Datagram Transport Layer Security (DTLS) over Stream Control Transmission Protocol (SCTP)

1.1. Overview

This document describes the usage of the Datagram Transport Layer Security (DTLS) protocol, as defined in DTLS 1.2 [RFC6347], and DTLS 1.3 [I-D.ietf-tls-dtls13], over the Stream Control Transmission Protocol (SCTP), as defined in [RFC4960] with Authenticated Chunks for SCTP (SCTP-AUTH) [RFC4895].¶

This specification provides mutual authentication of endpoints, confidentiality, integrity protection, and replay protection of user messages for applications that use SCTP as their transport protocol. Thus, it allows client/server applications to communicate in a way that is designed to give communications privacy and to prevent eavesdropping and detect tampering or message forgery. DTLS/SCTP uses DTLS for mutual authentication, key exchange with forward secrecy for SCTP-AUTH, and confidentiality of user messages. DTLS/SCTP use SCTP and SCTP-AUTH for integrity protection and replay protection of user messages.¶

Applications using DTLS over SCTP can use almost all transport features provided by SCTP and its extensions. DTLS/SCTP supports:¶

• preservation of message boundaries.¶
• a large number of unidirectional and bidirectional streams.¶
• ordered and unordered delivery of SCTP user messages.¶
• the partial reliability extension as defined in [RFC3758].¶
• the dynamic address reconfiguration extension as defined in [RFC5061].¶
• User messages of any size.¶

The method described in this document requires that the SCTP implementation supports the optional feature of fragmentation of SCTP user messages as defined in [RFC4960]. The implementation is required to have an SCTP API (for example the one described in [RFC6458]) that supports partial user message delivery and also recommended that I-DATA chunks as defined in [RFC8260] is used to efficiently implement and support larger user messages.¶

To simplify implementation and reduce the risk for security holes, limitations have been defined such that STARTTLS as specified in [RFC3788] is no longer supported.¶

1.2. DTLS Version

Using DTLS 1.2 instead of using DTLS 1.0 limits the lifetime of a DTLS connection and the data volume which can be transferred over a DTLS connection. This is caused by:¶

• The number of renegotiations in DTLS 1.2 is limited to 65534 compared to unlimited in DTLS 1.0.¶
• While the AEAD limits in DTLS 1.3 does not formally apply to DTLS 1.2 the mathematical limits apply equally well to DTLS 1.2.¶

DTLS 1.3 comes with a large number of significant changes.¶

• Renegotiations are not supported and instead partly replaced by KeyUpdates. The number of KeyUpdates is limited to 2^64.¶
• Strict AEAD significantly limits on how much many packets can be sent before rekeying.¶

Many applications using DTLS/SCTP are of semi-permanent nature and use SCTP associations with expected lifetimes of months or even years, and where there is a significant cost of bringing down the SCTP association in order to restart it. Such DTLS/SCTP usages that need:¶

• Periodic re-authentication and transfer of revocation information of both endpoints (not only the DTLS client).¶
• Periodic rerunning of Diffie-Hellman key-exchange to provide forward secrecy and mitigate static key exfiltration attacks.¶
• Perform SCTP-AUTH rekeying.¶

At the time of publication DTLS 1.3 does not support any of these, where DTLS 1.2 renegotiation functionality can provide this functionality in the context of DTLS/SCTP. To address these requirements from semi-permanent applications, this document use several overlapping DTLS connections with either DTLS 1.2 or 1.3. Having uniform procedures reduces the impact when upgrading from 1.2 to 1.3 and avoids using the renegotiation mechanism which is disabled by default in many DTLS implementations.¶

To address known vulnerabilities in DTLS 1.2 this document describes and mandates implementation constraints on ciphers and protocol options. The DTLS 1.2 renegotiation mechanism is forbidden to be used as it creates need for additional mechanism to handle race conditions and interactions between using DTLS connections in parallel.¶

In the rest of the document, unless the version of DTLS is specifically called out the text applies to both versions of DTLS.¶

1.3. Terminology

This document uses the following terms:¶

Association: An SCTP association.¶

Connection: An DTLS connection. It is uniquely identified by a connection identifier.¶

Stream: A unidirectional stream of an SCTP association. It is uniquely identified by a stream identifier.¶

1.4. Abbreviations

AEAD: Authenticated Encryption with Associated Data¶

DTLS: Datagram Transport Layer Security¶

HMAC: Keyed-Hash Message Authentication Code¶

MTU: Maximum Transmission Unit¶

PPID: Payload Protocol Identifier¶

SCTP: Stream Control Transmission Protocol¶

SCTP-AUTH: Authenticated Chunks for SCTP¶

TCP: Transmission Control Protocol¶

TLS: Transport Layer Security¶

ULP: Upper Layer Protocol¶

3.1. Version of DTLS

This document defines the usage of either DTLS 1.3 [I-D.ietf-tls-dtls13], or DTLS 1.2 [RFC6347]. Earlier versions of DTLS MUST NOT be used (see [RFC8996]). DTLS 1.3 is RECOMMENDED for security and performance reasons. It is expected that DTLS/SCTP as described in this document will work with future versions of DTLS.¶

3.2. Cipher Suites and Cryptographic Parameters

For DTLS 1.2, the cipher suites forbidden by [RFC7540] MUST NOT be used. For all versions of DTLS, cryptographic parameters giving confidentiality and forward secrecy MUST be used.¶

3.3. Message Sizes

DTLS/SCTP, automatically fragments and reassembles user messages. This specification defines how to fragment the user messages into DTLS records, where each DTLS record allows a maximum of 2^14 protected bytes. Each DTLS record adds some overhead, thus using records of maximum possible size are recommended to minimize the transmitted overhead. DTLS 1.3 has much less overhead than DTLS 1.2 per record.¶

The sequence of DTLS records is then fragmented into DATA or I-DATA Chunks to fit the path MTU by SCTP. These changes ensures that DTLS/SCTP has the same capability as SCTP to support user messages of any size. However, to simplify implementations it is OPTIONAL to support user messages larger than 2^64-1 bytes. This is to allow implementation to assume that 64-bit length fields and offset pointers will be sufficient.¶

Another implementation dependent exception to the support of any user message size is the SCTP-API defined in [RFC6458]. That API does not allow changing the SCTP-AUTH key used to send a particular user message. Thus, the user message size must be limited such that completion of the user message can occur within a short time frame from the establishment of the new DTLS connection (Section 4.7).¶

The security operations and reassembly process requires that the protected user message, i.e., with DTLS record overhead, is buffered in the receiver. This buffer space will thus put a limit on the largest size of plain text user message that can be transferred securely. However, by mandating the use of the partial delivery of user messages from SCTP and assuming that no two messages received on the same stream are interleaved (as it is the case when using the API defined in [RFC6458]) the required buffering prior to DTLS processing can be limited to a single DTLS record per used incoming stream. This enables the DTLS/SCTP implementation to provide the Upper Layer Protocol (ULP) with each DTLS record's content when it has been decrypted and its integrity been verified enabling partial user message delivery to the ULP. Implementations can trade-off buffer memory requirements in the DTLS layer with transport overhead by using smaller DTLS records.¶

The DTLS/SCTP implementation is expected to behave very similar to just SCTP when it comes to handling of user messages and dealing with large user messages and their reassembly and processing. Making it the ULP responsible for handling any resource contention related to large user messages.¶

3.4. Replay Protection

SCTP-AUTH [RFC4895] does not have explicit replay protection. However, the combination of SCTP-AUTH's protection of DATA or I-DATA chunks and SCTP user message handling will prevent third party attempts to inject or replay SCTP packets resulting in impact on the received protected user message. In fact, this document's solution is dependent on SCTP-AUTH and SCTP to prevent reordering, duplication, and removal of the DTLS records within each protected user message. This includes detection of changes to what DTLS records start and end the SCTP user message, and removal of DTLS records before an increment to the epoch. Without SCTP-AUTH, these would all have required explicit handling.¶

DTLS optionally supports record replay detection. Such replay detection could result in the DTLS layer dropping valid messages received outside of the DTLS replay window. As DTLS/SCTP provides replay protection even without DTLS replay protection, the replay detection of DTLS MUST NOT be used.¶

3.5. Path MTU Discovery

DTLS Path MTU Discovery MUST NOT be used. Since SCTP provides Path MTU discovery and fragmentation/reassembly for user messages, and specified in Section 3.3, DTLS can send maximum sized DTLS Records.¶

3.6. Retransmission of Messages

SCTP provides a reliable and in-sequence transport service for DTLS messages that require it. See Section 4.4. Therefore, DTLS procedures for retransmissions MUST NOT be used.¶

4.1. Mapping of DTLS Records

The SCTP implementation MUST support fragmentation of user messages using DATA [RFC4960], and optionally I-DATA [RFC8260] chunks.¶

DTLS/SCTP works as a shim layer between the user message API and SCTP. On the sender side a user message is split into fragments m0, m1, m2, each no larger than 2^14 = 16384 bytes.¶

   m0 | m1 | m2 | ... = user_message

The resulting fragments are protected with DTLS and the records are concatenated¶

   user_message' = DTLS( m0 ) | DTLS( m1 ) | DTLS( m2 ) ...

The new user_message', i.e., the protected user message, is the input to SCTP.¶

On the receiving side, the length field in each DTLS record can be used to determine the boundaries between DTLS records. DTLS can decrypt individual records or a concatenated sequence of records. The last DTLS record can be found by subtracting the length of individual records from the length of user_message'. Whether to decrypt individual records, sequences of records, or the whole user_message' is left to the implementation. The output from the DTLS decryption(s) is the fragments m0, m1, m2 ... The user_message is reassembled from decrypted DTLS records as user_message = m0 | m1 | m2 ... There are three failure cases an implementation needs to detect and then act on:¶

1. Failure in decryption and integrity verification process of any DTLS record. Due to SCTP-AUTH preventing delivery of injected or corrupt fragments of the protected user message this should only occur in case of implementation errors or internal hardware failures or the necessary security context has been prematurely discarded.¶
2. In case the SCTP layer indicates an end to a user message, e.g., when receiving a MSG_EOR in a recvmsg() call when using the API described in [RFC6458], and the last buffered DTLS record length field does not match, i.e., the DTLS record is incomplete.¶
3. Unable to perform the decryption processes due to lack of resources, such as memory, and have to abandon the user message fragment. This specification is defined such that the needed resources for the DTLS/SCTP operations are bounded for a given number of concurrent transmitted SCTP streams or unordered user messages.¶

The above failure cases all result in the receiver failing to recreate the full user message. This is a failure of the transport service that is not possible to recover from in the DTLS/SCTP layer and the sender could believe the complete message have been delivered. This error MUST NOT be ignored, as SCTP lacks any facility to declare a failure on a specific stream or user message, the DTLS connection and the SCTP association SHOULD be terminated. A valid exception to the termination of the SCTP association is if the receiver is capable of notifying the ULP about the failure in delivery and the ULP is capable of recovering from this failure.¶

Note that if the SCTP extension for Partial Reliability (PR-SCTP) [RFC3758] is used for a user message, user message may be partially delivered or abandoned. These failures are not a reason for terminating the DTLS connection and SCTP association.¶

The DTLS Connection ID MUST be negotiated ([I-D.ietf-tls-dtls-connection-id] or Section 9 of [I-D.ietf-tls-dtls13]). If DTLS 1.3 is used, the length field in the record layer MUST be included in all records. A 16-bit sequence number SHOULD be used rather than 8-bit to minimize issues with DTLS record sequence number wrapping.¶

The ULP may use multiple messages simultanous, and the progress and delivery of these messages are progressing indepentely, thus the recieving DTLS/SCTP implementation may not receive records in order in case of packet loss. Assuming that the sender will send the DTLS records in order the DTLS records where created (which may not be certain in some implementations), then there is a risk that DTLS sequence number have wrapped if the amount of data in flight is more than the sequence number covers. Thus, for 8-bit sequence number space with 16384 bytes records the receiver window only needs to be 256*16384 = 4,194,304 bytes for this risk to defintely exist. While a 16-bit sequence number should not have any sequence number wraps for receiver windows up to 1 Gbyte. The DTLS/SCTP may not be tightly integrated and the DTLS records may not be requested to be sent in strict sequence order, in these case additional guard ranges are needed.¶

Also, if smaller DTLS records are used, this limit will be correspondingly reduced. The DTLS/SCTP Sender needs to choose sequence number length and DTLS Record size so that the product is larger than the used receiver window, preferably twice as large. Receiver implementations that are offering receiver windows larger than the product 65536*16384 bytes MUST be capable of handling sequence number wraps through trial decoding with a lower values in the higher bits of the extended sequence number.¶

Section 4 of [I-D.ietf-tls-dtls-connection-id] states "If, however, an implementation chooses to receive different lengths of CID, the assigned CID values must be self-delineating since there is no other mechanism available to determine what connection (and thus, what CID length) is in use.". As this solution requires multiple connection IDs, using a zero-length CID will be highly problematic as it could result in that any DTLS records with a zero length CID ends up in another DTLS connection context, and there fail the decryption and integrity verification. And in that case to avoid losing the DTLS record, it would have to be forwarded to the zero-length CID using DTLS Connection and decryption and validation must be tried. Resulting in higher resource utilization. Thus, it is RECOMMENDED to not use the zero length CID values and instead use a single common length for the CID values. A single byte should be sufficient, as reuse of old CIDs is possible as long as the implementation ensure they are not used in near time to the previous usage.¶

4.2. DTLS Connection Handling

DTLS/SCTP is negotiated on SCTP level as an adaptation layer Section 5. After a succesful negotiation of the DTLS/SCTP during SCTP association establishment, a DTLS connection MUST be established prior to transmission of any ULP user messages. All DTLS connections are terminated when the SCTP association is terminated. A DTLS connection MUST NOT span multiple SCTP associations.¶

As it is required to establish the DTLS connection at the beginning of the SCTP association, either of the peers should never send any SCTP user messages that are not protected by DTLS. So, the case that an endpoint receives data that is not either DTLS messages or protected user messages in the form of a sequence of DTLS Records on any stream is a protocol violation. The receiver MAY terminate the SCTP association due to this protocol violation. Implementations that does not have a DTLS endpoint immediately ready on SCTP handshake completion will have to ensure correct caching of the messages until the DTLS endpoint is ready.¶

Whenever a mutual authentication, updated security parameters, rerun of Diffie-Hellman key-exchange , or SCTP-AUTH rekeying is needed, a new DTLS connection is instead setup in parallel with the old connection (i.e., there may be up to two simultaneous DTLS connections within one association).¶

4.3. Payload Protocol Identifier Usage

SCTP Payload Protocol Identifiers are assigned by IANA. Application protocols using DTLS over SCTP SHOULD register and use a separate Payload Protocol Identifier (PPID) and SHOULD NOT reuse the PPID that they registered for running directly over SCTP.¶

Using the same PPID does no harm as DTLS/SCTP requires all user mesages being DTLS protected and knows that DTLS is used. However, for protocol analyzers, for example, it is much easier if a separate PPID is used and avoids different behavior from [RFC6083]. This means, in particular, that there is no specific PPID for DTLS.¶

Messages that are exchanged between DTLS/SCTP peers not containing ULP user messages shall use PPID=0 according to section 3.3.1 of [RFC4960] as no application identifier can be specified by the upper layer for this payload data.¶

4.4. Stream Usage

DTLS 1.3 protects the actual content type of the DTLS record and have therefore omitted the non-protected content type field. Thus, it is not possible to determine which content type the DTLS record has on SCTP level. For DTLS 1.2 ULP user messages will be carried in DTLS records with content type "application_data".¶

DTLS Records carrying protected user message fragments MUST be sent in the by ULP indicated SCTP stream and user message. The ULP has no limitations in using SCTP facilities for stream and user messages. DTLS records of other types MAY be sent on any stream. It MAY also be sent in its own SCTP user message as well as interleaved with other DTLS records carrying protected user messages. Thus, it is allowed to insert between protected user message fragments DTLS records of other types as the DTLS receiver will process these and not result in any user message data being inserted into the ULP's user message. However, DTLS messages of other types than protected user message MUST be sent reliable, so the DTLS record can only be interleaved in case the ULP user message is sent as reliable.¶

DTLS is capable of handling reordering of the DTLS records. However, depending on stream properties and which user message DTLS records of other types are sent in may impact in which order and how quickly they are possible to process. Using a stream with in-order delivery will ensure that the DTLS Records are delivered in the order they are sent in user messages. Thus, ensuring that if there are DTLS records that need to be delivered in particular order it can be ensured. Alternatively, if it is desired that a DTLS record is delvired as early as possible avoiding in-order streams with queued messages and considering stream priorities can result in faster delviery.¶

A simple solution avoiding any protocol issue are to send all DTLS messages that are not protected user message fragments is to pick a stream not used by the ULP, send the DTLS messages in their own user messages with in order delivery. That mimics the RFC 6083 behavior without impacting the ULP.¶

4.5. Chunk Handling

DATA chunks of SCTP MUST be sent in an authenticated way as described in SCTP-AUTH [RFC4895]. All other chunks that can be authenticated, i.e., all chunk types that can be listed in the Chunk List Parameter [RFC4895], MUST also be sent in an authenticated way. This makes sure that an attacker cannot modify the stream in which a message is sent or affect the ordered/unordered delivery of the message.¶

If PR-SCTP as defined in [RFC3758] is used, FORWARD-TSN chunks MUST also be sent in an authenticated way as described in [RFC4895]. This makes sure that it is not possible for an attacker to drop messages and use forged FORWARD-TSN, SACK, and/or SHUTDOWN chunks to hide this dropping.¶

I-DATA chunk type as defined in [RFC8260] is RECOMMENDED to be supported to avoid some of the down sides that large user messages have on blocking transmission of later arriving high priority user messages. However, the support is not mandated and negotiated independently from DTLS/SCTP. If I-DATA chunks are used, then they MUST be sent in an authenticated way as described in [RFC4895].¶

4.6. SCTP-AUTH Hash Function

When using DTLS/SCTP, the SHA-256 Message Digest Algorithm MUST be supported in the SCTP-AUTH [RFC4895] implementation. SHA-1 MUST NOT be used when using DTLS/SCTP. [RFC4895] requires support and inclusion of SHA-1 in the HMAC-ALGO parameter, thus, to meet both requirements the HMAC-ALGO parameter will include both SHA-256 and SHA-1 with SHA-256 listed prior to SHA-1 to indicate the preference.¶

4.7. Parallel DTLS connections

To enable SCTP-AUTH rekeying, periodic authentication of both endpoints, and force attackers to dynamic key extraction [RFC7624], DTLS/SCTP per this specification defines the usage of parallel DTLS connections over the same SCTP association. This solution ensures that there are no limitations to the lifetime of the SCTP association due to DTLS, it also avoids dependency on version specific DTLS mechanisms such as renegotiation in DTLS 1.2, which is disabled by default in many DTLS implementations, or post-handshake messages in DTLS 1.3, which does not allow periodic mutual endpoint re-authentication or re-keying of SCTP-AUTH. Parallel DTLS connections enable opening a new DTLS connection performing a handshake, while the existing DTLS connection is kept in place. In DTLS 1.3 the handshake MAY be a full handshake or a resumption handshake and resumption can be done while the original connection is still open. In DTLS 1.2 the handshake MUST be a full handshake. On handshake completion switch to the security context of the new DTLS connection and then ensure delivery of all the SCTP chunks using the old DTLS connections security context. When that has been achieved close the old DTLS connection and discard the related security context.¶

As specified in Section 4.1 the usage of DTLS connection ID is required to ensure that the receiver can correctly identify the DTLS connection and its security context when performing its de-protection operations. There is also only a single SCTP-AUTH key exported per DTLS connection ensuring that there is clear mapping between the DTLS connection ID and the SCTP-AUTH security context for each key-id.¶

Application writers should be aware that establishing a new DTLS connections may result in changes of security parameters. See Section 9 for security considerations regarding rekeying.¶

A DTLS/SCTP Endpoint MUST NOT have more than two DTLS connections open at the same time. Either of the endpoints MAY initiate a new DTLS connection by performing a full DTLS handshake. As either endpoint can initiate a DTLS handshake on either side at the same time, either endpoint may receive a DTLS ClientHello when it has sent its own ClientHello. In this case the ClientHello from the endpoint that had the DTLS Client role in the establishment of the existing DTLS connection shall be continued to be processed and the other dropped.¶

When performing the DTLS handshake the endpoint MUST verify that the peer identifies using the same identity as in the previous DTLS connection.¶

When the DTLS handshake has been completed, a new SCTP-AUTH key will be exported per Section 4.10 and the new DTLS connection MUST be used for the DTLS protection operation of any future protected ULP user message. The endpoint is RECOMMENDED to use the security context of the new DTLS connection for any DTLS protection operation occurring after the completed handshake. The new SCTP-AUTH key SHALL be used for any SCTP user message being sent after the DTLS handshake has completed. There is a possibility to use the new SCTP-AUTH key for any SCTP packets part of an SCTP user message that was initiated but not yet fully transmitted prior to the completion of the new DTLS handshake, however the API defined in [RFC6458] is not supporting switching the SCTP-AUTH key on the sender side. Any SCTP-AUTH receiver implementation is expected to be able to select key on SCTP packet basis.¶

The DTLS/SCTP endpoint will indicate to its peer when the previous DTLS connection and its context are no longer needed for receiving any more data from this endpoint. This is done by having DTLS to send a DTLS close_notify alert. The endpoint MUST NOT send the close_notify until the following two conditions are fulfilled:¶

1. All SCTP packets containing part of any DTLS record or message protected using the security context of this DTLS connection have been acknowledged in a non-renegable way.¶
2. All SCTP packets using the SCTP-AUTH key associated with the security context of this DTLS connection have been acknowledged in a non-renegable way.¶

Note: For DTLS 1.2 receiving Close_notify will close the DTLS connection for further writes and requires the immediate generation of a Close_notify. Thus, this forces the DTLS/SCTP to protect any buffered data on DTLS/SCTP layer not yet protected to use the new DTLS connection. In addition the DTLS/SCTP layer will have to buffer the close_notify generated by the shuting down DTLS connection and also not discard the SCTP-AUTH key until it has fulfilled the delivery of the data protected by the closing DTLS connection security context.¶

SCTP implementations exposing APIs like [RFC6458] fulfilling these conditions requires draining the SCTP association of all outstanding data after having completed all the user messages using the previous SCTP-AUTH key identifier. Relying on the SCTP_SENDER_DRY_EVENT to know when delivery has been accomplished. A richer API could also be used that allows user message level tracking of delivery, see Section 6 for API considerations.¶

For SCTP implementations exposing APIs like [RFC6458] where it is not possible to change the SCTP-AUTH key for a partial SCTP message initiated before the change of security context will be forced to track the SCTP messages and determine when all using the old security context has been transmitted. This maybe be impossible to do completely reliable without tighter integration between the DTLS/SCTP layer and the SCTP implementation. This type of implementations also has an implicit limitation in how large SCTP messages it can support. Each SCTP message needs have completed delivery and enabling closing of the previous DTLS connection prior to the need to create yet another DTLS connection. Thus, SCTP messages can't be larger than that the transmission completes in less than the duration between the rekeying or re-authentications needed for this SCTP association.¶

The consequences of sending a DTLS close_notify alert in the old DTLS connection prior to the receiver having received the data can result in failure case 1 described in Section 4.1, which likely result in SCTP association termination.¶

4.8. Renegotiation and KeyUpdate

DTLS 1.2 renegotiation enables rekeying (with ephemeral Diffie- Hellman) of DTLS as well as mutual reauthentication and transfer of revocation information inside an DTLS 1.2 connection. Renegotiation has been removed from DTLS 1.3 and partly replaced with post-handshake messages such as KeyUpdate. The parallel DTLS connection solution was specified due to lack of necessary features with DTLS 1.3 considered needed for long lived SCTP associations, such as rekeying (with ephemeral Diffie-Hellman) as well as mutual reauthentication.¶

This specification do not allow usage of DTLS 1.2 renegotiation to avoid race conditions and corner cases in the interaction between the parallel DTLS connection mechanism and the keying of SCTP-AUTH. In addtion renegotiation is also disabled in implementation, as well as dealing with the epoch change reliable have similar or worse applicaiton impact.¶

This specification also recommends against using DTLS 1.3 KeyUpdate and instead rely on parallel DTLS connections. For DTLS 1.3 there isn't feature parity. It also have the issue that a DTLS implementation following the RFC may assume a too limited window for SCTP where the previous epoch's security context is maintained and thus changes to epoch handling (Section 4.9) are necessary. Thus, unless the below specified more application impacting draining is used there exist risk of losing data that the sender will have assumed has been reliably delivered.¶

4.9. DTLS Epochs

In general, DTLS implementations SHOULD discard records from earlier epochs. However, in the context of a reliable communication this is not appropriate.¶

4.10. Handling of Endpoint-Pair Shared Secrets

SCTP-AUTH [RFC4895] is keyed using Endpoint-Pair Shared Secrets. In SCTP associations where DTLS is used, DTLS is used to establish these secrets. The endpoints MUST NOT use another mechanism for establishing shared secrets for SCTP-AUTH. The endpoint-pair shared secret for Shared Key Identifier 0 is empty and MUST be used when establishing the first DTLS connection.¶

The initial DTLS connection will be used to establish a new shared secret as specified per DTLS version below, and which MUST use shared key identifier 1. After sending the DTLS Finished message for the initial DTLS connection, the active SCTP-AUTH key MUST be switched from key identifier 0 to key identifier 1. Once the initial Finished message from the peer has been processed by DTLS, the SCTP-AUTH key with Shared Key Identifier 0 MUST be removed.¶

When a subsequent DTLS connection is setup, a new a 64-byte shared secret is derived using the TLS-Exporter. The shared secret identifiers form a sequence. If the previous shared secret used Shared Key Identifier n, the new one MUST use Shared Key Identifier n+1, unless n= 65535, in which case the new Shared Key Identifier is 1.¶

After sending the DTLS Finished message, the active SCTP-AUTH key MUST be switched to the new one. When the endpoint has both sent and received a closeNotify on the old DTLS connection then the endpoint SHALL remove shared secret(s) related to old DTLS connection.¶

Whenever a new DTLS connection is established, a 64-byte
endpoint-pair shared secret is derived using the TLS-Exporter
described in {{RFC5705}}.

4.11. Shutdown

To prevent DTLS from discarding DTLS user messages while it is shutting down, the below procedure has been defined. Its goal is to avoid the need for APIs requiring per user message data level acknowledgments and utilizes existing SCTP protocol behavior to ensure delivery of the protected user messages data.¶

Note, this proceudre currenlty only works for DTLS 1.3. For DTLS 1.2 users the remote endpoint will be closed for sending more data with the reception of the close_notify in step 5, and step 6 will not be possible and that data will be lost.¶

The interaction between peers and protocol stacks shall be as follows:¶

1. Local instance of ULP asks for terminating the DTLS/SCTP Association.¶
2. Local DTLS/SCTP acknowledge the request, from this time on no new data from local instance of ULP will be accepted. In case a DTLS connection handshake is ongoing this needs to be aborted conclusively at this step to ensure that the necessary DTLS message exchange happens prior to draining any outstanding data in the SCTP association from this endpoint.¶
3. Local DTLS/SCTP finishes any protection operation on buffered user messages and ensures that all protected user message data has been successfully transferred to the remote ULP.¶
4. Local DTLS/SCTP sends DTLS Close_notify to remote instance of DTLS/SCTP on each and all DTLS connections, keys and session state are kept for processing packets received later on.¶
5. When receiving Close_notify on the last open DTLS connection, remote DTLS/SCTP instance informs its ULP that remote shutdown has been initiated. When two parallel DTLS connections are in place it is important to await Close_notify alert on both to not misstake a rekeying. No more ULP user message data to be sent to peer can be accepted by DTLS/SCTP. In case this endpoint has initiated and DTLS connection handshake this MUST be aborted as the peer is unable to respond.¶
6. Remote DTLS/SCTP finishes any protection operation on buffered user messages and ensures that all protected user message data has been successfully transferred to the remote ULP.¶
7. Remote DTLS/SCTP sends Close_notify to Local DTLS/SCTP entity for each and all DTLS connections.¶
8. When receiving Close_notify on the last open DTLS connection, local DTLS/SCTP instance initiates the SCTP shutdown procedure (section 9.2 of [RFC4960]).¶
9. Remote DTLS/SCTP replied to the SCTP shutdown procedure (section 9.2 of [RFC4960]).¶
10. Upon receiving the information that SCTP has closed the Association, independently the local and remote DTLS/SCTP entities destroy the DTLS connection.¶

The verification in step 3 and 6 that all user data message has been successfully delivered to the remote ULP can be provided by the SCTP stack that implements [RFC6458] by means of SCTP_SENDER_DRY event (section 6.1.9 of [RFC6458]).¶

A successful SCTP shutdown will indicate successful delivery of all data. However, in cases of communication failures and extensive packet loss the SCTP shutdown procedure can time out and result in SCTP association termination where its unknown if all data has been delivered. The DTLS/SCTP should indicate to ULP successful completion or failure to shutdown gracefully.¶

5.1. Adaptation Layer Indication in INIT/INIT-ACK

At the initialization of the association, a sender of the INIT or INIT ACK chunk that intends to use DTLS/SCTP as specified in this specification MUST include an Adaptation Layer Indication Parameter with the IANA assigned value TBD (Section 8.2) to inform its peer that it is able to support DTLS over SCTP per this specification.¶

5.2. DTLS over SCTP Initialization

Initialization of DTLS/SCTP requires all the following options to be part of the INIT/INIT-ACK handshake:¶

RANDOM: defined in [RFC4895]¶

CHUNKS: list of permitted chunks, defined in [RFC4895]¶

HMAC-ALGO: defined in [RFC4895]¶

ADAPTATION-LAYER-INDICATION: defined in [RFC5061]¶

When all the above options are present and having acceptable parameters, the Association will start with support of DTLS/SCTP. The set of options indicated are the DTLS/SCTP Mandatory Options. No data transfer is permitted before DTLS handshake is complete. Chunk bundling is permitted according to [RFC4960]. The DTLS handshake will enable authentication of both the peers.¶

The extension described in this document is given by the following message exchange.¶

   --- INIT[RANDOM; CHUNKS; HMAC-ALGO; ADAPTATION-LAYER-IND] --->
   <- INIT-ACK[RANDOM; CHUNKS; HMAC-ALGO; ADAPTATION-LAYER-IND] -
   ------------------------ COOKIE-ECHO ------------------------>
   <------------------------ COOKIE-ACK -------------------------
   ---------------- AUTH; DATA[DTLS Handshake] ----------------->
                               ...
                               ...
   <--------------- AUTH; DATA[DTLS Handshake] ------------------

5.3. Client Use Case

When a client initiates an SCTP Association with DTLS protection, i.e., the SCTP INIT containing DTSL/SCTP Mandatory Options, it can receive an INIT-ACK also containing DTLS/SCTP Mandatory Options, in that case the Association will proceed as specified in the previous Section 5.2 section. If the peer replies with an INIT-ACK not containing all DTLS/SCTP Mandatory Options, the client SHOULD reply with an SCTP ABORT.¶

5.4. Server Use Case

If a SCTP Server supports DTLS/SCTP, i.e., per this specification, when receiving an INIT chunk with all DTLS/SCTP Mandatory Options it will reply with an INIT-ACK also containing all the DTLS/SCTP Mandatory Options, following the sequence for DTLS initialization Section 5.2 and the related traffic case. If a SCTP Server that supports DTLS and configured to use it, receives an INIT chunk without all DTLS/SCTP Mandatory Options, it SHOULD reply with an SCTP ABORT.¶

5.5. RFC 6083 Fallback

This section discusses how an endpoint supporting this specification can fallback to follow the DTLS/SCTP behavior in RFC6083. It is recommended to define a setting that represents the policy to allow fallback or not. However, the possibility to use fallback is based on the ULP can operate using user messages that are no longer than 16384 bytes and where the security issues can be mitigated or considered acceptable. Fallback is NOT RECOMMEND to be enabled as it enables downgrade attacks to weaker algorithms and versions of DTLS.¶

An SCTP endpoint that receives an INIT chunk or an INIT-ACK chunk that does not contain the SCTP-Adaptation-Indication parameter with the DTLS/SCTP adaptation layer codepoint, see Section 8.2, may in certain cases potentially perform a fallback to RFC 6083 behavior. However, the fallback attempt should only be performed if policy says that is acceptable.¶

If fallback is allowed, it is possible that the client will send plain text user messages prior to DTLS handshake as it is allowed per RFC 6083. So that needs to be part of the consideration for a policy allowing fallback.¶

7.1. Socket Option to Get the HMAC Identifier being Sent (SCTP_SEND_HMAC_IDENT)

During the SCTP association establishment a HMAC Identifier is selected which is used by an SCTP endpoint when sending AUTH chunks. An application can access the result of this selection by using this read-only socket option, which uses the level IPPROTO_SCTP and the name SCTP_SEND_HMAC_IDENT.¶

The following structure is used to access HMAC Identifier used for sending AUTH chunks:¶

struct sctp_assoc_value {
    sctp_assoc_t assoc_id;
    uint32_t assoc_value;
};

assoc_id:

This parameter is ignored for one-to-one style sockets. For one-to-many style sockets, the application fills in an association identifier. It is an error to use SCTP_{FUTURE|CURRENT|ALL}_ASSOC in assoc_id.¶

assoc_value:

This parameter contains the HMAC Identifier used for sending AUTH chunks.¶

7.2. Exposing the HMAC Identifiers being Received

Section 6.1.8 of [RFC6458] defines the SCTP_AUTHENTICATION_EVENT event, which uses the following structure:¶

struct sctp_authkey_event {
    uint16_t auth_type;
    uint16_t auth_flags;
    uint32_t auth_length;
    uint16_t auth_keynumber;
    uint32_t auth_indication;
    sctp_assoc_t auth_assoc_id;
};

This document updates this structure to¶

struct sctp_authkey_event {
    uint16_t auth_type;
    uint16_t auth_flags;
    uint32_t auth_length;
    uint16_t auth_identifier; /* formerly auth_keynumber */
    uint32_t auth_indication;
    sctp_assoc_t auth_assoc_id;
};

by renaming auth_keynumber to auth_identifier. auth_identifier just replaces auth_keynumber in the context of [RFC6458]. In addition to that, the SCTP_AUTHENTICATION_EVENT event is extended to also indicate when a new HMAC Identifier is received and such reporting is explicitly enabled as described in Section 7.3. In this case auth_indication is SCTP_AUTH_NEW_HMAC and the new HMAC identifier is reported in auth_identifier.¶

7.3. Socket Option to Expose HMAC Identifier Usage (SCTP_EXPOSE_HMAC_IDENT_CHANGES)

This options allows the application to enable and disable the reception of SCTP_AUTHENTICATION_EVENT events when a new HMAC Identifiers has been received in an AUTH chunk (see Section 7.2). This read/write socket option uses the level IPPROTO_SCTP and the name SCTP_EXPOSE_HMAC_IDENT_CHANGES. It is needed to provide backwards compatibility and the default is that these events are not reported.¶

The following structure is used to enable or disable the reporting of newly received HMAC Identifiers in AUTH chunks:¶

struct sctp_assoc_value {
    sctp_assoc_t assoc_id;
    uint32_t assoc_value;
};

assoc_id:

This parameter is ignored for one-to-one style sockets. For one-to-many style sockets, the application may fill in an association identifier or SCTP_{FUTURE|CURRENT|ALL}_ASSOC.¶

assoc_value:

Newly received HMAC Identifiers are reported if, and only if, this parameter is non-zero.¶

8.1. TLS Exporter Label

RFC 6083 defined a TLS Exporter Label registry as described in [RFC5705]. IANA is requested to update the reference for the label "EXPORTER_DTLS_OVER_SCTP" to this specification.¶

8.2. SCTP Adaptation Layer Indication Code Point

[RFC5061] defined a IANA registry for Adaptation Code Points to be used in the Adaptation Layer Indication parameter. The registry was at time of writing located: https://www.iana.org/assignments/sctp-parameters/sctp-parameters.xhtml#sctp-parameters-27 IANA is requested to assign one Adaptation Code Point for DTLS/SCTP per the below proposed entry in Table 1.¶

RFC-Editor Note: Please replace [RFC-TBD] with the RFC number given to this specification.¶

9.1. Cryptographic Considerations

Over the years, there have been several serious attacks on earlier versions of Transport Layer Security (TLS), including attacks on its most commonly used ciphers and modes of operation. [RFC7457] summarizes the attacks that were known at the time of publishing and BCP 195 [RFC7525] [RFC8996] provide recommendations for improving the security of deployed services that use TLS.¶

When DTLS/SCTP is used with DTLS 1.2 [RFC6347], DTLS 1.2 MUST be configured to disable options known to provide insufficient security. HTTP/2 [RFC7540] gives good minimum requirements based on the attacks that where publicly known in 2015. DTLS 1.3 [I-D.ietf-tls-dtls13] only define strong algorithms without major weaknesses at the time of publication. Many of the TLS registries have a "Recommended" column. Parameters not marked as "Y" are NOT RECOMMENDED to support. DTLS 1.3 is preferred over DTLS 1.2 being a newer protocol that addresses known vulnerabilities and only defines strong algorithms without known major weaknesses at the time of publication.¶

DTLS 1.3 requires rekeying before algorithm specific AEAD limits have been reached. The AEAD limits equations are equally valid for DTLS 1.2 and SHOULD be followed for DTLS/SCTP, but are not mandated by the DTLS 1.2 specification.¶

HMAC-SHA-256 as used in SCTP-AUTH has a very large tag length and very good integrity properties. The SCTP-AUTH key can be used longer than the current algorithms in the TLS record layer. The SCTP-AUTH key is rekeyed when a new DTLS connection is set up at which point a new SCTP-AUTH key is derived using the TLS-Exporter.¶

(D)TLS 1.3 [RFC8446] discusses forward secrecy from EC(DHE), KeyUpdate, and tickets/resumption. Forward secrecy limits the effect of key leakage in one direction (compromise of a key at time T2 does not compromise some key at time T1 where T1 < T2). Protection in the other direction (compromise at time T1 does not compromise keys at time T2) can be achieved by rerunning EC(DHE). If a long-term authentication key has been compromised, a full handshake with EC(DHE) gives protection against passive attackers. If the resumption_master_secret has been compromised, a resumption handshake with EC(DHE) gives protection against passive attackers and a full handshake with EC(DHE) gives protection against active attackers. If a traffic secret has been compromised, any handshake with EC(DHE) gives protection against active attackers.¶

The document "Confidentiality in the Face of Pervasive Surveillance: A Threat Model and Problem Statement" [RFC7624] defines key exfiltration as the transmission of cryptographic keying material for an encrypted communication from a collaborator, deliberately or unwittingly, to an attacker. Using the terms in RFC 7624, forward secrecy without rerunning EC(DHE) still allows an attacker to do static key exfiltration. Rerunning EC(DHE) forces and attacker to dynamic key exfiltration (or content exfiltration).¶

When using DTLS 1.3 [I-D.ietf-tls-dtls13], AEAD limits and forward secrecy can be achieved by sending post-handshake KeyUpdate messages, which triggers rekeying of DTLS. Such symmetric rekeying gives significantly less protection against key leakage than re-running Diffie-Hellman as explained above. After leakage of application_traffic_secret_N, an attacker can passively eavesdrop on all future data sent on the connection including data encrypted with application_traffic_secret_N+1, application_traffic_secret_N+2, etc. Note that KeyUpdate does not update the exporter_secret.¶

DTLS/SCTP is in many deployments replacing IPsec. For IPsec, NIST (US), BSI (Germany), and ANSSI (France) recommends very frequent re-run of Diffie-Hellman to provide forward secrecy and force attackers to dynamic key extraction [RFC7624]. ANSSI writes "It is recommended to force the periodic renewal of the keys, e.g., every hour and every 100 GB of data, in order to limit the impact of a key compromise." [ANSSI-DAT-NT-003].¶

For many DTLS/SCTP deployments the SCTP association is expected to have a very long lifetime of months or even years. For associations with such long lifetimes there is a need to frequently re-authenticate both client and server. TLS Certificate lifetimes significantly shorter than a year are common which is shorter than many expected DTLS/SCTP associations.¶

SCTP-AUTH re-rekeying, periodic authentication of both endpoints, and frequent re-run of Diffie-Hellman to force attackers to dynamic key extraction is in DTLS/SCTP per this specification achieved by setting up new DTLS connections over the same SCTP association. Implementations SHOULD set up new connections frequently to force attackers to dynamic key extraction. Implementations MUST set up new connections before any of the certificates expire. It is RECOMMENDED that all negotiated and exchanged parameters are the same except for the timestamps in the certificates. Clients and servers MUST NOT accept a change of identity during the setup of a new connections, but MAY accept negotiation of stronger algorithms and security parameters, which might be motivated by new attacks.¶

Allowing new connections can enable denial-of-service attacks. The endpoints SHOULD limit the frequency of new connections.¶

When DTLS/SCTP is used with DTLS 1.2 [RFC6347], the TLS Session Hash and Extended Master Secret Extension [RFC7627] MUST be used to prevent unknown key-share attacks where an attacker establishes the same key on several connections. DTLS 1.3 always prevents these kinds of attacks. The use of SCTP-AUTH then cryptographically binds new connections to the old connection. This together with mandatory mutual authentication (on the DTLS layer) and a requirement to not accept new identities mitigates MITM attacks that have plagued renegotiation [TRISHAKE].¶

9.2. Downgrade Attacks

A peer supporting DTLS/SCTP according to this specification, DTLS/SCTP according to [RFC6083] and/or SCTP without DTLS may be vulnerable to downgrade attacks where on on-path attacker interferes with the protocol setup to lower or disable security. If possible, it is RECOMMENDED that the peers have a policy only allowing DTLS/SCTP according to this specification.¶

9.3. Targeting DTLS Messages

The DTLS handshake messages and other control messages, i.e. not application data can easily be identified when using DTLS 1.2 as their content type is not encrypted. With DTLS 1.3 there is no unprotected content type. However, they will sent with an PPID of 0 if sent in their own SCTP user messages. Section 4.4 proposes a basic behavior that will stil make it easily for anyone to detect the DTLS messages that are not proteceted user messages.¶

9.4. Authentication and Policy Decisions

DTLS/SCTP MUST be mutually authenticated. Authentication is the process of establishing the identity of a user or system and verifying that the identity is valid. DTLS only provides proof of possession of a key. DTLS/SCTP MUST perform identity authentication. It is RECOMMENDED that DTLS/SCTP is used with certificate-based authentication. When certificates are used the applicatication using DTLS/SCTP is reposible for certificate policies, certificate chain validation, and identity authentication (HTTPS does for example match the hostname with a subjectAltName of type dNSName). The application using DTLS/SCTP MUST define what the identity is and how it is encoded and the client and server MUST use the same identity format. Guidance on server certificate validation can be found in [RFC6125]. DTLS/SCTP enables periodic transfer of mutual revocation information (OSCP stapling) every time a new parallel connection is set up. All security decisions MUST be based on the peer's authenticated identity, not on its transport layer identity.¶

It is possible to authenticate DTLS endpoints based on IP addresses in certificates. SCTP associations can use multiple IP addresses per SCTP endpoint. Therefore, it is possible that DTLS records will be sent from a different source IP address or to a different destination IP address than that originally authenticated. This is not a problem provided that no security decisions are made based on the source or destination IP addresses.¶

9.5. Resumption and Tickets

In DTLS 1.3 any number of tickets can be issued in a connection and the tickets can be used for resumption as long as they are valid, which is up to seven days. The nodes in a resumed connection have the same roles (client or server) as in the connection where the ticket was issued. In DTLS/SCTP, there are no significant performance benefits with resumption and an implementation can chose to never issue any tickets. If tickets and resumption are used it is enough to issue a single ticket per connection.¶

9.6. Privacy Considerations

[RFC6973] suggests that the privacy considerations of IETF protocols be documented.¶

For each SCTP user message, the user also provides a stream identifier, a flag to indicate whether the message is sent ordered or unordered, and a payload protocol identifier. Although DTLS/SCTP provides privacy for the actual user message, the other three information fields are not confidentiality protected. They are sent as cleartext because they are part of the SCTP DATA chunk header.¶

It is RECOMMENDED that DTLS/SCTP is used with certificate based authentication in DTLS 1.3 [I-D.ietf-tls-dtls13] to provide identity protection. DTLS/SCTP MUST be used with a key exchange method providing forward secrecy.¶

9.7. Pervasive Monitoring

As required by [RFC7258], work on IETF protocols needs to consider the effects of pervasive monitoring and mitigate them when possible.¶

Pervasive Monitoring is widespread surveillance of users. By encrypting more information including user identities, DTLS 1.3 offers much better protection against pervasive monitoring.¶

Massive pervasive monitoring attacks relying on key exchange without forward secrecy has been reported. By mandating forward secrecy, DTLS/SCTP effectively mitigate many forms of passive pervasive monitoring and limits the amount of compromised data due to key compromise.¶

An important mitigation of pervasive monitoring is to force attackers to do dynamic key exfiltration instead of static key exfiltration. Dynamic key exfiltration increases the risk of discovery for the attacker [RFC7624]. DTLS/SCTP per this specification encourages implementations to frequently set up new DTLS connections with (EC)DHE over the same SCTP association to force attackers to do dynamic key exfiltration.¶

In addition to the privacy attacks discussed above, surveillance on a large scale may enable tracking of a user over a wider geographical area and across different access networks. Using information from DTLS/SCTP together with information gathered from other protocols increase the risk of identifying individual users.¶