A Datagram Transport Layer Security (DTLS) 1.2 Profile for the Internet of Things

This document defines a DTLS 1.2 profile that offers communication security for Internet of Things (IoT) applications and is reasonably implementable on many constrained devices. It aims to meet the following goals: Serves as a one-stop shop for implementers to know which pieces of the specification jungle contain relevant details. Does not alter the DTLS 1.2 specification. Does not introduce any new extensions. Aligns with the DTLS security modes of the Constrained Application Protocol (CoAP). DTLS is used to secure a number of applications run over an unreliable datagram transport. CoAP is one such protocol and has been designed specifically for use in IoT environments. CoAP can be secured a number of different ways, also called security modes. These security modes are as follows, see , , for additional details: No DTLS is used but instead application layer security functionality is assumed. DTLS supports the use of shared secrets . This mode is useful if the number of communication relationships between the IoT device and servers is small and for very constrained devices. Shared secret-based authentication mechanisms offer good performance and require a minimum of data to be exchanged. TLS supports client and server authentication using asymmetric cryptography. Two approaches for validating these public keys are available. First, allows raw public keys to be used in TLS without the overhead of certificates. This approach requires out-of-band validation of the public key. Second, the use of X.509 certificates with TLS is common on the Web today (at least for server-side authentication) and certain IoT environments may also re-use those capabilities. Certificates bind an identifier to the public key signed by a certification authority (CA). A trust anchor store has to be provisioned on the device to indicate what CAs are trusted. Furthermore, the certificate may contain a wealth of other information used to make authorization decisions. As described in , an application designer developing an IoT device needs to consider the security threats and the security services that can be used to mitigate the threats. Enabling devices to upload data and retrieve configuration information, inevitably requires that Internet-connected devices be able to authenticate themselves to servers and vice versa as well as to ensure that the data and information exchanged is integrity and confidentiality protected. While these security services can be provided at different layers in the protocol stack the use of communication security, as offered by DTLS, has been very popular on the Internet and it is likely to be useful for IoT scenarios as well. In case the communication security features offered by DTLS meet the security requirements of your application the remainder of the document might offer useful guidance. Not every IoT deployment will use CoAP but the discussion regarding choice of credentials and cryptographic algorithms will be very similar. As such, the discussions in this document are applicable beyond the use of the CoAP protocol. The design of DTLS is intentionally very similar to TLS. Since DTLS operates on top of an unreliable datagram transport a few enhancements to the TLS structure are, however necessary. RFC 6347 explains these differences in great detail. As a short summary, for those not familiar with DTLS the differences are: An explicit sequence number and an epoch field is included in the TLS Record Layer. Section 4.1 of RFC 6347 explains the processing rules for these two new fields. The value used to compute the MAC is the 64-bit value formed by concatenating the epoch and the sequence number. Stream ciphers must not be used with DTLS. The only stream cipher defined for TLS 1.2 is RC4 and due to cryptographic weaknesses it is not recommended anymore even for use with TLS. The TLS Handshake Protocol has been enhanced to include a stateless cookie exchange for Denial of Service (DoS) resistance. Furthermore, the header has been extended to deal with message loss, reordering, and fragmentation. Retransmission timers have been included to deal with message loss. For DoS protection a new handshake message, the HelloVerifyRequest, was added to DTLS. This handshake message is sent by the server and includes a stateless cookie, which is returned in a ClientHello message back to the server. This type of DoS protection mechanism has also been incorporated into the design of IKEv2. Although the exchange is optional for the server to execute, a client implementation has to be prepared to respond to it.

The key words "MUST", "MUST NOT", "REQUIRED", "MUST", "MUST NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in . Note that "Client" and "Server" in this document refer to TLS roles, where the Client initiates the TLS handshake. This does not restrict the interaction pattern of the protocols carried inside TLS as the record layer allows bi-directional communication. In the case of CoAP the "Client" can act as a CoAP Server or Client.

This document describes a profile of DTLS 1.2 and, to be useful, it has to make assumptions about the envisioned communication architecture. The communication architecture shown in assumes a uni-cast communication interaction with an IoT device utilizing a DTLS client and that client interacts with one or multiple DTLS servers. Clients are preconfigured with the address or addresses of servers (e.g., as part of the firmware) they will communicate with as well as authentication information: For PSK-based authentication (see ), this includes the paired "PSK identity" and shared secret to be used with each server. For raw public key-based authentication (see ), this includes either the server's public key or the hash of the server's public key. For certificate-based authentication (see ), this may include a pre-populated trust anchor store that allows the client to perform path validation for the certificate obtained during the handshake with the server. This document only focuses on the description of the DTLS client-side functionality.

TLS (and consequently DTLS) has the concept of ciphersuites and an IANA registry was created to register the suites. A ciphersuite (and the specification that defines it) contains the following information: Authentication and Key Exchange Algorithm (e.g., PSK) Cipher and Key Length (e.g., AES with 128 bit keys) Mode of operation (e.g., CBC) Hash Algorithm for Integrity Protection (e.g., SHA in combination with HMAC) Hash Algorithm for use with the Pseudorandom Function (e.g. HMAC with the SHA-256) Misc information (e.g., length of authentication tags) Information whether the ciphersuite is suitable for DTLS or only for TLS The TLS ciphersuite TLS_PSK_WITH_AES_128_CCM_8, for example, uses a pre-shared authentication and key exchange algorithm. RFC 6655 defines this ciphersuite. It uses the Advanced Encryption Standard (AES) encryption algorithm, which is a block cipher. Since the AES algorithm supports different key lengths (such as 128, 192 and 256 bits) this information has to be specified as well and the selected ciphersuite supports 128 bit keys. A block cipher encrypts plaintext in fixed-size blocks and AES operates on fixed block size of 128 bits. For messages exceeding 128 bits, the message is partitioned into 128-bit blocks and the AES cipher is applied to these input blocks with appropriate chaining, which is called mode of operation. TLS 1.2 introduced Authenticated Encryption with Associated Data (AEAD) ciphersuites . AEAD is a class of block cipher modes which encrypt (parts of) the message and authenticate the message simultaneously. Examples of such modes include the Counter with CBC-MAC (CCM) mode, and the Galois/Counter Mode (GCM). Some AEAD ciphersuites have shorter authentication tags and are therefore more suitable for networks with low bandwidth where small message size matters. The TLS_PSK_WITH_AES_128_CCM_8 ciphersuite that ends in "_8" has an 8-octet authentication tag, while the regular CCM ciphersuites have 16-octet authentication tags. TLS 1.2 also replaced the combination of MD5/SHA-1 hash functions in the TLS pseudo random function (PRF) with cipher-suite-specified PRFs. For this reason authors of more recent TLS 1.2 ciphersuite specifications explicitly indicate the MAC algorithm and the hash functions used with the TLS PRF. This document references the CoAP recommended ciphersuite choices, which have been selected based on implementation and deployment experience from the IoT community. Over time the preference for certain algorithms will, however, change. Not all components of a ciphersuite change at the same speed. Changes are more likely to expect for ciphers, the mode of operation, and the hash algorithms. Some deployment environments will also be impacted by local regulation, which might dictate a certain and less likely for public key algorithms (such as RSA vs. ECC).

The use of pre-shared secret credentials is one of the most basic techniques for DTLS since it is both computational efficient and bandwidth conserving. Pre-shared secret based authentication was introduced to TLS with RFC 4279 . The exchange shown in illustrates the DTLS exchange including the cookie exchange. While the server is not required to initiate a cookie exchange with every handshake, the client is required to implement and to react on it when challenged.

<-------- HelloVerifyRequest (contains cookie) ClientHello --------> (with cookie) ServerHello *ServerKeyExchange <-------- ServerHelloDone ClientKeyExchange ChangeCipherSpec Finished --------> ChangeCipherSpec <-------- Finished Application Data <-------> Application Data Legend: * indicates an optional message payload ]]> does not mandate the use of any particular type of identity. Hence, the TLS client and server clearly have to agree on the identities and keys to be used. The mandated encoding of identities in Section 5.1 of RFC 4279 aims to improve interoperability for those cases where the identity is configured by a person using some management interface. Many IoT devices do, however, not have a user interface and most of their credentials are bound to the device rather than the user. Furthermore, credentials are provisioned into trusted hardware modules or in the firmware by the developers. As such, the encoding considerations are not applicable to this usage environment. For use with this profile the PSK identities SHOULD NOT assume a structured format (as domain names, Distinguished Names, or IP addresses have) and a bit-by-bit comparison operation can then be used by the server-side infrastructure. As described in clients may have pre-shared keys with several different servers. The client indicates which key it uses by including a "PSK identity" in the ClientKeyExchange message. To help the client in selecting which PSK identity / PSK pair to use, the server can provide a "PSK identity hint" in the ServerKeyExchange message. For IoT environments a simplifying assumption is made that the hint for PSK key selection is based on the domain name of the server. Hence, servers SHOULD NOT send the "PSK identity hint" in the ServerKeyExchange message and client MUST ignore the message. This approach is inline with RFC 4279 . RFC 4279 requires TLS implementations supporting PSK ciphersuites to support arbitrary PSK identities up to 128 octets in length, and arbitrary PSKs up to 64 octets in length. This is a useful assumption for TLS stacks used in the desktop and mobile environment where management interfaces are used to provision identities and keys. For the IoT environment, however, many devices are not equipped with displays and input devices (e.g., keyboards). Hence, keys are distributed as part of hardware modules or are embedded into the firmware. As such, these restrictions are not applicable to this profile. Constrained Application Protocol (CoAP) currently specifies TLS_PSK_WITH_AES_128_CCM_8 as the mandatory to implement ciphersuite for use with shared secrets. This ciphersuite uses the AES algorithm with 128 bit keys and CCM as the mode of operation. The label "_8" indicates that an 8-octet authentication tag is used. This ciphersuite makes use of the default TLS 1.2 Pseudorandom Function (PRF), which uses HMAC with the SHA-256 hash function.

The use of raw public keys with DTLS, as defined in , is the first entry point into public key cryptography without having to pay the price of certificates and a PKI. The specification re-uses the existing Certificate message to convey the raw public key encoded in the SubjectPublicKeyInfo structure. To indicate support two new TLS extensions had been defined, as shown in , namely the server_certificate_type and the client_certificate_type. To operate this mechanism securely it is necessary to authenticate and authorize the public keys out-of-band. This document therefore assumes that a client implementation comes with one or multiple raw public keys of servers, it has to communicate with, pre-provisioned. Additionally, a device will have its own raw public key. To replace, delete, or add raw public key to this list requires a software update, for example using a firmware update mechanism.

client_certificate_type server_certificate_type <------- HelloVerifyRequest ClientHello --------> client_certificate_type server_certificate_type ServerHello client_certificate_type server_certificate_type Certificate ServerKeyExchange CertificateRequest <-------- ServerHelloDone Certificate ClientKeyExchange CertificateVerify [ChangeCipherSpec] Finished --------> [ChangeCipherSpec] <-------- Finished ]]> The CoAP recommended ciphersuite for use with this credential type is TLS_ECDHE_ECDSA_WITH_AES_128_CCM_8 . This elliptic curve cryptography (ECC) based AES-CCM TLS ciphersuite uses the Elliptic Curve Diffie Hellman (ECDHE) as the key establishment mechanism and an Elliptic Curve Digital Signature Algorithm (ECDSA) for authentication. This ciphersuite make use of the AEAD capability in DTLS 1.2 and utilizes an eight-octet authentication tag. Based on the Diffie-Hellman it provides perfect forward secrecy (PFS). More details about the PFS can be found in . RFC 6090 provides valuable information for implementing Elliptic Curve Cryptography algorithms. Since many IoT devices will either have limited ways to log error or no ability at all, any error will lead to implementations attempting to re-try the exchange.

The use of mutual certificate-based authentication is shown in , which makes use of the cached info extension . Support of the cached info extension is required. Caching certificate chains allows the client to reduce the communication overhead significantly since otherwise the server would provide the end entity certificate, and the certificate chain. Because certificate validation requires that root keys be distributed independently, the self-signed certificate that specifies the root certificate authority is omitted from the chain. Client implementations MUST be provisioned with a trust anchor store that contains the root certificates. The use of the Trust Anchor Management Protocol (TAMP) is, however, not envisioned. Instead IoT devices using this profile MUST rely a software update mechanism to provision these trust anchors. When DTLS is used to secure CoAP messages then the server provided certificates MUST contain the fully qualified DNS domain name or "FQDN". The coaps URI scheme is described in Section 6.2 of . This FQDN is stored in the SubjectAltName or in the CN, as explained in Section 9.1.3.3 of , and used by the client to match it against the FQDN used during the look-up process, as described in RFC 6125 . For the profile in this specification does not assume dynamic discovery of local servers. For client certificates the identifier used in the SubjectAltName or in the CN MUST be an EUI-64 , as mandated in Section 9.1.3.3 of . For certificate revocation neither the Online Certificate Status Protocol (OCSP) nor Certificate Revocation Lists (CRLs) are used. Instead, this profile relies on a software update mechanism. While multiple OCSP stapling has recently been introduced as a mechanism to piggyback OCSP request/responses inside the DTLS/TLS handshake to avoid the cost of a separate protocol handshake further investigations are needed to determine its suitability for the IoT environment.

cached_information <------- HelloVerifyRequest ClientHello --------> cached_information ServerHello cached_information Certificate ServerKeyExchange CertificateRequest <-------- ServerHelloDone Certificate ClientKeyExchange CertificateVerify [ChangeCipherSpec] Finished --------> [ChangeCipherSpec] <-------- Finished ]]> Regarding the ciphersuite choice the discussion in applies. Further details about X.509 certificates can be found in Section 9.1.3.3 of . QUESTION: What restrictions regarding the depth of the certificate chain should be made? Is one level enough?

DTLS uses the Alert protocol to convey error messages and specifies a longer list of errors. However, not all error messages defined in the TLS specification are applicable to this profile. All error messages marked as RESERVED are only supported for backwards compatibility with SSL and are therefore not applicable to this profile. Those include decryption_failed_RESERVED, no_certificate_RESERVE, and export_restriction_RESERVED. A number of the error messages are applicable only for certificate-based authentication ciphersuites. Hence, for PSK and raw public key use the following error messages are not applicable: bad_certificate, unsupported_certificate, certificate_revoked, certificate_expired, certificate_unknown, unknown_ca, and access_denied. Since this profile does not make use of compression at the TLS layer the decompression_failure error message is not applicable either. RFC 4279 introduced a new alert message unknown_psk_identity for PSK ciphersuites. As stated in Section 2 of RFC 4279 the decryption_error error message may also be used instead. For this profile the TLS server MUST return the decryption_error error message instead of the unknown_psk_identity. Furthermore, the following errors should not occur based on the description in this specification: This document only focuses on one version of the DTLS protocol. This error message indicates that the server requires ciphers to be more secure. This document does, however, specify the only acceptable ciphersuites and client implementations must support them. The IoT devices in focus of this specification are assumed to be unattended.

Session resumption is a feature of DTLS that allows a client to continue with an earlier established session state. The resulting exchange is shown in . In addition, the server may choose not to do a cookie exchange when a session is resumed. Still, clients have to be prepared to do a cookie exchange with every handshake.

ServerHello [ChangeCipherSpec] <-------- Finished [ChangeCipherSpec] Finished --------> Application Data <-------> Application Data ]]> Clients MUST implement session resumption to improve the performance of the handshake (in terms of reduced number of message exchanges, lower computational overhead, and less bandwidth conserved). Since the communication model described in does not assume that the server is constrained. RFC 5077 describing TLS session resumption without server-side state is not utilized by this profile.

recommends to always disable DTLS-level compression due to attacks. For IoT applications compression at the DTLS is not needed since application layer protocols are highly optimized and the compression algorithms at the DTLS layer increase code size and complexity. This DTLS client profile does not include DTLS layer compression.

Perfect forward secrecy (PFS) is designed to prevent the compromise of a long-term secret key from affecting the confidentiality of past conversations. The PSK ciphersuite recommended in the CoAP specification does not offer this property since it does not utilize a Diffie-Hellman exchange. on the other hand recommends using ciphersuites offering this security property and so do the public key-based ciphersuites recommended by the CoAP specification. The use of PFS is certainly a tradeoff decision since on one hand the compromise of long-term secrets of embedded devices is more likely than with many other Internet hosts but on the other hand a Diffie-Hellman exchange requires emphemeral key pairs to be generated, which can be demanding from a performance point of view. Finally, the impact of the disclosure of past conversations and the desire to increase the cost for pervasive monitoring (see ) has to be taken into account. Our recommendation is to stick with the ciphersuite suggested in the CoAP specification. New ciphersuites support PFS for pre-shared secret-based authentication, such as , and might be available as a standardized ciphersuite in the future.

RFC 6520 defines a heartbeat mechanism to test whether the other peer is still alive. The same mechanism can also be used to perform Path Maximum Transmission Unit (MTU) Discovery. A recommendation about the use of RFC 6520 depends on the type of message exchange an IoT device performs. There are three types of exchanges that need to be analysed: This is a common communication pattern where IoT devices upload data to a server on the Internet on an irregular basis. The communcation may be triggered by specific events, such as opening a door. Since the upload happens on an irregular and unpredicable basis and due to renumbering and Network Address Translation (NAT) a new DTLS session or DTLS session resumption can be used. In this case there is no use for a keep-alive extension for this scenario. This is a variation of the previous case whereby data gets uploaded on a regular basis, for example, based on frequent temperature readings. With such regular exchange it can be assumed that the DTLS context is still in kept at the IoT device. If neither NAT bindings nor IP address changes occurred then the DTLS record layer will not notice any changes. For the case where IP and port changes happened it is necessary to re-create the DTLS record layer using session resumption. In this scenario there is no use for a keep-alive extension. It is also very likely that the device will enter a sleep cycle in between data transmissions to keep power consumption low. In the two previous scenarios the client initiated the protocol interaction. In this case, we consider server-initiated messages. Since messages to the client may get blocked by intermediaries, such as NATs and stateful packet filtering firewalls, the initial connection setup is triggered by the client and then kept alive. Since state expires fairly quickly at middleboxes regular heartbeats are necessary whereby these keep-alive messages may be exchanged at the application layer or within DTLS itself. For this message exchange pattern the use of DTLS heartbeat messages is quite useful. The MTU discovery mechanism, on the other hand, is less likely to be relevant since for many IoT deployments the must constrained link is the wireless interface at the IoT device itself rather than somewhere in the network. Only in more complex network topologies the situation might be different. For server-initiated messages the heartbeat extension can be recommended.

The DTLS protocol requires random numbers to be available during the protocol run. For example, during the ClientHello and the ServerHello exchange the client and the server exchange random numbers. Also, the use of the Diffie Hellman exchange requires random numbers during the key pair generation. Special care has to be paid when generating random numbers in embedded systems as many entropy sources available on desktop operating systems or mobile devices might be missing, as described in . Consequently, if not enough time is given during system start time to fill the entropy pool then the output might be predictable and repeatable, for example leading to the same keys generated again and again. Recommendation: IoT devices using DTLS MUST offer ways to generate quality random numbers. Guidelines and requirements for random number generation can be found in RFC 4086 . It is important to note that sources contributing to the randomness pool on laptops, or desktop pcs are not available on many IoT device, such as mouse movement, timing of keystrokes, air turbulence on the movement of hard drive heads, etc. Other sources have to be found or dedicated hardware has to be added.

This RFC 6066 extension allows to avoid sending client-side certificates and URLs instead. This reduces the over-the-air transmission. This is certainly a useful extension when a certificate-based mode for DTLS is used since the TLS cached info extension does not provide any help with caching information on the server side. Recommendation: Add support for client certificate URLs for those environments where client-side certificates are used.

This RFC 6066 extension allows clients to indicate what trust anchor they support. With certificate-based authentication a DTLS server conveys its end entity certificate to the client during the DTLS exchange provides. Since the server does not necessarily know what trust anchors the client has stored it includes intermediate CA certs in the certificate payload as well to facilitate with certification path construction and path validation. Today, in most IoT deployments there is a fairly static relationship between the IoT device (and the software running on them) and the server- side infrastructure and no such dynamic indication of trust anchors is needed. Recommendation: For IoT deployments where clients talk to a fixed, pre-configured set of servers and where a software update mechanism is available this extension is not recommended. Environments where the client needs to interact with dynamically discovered DTLS servers this extension may be useful to reduce the communication overhead. Note, however, in that case the TLS cached info extension may help to reduce the communication overhead for everything but the first protocol interaction.

This RFC 6066 extension was introduced to reduces the size of the MAC used at the Record Layer. This extension was developed for TLS ciphersuites that used older modes of operation where the MAC and the encryption operation was performed independently. For CoAP, however, the recommended ciphersuites use the newer Authenticated Encryption with Associated Data (AEAD) construct, namely the CBC-MAC mode (CCM) with eight-octet authentication tags. Recommendation: Since this profile only supports AEAD ciphersuites this extension is not applicable.

This RFC 6066 extension defines a mechanism for a client to tell a TLS server the name of the server it wants to contact. This is a useful extension for many hosting environments where multiple virtual servers are run on single IP address. Recommendation: Unless it is known that a DTLS client does not interact with a server in a hosting environment that requires such an extension we advice to offer support for the SNI extension in this profile.

This RFC 6066 extension lowers the maximum fragment length support needed for the Record Layer from 2^14 bytes to 2^9 bytes. This is a very useful extension that allows the client to indicate to the server how much maximum memory buffers it uses for incoming messages. Ultimately, the main benefit of this extension is it to allows client implementations to lower their RAM requirements since the client does not need to accept packets of large size (such as 16k packets as required by plain TLS/DTLS). Recommendation: Client implementations must support this extension.

CoAP demands version 1.2 of DTLS to be used and the earlier version of DTLS is not supported. As such, there is no risk of downgrading to an older version of DTLS. The work described in is therefore also not applicable to this environment since there is no legacy server infrastructure to worry about. QUESTION: Should we say something for non-CoAP use of DTLS? To prevent the TLS renegotiation attack clients MUST respond to server-initiated renegotiation attempts with an Alert message (no_renegotiation) and clients MUST NOT initiate them. TLS and DTLS allows a client and a server who already have a TLS connection to negotiate new parameters, generate new keys, etc by initiating a TLS handshake using a ClientHello message. Renegotiation happens in the existing TLS connection, with the new handshake packets being encrypted along with application data.

The DTLS handshake exchange conveys various identifiers, which can be observed by an on-path eavesdropper. For example, the DTLS PSK exchange reveals the PSK identity, the supported extensions, the session id, algorithm parameters, etc. When session resumption is used then individual TLS sessions can be correlated by an on-path adversary. With many IoT deployments it is likely that keying material and their identifiers are persistent over a longer period of time due to the cost of updating software on these devices. User participation with many IoT deployments poses a challenge since many of the IoT devices operate unattended, even though they will initially be enabled by a human. The ability to control data sharing and to configure preference will have to be provided at a system level rather than at the level of a DTLS profile, which is the scope of this document. Quite naturally, the use of DTLS with mutual authentication will allow a TLS server to collect authentication information about the IoT device (potentially over a long period of time). While this strong form of authentication will prevent mis-attribution it also allows strong identification. This device-related data collection (e.g., sensor recordings) will be associated with other data to be truly useful and this extra data might include personal data about the owner of the device or data about the environment it senses. Consequently, the data stored on the server-side will be vulnerable to stored data compromise. For the communication between the client and the server this specification prevents eavesdroppers to gain access to the communication content. While the PSK-based ciphersuite does not provide PFS the asymmetric version does. No explicit techniques, such as extra padding, have been provided to make traffic analysis more difficult.

This entire document is about security. We would also like to point out that designing a software update mechanism into an IoT system is crucial to ensure that both functionality can be enhanced and that potential vulnerabilities can be fixed. This software update mechanism is also useful for changing configuration information, for example, trust anchors and other keying related information.

This document includes no request to IANA.

Thanks to Rene Hummen, Sye Loong Keoh, Sandeep Kumar, Eric Rescorla, Russ Housley, Michael Richardson, Zach Shelby, and Sean Turner for their helpful comments and discussions that have shaped the document. Big thanks also to Klaus Hartke, who wrote the initial version of this document.