Awesome

envelope spec | version 1.0.0

This is a spec for encrypting messages to groups of people. Initially it will support communication for large groups which share a public key (secret key cryptography / symmetric keys), but it has also been designed to support forward-secure secret-key cryptography (a little like Signal's double-ratchet).

envelope assumes each message is part of append-only chain (with a unique feed_id), made up of backlinked messages such that each message has a unique previous message with a unique id (prev_msg_id)

Anatomy

After boxing, a complete envelope message looks like this:

 +---------------------------------------+
 | ╔═══════════════════════════════════╗ |
 | ║            header_box             ║ |
 | ╚═══════════════════════════════════╝ |
 | ┌───────────────────────────────────┐ |
 | │            key_slot_1             │ |
 | ├───────────────────────────────────┤ |
 | │            key_slot_2             │ |
 | ├───────────────────────────────────┤ |
 | │               ...                 │ |
 | ├───────────────────────────────────┤ |
 | │            key_slot_n             │ |
 | └───────────────────────────────────┘ |
 | ╔═══════════════════════════════════╗ |
 | ║           extensions              ║ |
 | ║                                   ║ |
 | ╚═══════════════════════════════════╝ |
 | ╔═══════════════════════════════════╗ |
 | ║             body_box              ║ |
 | ║                                   ║ |
 | ║                                   ║ |
 | ║                                   ║ |
 | ║                                   ║ |
 | ║                         ╔═════════╝ |
 | ╚═════════════════════════╝           |
 +---------------------------------------+

header_box

A secretbox (refering to libsodium crypto_secretbox_easy), which describes the layout and configuration of the message to follow. Being able to decrypt this is required for being able to unbox the rest of the message.

 ╔═════════════════════════════════╗
 ║            header_box           ║
 ╚═════════════════════════════════╝ 
 |                32               |
 |                                 |

 ┌─────────────────┬───────────────┐
 │       HMAC      │    header*    │
 └─────────────────┴───────────────┘
         16       /       16        \
                 /                   \
                /                     \
               /                       \
              /                         \
             /                           \
   
    +----------------+-------+-------------------- ---+
    | offset         | flags | header_extensions      |
    +----------------+-------+-------------------- ---+
             2           1              13

HMAC - 16 bytes which allows authentication of the integrity of header*
header* - the header encrypted with header_key + zerod nonce
offset - 2 bytes which desribe the offset of the start of body_box in bytes
flags - 1 byte where each bit describes which extensions are active (if any)
header_extensions - 13 bytes for configuration of extensions

key_slot_n

Each of these slots is like a 'safety deposit box' which contains a copy of the top-level msg_key which allows decryption of everything in the message.

The slots contents are defined by

slot_content = xor(
  msg_key,
  Derive(recipient_key, ["slot_key", key_mgmt_scheme], 32)
)

Where

Derive is the same derivation function defined here
recipient_key is one of the shared keys you're encrypting to could be:
- a private key for a group (symmetric key)
- a double-ratchet derived key for an individual (this option requires more info in the header_extensions + extensions)
key_mgmt_scheme is the type of recipient_key, specifically what sort of key management it's involved in, e.g. :
- "envelope-large-symmetric-group"
- "envelope-id-based-dm-converted-ed25519"
- "envelope-signed-dh-key-curve25519"

Note these slots have no HMAC. This is because if you successfully extract msg_key from one of these slots you can immediately confirm if you can decrypt the header_box, which has an HMAC, which will confirm whether you have the correct key

extensions

...WIP

This is where things like keys for double-ratchet-like communication will go. This section might also contain padding.

body_box

The section which contains the plaintext which we've boxed.

 ╔═════════════════════════════════╗
 ║             body_box            ║
 ║                                 ║
 ║                                 ║
 ║                                 ║
 ║                                 ║
 ║                         ╔═══════╝
 ╚═════════════════════════╝
 |              >=16               |
 |                                 |

 ┌─────────────────┬───────────────┐
 │       HMAC      │               │
 ├─────────────────┘               │
 │                                 │
 │             body*               │
 │                                 │
 │                         ┌───────┘
 └─────────────────────────┘

HMAC - 16 bytes which allows authentication of the integrity of body*
body* - the body encrypted with body_key and a zerod nonce

Unboxing algorithm

When you receive a envelope message, the only things you know are:

the length of the whole box (doesn't tell you much, as there may be padding)
where the key-slots start (because the header_box is exactly 32 bytes)
where this message was posted (we call this it's "context", and the boxing is bound to this)
- which feed_id posted it
- what the prev_msg_id was (i.e. what was the message before it in this feed_ids chain?)

So starting after the header_box (32 bytes in), we lift out successive chunks of 32 bytes (the size of a key_slot) chunks and try and decrypt them.

The way we know if a key_slot has yielded us a valid key for the message is by trying to see if the "key" we've derived from a slot helps us decrypt the header_box. This works because the header_box has an HMAC, which is an authentication code which allows us know know if our decryption is valid.

If the first slot doesn't yield a valid key, we move to the second slot (starting 32 + 32 bytes into the box), and check the next slot. We either try incrementing through the whole box till we succeed (or reach the end), OR we set a "max depth" we want to try (e.g. if we think there will not be more than 10 slots, we can quit after (32 + 10 * 32 bytes).

Once we have the msg_key, we can decrypt the header_box. This reveals offset - the position of the start of the body_box in bytes. This allows us to proceed to decrypt the body of the original message.

Futher detail:

different keys are used to decrypt header_box, extensions, body_box, but they are all derived deterministically from msg_key
they are all encrypted with "zerod nonces", as the keys used for each are absolutely specific to the context (context = feed_id, prev_msg_id and msg_key)

Design

Original notes from a week long design session Dominic + Keks did. (scuttlebutt: %39f9I0e4bEln+yy6850joHRTqmEQfUyxssv54UANNuk=.sha256)

Key derivation

Keys are derived from msg_key as follows

msg_key
  │
  ├──> read_key = Derive(msg_key, "read_key", 32)
  │      │
  │      ├──> header_key = Derive(read_key, "header_key", 32)
  │      │
  │      └──> body_key = Derive(read_key, "body_key", 32)
  │
  └──> extensions = Derive(msg_key, "extentions", 32)
         │
         └──> TODO

Where the Derive function is defined here

msg_key is the symmetric key that is encrypted to each recipient or group. When entrusting the message, instead of sharing the msg_key instead the message read_key is shared. this gives access to header metadata and body but not ephemeral keys.