Specification of DNS over Dedicated QUIC Connections

4.1. Provide DNS Privacy

DoT [RFC7858] defines how to mitigate some of the issues described in "DNS Privacy Considerations" [RFC7626] by specifying how to transmit DNS messages over TLS. The "Usage Profiles for DNS over TLS and DNS over DTLS" [RFC8310] specify Strict and Opportunistic Usage Profiles for DoT including how stub resolvers can authenticate recursive resolvers.¶

QUIC connection setup includes the negotiation of security parameters using TLS, as specified in "Using TLS to Secure QUIC" [RFC9001], enabling encryption of the QUIC transport. Transmitting DNS messages over QUIC will provide essentially the same privacy protections as DoT [RFC7858] including Strict and Opportunistic Usage Profiles [RFC8310]. Further discussion on this is provided in Section 9.¶

4.2. Design for Minimum Latency

QUIC is specifically designed to reduce the delay between HTTP queries and HTTP responses. This is achieved through three main components:¶

1. Support for 0-RTT data during session resumption.¶
2. Support for advanced error recovery procedures as specified in "QUIC Loss Detection and Congestion Control" [RFC9002].¶
3. Mitigation of head-of-line blocking by allowing parallel delivery of data on multiple streams.¶

This mapping of DNS to QUIC will take advantage of these features in three ways:¶

1. Optional support for sending 0-RTT data during session resumption (the security and privacy implications of this are discussed in later sections).¶
2. Long-lived QUIC connections over which multiple DNS transactions are performed, generating the sustained traffic required to benefit from advanced recovery features.¶
3. Fast resumption of QUIC connections to manage the disconnect-on-idle feature of QUIC without incurring retransmission time-outs.¶
4. Mapping of each DNS Query/Response transaction to a separate stream, to mitigate head-of-line blocking. This enables servers to respond to queries "out of order". It also enables clients to process responses as soon as they arrive, without having to wait for in order delivery of responses previously posted by the server.¶

These considerations will be reflected in the mapping of DNS traffic to QUIC streams in Section 5.2.¶

4.3. No Specific Middlebox Bypass Mechanism

The mapping of DoQ is defined for minimal overhead and maximum performance. This means a different traffic profile than HTTP3 over QUIC. This difference can be noted by firewalls and middleboxes. There may be environments in which HTTP3 over QUIC will be able to pass through, but DoQ will be blocked by these middle boxes.¶

4.4. No Server Initiated Transactions

As stated in Section 1, this document does not specify support for server initiated transactions within established DoQ connections. That is, only the initiator of the DoQ connection may send queries over the connection.¶

DSO supports server-initiated transactions within existing connections, however DSO is not applicable to DNS over HTTP since HTTP has its own mechanism for managing sessions, and this is incompatible with the DSO; the same is true for DoQ.¶

5.1. Connection Establishment

DoQ connections are established as described in the QUIC transport specification [RFC9000]. During connection establishment, DoQ support is indicated by selecting the ALPN token "doq" in the crypto handshake.¶

5.2. Stream Mapping and Usage

The mapping of DNS traffic over QUIC streams takes advantage of the QUIC stream features detailed in Section 2 of the QUIC transport specification [RFC9000].¶

DNS traffic follows a simple pattern in which the client sends a query, and the server provides one or more responses (multiple can responses occur in zone transfers).¶

The mapping specified here requires that the client selects a separate QUIC stream for each query. The server then uses the same stream to provide all the response messages for that query. In order that multiple responses can be parsed, a 2-octet length field is used in exactly the same way as the 2-octet length field defined for DNS over TCP [RFC1035]. The practical result of this is that the content of each QUIC stream is exactly the same as the content of a TCP connection that would manage exactly one query.¶

All DNS messages (queries and responses) sent over DoQ connections MUST be encoded as a 2-octet length field followed by the message content as specified in [RFC1035].¶

The client MUST select the next available client-initiated bidirectional stream for each subsequent query on a QUIC connection, in conformance with the QUIC transport specification [RFC9000].¶

The client MUST send the DNS query over the selected stream, and MUST indicate through the STREAM FIN mechanism that no further data will be sent on that stream.¶

The server MUST send the response(s) on the same stream and MUST indicate, after the last response, through the STREAM FIN mechanism that no further data will be sent on that stream.¶

Therefore, a single client initiated DNS transaction consumes a single stream. This means that the client's first query occurs on QUIC stream 0, the second on 4, and so on.¶

For completeness it is noted that versions prior to -02 of this specification proposed a simpler mapping scheme which omitted the 2 byte length field and supported only a single response on a given stream. The more complex mapping above was adopted to specifically cater for XFR support, however it breaks compatibility with earlier versions.¶

5.3. DoQ Error Codes

The following error codes are defined for use when abruptly terminating streams, aborting reading of streams, or immediately closing connections:¶

DOQ_NO_ERROR (0x00):

No error. This is used when the connection or stream needs to be closed, but there is no error to signal.¶

DOQ_INTERNAL_ERROR (0x01):

The DoQ implementation encountered an internal error and is incapable of pursuing the transaction or the connection.¶

DOQ_PROTOCOL_ERROR (0x02):

The DoQ implementation encountered an protocol error and is forcibly aborting the connection.¶

DOQ_REQUEST_CANCELLED (0x03):

A DoQ client uses this to signal that it wants to cancel an outstanding transaction.¶

See Section 10.4 for details on registering new error codes.¶

5.4. Connection Management

Section 10 of the QUIC transport specification [RFC9000] specifies that connections can be closed in three ways:¶

• idle timeout¶
• immediate close¶
• stateless reset¶

Clients and servers implementing DoQ SHOULD negotiate use of the idle timeout. Closing on idle timeout is done without any packet exchange, which minimizes protocol overhead. Per section 10.1 of the QUIC transport specification, the effective value of the idle timeout is computed as the minimum of the values advertised by the two endpoints. Practical considerations on setting the idle timeout are discussed in Section 6.5.2.¶

Clients SHOULD monitor the idle time incurred on their connection to the server, defined by the time spent since the last packet from the server has been received. When a client prepares to send a new DNS query to the server, it will check whether the idle time is sufficient lower than the idle timer. If it is, the client will send the DNS query over the existing connection. If not, the client will establish a new connection and send the query over that connection.¶

Clients MAY discard their connection to the server before the idle timeout expires. If they do that, they SHOULD close the connection explicitly, using QUIC's CONNECTION_CLOSE mechanism, and use the DoQ error code DOQ_NO_ERROR.¶

Clients and servers MAY close the connection for a variety of other reasons, indicated using QUIC's CONNECTION_CLOSE. Client and servers that send packets over a connection discarded by their peer MAY receive a stateless reset indication. If a connection fails, all queries in progress over the connection MUST be considered failed, and a Server Failure (SERVFAIL, [RFC1035]) SHOULD be notified to the initiator of the transaction.¶

5.5. Session Resumption and 0-RTT

A client MAY take advantage of the session resumption mechanisms supported by QUIC transport [RFC9000] and QUIC TLS [RFC9001]. Clients SHOULD consider potential privacy issues associated with session resumption before deciding to use this mechanism. These privacy issues are detailed in Section 9.2 and Section 9.1, and the implementation considerations are discussed in Section 6.5.3.¶

The 0-RTT mechanism SHOULD NOT be used to send DNS requests that are not "replayable" transactions. Our analysis so far shows that such replayable transactions can only be QUERY requests, although we may need to also consider NOTIFY requests once the analysis of NOTIFY services is complete, see Appendix A.¶

Servers MUST NOT execute non replayable transactions received in 0-RTT data. Servers MUST adopt one of the following behaviors:¶

• Queue the offending transaction and only execute it after the QUIC handshake has been completed, as defined in section 4.1.1 of [RFC9001].¶
• Reply to the offending transaction with a response code REFUSED and an Extended DNS Error Code (EDE) "Too Early", see Section 10.3.¶
• Close the connection with the error code DOQ_PROTOCOL_ERROR.¶

For the zone transfer scenario, it would be possible to replay an XFR QUERY that had been sent in 0-RTT data. However the authentication mechanisms described in RFC9103 ("Zone transfer over TLS") will ensure that the response is not sent by the primary until the identity of the secondary has been verified i.e. the first behavior listed above.¶

5.6. Message Sizes

DoQ Queries and Responses are sent on QUIC streams, which in theory can carry up to 2^62 bytes. However, DNS messages are restricted in practice to a maximum size of 65535 bytes. This maximum size is enforced by the use of a two-octet message length field in DNS over TCP [RFC1035] and DNS over TLS [RFC7858], and by the definition of the "application/dns-message" for DNS over HTTP [RFC8484]. DoQ enforces the same restriction.¶

The Extension Mechanisms for DNS (EDNS) [RFC6891] allow peers to specify the UDP message size. This parameter is ignored by DoQ. DoQ implementations always assume that the maximum message size is 65535 bytes.¶

6.1. Authentication

For the stub to recursive resolver scenario, the authentication requirements are the same as described in DoT [RFC7858] and "Usage Profiles for DNS over TLS and DNS over DTLS" [RFC8310]. There is no need to authenticate the client's identity in either scenario.¶

For zone transfer, the requirements are the same as described in [RFC9103].¶

For the recursive resolver to authoritative nameserver scenario, authentication requirements are unspecified at the time of writing and are the subject on ongoing work in the DPRIVE WG.¶

6.2. Fall Back to Other Protocols on Connection Failure

If the establishment of the DoQ connection fails, clients MAY attempt to fall back to DoT and then potentially clear text, as specified in DoT [RFC7858] and "Usage Profiles for DNS over TLS and DNS over DTLS" [RFC8310], depending on their privacy profile.¶

DNS clients SHOULD remember server IP addresses that don't support DoQ, including timeouts, connection refusals, and QUIC handshake failures, and not request DoQ from them for a reasonable period (such as one hour per server). DNS clients following an out-of-band key-pinned privacy profile ([RFC7858]) MAY be more aggressive about retrying DoQ connection failures.¶

6.3. Address Validation

Section 8 of the QUIC transport specification [RFC9000] defines Address Validation procedures to avoid servers being used in address amplification attacks. DoQ implementations MUST conform to this specification, which limits the worst case amplification to a factor 3.¶

DoQ implementations SHOULD consider configuring servers to use the Address Validation using Retry Packets procedure defined in section 8.1.2 of the QUIC transport specification [RFC9000]). This procedure imposes a 1-RTT delay for verifying the return routability of the source address of a client, similar to the DNS Cookies mechanism [RFC7873].¶

DoQ implementations that configure Address Validation using Retry Packets SHOULD implement the Address Validation for Future Connections procedure defined in section 8.1.3 of the QUIC transport specification [RFC9000]). This defines how servers can send NEW TOKEN frames to clients after the client address is validated, in order to avoid the 1-RTT penalty during subsequent connections by the client from the same address.¶

6.4. Padding

Implementations SHOULD protect against the traffic analysis attacks described in Section 9.4 by the judicious injection of padding. This could be done either by padding individual DNS messages using the EDNS(0) Padding Option [RFC7830] and by padding QUIC packets (see Section 8.6 of the QUIC transport specification [RFC9000]).¶

In theory, padding at the QUIC level could result in better performance for the equivalent protection, because the amount of padding can take into account non-DNS frames such as acknowledgeemnts or flow control updates, and also because QUIC packets can carry multiple DNS messages. However, applications can only control the amount of padding in QUIC packets if the implementation of QUIC exposes adequate APIs. This leads to the following recommendation:¶

• if the implementation of QUIC exposes APIs to set a padding policy, DNS over QUIC SHOULD use that API to align the packet length to a small set of fixed sizes, aligned with the recommendations of the "Padding Policies for Extension Mechanisms for DNS (EDNS(0))" [RFC8467].¶
• if padding at the QUIC level is not available or not used, DNS over QUIC MUST ensure that all DNS queries and responses are padded to a small set of fixed sizes, using the EDNS padding extension as specified in "Padding Policies for Extension Mechanisms for DNS (EDNS(0))" [RFC8467].¶

6.5. Connection Handling

"DNS Transport over TCP - Implementation Requirements" [RFC7766] provides updated guidance on DNS over TCP, some of which is applicable to DoQ. This section attempts to specify which and how those considerations apply to DoQ.¶

6.6. Processing Queries in Parallel

As specified in Section 7 of "DNS Transport over TCP - Implementation Requirements" [RFC7766], resolvers are RECOMMENDED to support the preparing of responses in parallel and sending them out of order. In DoQ, they do that by sending responses on their specific stream as soon as possible, without waiting for availability of responses for previously opened streams.¶

6.7. Zone transfer

[RFC9103] specifies zone transfer over TLS (XoT) and includes updates to [RFC1995] (IXFR), [RFC5936] (AXFR) and [RFC7766]. Considerations relating to the re-use of XoT connections described there apply analogously to zone transfers performed using DoQ connections. For example:¶

• DoQ servers MUST be able to handle multiple concurrent IXFR requests on a single QUIC connection¶
• DoQ servers MUST be able to handle multiple concurrent AXFR requests on a single QUIC connection¶

• DoQ implementations SHOULD¶

  • use the same QUIC connection for both AXFR and IXFR requests to the same primary¶
  • pipeline such requests (if they pipeline XFR requests in general) and MAY intermingle them¶
  • send the response(s) for each request as soon as they are available i.e. responses MAY be sent intermingled¶

6.8. Flow Control Mechanisms

Servers and Clients manage flow control using the mechanisms defined in section 4 of [RFC9000]. These mechanisms allow clients and servers to specify how many streams can be created, how much data can be sent on a stream, and how much data can be sent on the union of all streams. For DNS over QUIC, controlling how many streams are created allows servers to control how many new requests the client can send on a given connection.¶

Flow control exists to protect endpoint resources. For servers, global and per-stream flow control limits control how much data can be sent by clients. The same mechanisms allow clients to control how much data can be sent by servers. Values that are too small will unnecessarily limit performance. Values that are too large might expose endpoints to overload or memory exhaustion. Implementations or deployments will need to adjust flow control limits to balance these concerns. In particular, zone transfer implementations will need to control these limits carefully to ensure both large and concurrent zone transfers are well managed.¶

Initial values of parameters control how many requests and how much data can be sent by clients and servers at the beginning of the connection. These values are specified in transport parameters exchanged during the connection handshake. The parameter values received in the initial connection also control how many requests and how much data can be sent by clients using 0-RTT data in a resumed connection. Using too small values of these initial parameters would restrict the usefulness of allowing 0-RTT data.¶

7.1. Performance Measurements

To our knowledge, no benchmarking studies comparing DoT, DoH and DoQ are published yet. However anecdotal evidence from the AdGuard DoQ recursive resolver deployment indicates that it performs well compared to the other encrypted protocols, particularly in mobile environments. Reasons given for this include that DoQ¶

• Uses less bandwidth due to a more efficient handshake (and due to less per message overhead when compared to DoH).¶
• Performs better in mobile environments due to the increased resilience to packet loss¶
• Can maintain connections as users move between mobile networks via its connection management¶

9.1. Privacy Issues With 0-RTT data

The 0-RTT data can be replayed by adversaries. That data may trigger queries by a recursive resolver to authoritative resolvers. Adversaries may be able to pick a time at which the recursive resolver outgoing traffic is observable, and thus find out what name was queried for in the 0-RTT data.¶

This risk is in fact a subset of the general problem of observing the behavior of the recursive resolver discussed in "DNS Privacy Considerations" [RFC7626]. The attack is partially mitigated by reducing the observability of this traffic. However, the risk is amplified for 0-RTT data, because the attacker might replay it at chosen times, several times.¶

The recommendation for TLS 1.3 [RFC8446] is that the capability to use 0-RTT data should be turned off by default, and only enabled if the user clearly understands the associated risks. In our case, allowing 0-RTT data provides significant performance gains, and we are concerned that a recommendation to not use it would simply be ignored. Instead, we provide a set of practical recommendations in Section 5.5 and Section 6.5.3.¶

The prevention on allowing replayable transactions in 0-RTT data expressed in Section 5.5 blocks the most obvious risks of replay attacks, as it only allows for transactions that will not change the long term state of the server.¶

Attacks trying to assess the state of the cache are more powerful if the attacker can choose the time at which the 0-RTT data will be replayed. Such attacks are blocked if the server enforces single-use tickets, or if the server implements a combination of Client Hello recording and freshness checks, as specified in section 8 of [RFC8446]. These blocking mechanisms rely on shared state between all server instances in a server system. In the case of DNS over QUIC, the protection against replay attacks on the DNS cache is achieved if this state is shared between all servers that share the same DNS cache.¶

The attacks described above apply to the stub resolver to recursive resolver scenario, but similar attacks might be envisaged in the recursive resolver to authoritative resolver scenario, and the same mitigations apply.¶

9.2. Privacy Issues With Session Resumption

The QUIC session resumption mechanism reduces the cost of re-establishing sessions and enables 0-RTT data. There is a linkability issue associated with session resumption, if the same resumption token is used several times. Attackers on path between client and server could observe repeated usage of the token and use that to track the client over time or over multiple locations.¶

The session resumption mechanism allows servers to correlate the resumed sessions with the initial sessions, and thus to track the client. This creates a virtual long duration session. The series of queries in that session can be used by the server to identify the client. Servers can most probably do that already if the client address remains constant, but session resumption tickets also enable tracking after changes of the client's address.¶

The recommendations in Section 6.5.3 are designed to mitigate these risks. Using session tickets only once mitigates the risk of tracking by third parties. Refusing to resume a session if addresses change mitigates the risk of tracking by the server.¶

The privacy trade-offs here may be context specific. Stub resolvers will have a strong motivation to prefer privacy over latency since they often change location. However, recursive resolvers that use a small set of static IP addresses are more likely to prefer the reduced latency provided by session resumption and may consider this a valid reason to use resumption tickets even if the IP address changed between sessions.¶

Encrypted zone transfer (RFC9103) explicitly does not attempt to hide the identity of the parties involved in the transfer, but at the same time such transfers are not particularly latency sensitive. This means that applications supporting zone transfers may decide to apply the same protections as stub to recursive applications.¶

9.3. Privacy Issues With New Tokens

QUIC specifies address validation mechanisms in section 8 of [RFC9000]. Use of an address validation token allows QUIC servers to avoid an extra RTT for new connections. Address validation tokens are typically tied to an IP address. QUIC clients normally only use these tokens when setting a new connection from a previously used address. However, due to the prevalence of NAT, clients are not always aware that they are using a new address. There is a linkability risk if clients mistakenly use address validation tokens after unknowingly moving to a new location.¶

The recommendations in Section 6.5.3 mitigates this risk by tying the usage of the NEW TOKEN to that of session resumption.¶

9.4. Traffic Analysis

Even though QUIC packets are encrypted, adversaries can gain information from observing packet lengths, in both queries and responses, as well as packet timing. Many DNS requests are emitted by web browsers. Loading a specific web page may require resolving dozen of DNS names. If an application adopts a simple mapping of one query or response per packet, or "one QUIC STREAM frame per packet", then the succession of packet lengths may provide enough information to identify the requested site.¶

Implementations SHOULD use the mechanisms defined in Section 6.4 to mitigate this attack.¶

10.1. Registration of DoQ Identification String

This document creates a new registration for the identification of DoQ in the "Application Layer Protocol Negotiation (ALPN) Protocol IDs" registry [RFC7301].¶

The "doq" string identifies DoQ:

    Protocol:                 DoQ
    Identification Sequence:  0x64 0x6F 0x71 ("doq")
    Specification:            This document

10.2. Reservation of Dedicated Port

Port 853 is currently reserved for 'DNS query-response protocol run over TLS/DTLS' [RFC7858]. However, the specification for DNS over DTLS (DoD) [RFC8094] is experimental, limited to stub to resolver, and no implementations or deployments currently exist to our knowledge (even though several years have passed since the specification was published).¶

This specification proposes to additionally reserve the use of port 853 for DoQ. QUIC was designed to be able to co-exist with other protocols on the same port, including DTLS , see Section 17.2 in [RFC9000].¶

IANA is requested to add the following value to the "Service Name and Transport Protocol Port Number Registry" in the System Range. The registry for that range requires IETF Review or IESG Approval [RFC6335].¶

   Service Name           dns-over-quic
   Port Number            853
   Transport Protocol(s)  UDP
   Assignee               IESG
   Contact                IETF Chair
   Description            DNS query-response protocol run over QUIC
   Reference              This document

10.3. Reservation of Extended DNS Error Code Too Early

IANA is requested to add the following value to the Extended DNS Error Codes registry [RFC8914]:¶

   INFO-CODE              TBD
   Purpose                Too Early
   Reference              This document

10.4. DNS over QUIC Error Codes Registry

IANA [SHALL add/has added] a registry for "DNS over QUIC Error Codes" on the "Domain Name System (DNS) Parameters" web page.¶

The "DNS over QUIC Error Codes" registry governs a 62-bit space. This space is split into three regions that are governed by different policies:¶

• Permanent registrations for values between 0x00 and 0x3f (in hexadecimal; inclusive), which are assigned using Standards Action or IESG Approval as defined in Section 4.9 and 4.10 of [RFC8126]¶
• Permanent registrations for values larger than 0x3f, which are assigned using the Specification Required policy ([RFC8126])¶
• Provisonal registrations for values larger than 0x3f, which require Expert Review, as defined in Section 4.5 of [RFC8126].¶

Provisional reservations share the range of values larger than 0x3f with some permanent registrations. This is by design, to enable conversion of provisional registrations into permanent registrations without requiring changes in deployed systems. (This design is aligned with the principles set in section 22 of [RFC9000].)¶

Registrations in this registry MUST include the following fields:¶

Value:

The assigned codepoint.¶

Status:

"Permanent" or "Provisional".¶

Contact:

Contact details for the registrant.¶

Notes:

Supplementary notes about the registration.¶

In addition, permanent registrations MUST include:¶

Error:

A short mnemonic for the parameter.¶

Specification:

A reference to a publicly available specification for the value (optional for provisional registrations).¶

Description:

A brief description of the error code semantics, which MAY be a summary if a specification reference is provided.¶

Provisional registrations of codepoints are intended to allow for private use and experimentation with extensions to DNS over QUIC. However, provisional registrations could be reclaimed and reassigned for another purpose. In addition to the parameters listed above, provisional registrations MUST include:¶

Date:

The date of last update to the registration.¶

A request to update the date on any provisional registration can be made without review from the designated expert(s).¶

The initial contents of this registry are shown in Table 1.¶