Return Routability Check for DTLS 1.2 and DTLS 1.3

Introduction A Connection ID (CID) is an identifier carried in the record layer header of a DTLS datagram that gives the receiver additional information for selecting the appropriate security context. The CID mechanism has been specified in for DTLS 1.2 and in for DTLS 1.3. Section 6 of describes how the use of CID increases the attack surface of DTLS 1.2 and 1.3 by providing both on-path and off-path attackers an opportunity for (D)DoS. It also describes the steps a DTLS principal must take when a record with a CID is received that has a source address (and/or port) different from the one currently associated with the DTLS connection. However, the actual mechanism for ensuring that the new peer address is willing to receive and process DTLS records is left open. To address the gap, this document defines a return routability check (RRC) sub-protocol for DTLS 1.2 and 1.3. The return routability check is performed by the receiving endpoint before the CID-address binding is updated in that endpoint's session state. This is done in order to give the receiving endpoint confidence that the sending peer is in fact reachable at the source address (and port) indicated in the received datagram. of this document explains the fundamental mechanism that aims to reduce the DDoS attack surface. Additionally, in , a more advanced address validation mechanism is discussed. This mechanism is designed to counteract off-path attackers trying to place themselves on-path by racing packets that trigger address rebinding at the receiver. To gain a detailed understanding of the attacker model, please refer to . Apart from of its use in the context of CID-address binding updates, the path validation capability offered by RRC can be used at any time by either endpoint. For instance, an endpoint might use RRC to check that a peer is still reachable at its last known address after a period of quiescence.

Conventions and 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 14 when, and only when, they appear in all capitals, as shown here. This document assumes familiarity with the CID format and protocol defined for DTLS 1.2 and for DTLS 1.3 . The presentation language used in this document is described in Section 4 of . This document reuses the definition of "anti-amplification limit" from to mean three times the amount of data received from an unvalidated address. This includes all DTLS records originating from that source address, excluding discarded ones. The terms "peer" and "endpoint" are defined in .

RRC Extension The use of RRC is negotiated via the rrc extension. The rrc extension is only defined for DTLS 1.2 and DTLS 1.3. On connecting, a client wishing to use RRC includes the rrc extension in its ClientHello. If the server is capable of meeting this requirement, it responds with a rrc extension in its ServerHello. The extension_type value for this extension is TBD1 and the extension_data field of this extension is empty. The client and server MUST NOT use RRC unless both sides have successfully exchanged rrc extensions. A client offering the rrc extension MUST also offer the connection_id extension . A client offering the connection_id extension SHOULD also offer the rrc extension, unless the application using DTLS has its own address validation mechanism.

RRC and CID Interplay RRC offers an in-protocol mechanism to perform peer address validation that complements the "peer address update" procedure described in . Specifically, when both CID and RRC have been successfully negotiated for the session, if a record with CID is received that has the source address and/or source port number of the enclosing UDP datagram different from what is currently associated with that CID value, the receiver SHOULD perform a return routability check as described in , unless an application layer specific address validation mechanism can be triggered instead (e.g., CoAP Echo ).

Return Routability Check Message Types This document defines the return_routability_check content type () to carry Return Routability Check messages. The RRC sub-protocol consists of three message types: path_challenge, path_response and path_drop that are used for path validation and selection as described in . Each message carries a Cookie, an 8-byte field containing 64 bits of entropy (e.g., obtained from the CSPRNG used by the TLS implementation, see ). The return_routability_check message MUST be authenticated and encrypted using the currently active security context.

Return Routability Check Message Future extensions to the Return Routability Check sub-protocol may define new message types. Implementations MUST be able to parse and understand the three RRC message types defined in this document. In addition, implementations MUST be able to parse and gracefully ignore messages with an unknown msg_type.

Attacker Model We define two classes of attackers, off-path and on-path, with increasing capabilities (see ) partly following terminology introduced in QUIC (:

An off-path attacker is not on the original path between the DTLS peers, but is able to observe packets on the original path and has faster routing compared to the DTLS peers, which allows it to make copies of the observed packets, race its copies to either peer and consistently win the race.
An on-path attacker is on the original path between the DTLS peers and is therefore capable, compared to the off-path attacker, to also drop and delay records at will.

Note that, in general, attackers cannot craft DTLS records in a way that would successfully pass verification, due to the cryptographic protections applied by the DTLS record layer.

On-path and off-path attackers that try to create an amplification attack by spoofing the source address of the victim ().
Off-path attackers that try to put themselves on-path (), provided that the enhanced path validation algorithm is used ().

Amplification Both on-path and off-path attackers can send a packet (either by modifying it on the fly, or by copying, injecting, and racing it, respectively) with the source address modified to that of a victim host. If the traffic generated by the server in response is larger compared to the received packet (e.g., a CoAP server returning an MTU's worth of data from a 20-bytes GET request ) the attacker can use the server as a traffic amplifier toward the victim.

Mitigation Strategy When receiving a packet with a known CID that has a source address different from the one currently associated with the DTLS connection, an RRC-capable endpoint will not send a (potentially large) response but instead a small path_challenge message to the victim host. Since the host is not able to decrypt it and generate a valid path_response, the address validation fails, which in turn keeps the original address binding unaltered. Note that in case of an off-path attacker, the original packet still reaches the intended destination; therefore, an implementation could use a different strategy to mitigate the attack.

Off-Path Packet Forwarding An off-path attacker that can observe packets might forward copies of genuine packets to endpoints over a different path. If the copied packet arrives before the genuine packet, this will appear as a path change, like in a genuine NAT rebinding occurrence. Any genuine packet will be discarded as a duplicate. If the attacker is able to continue forwarding packets, it might be able to cause migration to a path via the attacker. This places the attacker on-path, giving it the ability to observe or drop all subsequent packets. This style of attack relies on the attacker using a path that has the same or better characteristics (e.g., due to a more favourable service level agreements) as the direct path between endpoints. The attack is more effective if relatively few packets are sent or if packet loss coincides with the attempted attack. A data packet received on the original path that increases the maximum received packet number will cause the endpoint to move back to that path. Therefore, eliciting packets on this path increases the likelihood that the attack is unsuccessful. Note however that, unlike QUIC, DTLS has no "non-probing" packets so this would require application specific mechanisms.

Mitigation Strategy illustrates the case where a receiver receives a packet with a new source IP address and/or new port number. In order to determine that this path change was not triggered by an off-path attacker, the receiver will send a RRC message of type path_challenge (1) on the old path.

Off-Path Packet Forwarding Scenario Three cases need to be considered: Case 1: The old path is dead (e.g., due to a NAT rebinding), which leads to a timeout of (1). As shown in , a path_challenge (2) needs to be sent on the new path. If the sender replies with a path_response on the new path (3), the switch to the new path is considered legitimate.

Old path is dead | +-------. | .-----+ Receiver +...... | | | .---+ | . | | | | '----------' . | path- 3 | | . 1 path- response | | | . | challenge | | | . | .--|-+-|----------------------v--. / | | NAT X / timeout '----|-+-|-----------------------' | | | . | | 2 path- . | | | challenge . | | | .----------. . | | '-->| | . | '-----+ Sender +.....' '-------+ | '----------' ]]> Case 2: The old path is alive but not preferred. This case is shown in whereby the sender replies with a path_drop message (2) on the old path. This triggers the receiver to send a path_challenge (3) on the new path. The sender will reply with a path_response (4) on the new path, thus providing confirmation for the path migration.

Old path is not preferred | |<------. | .-----+ Receiver +-----. | | | .---+ +---. | | | | | '----------' | | | path- 4 | | path- 1 | | response | | | challenge | | | | | | | | | .---------|-+-|----. .--|-+-|-----------. / AP/NAT A | | / / | | AP/NAT B / '-----------|---|--' '----|-+-|---------' | | | | | | | | 3 path- | | 2 path- | | | challenge | | | drop | | | .----------. | | | | | '-->| |<--' | | | '-----+ Sender +-----' | '-------+ +-------' '----------' ]]> Case 3: The old path is alive and preferred. This is most likely the result of an off-path attacker trying to place itself on path. The receiver sends a path_challenge on the old path and the sender replies with a path_response (2) on the old path. The interaction is shown in . This results in the connection not being migrated to the new path, thus thwarting the attack.

Old path is preferred Note that this defense is imperfect, but this is not considered a serious problem. If the path via the attacker is reliably faster than the old path despite multiple attempts to use that old path, it is not possible to distinguish between an attack and an improvement in routing. An endpoint could also use heuristics to improve detection of this style of attack. For instance, NAT rebinding is improbable if packets were recently received on the old path; similarly, rebinding is rare on IPv6 paths. Endpoints can also look for duplicated packets. Conversely, a change in connection ID is more likely to indicate an intentional migration rather than an attack. Note that changes in connection IDs are supported in DTLS 1.3 but not in DTLS 1.2.

Path Validation Procedure The receiver that observes the peer's address or port change MUST stop sending any buffered application data, or limit the data sent to the unvalidated address to the anti-amplification limit. It then initiates the return routability check. This document describes two kinds of checks: basic () and enhanced (). The choice of one or the other depends on whether the off-path attacker scenario described in is to be considered. (The decision on what strategy to choose depends mainly on the threat model, but may also be influenced by other considerations. Examples of impacting factors include: the need to minimise implementation complexity, privacy concerns, and the need to reduce the time it takes to switch path. The choice may be offered as a configuration option to the user of the TLS implementation.) After the path validation procedure is completed, any pending send operation is resumed to the bound peer address. and list the requirements for the initiator and responder roles, broken down per protocol phase.

Basic The basic return routability check comprises the following steps:

The receiver (i.e., the initiator) creates a return_routability_check message of type path_challenge and places the unpredictable cookie into the message.
The message is sent to the observed new address and a timer T (see ) is started.
The peer (i.e., the responder) cryptographically verifies the received return_routability_check message of type path_challenge and responds by echoing the cookie value in a return_routability_check message of type path_response.
When the initiator receives the return_routability_check message of type path_response and verifies that it contains the sent cookie, it updates the peer address binding.
If T expires the peer address binding is not updated.

Enhanced The enhanced return routability check comprises the following steps:

The receiver (i.e., the initiator) creates a return_routability_check message of type path_challenge and places the unpredictable cookie into the message.
The message is sent to the previously valid address, which corresponds to the old path. Additionally, a timer T is started, see .
If the path is still functional, the peer (i.e., the responder) cryptographically verifies the received return_routability_check message of type path_challenge. The action to be taken depends on whether the path through which the message was received is the preferred one or not anymore:
- If the path through which the message was received is preferred, a return_routability_check message of type path_response MUST be returned.
- If the path through which the message was received is not preferred, a return_routability_check message of type path_drop MUST be returned. In either case, the peer echoes the cookie value in the response.
The initiator receives and verifies that the return_routability_check message contains the previously sent cookie. The actions taken by the initiator differ based on the received message:
- When a return_routability_check message of type path_response was received, the initiator MUST continue using the previously valid address, i.e., no switch to the new path takes place and the peer address binding is not updated.
- When a return_routability_check message of type path_drop was received, the initiator MUST perform a return routability check on the observed new address, as described in .
If T expires the peer address binding is not updated. In this case, the initiator MUST perform a return routability check on the observed new address, as described in .

Path Challenge Requirements

The initiator MAY send multiple return_routability_check messages of type path_challenge to cater for packet loss on the probed path.
- Each path_challenge SHOULD go into different transport packets. (Note that the DTLS implementation may not have control over the packetization done by the transport layer.)
- The transmission of subsequent path_challenge messages SHOULD be paced to decrease the chance of loss.
- Each path_challenge message MUST contain random data.
The initiator MAY use padding using the record padding mechanism available in DTLS 1.3 (and in DTLS 1.2, when CID is enabled on the sending direction) up to the anti-amplification limit to probe if the path MTU (PMTU) for the new path is still acceptable.

Path Response/Drop Requirements

The responder MUST NOT delay sending an elicited path_response or path_drop messages.
The responder MUST send exactly one path_response or path_drop message for each received path_challenge.
The responder MUST send the path_response or the path_drop on the path where the corresponding path_challenge has been received, so that validation succeeds only if the path is functional in both directions. The initiator MUST NOT enforce this behaviour.
The initiator MUST silently discard any invalid path_response or path_drop it receives.

Note that RRC does not cater for PMTU discovery on the reverse path. If the responder wants to do PMTU discovery using RRC, it should initiate a new path validation procedure.

Timer Choice When setting T, implementations are cautioned that the new path could have a longer round-trip time (RTT) than the original. In settings where there is external information about the RTT of the active path, implementations SHOULD use T = 3xRTT. If an implementation has no way to obtain information regarding the RTT of the active path, T SHOULD be set to 1s. Profiles for specific deployment environments -- for example, constrained networks -- MAY specify a different, more suitable value for T.

Example In the example DTLS 1.3 handshake shown in , a client and a server successfully negotiate support for both CID and the RRC extension.

Message Flow for Full DTLS Handshake ServerHello ^ Key + key_share | Exch + connection_id=100 | + rrc v {EncryptedExtensions} ^ Server {CertificateRequest} v Params {Certificate} ^ {CertificateVerify} | Auth <-------- {Finished} v ^ {Certificate} Auth | {CertificateVerify} v {Finished} --------> [Application Data] <-------> [Application Data] + Indicates noteworthy extensions sent in the previously noted message. * Indicates optional or situation-dependent messages/extensions that are not always sent. {} Indicates messages protected using keys derived from a [sender]_handshake_traffic_secret. [] Indicates messages protected using keys derived from [sender]_application_traffic_secret_N. ]]> Once a connection has been established, the client and the server exchange application payloads protected by DTLS with a unilaterally used CID. In our case, the client is requested to use CID 100 for records sent to the server. At some point in the communication interaction, the IP address used by the client changes and, thanks to the CID usage, the security context to interpret the record is successfully located by the server. However, the server wants to test the reachability of the client at its new IP address. shows the server initiating a "basic" RRC exchange (see ) that establishes reachability of the client at the new IP address.

"Basic" Return Routability Example Src-IP=A Dst-IP=Z <======== Application Data Src-IP=Z Dst-IP=A <<------------->> << Some >> << Time >> << Later >> <<------------->> Application Data ========> Src-IP=B Dst-IP=Z <<< Unverified IP Address B >> <-------- Return Routability Check path_challenge(cookie) Src-IP=Z Dst-IP=B Return Routability Check --------> path_response(cookie) Src-IP=B Dst-IP=Z <<< IP Address B Verified >> <======== Application Data Src-IP=Z Dst-IP=B ]]>

Security Considerations Note that the return routability checks do not protect against flooding of third-parties if the attacker is on-path, as the attacker can redirect the return routability checks to the real peer (even if those datagrams are cryptographically authenticated). On-path adversaries can, in general, pose a harm to connectivity.

Privacy Considerations When using DTLS 1.3, peers SHOULD avoid using the same CID on multiple network paths, in particular when initiating connection migration or when probing a new network path, as described in , as an adversary can otherwise correlate the communication interaction across those different paths. DTLS 1.3 provides mechanisms to ensure that a new CID can always be used. In general, an endpoint should proactively send a RequestConnectionId message to ask for new CIDs as soon as the pool of spare CIDs is depleted (or goes below a threshold). Also, in case a peer might have exhausted available CIDs, a migrating endpoint could include NewConnectionId in packets sent on the new path to make sure that the subsequent path validation can use fresh CIDs. Note that DTLS 1.2 does not offer the ability to request new CIDs during the session lifetime since CIDs have the same life-span of the connection. Therefore, deployments that use DTLS in multihoming environments SHOULD refuse to use CIDs with DTLS 1.2 and switch to DTLS 1.3 if the correlation privacy threat is a concern.

Operational Considerations

Logging Anomalous Events Logging of RRC operations at both ends of the protocol can be generally useful for the users of an implementation. In particular, for security information and event management (SIEM) and troubleshooting purposes, it is strongly advised that implementations collect statistics about any unsuccessful RRC operations, as they could represent security-relevant events when they coincide with attempts by an attacker to interfere with the end-to-end path.

Middlebox Interference Since the DTLS 1.3 encrypted packet's record type is opaque to on-path observers, RRC messages are immune to middlebox interference when using DTLS 1.3. In contrast, DTLS 1.2 RRC messages that are not wrapped in the tls12_cid record (e.g., in the server-to-client direction if the server didn't negotiate a CID or it negotiated a zero-length CID) have the return_routability_check content type in plain text, making them susceptible to interference (e.g., dropping of path_challenge messages), which would hinder the RRC functionality altogether. Therefore, when using RRC in DTLS 1.2, it is recommended to enable CID in both directions.

IANA Considerations RFC Editor: please replace RFCthis with this RFC number and remove this note.

New TLS ContentType IANA is requested to allocate an entry to the TLS ContentType registry, for the return_routability_check(TBD2) message defined in this document. IANA is requested to set the DTLS_OK column to Y and to add the following note prior to the table:

NOTE: The return_routability_check content type is only applicable to DTLS 1.2 and 1.3.

New TLS ExtensionType IANA is requested to allocate the extension code point (TBD1) for the rrc extension to the TLS ExtensionType Values registry as described in . rrc entry in the TLS ExtensionType Values registry

Value	Extension Name	TLS 1.3	DTLS-Only	Recommended	Reference	Comment
TBD1	rrc	CH, SH	Y	N	RFCthis

New RRC Message Type Sub-registry IANA is requested to create a new registry for RRC Message Types within the TLS Parameters registry group . This registry will be administered under the "Expert Review" policy (). Follow the procedures in to submit registration requests. Each entry in the registry must include the following fields:

Value:: A (decimal) number in the range 0 to 253
Description:: A brief description of the RRC message
DTLS-Only:: Whether the message applies only to DTLS. Since RRC is only available in DTLS, this column will be set to Y for all the entries in this registry
Recommended:: Whether the message is recommended for implementations to support. The semantics for this field is defined in and updated in
Reference:: A reference to a publicly available specification for the value
Comment:: Any relevant notes or comments that relate to this entry

The initial state of this sub-registry is as follows: Initial Entries in RRC Message Type registry

Value	Description	DTLS-Only	Recommended	Reference
0	path_challenge	Y	Y	RFCthis
1	path_response	Y	Y	RFCthis
2	path_drop	Y	Y	RFCthis
3-253	Unassigned
254-255	Reserved for Private Use	Y		RFCthis

IANA is requested to add the following note for additional information regarding the use of RRC message codepoints in experiments:

Note:: As specified in , assignments made in the Private Use space are not generally useful for broad interoperability. Those making use of the Private Use range are responsible for ensuring that no conflicts occur within the intended scope of use. For widespread experiments, provisional registrations () are available.

Designated Expert Instructions To enable a broadly informed review of registration decisions, it is recommended that multiple Designated Experts be appointed who are able to represent the perspectives of both the transport and security areas. In cases where a registration decision could be perceived as creating a conflict of interest for a particular Expert, that Expert SHOULD defer to the judgment of the other Experts.

Acknowledgments We would like to thank Eric Rescorla, Hanno Becker, , Joe Clarke, , Marco Tiloca, Martin Thomson, Mike Ounsworth, Mohit Sahni, Rich Salz, Russ Housley, Sean Turner, and Yaron Sheffer for their input to this document.