ssbc / ssb-meta-feeds-group-spec

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

ssb-meta-feeds-group-spec

Version: 1.0

Authors: Mix Irving mix@protozoa.nz, Andre Staltz contact@staltz.com

License: This work is licensed under a Creative Commons Attribution 4.0 International License.

Abstract

This document specifies how SSB private group content is organized in a metafeed tree, what data must be encrypted and to whom, and how peers replicate group-related portions of the tree.

1. Introduction

SSB Private Groups is a symmetric encryption format that allows a large number of peers to share a symmetric key and use it to encrypt messages to each other. With the wide adoption of private groups, there would be a large volume of content that is not readable by the general public, which can cause storage problems as that content is replicated in the network.

In order to support partial replication, it is desireable to put different group content in different subfeeds. However, we need to have a clear way to discover how you've been invited to a group, without replicating the whole group's content. We also need to consider how to ensure our group data is replicated enough so it's readily available for anyone who needs access to it. For example, if a group contains only three people, and if only two people have copies of the group data, then there's little or no gossip propagation, and the third member can only get updates when it is directly connected one of the other two members. Thus we need to enable sympathetic replication, such that peers who don't strictly need the group content are incentivized to replicate it anyway.

Additionally, our metafeed structure should not allow other peers to learn which groups a peer is in, unless they are also members of the group.

Use of metafeeds for groups is described formally in Section 3 and illustrated in the examples in Section 4.

2. Terminology

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119.

3. Functional Specification

Implementations supporting this specification MUST follow the ssb-meta-feeds-spec version 1.0.

We define two types of feeds that each peer will have:

  1. A group/additions feed
  2. A "group feed" for each group

We then define three different flows for how peers can interact with these feeds when:

  1. Creating a new group
  2. Adding a peer to a group
  3. Discovering group membership

3.1. The group/additions feed

The purpose of this feed is to hold messages for coordination of joining groups. It represents the record of all peers the user had added to groups.

  • 3.1.1 Each peer running this spec MUST have an additions feed
  • 3.1.2 Each peer MUST deterministically place their additions feed as a subfeed of a shard feed, such that:
    • The shard feed is derived from the string "group/additions" according to the v1 tree structure specified in ssb-meta-feeds-spec.
    • The metafeeds/add/derived message announcing the additions feed
      • MUST have feedpurpose equal to the string group/additions,
      • MUST be of feed format classic
      • MUST have metadata following the ssb-meta-feeds-dm-spec Section 1
      • MUST NOT be encrypted
  • 3.1.3 If a peer A wants to add another peer B in a group, then A MUST replicate B's additions feed and B must replicate A's additions feed
  • 3.1.4 All messages published on the additions feed MUST be "group/add-member" messages encrypted with envelope-spec encryption
    • See details below
  • 3.1.5 There MUST be at most one additions feed per metafeed tree

group/add-member messages

This is the only type of message currently expected in the additions feed. It's defined in the private-group-spec and MUST have at least the following fields in its message content:

  • type property equal to the string "group/add-member"
  • version property equal to the string "v2"
  • secret property equal to the base64-encoded string of the group secret key
  • root property equal to the ID of the group/init message, as an SSB URI
  • recps property containing an array of feed IDs
    • The first feed ID MUST be the ID of the group feed, as a ssb:identity/group/ URI
    • The subsequent feed IDs (at most 15 of them) MUST be the root metafeed ID of the peer(s) being added to the group, all as SSB URIs

The encryption of this message on the additions feed MUST follow the ssb-meta-feeds-dm-spec.

3.2. Group feeds

The purpose of this feed is to hold the messages the user publishes for a specific group. It represents all the activity this user has in a group. The user has at most one group feed for each group they are a member of.

Each group feed MUST be a direct subfeed of a shard feed, where the shard is derived using the base64 encoded group secret.

  • 3.2.1 Each peer that is a member of a group MAY have a group feed for that group
    • It is not compulsory to have a group feed for a group the user is a member of, because the user may not have confirmed their participation in the group after being added to it.
  • 3.2.2 Each peer MUST deterministically place their group feed as a subfeed of a shard feed, such that:
    • The shard feed is derived from the base64 encoded string of the group secret according to the v1 tree structure specified in ssb-meta-feeds-spec.
    • The metafeeds/add/derived message announcing the group feed
      • MUST have feedpurpose equal to the base64 encoded group secret
      • MUST be encrypted with the group secret, using envelope-spec encryption
      • SHOULD be encrypted as well to your own_key, for recovery purposes
  • 3.2.3 Each group member SHOULD replicate the group feeds of all other group members they know of
  • 3.2.4 All content on the group feed MUST be encrypted with the group secret key, using envelope-spec encryption
Details about the shard feed
The shard feed is derived by the base64 encoded group secret.

We cannot use the group id, as this is publicly known, which would give attackers a way to test if people are in the group (breaking membership confidentiality).

We choose the group secret because it is a value known only to those already in the group.

Details about the group feed

We need a feedpurpose which is unique to the group, which the group secret is.

We cannot use the group id, because this is derived using the group init message, which does not exist until our feed exists. We encrypt this announce message so as not to leak the secret and to protect membership confidentiality.

For sympathetic replication we will therefore need a distinct type of announcement.

3.3. Creating a new group

To create a new group, a peer should perform all of the following steps successfully:

  • 3.3.1. MUST create a new symmetric group key, also known as the group secret, which MUST have at least 32 bytes of cryptographically secure random data
  • 3.3.2. MUST create a new group feed using this symmetric group key, as described in Section 3.2
  • 3.3.3. MUST publish a group/init message on the group feed, as described in the private-group-spec
  • 3.3.3. SHOULD create a group/additions feed, as described in Section 3.1, unless there is already one for this metafeed
  • 3.3.4. SHOULD publish a group/add-member message on the group/additions feed to add themselves to the group, as described in Section 3.1

3.4. Adding a peer to a group

To add other peers to a group, a peer MUST publish a group/add-member message on the group/additions feed addressed to the other peer, as described in Section 3.1. This message must be encrypted following the ssb-meta-feeds-dm-spec.

3.5. Discovering group membership

To discover the members of a group, a peer MUST perform the following steps:

  • 3.5.1. MUST replicate the group/additions feed of other peers, either on the basis of friendship or because they are a known member of a group
  • 3.5.2. MUST decrypt some or all group/add-member messages on the group/additions feed
  • 3.5.3. Each decrypted group/add-member message reveals, in the recps field, the newly added member(s) and the group identifier, allowing the user to infer that the newly added member(s) are members of the group

Upon a peer discovering that they are a member of a group, they MAY create a group feed if they do not already have one. This allows them to publish messages to the group.

4. Examples

The following diagram illustrates the structure of a metafeed tree for a peer, "Staltz", who is a member of three groups: "helsinki", "aalborg", and "wellington".

graph TB

root(root)
v1(v1)
5(5) & b(b) & f(f)
additions(group/additions):::additionsClass
aalborg(aalborg):::group
helsinki(helsinki):::group
wellington(wellington):::group

root --> v1 --> 5 --> helsinki
         v1 --> b --> additions
         v1 --> f --> aalborg
                f --> wellington

classDef default stroke:none;
classDef additionsClass fill: #BF2669, stroke:none, color:white;
classDef group fill: #702A8C, stroke: none, color:white;

4.1. Creating a group

Staltz starts up his application. We assume he has already created his group/additions feed (following the spec above). In his application he creates a new "helsinki" group, which means he:

  1. Creates a new symmetric groupKey, also known as "group secret"
  2. Creates a content feed under some shard (using the groupKey following the spec above)
  3. Publishes a box2-encrypted group/init message on that new "helsinki" content feed
  4. Publishes a box2-encrypted group/add-member message on his "group/additions" feed
details
This helps new members quickly see he is a member of the group, and also ensures he has a copy of the groupKey persisted in his records (encrypted to him and the group)
graph TB

root(root)
v1(v1)
4(4)
d(d)
additions(group/additions):::additionsClass
helsinki(helsinki):::group

subgraph Staltz
  root --> v1 --> 4 --> helsinki
           v1 --> d --> additions
end

classDef default stroke:none;
classDef additionsClass fill: #BF2669, stroke:none, color:white;
classDef group fill: #702A8C, stroke: none, color:white;

Diagram showing Staltz feed state from his perspective

4.2. Group creator invites someone

Staltz wants to invite his friend Arj to the group he set up, so he publishes a group/add-member message (which contains the group secret) on his "group/additions" feed.

When Arj next starts up his application and replicates Staltz's feed tree (they are friends), he discovers the new group/add-member for him on Staltz's "group/additions" feed (because peers must replicate their friends' "group/additions" feeds).

graph TB

rootA(root)
v1A(v1)
4A(4):::unreplicated
dA(d)
additionsA(group/additions):::additionsClass
helsinkiA(helsinki):::unreplicated

rootB(root)
v1B(v1)
9B(9)
additionsB(group/additions):::additionsClass

subgraph Staltz
  rootA --> v1A --> 4A --> helsinkiA
            v1A --> dA --> additionsA
end

subgraph Arj
  rootB --> v1B --> 9B --> additionsB
end

classDef default stroke:none;
classDef additionsClass fill: #BF2669, stroke:none, color:white;
classDef group fill: #702A8C, stroke: none, color:white;
classDef unreplicated opacity: 0.4, stroke: none;

Diagram showing feed state of Arj and Staltz from Arj's perspective. The greyed out feeds show feeds that exist for Staltz but which Arj has yet to want to replicate.

Assuming he accepts this invitation, Arj then does the following:

  1. Calculates the shard for the "helsinki" group for staltz, and starts replicating that shard feed and the "helsinki" feed
  2. Creates a "helsinki" feed for himself
graph TB

rootA(root)
v1A(v1)
4A(4) & dA(d)
additionsA(group/additions):::additionsClass
helsinkiA(helsinki):::group

rootB(root)
v1B(v1)
9B(9) & cB(c)
additionsB(group/additions):::additionsClass
helsinkiB(helsinki):::group

subgraph Staltz
  rootA --> v1A --> 4A --> helsinkiA
            v1A --> dA --> additionsA
end

subgraph Arj
  rootB --> v1B --> 9B --> additionsB
            v1B --> cB --> helsinkiB
end

classDef default stroke:none;
classDef additionsClass fill: #BF2669, stroke:none, color:white;
classDef group fill: #702A8C, stroke: none, color:white;

Diagram showing the updated state for Arj after he joins the group. Note the shards each feed lands in are different for each person (but deterministic if you know the groupKey).

Staltz can see that Arj has accepted the invitation because he is able to decrypt the feed announcement message for Arj's "helsinki" feed on the shard feed, and read that the feedpurpose is the groupKey. Staltz knows which shard feed to watch for the announcement, because Arj's shard feed is deterministically derived with information Staltz is aware of.

4.3. Non-group creator invites someone

Arj now wants to invite Mix to the "helsinki" group. He follows the same pattern as in (2), but now as the inviter.

Mix knows Arj is a part of the group because he was invited by them. Mix also knows Staltz is part of the group because all group/add-member messages have

Staltz can see Arj has invited Mix because he's replicating Arj's "group/additions" feed, so Staltz starts replicating Mix's group feed.

5. Security Considerations

5.1. Degraded confidentiality at scale

As noted in the private-groups-original-notes, SSB Private Groups are based on symmetric encryption and static keys (no key rotation), which means it cannot guarantee perfect-forward-secrecy neither post-compromise-security. This makes SSB Private Groups less likely to guarantee confidentiality at scale, when a group has dozens or hundreds of members. If any of the members leaks the group secret, all past communication in the group is compromised.

This also means that any group member can add new members to the group. They may even do so without publishing a group/add-member message, which effectively allows any group member to invite a third party that will remain undetected by existing group members. The only signal group members may get is that this third party is requesting replication for group feeds that they should not know about.

SSB Private Groups with several members should thus be used with caution. Our security model assumes that the group is as trustworthy as the least trustworthy member in the group.

The choice of symmetric encryption (as opposed to, e.g. hash-ratchet or double-ratchet) was made because of efficiency. Symmetric encryption means adding new members or publishing new content is O(1) fast, and the system is overall simple to implement.

5.2. Membership analysis attacks

A group feed is discoverable only by the group members, because its announcement is encrypted. However, an eavesdropper who can replicate a peer's metafeed tree will notice the presence of these encrypted message on shard feeds. If the eavesdropper can do that for several peers, then they may perform a metadata analysis attack where they try to correlate the encrypted messages on the shard feeds to the group members.

This can be mitigated by publishing dummy encrypted messages on the shard feed at random intervals, with the downside of increasing the size of the shard feed, and thus making partial replication heavier.

Another mitigation, which is covered by this specification and by ssb-meta-feeds-spec, is how shard feeds are derived for each peer. The derivation involves a hash function and a piece of metadata unique to the peer, which means that group members will have their group feeds located in a different shard. This makes it harder for an eavesdropper to correlate the encrypted messages on the shard feeds of the peers it is eavesdropping on.

Another way membership information can leak is through replication requests: if only Alice, Bob, and Carol request for feeds A, B, C, and if these feeds don't seem to be publicly declared in any metafeed tree, it might be reasonable for an eavesdropper to guess that Alice, Bob, Carol are all in a group and that A, B, C are group feeds. We think this can be mitigated through "sympathetic replication". Sympathetic replication involves asking friendly peers who aren't in our groups to replicate some of our group feeds. This adds more noise in metadata analysis attacks, and makes it harder to guess who is in a group. It also improves replication resilience, because there are more peers replicating group feeds, making them more available in the network. Sympathetic replication needs further investigation and specification.

It is currently unknown how much can be learned from metadata analysis attacks, as it depends on these mitigations and the network's complexity, such as number of peers, number of groups, and number of relationships between peers.

6. References

6.1. Normative References

6.2. Informative References

Acknowledgements

This project has received funding from the European Union’s Horizon 2020 research and innovation programme within the framework of the NGI-Assure Project funded under grant agreement No 957073.

About