Internet Engineering Task Force (IETF)                        Y. Sheffer
Request for Comments: 9325                                        Intuit
BCP: 195                                                  P. Saint-Andre
Obsoletes: 7525                                              Independent
Updates: 5288, 6066                                           T. Fossati
Category: Best Current Practice                              ARM Limited
ISSN: 2070-1721                                            November 2022


 Recommendations for Secure Use of Transport Layer Security (TLS) and
               Datagram Transport Layer Security (DTLS)

Abstract

  Transport Layer Security (TLS) and Datagram Transport Layer Security
  (DTLS) are used to protect data exchanged over a wide range of
  application protocols and can also form the basis for secure
  transport protocols.  Over the years, the industry has witnessed
  several serious attacks on TLS and DTLS, including attacks on the
  most commonly used cipher suites and their modes of operation.  This
  document provides the latest recommendations for ensuring the
  security of deployed services that use TLS and DTLS.  These
  recommendations are applicable to the majority of use cases.

  RFC 7525, an earlier version of the TLS recommendations, was
  published when the industry was transitioning to TLS 1.2.  Years
  later, this transition is largely complete, and TLS 1.3 is widely
  available.  This document updates the guidance given the new
  environment and obsoletes RFC 7525.  In addition, this document
  updates RFCs 5288 and 6066 in view of recent attacks.

Status of This Memo

  This memo documents an Internet Best Current Practice.

  This document is a product of the Internet Engineering Task Force
  (IETF).  It represents the consensus of the IETF community.  It has
  received public review and has been approved for publication by the
  Internet Engineering Steering Group (IESG).  Further information on
  BCPs is available in Section 2 of RFC 7841.

  Information about the current status of this document, any errata,
  and how to provide feedback on it may be obtained at
  https://www.rfc-editor.org/info/rfc9325.

Copyright Notice

  Copyright (c) 2022 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.

Table of Contents

  1.  Introduction
  2.  Terminology
  3.  General Recommendations
    3.1.  Protocol Versions
      3.1.1.  SSL/TLS Protocol Versions
      3.1.2.  DTLS Protocol Versions
      3.1.3.  Fallback to Lower Versions
    3.2.  Strict TLS
    3.3.  Compression
      3.3.1.  Certificate Compression
    3.4.  TLS Session Resumption
    3.5.  Renegotiation in TLS 1.2
    3.6.  Post-Handshake Authentication
    3.7.  Server Name Indication (SNI)
    3.8.  Application-Layer Protocol Negotiation (ALPN)
    3.9.  Multi-Server Deployment
    3.10. Zero Round-Trip Time (0-RTT) Data in TLS 1.3
  4.  Recommendations: Cipher Suites
    4.1.  General Guidelines
    4.2.  Cipher Suites for TLS 1.2
      4.2.1.  Implementation Details
    4.3.  Cipher Suites for TLS 1.3
    4.4.  Limits on Key Usage
    4.5.  Public Key Length
    4.6.  Truncated HMAC
  5.  Applicability Statement
    5.1.  Security Services
    5.2.  Opportunistic Security
  6.  IANA Considerations
  7.  Security Considerations
    7.1.  Host Name Validation
    7.2.  AES-GCM
      7.2.1.   Nonce Reuse in TLS 1.2
    7.3.  Forward Secrecy
    7.4.  Diffie-Hellman Exponent Reuse
    7.5.  Certificate Revocation
  8.  References
    8.1.  Normative References
    8.2.  Informative References
  Appendix A.  Differences from RFC 7525
  Acknowledgments
  Authors' Addresses

1.  Introduction

  Transport Layer Security (TLS) and Datagram Transport Layer Security
  (DTLS) are used to protect data exchanged over a wide variety of
  application protocols, including HTTP [RFC9112] [RFC9113], IMAP
  [RFC9051], Post Office Protocol (POP) [STD53], SIP [RFC3261], SMTP
  [RFC5321], and the Extensible Messaging and Presence Protocol (XMPP)
  [RFC6120].  Such protocols use both the TLS or DTLS handshake
  protocol and the TLS or DTLS record layer.  Although the TLS
  handshake protocol can also be used with different record layers to
  define secure transport protocols (the most prominent example is QUIC
  [RFC9000]), such transport protocols are not directly in scope for
  this document; nevertheless, many of the recommendations here might
  apply insofar as such protocols use the TLS handshake protocol.

  Over the years leading to 2015, the industry had witnessed serious
  attacks on TLS and DTLS, including attacks on the most commonly used
  cipher suites and their modes of operation.  For instance, both the
  AES-CBC [RFC3602] and RC4 [RFC7465] encryption algorithms, which
  together were once the most widely deployed ciphers, were attacked in
  the context of TLS.  Detailed information about the attacks known
  prior to 2015 is provided in a companion document [RFC7457] to the
  previous version of the TLS recommendations [RFC7525], which will
  help the reader understand the rationale behind the recommendations
  provided here.  That document has not been updated in concert with
  this one; instead, newer attacks are described in this document, as
  are mitigations for those attacks.

  The TLS community reacted to the attacks described in [RFC7457] in
  several ways:

  *  Detailed guidance was published on the use of TLS 1.2 [RFC5246]
     and DTLS 1.2 [RFC6347] along with earlier protocol versions.  This
     guidance is included in the original [RFC7525] and mostly retained
     in this revised version; note that this guidance was mostly
     adopted by the industry since the publication of RFC 7525 in 2015.

  *  Versions of TLS earlier than 1.2 were deprecated [RFC8996].

  *  Version 1.3 of TLS [RFC8446] was released, followed by version 1.3
     of DTLS [RFC9147]; these versions largely mitigate or resolve the
     described attacks.

  Those who implement and deploy TLS and TLS-based protocols need
  guidance on how they can be used securely.  This document provides
  guidance for deployed services as well as for software
  implementations, assuming the implementer expects their code to be
  deployed in the environments defined in Section 5.  Concerning
  deployment, this document targets a wide audience, namely all
  deployers who wish to add authentication (be it one-way only or
  mutual), confidentiality, and data integrity protection to their
  communications.

  The recommendations herein take into consideration the security of
  various mechanisms, their technical maturity and interoperability,
  and their prevalence in implementations at the time of writing.
  Unless it is explicitly called out that a recommendation applies to
  TLS alone or to DTLS alone, each recommendation applies to both TLS
  and DTLS.

  This document attempts to minimize new guidance to TLS 1.2
  implementations, and the overall approach is to encourage systems to
  move to TLS 1.3.  However, this is not always practical.  Newly
  discovered attacks, as well as ecosystem changes, necessitated some
  new requirements that apply to TLS 1.2 environments.  Those are
  summarized in Appendix A.

  Naturally, future attacks are likely, and this document cannot
  address them.  Those who implement and deploy TLS/DTLS and protocols
  based on TLS/DTLS are strongly advised to pay attention to future
  developments.  In particular, although it is known that the creation
  of quantum computers will have a significant impact on the security
  of cryptographic primitives and the technologies that use them,
  currently post-quantum cryptography is a work in progress and it is
  too early to make recommendations; once the relevant specifications
  are standardized in the IETF or elsewhere, this document should be
  updated to reflect best practices at that time.

  As noted, the TLS 1.3 specification resolves many of the
  vulnerabilities listed in this document.  A system that deploys TLS
  1.3 should have fewer vulnerabilities than TLS 1.2 or below.
  Therefore, this document replaces [RFC7525], with an explicit goal to
  encourage migration of most uses of TLS 1.2 to TLS 1.3.

  These are minimum recommendations for the use of TLS in the vast
  majority of implementation and deployment scenarios, with the
  exception of unauthenticated TLS (see Section 5).  Other
  specifications that reference this document can have stricter
  requirements related to one or more aspects of the protocol, based on
  their particular circumstances (e.g., for use with a specific
  application protocol); when that is the case, implementers are
  advised to adhere to those stricter requirements.  Furthermore, this
  document provides a floor, not a ceiling: where feasible,
  administrators of services are encouraged to go beyond the minimum
  support available in implementations to provide the strongest
  security possible.  For example, based on knowledge about the
  deployed base for an existing application protocol and a cost-benefit
  analysis regarding security strength vs. interoperability, a given
  service provider might decide to disable TLS 1.2 entirely and offer
  only TLS 1.3.

  Community knowledge about the strength of various algorithms and
  feasible attacks can change quickly, and experience shows that a Best
  Current Practice (BCP) document about security is a point-in-time
  statement.  Readers are advised to seek out any errata or updates
  that apply to this document.

  This document updates [RFC5288] in view of the [Boeck2016] attack.
  See Section 7.2.1 for the details.

  This document updates [RFC6066] in view of the [ALPACA] attack.  See
  Section 3.7 for the details.

2.  Terminology

  A number of security-related terms in this document are used in the
  sense defined in [RFC4949], including "attack", "authentication",
  "certificate", "cipher", "compromise", "confidentiality",
  "credential", "data integrity", "encryption", "forward secrecy",
  "key", "key length", "self-signed certificate", "strength", and
  "strong".

  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.

3.  General Recommendations

  This section provides general recommendations on the secure use of
  TLS.  Recommendations related to cipher suites are discussed in the
  following section.

3.1.  Protocol Versions

3.1.1.  SSL/TLS Protocol Versions

  It is important both to stop using old, less secure versions of SSL/
  TLS and to start using modern, more secure versions; therefore, the
  following are the recommendations concerning TLS/SSL protocol
  versions:

  *  Implementations MUST NOT negotiate SSL version 2.

     Rationale: Today, SSLv2 is considered insecure [RFC6176].

  *  Implementations MUST NOT negotiate SSL version 3.

     Rationale: SSLv3 [RFC6101] was an improvement over SSLv2 and
     plugged some significant security holes but did not support strong
     cipher suites.  SSLv3 does not support TLS extensions, some of
     which (e.g., renegotiation_info [RFC5746]) are security critical.
     In addition, with the emergence of the Padding Oracle On
     Downgraded Legacy Encryption (POODLE) attack [POODLE], SSLv3 is
     now widely recognized as fundamentally insecure.  See [RFC7568]
     for further details.

  *  Implementations MUST NOT negotiate TLS version 1.0 [RFC2246].

     Rationale: TLS 1.0 (published in 1999) does not support many
     modern, strong cipher suites.  In addition, TLS 1.0 lacks a per-
     record Initialization Vector (IV) for cipher suites based on
     cipher block chaining (CBC) and does not warn against common
     padding errors.  This and other recommendations in this section
     are in line with [RFC8996].

  *  Implementations MUST NOT negotiate TLS version 1.1 [RFC4346].

     Rationale: TLS 1.1 (published in 2006) is a security improvement
     over TLS 1.0 but still does not support certain stronger cipher
     suites that were introduced with the standardization of TLS 1.2 in
     2008, including the cipher suites recommended for TLS 1.2 by this
     document (see Section 4.2 below).

  *  Implementations MUST support TLS 1.2 [RFC5246].

     Rationale: TLS 1.2 is implemented and deployed more widely than
     TLS 1.3 at this time, and when the recommendations in this
     document are followed to mitigate known attacks, the use of TLS
     1.2 is as safe as the use of TLS 1.3.  In most application
     protocols that reuse TLS and DTLS, there is no immediate need to
     migrate solely to TLS 1.3.  Indeed, because many application
     clients are dependent on TLS libraries or operating systems that
     do not yet support TLS 1.3, proactively deprecating TLS 1.2 would
     introduce significant interoperability issues, thus harming
     security more than helping it.  Nevertheless, it is expected that
     a future version of this BCP will deprecate the use of TLS 1.2
     when it is appropriate to do so.

  *  Implementations SHOULD support TLS 1.3 [RFC8446] and, if
     implemented, MUST prefer to negotiate TLS 1.3 over earlier
     versions of TLS.

     Rationale: TLS 1.3 is a major overhaul to the protocol and
     resolves many of the security issues with TLS 1.2.  To the extent
     that an implementation supports TLS 1.2 (even if it defaults to
     TLS 1.3), it MUST follow the recommendations regarding TLS 1.2
     specified in this document.

  *  New transport protocols that integrate the TLS/DTLS handshake
     protocol and/or record layer MUST use only TLS/DTLS 1.3 (for
     instance, QUIC [RFC9001] took this approach).  New application
     protocols that employ TLS/DTLS for channel or session encryption
     MUST integrate with both TLS/DTLS versions 1.2 and 1.3;
     nevertheless, in rare cases where broad interoperability is not a
     concern, application protocol designers MAY choose to forego TLS
     1.2.

     Rationale: Secure deployment of TLS 1.3 is significantly easier
     and less error prone than secure deployment of TLS 1.2.  When
     designing a new secure transport protocol such as QUIC, there is
     no reason to support TLS 1.2.  By contrast, new application
     protocols that reuse TLS need to support both TLS 1.3 and TLS 1.2
     in order to take advantage of underlying library or operating
     system support for both versions.

  This BCP applies to TLS 1.3, TLS 1.2, and earlier versions.  It is
  not safe for readers to assume that the recommendations in this BCP
  apply to any future version of TLS.

3.1.2.  DTLS Protocol Versions

  DTLS, an adaptation of TLS for UDP datagrams, was introduced when TLS
  1.1 was published.  The following are the recommendations with
  respect to DTLS:

  *  Implementations MUST NOT negotiate DTLS version 1.0 [RFC4347].

     Version 1.0 of DTLS correlates to version 1.1 of TLS (see above).

  *  Implementations MUST support DTLS 1.2 [RFC6347].

     Version 1.2 of DTLS correlates to version 1.2 of TLS (see above).
     (There is no version 1.1 of DTLS.)

  *  Implementations SHOULD support DTLS 1.3 [RFC9147] and, if
     implemented, MUST prefer to negotiate DTLS version 1.3 over
     earlier versions of DTLS.

     Version 1.3 of DTLS correlates to version 1.3 of TLS (see above).

3.1.3.  Fallback to Lower Versions

  TLS/DTLS 1.2 clients MUST NOT fall back to earlier TLS versions,
  since those versions have been deprecated [RFC8996].  As a result,
  the downgrade-protection Signaling Cipher Suite Value (SCSV)
  mechanism [RFC7507] is no longer needed for clients.  In addition,
  TLS 1.3 implements a new version-negotiation mechanism.

3.2.  Strict TLS

  The following recommendations are provided to help prevent "SSL
  Stripping" and STARTTLS command injection (attacks that are
  summarized in [RFC7457]):

  *  Many existing application protocols were designed before the use
     of TLS became common.  These protocols typically support TLS in
     one of two ways: either via a separate port for TLS-only
     communication (e.g., port 443 for HTTPS) or via a method for
     dynamically upgrading a channel from unencrypted to TLS protected
     (e.g., STARTTLS, which is used in protocols such as IMAP and
     XMPP).  Regardless of the mechanism for protecting the
     communication channel (TLS-only port or dynamic upgrade), what
     matters is the end state of the channel.  When a protocol defines
     both a dynamic upgrade method and a separate TLS-only method, then
     the separate TLS-only method MUST be supported by implementations
     and MUST be configured by administrators to be used in preference
     to the dynamic upgrade method.  When a protocol supports only a
     dynamic upgrade method, implementations MUST provide a way for
     administrators to set a strict local policy that forbids use of
     plaintext in the absence of a negotiated TLS channel, and
     administrators MUST use this policy.

  *  HTTP client and server implementations intended for use in the
     World Wide Web (see Section 5) MUST support the HTTP Strict
     Transport Security (HSTS) header field [RFC6797] so that web
     servers can advertise that they are willing to accept TLS-only
     clients.  Web servers SHOULD use HSTS to indicate that they are
     willing to accept TLS-only clients, unless they are deployed in
     such a way that using HSTS would in fact weaken overall security
     (e.g., it can be problematic to use HSTS with self-signed
     certificates, as described in Section 11.3 of [RFC6797]).  Similar
     technologies exist for non-HTTP application protocols, such as
     Mail Transfer Agent Strict Transport Security (MTA-STS) for mail
     transfer agents [RFC8461] and methods based on DNS-Based
     Authentication of Named Entities (DANE) [RFC6698] for SMTP
     [RFC7672] and XMPP [RFC7712].

  Rationale: Combining unprotected and TLS-protected communication
  opens the way to SSL Stripping and similar attacks, since an initial
  part of the communication is not integrity protected and therefore
  can be manipulated by an attacker whose goal is to keep the
  communication in the clear.

3.3.  Compression

  In order to help prevent compression-related attacks (summarized in
  Section 2.6 of [RFC7457]) when using TLS 1.2, implementations and
  deployments SHOULD NOT support TLS-level compression (Section 6.2.2
  of [RFC5246]); the only exception is when the application protocol in
  question has been proven not to be open to such attacks.  However,
  even in this case, extreme caution is warranted because of the
  potential for future attacks related to TLS compression.  More
  specifically, the HTTP protocol is known to be vulnerable to
  compression-related attacks.  (This recommendation applies to TLS 1.2
  only, because compression has been removed from TLS 1.3.)

  Rationale: TLS compression has been subject to security attacks such
  as the Compression Ratio Info-leak Made Easy (CRIME) attack.

  Implementers should note that compression at higher protocol levels
  can allow an active attacker to extract cleartext information from
  the connection.  The Browser Reconnaissance and Exfiltration via
  Adaptive Compression of Hypertext (BREACH) attack is one such case.
  These issues can only be mitigated outside of TLS and are thus
  outside the scope of this document.  See Section 2.6 of [RFC7457] for
  further details.

3.3.1.  Certificate Compression

  Certificate chains often take up most of the bytes transmitted during
  the handshake.  In order to manage their size, some or all of the
  following methods can be employed (see also Section 4 of [RFC9191]
  for further suggestions):

  *  Limit the number of names or extensions.

  *  Use keys with small public key representations, like the Elliptic
     Curve Digital Signature Algorithm (ECDSA).

  *  Use certificate compression.

  To achieve the latter, TLS 1.3 defines the compress_certificate
  extension in [RFC8879].  See also Section 5 of [RFC8879] for security
  and privacy considerations associated with its use.  For the
  avoidance of doubt, CRIME-style attacks on TLS compression do not
  apply to certificate compression.

  Due to the strong likelihood of middlebox interference, compression
  in the style of [RFC8879] has not been made available in TLS 1.2.  In
  theory, the cached_info extension defined in [RFC7924] could be used,
  but it is not supported widely enough to be considered a practical
  alternative.

3.4.  TLS Session Resumption

  Session resumption drastically reduces the number of full TLS
  handshakes and thus is an essential performance feature for most
  deployments.

  Stateless session resumption with session tickets is a popular
  strategy.  For TLS 1.2, it is specified in [RFC5077].  For TLS 1.3, a
  more secure mechanism based on the use of a pre-shared key (PSK) is
  described in Section 4.6.1 of [RFC8446].  See [Springall16] for a
  quantitative study of the risks induced by TLS cryptographic
  "shortcuts", including session resumption.

  When it is used, the resumption information MUST be authenticated and
  encrypted to prevent modification or eavesdropping by an attacker.
  Further recommendations apply to session tickets:

  *  A strong cipher MUST be used when encrypting the ticket (at least
     as strong as the main TLS cipher suite).

  *  Ticket-encryption keys MUST be changed regularly, e.g., once every
     week, so as not to negate the benefits of forward secrecy (see
     Section 7.3 for details on forward secrecy).  Old ticket-
     encryption keys MUST be destroyed at the end of the validity
     period.

  *  For similar reasons, session ticket validity MUST be limited to a
     reasonable duration (e.g., half as long as ticket-encryption key
     validity).

  *  TLS 1.2 does not roll the session key forward within a single
     session.  Thus, to prevent an attack where the server's ticket-
     encryption key is stolen and used to decrypt the entire content of
     a session (negating the concept of forward secrecy), a TLS 1.2
     server SHOULD NOT resume sessions that are too old, e.g., sessions
     that have been open longer than two ticket-encryption key rotation
     periods.

  Rationale: Session resumption is another kind of TLS handshake and
  therefore must be as secure as the initial handshake.  This document
  (Section 4) recommends the use of cipher suites that provide forward
  secrecy, i.e., that prevent an attacker who gains momentary access to
  the TLS endpoint (either client or server) and its secrets from
  reading either past or future communication.  The tickets must be
  managed so as not to negate this security property.

  TLS 1.3 provides the powerful option of forward secrecy even within a
  long-lived connection that is periodically resumed.  Section 2.2 of
  [RFC8446] recommends that clients SHOULD send a "key_share" when
  initiating session resumption.  In order to gain forward secrecy,
  this document recommends that server implementations SHOULD select
  the "psk_dhe_ke" PSK key exchange mode and respond with a "key_share"
  to complete an Ephemeral Elliptic Curve Diffie-Hellman (ECDHE)
  exchange on each session resumption.  As a more performant
  alternative, server implementations MAY refrain from responding with
  a "key_share" until a certain amount of time (e.g., measured in
  hours) has passed since the last ECDHE exchange; this implies that
  the "key_share" operation would not occur for the presumed majority
  of session resumption requests (which would occur within a few hours)
  while still ensuring forward secrecy for longer-lived sessions.

  TLS session resumption introduces potential privacy issues where the
  server is able to track the client, in some cases indefinitely.  See
  [Sy2018] for more details.

3.5.  Renegotiation in TLS 1.2

  The recommendations in this section apply to TLS 1.2 only, because
  renegotiation has been removed from TLS 1.3.

  Renegotiation in TLS 1.2 is a handshake that establishes new
  cryptographic parameters for an existing session.  The mechanism
  existed in TLS 1.2 and in earlier protocol versions and was improved
  following several major attacks including a plaintext injection
  attack, CVE-2009-3555 [CVE].

  TLS 1.2 clients and servers MUST implement the renegotiation_info
  extension, as defined in [RFC5746].

  TLS 1.2 clients MUST send renegotiation_info in the Client Hello.  If
  the server does not acknowledge the extension, the client MUST
  generate a fatal handshake_failure alert prior to terminating the
  connection.

  Rationale: It is not safe for a client to connect to a TLS 1.2 server
  that does not support renegotiation_info regardless of whether either
  endpoint actually implements renegotiation.  See also Section 4.1 of
  [RFC5746].

  A related attack resulting from TLS session parameters not being
  properly authenticated is a Triple Handshake [Triple-Handshake].  To
  address this attack, TLS 1.2 implementations MUST support the
  extended_master_secret extension defined in [RFC7627].

3.6.  Post-Handshake Authentication

  Renegotiation in TLS 1.2 was (partially) replaced in TLS 1.3 by
  separate post-handshake authentication and key update mechanisms.  In
  the context of protocols that multiplex requests over a single
  connection (such as HTTP/2 [RFC9113]), post-handshake authentication
  has the same problems as TLS 1.2 renegotiation.  Multiplexed
  protocols SHOULD follow the advice provided for HTTP/2 in
  Section 9.2.3 of [RFC9113].

3.7.  Server Name Indication (SNI)

  TLS implementations MUST support the Server Name Indication (SNI)
  extension defined in Section 3 of [RFC6066] for those higher-level
  protocols that would benefit from it, including HTTPS.  However, the
  actual use of SNI in particular circumstances is a matter of local
  policy.  At the time of writing, a technology for encrypting the SNI
  (called Encrypted Client Hello) is being worked on in the TLS Working
  Group [TLS-ECH].  Once that method has been standardized and widely
  implemented, it will likely be appropriate to recommend its usage in
  a future version of this BCP.

  Rationale: SNI supports deployment of multiple TLS-protected virtual
  servers on a single address, and therefore enables fine-grained
  security for these virtual servers, by allowing each one to have its
  own certificate.  However, SNI also leaks the target domain for a
  given connection; this information leak will be closed by use of TLS
  Encrypted Client Hello once that method has been standardized.

  In order to prevent the attacks described in [ALPACA], a server that
  does not recognize the presented server name SHOULD NOT continue the
  handshake and instead SHOULD fail with a fatal-level
  unrecognized_name(112) alert.  Note that this recommendation updates
  Section 3 of [RFC6066], which stated:

  |  If the server understood the ClientHello extension but does not
  |  recognize the server name, the server SHOULD take one of two
  |  actions: either abort the handshake by sending a fatal-level
  |  unrecognized_name(112) alert or continue the handshake.

  Clients SHOULD abort the handshake if the server acknowledges the SNI
  extension but presents a certificate with a different hostname than
  the one sent by the client.

3.8.  Application-Layer Protocol Negotiation (ALPN)

  TLS implementations (both client- and server-side) MUST support the
  Application-Layer Protocol Negotiation (ALPN) extension [RFC7301].

  In order to prevent "cross-protocol" attacks resulting from failure
  to ensure that a message intended for use in one protocol cannot be
  mistaken for a message for use in another protocol, servers are
  advised to strictly enforce the behavior prescribed in Section 3.2 of
  [RFC7301]:

  |  In the event that the server supports no protocols that the client
  |  advertises, then the server SHALL respond with a fatal
  |  'no_application_protocol' alert.

  Clients SHOULD abort the handshake if the server acknowledges the
  ALPN extension but does not select a protocol from the client list.
  Failure to do so can result in attacks such those described in
  [ALPACA].

  Protocol developers are strongly encouraged to register an ALPN
  identifier for their protocols.  This applies both to new protocols
  and to well-established protocols; however, because the latter might
  have a large deployed base, strict enforcement of ALPN usage may not
  be feasible when an ALPN identifier is registered for a well-
  established protocol.

3.9.  Multi-Server Deployment

  Deployments that involve multiple servers or services can increase
  the size of the attack surface for TLS.  Two scenarios are of
  interest:

  1.  Deployments in which multiple services handle the same domain
      name via different protocols (e.g., HTTP and IMAP).  In this
      case, an attacker might be able to direct a connecting endpoint
      to the service offering a different protocol and mount a cross-
      protocol attack.  In a cross-protocol attack, the client and
      server believe they are using different protocols, which the
      attacker might exploit if messages sent in one protocol are
      interpreted as messages in the other protocol with undesirable
      effects (see [ALPACA] for more detailed information about this
      class of attacks).  To mitigate this threat, service providers
      SHOULD deploy ALPN (see Section 3.8).  In addition, to the extent
      possible, they SHOULD ensure that multiple services handling the
      same domain name provide equivalent levels of security that are
      consistent with the recommendations in this document; such
      measures SHOULD include the handling of configurations across
      multiple TLS servers and protections against compromise of
      credentials held by those servers.

  2.  Deployments in which multiple servers providing the same service
      have different TLS configurations.  In this case, an attacker
      might be able to direct a connecting endpoint to a server with a
      TLS configuration that is more easily exploitable (see [DROWN]
      for more detailed information about this class of attacks).  To
      mitigate this threat, service providers SHOULD ensure that all
      servers providing the same service provide equivalent levels of
      security that are consistent with the recommendations in this
      document.

3.10.  Zero Round-Trip Time (0-RTT) Data in TLS 1.3

  The 0-RTT early data feature is new in TLS 1.3.  It provides reduced
  latency when TLS connections are resumed, at the potential cost of
  certain security properties.  As a result, it requires special
  attention from implementers on both the server and the client side.
  Typically, this extends to the TLS library as well as protocol layers
  above it.

  For HTTP over TLS, refer to [RFC8470] for guidance.

  For QUIC on TLS, refer to Section 9.2 of [RFC9001].

  For other protocols, generic guidance is given in Section 8 and
  Appendix E.5 of [RFC8446].  To paraphrase Appendix E.5, applications
  MUST avoid this feature unless an explicit specification exists for
  the application protocol in question to clarify when 0-RTT is
  appropriate and secure.  This can take the form of an IETF RFC, a
  non-IETF standard, or documentation associated with a non-standard
  protocol.

4.  Recommendations: Cipher Suites

  TLS 1.2 provided considerable flexibility in the selection of cipher
  suites.  Unfortunately, the security of some of these cipher suites
  has degraded over time to the point where some are known to be
  insecure (this is one reason why TLS 1.3 restricted such
  flexibility).  Incorrectly configuring a server leads to no or
  reduced security.  This section includes recommendations on the
  selection and negotiation of cipher suites.

4.1.  General Guidelines

  Cryptographic algorithms weaken over time as cryptanalysis improves:
  algorithms that were once considered strong become weak.
  Consequently, cipher suites using weak algorithms need to be phased
  out and replaced with more secure cipher suites.  This helps to
  ensure that the desired security properties still hold.  SSL/TLS has
  been in existence for well over 20 years and many of the cipher
  suites that have been recommended in various versions of SSL/TLS are
  now considered weak or at least not as strong as desired.  Therefore,
  this section modernizes the recommendations concerning cipher suite
  selection.

  *  Implementations MUST NOT negotiate the cipher suites with NULL
     encryption.

     Rationale: The NULL cipher suites do not encrypt traffic and so
     provide no confidentiality services.  Any entity in the network
     with access to the connection can view the plaintext of contents
     being exchanged by the client and server.  Nevertheless, this
     document does not discourage software from implementing NULL
     cipher suites, since they can be useful for testing and debugging.

  *  Implementations MUST NOT negotiate RC4 cipher suites.

     Rationale: The RC4 stream cipher has a variety of cryptographic
     weaknesses, as documented in [RFC7465].  Note that DTLS
     specifically forbids the use of RC4 already.

  *  Implementations MUST NOT negotiate cipher suites offering less
     than 112 bits of security, including so-called "export-level"
     encryption (which provides 40 or 56 bits of security).

     Rationale: Based on [RFC3766], at least 112 bits of security is
     needed.  40-bit and 56-bit security (found in so-called "export
     ciphers") are considered insecure today.

  *  Implementations SHOULD NOT negotiate cipher suites that use
     algorithms offering less than 128 bits of security.

     Rationale: Cipher suites that offer 112 or more bits but less than
     128 bits of security are not considered weak at this time;
     however, it is expected that their useful lifespan is short enough
     to justify supporting stronger cipher suites at this time.
     128-bit ciphers are expected to remain secure for at least several
     years and 256-bit ciphers until the next fundamental technology
     breakthrough.  Note that, because of so-called "meet-in-the-
     middle" attacks [Multiple-Encryption], some legacy cipher suites
     (e.g., 168-bit Triple DES (3DES)) have an effective key length
     that is smaller than their nominal key length (112 bits in the
     case of 3DES).  Such cipher suites should be evaluated according
     to their effective key length.

  *  Implementations SHOULD NOT negotiate cipher suites based on RSA
     key transport, a.k.a. "static RSA".

     Rationale: These cipher suites, which have assigned values
     starting with the string "TLS_RSA_WITH_*", have several drawbacks,
     especially the fact that they do not support forward secrecy.

  *  Implementations SHOULD NOT negotiate cipher suites based on non-
     ephemeral (static) finite-field Diffie-Hellman (DH) key agreement.
     Similarly, implementations SHOULD NOT negotiate non-ephemeral
     Elliptic Curve DH key agreement.

     Rationale: The former cipher suites, which have assigned values
     prefixed by "TLS_DH_*", have several drawbacks, especially the
     fact that they do not support forward secrecy.  The latter
     ("TLS_ECDH_*") also lack forward secrecy and are subject to
     invalid curve attacks [Jager2015].

  *  Implementations MUST support and prefer to negotiate cipher suites
     offering forward secrecy.  However, TLS 1.2 implementations SHOULD
     NOT negotiate cipher suites based on ephemeral finite-field
     Diffie-Hellman key agreement (i.e., "TLS_DHE_*" suites).  This is
     justified by the known fragility of the construction (see
     [RACCOON]) and the limitation around negotiation, including using
     [RFC7919], which has seen very limited uptake.

     Rationale: Forward secrecy (sometimes called "perfect forward
     secrecy") prevents the recovery of information that was encrypted
     with older session keys, thus limiting how far back in time data
     can be decrypted when an attack is successful.  See Sections 7.3
     and 7.4 for a detailed discussion.

4.2.  Cipher Suites for TLS 1.2

  Given the foregoing considerations, implementation and deployment of
  the following cipher suites is RECOMMENDED:

  *  TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256

  *  TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384

  *  TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256

  *  TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384

  As these are Authenticated Encryption with Associated Data (AEAD)
  algorithms [RFC5116], these cipher suites are supported only in TLS
  1.2 and not in earlier protocol versions.

  Typically, to prefer these suites, the order of suites needs to be
  explicitly configured in server software.  It would be ideal if
  server software implementations were to prefer these suites by
  default.

  Some devices have hardware support for AES Counter Mode with CBC-MAC
  (AES-CCM) but not AES Galois/Counter Mode (AES-GCM), so they are
  unable to follow the foregoing recommendations regarding cipher
  suites.  There are even devices that do not support public key
  cryptography at all, but these are out of scope entirely.

  A cipher suite that operates in CBC (cipher block chaining) mode
  (e.g., TLS_ECDHE_RSA_WITH_AES_128_CBC_SHA256) SHOULD NOT be used
  unless the encrypt_then_mac extension [RFC7366] is also successfully
  negotiated.  This requirement applies to both client and server
  implementations.

  When using ECDSA signatures for authentication of TLS peers, it is
  RECOMMENDED that implementations use the NIST curve P-256.  In
  addition, to avoid predictable or repeated nonces (which could reveal
  the long-term signing key), it is RECOMMENDED that implementations
  implement "deterministic ECDSA" as specified in [RFC6979] and in line
  with the recommendations in [RFC8446].

  Note that implementations of "deterministic ECDSA" may be vulnerable
  to certain side-channel and fault injection attacks precisely because
  of their determinism.  While most fault injection attacks described
  in the literature assume physical access to the device (and therefore
  are more relevant in Internet of Things (IoT) deployments with poor
  or non-existent physical security), some can be carried out remotely
  [Poddebniak2017], e.g., as Rowhammer [Kim2014] variants.  In
  deployments where side-channel attacks and fault injection attacks
  are a concern, implementation strategies combining both randomness
  and determinism (for example, as described in [CFRG-DET-SIGS]) can be
  used to avoid the risk of successful extraction of the signing key.

4.2.1.  Implementation Details

  Clients SHOULD include TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 as the
  first proposal to any server.  Servers MUST prefer this cipher suite
  over weaker cipher suites whenever it is proposed, even if it is not
  the first proposal.  Clients are of course free to offer stronger
  cipher suites, e.g., using AES-256; when they do, the server SHOULD
  prefer the stronger cipher suite unless there are compelling reasons
  (e.g., seriously degraded performance) to choose otherwise.

  The previous version of the TLS recommendations [RFC7525] implicitly
  allowed the old RFC 5246 mandatory-to-implement cipher suite,
  TLS_RSA_WITH_AES_128_CBC_SHA.  At the time of writing, this cipher
  suite does not provide additional interoperability, except with very
  old clients.  As with other cipher suites that do not provide forward
  secrecy, implementations SHOULD NOT support this cipher suite.  Other
  application protocols specify other cipher suites as mandatory to
  implement (MTI).

  [RFC8422] allows clients and servers to negotiate ECDH parameters
  (curves).  Both clients and servers SHOULD include the "Supported
  Elliptic Curves Extension" [RFC8422].  Clients and servers SHOULD
  support the NIST P-256 (secp256r1) [RFC8422] and X25519 (x25519)
  [RFC7748] curves.  Note that [RFC8422] deprecates all but the
  uncompressed point format.  Therefore, if the client sends an
  ec_point_formats extension, the ECPointFormatList MUST contain a
  single element, "uncompressed".

4.3.  Cipher Suites for TLS 1.3

  This document does not specify any cipher suites for TLS 1.3.
  Readers are referred to Section 9.1 of [RFC8446] for cipher suite
  recommendations.

4.4.  Limits on Key Usage

  All ciphers have an upper limit on the amount of traffic that can be
  securely protected with any given key.  In the case of AEAD cipher
  suites, two separate limits are maintained for each key:

  1.  Confidentiality limit (CL), i.e., the number of records that can
      be encrypted.

  2.  Integrity limit (IL), i.e., the number of records that are
      allowed to fail authentication.

  The latter applies to DTLS (and also to QUIC) but not to TLS itself,
  since TLS connections are torn down on the first decryption failure.

  When a sender is approaching CL, the implementation SHOULD initiate a
  new handshake (in TLS 1.3, this can be achieved by sending a
  KeyUpdate message on the established session) to rotate the session
  key.  When a receiver has reached IL, the implementation SHOULD close
  the connection.  Although these recommendations are a best practice,
  implementers need to be aware that it is not always easy to
  accomplish them in protocols that are built on top of TLS/DTLS
  without introducing coordination across layer boundaries.  See
  Section 6 of [RFC9001] for an example of the cooperation that was
  necessary in QUIC between the crypto and transport layers to support
  key updates.  Note that in general, application protocols might not
  be able to emulate that method given their more constrained
  interaction with TLS/DTLS.  As a result of these complexities, these
  recommendations are not mandatory.

  For all TLS 1.3 cipher suites, readers are referred to Section 5.5 of
  [RFC8446] for the values of CL and IL.  For all DTLS 1.3 cipher
  suites, readers are referred to Section 4.5.3 of [RFC9147].

  For all AES-GCM cipher suites recommended for TLS 1.2 and DTLS 1.2 in
  this document, CL can be derived by plugging the corresponding
  parameters into the inequalities in Section 6.1 of [AEAD-LIMITS] that
  apply to random, partially implicit nonces, i.e., the nonce
  construction used in TLS 1.2.  Although the obtained figures are
  slightly higher than those for TLS 1.3, it is RECOMMENDED that the
  same limit of 2^24.5 records is used for both versions.

  For all AES-GCM cipher suites recommended for DTLS 1.2, IL (obtained
  from the same inequalities referenced above) is 2^28.

4.5.  Public Key Length

  When using the cipher suites recommended in this document, two public
  keys are normally used in the TLS handshake: one for the Diffie-
  Hellman key agreement and one for server authentication.  Where a
  client certificate is used, a third public key is added.

  With a key exchange based on modular exponential (MODP) Diffie-
  Hellman groups ("DHE" cipher suites), DH key lengths of at least 2048
  bits are REQUIRED.

  Rationale: For various reasons, in practice, DH keys are typically
  generated in lengths that are powers of two (e.g., 2^10 = 1024 bits,
  2^11 = 2048 bits, 2^12 = 4096 bits).  Because a DH key of 1228 bits
  would be roughly equivalent to only an 80-bit symmetric key
  [RFC3766], it is better to use keys longer than that for the "DHE"
  family of cipher suites.  A DH key of 1926 bits would be roughly
  equivalent to a 100-bit symmetric key [RFC3766].  A DH key of 2048
  bits (equivalent to a 112-bit symmetric key) is the minimum allowed
  by the latest revision of [NIST.SP.800-56A] as of this writing (see
  in particular Appendix D of that document).

  As noted in [RFC3766], correcting for the emergence of The Weizmann
  Institute Relation Locator (TWIRL) machine [TWIRL] would imply that
  1024-bit DH keys yield about 61 bits of equivalent strength and that
  a 2048-bit DH key would yield about 92 bits of equivalent strength.
  The Logjam attack [Logjam] further demonstrates that 1024-bit Diffie-
  Hellman parameters should be avoided.

  With regard to ECDH keys, implementers are referred to the IANA "TLS
  Supported Groups" registry (formerly known as the "EC Named Curve
  Registry") within the "Transport Layer Security (TLS) Parameters"
  registry [IANA_TLS] and in particular to the "recommended" groups.
  Curves of less than 224 bits MUST NOT be used.  This recommendation
  is in line with the latest revision of [NIST.SP.800-56A].

  When using RSA, servers MUST authenticate using certificates with at
  least a 2048-bit modulus for the public key.  In addition, the use of
  the SHA-256 hash algorithm is RECOMMENDED and SHA-1 or MD5 MUST NOT
  be used [RFC9155] (for more details, see also [CAB-Baseline], for
  which the current version at the time of writing is 1.8.4).  Clients
  MUST indicate to servers that they request SHA-256 by using the
  "Signature Algorithms" extension defined in TLS 1.2.  For TLS 1.3,
  the same requirement is already specified by [RFC8446].


4.6.  Truncated HMAC

  Implementations MUST NOT use the Truncated HMAC Extension, defined in
  Section 7 of [RFC6066].

  Rationale: The extension does not apply to the AEAD cipher suites
  recommended above.  However, it does apply to most other TLS cipher
  suites.  Its use has been shown to be insecure in [PatersonRS11].

5.  Applicability Statement

  The recommendations of this document primarily apply to the
  implementation and deployment of application protocols that are most
  commonly used with TLS and DTLS on the Internet today.  Examples
  include, but are not limited to:

  *  Web software and services that wish to protect HTTP traffic with
     TLS.

  *  Email software and services that wish to protect IMAP, Post Office
     Protocol version 3 (POP3), or SMTP traffic with TLS.

  *  Instant-messaging software and services that wish to protect
     Extensible Messaging and Presence Protocol (XMPP) or Internet
     Relay Chat (IRC) traffic with TLS.

  *  Realtime media software and services that wish to protect Secure
     Realtime Transport Protocol (SRTP) traffic with DTLS.

  This document does not modify the implementation and deployment
  recommendations (e.g., mandatory-to-implement cipher suites)
  prescribed by existing application protocols that employ TLS or DTLS.
  If the community that uses such an application protocol wishes to
  modernize its usage of TLS or DTLS to be consistent with the best
  practices recommended here, it needs to explicitly update the
  existing application protocol definition (one example is [RFC7590],
  which updates [RFC6120]).

  Designers of new application protocols developed through the Internet
  Standards Process [RFC2026] are expected at minimum to conform to the
  best practices recommended here, unless they provide documentation of
  compelling reasons that would prevent such conformance (e.g.,
  widespread deployment on constrained devices that lack support for
  the necessary algorithms).

  Although many of the recommendations provided here might also apply
  to QUIC insofar that it uses the TLS 1.3 handshake protocol, QUIC and
  other such secure transport protocols are out of scope of this
  document.  For QUIC specifically, readers are referred to Section 9.2
  of [RFC9001].

  This document does not address the use of TLS in constrained-node
  networks [RFC7228].  For recommendations regarding the profiling of
  TLS and DTLS for small devices with severe constraints on power,
  memory, and processing resources, the reader is referred to [RFC7925]
  and [IOT-PROFILE].

5.1.  Security Services

  This document provides recommendations for an audience that wishes to
  secure their communication with TLS to achieve the following:

  Confidentiality:  all application-layer communication is encrypted
     with the goal that no party should be able to decrypt it except
     the intended receiver.

  Data integrity:  any changes made to the communication in transit are
     detectable by the receiver.

  Authentication:  an endpoint of the TLS communication is
     authenticated as the intended entity to communicate with.

  With regard to authentication, TLS enables authentication of one or
  both endpoints in the communication.  In the context of opportunistic
  security [RFC7435], TLS is sometimes used without authentication.  As
  discussed in Section 5.2, considerations for opportunistic security
  are not in scope for this document.

  If deployers deviate from the recommendations given in this document,
  they need to be aware that they might lose access to one of the
  foregoing security services.

  This document applies only to environments where confidentiality is
  required.  It requires algorithms and configuration options that
  enforce secrecy of the data in transit.

  This document also assumes that data integrity protection is always
  one of the goals of a deployment.  In cases where integrity is not
  required, it does not make sense to employ TLS in the first place.
  There are attacks against confidentiality-only protection that
  utilize the lack of integrity to also break confidentiality (see, for
  instance, [DegabrieleP07] in the context of IPsec).

  This document addresses itself to application protocols that are most
  commonly used on the Internet with TLS and DTLS.  Typically, all
  communication between TLS clients and TLS servers requires all three
  of the above security services.  This is particularly true where TLS
  clients are user agents like web browsers or email clients.

  This document does not address the rarer deployment scenarios where
  one of the above three properties is not desired, such as the use
  case described in Section 5.2.  As another scenario where
  confidentiality is not needed, consider a monitored network where the
  authorities in charge of the respective traffic domain require full
  access to unencrypted (plaintext) traffic and where users collaborate
  and send their traffic in the clear.

5.2.  Opportunistic Security

  There are several important scenarios in which the use of TLS is
  optional, i.e., the client decides dynamically ("opportunistically")
  whether to use TLS with a particular server or to connect in the
  clear.  This practice, often called "opportunistic security", is
  described at length in [RFC7435] and is often motivated by a desire
  for backward compatibility with legacy deployments.

  In these scenarios, some of the recommendations in this document
  might be too strict, since adhering to them could cause fallback to
  cleartext, a worse outcome than using TLS with an outdated protocol
  version or cipher suite.

6.  IANA Considerations

  This document has no IANA actions.

7.  Security Considerations

  This entire document discusses the security practices directly
  affecting applications using the TLS protocol.  This section contains
  broader security considerations related to technologies used in
  conjunction with or by TLS.  The reader is referred to the Security
  Considerations sections of TLS 1.3 [RFC8446], DTLS 1.3 [RFC9147], TLS
  1.2 [RFC5246], and DTLS 1.2 [RFC6347] for further context.

7.1.  Host Name Validation

  Application authors should take note that some TLS implementations do
  not validate host names.  If the TLS implementation they are using
  does not validate host names, authors might need to write their own
  validation code or consider using a different TLS implementation.

  It is noted that the requirements regarding host name validation
  (and, in general, binding between the TLS layer and the protocol that
  runs above it) vary between different protocols.  For HTTPS, these
  requirements are defined by Sections 4.3.3, 4.3.4, and 4.3.5 of
  [RFC9110].

  Host name validation is security-critical for all common TLS use
  cases.  Without it, TLS ensures that the certificate is valid and
  guarantees possession of the private key but does not ensure that the
  connection terminates at the desired endpoint.  Readers are referred
  to [RFC6125] for further details regarding generic host name
  validation in the TLS context.  In addition, that RFC contains a long
  list of application protocols, some of which implement a policy very
  different from HTTPS.

  If the host name is discovered indirectly and insecurely (e.g., by a
  cleartext DNS query for an SRV or Mail Exchange (MX) record), it
  SHOULD NOT be used as a reference identifier [RFC6125] even when it
  matches the presented certificate.  This proviso does not apply if
  the host name is discovered securely (for further discussion, see
  [RFC7673] and [RFC7672]).

  Host name validation typically applies only to the leaf "end entity"
  certificate.  Naturally, in order to ensure proper authentication in
  the context of the PKI, application clients need to verify the entire
  certification path in accordance with [RFC5280].

7.2.  AES-GCM

  Section 4.2 recommends the use of the AES-GCM authenticated
  encryption algorithm.  Please refer to Section 6 of [RFC5288] for
  security considerations that apply specifically to AES-GCM when used
  with TLS.

7.2.1.   Nonce Reuse in TLS 1.2

  The existence of deployed TLS stacks that mistakenly reuse the AES-
  GCM nonce is documented in [Boeck2016], showing there is an actual
  risk of AES-GCM getting implemented insecurely and thus making TLS
  sessions that use an AES-GCM cipher suite vulnerable to attacks such
  as [Joux2006].  (See [CVE] records: CVE-2016-0270, CVE-2016-10213,
  CVE-2016-10212, and CVE-2017-5933.)

  While this problem has been fixed in TLS 1.3, which enforces a
  deterministic method to generate nonces from record sequence numbers
  and shared secrets for all its AEAD cipher suites (including AES-
  GCM), TLS 1.2 implementations could still choose their own
  (potentially insecure) nonce generation methods.

  It is therefore RECOMMENDED that TLS 1.2 implementations use the
  64-bit sequence number to populate the nonce_explicit part of the GCM
  nonce, as described in the first two paragraphs of Section 5.3 of
  [RFC8446].  This stronger recommendation updates Section 3 of
  [RFC5288], which specifies that the use of 64-bit sequence numbers to
  populate the nonce_explicit field is optional.

  We note that at the time of writing, there are no cipher suites
  defined for nonce-reuse-resistant algorithms such as AES-GCM-SIV
  [RFC8452].

7.3.  Forward Secrecy

  Forward secrecy (also called "perfect forward secrecy" or "PFS" and
  defined in [RFC4949]) is a defense against an attacker who records
  encrypted conversations where the session keys are only encrypted
  with the communicating parties' long-term keys.

  Should the attacker be able to obtain these long-term keys at some
  point later in time, the session keys and thus the entire
  conversation could be decrypted.

  In the context of TLS and DTLS, such compromise of long-term keys is
  not entirely implausible.  It can happen, for example, due to:

  *  A client or server being attacked by some other attack vector, and
     the private key retrieved.

  *  A long-term key retrieved from a device that has been sold or
     otherwise decommissioned without prior wiping.

  *  A long-term key used on a device as a default key [Heninger2012].

  *  A key generated by a trusted third party like a CA and later
     retrieved from it by either extortion or compromise
     [Soghoian2011].

  *  A cryptographic breakthrough or the use of asymmetric keys with
     insufficient length [Kleinjung2010].

  *  Social engineering attacks against system administrators.

  *  Collection of private keys from inadequately protected backups.

  Forward secrecy ensures in such cases that it is not feasible for an
  attacker to determine the session keys even if the attacker has
  obtained the long-term keys some time after the conversation.  It
  also protects against an attacker who is in possession of the long-
  term keys but remains passive during the conversation.

  Forward secrecy is generally achieved by using the Diffie-Hellman
  scheme to derive session keys.  The Diffie-Hellman scheme has both
  parties maintain private secrets and send parameters over the network
  as modular powers over certain cyclic groups.  The properties of the
  so-called Discrete Logarithm Problem (DLP) allow the parties to
  derive the session keys without an eavesdropper being able to do so.
  There is currently no known attack against DLP if sufficiently large
  parameters are chosen.  A variant of the Diffie-Hellman scheme uses
  elliptic curves instead of the originally proposed modular
  arithmetic.  Given the current state of the art, Elliptic Curve
  Diffie-Hellman appears to be more efficient, permits shorter key
  lengths, and allows less freedom for implementation errors than
  finite-field Diffie-Hellman.

  Unfortunately, many TLS/DTLS cipher suites were defined that do not
  feature forward secrecy, e.g., TLS_RSA_WITH_AES_256_CBC_SHA256.  This
  document therefore advocates strict use of forward-secrecy-only
  ciphers.

7.4.  Diffie-Hellman Exponent Reuse

  For performance reasons, it is not uncommon for TLS implementations
  to reuse Diffie-Hellman and Elliptic Curve Diffie-Hellman exponents
  across multiple connections.  Such reuse can result in major security
  issues:

  *  If exponents are reused for too long (in some cases, even as
     little as a few hours), an attacker who gains access to the host
     can decrypt previous connections.  In other words, exponent reuse
     negates the effects of forward secrecy.

  *  TLS implementations that reuse exponents should test the DH public
     key they receive for group membership, in order to avoid some
     known attacks.  These tests are not standardized in TLS at the
     time of writing, although general guidance in this area is
     provided by [NIST.SP.800-56A] and available in many protocol
     implementations.

  *  Under certain conditions, the use of static finite-field DH keys,
     or of ephemeral finite-field DH keys that are reused across
     multiple connections, can lead to timing attacks (such as those
     described in [RACCOON]) on the shared secrets used in Diffie-
     Hellman key exchange.

  *  An "invalid curve" attack can be mounted against Elliptic Curve DH
     if the victim does not verify that the received point lies on the
     correct curve.  If the victim is reusing the DH secrets, the
     attacker can repeat the probe varying the points to recover the
     full secret (see [Antipa2003] and [Jager2015]).

  To address these concerns:

  *  TLS implementations SHOULD NOT use static finite-field DH keys and
     SHOULD NOT reuse ephemeral finite-field DH keys across multiple
     connections.

  *  Server implementations that want to reuse Elliptic Curve DH keys
     SHOULD either use a "safe curve" [SAFECURVES] (e.g., X25519) or
     perform the checks described in [NIST.SP.800-56A] on the received
     points.

7.5.  Certificate Revocation

  The following considerations and recommendations represent the
  current state of the art regarding certificate revocation, even
  though no complete and efficient solution exists for the problem of
  checking the revocation status of common public key certificates
  [RFC5280]:

  *  Certificate revocation is an important tool when recovering from
     attacks on the TLS implementation as well as cases of misissued
     certificates.  TLS implementations MUST implement a strategy to
     distrust revoked certificates.

  *  Although Certificate Revocation Lists (CRLs) are the most widely
     supported mechanism for distributing revocation information, they
     have known scaling challenges that limit their usefulness, despite
     workarounds such as partitioned CRLs and delta CRLs.  The more
     modern [CRLite] and the follow-on Let's Revoke [LetsRevoke] build
     on the availability of Certificate Transparency [RFC9162] logs and
     aggressive compression to allow practical use of the CRL
     infrastructure, but at the time of writing, neither solution is
     deployed for client-side revocation processing at scale.

  *  Proprietary mechanisms that embed revocation lists in the web
     browser's configuration database cannot scale beyond the few most
     heavily used web servers.

  *  The Online Certification Status Protocol (OCSP) [RFC6960] in its
     basic form presents both scaling and privacy issues.  In addition,
     clients typically "soft-fail", meaning that they do not abort the
     TLS connection if the OCSP server does not respond.  (However,
     this might be a workaround to avoid denial-of-service attacks if
     an OCSP responder is taken offline.)  For a recent survey of the
     status of OCSP deployment in the web PKI, see [Chung18].

  *  The TLS Certificate Status Request extension (Section 8 of
     [RFC6066]), commonly called "OCSP stapling", resolves the
     operational issues with OCSP.  However, it is still ineffective in
     the presence of an active on-path attacker because the attacker
     can simply ignore the client's request for a stapled OCSP
     response.

  *  [RFC7633] defines a certificate extension that indicates that
     clients must expect stapled OCSP responses for the certificate and
     must abort the handshake ("hard-fail") if such a response is not
     available.

  *  OCSP stapling as used in TLS 1.2 does not extend to intermediate
     certificates within a certificate chain.  The Multiple Certificate
     Status extension [RFC6961] addresses this shortcoming, but it has
     seen little deployment and had been deprecated by [RFC8446].  As a
     result, although this extension was recommended for TLS 1.2 in
     [RFC7525], it is no longer recommended by this document.

  *  TLS 1.3 (Section 4.4.2.1 of [RFC8446]) allows the association of
     OCSP information with intermediate certificates by using an
     extension to the CertificateEntry structure.  However, using this
     facility remains impractical because many certification
     authorities (CAs) either do not publish OCSP for CA certificates
     or publish OCSP reports with a lifetime that is too long to be
     useful.

  *  Both CRLs and OCSP depend on relatively reliable connectivity to
     the Internet, which might not be available to certain kinds of
     nodes.  A common example is newly provisioned devices that need to
     establish a secure connection in order to boot up for the first
     time.

  For the common use cases of public key certificates in TLS, servers
  SHOULD support the following as a best practice given the current
  state of the art and as a foundation for a possible future solution:
  OCSP [RFC6960] and OCSP stapling using the status_request extension
  defined in [RFC6066].  Note that the exact mechanism for embedding
  the status_request extension differs between TLS 1.2 and 1.3.  As a
  matter of local policy, server operators MAY request that CAs issue
  must-staple [RFC7633] certificates for the server and/or for client
  authentication, but we recommend reviewing the operational conditions
  before deciding on this approach.

  The considerations in this section do not apply to scenarios where
  the DNS-Based Authentication of Named Entities (DANE) TLSA resource
  record [RFC6698] is used to signal to a client which certificate a
  server considers valid and good to use for TLS connections.

8.  References

8.1.  Normative References

  [RFC2119]  Bradner, S., "Key words for use in RFCs to Indicate
             Requirement Levels", BCP 14, RFC 2119,
             DOI 10.17487/RFC2119, March 1997,
             <https://www.rfc-editor.org/info/rfc2119>.

  [RFC3766]  Orman, H. and P. Hoffman, "Determining Strengths For
             Public Keys Used For Exchanging Symmetric Keys", BCP 86,
             RFC 3766, DOI 10.17487/RFC3766, April 2004,
             <https://www.rfc-editor.org/info/rfc3766>.

  [RFC5246]  Dierks, T. and E. Rescorla, "The Transport Layer Security
             (TLS) Protocol Version 1.2", RFC 5246,
             DOI 10.17487/RFC5246, August 2008,
             <https://www.rfc-editor.org/info/rfc5246>.

  [RFC5288]  Salowey, J., Choudhury, A., and D. McGrew, "AES Galois
             Counter Mode (GCM) Cipher Suites for TLS", RFC 5288,
             DOI 10.17487/RFC5288, August 2008,
             <https://www.rfc-editor.org/info/rfc5288>.

  [RFC5746]  Rescorla, E., Ray, M., Dispensa, S., and N. Oskov,
             "Transport Layer Security (TLS) Renegotiation Indication
             Extension", RFC 5746, DOI 10.17487/RFC5746, February 2010,
             <https://www.rfc-editor.org/info/rfc5746>.

  [RFC6066]  Eastlake 3rd, D., "Transport Layer Security (TLS)
             Extensions: Extension Definitions", RFC 6066,
             DOI 10.17487/RFC6066, January 2011,
             <https://www.rfc-editor.org/info/rfc6066>.

  [RFC6125]  Saint-Andre, P. and J. Hodges, "Representation and
             Verification of Domain-Based Application Service Identity
             within Internet Public Key Infrastructure Using X.509
             (PKIX) Certificates in the Context of Transport Layer
             Security (TLS)", RFC 6125, DOI 10.17487/RFC6125, March
             2011, <https://www.rfc-editor.org/info/rfc6125>.

  [RFC6176]  Turner, S. and T. Polk, "Prohibiting Secure Sockets Layer
             (SSL) Version 2.0", RFC 6176, DOI 10.17487/RFC6176, March
             2011, <https://www.rfc-editor.org/info/rfc6176>.

  [RFC6347]  Rescorla, E. and N. Modadugu, "Datagram Transport Layer
             Security Version 1.2", RFC 6347, DOI 10.17487/RFC6347,
             January 2012, <https://www.rfc-editor.org/info/rfc6347>.

  [RFC6979]  Pornin, T., "Deterministic Usage of the Digital Signature
             Algorithm (DSA) and Elliptic Curve Digital Signature
             Algorithm (ECDSA)", RFC 6979, DOI 10.17487/RFC6979, August
             2013, <https://www.rfc-editor.org/info/rfc6979>.

  [RFC7301]  Friedl, S., Popov, A., Langley, A., and E. Stephan,
             "Transport Layer Security (TLS) Application-Layer Protocol
             Negotiation Extension", RFC 7301, DOI 10.17487/RFC7301,
             July 2014, <https://www.rfc-editor.org/info/rfc7301>.

  [RFC7366]  Gutmann, P., "Encrypt-then-MAC for Transport Layer
             Security (TLS) and Datagram Transport Layer Security
             (DTLS)", RFC 7366, DOI 10.17487/RFC7366, September 2014,
             <https://www.rfc-editor.org/info/rfc7366>.

  [RFC7465]  Popov, A., "Prohibiting RC4 Cipher Suites", RFC 7465,
             DOI 10.17487/RFC7465, February 2015,
             <https://www.rfc-editor.org/info/rfc7465>.

  [RFC7627]  Bhargavan, K., Ed., Delignat-Lavaud, A., Pironti, A.,
             Langley, A., and M. Ray, "Transport Layer Security (TLS)
             Session Hash and Extended Master Secret Extension",
             RFC 7627, DOI 10.17487/RFC7627, September 2015,
             <https://www.rfc-editor.org/info/rfc7627>.

  [RFC7748]  Langley, A., Hamburg, M., and S. Turner, "Elliptic Curves
             for Security", RFC 7748, DOI 10.17487/RFC7748, January
             2016, <https://www.rfc-editor.org/info/rfc7748>.

  [RFC8174]  Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC
             2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174,
             May 2017, <https://www.rfc-editor.org/info/rfc8174>.

  [RFC8422]  Nir, Y., Josefsson, S., and M. Pegourie-Gonnard, "Elliptic
             Curve Cryptography (ECC) Cipher Suites for Transport Layer
             Security (TLS) Versions 1.2 and Earlier", RFC 8422,
             DOI 10.17487/RFC8422, August 2018,
             <https://www.rfc-editor.org/info/rfc8422>.

  [RFC8446]  Rescorla, E., "The Transport Layer Security (TLS) Protocol
             Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018,
             <https://www.rfc-editor.org/info/rfc8446>.

  [RFC8996]  Moriarty, K. and S. Farrell, "Deprecating TLS 1.0 and TLS
             1.1", BCP 195, RFC 8996, DOI 10.17487/RFC8996, March 2021,
             <https://www.rfc-editor.org/info/rfc8996>.

  [RFC9147]  Rescorla, E., Tschofenig, H., and N. Modadugu, "The
             Datagram Transport Layer Security (DTLS) Protocol Version
             1.3", RFC 9147, DOI 10.17487/RFC9147, April 2022,
             <https://www.rfc-editor.org/info/rfc9147>.

  [RFC9155]  Velvindron, L., Moriarty, K., and A. Ghedini, "Deprecating
             MD5 and SHA-1 Signature Hashes in TLS 1.2 and DTLS 1.2",
             RFC 9155, DOI 10.17487/RFC9155, December 2021,
             <https://www.rfc-editor.org/info/rfc9155>.

8.2.  Informative References

  [AEAD-LIMITS]
             Günther, F., Thomson, M., and C. A. Wood, "Usage Limits on
             AEAD Algorithms", Work in Progress, Internet-Draft, draft-
             irtf-cfrg-aead-limits-05, 11 July 2022,
             <https://datatracker.ietf.org/doc/html/draft-irtf-cfrg-
             aead-limits-05>.

  [ALPACA]   Brinkmann, M., Dresen, C., Merget, R., Poddebniak, D.,
             Müller, J., Somorovsky, J., Schwenk, J., and S. Schinzel,
             "ALPACA: Application Layer Protocol Confusion - Analyzing
             and Mitigating Cracks in TLS Authentication", 30th USENIX
             Security Symposium (USENIX Security 21), August 2021,
             <https://www.usenix.org/conference/usenixsecurity21/
             presentation/brinkmann>.

  [Antipa2003]
             Antipa, A., Brown, D. R. L., Menezes, A., Struik, R., and
             S. Vanstone, "Validation of Elliptic Curve Public Keys",
             Public Key Cryptography - PKC 2003, December 2003,
             <https://doi.org/10.1007/3-540-36288-6_16>.

  [Boeck2016]
             Böck, H., Zauner, A., Devlin, S., Somorovsky, J., and P.
             Jovanovic, "Nonce-Disrespecting Adversaries: Practical
             Forgery Attacks on GCM in TLS", May 2016,
             <https://eprint.iacr.org/2016/475.pdf>.

  [CAB-Baseline]
             CA/Browser Forum, "Baseline Requirements for the Issuance
             and Management of Publicly-Trusted Certificates",
             Version 1.8.4, April 2022,
             <https://cabforum.org/documents/>.

  [CFRG-DET-SIGS]
             Preuß Mattsson, J., Thormarker, E., and S. Ruohomaa,
             "Deterministic ECDSA and EdDSA Signatures with Additional
             Randomness", Work in Progress, Internet-Draft, draft-irtf-
             cfrg-det-sigs-with-noise-00, 8 August 2022,
             <https://datatracker.ietf.org/doc/html/draft-irtf-cfrg-
             det-sigs-with-noise-00>.

  [Chung18]  Chung, T., Lok, J., Chandrasekaran, B., Choffnes, D.,
             Levin, D., Maggs, B., Mislove, A., Rula, J., Sullivan, N.,
             and C. Wilson, "Is the Web Ready for OCSP Must-Staple?",
             Proceedings of the Internet Measurement Conference 2018,
             DOI 10.1145/3278532.3278543, October 2018,
             <https://doi.org/10.1145/3278532.3278543>.

  [CRLite]   Larisch, J., Choffnes, D., Levin, D., Maggs, B., Mislove,
             A., and C. Wilson, "CRLite: A Scalable System for Pushing
             All TLS Revocations to All Browsers", 2017 IEEE Symposium
             on Security and Privacy (SP), DOI 10.1109/sp.2017.17, May
             2017, <https://doi.org/10.1109/sp.2017.17>.

  [CVE]      MITRE, "Common Vulnerabilities and Exposures",
             <https://cve.mitre.org>.

  [DegabrieleP07]
             Degabriele, J. and K. Paterson, "Attacking the IPsec
             Standards in Encryption-only Configurations", 2007 IEEE
             Symposium on Security and Privacy (SP '07),
             DOI 10.1109/sp.2007.8, May 2007,
             <https://doi.org/10.1109/sp.2007.8>.

  [DROWN]    Aviram, N., Schinzel, S., Somorovsky, J., Heninger, N.,
             Dankel, M., Steube, J., Valenta, L., Adrian, D.,
             Halderman, J., Dukhovni, V., Käsper, E., Cohney, S.,
             Engels, S., Paar, C., and Y. Shavitt, "DROWN: Breaking TLS
             using SSLv2", 25th USENIX Security Symposium (USENIX
             Security 16), August 2016,
             <https://www.usenix.org/conference/usenixsecurity16/
             technical-sessions/presentation/aviram>.

  [Heninger2012]
             Heninger, N., Durumeric, Z., Wustrow, E., and J. A.
             Halderman, "Mining Your Ps and Qs: Detection of Widespread
             Weak Keys in Network Devices", 21st Usenix Security
             Symposium, August 2012.

  [IANA_TLS] IANA, "Transport Layer Security (TLS) Parameters",
             <https://www.iana.org/assignments/tls-parameters>.

  [IOT-PROFILE]
             Tschofenig, H. and T. Fossati, "TLS/DTLS 1.3 Profiles for
             the Internet of Things", Work in Progress, Internet-Draft,
             draft-ietf-uta-tls13-iot-profile-05, 6 July 2022,
             <https://datatracker.ietf.org/doc/html/draft-ietf-uta-
             tls13-iot-profile-05>.

  [Jager2015]
             Jager, T., Schwenk, J., and J. Somorovsky, "Practical
             Invalid Curve Attacks on TLS-ECDH", Computer Security --
             ESORICS 2015, pp. 407-425,
             DOI 10.1007/978-3-319-24174-6_21, 2015,
             <https://doi.org/10.1007/978-3-319-24174-6_21>.

  [Joux2006] Joux, A., "Authentication Failures in NIST version of
             GCM", 2006, <https://csrc.nist.gov/csrc/media/projects/
             block-cipher-techniques/documents/bcm/comments/800-38-
             series-drafts/gcm/joux_comments.pdf>.

  [Kim2014]  Kim, Y., Daly, R., Kim, J., Fallin, C., Lee, J. H., Lee,
             D., Wilkerson, C., Lai, K., and O. Mutlu, "Flipping Bits
             in Memory Without Accessing Them: An Experimental Study of
             DRAM Disturbance Errors", DOI 10.1109/ISCA.2014.6853210,
             July 2014, <https://users.ece.cmu.edu/~yoonguk/papers/kim-
             isca14.pdf>.

  [Kleinjung2010]
             Kleinjung, T., Aoki, K., Franke, J., Lenstra, A., Thomé,
             E., Bos, J., Gaudry, P., Kruppa, A., Montgomery, P.,
             Osvik, D., te Riele, H., Timofeev, A., and P. Zimmermann,
             "Factorization of a 768-Bit RSA Modulus", Advances in
             Cryptology - CRYPTO 2010, pp. 333-350,
             DOI 10.1007/978-3-642-14623-7_18, 2010,
             <https://doi.org/10.1007/978-3-642-14623-7_18>.

  [LetsRevoke]
             Smith, T., Dickinson, L., and K. Seamons, "Let's Revoke:
             Scalable Global Certificate Revocation", Proceedings 2020
             Network and Distributed System Security Symposium,
             DOI 10.14722/ndss.2020.24084, February 2020,
             <https://doi.org/10.14722/ndss.2020.24084>.

  [Logjam]   Adrian, D., Bhargavan, K., Durumeric, Z., Gaudry, P.,
             Green, M., Halderman, J., Heninger, N., Springall, D.,
             Thomé, E., Valenta, L., VanderSloot, B., Wustrow, E.,
             Zanella-Béguelin, S., and P. Zimmermann, "Imperfect
             Forward Secrecy: How Diffie-Hellman Fails in Practice",
             Proceedings of the 22nd ACM SIGSAC Conference on Computer
             and Communications Security, pp. 5-17,
             DOI 10.1145/2810103.2813707, October 2015,
             <https://doi.org/10.1145/2810103.2813707>.

  [Multiple-Encryption]
             Merkle, R. and M. Hellman, "On the security of multiple
             encryption", Communications of the ACM, Vol. 24, Issue 7,
             pp. 465-467, DOI 10.1145/358699.358718, July 1981,
             <https://doi.org/10.1145/358699.358718>.

  [NIST.SP.800-56A]
             National Institute of Standards and Technology,
             "Recommendation for Pair-Wise Key-Establishment Schemes
             Using Discrete Logarithm Cryptography", Revision 3, NIST
             Special Publication 800-56A,
             DOI 10.6028/NIST.SP.800-56Ar3, April 2018,
             <https://doi.org/10.6028/NIST.SP.800-56Ar3>.

  [PatersonRS11]
             Paterson, K., Ristenpart, T., and T. Shrimpton, "Tag Size
             Does Matter: Attacks and Proofs for the TLS Record
             Protocol", Proceedings of the 17th International
             conference on The Theory and Application of Cryptology and
             Information Security, pp. 372-389,
             DOI 10.1007/978-3-642-25385-0_20, December 2011,
             <https://doi.org/10.1007/978-3-642-25385-0_20>.

  [Poddebniak2017]
             Poddebniak, D., Somorovsky, J., Schinzel, S., Lochter, M.,
             and P. Rösler, "Attacking Deterministic Signature Schemes
             using Fault Attacks", Conference: 2018 IEEE European
             Symposium on Security and Privacy,
             DOI 10.1109/EuroSP.2018.00031, April 2018,
             <https://eprint.iacr.org/2017/1014.pdf>.

  [POODLE]   US-CERT, "SSL 3.0 Protocol Vulnerability and POODLE
             Attack", October 2014,
             <https://www.us-cert.gov/ncas/alerts/TA14-290A>.

  [RACCOON]  Merget, R., Brinkmann, M., Aviram, N., Somorovsky, J.,
             Mittmann, J., and J. Schwenk, "Raccoon Attack: Finding and
             Exploiting Most-Significant-Bit-Oracles in TLS-DH(E)",
             30th USENIX Security Symposium (USENIX Security 21), 2021,
             <https://www.usenix.org/conference/usenixsecurity21/
             presentation/merget>.

  [RFC2026]  Bradner, S., "The Internet Standards Process -- Revision
             3", BCP 9, RFC 2026, DOI 10.17487/RFC2026, October 1996,
             <https://www.rfc-editor.org/info/rfc2026>.

  [RFC2246]  Dierks, T. and C. Allen, "The TLS Protocol Version 1.0",
             RFC 2246, DOI 10.17487/RFC2246, January 1999,
             <https://www.rfc-editor.org/info/rfc2246>.

  [RFC3261]  Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston,
             A., Peterson, J., Sparks, R., Handley, M., and E.
             Schooler, "SIP: Session Initiation Protocol", RFC 3261,
             DOI 10.17487/RFC3261, June 2002,
             <https://www.rfc-editor.org/info/rfc3261>.

  [RFC3602]  Frankel, S., Glenn, R., and S. Kelly, "The AES-CBC Cipher
             Algorithm and Its Use with IPsec", RFC 3602,
             DOI 10.17487/RFC3602, September 2003,
             <https://www.rfc-editor.org/info/rfc3602>.

  [RFC4346]  Dierks, T. and E. Rescorla, "The Transport Layer Security
             (TLS) Protocol Version 1.1", RFC 4346,
             DOI 10.17487/RFC4346, April 2006,
             <https://www.rfc-editor.org/info/rfc4346>.

  [RFC4347]  Rescorla, E. and N. Modadugu, "Datagram Transport Layer
             Security", RFC 4347, DOI 10.17487/RFC4347, April 2006,
             <https://www.rfc-editor.org/info/rfc4347>.

  [RFC4949]  Shirey, R., "Internet Security Glossary, Version 2",
             FYI 36, RFC 4949, DOI 10.17487/RFC4949, August 2007,
             <https://www.rfc-editor.org/info/rfc4949>.

  [RFC5077]  Salowey, J., Zhou, H., Eronen, P., and H. Tschofenig,
             "Transport Layer Security (TLS) Session Resumption without
             Server-Side State", RFC 5077, DOI 10.17487/RFC5077,
             January 2008, <https://www.rfc-editor.org/info/rfc5077>.

  [RFC5116]  McGrew, D., "An Interface and Algorithms for Authenticated
             Encryption", RFC 5116, DOI 10.17487/RFC5116, January 2008,
             <https://www.rfc-editor.org/info/rfc5116>.

  [RFC5280]  Cooper, D., Santesson, S., Farrell, S., Boeyen, S.,
             Housley, R., and W. Polk, "Internet X.509 Public Key
             Infrastructure Certificate and Certificate Revocation List
             (CRL) Profile", RFC 5280, DOI 10.17487/RFC5280, May 2008,
             <https://www.rfc-editor.org/info/rfc5280>.

  [RFC5321]  Klensin, J., "Simple Mail Transfer Protocol", RFC 5321,
             DOI 10.17487/RFC5321, October 2008,
             <https://www.rfc-editor.org/info/rfc5321>.

  [RFC6101]  Freier, A., Karlton, P., and P. Kocher, "The Secure
             Sockets Layer (SSL) Protocol Version 3.0", RFC 6101,
             DOI 10.17487/RFC6101, August 2011,
             <https://www.rfc-editor.org/info/rfc6101>.

  [RFC6120]  Saint-Andre, P., "Extensible Messaging and Presence
             Protocol (XMPP): Core", RFC 6120, DOI 10.17487/RFC6120,
             March 2011, <https://www.rfc-editor.org/info/rfc6120>.

  [RFC6698]  Hoffman, P. and J. Schlyter, "The DNS-Based Authentication
             of Named Entities (DANE) Transport Layer Security (TLS)
             Protocol: TLSA", RFC 6698, DOI 10.17487/RFC6698, August
             2012, <https://www.rfc-editor.org/info/rfc6698>.

  [RFC6797]  Hodges, J., Jackson, C., and A. Barth, "HTTP Strict
             Transport Security (HSTS)", RFC 6797,
             DOI 10.17487/RFC6797, November 2012,
             <https://www.rfc-editor.org/info/rfc6797>.

  [RFC6960]  Santesson, S., Myers, M., Ankney, R., Malpani, A.,
             Galperin, S., and C. Adams, "X.509 Internet Public Key
             Infrastructure Online Certificate Status Protocol - OCSP",
             RFC 6960, DOI 10.17487/RFC6960, June 2013,
             <https://www.rfc-editor.org/info/rfc6960>.

  [RFC6961]  Pettersen, Y., "The Transport Layer Security (TLS)
             Multiple Certificate Status Request Extension", RFC 6961,
             DOI 10.17487/RFC6961, June 2013,
             <https://www.rfc-editor.org/info/rfc6961>.

  [RFC7228]  Bormann, C., Ersue, M., and A. Keranen, "Terminology for
             Constrained-Node Networks", RFC 7228,
             DOI 10.17487/RFC7228, May 2014,
             <https://www.rfc-editor.org/info/rfc7228>.

  [RFC7435]  Dukhovni, V., "Opportunistic Security: Some Protection
             Most of the Time", RFC 7435, DOI 10.17487/RFC7435,
             December 2014, <https://www.rfc-editor.org/info/rfc7435>.

  [RFC7457]  Sheffer, Y., Holz, R., and P. Saint-Andre, "Summarizing
             Known Attacks on Transport Layer Security (TLS) and
             Datagram TLS (DTLS)", RFC 7457, DOI 10.17487/RFC7457,
             February 2015, <https://www.rfc-editor.org/info/rfc7457>.

  [RFC7507]  Moeller, B. and A. Langley, "TLS Fallback Signaling Cipher
             Suite Value (SCSV) for Preventing Protocol Downgrade
             Attacks", RFC 7507, DOI 10.17487/RFC7507, April 2015,
             <https://www.rfc-editor.org/info/rfc7507>.

  [RFC7525]  Sheffer, Y., Holz, R., and P. Saint-Andre,
             "Recommendations for Secure Use of Transport Layer
             Security (TLS) and Datagram Transport Layer Security
             (DTLS)", BCP 195, RFC 7525, DOI 10.17487/RFC7525, May
             2015, <https://www.rfc-editor.org/info/rfc7525>.

  [RFC7568]  Barnes, R., Thomson, M., Pironti, A., and A. Langley,
             "Deprecating Secure Sockets Layer Version 3.0", RFC 7568,
             DOI 10.17487/RFC7568, June 2015,
             <https://www.rfc-editor.org/info/rfc7568>.

  [RFC7590]  Saint-Andre, P. and T. Alkemade, "Use of Transport Layer
             Security (TLS) in the Extensible Messaging and Presence
             Protocol (XMPP)", RFC 7590, DOI 10.17487/RFC7590, June
             2015, <https://www.rfc-editor.org/info/rfc7590>.

  [RFC7633]  Hallam-Baker, P., "X.509v3 Transport Layer Security (TLS)
             Feature Extension", RFC 7633, DOI 10.17487/RFC7633,
             October 2015, <https://www.rfc-editor.org/info/rfc7633>.

  [RFC7672]  Dukhovni, V. and W. Hardaker, "SMTP Security via
             Opportunistic DNS-Based Authentication of Named Entities
             (DANE) Transport Layer Security (TLS)", RFC 7672,
             DOI 10.17487/RFC7672, October 2015,
             <https://www.rfc-editor.org/info/rfc7672>.

  [RFC7673]  Finch, T., Miller, M., and P. Saint-Andre, "Using DNS-
             Based Authentication of Named Entities (DANE) TLSA Records
             with SRV Records", RFC 7673, DOI 10.17487/RFC7673, October
             2015, <https://www.rfc-editor.org/info/rfc7673>.

  [RFC7712]  Saint-Andre, P., Miller, M., and P. Hancke, "Domain Name
             Associations (DNA) in the Extensible Messaging and
             Presence Protocol (XMPP)", RFC 7712, DOI 10.17487/RFC7712,
             November 2015, <https://www.rfc-editor.org/info/rfc7712>.

  [RFC7919]  Gillmor, D., "Negotiated Finite Field Diffie-Hellman
             Ephemeral Parameters for Transport Layer Security (TLS)",
             RFC 7919, DOI 10.17487/RFC7919, August 2016,
             <https://www.rfc-editor.org/info/rfc7919>.

  [RFC7924]  Santesson, S. and H. Tschofenig, "Transport Layer Security
             (TLS) Cached Information Extension", RFC 7924,
             DOI 10.17487/RFC7924, July 2016,
             <https://www.rfc-editor.org/info/rfc7924>.

  [RFC7925]  Tschofenig, H., Ed. and T. Fossati, "Transport Layer
             Security (TLS) / Datagram Transport Layer Security (DTLS)
             Profiles for the Internet of Things", RFC 7925,
             DOI 10.17487/RFC7925, July 2016,
             <https://www.rfc-editor.org/info/rfc7925>.

  [RFC8452]  Gueron, S., Langley, A., and Y. Lindell, "AES-GCM-SIV:
             Nonce Misuse-Resistant Authenticated Encryption",
             RFC 8452, DOI 10.17487/RFC8452, April 2019,
             <https://www.rfc-editor.org/info/rfc8452>.

  [RFC8461]  Margolis, D., Risher, M., Ramakrishnan, B., Brotman, A.,
             and J. Jones, "SMTP MTA Strict Transport Security (MTA-
             STS)", RFC 8461, DOI 10.17487/RFC8461, September 2018,
             <https://www.rfc-editor.org/info/rfc8461>.

  [RFC8470]  Thomson, M., Nottingham, M., and W. Tarreau, "Using Early
             Data in HTTP", RFC 8470, DOI 10.17487/RFC8470, September
             2018, <https://www.rfc-editor.org/info/rfc8470>.

  [RFC8879]  Ghedini, A. and V. Vasiliev, "TLS Certificate
             Compression", RFC 8879, DOI 10.17487/RFC8879, December
             2020, <https://www.rfc-editor.org/info/rfc8879>.

  [RFC9000]  Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based
             Multiplexed and Secure Transport", RFC 9000,
             DOI 10.17487/RFC9000, May 2021,
             <https://www.rfc-editor.org/info/rfc9000>.

  [RFC9001]  Thomson, M., Ed. and S. Turner, Ed., "Using TLS to Secure
             QUIC", RFC 9001, DOI 10.17487/RFC9001, May 2021,
             <https://www.rfc-editor.org/info/rfc9001>.

  [RFC9051]  Melnikov, A., Ed. and B. Leiba, Ed., "Internet Message
             Access Protocol (IMAP) - Version 4rev2", RFC 9051,
             DOI 10.17487/RFC9051, August 2021,
             <https://www.rfc-editor.org/info/rfc9051>.

  [RFC9110]  Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke,
             Ed., "HTTP Semantics", STD 97, RFC 9110,
             DOI 10.17487/RFC9110, June 2022,
             <https://www.rfc-editor.org/info/rfc9110>.

  [RFC9112]  Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke,
             Ed., "HTTP/1.1", STD 99, RFC 9112, DOI 10.17487/RFC9112,
             June 2022, <https://www.rfc-editor.org/info/rfc9112>.

  [RFC9113]  Thomson, M., Ed. and C. Benfield, Ed., "HTTP/2", RFC 9113,
             DOI 10.17487/RFC9113, June 2022,
             <https://www.rfc-editor.org/info/rfc9113>.

  [RFC9162]  Laurie, B., Messeri, E., and R. Stradling, "Certificate
             Transparency Version 2.0", RFC 9162, DOI 10.17487/RFC9162,
             December 2021, <https://www.rfc-editor.org/info/rfc9162>.

  [RFC9191]  Sethi, M., Preuß Mattsson, J., and S. Turner, "Handling
             Large Certificates and Long Certificate Chains in TLS-
             Based EAP Methods", RFC 9191, DOI 10.17487/RFC9191,
             February 2022, <https://www.rfc-editor.org/info/rfc9191>.

  [SAFECURVES]
             Bernstein, D. J. and T. Lange, "SafeCurves: choosing safe
             curves for elliptic-curve cryptography", December 2014,
             <https://safecurves.cr.yp.to>.

  [Soghoian2011]
             Soghoian, C. and S. Stamm, "Certified Lies: Detecting and
             Defeating Government Interception Attacks Against SSL",
             SSRN Electronic Journal, DOI 10.2139/ssrn.1591033, April
             2010, <https://doi.org/10.2139/ssrn.1591033>.

  [Springall16]
             Springall, D., Durumeric, Z., and J. Halderman, "Measuring
             the Security Harm of TLS Crypto Shortcuts", Proceedings of
             the 2016 Internet Measurement Conference, pp. 33-47,
             DOI 10.1145/2987443.2987480, November 2016,
             <https://doi.org/10.1145/2987443.2987480>.

  [STD53]    Myers, J. and M. Rose, "Post Office Protocol - Version 3",
             STD 53, RFC 1939, May 1996.

             <https://www.rfc-editor.org/info/std53>

  [Sy2018]   Sy, E., Burkert, C., Federrath, H., and M. Fischer,
             "Tracking Users across the Web via TLS Session
             Resumption", Proceedings of the 34th Annual Computer
             Security Applications Conference, pp. 289-299,
             DOI 10.1145/3274694.3274708, December 2018,
             <https://doi.org/10.1145/3274694.3274708>.

  [TLS-ECH]  Rescorla, E., Oku, K., Sullivan, N., and C. A. Wood, "TLS
             Encrypted Client Hello", Work in Progress, Internet-Draft,
             draft-ietf-tls-esni-15, 3 October 2022,
             <https://datatracker.ietf.org/doc/html/draft-ietf-tls-
             esni-15>.

  [Triple-Handshake]
             Bhargavan, K., Lavaud, A., Fournet, C., Pironti, A., and
             P. Strub, "Triple Handshakes and Cookie Cutters: Breaking
             and Fixing Authentication over TLS", 2014 IEEE Symposium
             on Security and Privacy, DOI 10.1109/sp.2014.14, May 2014,
             <https://doi.org/10.1109/sp.2014.14>.

  [TWIRL]    Shamir, A. and E. Tromer, "Factoring Large Numbers with
             the TWIRL Device", 2014 IEEE Symposium on Security and
             Privacy, DOI 10.1007/978-3-540-45146-4_1, 2004,
             <https://cs.tau.ac.il/~tromer/papers/twirl.pdf>.

Appendix A.  Differences from RFC 7525

  This revision of the Best Current Practices contains numerous
  changes, and this section is focused on the normative changes.

  *  High-level differences:

     -  Described the expectations from new TLS-incorporating transport
        protocols and from new application protocols layered on TLS.

     -  Clarified items (e.g., renegotiation) that only apply to TLS
        1.2.

     -  Changed the status of TLS 1.0 and 1.1 from "SHOULD NOT" to
        "MUST NOT".

     -  Added TLS 1.3 at a "SHOULD" level.

     -  Made similar changes to DTLS.

     -  Included specific guidance for multiplexed protocols.

     -  MUST-level implementation requirement for ALPN and more
        specific SHOULD-level guidance for ALPN and SNI.

     -  Clarified discussion of strict TLS policies, including MUST-
        level recommendations.

     -  Limits on key usage.

     -  New attacks since [RFC7457]: ALPACA, Raccoon, Logjam, and
        "Nonce-Disrespecting Adversaries".

     -  RFC 6961 (OCSP status_request_v2) has been deprecated.

     -  MUST-level requirement for server-side RSA certificates to have
        a 2048-bit modulus at a minimum, replacing a "SHOULD".

  *  Differences specific to TLS 1.2:

     -  SHOULD-level guidance on AES-GCM nonce generation.

     -  SHOULD NOT use (static or ephemeral) finite-field DH key
        agreement.

     -  SHOULD NOT reuse ephemeral finite-field DH keys across multiple
        connections.

     -  SHOULD NOT use static Elliptic Curve DH key exchange.

     -  2048-bit DH is now a "MUST" and ECDH minimal curve size is 224
        (vs. 192 previously).

     -  Support for extended_master_secret is now a "MUST" (previously
        it was a soft recommendation, as the RFC had not been published
        at the time).  Also removed other, more complicated, related
        mitigations.

     -  MUST-level restriction on session ticket validity, replacing a
        "SHOULD".

     -  SHOULD-level restriction on the TLS session duration, depending
        on the rotation period of an [RFC5077] ticket key.

     -  Dropped TLS_DHE_RSA_WITH_AES from the recommended ciphers.

     -  Added TLS_ECDHE_ECDSA_WITH_AES to the recommended ciphers.

     -  SHOULD NOT use the old MTI cipher suite,
        TLS_RSA_WITH_AES_128_CBC_SHA.

     -  Recommended curve X25519 alongside NIST P-256.

  *  Differences specific to TLS 1.3:

     -  New TLS 1.3 capabilities: 0-RTT.

     -  Removed capabilities: renegotiation and compression.

     -  Added mention of TLS Encrypted Client Hello, but no
        recommendation for use until it is finalized.

     -  SHOULD-level requirement for forward secrecy in TLS 1.3 session
        resumption.

     -  Generic MUST-level guidance to avoid 0-RTT unless it is
        documented for the particular protocol.

Acknowledgments

  Thanks to Alexey Melnikov, Alvaro Retana, Andrei Popov, Ben Kaduk,
  Christian Huitema, Corey Bonnell, Cullen Jennings, Daniel Kahn
  Gillmor, David Benjamin, Eric Rescorla, Éric Vyncke, Francesca
  Palombini, Hannes Tschofenig, Hubert Kario, Ilari Liusvaara, John
  Preuß Mattsson, John R. Levine, Julien Élie, Lars Eggert, Leif
  Johansson, Magnus Westerlund, Martin Duke, Martin Thomson, Mohit
  Sahni, Nick Sullivan, Nimrod Aviram, Paul Wouters, Peter Gutmann,
  Rich Salz, Robert Sayre, Robert Wilton, Roman Danyliw, Ryan Sleevi,
  Sean Turner, Stephen Farrell, Tim Evans, Valery Smyslov, Viktor
  Dukhovni, and Warren Kumari for helpful comments and discussions that
  have shaped this document.

  The authors gratefully acknowledge the contribution of Ralph Holz,
  who was a coauthor of RFC 7525, the previous version of the TLS
  recommendations.

  See RFC 7525 for additional acknowledgments specific to the previous
  version of the TLS recommendations.

Authors' Addresses

  Yaron Sheffer
  Intuit
  Email: [email protected]


  Peter Saint-Andre
  Independent
  Email: [email protected]


  Thomas Fossati
  ARM Limited
  Email: [email protected]

You are viewing proxied material from gopher.fnord.one. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.