Network Working Group R. Barnes
Internet-Draft Cisco
Intended status: Informational J. Millican
Expires: November 3, 2019 Facebook
E. Omara
Google
K. Cohn-Gordon
University of Oxford
R. Robert
Wire
May 02, 2019
The Messaging Layer Security (MLS) Protocol
draft-ietf-mls-protocol-05
Abstract
Messaging applications are increasingly making use of end-to-end
security mechanisms to ensure that messages are only accessible to
the communicating endpoints, and not to any servers involved in
delivering messages. Establishing keys to provide such protections
is challenging for group chat settings, in which more than two
clients need to agree on a key but may not be online at the same
time. In this document, we specify a key establishment protocol that
provides efficient asynchronous group key establishment with forward
secrecy and post-compromise security for groups in size ranging from
two to thousands.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on November 3, 2019.
Barnes, et al. Expires November 3, 2019 [Page 1]
Internet-Draft MLS May 2019
Copyright Notice
Copyright (c) 2019 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(https://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1. Change Log . . . . . . . . . . . . . . . . . . . . . . . 4
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 5
3. Basic Assumptions . . . . . . . . . . . . . . . . . . . . . . 6
4. Protocol Overview . . . . . . . . . . . . . . . . . . . . . . 7
5. Ratchet Trees . . . . . . . . . . . . . . . . . . . . . . . . 10
5.1. Tree Computation Terminology . . . . . . . . . . . . . . 10
5.2. Ratchet Tree Nodes . . . . . . . . . . . . . . . . . . . 13
5.3. Views of a Ratchet Tree . . . . . . . . . . . . . . . . . 14
5.4. Ratchet Tree Updates . . . . . . . . . . . . . . . . . . 15
5.5. Synchronizing Views of the Tree . . . . . . . . . . . . . 16
6. Cryptographic Objects . . . . . . . . . . . . . . . . . . . . 17
6.1. Curve25519, SHA-256, and AES-128-GCM . . . . . . . . . . 18
6.1.1. P-256, SHA-256, and AES-128-GCM . . . . . . . . . . . 18
6.2. Credentials . . . . . . . . . . . . . . . . . . . . . . . 19
6.3. Tree Hashes . . . . . . . . . . . . . . . . . . . . . . . 20
6.4. Group State . . . . . . . . . . . . . . . . . . . . . . . 21
6.5. Direct Paths . . . . . . . . . . . . . . . . . . . . . . 23
6.6. Key Schedule . . . . . . . . . . . . . . . . . . . . . . 23
6.7. Encryption Keys . . . . . . . . . . . . . . . . . . . . . 25
7. Initialization Keys . . . . . . . . . . . . . . . . . . . . . 27
8. Message Framing . . . . . . . . . . . . . . . . . . . . . . . 28
8.1. Metadata Encryption . . . . . . . . . . . . . . . . . . . 30
8.2. Content Signing and Encryption . . . . . . . . . . . . . 31
9. Handshake Messages . . . . . . . . . . . . . . . . . . . . . 31
9.1. Init . . . . . . . . . . . . . . . . . . . . . . . . . . 33
9.2. Add . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
9.3. Update . . . . . . . . . . . . . . . . . . . . . . . . . 36
9.4. Remove . . . . . . . . . . . . . . . . . . . . . . . . . 37
10. Sequencing of State Changes . . . . . . . . . . . . . . . . . 37
10.1. Server-Enforced Ordering . . . . . . . . . . . . . . . . 38
Barnes, et al. Expires November 3, 2019 [Page 2]
Internet-Draft MLS May 2019
10.2. Client-Enforced Ordering . . . . . . . . . . . . . . . . 39
10.3. Merging Updates . . . . . . . . . . . . . . . . . . . . 39
11. Application Messages . . . . . . . . . . . . . . . . . . . . 40
11.1. Message Encryption and Decryption . . . . . . . . . . . 41
11.2. Delayed and Reordered Application messages . . . . . . . 42
12. Security Considerations . . . . . . . . . . . . . . . . . . . 42
12.1. Confidentiality of the Group Secrets . . . . . . . . . . 43
12.2. Authentication . . . . . . . . . . . . . . . . . . . . . 43
12.3. Forward and post-compromise security . . . . . . . . . . 43
12.4. Init Key Reuse . . . . . . . . . . . . . . . . . . . . . 44
13. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 44
14. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 44
15. References . . . . . . . . . . . . . . . . . . . . . . . . . 45
15.1. Normative References . . . . . . . . . . . . . . . . . . 45
15.2. Informative References . . . . . . . . . . . . . . . . . 46
Appendix A. Tree Math . . . . . . . . . . . . . . . . . . . . . 47
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 50
1. Introduction
DISCLAIMER: This is a work-in-progress draft of MLS and has not yet
seen significant security analysis. It should not be used as a basis
for building production systems.
RFC EDITOR: PLEASE REMOVE THE FOLLOWING PARAGRAPH The source for this
draft is maintained in GitHub. Suggested changes should be submitted
as pull requests at https://github.com/mlswg/mls-protocol.
Instructions are on that page as well. Editorial changes can be
managed in GitHub, but any substantive change should be discussed on
the MLS mailing list.
A group of users who want to send each other encrypted messages needs
a way to derive shared symmetric encryption keys. For two parties,
this problem has been studied thoroughly, with the Double Ratchet
emerging as a common solution [doubleratchet] [signal]. Channels
implementing the Double Ratchet enjoy fine-grained forward secrecy as
well as post-compromise security, but are nonetheless efficient
enough for heavy use over low-bandwidth networks.
For a group of size greater than two, a common strategy is to
unilaterally broadcast symmetric "sender" keys over existing shared
symmetric channels, and then for each member to send messages to the
group encrypted with their own sender key. Unfortunately, while this
improves efficiency over pairwise broadcast of individual messages
and provides forward secrecy (with the addition of a hash ratchet),
it is difficult to achieve post-compromise security with sender keys.
An adversary who learns a sender key can often indefinitely and
passively eavesdrop on that member's messages. Generating and
Barnes, et al. Expires November 3, 2019 [Page 3]
Internet-Draft MLS May 2019
distributing a new sender key provides a form of post-compromise
security with regard to that sender. However, it requires
computation and communications resources that scale linearly with the
size of the group.
In this document, we describe a protocol based on tree structures
that enable asynchronous group keying with forward secrecy and post-
compromise security. Based on earlier work on "asynchronous
ratcheting trees" [art], the mechanism presented here use a
asynchronous key-encapsulation mechanism for tree structures. This
mechanism allows the members of the group to derive and update shared
keys with costs that scale as the log of the group size.
1.1. Change Log
RFC EDITOR PLEASE DELETE THIS SECTION.
draft-05
o Common framing for handshake and application messages (*)
o Handshake message encryption (*)
o Convert from literal state to a commitment via the "tree hash" (*)
o Add credentials to the tree and remove the "roster" concept (*)
o Remove the secret field from tree node values
draft-04
o Updating the language to be similar to the Architecture document
o ECIES is now renamed in favor of HPKE (*)
o Using a KDF instead of a Hash in TreeKEM (*)
draft-03
o Added ciphersuites and signature schemes (*)
o Re-ordered fields in UserInitKey to make parsing easier (*)
o Fixed inconsistencies between Welcome and GroupState (*)
o Added encryption of the Welcome message (*)
draft-02
Barnes, et al. Expires November 3, 2019 [Page 4]
Internet-Draft MLS May 2019
o Removed ART (*)
o Allowed partial trees to avoid double-joins (*)
o Added explicit key confirmation (*)
draft-01
o Initial description of the Message Protection mechanism. (*)
o Initial specification proposal for the Application Key Schedule
using the per-participant chaining of the Application Secret
design. (*)
o Initial specification proposal for an encryption mechanism to
protect Application Messages using an AEAD scheme. (*)
o Initial specification proposal for an authentication mechanism of
Application Messages using signatures. (*)
o Initial specification proposal for a padding mechanism to
improving protection of Application Messages against traffic
analysis. (*)
o Inversion of the Group Init Add and Application Secret derivations
in the Handshake Key Schedule to be ease chaining in case we
switch design. (*)
o Removal of the UserAdd construct and split of GroupAdd into Add
and Welcome messages (*)
o Initial proposal for authenticating handshake messages by signing
over group state and including group state in the key schedule (*)
o Added an appendix with example code for tree math
o Changed the ECIES mechanism used by TreeKEM so that it uses nonces
generated from the shared secret
draft-00
o Initial adoption of draft-barnes-mls-protocol-01 as a WG item.
2. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in BCP
Barnes, et al. Expires November 3, 2019 [Page 5]
Internet-Draft MLS May 2019
14 [RFC2119] [RFC8174] when, and only when, they appear in all
capitals, as shown here.
Client: An agent that uses this protocol to establish shared
cryptographic state with other clients. A client is defined by
the cryptographic keys it holds. An application or user may use
one client per device (keeping keys local to each device) or sync
keys among a user's devices so that each user appears as a single
client.
Group: A collection of clients with shared cryptographic state.
Member: A client that is included in the shared state of a group,
hence has access to the group's secrets.
Initialization Key: A short-lived HPKE key pair used to introduce a
new client to a group. Initialization keys are published for each
client (UserInitKey).
Leaf Key: A secret that represent a member's contribution to the
group secret (so called because the members' leaf keys are the
leaves in the group's ratchet tree).
Identity Key: A long-lived signing key pair used to authenticate the
sender of a message.
Terminology specific to tree computations is described in Section 5.
We use the TLS presentation language [RFC8446] to describe the
structure of protocol messages.
3. Basic Assumptions
This protocol is designed to execute in the context of a Messaging
Service (MS) as described in [I-D.ietf-mls-architecture]. In
particular, we assume the MS provides the following services:
o A long-term identity key provider which allows clients to
authenticate protocol messages in a group. These keys MUST be
kept for the lifetime of the group as there is no mechanism in the
protocol for changing a client's identity key.
o A broadcast channel, for each group, which will relay a message to
all members of a group. For the most part, we assume that this
channel delivers messages in the same order to all participants.
(See Section 10 for further considerations.)
Barnes, et al. Expires November 3, 2019 [Page 6]
Internet-Draft MLS May 2019
o A directory to which clients can publish initialization keys and
download initialization keys for other participants.
4. Protocol Overview
The goal of this protocol is to allow a group of clients to exchange
confidential and authenticated messages. It does so by deriving a
sequence of secrets and keys known only to members. Those should be
secret against an active network adversary and should have both
forward and post-compromise secrecy with respect to compromise of a
participant.
We describe the information stored by each client as a _state_, which
includes both public and private data. An initial state, including
an initial set of clients, is set up by a group creator using the
_Init_ algorithm and based on information pre-published by clients.
The creator sends the _Init_ message to the clients, who can then set
up their own group state and derive the same shared secret. Clients
then exchange messages to produce new shared states which are
causally linked to their predecessors, forming a logical Directed
Acyclic Graph (DAG) of states. Members can send _Update_ messages
for post-compromise secrecy and new clients can be added or existing
members removed from the group.
The protocol algorithms we specify here follow. Each algorithm
specifies both (i) how a client performs the operation and (ii) how
other clients update their state based on it.
There are four major operations in the lifecycle of a group:
o Adding a member, initiated by a current member;
o Updating the leaf secret of a member;
o Removing a member.
Before the initialization of a group, clients publish UserInitKey
objects to a directory provided to the Messaging Service.
Barnes, et al. Expires November 3, 2019 [Page 7]
Internet-Draft MLS May 2019
Group
A B C Directory Channel
| | | | |
| UserInitKeyA | | | |
|------------------------------------------->| |
| | | | |
| | UserInitKeyB | | |
| |---------------------------->| |
| | | | |
| | | UserInitKeyC | |
| | |------------->| |
| | | | |
When a client A wants to establish a group with B and C, it first
downloads UserInitKeys for B and C. It then initializes a group
state containing only itself and uses the UserInitKeys to compute
Welcome and Add messages to add B and C, in a sequence chosen by A.
The Welcome messages are sent directly to the new members (there is
no need to send them to the group). The Add messages are broadcasted
to the group, and processed in sequence by B and C. Messages
received before a client has joined the group are ignored. Only
after A has received its Add messages back from the server does it
update its state to reflect their addition.
Barnes, et al. Expires November 3, 2019 [Page 8]
Internet-Draft MLS May 2019
Group
A B C Directory Channel
| | | | |
| UserInitKeyB, UserInitKeyC | |
|<-------------------------------------------| |
|state.init() | | | |
| | | | |
| | | | Add(A->AB) |
|--------------------------------------------------------------->|
| | | | |
| Welcome(B) | | | |
|------------->|state.init() | | |
| | | | |
| | | | Add(A->AB) |
|<---------------------------------------------------------------|
|state.add(B) |<------------------------------------------------|
| |state.join() | | |
| | | | |
| | | | Add(AB->ABC) |
|--------------------------------------------------------------->|
| | | | |
| | Welcome(C) | | |
|---------------------------->|state.init() | |
| | | | |
| | | | Add(AB->ABC) |
|<---------------------------------------------------------------|
|state.add(C) |<------------------------------------------------|
| |state.add(C) |<---------------------------------|
| | |state.join() | |
Subsequent additions of group members proceed in the same way. Any
member of the group can download an UserInitKey for a new client and
broadcast an Add message that the current group can use to update
their state and the new client can use to initialize its state.
To enforce forward secrecy and post-compromise security of messages,
each member periodically updates its leaf secret which represents its
contribution to the group secret. Any member of the group can send
an Update at any time by generating a fresh leaf secret and sending
an Update message that describes how to update the group secret with
that new information. Once all members have processed this message,
the group's secrets will be unknown to an attacker that had
compromised the sender's prior leaf secret.
It is left to the application to determine the interval of time
between Update messages. This policy could require a change for each
message, or it could require sending an update every week or more.
Barnes, et al. Expires November 3, 2019 [Page 9]
Internet-Draft MLS May 2019
Group
A B ... Z Directory Channel
| | | | |
| Update(A) | | | |
|---------------------------------------------------------->|
| | | | |
| | | | Update(A) |
|<----------------------------------------------------------|
|state.upd(A) |<-------------------------------------------|
| |state.upd(A) |<----------------------------|
| | |state.upd(A) | |
| | | | |
Members are removed from the group in a similar way, as an update is
effectively removing the old leaf from the group. Any member of the
group can generate a Remove message that adds new entropy to the
group state that is known to all members except the removed member.
After other participants have processed this message, the group's
secrets will be unknown to the removed participant. Note that this
does not necessarily imply that any member is actually allowed to
evict other members; groups can layer authentication-based access
control policies on top of these basic mechanism.
Group
A B ... Z Directory Channel
| | | | |
| | | Remove(B) | |
| | |---------------------------->|
| | | | |
| | | | Remove(B) |
|<----------------------------------------------------------|
|state.del(B) | |<----------------------------|
| | |state.del(B) | |
| | | | |
| | | | |
5. Ratchet Trees
The protocol uses "ratchet trees" for deriving shared secrets among a
group of clients.
5.1. Tree Computation Terminology
Trees consist of _nodes_. A node is a _leaf_ if it has no children,
and a _parent_ otherwise; note that all parents in our trees have
precisely two children, a _left_ child and a _right_ child. A node
is the _root_ of a tree if it has no parents, and _intermediate_ if
it has both children and parents. The _descendants_ of a node are
Barnes, et al. Expires November 3, 2019 [Page 10]
Internet-Draft MLS May 2019
that node, its children, and the descendants of its children, and we
say a tree _contains_ a node if that node is a descendant of the root
of the tree. Nodes are _siblings_ if they share the same parent.
A _subtree_ of a tree is the tree given by the descendants of any
node, the _head_ of the subtree. The _size_ of a tree or subtree is
the number of leaf nodes it contains. For a given parent node, its
_left subtree_ is the subtree with its left child as head
(respectively _right subtree_).
All trees used in this protocol are left-balanced binary trees. A
binary tree is _full_ (and _balanced_) if its size is a power of two
and for any parent node in the tree, its left and right subtrees have
the same size. If a subtree is full and it is not a subset of any
other full subtree, then it is _maximal_.
A binary tree is _left-balanced_ if for every parent, either the
parent is balanced, or the left subtree of that parent is the largest
full subtree that could be constructed from the leaves present in the
parent's own subtree. Note that given a list of "n" items, there is
a unique left-balanced binary tree structure with these elements as
leaves. In such a left-balanced tree, the "k-th" leaf node refers to
the "k-th" leaf node in the tree when counting from the left,
starting from 0.
The _direct path_ of a root is the empty list, and of any other node
is the concatenation of that node with the direct path of its parent.
The _copath_ of a node is the list of siblings of nodes in its direct
path. The _frontier_ of a tree is the list of heads of the maximal
full subtrees of the tree, ordered from left to right.
For example, in the below tree:
o The direct path of C is (C, CD, ABCD)
o The copath of C is (D, AB, EFG)
o The frontier of the tree is (ABCD, EF, G)
Barnes, et al. Expires November 3, 2019 [Page 11]
Internet-Draft MLS May 2019
ABCDEFG
/ \
/ \
/ \
ABCD EFG
/ \ / \
/ \ / \
AB CD EF |
/ \ / \ / \ |
A B C D E F G
1 1 1
0 1 2 3 4 5 6 7 8 9 0 1 2
Each node in the tree is assigned an _node index_, starting at zero
and running from left to right. A node is a leaf node if and only if
it has an even index. The node indices for the nodes in the above
tree are as follows:
o 0 = A
o 1 = AB
o 2 = B
o 3 = ABCD
o 4 = C
o 5 = CD
o 6 = D
o 7 = ABCDEFG
o 8 = E
o 9 = EF
o 10 = F
o 11 = EFG
o 12 = G
(Note that left-balanced binary trees are the same structure that is
used for the Merkle trees in the Certificate Transparency protocol
[I-D.ietf-trans-rfc6962-bis].)
Barnes, et al. Expires November 3, 2019 [Page 12]
Internet-Draft MLS May 2019
The leaves of the tree are indexed separately, using a _leaf index_,
since the protocol messages only need to refer to leaves in the tree.
Like nodes, leaves are numbered left to right. Note that given the
above numbering, a node is a leaf node if and only if it has an even
node index, and a leaf node's leaf index is half its node index. The
leaf indices in the above tree are as follows:
o 0 = A
o 1 = B
o 2 = C
o 3 = D
o 4 = E
o 5 = F
o 6 = G
5.2. Ratchet Tree Nodes
A particular instance of a ratchet tree is based on the following
cryptographic primitives, defined by the ciphersuite in use:
o An HPKE ciphersuite, which specifies a Key Encapsulation Method
(KEM), an AEAD encryption scheme, and a hash function
o A Derive-Key-Pair function that produces an asymmetric key pair
for the specified KEM from a symmetric secret, using the specified
hash function.
Each node in a ratchet tree contains up to three values:
o A private key (only within direct path, see below)
o A public key
o A credential (only for leaf nodes)
The conditions under which each of these values must or must not be
present are laid out in Section 5.3.
A node in the tree may also be _blank_, indicating that no value is
present at that node. The _resolution_ of a node is an ordered list
of non-blank nodes that collectively cover all non-blank descendants
Barnes, et al. Expires November 3, 2019 [Page 13]
Internet-Draft MLS May 2019
of the node. The nodes in a resolution are ordered according to
their indices.
o The resolution of a non-blank node is a one element list
containing the node itself
o The resolution of a blank leaf node is the empty list
o The resolution of a blank intermediate node is the result of
concatinating the resolution of its left child with the resolution
of its right child, in that order
For example, consider the following tree, where the "_" character
represents a blank node:
_
/ \
/ \
_ CD
/ \ / \
A _ C D
0 1 2 3 4 5 6
In this tree, we can see all three of the above rules in play:
o The resolution of node 5 is the list [CD]
o The resolution of node 2 is the empty list []
o The resolution of node 3 is the list [A, CD]
Every node, regardless of whether a node is blank or populated, has a
corresponding _hash_ that summarizes the contents of the subtree
below that node. The rules for computing these hashes are described
in Section 6.3.
5.3. Views of a Ratchet Tree
We generally assume that each participant maintains a complete and
up-to-date view of the public state of the group's ratchet tree,
including the public keys for all nodes and the credentials
associated with the leaf nodes.
No participant in an MLS group has full knowledge of the secret state
of the tree, i.e., private keys associated to the nodes. Instead,
each member is assigned to a leaf of the tree, which determines the
Barnes, et al. Expires November 3, 2019 [Page 14]
Internet-Draft MLS May 2019
set of secret state known to the member. The credential stored at
that leaf is one provided by the member.
In particular, MLS maintains the members' views of the tree in such a
way as to maintain the _tree invariant_:
The private key for a node in the tree is known to a member of
the group if and only if that member's leaf is a descendant of
the node or equal to it.
In other words, each member holds the private keys for nodes in its
direct path, and no others.
5.4. Ratchet Tree Updates
Nodes in a tree are always updated along the direct path from a leaf
to the root. The generator of the update chooses a random secret
value "path_secret[0]", and generates a sequence of "path secrets",
one for each node from the leaf to the root. That is, path_secret[0]
is used for the leaf, path_secret[1] for its parent, and so on. At
each step, the path secret is used to derive a new secret value for
the corresponding node, from which the node's key pair is derived.
path_secret[n] = HKDF-Expand-Label(path_secret[n-1],
"path", "", Hash.Length)
node_secret[n] = HKDF-Expand-Label(path_secret[n],
"node", "", Hash.Length)
node_priv[n], node_pub[n] = Derive-Key-Pair(node_secret[n])
For example, suppose there is a group with four members:
G
/ \
/ \
/ \
E F
/ \ / \
A B C D
If the second participant (B) subsequently generates an update based
on a secret X, then the sender would generate the following sequence
of path secrets and node secrets:
Barnes, et al. Expires November 3, 2019 [Page 15]
Internet-Draft MLS May 2019
path_secret[2] ---> node_secret[2]
^
|
path_secret[1] ---> node_secret[1]
^
|
X = path_secret[0] ---> node_secret[0]
After the update, the tree will have the following structure, where
"ns[i]" represents the node_secret values generated as described
above:
ns[2]
/ \
ns[1] F
/ \ / \
A ns[0] C D
5.5. Synchronizing Views of the Tree
The members of the group need to keep their views of the tree in sync
and up to date. When a client proposes a change to the tree (e.g.,
to add or remove a member), it transmits a handshake message
containing a set of public values for intermediate nodes in the
direct path of a leaf. The other members of the group can use these
public values to update their view of the tree, aligning their copy
of the tree to the sender's.
To perform an update for a leaf, the sender broadcasts to the group
the following information for each node in the direct path of the
leaf, as well as the root:
o The public key for the node
o Zero or more encrypted copies of the path secret corresponding to
the node
The path secret value for a given node is encrypted for the subtree
corresponding to the parent's non-updated child, i.e., the child on
the copath of the leaf node. There is one encrypted path secret for
each public key in the resolution of the non-updated child. In
particular, for the leaf node, there are no encrypted secrets, since
a leaf node has no children.
The recipient of an update processes it with the following steps:
1. Compute the updated path secrets * Identify a node in the direct
path for which the local member is in the subtree of the non-
Barnes, et al. Expires November 3, 2019 [Page 16]
Internet-Draft MLS May 2019
updated child * Identify a node in the resolution of the copath
node for which this node has a private key * Decrypt the path
secret for the parent of the copath node using the private key
from the resolution node * Derive path secrets for ancestors of
that node using the algorithm described above * The recipient
SHOULD verify that the received public keys agree with the public
keys derived from the new node_secret values
2. Merge the updated path secrets into the tree * Replace the public
keys for nodes on the direct path with the received public keys *
For nodes where an updated path secret was computed in step 1,
compute the corresponding node secret and node key pair and
replace the values stored at the node with the computed values.
For example, in order to communicate the example update described in
the previous section, the sender would transmit the following values:
+------------+----------------------------------+
| Public Key | Ciphertext(s) |
+------------+----------------------------------+
| pk(ns[2]) | E(pk(C), ps[2]), E(pk(D), ps[2]) |
| | |
| pk(ns[1]) | E(pk(A), ps[1]) |
| | |
| pk(ns[0]) | |
+------------+----------------------------------+
In this table, the value pk(X) represents the public key
corresponding derived from the node secret X. The value E(K, S)
represents the public-key encryption of the path secret S to the
public key K.
6. Cryptographic Objects
Each MLS session uses a single ciphersuite that specifies the
following primitives to be used in group key computations:
o A hash function
o A Diffie-Hellman finite-field group or elliptic curve
o An AEAD encryption algorithm [RFC5116]
The ciphersuite must also specify an algorithm "Derive-Key-Pair" that
maps octet strings with the same length as the output of the hash
function to key pairs for the asymmetric encryption scheme.
Barnes, et al. Expires November 3, 2019 [Page 17]
Internet-Draft MLS May 2019
Public keys used in the protocol are opaque values in a format
defined by the ciphersuite, using the following types:
opaque HPKEPublicKey<1..2^16-1>;
opaque SignaturePublicKey<1..2^16-1>;
Cryptographic algorithms are indicated using the following types:
enum {
ecdsa_secp256r1_sha256(0x0403),
ed25519(0x0807),
(0xFFFF)
} SignatureScheme;
enum {
P256_SHA256_AES128GCM(0x0000),
X25519_SHA256_AES128GCM(0x0001),
(0xFFFF)
} CipherSuite;
6.1. Curve25519, SHA-256, and AES-128-GCM
This ciphersuite uses the following primitives:
o Hash function: SHA-256
o Diffie-Hellman group: Curve25519 [RFC7748]
o AEAD: AES-128-GCM
Given an octet string X, the private key produced by the Derive-Key-
Pair operation is SHA-256(X). (Recall that any 32-octet string is a
valid Curve25519 private key.) The corresponding public key is
X25519(SHA-256(X), 9).
Implementations SHOULD use the approach specified in [RFC7748] to
calculate the Diffie-Hellman shared secret. Implementations MUST
check whether the computed Diffie-Hellman shared secret is the all-
zero value and abort if so, as described in Section 6 of [RFC7748].
If implementers use an alternative implementation of these elliptic
curves, they SHOULD perform the additional checks specified in
Section 7 of [RFC7748]
6.1.1. P-256, SHA-256, and AES-128-GCM
This ciphersuite uses the following primitives:
o Hash function: SHA-256
Barnes, et al. Expires November 3, 2019 [Page 18]
Internet-Draft MLS May 2019
o Diffie-Hellman group: secp256r1 (NIST P-256)
o AEAD: AES-128-GCM
Given an octet string X, the private key produced by the Derive-Key-
Pair operation is SHA-256(X), interpreted as a big-endian integer.
The corresponding public key is the result of multiplying the
standard P-256 base point by this integer.
P-256 ECDH calculations (including parameter and key generation as
well as the shared secret calculation) are performed according to
[IEEE1363] using the ECKAS-DH1 scheme with the identity map as key
derivation function (KDF), so that the shared secret is the
x-coordinate of the ECDH shared secret elliptic curve point
represented as an octet string. Note that this octet string (Z in
IEEE 1363 terminology) as output by FE2OSP, the Field Element to
Octet String Conversion Primitive, has constant length for any given
field; leading zeros found in this octet string MUST NOT be
truncated.
(Note that this use of the identity KDF is a technicality. The
complete picture is that ECDH is employed with a non-trivial KDF
because MLS does not directly use this secret for anything other than
for computing other secrets.)
Clients MUST validate remote public values by ensuring that the point
is a valid point on the elliptic curve. The appropriate validation
procedures are defined in Section 4.3.7 of [X962] and alternatively
in Section 5.6.2.3 of [keyagreement]. This process consists of three
steps: (1) verify that the value is not the point at infinity (O),
(2) verify that for Y = (x, y) both integers are in the correct
interval, (3) ensure that (x, y) is a correct solution to the
elliptic curve equation. For these curves, implementers do not need
to verify membership in the correct subgroup.
6.2. Credentials
A member of a group authenticates the identities of other
participants by means of credentials issued by some authentication
system, e.g., a PKI. Each type of credential MUST express the
following data:
o The public key of a signature key pair
o The identity of the holder of the private key
o The signature scheme that the holder will use to sign MLS messages
Barnes, et al. Expires November 3, 2019 [Page 19]
Internet-Draft MLS May 2019
Credentials MAY also include information that allows a relying party
to verify the identity / signing key binding.
enum {
basic(0),
x509(1),
(255)
} CredentialType;
struct {
opaque identity<0..2^16-1>;
SignatureScheme algorithm;
SignaturePublicKey public_key;
} BasicCredential;
struct {
CredentialType credential_type;
select (credential_type) {
case basic:
BasicCredential;
case x509:
opaque cert_data<1..2^24-1>;
};
} Credential;
6.3. Tree Hashes
To allow group members to verify that they agree on the cryptographic
state of the group, this section defines a scheme for generating a
hash value that represents the contents of the group's ratchet tree
and the members' credentials.
The hash of a tree is the hash of its root node, which we define
recursively, starting with the leaves. The hash of a leaf node is
the hash of a "LeafNodeHashInput" object:
Barnes, et al. Expires November 3, 2019 [Page 20]
Internet-Draft MLS May 2019
struct {
uint8 present;
switch (present) {
case 0: struct{};
case 1: T value;
}
} optional<T>;
struct {
HPKEPublicKey public_key;
Credential credential;
} LeafNodeInfo;
struct {
uint8 hash_type = 0;
optional<LeafNodeInfo> info;
} LeafNodeHashInput;
The "public_key" and "credential" fields represent the leaf public
key and the credential for the member holding that leaf,
respectively. The "info" field is equal to the null optional value
when the leaf is blank (i.e., no member occupies that leaf).
Likewise, the hash of a parent node (including the root) is the hash
of a "ParentNodeHashInput" struct:
struct {
uint8 hash_type = 1;
optional<HPKEPublicKey> public_key;
opaque left_hash<0..255>;
opaque right_hash<0..255>;
} ParentNodeHashInput
The "left_hash" and "right_hash" fields hold the hashes of the node's
left and right children, respectively. The "public_key" field holds
the hash of the public key stored at this node, represented as an
"optional<HPKEPublicKey>" object, which is null if and only if the
node is blank.
6.4. Group State
Each member of the group maintains a representation of the state of
the group:
Barnes, et al. Expires November 3, 2019 [Page 21]
Internet-Draft MLS May 2019
struct {
opaque group_id<0..255>;
uint32 epoch;
opaque tree_hash<0..255>;
opaque transcript_hash<0..255>;
} GroupState;
The fields in this state have the following semantics:
o The "group_id" field is an application-defined identifier for the
group.
o The "epoch" field represents the current version of the group key.
o The "tree_hash" field contains a commitment to the contents of the
group's rachet tree and the credentials for the members of the
group, as described in Section 6.3.
o The "transcript_hash" field contains the list of "GroupOperation"
messages that led to this state.
When a new member is added to the group, an existing member of the
group provides the new member with a Welcome message. The Welcome
message provides the information the new member needs to initialize
its GroupState.
Different group operations will have different effects on the group
state. These effects are described in their respective subsections
of Section 9. The following rules apply to all operations:
o The "group_id" field is constant
o The "epoch" field increments by one for each GroupOperation that
is processed
o The "tree_hash" is updated to represent the current tree and
credentials
o The "transcript_hash" is updated by a GroupOperation message
"operation" in the following way:
transcript_hash_[n] = Hash(transcript_hash_[n-1] || operation)
When a new one-member group is created (which requires no
GroupOperation), the "transcript_hash" field is set to an all-zero
vector of length Hash.length, where the Hash algorithm is defined by
the ciphersuite.
Barnes, et al. Expires November 3, 2019 [Page 22]
Internet-Draft MLS May 2019
6.5. Direct Paths
As described in Section 5.4, each MLS message needs to transmit node
values along the direct path of a leaf. The path contains a public
key for the leaf node, and a public key and encrypted secret value
for intermediate nodes in the path. In both cases, the path is
ordered from the leaf to the root; each node MUST be the parent of
its predecessor.
struct {
HPKEPublicKey ephemeral_key;
opaque ciphertext<0..2^16-1>;
} HPKECiphertext;
struct {
HPKEPublicKey public_key;
HPKECiphertext encrypted_path_secrets<0..2^16-1>;
} DirectPathNode;
struct {
DirectPathNode nodes<0..2^16-1>;
} DirectPath;
The length of the "node\_secrets" vector MUST be zero for the first
node in the path. For the remaining elements in the vector, the
number of ciphertexts in the "node\_secrets" vector MUST be equal to
the length of the resolution of the corresponding copath node. Each
ciphertext in the list is the encryption to the corresponding node in
the resolution.
The HPKECiphertext values are computed according to the Encrypt
function defined in [I-D.barnes-cfrg-hpke].
Decryption is performed in the corresponding way, using the private
key of the resolution node and the ephemeral public key transmitted
in the message.
6.6. Key Schedule
Group keys are derived using the HKDF-Extract and HKDF-Expand
functions as defined in [RFC5869], as well as the functions defined
below:
Barnes, et al. Expires November 3, 2019 [Page 23]
Internet-Draft MLS May 2019
HKDF-Expand-Label(Secret, Label, Context, Length) =
HKDF-Expand(Secret, HkdfLabel, Length)
Where HkdfLabel is specified as:
struct {
uint16 length = Length;
opaque label<7..255> = "mls10 " + Label;
opaque context<0..2^32-1> = Context;
} HkdfLabel;
Derive-Secret(Secret, Label, Context) =
HKDF-Expand-Label(Secret, Label, Hash(Context), Hash.length)
The Hash function used by HKDF is the ciphersuite hash algorithm.
Hash.length is its output length in bytes. In the below diagram:
o HKDF-Extract takes its salt argument from the top and its IKM
argument from the left
o Derive-Secret takes its Secret argument from the incoming arrow
When processing a handshake message, a client combines the following
information to derive new epoch secrets:
o The init secret from the previous epoch
o The update secret for the current epoch
o The GroupState object for current epoch
Given these inputs, the derivation of secrets for an epoch proceeds
as shown in the following diagram:
Barnes, et al. Expires November 3, 2019 [Page 24]
Internet-Draft MLS May 2019
init_secret_[n-1] (or 0)
|
V
update_secret -> HKDF-Extract = epoch_secret
|
+--> Derive-Secret(., "sender data", GroupState_[n])
| = sender_data_secret
|
+--> Derive-Secret(., "handshake", GroupState_[n])
| = handshake_secret
|
+--> Derive-Secret(., "app", GroupState_[n])
| = application_secret
|
+--> Derive-Secret(., "confirm", GroupState_[n])
| = confirmation_key
|
V
Derive-Secret(., "init", GroupState_[n])
|
V
init_secret_[n]
6.7. Encryption Keys
As described in Section 8, MLS encrypts three different types of
information:
o Metadata (sender information)
o Handshake messages
o Application messages
The sender information used to look up the key for the content
encryption is encrypted under AEAD using a random nonce and the
sender_data_key which is derived from the sender_data_secret as
follows:
sender_data_key =
HKDF-Expand-Label(sender_data_secret, "sd key", "", key_length)
Each handshake message is encrypted using a key and a nonce derived
from the handshake_secret for a specific sender to prevent two
senders to perform in the following way:
Barnes, et al. Expires November 3, 2019 [Page 25]
Internet-Draft MLS May 2019
handshake_nonce_[sender] =
HKDF-Expand-Label(handshake_secret, "hs nonce", [sender], nonce_length)
handshake_key_[sender] =
HKDF-Expand-Label(handshake_secret, "hs key", [sender], key_length)
Here the value [sender] represents the index of the member that will
use this key to send, encoded as a uint32.
For application messages, a chain of keys is derived for each sender
in a similar fashion. This allows forward secrecy at the level of
application messages within and out of an epoch. A step in this
chain (the second subscript) is called a "generation".
application_secret
|
V
HKDF-Expand-Label(., "app sender", [sender], Hash.length)
|
V
application_secret_[sender]_[0]
|
...
|
V
application_secret_[sender]_[N-1]
|
+--> HKDF-Expand-Label(.,"nonce", "", nonce_length)
| = write_nonce_[sender]_[N-1]
|
+--> HKDF-Expand-Label(.,"key", "", key_length)
| = write_key_[sender]_[N-1]
V
HKDF-Expand-Label(., "app sender", [sender], Hash.length)
|
V
application_secret_[sender]_[N]
As before the value [sender] represents the index of the member that
will use this key to send, encoded as a uint32.
[[ OPEN ISSUE: The HKDF context field is left empty for now. A
proper security study is needed to make sure that we do not need more
information in the context to achieve the security goals.]]
[[ OPEN ISSUE: At the moment there is no contributivity of
Application secrets chained from the initial one to the next
generation of Epoch secret. While this seems safe because
Barnes, et al. Expires November 3, 2019 [Page 26]
Internet-Draft MLS May 2019
cryptographic operations using the application secrets can't affect
the group init_secret, it remains to be proven correct. ]]
The following rules apply to the usage of the secrets, keys, and
nonces derived above:
o Senders MUST only use a given secret once and monotonically
increment the generation of their secret. This is important to
provide Forward Secrecy at the level of Application messages. An
attacker getting hold of a member specific Application Secret at
generation [N+1] will not be able to derive the member's
Application Secret [N] nor the associated AEAD key and nonce.
o Receivers MUST delete an Application Secret once it has been used
to derive the corresponding AEAD key and nonce as well as the next
Application Secret. Receivers MAY keep the AEAD key and nonce
around for some reasonable period.
o Receivers MUST delete AEAD keys and nonces once they have been
used to successfully decrypt a message.
7. Initialization Keys
In order to facilitate asynchronous addition of clients to a group,
it is possible to pre-publish initialization keys that provide some
public information about a user. UserInitKey messages provide
information about a client that any existing member can use to add
this client to the group asynchronously.
A UserInitKey object specifies what ciphersuites a client supports,
as well as providing public keys that the client can use for key
derivation and signing. The client's identity key is intended to be
stable throughout the lifetime of the group; there is no mechanism to
change it. Init keys are intended to be used a very limited number
of times, potentially once. (see Section 12.4). UserInitKeys also
contain an identifier chosen by the client, which the client MUST
assure uniquely identifies a given UserInitKey object among the set
of UserInitKeys created by this client.
The init_keys array MUST have the same length as the cipher_suites
array, and each entry in the init_keys array MUST be a public key for
the asymmetric encryption scheme defined in the cipher_suites array
and used in the HPKE construction for TreeKEM.
The whole structure is signed using the client's identity key. A
UserInitKey object with an invalid signature field MUST be considered
malformed. The input to the signature computation comprises all of
the fields except for the signature field.
Barnes, et al. Expires November 3, 2019 [Page 27]
Internet-Draft MLS May 2019
uint8 ProtocolVersion;
struct {
opaque user_init_key_id<0..255>;
ProtocolVersion supported_versions<0..255>;
CipherSuite cipher_suites<0..255>;
HPKEPublicKey init_keys<1..2^16-1>;
Credential credential;
opaque signature<0..2^16-1>;
} UserInitKey;
8. Message Framing
Handshake and application messages use a common framing structure.
This framing provides encryption to assure confidentiality within the
group, as well as signing to authenticate the sender within the
group.
The two main structures involved are MLSPlaintext and MLSCiphertext.
MLSCiphertext represents a signed and encrypted message, with
protections for both the content of the message and related metadata.
MLSPlaintext represents a message that is only signed, and not
encrypted. Applications SHOULD use MLSCiphertext to encode both
application and handshake messages, but MAY transmit handshake
messages encoded as MLSPlaintext objects in cases where it is
necessary for the delivery service to examine such messages.
Barnes, et al. Expires November 3, 2019 [Page 28]
Internet-Draft MLS May 2019
enum {
invalid(0),
handshake(1),
application(2),
(255)
} ContentType;
struct {
opaque group_id<0..255>;
uint32 epoch;
uint32 sender;
ContentType content_type;
select (MLSPlaintext.content_type) {
case handshake:
GroupOperation operation;
case application:
opaque application_data<0..2^32-1>;
}
opaque signature<0..2^16-1>;
} MLSPlaintext;
struct {
opaque group_id<0..255>;
uint32 epoch;
ContentType content_type;
opaque sender_data_nonce<0..255>;
opaque encrypted_sender_data<0..255>;
opaque ciphertext<0..2^32-1>;
} MLSCiphertext;
The remainder of this section describe how to compute the signature
of an MLSPlaintext object and how to convert it to an MLSCiphertext
object. The overall process is as follows:
o Gather the required metadata:
* Group ID
* Epoch
* Content Type
* Nonce
* Sender index
Barnes, et al. Expires November 3, 2019 [Page 29]
Internet-Draft MLS May 2019
* Key generation
o Sign the protected content and metadata
o Encrypt the sender information using the random nonce and the key
derived from the sender_data_secret
o Encrypt the content using a content encryption key identified by
the metadata
The group identifier, epoch and content_type fields are copied from
the MLSPlaintext object directly. The content encryption process
populates the ciphertext field of the MLSCiphertext object. The
metadata encryption step populates the encrypted_sender_data field.
Decryption follows the same step in reverse: Decrypt the metadata,
then the message and verify the content signature.
8.1. Metadata Encryption
The "sender data" used to look up the key for the content encryption
is encrypted under AEAD using the MLSCiphertext sender_data_nonce and
the sender_data_key from the keyschedule. It is encoded as an object
of the following form:
struct {
uint32 sender;
uint32 generation;
} MLSSenderData;
The Additional Authenticated Data (AAD) for the SenderData ciphertext
computation is its prefix in the MLSCiphertext, namely:
struct {
opaque group_id<0..255>;
uint32 epoch;
ContentType content_type;
opaque sender_data_nonce<0..255>;
} MLSCiphertextSenderDataAAD;
When parsing a SenderData struct as part of message decryption, the
recipient MUST verify that the sender field represents an occupied
leaf in the ratchet tree. In particular, the sender index value MUST
be less than the number of leaves in the tree.
Barnes, et al. Expires November 3, 2019 [Page 30]
Internet-Draft MLS May 2019
8.2. Content Signing and Encryption
The signature field in an MLSPlaintext object is computed using the
signing private key corresponding to the credential at the leaf in
the tree indicated by the sender field. The signature covers the
metadata and message content, with the signature field truncated.
The ciphertext field of the MLSCiphertext object is produced by
supplying the inputs described below to the AEAD function specified
by the ciphersuite in use. The plaintext input contains content and
signature of the MLSPlaintext, plus optional padding. These values
are encoded in the following form:
struct {
opaque content[length\_of\_content];
uint8 signature[MLSInnerPlaintext.sig_len];
uint16 sig_len;
uint8 marker = 1;
uint8 zero\_padding[length\_of\_padding];
} MLSCiphertextContent;
The key and nonce used for the encryption of the message depend on
the content type of the message. The sender chooses the handshake
key for a handshake message or an ununsed generation from its (per-
sender) application key chain for the current epoch, according to the
type of message being encrypted.
The Additional Authenticated Data (AAD) input to the encryption
contains an object of the following form, with the values used to
identify the key and nonce:
struct {
opaque group_id<0..255>;
uint32 epoch;
ContentType content_type;
opaque sender_data_nonce<0..255>;
opaque encrypted_sender_data<0..255>;
} MLSCiphertextContentAAD;
The ciphertext field of the MLSCiphertext object is produced by
supplying these inputs to the AEAD function specified by the
ciphersuite in use.
9. Handshake Messages
Over the lifetime of a group, its state will change for:
o Group initialization
Barnes, et al. Expires November 3, 2019 [Page 31]
Internet-Draft MLS May 2019
o A member adding a new client
o A member updating its leaf key
o A member deleting another member
In MLS, these changes are accomplished by broadcasting "handshake"
messages to the group. Note that unlike TLS and DTLS, there is not a
consolidated handshake phase to the protocol. Rather, handshake
messages are exchanged throughout the lifetime of a group, whenever a
change is made to the group state. This means an unbounded number of
interleaved application and handshake messages.
An MLS handshake message encapsulates a specific GroupOperation
message that accomplishes a change to the group state. It is carried
in an MLSPlaintext message that provides a signature by the sender of
the message. Applications may choose to send handshake messages in
encrypted form, as MLSCiphertext messages.
enum {
init(0),
add(1),
update(2),
remove(3),
(255)
} GroupOperationType;
struct {
GroupOperationType msg_type;
select (GroupOperation.msg_type) {
case init: Init;
case add: Add;
case update: Update;
case remove: Remove;
};
opaque confirmation<0..255>;
} GroupOperation;
The high-level flow for processing a handshake message is as follows:
1. If the handshake message is encrypted (i.e., encoded as an
MLSCiphertext object), decrypt it following the procedures
described in Section 8.
2. Verify that the "epoch" field of enclosing MLSPlaintext message
is equal the "epoch" field of the current GroupState object.
Barnes, et al. Expires November 3, 2019 [Page 32]
Internet-Draft MLS May 2019
3. Verify that the signature on the MLSPlaintext message verifies
using the public key from the credential stored at the leaf in
the tree indicated by the "sender" field.
4. Use the "operation" message to produce an updated, provisional
GroupState object incorporating the proposed changes.
5. Use the "confirmation_key" for the new epoch to compute the
confirmation MAC for this message, as described below, and verify
that it is the same as the "confirmation" field in the
GroupOperation object.
6. If the the above checks are successful, consider the updated
GroupState object as the current state of the group.
The confirmation value confirms that the members of the group have
arrived at the same state of the group:
GroupOperation.confirmation =
HMAC(confirmation_key, GroupState.transcript\_hash)
HMAC [RFC2104] uses the Hash algorithm for the ciphersuite in use.
Sign uses the signature algorithm indicated by the signer's
credential.
[[ OPEN ISSUE: It is not possible for the recipient of a handshake
message to verify that ratchet tree information in the message is
accurate, because each node can only compute the secret and private
key for nodes in its direct path. This creates the possibility that
a malicious participant could cause a denial of service by sending a
handshake message with invalid values for public keys in the ratchet
tree. ]]
9.1. Init
[[ OPEN ISSUE: Direct initialization is currently undefined. A
client can create a group by initializing its own state to reflect a
group including only itself, then adding the initial members. This
has computation and communication complexity O(N log N) instead of
the O(N) complexity of direct initialization. ]]
9.2. Add
In order to add a new member to the group, an existing member of the
group must take two actions:
1. Send a Welcome message to the new member
Barnes, et al. Expires November 3, 2019 [Page 33]
Internet-Draft MLS May 2019
2. Send an Add message to the group (including the new member)
The Welcome message contains the information that the new member
needs to initialize a GroupState object that can be updated to the
current state using the Add message. This information is encrypted
for the new member using HPKE. The recipient key pair for the HPKE
encryption is the one included in the indicated UserInitKey,
corresponding to the indicated ciphersuite.
struct {
HPKEPublicKey public_key;
optional<Credential> credential;
} RatchetNode;
struct {
ProtocolVersion version;
opaque group_id<0..255>;
uint32 epoch;
optional<RatchetNode> tree<1..2^32-1>;
opaque transcript_hash<0..255>;
opaque init_secret<0..255>;
} WelcomeInfo;
struct {
opaque user_init_key_id<0..255>;
CipherSuite cipher_suite;
HPKECiphertext encrypted_welcome_info;
} Welcome;
In the description of the tree as a list of nodes, the "credential"
field for a node MUST be populated if and only if that node is a leaf
in the tree.
Note that the "init_secret" in the Welcome message is the
"init_secret" at the output of the key schedule diagram in
Section 6.6. That is, if the "epoch" value in the Welcome message is
"n", then the "init_secret" value is "init_secret_[n]". The new
member can combine this init secret with the update secret
transmitted in the corresponding Add message to get the epoch secret
for the epoch in which it is added. No secrets from prior epochs are
revealed to the new member.
Since the new member is expected to process the Add message for
itself, the Welcome message should reflect the state of the group
before the new user is added. The sender of the Welcome message can
simply copy all fields from their GroupState object.
Barnes, et al. Expires November 3, 2019 [Page 34]
Internet-Draft MLS May 2019
[[ OPEN ISSUE: The Welcome message needs to be synchronized in the
same way as the Add. That is, the Welcome should be sent only if the
Add succeeds, and is not in conflict with another, simultaneous Add.
]]
An Add message provides existing group members with the information
they need to update their GroupState with information about the new
member:
struct {
uint32 index;
UserInitKey init_key;
opaque welcome_info_hash<0..255>;
} Add;
The "index" field indicates where in the tree the new member should
be added. The new member can be added at an existing, blank leaf
node, or at the right edge of the tree. In any case, the "index"
value MUST satisfy "0 <= index <= n", where "n" is the size of the
group. The case "index = n" indicates an add at the right edge of
the tree). If "index < n" and the leaf node at position "index" is
not blank, then the recipient MUST reject the Add as malformed.
The "welcome_info_hash" field contains a hash of the WelcomeInfo
object sent in a Welcome message to the new member.
A group member generates this message by requesting a UserInitKey
from the directory for the user to be added, and encoding it into an
Add message.
The client joining the group processes Welcome and Add messages
together as follows:
o Prepare a new GroupState object based on the Welcome message
o Process the Add message as an existing member would
An existing member receiving a Add message first verifies the
signature on the message, then updates its state as follows:
o If the "index" value is equal to the size of the group, increment
the size of the group, and extend the tree accordingly
o Verify the signature on the included UserInitKey; if the signature
verification fails, abort
Barnes, et al. Expires November 3, 2019 [Page 35]
Internet-Draft MLS May 2019
o Generate a WelcomeInfo object describing the state prior to the
add, and verify that its hash is the same as the value of the
"welcome_info_hash" field
o Update the ratchet tree by setting to blank all nodes in the
direct path of the new node
o Set the leaf node in the tree at position "index" to a new node
containing the public key from the UserInitKey in the Add
corresponding to the ciphersuite in use, as well as the credential
under which the UserInitKey was signed
The update secret resulting from this change is an all-zero octet
string of length Hash.length.
After processing an Add message, the new member SHOULD send an Update
immediately to update its key. This will help to limit the tree
structure degrading into subtrees, and thus maintain the protocol's
efficiency.
9.3. Update
An Update message is sent by a group member to update its leaf secret
and key pair. This operation provides post-compromise security with
regard to the member's prior leaf private key.
struct {
DirectPath path;
} Update;
The sender of an Update message creates it in the following way:
o Generate a fresh leaf key pair
o Compute its direct path in the current ratchet tree
A member receiving a Update message first verifies the signature on
the message, then updates its state as follows:
o Update the cached ratchet tree by replacing nodes in the direct
path from the updated leaf using the information contained in the
Update message
The update secret resulting from this change is the path secret for
the root node of the ratchet tree.
Barnes, et al. Expires November 3, 2019 [Page 36]
Internet-Draft MLS May 2019
9.4. Remove
A Remove message is sent by a group member to remove one or more
other members from the group. A member MUST NOT use a Remove message
to remove themselves from the group. If a member of a group receives
a Remove message where the removed index is equal to the signer
index, the recipient MUST reject the message as malformed.
struct {
uint32 removed;
DirectPath path;
} Remove;
The sender of a Remove message generates it as as follows:
o Generate a fresh leaf key pair
o Compute its direct path in the current ratchet tree, starting from
the removed leaf
A member receiving a Remove message first verifies the signature on
the message. The member then updates its state as follows:
o Update the ratchet tree by replacing nodes in the direct path from
the removed leaf using the information in the Remove message
o Update the ratchet tree by setting to blank all nodes in the
direct path of the removed leaf, and also setting the root node to
blank
o Truncate the tree such that the rightmost non-blank leaf is the
last node of the tree
Note that, in step 4, there must be at least one non-null element in
the tree, since any valid GroupState must have the current member in
the tree and self-removal is prohibited. The same reasoning
justifies the existence of a non-blank leaf in the ratchet tree in
step 5.
The update secret resulting from this change is the path secret
computed for the root node of the ratchet tree in the first step.
10. Sequencing of State Changes
[[ OPEN ISSUE: This section has an initial set of considerations
regarding sequencing. It would be good to have some more detailed
discussion, and hopefully have a mechanism to deal with this issue.
]]
Barnes, et al. Expires November 3, 2019 [Page 37]
Internet-Draft MLS May 2019
Each handshake message is premised on a given starting state,
indicated in its "prior_epoch" field. If the changes implied by a
handshake messages are made starting from a different state, the
results will be incorrect.
This need for sequencing is not a problem as long as each time a
group member sends a handshake message, it is based on the most
current state of the group. In practice, however, there is a risk
that two members will generate handshake messages simultaneously,
based on the same state.
When this happens, there is a need for the members of the group to
deconflict the simultaneous handshake messages. There are two
general approaches:
o Have the delivery service enforce a total order
o Have a signal in the message that clients can use to break ties
As long as handshake messages cannot be merged, there is a risk of
starvation. In a sufficiently busy group, a given member may never
be able to send a handshake message, because he always loses to other
members. The degree to which this is a practical problem will depend
on the dynamics of the application.
It might be possible, because of the non-contributivity of
intermediate nodes, that update messages could be applied one after
the other without the Delivery Service having to reject any handshake
message, which would make MLS more resilient regarding the
concurrency of handshake messages. The Messaging system can decide
to choose the order for applying the state changes. Note that there
are certain cases (if no total ordering is applied by the Delivery
Service) where the ordering is important for security, ie. all
updates must be executed before removes.
Regardless of how messages are kept in sequence, implementations MUST
only update their cryptographic state when valid handshake messages
are received. Generation of handshake messages MUST be stateless,
since the endpoint cannot know at that time whether the change
implied by the handshake message will succeed or not.
10.1. Server-Enforced Ordering
With this approach, the delivery service ensures that incoming
messages are added to an ordered queue and outgoing messages are
dispatched in the same order. The server is trusted to resolve
conflicts during race-conditions (when two members send a message at
Barnes, et al. Expires November 3, 2019 [Page 38]
Internet-Draft MLS May 2019
the same time), as the server doesn't have any additional knowledge
thanks to the confidentiality of the messages.
Messages should have a counter field sent in clear-text that can be
checked by the server and used for tie-breaking. The counter starts
at 0 and is incremented for every new incoming message. If two group
members send a message with the same counter, the first message to
arrive will be accepted by the server and the second one will be
rejected. The rejected message needs to be sent again with the
correct counter number.
To prevent counter manipulation by the server, the counter's
integrity can be ensured by including the counter in a signed message
envelope.
This applies to all messages, not only state changing messages.
10.2. Client-Enforced Ordering
Order enforcement can be implemented on the client as well, one way
to achieve it is to use a two step update protocol: the first client
sends a proposal to update and the proposal is accepted when it gets
50%+ approval from the rest of the group, then it sends the approved
update. Clients which didn't get their proposal accepted, will wait
for the winner to send their update before retrying new proposals.
While this seems safer as it doesn't rely on the server, it is more
complex and harder to implement. It also could cause starvation for
some clients if they keep failing to get their proposal accepted.
10.3. Merging Updates
It is possible in principle to partly address the problem of
concurrent changes by having the recipients of the changes merge
them, rather than having the senders retry. Because the value of
intermediate node is determined by its last updated child, updates
can be merged by recipients as long as the recipients agree on an
order - the only question is which node was last updated.
Recall that the processing of an update proceeds in two steps:
1. Compute updated secret values by hashing up the tree
2. Update the tree with the new secret and public values
To merge an ordered list of updates, a recipient simply performs
these updates in the specified order.
Barnes, et al. Expires November 3, 2019 [Page 39]
Internet-Draft MLS May 2019
For example, suppose we have a tree in the following configuration:
KDF(KDF(D))
/ \
KDF(B) KDF(D)
/ \ / \
A B C D
Now suppose B and C simultaneously decide to update to X and Y,
respectively. They will send out updates of the following form:
Update from B Update from C
============= =============
KDF(KDF(X)) KDF(KDF(Y))
/ \
KDF(X) KDF(Y)
\ /
X Y
Assuming that the ordering agreed by the group says that B's update
should be processed before C's, the other members in the group will
overwrite the root value for B with the root value from C, and all
arrive at the following state:
KDF(KDF(Y))
/ \
KDF(X) KDF(Y)
/ \ / \
A X Y D
11. Application Messages
The primary purpose of the handshake protocol is to provide an
authenticated group key exchange to clients. In order to protect
Application messages sent among those members of a group, the
Application secret provided by the handshake key schedule is used to
derive encryption keys for the Message Protection Layer.
Application messages MUST be protected with the Authenticated-
Encryption with Associated-Data (AEAD) encryption scheme associated
with the MLS ciphersuite. Note that "Authenticated" in this context
does not mean messages are known to be sent by a specific client but
only from a legitimate member of the group. To authenticate a
message from a particular member, signatures are required. Handshake
messages MUST use asymmetric signatures to strongly authenticate the
sender of a message.
Barnes, et al. Expires November 3, 2019 [Page 40]
Internet-Draft MLS May 2019
Each member maintains their own chain of Application secrets, where
the first one is derived based on a secret chained from the Epoch
secret. As shown in Section 6.6, the initial Application secret is
bound to the identity of each client to avoid collisions and allow
support for decryption of reordered messages.
Subsequent Application secrets MUST be rotated for each message sent
in order to provide stronger cryptographic security guarantees. The
Application Key Schedule use this rotation to generate fresh AEAD
encryption keys and nonces used to encrypt and decrypt future
Application messages. In all cases, a participant MUST NOT encrypt
more than expected by the security bounds of the AEAD scheme used.
Note that each change to the Group through a handshake message will
cause a change of the group Secret. Hence this change MUST be
applied before encrypting any new Application message. This is
required for confidentiality reasons in order for members to avoid
receiving messages from the group after leaving, being added to, or
excluded from the group.
11.1. Message Encryption and Decryption
The group members MUST use the AEAD algorithm associated with the
negotiated MLS ciphersuite to AEAD encrypt and decrypt their
Application messages according to the Message Framing section.
The group identifier and epoch allow a device to know which group
secrets should be used and from which Epoch secret to start computing
other secrets and keys. The "sender" identifier is used to derive
the member's Application secret chain from the initial group
Application secret. The application generation field is used to
determine which Application secret should be used from the chain to
compute the correct AEAD keys before performing decryption.
Application messages SHOULD be padded to provide some resistance
against traffic analysis techniques over encrypted traffic. [CLINIC]
[HCJ16] While MLS might deliver the same payload less frequently
across a lot of ciphertexts than traditional web servers, it might
still provide the attacker enough information to mount an attack. If
Alice asks Bob: "When are we going to the movie ?" the answer
"Wednesday" might be leaked to an adversary by the ciphertext length.
An attacker expecting Alice to answer Bob with a day of the week
might find out the plaintext by correlation between the question and
the length.
Similarly to TLS 1.3, if padding is used, the MLS messages MUST be
padded with zero-valued bytes before AEAD encryption. Upon AEAD
decryption, the length field of the plaintext is used to compute the
Barnes, et al. Expires November 3, 2019 [Page 41]
Internet-Draft MLS May 2019
number of bytes to be removed from the plaintext to get the correct
data. As the padding mechanism is used to improve protection against
traffic analysis, removal of the padding SHOULD be implemented in a
"constant-time" manner at the MLS layer and above layers to prevent
timing side-channels that would provide attackers with information on
the size of the plaintext. The padding length length_of_padding can
be chosen at the time of the message encryption by the sender.
Recipients can calculate the padding size from knowing the total size
of the ApplicationPlaintext and the length of the content.
[[ TODO: A preliminary formal security analysis has yet to be
performed on this authentication scheme.]]
[[ OPEN ISSUE: Currently, the group identifier, epoch and generation
are contained as meta-data of the Signature. A different solution
could be to include the GroupState instead, if more information is
required to achieve the security goals regarding cross-group attacks.
]]
[[ OPEN ISSUE: Should the padding be required for handshake messages
? Can an adversary get more than the position of a participant in the
tree without padding ? Should the base ciphertext block length be
negotiated or is is reasonable to allow to leak a range for the
length of the plaintext by allowing to send a variable number of
ciphertext blocks ? ]]
11.2. Delayed and Reordered Application messages
Since each Application message contains the group identifier, the
epoch and a message counter, a client can receive messages out of
order. If they are able to retrieve or recompute the correct AEAD
decryption key from currently stored cryptographic material clients
can decrypt these messages.
For usability, MLS clients might be required to keep the AEAD key and
nonce for a certain amount of time to retain the ability to decrypt
delayed or out of order messages, possibly still in transit while a
decryption is being done.
[[TODO: Describe here or in the Architecture spec the details.
Depending on which Secret or key is kept alive, the security
guarantees will vary.]]
12. Security Considerations
The security goals of MLS are described in [I-D.ietf-mls-
architecture]. We describe here how the protocol achieves its goals
Barnes, et al. Expires November 3, 2019 [Page 42]
Internet-Draft MLS May 2019
at a high level, though a complete security analysis is outside of
the scope of this document.
12.1. Confidentiality of the Group Secrets
Group secrets are derived from (i) previous group secrets, and (ii)
the root key of a ratcheting tree. Only group members know their
leaf private key in the group, therefore, the root key of the group's
ratcheting tree is secret and thus so are all values derived from it.
Initial leaf keys are known only by their owner and the group
creator, because they are derived from an authenticated key exchange
protocol. Subsequent leaf keys are known only by their owner.
[[TODO: or by someone who replaced them.]]
Note that the long-term identity keys used by the protocol MUST be
distributed by an "honest" authentication service for clients to
authenticate their legitimate peers.
12.2. Authentication
There are two forms of authentication we consider. The first form
considers authentication with respect to the group. That is, the
group members can verify that a message originated from one of the
members of the group. This is implicitly guaranteed by the secrecy
of the shared key derived from the ratcheting trees: if all members
of the group are honest, then the shared group key is only known to
the group members. By using AEAD or appropriate MAC with this shared
key, we can guarantee that a member in the group (who knows the
shared secret key) has sent a message.
The second form considers authentication with respect to the sender,
meaning the group members can verify that a message originated from a
particular member of the group. This property is provided by digital
signatures on the messages under identity keys.
[[ OPEN ISSUE: Signatures under the identity keys, while simple, have
the side-effect of preclude deniability. We may wish to allow other
options, such as (ii) a key chained off of the identity key, or (iii)
some other key obtained through a different manner, such as a
pairwise channel that provides deniability for the message
contents.]]
12.3. Forward and post-compromise security
Message encryption keys are derived via a hash ratchet, which
provides a form of forward secrecy: learning a message key does not
reveal previous message or root keys. Post-compromise security is
Barnes, et al. Expires November 3, 2019 [Page 43]
Internet-Draft MLS May 2019
provided by Update operations, in which a new root key is generated
from the latest ratcheting tree. If the adversary cannot derive the
updated root key after an Update operation, it cannot compute any
derived secrets.
12.4. Init Key Reuse
Initialization keys are intended to be used only once and then
deleted. Reuse of init keys is not believed to be inherently
insecure [dhreuse], although it can complicate protocol analyses.
13. IANA Considerations
TODO: Registries for protocol parameters, e.g., ciphersuites
14. Contributors
o Benjamin Beurdouche
INRIA
benjamin.beurdouche@ens.fr
o Karthikeyan Bhargavan
INRIA
karthikeyan.bhargavan@inria.fr
o Cas Cremers
University of Oxford
cas.cremers@cs.ox.ac.uk
o Alan Duric
Wire
alan@wire.com
o Srinivas Inguva
Twitter
singuva@twitter.com
o Albert Kwon
MIT
kwonal@mit.edu
o Eric Rescorla
Mozilla
ekr@rtfm.com
o Thyla van der Merwe
Royal Holloway, University of London
thyla.van.der@merwe.tech
Barnes, et al. Expires November 3, 2019 [Page 44]
Internet-Draft MLS May 2019
15. References
15.1. Normative References
[I-D.barnes-cfrg-hpke]
Barnes, R. and K. Bhargavan, "Hybrid Public Key
Encryption", draft-barnes-cfrg-hpke-01 (work in progress),
March 2019.
[IEEE1363]
"IEEE Standard Specifications for Password-Based Public-
Key Cryptographic Techniques", IEEE standard,
DOI 10.1109/ieeestd.2009.4773330, n.d..
[RFC2104] Krawczyk, H., Bellare, M., and R. Canetti, "HMAC: Keyed-
Hashing for Message Authentication", RFC 2104,
DOI 10.17487/RFC2104, February 1997,
<https://www.rfc-editor.org/info/rfc2104>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC5116] McGrew, D., "An Interface and Algorithms for Authenticated
Encryption", RFC 5116, DOI 10.17487/RFC5116, January 2008,
<https://www.rfc-editor.org/info/rfc5116>.
[RFC5869] Krawczyk, H. and P. Eronen, "HMAC-based Extract-and-Expand
Key Derivation Function (HKDF)", RFC 5869,
DOI 10.17487/RFC5869, May 2010,
<https://www.rfc-editor.org/info/rfc5869>.
[RFC7748] Langley, A., Hamburg, M., and S. Turner, "Elliptic Curves
for Security", RFC 7748, DOI 10.17487/RFC7748, January
2016, <https://www.rfc-editor.org/info/rfc7748>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
May 2017, <https://www.rfc-editor.org/info/rfc8174>.
[RFC8446] Rescorla, E., "The Transport Layer Security (TLS) Protocol
Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018,
<https://www.rfc-editor.org/info/rfc8446>.
[X962] ANSI, "Public Key Cryptography For The Financial Services
Industry: The Elliptic Curve Digital Signature Algorithm
(ECDSA)", ANSI X9.62, 1998.
Barnes, et al. Expires November 3, 2019 [Page 45]
Internet-Draft MLS May 2019
15.2. Informative References
[art] Cohn-Gordon, K., Cremers, C., Garratt, L., Millican, J.,
and K. Milner, "On Ends-to-Ends Encryption: Asynchronous
Group Messaging with Strong Security Guarantees", January
2018, <https://eprint.iacr.org/2017/666.pdf>.
[CLINIC] Miller, B., Huang, L., Joseph, A., and J. Tygar, "I Know
Why You Went to the Clinic: Risks and Realization of HTTPS
Traffic Analysis", Privacy Enhancing Technologies pp.
143-163, DOI 10.1007/978-3-319-08506-7_8, 2014.
[dhreuse] Menezes, A. and B. Ustaoglu, "On reusing ephemeral keys in
Diffie-Hellman key agreement protocols", International
Journal of Applied Cryptography Vol. 2, pp. 154,
DOI 10.1504/ijact.2010.038308, 2010.
[doubleratchet]
Cohn-Gordon, K., Cremers, C., Dowling, B., Garratt, L.,
and D. Stebila, "A Formal Security Analysis of the Signal
Messaging Protocol", 2017 IEEE European Symposium on
Security and Privacy (EuroS&P),
DOI 10.1109/eurosp.2017.27, April 2017.
[HCJ16] Husak, M., Čermak, M., Jirsik, T., and P.
Čeleda, "HTTPS traffic analysis and client
identification using passive SSL/TLS fingerprinting",
EURASIP Journal on Information Security Vol. 2016,
DOI 10.1186/s13635-016-0030-7, February 2016.
[I-D.ietf-trans-rfc6962-bis]
Laurie, B., Langley, A., Kasper, E., Messeri, E., and R.
Stradling, "Certificate Transparency Version 2.0", draft-
ietf-trans-rfc6962-bis-31 (work in progress), February
2019.
[keyagreement]
Barker, E., Chen, L., Roginsky, A., and M. Smid,
"Recommendation for Pair-Wise Key Establishment Schemes
Using Discrete Logarithm Cryptography", National Institute
of Standards and Technology report,
DOI 10.6028/nist.sp.800-56ar2, May 2013.
[signal] Perrin(ed), T. and M. Marlinspike, "The Double Ratchet
Algorithm", n.d.,
<https://www.signal.org/docs/specifications/
doubleratchet/>.
Barnes, et al. Expires November 3, 2019 [Page 46]
Internet-Draft MLS May 2019
Appendix A. Tree Math
One benefit of using left-balanced trees is that they admit a simple
flat array representation. In this representation, leaf nodes are
even-numbered nodes, with the n-th leaf at 2*n. Intermediate nodes
are held in odd-numbered nodes. For example, a 11-element tree has
the following structure:
X
X
X X X
X X X X X
X X X X X X X X X X X
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
This allows us to compute relationships between tree nodes simply by
manipulating indices, rather than having to maintain complicated
structures in memory, even for partial trees. The basic rule is that
the high-order bits of parent and child nodes have the following
relation (where "x" is an arbitrary bit string):
parent=01x => left=00x, right=10x
The following python code demonstrates the tree computations
necessary for MLS. Test vectors can be derived from the diagram
above.
# The largest power of 2 less than n. Equivalent to:
# int(math.floor(math.log(x, 2)))
def log2(x):
if x == 0:
return 0
k = 0
while (x >> k) > 0:
k += 1
return k-1
# The level of a node in the tree. Leaves are level 0, their
# parents are level 1, etc. If a node's children are at different
# level, then its level is the max level of its children plus one.
def level(x):
if x & 0x01 == 0:
return 0
k = 0
while ((x >> k) & 0x01) == 1:
k += 1
Barnes, et al. Expires November 3, 2019 [Page 47]
Internet-Draft MLS May 2019
return k
# The number of nodes needed to represent a tree with n leaves
def node_width(n):
return 2*(n - 1) + 1
# The index of the root node of a tree with n leaves
def root(n):
w = node_width(n)
return (1 << log2(w)) - 1
# The left child of an intermediate node. Note that because the
# tree is left-balanced, there is no dependency on the size of the
# tree. The child of a leaf node is itself.
def left(x):
k = level(x)
if k == 0:
return x
return x ^ (0x01 << (k - 1))
# The right child of an intermediate node. Depends on the size of
# the tree because the straightforward calculation can take you
# beyond the edge of the tree. The child of a leaf node is itself.
def right(x, n):
k = level(x)
if k == 0:
return x
r = x ^ (0x03 << (k - 1))
while r >= node_width(n):
r = left(r)
return r
# The immediate parent of a node. May be beyond the right edge of
# the tree.
def parent_step(x):
k = level(x)
b = (x >> (k + 1)) & 0x01
return (x | (1 << k)) ^ (b << (k + 1))
# The parent of a node. As with the right child calculation, have
# to walk back until the parent is within the range of the tree.
def parent(x, n):
if x == root(n):
return x
p = parent_step(x)
Barnes, et al. Expires November 3, 2019 [Page 48]
Internet-Draft MLS May 2019
while p >= node_width(n):
p = parent_step(p)
return p
# The other child of the node's parent. Root's sibling is itself.
def sibling(x, n):
p = parent(x, n)
if x < p:
return right(p, n)
elif x > p:
return left(p)
return p
# The direct path of a node, ordered from the root
# down, not including the root or the terminal node
def direct_path(x, n):
d = []
p = parent(x, n)
r = root(n)
while p != r:
d.append(p)
p = parent(p, n)
return d
# The copath of the node is the siblings of the nodes on its direct
# path (including the node itself)
def copath(x, n):
d = dirpath(x, n)
if x != sibling(x, n):
d.append(x)
return [sibling(y, n) for y in d]
# Frontier is is the list of full subtrees, from left to right. A
# balance binary tree with n leaves has a full subtree for every
# power of two where n has a bit set, with the largest subtrees
# furthest to the left. For example, a tree with 11 leaves has full
# subtrees of size 8, 2, and 1.
def frontier(n):
st = [1 << k for k in range(log2(n) + 1) if n & (1 << k) != 0]
st = reversed(st)
base = 0
f = []
for size in st:
f.append(root(size) + base)
base += 2*size
Barnes, et al. Expires November 3, 2019 [Page 49]
Internet-Draft MLS May 2019
return f
# Leaves are in even-numbered nodes
def leaves(n):
return [2*i for i in range(n)]
# The resolution of a node is the collection of non-blank
# descendants of this node. Here the tree is represented by a list
# of nodes, where blank nodes are represented by None
def resolve(tree, x, n):
if tree[x] != None:
return [x]
if level(x) == 0:
return []
L = resolve(tree, left(x), n)
R = resolve(tree, right(x, n), n)
return L + R
Authors' Addresses
Richard Barnes
Cisco
Email: rlb@ipv.sx
Jon Millican
Facebook
Email: jmillican@fb.com
Emad Omara
Google
Email: emadomara@google.com
Katriel Cohn-Gordon
University of Oxford
Email: me@katriel.co.uk
Barnes, et al. Expires November 3, 2019 [Page 50]
Internet-Draft MLS May 2019
Raphael Robert
Wire
Email: raphael@wire.com
Barnes, et al. Expires November 3, 2019 [Page 51]