CoAP: Echo, Request-Tag, and Token Processing

1.1. 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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

Like [RFC7252], this document is relying on the Representational State Transfer [REST] architecture of the Web.¶

Unless otherwise specified, the terms "client" and "server" refer to "CoAP client" and "CoAP server", respectively, as defined in [RFC7252]. The term "origin server" is used as in [RFC7252]. The term "origin client" is used in this document to denote the client from which a request originates; to distinguish from clients in proxies.¶

A message's "freshness" is a measure of when a message was sent on a time scale of the recipient. A server that receives a request can either verify that the request is fresh or determine that it cannot be verified that the request is fresh. What is considered a fresh message is application dependent; exemplary uses are "no more than one hour ago" or "after this server's last reboot".¶

The terms "payload" and "body" of a message are used as in [RFC7959]. The complete interchange of a request and a response body is called a (REST) "operation". An operation fragmented using [RFC7959] is called a "block-wise operation". A block-wise operation which is fragmenting the request body is called a "block-wise request operation". A block-wise operation which is fragmenting the response body is called a "block-wise response operation".¶

Two request messages are said to be "matchable" if they occur between the same endpoint pair, have the same code, and have the same set of options, with the exception that elective NoCacheKey options and options involved in block-wise transfer (Block1, Block2 and Request-Tag) need not be the same. Two operations are said to be matchable if any of their messages are.¶

Two matchable block-wise operations are said to be "concurrent" if a block of the second request is exchanged even though the client still intends to exchange further blocks in the first operation. (Concurrent block-wise request operations from a single endpoint are impossible with the options of [RFC7959] (see the last paragraphs of Sections 2.4 and 2.5) because the second operation's block overwrites any state of the first exchange.).¶

The Echo and Request-Tag options are defined in this document.¶

2.1. Request Freshness

A CoAP server receiving a request is in general not able to verify when the request was sent by the CoAP client. This remains true even if the request was protected with a security protocol, such as DTLS. This makes CoAP requests vulnerable to certain delay attacks which are particularly perilous in the case of actuators ([I-D.mattsson-core-coap-attacks]). Some attacks can be mitigated by establishing fresh session keys, e.g. performing a DTLS handshake for each request, but in general this is not a solution suitable for constrained environments, for example, due to increased message overhead and latency. Additionally, if there are proxies, fresh DTLS session keys between server and proxy does not say anything about when the client made the request. In a general hop-by-hop setting, freshness may need to be verified in each hop.¶

A straightforward mitigation of potential delayed requests is that the CoAP server rejects a request the first time it appears and asks the CoAP client to prove that it intended to make the request at this point in time.¶

2.2. The Echo Option

This document defines the Echo option, a lightweight challenge-response mechanism for CoAP that enables a CoAP server to verify the freshness of a request. A fresh request is one whose age has not yet exceeded the freshness requirements set by the server. The freshness requirements are application specific and may vary based on resource, method, and parameters outside of CoAP such as policies. The Echo option value is a challenge from the server to the client included in a CoAP response and echoed back to the server in one or more CoAP requests.¶

This mechanism is not only important in the case of actuators, or other use cases where the CoAP operations require freshness of requests, but also in general for synchronizing state between CoAP client and server, cryptographically verifying the aliveness of the client, or forcing a client to demonstrate reachability at its claimed network address. The same functionality can be provided by echoing freshness indicators in CoAP payloads, but this only works for methods and response codes defined to have a payload. The Echo option provides a convention to transfer freshness indicators that works for all methods and response codes.¶

+--------+---+---+---+---+-------------+--------+------+---------+
| No.    | C | U | N | R | Name        | Format | Len. | Default |
+--------+---+---+---+---+-------------+--------+------+---------+
| TBD252 |   |   | x |   | Echo        | opaque | 1-40 | (none)  |
+--------+---+---+---+---+-------------+--------+------+---------+

      C = Critical, U = Unsafe, N = NoCacheKey, R = Repeatable

2.3. Echo Processing

The Echo option MAY be included in any request or response (see Section 2.4 for different applications).¶

The application decides under what conditions a CoAP request to a resource is required to be fresh. These conditions can for example include what resource is requested, the request method and other data in the request, and conditions in the environment such as the state of the server or the time of the day.¶

If a certain request is required to be fresh, the request does not contain a fresh Echo option value, and the server cannot verify the freshness of the request in some other way, the server MUST NOT process the request further and SHOULD send a 4.01 Unauthorized response with an Echo option. The server MAY include the same Echo option value in several different response messages and to different clients. Examples of this could be time-based freshness when several responses are sent closely after each other or event-based freshness with no event taking place between the responses.¶

The server may use request freshness provided by the Echo option to verify the aliveness of a client or to synchronize state. The server may also include the Echo option in a response to force a client to demonstrate reachability at its claimed network address. Note that the Echo option does not bind a request to any particular previous response, but provides an indication that the client had access to the previous response at the time when it created the request.¶

Upon receiving a 4.01 Unauthorized response with the Echo option, the client SHOULD resend the original request with the addition of an Echo option with the received Echo option value. The client MAY send a different request compared to the original request. Upon receiving any other response with the Echo option, the client SHOULD echo the Echo option value in the next request to the server. The client MAY include the same Echo option value in several different requests to the server, or discard it at any time (especially to avoid tracking, see Section 6).¶

A client MUST only send Echo values to endpoints it received them from (where as defined in [RFC7252] Section 1.2, the security association is part of the endpoint). In OSCORE processing, that means sending Echo values from Outer options (or from non-OSCORE responses) back in Outer options, and those from Inner options in Inner options in the same security context.¶

Upon receiving a request with the Echo option, the server determines if the request is required to be fresh. If not, the Echo option MAY be ignored. If the request is required to be fresh and the server cannot verify the freshness of the request in some other way, the server MUST use the Echo option to verify that the request is fresh. If the server cannot verify that the request is fresh, the request is not processed further, and an error message MAY be sent. The error message SHOULD include a new Echo option.¶

One way for the server to verify freshness is to bind the Echo value to a specific point in time and verify that the request is not older than a certain threshold T. The server can verify this by checking that (t1 - t0) < T, where t1 is the request receive time and t0 is the time when the Echo option value was generated. An example message flow over DTLS is shown Figure 2.¶

Client   Server
   |       |
   +------>|        Code: 0.03 (PUT)
   |  PUT  |       Token: 0x41
   |       |    Uri-Path: lock
   |       |     Payload: 0 (Unlock)
   |       |
   |<------+        Code: 4.01 (Unauthorized)
   |  4.01 |       Token: 0x41
   |       |        Echo: 0x00000009437468756c687521 (t0 = 9, +MAC)
   |       |
   | ...   | The round trips take 1 second, time is now t1 = 10.
   |       |
   +------>|        Code: 0.03 (PUT)
   |  PUT  |       Token: 0x42
   |       |    Uri-Path: lock
   |       |        Echo: 0x00000009437468756c687521 (t0 = 9, +MAC)
   |       |     Payload: 0 (Unlock)
   |       |
   |       | Verify MAC, compare t1 - t0 = 1 < T => permitted.
   |       |
   |<------+        Code: 2.04 (Changed)
   |  2.04 |       Token: 0x42
   |       |

Another way for the server to verify freshness is to maintain a cache of values associated to events. The size of the cache is defined by the application. In the following we assume the cache size is 1, in which case freshness is defined as no new event has taken place. At each event a new value is written into the cache. The cache values MUST be different except with negligible probability. The server verifies freshness by checking that e0 equals e1, where e0 is the cached value when the Echo option value was generated, and e1 is the cached value at the reception of the request. An example message flow over DTLS is shown in Figure 3.¶

Client   Server
   |       |
   +------>|        Code: 0.03 (PUT)
   |  PUT  |       Token: 0x41
   |       |    Uri-Path: lock
   |       |     Payload: 0 (Unlock)
   |       |
   |<------+        Code: 4.01 (Unauthorized)
   |  4.01 |       Token: 0x41
   |       |        Echo: 0x05 (e0 = 5, number of total lock
   |       |                            operations performed)
   |       |
   | ...   | No alterations happen to the lock state, e1 has the
   |       | same value e1 = 5.
   |       |
   +------>|        Code: 0.03 (PUT)
   |  PUT  |       Token: 0x42
   |       |    Uri-Path: lock
   |       |        Echo: 0x05
   |       |     Payload: 0 (Unlock)
   |       |
   |       | Compare e1 = e0 => permitted.
   |       |
   |<------+        Code: 2.04 (Changed)
   |  2.04 |       Token: 0x42
   |       |        Echo: 0x06 (e2 = 6, to allow later locking
   |       |                            without more round-trips)
   |       |

When used to serve freshness requirements (including client aliveness and state synchronizing), the Echo option value MUST be integrity protected between the intended endpoints, e.g. using DTLS, TLS, or an OSCORE Inner option ([RFC8613]). When used to demonstrate reachability at a claimed network address, the Echo option SHOULD be a MAC of the claimed address, but MAY be unprotected. Combining different Echo applications can necessitate different choices, see Appendix A item 2 for an example.¶

An Echo option MAY be sent with a successful response, i.e., even though the request satisfied any freshness requirements on the operation. This is called a "preemptive" Echo value, and useful when the server anticipates that the client will need to demonstrate freshness relative to the current response the near future.¶

A CoAP-to-CoAP proxy MAY set an Echo option on responses, both on forwarded ones that had no Echo option or ones generated by the proxy (from cache or as an error). If it does so, it MUST remove the Echo option it recognizes as one generated by itself on follow-up requests. When it receives an Echo option in a response, it MAY forward it to the client (and, not recognizing it as an own in future requests, relay it in the other direction as well) or process it on its own. If it does so, it MUST ensure that the client's request was generated (or is re-generated) after the Echo value used to send to the server was first seen. (In most cases, this means that the proxy needs to ask the client to repeat the request with a new Echo value.)¶

The CoAP server side of CoAP-to-HTTP proxies MAY request freshness, especially if they have reason to assume that access may require it (e.g. because it is a PUT or POST); how this is determined is out of scope for this document. The CoAP client side of HTTP-to-CoAP proxies MUST respond to Echo challenges itself if the proxy knows from the recent establishing of the connection that the HTTP request is fresh. Otherwise, it MUST NOT repeat an unsafe request and SHOULD respond with 503 Service Unavailable, Retry-After: 0 and terminate any underlying Keep-Alive connection. If the HTTP request arrived in Early Data, the proxy SHOULD use a 425 Too Early response instead (see [RFC8470]). They MAY also use other mechanisms to establish freshness of the HTTP request that are not specified here.¶

2.4. Applications of the Echo Option

Unless otherwise noted, all these applications require a security protocol to be used, and the Echo option to be protected by it.¶

1. Actuation requests often require freshness guarantees to avoid accidental or malicious delayed actuator actions. In general, all non-safe methods (e.g. POST, PUT, DELETE) may require freshness guarantees for secure operation.¶

   • The same Echo value may be used for multiple actuation requests to the same server, as long as the total time since the Echo option value was generated is below the freshness threshold.¶
   • For actuator applications with low delay tolerance, to avoid additional round-trips for multiple requests in rapid sequence, the server may send preemptive Echo values in successful requests, irrespectively of whether the request contained an Echo option or not. The client then uses the Echo option with the new value in the next actuation request, and the server compares the receive time accordingly.¶

2. A server may use the Echo option to synchronize properties (such as state or time) with a requesting client. A server MUST NOT synchronize a property with a client which is not the authority of the property being synchronized. E.g. if access to a server resource is dependent on time, then server MUST NOT synchronize time with a client requesting access unless the client is time authority for the server.¶

   Note that the state to be synchronized is not carried inside the Echo option. Any explicit state information needs to be carried along in the messages the Echo value is sent in; the Echo mechanism only provides a partial order on the messages' processing.¶

   • If a server reboots during operation it may need to synchronize state or time before continuing the interaction. For example, with OSCORE it is possible to reuse a partly persistently stored security context by synchronizing the Partial IV (sequence number) using the Echo option as specified in Section 7.5 of [RFC8613].¶
   • A device joining a CoAP group communication [I-D.ietf-core-groupcomm-bis] protected with OSCORE [I-D.ietf-core-oscore-groupcomm] may be required to initially synchronize its replay window state with a client by using the Echo option in a unicast response to a multicast request. The client receiving the response with the Echo option includes the Echo value in a subsequent unicast request to the responding server.¶

3. An attacker can perform a denial-of-service attack by putting a victim's address in the source address of a CoAP request and sending the request to a resource with a large amplification factor. The amplification factor is the ratio between the size of the request and the total size of the response(s) to that request. A server that provides a large amplification factor to an unauthenticated peer SHOULD mitigate amplification attacks as described in Section 11.3 of [RFC7252]. One way to mitigate such attacks is that the server responds to the alleged source address of the request with an Echo option in short response message (e.g. 4.01 Unauthorized), thereby requesting the client to verify its source address. This needs to be done only once per endpoint and limits the range of potential victims from the general Internet to endpoints that have been previously in contact with the server. For this application, the Echo option can be used in messages that are not integrity protected, for example during discovery. (This is formally recommended in Section 2.6).¶

   • In the presence of a proxy, a server will not be able to distinguish different origin client endpoints. Following from the recommendation above, a proxy that provides a large amplification factor to unauthenticated peers SHOULD mitigate amplification attacks. The proxy SHOULD use Echo to verify origin reachability as described in Section 2.3. The proxy MAY forward safe requests immediately to have a cached result available when the client's repeated request arrives.¶

   • Amplification mitigation is a trade-off between giving leverage to an attacker and causing overhead. An amplification factor of 3 (i.e., don't send more than three times the number of bytes received until the peer's address is confirmed) is considered acceptable for unconstrained applications in [RFC9000] Section 8.¶

     When that limit is applied and no further context is available, a safe default is sending initial responses no larger than 136 Bytes in CoAP serialization. (The number is assuming a 14 + 40 + 8 Bytes Ethernet, IP and UDP header with 4 Bytes added for the CoAP header. Triple that minus the non-CoAP headers gives the 136 Bytes). Given the token also takes up space in the request, responding with 132 Bytes after the token is safe as well.¶

   • When an Echo response is sent to mitigate amplification, it MUST be sent as a piggybacked or Non-confirmable response, never as a separate one (which would cause amplification due to retransmission).¶
4. A server may want to use the request freshness provided by the Echo to verify the aliveness of a client. Note that in a deployment with hop-by-hop security and proxies, the server can only verify aliveness of the closest proxy.¶

2.5. Characterization of Echo Applications

Use cases for the Echo option can be characterized by several criteria that help determine the required properties of the Echo value. These criteria apply both to those listed in Section 2.4 and any novel applications. They provide rationale for the statements in the former, and guidance for the latter.¶

2.6. Updated Amplification Mitigation Requirements for Servers

This section updates the amplification mitigation requirements for servers in [RFC7252] to recommend use of the Echo option to mitigate amplification attacks. The requirements for clients are not updated. Section 11.3 of [RFC7252] is updated by adding the following text:¶

A CoAP server SHOULD mitigate potential amplification attacks by responding to unauthenticated clients with 4.01 Unauthorized including an Echo option, as described in Section 2.4 item 3 of [[this document]].¶

3.1. Fragmented Message Body Integrity

CoAP was designed to work over unreliable transports, such as UDP, and includes a lightweight reliability feature to handle messages which are lost or arrive out of order. In order for a security protocol to support CoAP operations over unreliable transports, it must allow out-of-order delivery of messages.¶

The block-wise transfer mechanism [RFC7959] extends CoAP by defining the transfer of a large resource representation (CoAP message body) as a sequence of blocks (CoAP message payloads). The mechanism uses a pair of CoAP options, Block1 and Block2, pertaining to the request and response payload, respectively. The block-wise functionality does not support the detection of interchanged blocks between different message bodies to the same resource having the same block number. This remains true even when CoAP is used together with a security protocol such as DTLS or OSCORE, within the replay window ([I-D.mattsson-core-coap-attacks]), which is a vulnerability of CoAP when using RFC7959.¶

A straightforward mitigation of mixing up blocks from different messages is to use unique identifiers for different message bodies, which would provide equivalent protection to the case where the complete body fits into a single payload. The ETag option [RFC7252], set by the CoAP server, identifies a response body fragmented using the Block2 option.¶

3.2. The Request-Tag Option

This document defines the Request-Tag option for identifying request bodies, similar to ETag, but ephemeral and set by the CoAP client. The Request-Tag is intended for use as a short-lived identifier for keeping apart distinct block-wise request operations on one resource from one client, addressing the issue described in Section 3.1. It enables the receiving server to reliably assemble request payloads (blocks) to their message bodies, and, if it chooses to support it, to reliably process simultaneous block-wise request operations on a single resource. The requests must be integrity protected if they should protect against interchange of blocks between different message bodies. The Request-Tag option is only used in requests that carry the Block1 option, and in Block2 requests following these.¶

In essence, it is an implementation of the "proxy-safe elective option" used just to "vary the cache key" as suggested in [RFC7959] Section 2.4.¶

+--------+---+---+---+---+-------------+--------+------+---------+
| No.    | C | U | N | R | Name        | Format | Len. | Default |
+--------+---+---+---+---+-------------+--------+------+---------+
| TBD292 |   |   |   | x | Request-Tag | opaque |  0-8 | (none)  |
+--------+---+---+---+---+-------------+--------+------+---------+

      C = Critical, U = Unsafe, N = NoCacheKey, R = Repeatable

3.3. Request-Tag Processing by Servers

The Request-Tag option does not require any particular processing on the server side outside of the processing already necessary for any unknown elective proxy-safe cache-key option: The option varies the properties that distinguish block-wise operations (which includes all options except elective NoCacheKey and except Block1/2), and thus the server cannot treat messages with a different list of Request-Tag options as belonging to the same operation.¶

To keep utilizing the cache, a server (including proxies) MAY discard the Request-Tag option from an assembled block-wise request when consulting its cache, as the option relates to the operation-on-the-wire and not its semantics. For example, a FETCH request with the same body as an older one can be served from the cache if the older's Max-Age has not expired yet, even if the second operation uses a Request-Tag and the first did not. (This is similar to the situation about ETag in that it is formally part of the cache key, but implementations that are aware of its meaning can cache more efficiently, see [RFC7252] Section 5.4.2).¶

A server receiving a Request-Tag MUST treat it as opaque and make no assumptions about its content or structure.¶

Two messages carrying the same Request-Tag is a necessary but not sufficient condition for being part of the same operation. For one, a server may still treat them as independent messages when it sends 2.01/2.04 responses for every block. Also, a client that lost interest in an old operation but wants to start over can overwrite the server's old state with a new initial (num=0) Block1 request and the same Request-Tag under some circumstances. Likewise, that results in the new message not being part of the old operation.¶

As it has always been, a server that can only serve a limited number of block-wise operations at the same time can delay the start of the operation by replying with 5.03 (Service unavailable) and a Max-Age indicating how long it expects the existing operation to go on, or it can forget about the state established with the older operation and respond with 4.08 (Request Entity Incomplete) to later blocks on the first operation.¶

3.4. Setting the Request-Tag

For each separate block-wise request operation, the client can choose a Request-Tag value, or choose not to set a Request-Tag. It needs to be set to the same value (or unset) in all messages belonging to the same operation, as otherwise they are treated as separate operations by the server.¶

Starting a request operation matchable to a previous operation and even using the same Request-Tag value is called request tag recycling. The absence of a Request-Tag option is viewed as a value distinct from all values with a single Request-Tag option set; starting a request operation matchable to a previous operation where neither has a Request-Tag option therefore constitutes request tag recycling just as well (also called "recycling the absent option").¶

Clients that use Request-Tag for a particular purpose (like in Section 3.5) MUST NOT recycle a request tag unless the first operation has concluded. What constitutes a concluded operation depends on the purpose, and is defined accordingly; see examples in Section 3.5.¶

When Block1 and Block2 are combined in an operation, the Request-Tag of the Block1 phase is set in the Block2 phase as well for otherwise the request would have a different set of options and would not be recognized any more.¶

Clients are encouraged to generate compact messages. This means sending messages without Request-Tag options whenever possible, and using short values when the absent option cannot be recycled.¶

Note that Request-Tag options can be present in request messages that carry no Block option (for example, because a Request-Tag unaware proxy reassembled them), and MUST be ignored in those.¶

The Request-Tag option MUST NOT be present in response messages.¶

3.5. Applications of the Request-Tag Option

3.6. Rationale for the Option Properties

The Request-Tag option can be elective, because to servers unaware of the Request-Tag option, operations with differing request tags will not be matchable.¶

The Request-Tag option can be safe to forward but part of the cache key, because proxies unaware of the Request-Tag option will consider operations with differing request tags unmatchable but can still forward them.¶

The Request-Tag option is repeatable because this easily allows several cascaded stateless proxies to each put in an origin address. They can perform the steps of Section 3.5.3 without the need to create an option value that is the concatenation of the received option and their own value, and can simply add a new Request-Tag option unconditionally.¶

In draft versions of this document, the Request-Tag option used to be critical and unsafe to forward. That design was based on an erroneous understanding of which blocks could be composed according to [RFC7959].¶

3.7. Rationale for Introducing the Option

An alternative that was considered to the Request-Tag option for coping with the problem of fragmented message body integrity (Section 3.5.1) was to update [RFC7959] to say that blocks could only be assembled if their fragments' order corresponded to the sequence numbers.¶

That approach would have been difficult to roll out reliably on DTLS where many implementations do not expose sequence numbers, and would still not prevent attacks like in [I-D.mattsson-core-coap-attacks] Section 2.5.2.¶

3.8. Block2 / ETag Processing

The same security properties as in Section 3.5.1 can be obtained for block-wise response operations. The threat model here does not depend on an attacker: a client can construct a wrong representation by assembling it from blocks from different resource states. That can happen when a resource is modified during a transfer, or when some blocks are still valid in the client's cache.¶

Rules stating that response body reassembly is conditional on matching ETag values are already in place from Section 2.4 of [RFC7959].¶

To gain equivalent protection to Section 3.5.1, a server MUST use the Block2 option in conjunction with the ETag option ([RFC7252], Section 5.10.6), and MUST NOT use the same ETag value for different representations of a resource.¶

4.1. Request-Response Binding

A fundamental requirement of secure REST operations is that the client can bind a response to a particular request. If this is not ensured, a client may erroneously associate the wrong response to a request. The wrong response may be an old response for the same resource or a response for a completely different resource (see e.g. Section 2.3 of [I-D.mattsson-core-coap-attacks]). For example, a request for the alarm status "GET /status" may be associated to a prior response "on", instead of the correct response "off".¶

In HTTP/1.1, this type of binding is always assured by the ordered and reliable delivery as well as mandating that the server sends responses in the same order that the requests were received. The same is not true for CoAP where the server (or an attacker) can return responses in any order and where there can be any number of responses to a request (see e.g. [RFC7641]). In CoAP, concurrent requests are differentiated by their Token. Note that the CoAP Message ID cannot be used for this purpose since those are typically different for REST request and corresponding response in case of "separate response", see Section 2.2 of [RFC7252].¶

CoAP [RFC7252] does not treat Token as a cryptographically important value and does not give stricter guidelines than that the Tokens currently "in use" SHOULD (not SHALL) be unique. If used with a security protocol not providing bindings between requests and responses (e.g. DTLS and TLS) Token reuse may result in situations where a client matches a response to the wrong request. Note that mismatches can also happen for other reasons than a malicious attacker, e.g. delayed delivery or a server sending notifications to an uninterested client.¶

A straightforward mitigation is to mandate clients to not reuse Tokens until the traffic keys have been replaced. The following section formalizes that.¶

4.2. Updated Token Processing Requirements for Clients

As described in Section 4.1, the client must be able to verify that a response corresponds to a particular request. This section updates the Token processing requirements for clients in [RFC7252] to always assure a cryptographically secure binding of responses to requests for secure REST operations like "coaps". The Token processing for servers is not updated. Token processing in Section 5.3.1 of [RFC7252] is updated by adding the following text:¶

When CoAP is used with a security protocol not providing bindings between requests and responses, the Tokens have cryptographic importance. The client MUST make sure that Tokens are not used in a way so that responses risk being associated with the wrong request.¶

One easy way to accomplish this is to implement the Token (or part of the Token) as a sequence number starting at zero for each new or rekeyed secure connection. This approach SHOULD be followed.¶

5.1. Token reuse

Reusing Tokens in a way so that responses are guaranteed to not be associated with the wrong request is not trivial: The server may process requests in any order, and send multiple responses to the same request. An attacker may block, delay, and reorder messages. The use of a sequence number is therefore recommended when CoAP is used with a security protocol that does not provide bindings between requests and responses such as DTLS or TLS.¶

For a generic response to a Confirmable request over DTLS, binding can only be claimed without out-of-band knowledge if¶

• the original request was never retransmitted,¶
• the response was piggybacked in an Acknowledgement message (as a Confirmable or Non-confirmable response may have been transmitted multiple times), and¶
• if observation was used, the same holds for the registration, all re-registrations, and the cancellation.¶

(In addition, for observations, any responses using that Token and a DTLS sequence number earlier than the cancellation Acknowledgement message need to be discarded. This is typically not supported in DTLS implementations.)¶

In some setups, Tokens can be reused without the above constraints, as a different component in the setup provides the associations:¶

• In CoAP over TLS, retransmissions are not handled by the CoAP layer and behaves like a replay window size of 1. When a client is sending TLS-protected requests without Observe to a single server, the client can reuse a Token as soon as the previous response with that Token has been received.¶
• Requests whose responses are cryptographically bound to the requests (like in OSCORE) can reuse Tokens indefinitely.¶

In all other cases, a sequence number approach is RECOMMENDED as per Section 4.¶

Tokens that cannot be reused need to be handled appropriately. This could be solved by increasing the Token as soon as the currently used Token cannot be reused, or by keeping a list of all Tokens unsuitable for reuse.¶

When the Token (or part of the Token) contains a sequence number, the encoding of the sequence number has to be chosen in a way to avoid any collisions. This is especially true when the Token contains more information than just the sequence number, e.g. serialized state as in [RFC8974].¶