Internet-Draft | TLS/DTLS 1.3 IoT Profiles | March 2024 |
Tschofenig, et al. | Expires 4 September 2024 | [Page] |
This document is a companion to RFC 7925 and defines TLS/DTLS 1.3 profiles for Internet of Things devices. It also updates RFC 7925 with regards to the X.509 certificate profile.¶
This note is to be removed before publishing as an RFC.¶
Source for this draft and an issue tracker can be found at https://github.com/thomas-fossati/draft-tls13-iot.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 4 September 2024.¶
Copyright (c) 2024 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
This document defines a profile of DTLS 1.3 [DTLS13] and TLS 1.3 [RFC8446] that offers communication security services for IoT applications and is reasonably implementable on many constrained devices. Profile thereby means that available configuration options and protocol extensions are utilized to best support the IoT environment.¶
For IoT profiles using TLS/DTLS 1.2 please consult [RFC7925]. This document re-uses the communication pattern defined in [RFC7925] and makes IoT-domain specific recommendations for version 1.3 (where necessary).¶
TLS 1.3 has been re-designed and several previously defined extensions are not applicable to the new version of TLS/DTLS anymore. The following features changed with the transition from TLS 1.2 to 1.3:¶
Finally, ciphersuites were depreciated and the RSA-based key transport is not yet supported in TLS 1.3.¶
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.¶
This document reuses the terms "SHOULD+", "SHOULD-" and "MUST-" from [RFC8221].¶
In accordance with the recommendations in [RFC7925], a compliant implementation MUST implement TLS_AES_128_CCM_8_SHA256. It SHOULD implement TLS_CHACHA20_POLY1305_SHA256.¶
Pre-shared key based authentication is integrated into the main TLS/DTLS 1.3 specification and has been harmonized with session resumption.¶
A compliant implementation supporting authentication based on certificates and raw public keys MUST support digital signatures with ecdsa_secp256r1_sha256. A compliant implementation MUST support the key exchange with secp256r1 (NIST P-256) and SHOULD support key exchange with X25519.¶
A plain PSK-based TLS/DTLS client or server MUST implement the following extensions:¶
For use of external pre-shared keys [RFC9258] makes the following recommendation:¶
Applications SHOULD provision separate PSKs for (D)TLS 1.3 and prior versions.¶
Where possible, the importer interface defined in [RFC9258] MUST be used for external PSKs. This ensures that external PSKs used in (D)TLS 1.3 are bound to a specific key derivation function (KDF) and hash function.¶
The SNI extension is discussed in this document and the justification for implementing and using the ALPN extension can be found in [RFC9325].¶
For TLS/DTLS clients and servers implementing raw public keys and/or certificates the guidance for mandatory-to-implement extensions described in Section 9.2 of [RFC8446] MUST be followed.¶
TLS 1.3 simplified the Alert protocol but the underlying challenge in an embedded context remains unchanged, namely what should an IoT device do when it encounters an error situation. The classical approach used in a desktop environment where the user is prompted is often not applicable with unattended devices. Hence, it is more important for a developer to find out from which error cases a device can recover from.¶
TLS 1.3 has built-in support for session resumption by utilizing PSK-based credentials established in an earlier exchange.¶
TLS 1.3 does not have support for compression of application data traffic, as offered by previous versions of TLS. Applications are therefore responsible for transmitting payloads that are either compressed or use a more efficient encoding otherwise.¶
With regards to the handshake itself, various strategies have been applied to reduce the size of the exchanged payloads. TLS and DTLS 1.3 use less overhead, depending on the type of key confirmations, when compared to previous versions of the protocol. Additionally, the work on Compact TLS (cTLS) [I-D.ietf-tls-ctls] has taken compression of the handshake a step further by utilizing out-of-band knowledge between the communication parties to reduce the amount of data to be transmitted at each individual handshake, among applying other techniques.¶
TLS 1.3 allows the use of PFS with all ciphersuites since the support for it is negotiated independently.¶
Compared to DTLS 1.2 timeout-based whole flight retransmission, DTLS 1.3 ACKs sensibly decrease the risk of congestion collapse which was the basis for the very conservative recommendations given in Section 11 of [RFC7925].¶
In general, the recommendations in Section 7.3 of [DTLS13] regarding ACKs apply. In particular, "[w]hen DTLS 1.3 is used in deployments with lossy networks, such as low-power, long-range radio networks as well as low-power mesh networks, the use of ACKs is recommended" to signal any sign of disruption or lack of progress. This allows for selective or early retransmission, which leads to more efficient use of bandwidth and memory resources.¶
Due to the vast range of network technologies used in IoT deployments, from wired LAN to GSM-SMS, it's not possible to provide a universal recommendation for an initial timeout. Therefore, it is RECOMMENDED that DTLS 1.3 implementations allow developers to explicitly set the initial timer value. Developers SHOULD set the initial timeout to be twice the expected round-trip time (RTT), but no less than 1000ms. For specific application/network combinations, a sub-second initial timeout MAY be set. In cases where no RTT estimates are available, a 1000ms initial timeout is suitable for the general Internet.¶
For RRC, the recommendations in Section 7.5 of [I-D.ietf-tls-dtls-rrc] apply. Just like the handshake initial timers, it is RECOMMENDED that DTLS 1.2 and 1.3 implementations offer an option for their developers to explicitly set the RRC timer.¶
The discussion in Section 12 of [RFC7925] is applicable with one exception: the ClientHello and the ServerHello messages in TLS 1.3 do not contain gmt_unix_time component anymore.¶
This specification mandates the implementation of the Server Name Indication (SNI) extension. Where privacy requirements require it, the ECH (Encrypted Client Hello) extension [I-D.ietf-tls-esni] prevents an on-path attacker to determine the domain name the client is trying to connect to.¶
Since the Encrypted Client Hello extension requires use of Hybrid Public Key Encryption (HPKE) [I-D.irtf-cfrg-hpke] and additional protocols require further protocol exchanges and cryptographic operations, there is a certain overhead associated with this privacy feature.¶
Note that in industrial IoT deployments the use of ECH may not be an option because network administrators inspect DNS traffic generated by IoT devices in order to detect malicious behaviour.¶
Besides, to avoid leaking DNS lookups from network inspection altogether further protocols are needed, including DoH [RFC8484] and DPRIVE [RFC7858] [RFC8094]. For use of such techniques in managed networks, the reader is advised to keep up to date with the protocols defined by the Adaptive DNS Discovery (add) working group [ADD].¶
The Maximum Fragment Length Negotiation (MFL) extension has been superseded by the Record Size Limit (RSL) extension [RFC8449]. Implementations in compliance with this specification MUST implement the RSL extension and SHOULD use it to indicate their RAM limitations.¶
The recommendations in Section 19 of [RFC7925] are applicable.¶
The recommendations in Section 20 of [RFC7925] are applicable.¶
Appendix E.5 of [TLS13] establishes that:¶
Application protocols MUST NOT use 0-RTT data without a profile that defines its use. That profile needs to identify which messages or interactions are safe to use with 0-RTT and how to handle the situation when the server rejects 0-RTT and falls back to 1-RTT.¶
At the time of writing, no such profile has been defined for CoAP [CoAP]. Therefore, 0-RTT MUST NOT be used by CoAP applications.¶
This section contains updates and clarifications to the certificate profile defined in [RFC7925]. The content of Table 1 of [RFC7925] has been split by certificate "type" in order to clarify exactly what requirements and recommendations apply to which entity in the PKI hierarchy.¶
The content is also better aligned with the IEEE 802.1AR [_8021AR] specification, which introduces the terms Initial Device Identifier (IDevID) and Locally Significant Device Identifiers (LDevIDs). IDevIDs and LDevIDs are Device Identifier (DevID) and a DevID consists of¶
The IDevID is typically provisioned by a manufacturer and signed by the manufacturer CA. It is then used to obtain operational certificates, the LDevIDs, from the operator or owner of the device. Some protocols also introduce an additional hierarchy with application instance certificates, which are obtained for use with specific applications.¶
IDevIDs are primarily used with device onboarding or bootstrapping protocols, such as the Bootstrapping Remote Secure Key Infrastructure (BRSKI) protocol [RFC8995] or by LwM2M Bootstrap [LwM2M]. Hence, the use of IDevIDs is limited in purpose even though they have a long lifetime, or do not expire at all. While some bootstrapping protocols use TLS (and therefore make use of the IDevID as part of client authentication) there are other bootstrapping protocols that do not use TLS/DTLS for client authentication, such as FIDO Device Onboarding (FDO) [FDO]. In many cases, a profile for the certificate content is provided by those specifications. For these reasons, this specification focuses on the description of LDevIDs.¶
While the IEEE 802.1AR terminology is utilized in this document, this specification does not claim conformance to IEEE 802.1AR since such a compliance statement goes beyond the use of the terminology and the certificate content and would include the use of management protocols, fulfillment of certain hardware security requirements, and interfaces to access these hardware security modules. Placing these requirements on network equipment like routers may be appropriate but designers of constrained IoT devices have opted for different protocols and hardware security choices.¶
To avoid repetition, this section outlines requirements on X.509 certificates applicable to all PKI entities.¶
Certificates MUST be of type X.509 v3. Note that TLS 1.3 allows to convey payloads other than X.509 certificates in the Certificate message. The description in this section only focuses on X.509 v3 certificates and leaves the description of other formats to other sections or even other specifications.¶
CAs MUST generate non-sequential serial numbers greater than or equal to eight (8) octets from a cryptographically secure pseudo-random number generator. [RFC5280] limits this field to a maximum of 20 octets. The serial number MUST be unique for each certificate issued by a given CA (i.e., the issuer name and the serial number uniquely identify a certificate).¶
The signature MUST be ecdsa-with-SHA256 or stronger [RFC5758].¶
Note: In contrast to IEEE 802.1AR this specification does not require end entity certificates, subordinate CA certificates, and CA certificates to use the same signature algorithm. Furthermore, this specification does not utlize RSA for use with constrained IoT devices and networks.¶
The issuer field MUST contain a non-empty distinguished name (DN) of the entity that has signed and issued the certificate in accordance to [RFC5280].¶
In IoT deployment scenarios it is often expected that the IDevIDs have no maximum validity period. For this purpose the use of a special value for the notAfter date field, the GeneralizedTime value of 99991231235959Z, is utilized. If this is done, then CA certificates and certificates of subordinate CAs cannot have a maximum validity period either. Hence, it requires careful consideration whether it is appropriate to issue IDevID certificates with no maximum validity period.¶
LDevID certificates are, however, issued by the operator or owner, and may be renewed at a regular interval using protocols, such as Enrollment over Secure Transport (EST) [RFC7030] or the Certificate Management Protocol (CMP) [RFC9483]. It is therefore RECOMMENDED to limit the lifetime of these LDevID certificates using the notBefore and notAfter fields, as described in Section 4.1.2.5 of [RFC5280]. Values MUST be expressed in Greenwich Mean Time (Zulu) and MUST include seconds even where the number of seconds is zero.¶
Note that the validity period is defined as the period of time from notBefore through notAfter, inclusive. This means that a hypothetical certificate with a notBefore date of 9 June 2021 at 03:42:01 and a notAfter date of 7 September 2021 at 03:42:01 becomes valid at the beginning of the :01 second, and only becomes invalid at the :02 second, a period that is 90 days plus 1 second. So for a 90-day, notAfter must actually be 03:42:00.¶
For devices without a reliable source of time we advise the use of a device management solution, which typically includes a certificate management protocol, to manage the lifetime of all the certificates used by the device. While this approach does not utilize certificates to its widest extent, it is a solution that extends the capabilities offered by a raw public key approach.¶
The SubjectPublicKeyInfo structure indicates the algorithm and any associated parameters for the ECC public key. This profile uses the id-ecPublicKey algorithm identifier for ECDSA signature keys, as defined and specified in [RFC5480]. This specification assumes that devices supported one of the following algorithms:¶
There is no requirement to use CA certificates and certificates of subordinate CAs to use the same algorithm as the end-entity certificate. Certificates with longer lifetime may well use a cryptographic stronger algorithm.¶
The guidance in Section 4.4.3 of [RFC7925] still holds: for certificate revocation, neither the Online Certificate Status Protocol (OCSP) nor Certificate Revocation Lists (CRLs) are used by constrained IoT devices.¶
Since the publication of RFC 7925 the need for firmware update mechanisms has been reinforced and the work on standardizing a secure and interoperable firmware update mechanism has made substantial progress, see [RFC9019]. RFC 7925 recommends to use a software / firmware update mechanism to provision devices with new trust anchors.¶
The use of device management protocols for IoT devices, which often include an onboarding or bootstrapping mechanism, has also seen considerable uptake in deployed devices and these protocols, some of which are standardized, allow updates of certificates on demand. This enables a deployment model where IoT device utilize end-entity certificates with shorter lifetime making certificate revocation protocols, like OCSP and CRLs, less relevant. Whenever certificates are updated the TLS stack needs to be informed since the communication endpoints need to be aware of the new certificates. This is particularly important when long-lived TLS connections are used. In such a case, the a post-handshake authentication exchange needs to be triggered. TLS 1.3 provides client-to-server post-handshake authentication only. Mutual authentication via post-handshake messages is available via [RFC9261] but requires the application layer protocol to carry the payloads.¶
Hence, instead of performing certificate revocation checks on the IoT device itself this specification recommends to delegate this task to the IoT device operator and to take the necessary action to allow IoT devices to remain operational.¶
The CRL distribution points extension has been defined in RFC 5280 to identify how CRL information is obtained. The authority information access extension indicates how to access information like the online certificate status service (OCSP). Both extensions SHOULD NOT be set. If set, they MUST NOT be marked critical.¶
This section outlines the requirements for root CA certificates.¶
[RFC5280] defines the Subject field as follows: "The subject field identifies the entity associated with the public key stored in the subject public key field." RFC 5280 adds "If the subject is a CA then the subject field MUST be populated with a non-empty distinguished name matching the contents of the issuer field in all certificates issued by the subject CA."¶
The Subject field MUST be present and MUST contain the commonName, the organizationName, and the countryName attribute and MAY contain an organizationalUnitName attribute.¶
Section 4.2.1.2 of [RFC5280] defines the Subject Key Identifier as follows: "The subject key identifier extension provides a means of identifying certificates that contain a particular public key."¶
The Subject Key Identifier extension MUST be set, MUST NOT be marked critical, and MUST contain the key identifier of the public key contained in the subject public key info field.¶
[Editor's Note: Do we need to set the Subject Key Identifier in the CA certificate?]¶
[RFC5280] defines the key usage field as follows: "The key usage extension defines the purpose (e.g., encipherment, signature, certificate signing) of the key contained in the certificate."¶
The Key Usage extension SHOULD be set. If it is set, it MUST be marked critical and the keyCertSign or cRLSign purposes MUST be set. Additional key usages MAY be set depending on the intended usage of the public key.¶
[Editor's Note: Should we harden the requirement to "The Key Usage extension MUST be set.]¶
[RFC5280] defines the extended key usage as follows: "This extension indicates one or more purposes for which the certified public key may be used, in addition to or in place of the basic purposes indicated in the key usage extension."¶
This extendedKeyUsage extension MUST NOT be set.¶
[RFC5280] states that "The Basic Constraints extension identifies whether the subject of the certificate is a CA and the maximum depth of valid certification paths that include this certificate. The cA boolean indicates whether the certified public key may be used to verify certificate signatures."¶
For the pathLenConstraint RFC 5280 makes further statements:¶
The Basic Constraints extension MUST be set, MUST be marked critical, the cA flag MUST be set to true and the pathLenConstraint MUST be omitted.¶
[Editor's Note: Should we soften the requirement to: "The pathLenConstraint field SHOULD NOT be present."]¶
This section outlines the requirements for subordinate CA certificates.¶
The Subject field MUST be set and MUST contain the commonName, the organizationName, and the countryName attribute and MAY contain an organizationalUnitName attribute.¶
The Subject Key Identifier extension MUST be set, MUST NOT be marked critical, and MUST contain the key identifier of the public key contained in the subject public key info field.¶
The Key Usage extension MUST be set, MUST be marked critical, the keyCertSign or cRLSign purposes MUST be set, and the digitalSignature purpose SHOULD be set.¶
The extendedKeyUsage extensed MAY be set depending on the intended usage of the public key.¶
[Editor's Note: Should we harden the requirement to "The extendedKeyUsage MUST NOT be present."]¶
The Basic Constraints extension MUST be set, MUST be marked critical, the cA flag MUST be set to true and the pathLenConstraint MUST be set to 0.¶
[Editor's Note: Should we soften the requriement to "The pathLenConstraint field MAY be present."]¶
The CRL Distribution Point extension SHOULD NOT be set. If it is set, it MUST NOT be marked critical and MUST identify the CRL relevant for this certificate.¶
This section outlines the requirements for end entity certificates.¶
The requirement in Section 4.4.2 of [RFC7925] to only use EUI-64 for end entity certificates as a Subject name is lifted.¶
Two fields are typically used to encode a device identifer, namely the Subject and the subjectAltName fields. Protocol specifications tend to offer recommendations what identifiers to use and the deployment situation is fragmented.¶
The Subject field MAY include a unique device serial number. If the serial number is included, it MUST be encoded in the serialNumber attribute.¶
[RFC5280] defines: "The subject alternative name extension allows identities to be bound to the subject of the certificate. These identities may be included in addition to or in place of the identity in the subject field of the certificate."¶
The subject alternative name extension MAY be set. If it is set, it MUST NOT be marked critical, except when the subject DN contains an empty sequence.¶
If the EUI-64 format is used to identify the subject of an end entity
certificate, it MUST be encoded in a subjectAltName of type DNS-ID as a string
of the form HH-HH-HH-HH-HH-HH-HH-HH
where 'H' is one of the symbols '0'-'9'
or 'A'-'F'.¶
Domain names MUST NOT be encoded in the subject commonName. Instead they
MUST be encoded in a subjectAltName of type DNS-ID. Domain names MUST NOT
contain wildcard (*
) characters. The subjectAltName MUST NOT contain multiple
names.¶
Note: The IEEE 802.1AR recomments to encode information about a Trusted Platform Module (TPM), if present, in the HardwareModuleName. This specification does not follow this recommendation.¶
The Subject Key Identifier SHOULD NOT be included in end-entity certificates. If it is included, then the Subject Key Identifier extension MUST NOT be marked critical, and MUST contain the key identifier of the public key contained in the subject public key info field.¶
[Editor's Note: Should we harden the requirement and state: "The Subject Key Identifier MUST NOT be included in end-entity certificates."]¶
The key usage extension MUST be set and MUST be marked as critical. For signature verification keys the digitialSignature key usage purpose MUST be specified. Other key usages are set according to the intended usage of the key.¶
If enrollment of new certificates uses server-side key generation, encrypted delivery of the private key is required. In such cases the key usage keyEncipherment or keyAgreement MUST be set because the encrypted delivery of the newly generated key involves encryption or agreement of a symmetric key. On-device key generation is, however, the preferred approach.¶
The extendedKeyUsage MUST be present and contain at least one of id-kp-serverAuth or id-kp-clientAuth.¶
In a public key-based key exchange, certificates and public keys are a major contributor to the size of the overall handshake. For example, in a regular TLS 1.3 handshake with minimal ECC certificates and no subordinate CA utilizing the secp256r1 curve with mutual authentication, around 40% of the entire handshake payload is consumed by the two exchanged certificates.¶
Hence, it is not surprising that there is a strong desire to reduce the size of certificates and certificate chains. This has lead to various standardization efforts. Below is a brief summary of what options an implementer has to reduce the bandwidth requirements of a public key-based key exchange. Note that many of the standardized extensions are not readily available in TLS/DTLS stacks since optimizations typically get implemented last.¶
The use of certificate handles, as introduced in cTLS [I-D.ietf-tls-ctls], is a form of caching or compressing certificates as well.¶
Whether to utilize any of the above extensions or a combination of them depends on the anticipated deployment environment, the availability of code, and the constraints imposed by already deployed infrastructure (e.g., CA infrastructure, tool support).¶
According to Section 4.5.3 of [DTLS13], the use of AES-CCM with 8-octet authentication tags (CCM_8) is considered unsuitable for general use with DTLS. This is because it has low integrity limits (i.e., high sensitivity to forgeries) which makes endpoints that negotiate ciphersuites based on such AEAD vulnerable to a trivial DoS attack. See also Sections 5.3 and 5.4 of [I-D.irtf-cfrg-aead-limits] for further discussion on this topic, as well as references to the analysis supporting these conclusions.¶
Specifically, [DTLS13] warns that:¶
> TLS_AES_128_CCM_8_SHA256 MUST NOT be used in DTLS without additional > safeguards against forgery. Implementations MUST set usage limits for > AEAD_AES_128_CCM_8 based on an understanding of any additional forgery > protections that are used.¶
Since all the ciphersuites required by [RFC7925] and [CoAP] rely on CCM_8, there is no alternate ciphersuite available for applications that aim to eliminate the security and availability threats related to CCM_8 while retaining interoperability with the larger ecosystem.¶
In order to ameliorate the situation, this document RECOMMENDS that implementations support the following two ciphersuites:¶
and offer them as their first choice. These ciphersuites provide confidentiality and integrity limits that are considered acceptable in the most general settings. For the details on the exact bounds of both ciphersuites see Section 4.5.3 of [DTLS13]. Note that the GCM-based ciphersuite offers superior interoperability with cloud services at the cost of a slight increase in the wire and peak RAM footprints.¶
When the GCM-based ciphersuite is used with TLS 1.2, the recommendations in Section 7.2.1 of [RFC9325] related to deterministic nonce generation apply. In addition, the integrity limits on key usage detailed in Section 4.4 of [RFC9325] also apply.¶
Table 1 summarizes the recommendations regarding ciphersuites:¶
Ciphersuite | Requirement |
---|---|
TLS_AES_128_CCM_8_SHA256
|
MUST- |
TLS_ECDHE_ECDSA_WITH_AES_128_CCM
|
SHOULD+ |
TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256
|
SHOULD+ |
A number of passive side-channel attacks as well as active fault-injection attacks (e.g., [Ambrose2017]) have been demonstrated to be successful in allowing a malicious third party to gain information about the signing key if a fully deterministic signature scheme (e.g., [RFC6979] ECDSA or EdDSA [RFC8032]) is used.¶
Most of these attacks assume physical access to the device and are therefore especially relevant to smart cards as well as IoT deployments with poor or non-existent physical security.¶
In this security model, it is recommended to combine both randomness and determinism, for example, as described in [I-D.irtf-cfrg-det-sigs-with-noise].¶
A list of open issues can be found at https://github.com/thomas-fossati/draft-tls13-iot/issues¶
This entire document is about security.¶
This document makes no requests to IANA.¶
We would like to thank Ben Kaduk, Hendrik Brockhaus, and John Mattsson.¶