Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Process comments folkert #3

Merged
merged 3 commits into from
Dec 21, 2023
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 18 additions & 13 deletions draft-venhoek-nts-pool.md
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,10 @@ author:
fullname: "David Venhoek"
organization: Tweede golf B.V.
email: "david@tweedegolf.com"
-
fullname: "Folkert de Vries"
organization: Tweede golf B.V.
email: "folkert@tweedegolf.com"

normative:
RFC8915:
Expand All @@ -49,7 +53,7 @@ NTS {{RFC8915}} provides authenticity and limited confidentiality for NTP {{RFC5

This document aims to provide extensions to the NTS Key Exchange sessions that allow for an implementation of a pool for NTS that:

- Is usable without changes to client,
- is usable without changes to the client,
- avoids constraining the downstream time source's cookie format,
- avoids downstream time sources having potential access to all traffic.

Expand All @@ -63,14 +67,13 @@ Where further specificity of the role of a participant is needed, we will use th

# General pool architecture

We propose a pool model where the pool provides an NTS Key Exchange service to the outside world. This allows the pool to terminate the TLS connection and avoids having to distribute certificates to all downstream time servers. However, that also implies that the pool needs to extract the keys and get valid cookies for the selected downstream time server.
We propose a pool model where the pool provides an NTS Key Exchange service to the outside world. A major advantage of this model is that it avoids having to distribute certificates to all downstream time servers. Contrary to {{RFC8915}}, there is no direct TLS connection between the client and the selected downstream time service.

To solve this, we ask downstream servers to provide an extension of the NTS Key Exchange protocol that allows the pool to directly communicate the keys to the downstream server, instead of having the downstream server extract this from the TLS session. The explicit communication of keys allows the pool to do the extraction from the TLS server whilst remaining oblivious to the cookie format of the downstream server.
In {{RFC8915}}, cookies are generated based on key material that is extracted from this TLS connection. Our proposed model instead establishes two TLS connections: between the client and the pool, and between the pool and the downstream time server. Because cookies need to be generated using key material from the client, the pool extracts this key material and sends it to the server. The server uses this key material (rather than key material extracted from its connection with the pool) to generate cookies. This way, the pool can remain oblivious to the cookie format of the downstream time server.

# Client facilities for pools

One challenge with a pool through NTS Key Exchange is that clients that allow for explicit pool configuration want to end up with multiple independent time sources. To ensure they won't have to do multiple NTS Key Exchange sessions just to discard the result because they already have the time server they
result in, we also introduce a record clients can use to indicate which downstream time servers they don't want.
One challenge with getting multiple time sources from a single NTS Key Exchange server is that clients that allow for explicit pool configuration want to end up with multiple independent time sources. Without additional support, a user of a pool might receive a downstream time source it already has from an NTS Key Exchange session, resulting in that session being a waste of time. To avoid this, we also introduce a record that clients can use to indicate which downstream time servers they don't want, because they already have them.

# New NTS record types

Expand All @@ -80,9 +83,9 @@ Critical bit: 0

Indicates a desire to keep the TLS connection active for more than one message exchange. This can be used by a pool to reuse connections to downstream NTS Key Exchange servers multiple times, reducing load on both the pool and downstream servers.

Client MUST send this record with a 0 sized body. Client MUST NOT use Keep Alive unless the request contains a record type allowing the use of Keep Alive. Within this specification, that is limited to the Supported Protocol List and Fixed Key Request records. A server SHOULD ignore any body for the Keep Alive record.
Client MUST send this record with a body of size 0. Client MUST NOT use Keep Alive unless the request contains a record type allowing the use of Keep Alive. Within this specification, that is limited to the Supported Protocol List and Fixed Key Request records. A server SHOULD ignore any body for the Keep Alive record.

When supported by server and allowed for the request in question, the server MUST include a Keep Alive record with 0 sized body in the response and keep the TLS connection active after the response to handle further requests from the client. A client SHOULD ignore any body for the Keep Alive record.
When supported by server and allowed for the request in question, the server MUST include a Keep Alive record with a body of size 0 in the response and keep the TLS connection active after the response to handle further requests from the client. A client SHOULD ignore any body for the Keep Alive record.

When included in the request or response, the client respectively server MAY, contrary to the requirements in {{RFC8915}}, send another request or response. Any TLS "close_notify" SHALL be sent only after the last request or response respectively to use the connection.

Expand All @@ -94,9 +97,9 @@ Critical bit: 1

This record can be used by a pool to query downstream servers about which next protocols they support.

Client MUST send with no body. Clients MAY use Keep Alive in combination with this record. A request with this record SHOULD NOT include a "Next Protocol Negotiation", "AEAD Algorithm Negotiation" or "Fixed Key Request" record.
Client MUST send with no body. Clients MAY use Keep Alive in combination with this record. Contrary to {{RFC8915}}, a request with this record SHOULD NOT include a "Next Protocol Negotiation", "AEAD Algorithm Negotiation" or "Fixed Key Request" record.

Server MUST ignore any client body sent and MUST send in response a Supported Next Protocol List record with as data a list of 16 bit integers, giving the protocol IDs the server supports.
Server MUST ignore any client body sent and MUST send in response a Supported Next Protocol List record with as data a list of 16-bit integers, giving the protocol IDs the server supports.

When included, the server MUST NOT negotiate a next protocol, AEAD algorithm, or keys for this request.

Expand All @@ -106,12 +109,14 @@ Critical bit: 1

This record can be used by a pool to query downstream servers about which AEAD algorithms they support.

Client MUST send with no body. Clients MAY use Keep Alive in combination with this record. A request with this record SHOULD NOT include a "Next Protocol Negotiation", "AEAD Algorithm Negotiation" or "Fixed Key Request" record.
Client MUST send with no body. Clients MAY use Keep Alive in combination with this record. Contrary to {{RFC8915}}, a request with this record SHOULD NOT include a "Next Protocol Negotiation", "AEAD Algorithm Negotiation" or "Fixed Key Request" record.

Server MUST ignore any client body sent and MUST send in response a Supported Algorithm List record with as data a list of tuples of two 16 bit integers, the first giving an algorithm ID for the AEAD and the second giving the length of the key for that algorithm ID.
Server MUST ignore any client body sent and MUST send in response a Supported Algorithm List record with as data a list of tuples of two 16-bit integers, the first giving an algorithm ID for the AEAD and the second giving the length of the key for that algorithm ID.

When included, the server MUST NOT negotiate a next protocol, AEAD algorithm, or keys for this request.

We include algorithm key size in the response so that a pool does not itself need knowledge of which AEAD algorithm's exist, and what their key sizes are. Instead, it can use the provided key length when extracting keys from the TLS connection between end user and pool. This allows adoption of new AEAD algorithms without any software changes being needed in pool software.

## Fixed Key Request {#fixedkey}
Record Type Number: To be assigned by IANA (draft implementations: 0x4002)
Critical Bit: 1
Expand All @@ -136,15 +141,15 @@ MUST NOT be sent by a server. Server MAY at its discretion ignore the request fr

## Pool's position

In the pool design presented above, the pool effectively acts as a man in the middle between the user and the ultimate time source during the NTS Key Exchange portion of the session. This means that the pool has access to the key material of all these sessions. Although this is a small additional risk, we consider this acceptable as the pool could already always assign sessions for a user to time servers it controls anyway.
In the pool design presented above, the pool effectively acts as a man in the middle between the user and the ultimate time source during the NTS Key Exchange portion of the session. This means that the pool has access to the key material of these sessions. Although this is a small additional risk, we consider this acceptable because the pool could already always assign sessions for a user to time servers it controls anyway.

The fact that the pool also gets access to key material makes it less advisable to have a pool as a downstream time source for another pool, as this increases the number of actors with access to the key material even further.

The design above does avoid sharing key material between all downstream time sources. As a consequence, a downstream time source in the pool will not be able to break confidentiality or authenticity of traffic with other downstream time sources of the pool. Furthermore, any traffic directly with the downstream time source has no key material involved that is known to the pool.

## Error handling

To avoid giving multiple downstream time sources access to the key material of the end user, it is important that the keys extracted from the TLS session between the user and the pool are sent to at most one downstream time source. If an error occurs after sending the Fixed Key Request record, either with the TLS connection between the pool and the downstream time source, or by being explicitly reported by the downstream time source to the pool, the pool SHOULD return an error to the user. Retrying with a different downstream time source may unintentionally leave the user vulnerable to the provider of the originally selected downstream time source.
To avoid giving multiple downstream time sources access to the key material of the end user, it is important that the keys extracted from the TLS session between the user and the pool are sent to at most one downstream time source. If an error occurs after sending the Fixed Key Request record, either with the TLS connection between the pool and the downstream time source, or by being explicitly reported by the downstream time source to the pool, the pool SHOULD return an error to the user. Retrying with a different downstream time source may unintentionally leave the user vulnerable to the operator of the originally selected downstream time source.

# IANA Considerations

Expand Down