given the following JSON object:

{
    "a": null
}

The following canonical JSON should be produced:

{"a":null}

3.2   Signing Details

JSON is signed by encoding the JSON object without signatures or keys grouped as unsigned, using the canonical encoding described above. The JSON bytes are then signed using the signature algorithm and the signature is encoded using unpadded Base64. The resulting base64 signature is added to an object under the signing key identifier which is added to the signatures object under the name of the entity signing it which is added back to the original JSON object along with the unsigned object.

The signing key identifier is the concatenation of the signing algorithm and a key identifier. The signing algorithm identifies the algorithm used to sign the JSON. The currently supported value for signing algorithm is ed25519 as implemented by NACL (http://nacl.cr.yp.to/). The key identifier is used to distinguish between different signing keys used by the same entity.

The unsigned object and the signatures object are not covered by the signature. Therefore intermediate entities can add unsigned data such as timestamps and additional signatures.

{
   "name": "example.org",
   "signing_keys": {
     "ed25519:1": "XSl0kuyvrXNj6A+7/tkrB9sxSbRi08Of5uRhxOqZtEQ"
   },
   "unsigned": {
      "age_ts": 922834800000
   },
   "signatures": {
      "example.org": {
         "ed25519:1": "s76RUgajp8w172am0zQb/iPTHsRnb4SkrzGoeCOSFfcBY2V/1c8QfrmdXHpvnc2jK5BD1WiJIxiMW95fMjK7Bw"
      }
   }
}
def sign_json(json_object, signing_key, signing_name):
    signatures = json_object.pop("signatures", {})
    unsigned = json_object.pop("unsigned", None)

    signed = signing_key.sign(encode_canonical_json(json_object))
    signature_base64 = encode_base64(signed.signature)

    key_id = "%s:%s" % (signing_key.alg, signing_key.version)
    signatures.setdefault(signing_name, {})[key_id] = signature_base64

    json_object["signatures"] = signatures
    if unsigned is not None:
        json_object["unsigned"] = unsigned

    return json_object

3.3   Checking for a Signature

To check if an entity has signed a JSON object an implementation does the following:

  1. Checks if the signatures member of the object contains an entry with the name of the entity. If the entry is missing then the check fails.
  2. Removes any signing key identifiers from the entry with algorithms it doesn't understand. If there are no signing key identifiers left then the check fails.
  3. Looks up verification keys for the remaining signing key identifiers either from a local cache or by consulting a trusted key server. If it cannot find a verification key then the check fails.
  4. Decodes the base64 encoded signature bytes. If base64 decoding fails then the check fails.
  5. Removes the signatures and unsigned members of the object.
  6. Encodes the remainder of the JSON object using the Canonical JSON encoding.
  7. Checks the signature bytes against the encoded object using the verification key. If this fails then the check fails. Otherwise the check succeeds.

4   Identifier Grammar

4.1   Server Name

A homeserver is uniquely identified by its server name. This value is used in a number of identifiers, as described below.

The server name represents the address at which the homeserver in question can be reached by other homeservers. The complete grammar is:

server_name = dns_name [ ":" port]
dns_name = host
port = *DIGIT

where host is as defined by RFC3986, section 3.2.2.

Examples of valid server names are:

  • matrix.org
  • matrix.org:8888
  • 1.2.3.4 (IPv4 literal)
  • 1.2.3.4:1234 (IPv4 literal with explicit port)
  • [1234:5678::abcd] (IPv6 literal)
  • [1234:5678::abcd]:5678 (IPv6 literal with explicit port)

4.2   Common Identifier Format

The Matrix protocol uses a common format to assign unique identifiers to a number of entities, including users, events and rooms. Each identifier takes the form:

&localpart:domain

where & represents a 'sigil' character; domain is the server name of the homeserver which allocated the identifier, and localpart is an identifier allocated by that homeserver.

The sigil characters are as follows:

  • @: User ID
  • !: Room ID
  • $: Event ID
  • #: Room alias

The precise grammar defining the allowable format of an identifier depends on the type of identifier.

4.2.1   User Identifiers

Users within Matrix are uniquely identified by their Matrix user ID. The user ID is namespaced to the homeserver which allocated the account and has the form:

@localpart:domain

The localpart of a user ID is an opaque identifier for that user. It MUST NOT be empty, and MUST contain only the characters a-z, 0-9, ., _, =, -, and /.

The domain of a user ID is the server name of the homeserver which allocated the account.

The length of a user ID, including the @ sigil and the domain, MUST NOT exceed 255 characters.

The complete grammar for a legal user ID is:

user_id = "@" user_id_localpart ":" server_name
user_id_localpart = 1*user_id_char
user_id_char = DIGIT
             / %x61-7A                   ; a-z
             / "-" / "." / "=" / "_" / "/"

Rationale

A number of factors were considered when defining the allowable characters for a user ID.

Firstly, we chose to exclude characters outside the basic US-ASCII character set. User IDs are primarily intended for use as an identifier at the protocol level, and their use as a human-readable handle is of secondary benefit. Furthermore, they are useful as a last-resort differentiator between users with similar display names. Allowing the full unicode character set would make very difficult for a human to distinguish two similar user IDs. The limited character set used has the advantage that even a user unfamiliar with the Latin alphabet should be able to distinguish similar user IDs manually, if somewhat laboriously.

We chose to disallow upper-case characters because we do not consider it valid to have two user IDs which differ only in case: indeed it should be possible to reach @user:matrix.org as @USER:matrix.org. However, user IDs are necessarily used in a number of situations which are inherently case-sensitive (notably in the state_key of m.room.member events). Forbidding upper-case characters (and requiring homeservers to downcase usernames when creating user IDs for new users) is a relatively simple way to ensure that @USER:matrix.org cannot refer to a different user to @user:matrix.org.

Finally, we decided to restrict the allowable punctuation to a very basic set to reduce the possibility of conflicts with special characters in various situations. For example, "*" is used as a wildcard in some APIs (notably the filter API), so it cannot be a legal user ID character.

The length restriction is derived from the limit on the length of the sender key on events; since the user ID appears in every event sent by the user, it is limited to ensure that the user ID does not dominate over the actual content of the events.

Matrix user IDs are sometimes informally referred to as MXIDs.

4.2.1.1   Historical User IDs

Older versions of this specification were more tolerant of the characters permitted in user ID localparts. There are currently active users whose user IDs do not conform to the permitted character set, and a number of rooms whose history includes events with a sender which does not conform. In order to handle these rooms successfully, clients and servers MUST accept user IDs with localparts from the expanded character set:

extended_user_id_char = %x21-7E

4.2.1.2   Mapping from other character sets

In certain circumstances it will be desirable to map from a wider character set onto the limited character set allowed in a user ID localpart. Examples include a homeserver creating a user ID for a new user based on the username passed to /register, or a bridge mapping user ids from another protocol.

Implementations are free to do this mapping however they choose. Since the user ID is opaque except to the implementation which created it, the only requirement is that the implemention can perform the mapping consistently. However, we suggest the following algorithm:

  1. Encode character strings as UTF-8.
  2. Convert the bytes A-Z to lower-case.
    • In the case where a bridge must be able to distinguish two different users with ids which differ only by case, escape upper-case characters by prefixing with _ before downcasing. For example, A becomes _a. Escape a real _ with a second _.
  3. Encode any remaining bytes outside the allowed character set, as well as =, as their hexadecimal value, prefixed with =. For example, # becomes =23; á becomes =c3=a1.

Rationale

The suggested mapping is an attempt to preserve human-readability of simple ASCII identifiers (unlike, for example, base-32), whilst still allowing representation of any character (unlike punycode, which provides no way to encode ASCII punctuation).

4.2.2   Room IDs and Event IDs

A room has exactly one room ID. A room ID has the format:

!opaque_id:domain

An event has exactly one event ID. An event ID has the format:

$opaque_id:domain

The domain of a room/event ID is the server name of the homeserver which created the room/event. The domain is used only for namespacing to avoid the risk of clashes of identifiers between different homeservers. There is no implication that the room or event in question is still available at the corresponding homeserver.

Event IDs and Room IDs are case-sensitive. They are not meant to be human readable.

4.2.3   Room Aliases

A room may have zero or more aliases. A room alias has the format:

#room_alias:domain

The domain of a room alias is the server name of the homeserver which created the alias. Other servers may contact this homeserver to look up the alias.

Room aliases MUST NOT exceed 255 bytes (including the # sigil and the domain).

5   3PID Types

Third Party Identifiers (3PIDs) represent identifiers on other namespaces that might be associated with a particular person. They comprise a tuple of medium which is a string that identifies the namespace in which the identifier exists, and an address: a string representing the identifier in that namespace. This must be a canonical form of the identifier, i.e. if multiple strings could represent the same identifier, only one of these strings must be used in a 3PID address, in a well-defined manner.

For example, for e-mail, the medium is 'email' and the address would be the email address, e.g. the string bob@example.com. Since domain resolution is case-insensitive, the email address bob@Example.com is also has the 3PID address of bob@example.com (without the capital 'e') rather than bob@Example.com.

The namespaces defined by this specification are listed below. More namespaces may be defined in future versions of this specification.

5.1   E-Mail

Medium: email

Represents E-Mail addresses. The address is the raw email address in user@domain form with the domain in lowercase. It must not contain other text such as real name, angle brackets or a mailto: prefix.

5.2   PSTN Phone numbers

Medium: msisdn

Represents telephone numbers on the public switched telephone network. The address is the telephone number represented as a MSISDN (Mobile Station International Subscriber Directory Number) as defined by the E.164 numbering plan. Note that MSISDNs do not include a leading '+'.

6   Security Threat Model

6.1   Denial of Service

The attacker could attempt to prevent delivery of messages to or from the victim in order to:

  • Disrupt service or marketing campaign of a commercial competitor.
  • Censor a discussion or censor a participant in a discussion.
  • Perform general vandalism.

6.1.1   Threat: Resource Exhaustion

An attacker could cause the victims server to exhaust a particular resource (e.g. open TCP connections, CPU, memory, disk storage)

6.1.2   Threat: Unrecoverable Consistency Violations

An attacker could send messages which created an unrecoverable "split-brain" state in the cluster such that the victim's servers could no longer derive a consistent view of the chatroom state.

6.1.3   Threat: Bad History

An attacker could convince the victim to accept invalid messages which the victim would then include in their view of the chatroom history. Other servers in the chatroom would reject the invalid messages and potentially reject the victims messages as well since they depended on the invalid messages.

6.1.4   Threat: Block Network Traffic

An attacker could try to firewall traffic between the victim's server and some or all of the other servers in the chatroom.

6.1.5   Threat: High Volume of Messages

An attacker could send large volumes of messages to a chatroom with the victim making the chatroom unusable.

6.1.6   Threat: Banning users without necessary authorisation

An attacker could attempt to ban a user from a chatroom without the necessary authorisation.

6.2   Spoofing

An attacker could try to send a message claiming to be from the victim without the victim having sent the message in order to:

  • Impersonate the victim while performing illicit activity.
  • Obtain privileges of the victim.

6.2.1   Threat: Altering Message Contents

An attacker could try to alter the contents of an existing message from the victim.

6.2.2   Threat: Fake Message "origin" Field

An attacker could try to send a new message purporting to be from the victim with a phony "origin" field.

6.3   Spamming

The attacker could try to send a high volume of solicited or unsolicited messages to the victim in order to:

  • Find victims for scams.
  • Market unwanted products.

6.3.1   Threat: Unsolicited Messages

An attacker could try to send messages to victims who do not wish to receive them.

6.3.2   Threat: Abusive Messages

An attacker could send abusive or threatening messages to the victim

6.4   Spying

The attacker could try to access message contents or metadata for messages sent by the victim or to the victim that were not intended to reach the attacker in order to:

  • Gain sensitive personal or commercial information.
  • Impersonate the victim using credentials contained in the messages. (e.g. password reset messages)
  • Discover who the victim was talking to and when.

6.4.1   Threat: Disclosure during Transmission

An attacker could try to expose the message contents or metadata during transmission between the servers.

6.4.2   Threat: Disclosure to Servers Outside Chatroom

An attacker could try to convince servers within a chatroom to send messages to a server it controls that was not authorised to be within the chatroom.

6.4.3   Threat: Disclosure to Servers Within Chatroom

An attacker could take control of a server within a chatroom to expose message contents or metadata for messages in that room.

7   Cryptographic Test Vectors

To assist in the development of compatible implementations, the following test values may be useful for verifying the cryptographic event signing code.

7.1   Signing Key

The following test vectors all use the 32-byte value given by the following Base64-encoded string as the seed for generating the ed25519 signing key:

SIGNING_KEY_SEED = decode_base64(
    "YJDBA9Xnr2sVqXD9Vj7XVUnmFZcZrlw8Md7kMW+3XA1"
)

In each case, the server name and key ID are as follows:

SERVER_NAME = "domain"

KEY_ID = "ed25519:1"

7.2   JSON Signing

Given an empty JSON object:

{}

The JSON signing algorithm should emit the following signed data:

{
    "signatures": {
        "domain": {
            "ed25519:1": "K8280/U9SSy9IVtjBuVeLr+HpOB4BQFWbg+UZaADMtTdGYI7Geitb76LTrr5QV/7Xg4ahLwYGYZzuHGZKM5ZAQ"
        }
    }
}

Given the following JSON object with data values in it:

{
    "one": 1,
    "two": "Two"
}

The JSON signing algorithm should emit the following signed JSON:

{
    "one": 1,
    "signatures": {
        "domain": {
            "ed25519:1": "KqmLSbO39/Bzb0QIYE82zqLwsA+PDzYIpIRA2sRQ4sL53+sN6/fpNSoqE7BP7vBZhG6kYdD13EIMJpvhJI+6Bw"
        }
    },
    "two": "Two"
}

7.3   Event Signing

Given the following minimally-sized event:

{
    "event_id": "$0:domain",
    "origin": "domain",
    "origin_server_ts": 1000000,
    "signatures": {},
    "type": "X",
    "unsigned": {
        "age_ts": 1000000
    }
}

The event signing algorithm should emit the following signed event:

{
    "event_id": "$0:domain",
    "hashes": {
        "sha256": "6tJjLpXtggfke8UxFhAKg82QVkJzvKOVOOSjUDK4ZSI"
    },
    "origin": "domain",
    "origin_server_ts": 1000000,
    "signatures": {
        "domain": {
            "ed25519:1": "2Wptgo4CwmLo/Y8B8qinxApKaCkBG2fjTWB7AbP5Uy+aIbygsSdLOFzvdDjww8zUVKCmI02eP9xtyJxc/cLiBA"
        }
    },
    "type": "X",
    "unsigned": {
        "age_ts": 1000000
    }
}

Given the following event containing redactable content:

{
    "content": {
        "body": "Here is the message content",
    },
    "event_id": "$0:domain",
    "origin": "domain",
    "origin_server_ts": 1000000,
    "type": "m.room.message",
    "room_id": "!r:domain",
    "sender": "@u:domain",
    "signatures": {},
    "unsigned": {
        "age_ts": 1000000
    }
}

The event signing algorithm should emit the following signed event:

{
    "content": {
        "body": "Here is the message content",
    },
    "event_id": "$0:domain",
    "hashes": {
        "sha256": "onLKD1bGljeBWQhWZ1kaP9SorVmRQNdN5aM2JYU2n/g"
    },
    "origin": "domain",
    "origin_server_ts": 1000000,
    "type": "m.room.message",
    "room_id": "!r:domain",
    "sender": "@u:domain",
    "signatures": {
        "domain": {
            "ed25519:1": "Wm+VzmOUOz08Ds+0NTWb1d4CZrVsJSikkeRxh6aCcUwu6pNC78FunoD7KNWzqFn241eYHYMGCA5McEiVPdhzBA"
        }
    },
    "unsigned": {
        "age_ts": 1000000
    }
}