RFC 9289

Internet Engineering Task Force (IETF)                      T. Myklebust
Request for Comments: 9289                                   Hammerspace
Updates: 5531                                              C. Lever, Ed.
Category: Standards Track                                         Oracle
ISSN: 2070-1721                                           September 2022

          Towards Remote Procedure Call Encryption by Default


   This document describes a mechanism that, through the use of
   opportunistic Transport Layer Security (TLS), enables encryption of
   Remote Procedure Call (RPC) transactions while they are in transit.
   The proposed mechanism interoperates with Open Network Computing
   (ONC) RPC implementations that do not support it.  This document
   updates RFC 5531.

Status of This Memo

   This is an Internet Standards Track document.

   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
   Internet Standards 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

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.  Requirements Language
   3.  Terminology
   4.  RPC-with-TLS in Operation
     4.1.  Discovering Server-Side TLS Support
     4.2.  Authentication
       4.2.1.  Using TLS with RPCSEC_GSS
   5.  TLS Requirements
     5.1.  Base Transport Considerations
       5.1.1.  Protected Operation on TCP
       5.1.2.  Protected Operation on UDP
       5.1.3.  Protected Operation on Other Transports
     5.2.  TLS Peer Authentication
       5.2.1.  X.509 Certificates Using PKIX Trust  Extended Key Usage Values
       5.2.2.  Pre-shared Keys
   6.  Security Considerations
     6.1.  The Limitations of Opportunistic Security
       6.1.1.  STRIPTLS Attacks
       6.1.2.  Privacy Leakage before Session Establishment
     6.2.  TLS Identity Management on Clients
     6.3.  Security Considerations for AUTH_SYS on TLS
     6.4.  Best Security Policy Practices
   7.  IANA Considerations
     7.1.  RPC Authentication Flavor
     7.2.  ALPN Identifier for SunRPC
     7.3.  Object Identifier for PKIX Extended Key Usage
     7.4.  Object Identifier for ASN.1 Module
   8.  References
     8.1.  Normative References
     8.2.  Informative References
   Appendix A.  Known Weaknesses of the AUTH_SYS Authentication Flavor
   Appendix B.  ASN.1 Module

   Authors' Addresses

1.  Introduction

   In 2014 the IETF published a document entitled "Pervasive Monitoring
   Is an Attack" [RFC7258], which recognized that unauthorized
   observation of network traffic had become widespread and was a
   subversive threat to all who make use of the Internet at large.  It
   strongly recommended that newly defined Internet protocols should
   make a genuine effort to mitigate monitoring attacks.  Typically,
   this mitigation includes encrypting data in transit.

   The Remote Procedure Call version 2 protocol has been a Proposed
   Standard for three decades (see [RFC5531] and its antecedents).  Over
   twenty years ago, Eisler et al. first introduced RPCSEC_GSS as an in-
   transit encryption mechanism for RPC [RFC2203].  However, experience
   has shown that RPCSEC_GSS with in-transit encryption can be
   challenging to use in practice due to the following:

   *  Parts of each RPC header remain in cleartext, constituting a loss
      of metadata confidentiality.

   *  Offloading the Generic Security Service (GSS) privacy service is
      not practical in large multi-user deployments since each message
      is encrypted using a key based on the issuing RPC user.

   However strong GSS-provided confidentiality is, it cannot provide any
   security if the challenges of using it result in choosing not to
   deploy it at all.

   Moreover, the use of AUTH_SYS remains common despite the adverse
   effects that acceptance of User Identifiers (UIDs) and Group
   Identifiers (GIDs) from unauthenticated clients brings with it.
   Continued use is in part because:

   *  Per-client deployment and administrative costs for the only well-
      defined alternative to AUTH_SYS are expensive at scale.  For
      instance, administrators must provide keying material for each RPC
      client, including transient clients.

   *  GSS host identity management and user identity management
      typically must be enforced in the same security realm.  However,
      cloud providers, for instance, might prefer to remain
      authoritative for host identity but allow tenants to manage user
      identities within their private networks.

   In view of the challenges with the currently available mechanisms for
   authenticating and protecting the confidentiality of RPC
   transactions, this document specifies a transport-layer security
   mechanism that complements the existing ones.  The TLS [RFC8446] and
   Datagram Transport Layer Security (DTLS) [RFC9147] protocols are
   well-established Internet building blocks that protect many standard
   Internet protocols such as the Hypertext Transfer Protocol (HTTP)

   Encrypting at the RPC transport layer accords several significant

   Encryption by Default:  Transport encryption can be enabled without
      additional administrative tasks such as identifying client systems
      to a trust authority and providing each with keying material.

   Encryption Offload:  Hardware support for the GSS privacy service has
      not appeared in the marketplace.  However, the use of a well-
      established transport encryption mechanism that is employed by
      other ubiquitous network protocols makes it more likely that
      encryption offload for RPC is practicable.

   Securing AUTH_SYS:  Most critically, transport encryption can
      significantly reduce several security issues inherent in the
      current widespread use of AUTH_SYS (i.e., acceptance of UIDs and
      GIDs generated by an unauthenticated client).

   Decoupled User and Host Identities:  TLS can be used to authenticate
      peer hosts while other security mechanisms can handle user

   Compatibility:  The imposition of encryption at the transport layer
      protects any upper-layer protocol that employs RPC, without
      alteration of the upper-layer protocol.

   Further, Section 6 of the current document defines policies in line
   with [RFC7435] that enable RPC-with-TLS to be deployed
   opportunistically in environments that contain RPC implementations
   that do not support TLS.  However, specifications for RPC-based
   upper-layer protocols should choose to require even stricter policies
   that guarantee encryption and host authentication are used for all
   RPC transactions to mitigate against pervasive monitoring attacks
   [RFC7258].  Enforcing the use of RPC-with-TLS is of particular
   importance for existing upper-layer protocols whose security
   infrastructure is weak.

   The protocol specification in the current document assumes that
   support for ONC RPC [RFC5531], TLS [RFC8446], PKIX [RFC5280], DNSSEC/
   DNS-Based Authentication of Named Entities (DANE) [RFC6698], and
   optionally RPCSEC_GSS [RFC2203] is available within the platform
   where RPC-with-TLS support is to be added.

2.  Requirements Language

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "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.  Terminology

   This document adopts the terminology introduced in Section 3 of
   [RFC6973] and assumes a working knowledge of the RPC version 2
   protocol [RFC5531] and the TLS version 1.3 protocol [RFC8446].

   Note also that the NFS community long ago adopted the use of the term
   "privacy" from documents such as [RFC2203].  In the current document,
   the authors use the term "privacy" only when referring specifically
   to the historic GSS privacy service defined in [RFC2203].  Otherwise,
   the authors use the term "confidentiality", following the practices
   of contemporary security communities.

   We adhere to the convention that a "client" is a network host that
   actively initiates an association, and a "server" is a network host
   that passively accepts an association request.

   RPC documentation historically refers to the authentication of a
   connecting host as "machine authentication" or "host authentication".
   TLS documentation refers to the same as "peer authentication".  In
   the current document, there is little distinction between these

   The term "user authentication" in the current document refers
   specifically to the RPC caller's credential, provided in the "cred"
   and "verf" fields in each RPC Call.

4.  RPC-with-TLS in Operation

4.1.  Discovering Server-Side TLS Support

   The mechanism described in the current document interoperates fully
   with RPC implementations that do not support RPC-with-TLS.  When an
   RPC-with-TLS-enabled peer encounters a peer that does not support
   RPC-with-TLS, policy settings on the RPC-with-TLS-enabled peer
   determine whether RPC operation continues without the use of TLS or
   is discontinued altogether.

   To achieve this interoperability, we introduce a new RPC
   authentication flavor called AUTH_TLS.  The AUTH_TLS authentication
   flavor signals that the client wants to initiate TLS negotiation if
   the server supports it.  Except for the modifications described in
   this section, the RPC protocol is unaware of security encapsulation
   at the transport layer.  The value of AUTH_TLS is defined in
   Section 7.1.

   An RPC client begins its communication with an RPC server by
   selecting a transport and destination port.  The choice of transport
   and port is typically based on the RPC program that is to be used.
   The RPC client might query the RPC server's RPCBIND service to make
   this selection (The RPCBIND service is described in [RFC1833]).  The
   mechanism described in the current document does not support RPC
   transports other than TCP and UDP.  In all cases, an RPC server MUST
   listen on the same ports for (D)TLS-protected RPC programs as the
   ports used when (D)TLS is not available.

   To protect RPC traffic to a TCP port, the RPC client opens a TCP
   connection to that port and sends a NULL RPC procedure with an
   auth_flavor of AUTH_TLS on that connection.  To protect RPC traffic
   to a UDP port, the RPC client sends a UDP datagram to that port
   containing a NULL RPC procedure with an auth_flavor of AUTH_TLS.  The
   client constructs this RPC procedure as follows:

   *  The length of the opaque data constituting the credential sent in
      the RPC Call message MUST be zero.

   *  The verifier accompanying the credential MUST be an AUTH_NONE
      verifier of length zero.

   *  The flavor value of the verifier in the RPC Reply message received
      from the server MUST be AUTH_NONE.

   *  The length of the verifier's body field is eight.

   *  The bytes of the verifier's body field encode the ASCII characters
      "STARTTLS" as a fixed-length opaque.

   The RPC server signals its corresponding support for RPC-with-TLS by
   replying with a reply_stat of MSG_ACCEPTED and an AUTH_NONE verifier
   containing the "STARTTLS" token.  The client SHOULD proceed with TLS
   session establishment, even if the Reply's accept_stat is not
   SUCCESS.  If the AUTH_TLS probe was done via TCP, the RPC client MUST
   send the "ClientHello" message on the same connection.  If the
   AUTH_TLS probe was done via UDP, the RPC client MUST send the
   "ClientHello" message to the same UDP destination port.

   Conversely, if the Reply's reply_stat is not MSG_ACCEPTED, if its
   verifier flavor is not AUTH_NONE, or if its verifier does not contain
   the "STARTTLS" token, the RPC client MUST NOT send a "ClientHello"
   message.  RPC operation may continue, depending on local policy, but
   without confidentiality, integrity, or peer authentication protection
   from (D)TLS.

   If, after a successful RPC AUTH_TLS probe, the subsequent (D)TLS
   handshake should fail for any reason, the RPC client reports this
   failure to the upper-layer application the same way it reports an
   AUTH_ERROR rejection from the RPC server.

   If an RPC client uses the AUTH_TLS authentication flavor on any
   procedure other than the NULL procedure, or an RPC client sends an
   RPC AUTH_TLS probe within an existing (D)TLS session, the RPC server
   MUST reject that RPC Call by returning a reply_stat of MSG_DENIED
   with a reject_stat of AUTH_ERROR and an auth_stat of AUTH_BADCRED.

   Once the TLS session handshake is complete, the RPC client and server
   have established a secure channel for exchanging RPC transactions.  A
   successful AUTH_TLS probe on one particular port/transport tuple does
   not imply that RPC-with-TLS is available on that same server using a
   different port/transport tuple, nor does it imply that RPC-with-TLS
   will be available in the future using the successfully probed port.

4.2.  Authentication

   There is some overlap between the authentication capabilities of RPC
   and TLS.  The goal of interoperability with implementations that do
   not support TLS requires limiting the combinations that are allowed
   and precisely specifying the role that each layer plays.

   Each RPC server that supports RPC-with-TLS MUST possess a unique
   global identity (e.g., a certificate that is signed by a well-known
   trust anchor).  Such an RPC server MUST request a TLS peer identity
   from each client upon first contact.  There are two different modes
   of client deployment:

   Server-Only Host Authentication
      In this type of deployment, the client can authenticate the server
      host using the presented server peer TLS identity, but the server
      cannot authenticate the client.  In this situation, RPC-with-TLS
      clients are anonymous.  They present no globally unique identifier
      to the server peer.

   Mutual Host Authentication
      In this type of deployment, the client possesses an identity that
      is backed by a trusted entity (e.g., a pre-shared key or a
      certificate validated with a certification path).  As part of the
      TLS handshake, both peers authenticate using the presented TLS
      identities.  If authentication of either peer fails, or if
      authorization based on those identities blocks access to the
      server, the peers MUST reject the association.  Further
      explanation appears in Section 5.2.

   In either of these modes, RPC user authentication is not affected by
   the use of transport layer security.  When a client presents a TLS
   peer identity to an RPC server, the protocol extension described in
   the current document provides no way for the server to know whether
   that identity represents one RPC user on that client or is shared
   amongst many RPC users.  Therefore, a server implementation cannot
   utilize the remote TLS peer identity to authenticate RPC users.

4.2.1.  Using TLS with RPCSEC_GSS

   To use GSS, an RPC server has to possess a GSS service principal.  On
   a TLS session, GSS mutual (peer) authentication occurs as usual, but
   only after a TLS session has been established for communication.
   Authentication of RPCSEC_GSS users is unchanged by the use of TLS.

   RPCSEC_GSS can also perform per-request integrity or confidentiality
   protection.  When operating over a TLS session, these GSS services
   become largely redundant.  An RPC implementation capable of
   concurrently using TLS and RPCSEC_GSS MUST use Generic Security
   Service Application Program Interface (GSS-API) channel binding, as
   defined in [RFC5056], to determine when an underlying transport
   provides a sufficient degree of confidentiality.  RPC-with-TLS
   implementations MUST provide the "tls-exporter" channel binding type,
   as defined in [RFC9266].

5.  TLS Requirements

   When peers negotiate a TLS session that is to transport RPC, the
   following restrictions apply:

   *  Implementations MUST NOT negotiate TLS versions prior to 1.3 (for
      TLS [RFC8446] or DTLS [RFC9147], respectively).  Support for
      mandatory-to-implement cipher suites for the negotiated TLS
      version is REQUIRED.

   *  Implementations MUST conform to the recommendations for TLS usage
      specified in BCP 195 [RFC7525].  Although RFC 7525 permits the use
      of TLS 1.2, the requirement to use TLS 1.3 or later for RPC-with-
      TLS takes precedence.  Further, because TLS 1.3 ciphers are
      qualitatively different than cipher suites in previous versions of
      TLS, and RFC 7525 predates TLS 1.3, the cipher suite
      recommendations in RFC 7525 do not apply to RPC-with-(D)TLS.  A
      strict TLS mode for RPC-with-TLS that protects against STRIPTLS
      attacks is discussed in detail in Section 6.1.1.

   *  Implementations MUST support certificate-based mutual
      authentication.  Support for Pre-Shared Key (PSK) mutual
      authentication is OPTIONAL; see Section 5.2.2 for further details.

   *  Negotiation of a cipher suite providing confidentiality as well as
      integrity protection is REQUIRED.

   Client implementations MUST include the
   "application_layer_protocol_negotiation(16)" extension [RFC7301] in
   their "ClientHello" message and MUST include the protocol identifier
   defined in Section 7.2 in that message's ProtocolNameList value.

   Similarly, in response to the "ClientHello" message, server
   implementations MUST include the
   "application_layer_protocol_negotiation(16)" extension [RFC7301] in
   their "ServerHello" message and MUST include only the protocol
   identifier defined in Section 7.2 in that message's ProtocolNameList

   If the server responds incorrectly (for instance, if the
   "ServerHello" message does not conform to the above requirements),
   the client MUST NOT establish a TLS session for use with RPC on this
   connection.  See [RFC7301] for further details about how to form
   these messages properly.

5.1.  Base Transport Considerations

   There is frequently a strong association between an RPC program and a
   particular destination port number.  The use of TLS or DTLS does not
   change that association.  Thus, it is frequently, though not always,
   the case that a single TLS session carries traffic for only one RPC

5.1.1.  Protected Operation on TCP

   The use of the TLS protocol [RFC8446] protects RPC on TCP
   connections.  Typically, once an RPC client completes the TCP
   handshake, it uses the mechanism described in Section 4.1 to discover
   RPC-with-TLS support for that RPC program on that connection.  Until
   an AUTH_TLS probe is done on a connection, the RPC server treats all
   traffic as RPC messages.  If spurious traffic appears on a TCP
   connection between the initial cleartext AUTH_TLS probe and the TLS
   session handshake, receivers MUST discard that data without response
   and then SHOULD drop the connection.

   The protocol convention specified in the current document assumes
   there can be no more than one concurrent TLS session per TCP
   connection.  This is true of current generations of TLS, but might be
   different in a future version of TLS.

   Once a TLS session is established on a TCP connection, no further
   cleartext communication can occur on that connection until the
   session is terminated.  The use of TLS does not alter RPC record
   framing used on TCP transports.

   Furthermore, if an RPC server responds with PROG_UNAVAIL to an RPC
   Call within an established TLS session, that does not imply that RPC
   server will subsequently reject the same RPC program on a different
   TCP connection.

   Reverse-direction operation occurs only on connected transports such
   as TCP (see Section 2 of [RFC8167]).  To protect reverse-direction
   RPC operations, the RPC server does not establish a separate TLS
   session on the TCP connection but instead uses the existing TLS
   session on that connection to protect these operations.

   When operation is complete, an RPC peer terminates a TLS session by
   sending a TLS closure alert.  It may then close the TCP connection.

5.1.2.  Protected Operation on UDP

   The use of the DTLS protocol [RFC9147] protects RPC carried in UDP
   datagrams.  As soon as a client initializes a UDP socket for use with
   an RPC service, it uses the mechanism described in Section 4.1 to
   discover RPC-with-DTLS support for that RPC program on that port.  If
   spurious traffic appears on a 5-tuple between the initial cleartext
   AUTH_TLS probe and the DTLS association handshake, receivers MUST
   discard that traffic without response.

   Using DTLS does not introduce reliable or in-order semantics to RPC
   on UDP.  The use of DTLS record replay protection is REQUIRED when
   transporting RPC traffic.

   Each RPC message MUST fit in a single DTLS record.  DTLS
   encapsulation has overhead, which reduces the Packetization Layer
   Path MTU (PLPMTU) and thus the maximum RPC payload size.  A possible
   PLPMTU discovery mechanism is offered in [RFC8899].

   The current document does not specify a mechanism that enables a
   server to distinguish between DTLS traffic and unprotected RPC
   traffic directed to the same port.  To make this distinction, each
   peer matches ingress datagrams that appear to be DTLS traffic to
   existing DTLS session state.  A peer treats any datagram that fails
   the matching process as an RPC message.

   Multihomed RPC clients and servers may send protected RPC messages
   via network interfaces that were not involved in the handshake that
   established the DTLS session.  Therefore, when protecting RPC
   traffic, each DTLS handshake MUST include the "connection_id(54)"
   extension described in Section 9 of [RFC9147], and RPC-with-DTLS peer
   endpoints MUST provide a ConnectionID with a nonzero length.
   Endpoints implementing RPC programs that expect a significant number
   of concurrent clients SHOULD employ ConnectionIDs of at least 4 bytes
   in length.

   Sending a TLS closure alert terminates a DTLS session.  Because
   neither DTLS nor UDP provide in-order delivery, after session closure
   there can be ambiguity as to whether a datagram should be interpreted
   as DTLS protected or not.  Therefore, receivers MUST discard
   datagrams exchanged using the same 5-tuple that just terminated the
   DTLS session for a sufficient length of time to ensure that
   retransmissions have ceased and packets already in the network have
   been delivered.  In the absence of more specific data, a period of 60
   seconds is expected to suffice.

5.1.3.  Protected Operation on Other Transports

   Transports that provide intrinsic TLS-level security (e.g., QUIC)
   need to be addressed separately from the current document.  In such
   cases, the use of TLS is not opportunistic as it can be for TCP or

   RPC-over-RDMA can make use of transport layer security below the RDMA
   transport layer [RFC8166].  The exact mechanism is not within the
   scope of the current document.  Because there might not be other
   provisions to exchange client and server certificates, authentication
   material exchange needs to be provided by facilities within a future
   version of the RPC-over-RDMA transport protocol.

5.2.  TLS Peer Authentication

   TLS can perform peer authentication using any of the following

5.2.1.  X.509 Certificates Using PKIX Trust

   X.509 certificates are specified in [X.509].  [RFC5280] provides a
   profile of Internet PKI X.509 public key infrastructure.  RPC-with-
   TLS implementations are REQUIRED to support the PKIX mechanism
   described in [RFC5280].

   The rules and guidelines defined in [RFC6125] apply to RPC-with-TLS
   certificates with the following considerations:

   *  The DNS-ID identifier type is a subjectAltName extension that
      contains a dNSName, as defined in Section of [RFC5280].
      Support for the DNS-ID identifier type is REQUIRED in RPC-with-TLS
      client and server implementations.  Certification authorities that
      issue such certificates MUST support the DNS-ID identifier type.

   *  To specify the identity of an RPC peer as a domain name, the
      certificate MUST contain a subjectAltName extension that contains
      a dNSName.  DNS domain names in RPC-with-TLS certificates MUST NOT
      contain the wildcard character '*' within the identifier.

   *  To specify the identity of an RPC peer as a network identifier
      (netid) or a universal network address (uaddr), the certificate
      MUST contain a subjectAltName extension that contains an

   When validating a server certificate, an RPC-with-TLS client
   implementation takes the following into account:

   *  Certificate validation MUST include the verification rules as per
      Section 6 of [RFC5280] and Section 6 of [RFC6125].

   *  Server certificate validation MUST include a check on whether the
      locally configured expected DNS-ID or iPAddress subjectAltName of
      the server that is contacted matches its presented certificate.

   *  For RPC services accessed by their netids and uaddrs, the
      iPAddress subjectAltName MUST be present in the certificate and
      MUST exactly match the address represented by the universal
      network address.

   An RPC client's domain name and IP address are often assigned
   dynamically; thus, RPC servers cannot rely on those to verify client
   certificates.  Therefore, when an RPC-with-TLS client presents a
   certificate to an RPC-with-TLS server, the server takes the following
   into account:

   *  The server MUST use a procedure conformant to Section 6 of
      [RFC5280] to validate the client certificate's certification path.

   *  The tuple (serial number of the presented certificate; Issuer)
      uniquely identifies the RPC client.  The meaning and syntax of
      these fields is defined in Section 4 of [RFC5280].

   RPC-with-TLS implementations MAY allow the configuration of a set of
   additional properties of the certificate to check for a peer's
   authorization to communicate (e.g., a set of allowed values in
   subjectAltName:URI, a set of allowed X.509v3 Certificate Policies, or
   a set of extended key usages).

   When the configured set of trust anchors changes (e.g., removal of a
   Certification Authority (CA) from the list of trusted CAs; issuance
   of a new Certificate Revocation List (CRL) for a given CA),
   implementations SHOULD reevaluate the certificate originally
   presented in the context of the new configuration and terminate the
   TLS session if the certificate is no longer trustworthy.  Extended Key Usage Values

   Section of [RFC5280] specifies the extended key usage X.509
   certificate extension.  This extension, which may appear in end-
   entity certificates, indicates one or more purposes for which the
   certified public key may be used in addition to or in place of the
   basic purposes indicated in the key usage extension.

   The current document defines two new KeyPurposeId values: one that
   identifies the RPC-with-TLS peer as an RPC client, and one that
   identifies the RPC-with-TLS peer as an RPC server.

   The inclusion of the RPC server value (id-kp-rpcTLSServer) indicates
   that the certificate has been issued for allowing the holder to
   process RPC transactions.

   The inclusion of the RPC client value (id-kp-rpcTLSClient) indicates
   that the certificate has been issued for allowing the holder to
   request RPC transactions.

5.2.2.  Pre-shared Keys

   This mechanism is OPTIONAL to implement.  In this mode, the RPC peer
   can be uniquely identified by keying material that has been shared
   out of band (see Section 2.2 of [RFC8446]).  The PSK Identifier
   SHOULD be exposed at the RPC layer.

6.  Security Considerations

   One purpose of the mechanism described in the current document is to
   protect RPC-based applications against threats to the confidentiality
   of RPC transactions and RPC user identities.  A taxonomy of these
   threats appears in Section 5 of [RFC6973].  Also, Section 6 of
   [RFC7525] contains a detailed discussion of technologies used in
   conjunction with TLS.  Section 8 of [RFC5280] covers important
   considerations about handling certificate material securely.
   Implementers should familiarize themselves with these materials.

   Once a TLS session is established, the RPC payload carried on TLS
   version 1.3 is forward secure.  However, implementers need to be
   aware that replay attacks can occur during session establishment.
   Remedies for such attacks are discussed in detail in Section 8 of
   [RFC8446].  Further, the current document does not provide a profile
   that defines the use of 0-RTT data (see Appendix E.5 of [RFC8446]).
   Therefore, RPC-with-TLS implementations MUST NOT use 0-RTT data.

6.1.  The Limitations of Opportunistic Security

   Readers can find the definition of Opportunistic Security in
   [RFC7435].  A discussion of its underlying principles appears in
   Section 3 of that document.

   The purpose of using an explicitly opportunistic approach is to
   enable interoperation with implementations that do not support RPC-
   with-TLS.  A range of options is allowed by this approach, from "no
   peer authentication or encryption" to "server-only authentication
   with encryption" to "mutual authentication with encryption".  The
   actual security level may indeed be selected based on policy and
   without user intervention.

   In environments where interoperability is a priority, the security
   benefits of TLS are partially or entirely waived.  Implementations of
   the mechanism described in the current document must take care to
   accurately represent to all RPC consumers the level of security that
   is actually in effect, and are REQUIRED to provide an audit log of
   RPC-with-TLS security mode selection.

   In all other cases, the adoption, implementation, and deployment of
   RPC-based upper-layer protocols that enforce the use of TLS
   authentication and encryption (when similar RPCSEC_GSS services are
   not in use) is strongly encouraged.

6.1.1.  STRIPTLS Attacks

   The initial AUTH_TLS probe occurs in cleartext.  An on-path attacker
   can alter a cleartext handshake to make it appear as though TLS
   support is not available on one or both peers.  Client implementers
   can choose from the following to mitigate STRIPTLS attacks:

   *  A TLSA record [RFC6698] can alert clients that TLS is expected to
      work, and provide a binding of a hostname to the X.509 identity.
      If TLS cannot be negotiated or authentication fails, the client
      disconnects and reports the problem.  When an opportunistic
      security policy is in place, a client SHOULD check for the
      existence of a TLSA record for the target server before initiating
      an RPC-with-TLS association.

   *  Client security policy can require that a TLS session is
      established on every connection.  If an attacker spoofs the
      handshake, the client disconnects and reports the problem.  This
      policy prevents an attacker from causing the association to fall
      back to cleartext silently.  If TLSA records are not available,
      this approach is strongly encouraged.

6.1.2.  Privacy Leakage before Session Establishment

   As mentioned earlier, communication between an RPC client and server
   appears in the clear on the network prior to the establishment of a
   TLS session.  This cleartext information usually includes transport
   connection handshake exchanges, the RPC NULL procedure probing
   support for TLS, and the initial parts of TLS session establishment.
   Appendix C of [RFC8446] discusses precautions that can mitigate
   exposure during the exchange of connection handshake information and
   TLS certificate material that might enable attackers to track the RPC
   client.  Note that when PSK authentication is used, the PSK
   identifier is exposed during the TLS handshake and can be used to
   track the RPC client.

   Any RPC traffic that appears on the network before a TLS session has
   been established is vulnerable to monitoring or undetected
   modification.  A secure client implementation limits or prevents any
   RPC exchanges that are not protected.

   The exception to this edict is the initial RPC NULL procedure that
   acts as a STARTTLS message, which cannot be protected.  This RPC NULL
   procedure contains no arguments or results, and the AUTH_TLS
   authentication flavor it uses does not contain user information, so
   there is negligible privacy impact from this exception.

6.2.  TLS Identity Management on Clients

   The goal of RPC-with-TLS is to hide the content of RPC requests while
   they are in transit.  RPC-with-TLS protocol by itself cannot protect
   against exposure of a user's RPC requests to other users on the same

   Moreover, client implementations are free to transmit RPC requests
   for more than one RPC user using the same TLS session.  Depending on
   the details of the client RPC implementation, this means that the
   client's TLS credentials are potentially visible to every RPC user
   that shares a TLS session.  Privileged users may also be able to
   access this TLS identity.

   As a result, client implementations need to carefully segregate TLS
   credentials so that local access to it is restricted to only the
   local users that are authorized to perform operations on the remote
   RPC server.

6.3.  Security Considerations for AUTH_SYS on TLS

   Using a TLS-protected transport when the AUTH_SYS authentication
   flavor is in use addresses several longstanding weaknesses in
   AUTH_SYS (as detailed in Appendix A).  TLS augments AUTH_SYS by
   providing both integrity protection and confidentiality that AUTH_SYS
   lacks.  TLS protects data payloads, RPC headers, and user identities
   against monitoring and alteration while in transit.

   TLS guards against in-transit insertion and deletion of RPC messages,
   thus ensuring the integrity of the message stream between RPC client
   and server.  DTLS does not provide full message stream protection,
   but it does enable receivers to reject nonparticipant messages.  In
   particular, transport-layer encryption plus peer authentication
   protects receiving eXternal Data Representation (XDR) decoders from
   deserializing untrusted data, a common coding vulnerability.
   However, these decoders would still be exposed to untrusted input in
   the case of the compromise of a trusted peer or Certification

   The use of TLS enables strong authentication of the communicating RPC
   peers, providing a degree of non-repudiation.  When AUTH_SYS is used
   with TLS, but the RPC client is unauthenticated, the RPC server still
   acts on RPC requests for which there is no trustworthy
   authentication.  In-transit traffic is protected, but the RPC client
   itself can still misrepresent user identity without server detection.
   TLS without authentication is an improvement from AUTH_SYS without
   encryption, but it leaves a critical security exposure.

   In light of the above, when AUTH_SYS is used, the use of a TLS mutual
   authentication mechanism is RECOMMENDED to prove that the RPC client
   is known to the RPC server.  The server can then determine whether
   the UIDs and GIDs in AUTH_SYS requests from that client can be
   accepted, based on the authenticated identity of the client.

   The use of TLS does not enable RPC clients to detect compromise that
   leads to the impersonation of RPC users.  Also, there continues to be
   a requirement that the mapping of 32-bit user and group ID values to
   user identities is the same on both the RPC client and server.

6.4.  Best Security Policy Practices

   RPC-with-TLS implementations and deployments are strongly encouraged
   to adhere to the following policies to achieve the strongest possible
   security with RPC-with-TLS.

   *  When using AUTH_NULL or AUTH_SYS, both peers are RECOMMENDED to
      have DNSSEC TLSA records, keys with which to perform mutual peer
      authentication using one of the methods described in Section 5.2,
      and a security policy that requires mutual peer authentication and
      rejection of a connection when host authentication fails.

   *  RPCSEC_GSS provides integrity and privacy services that are
      largely redundant when TLS is in use.  These services SHOULD be
      disabled in that case.

7.  IANA Considerations

7.1.  RPC Authentication Flavor

   Following Appendix B of [RFC5531], an entry has been added to the
   "RPC Authentication Flavor Numbers" registry.  The purpose of the new
   authentication flavor is to signal the use of TLS with RPC.  This new
   flavor is not a pseudo-flavor.

   The fields in the new entry have been assigned as follows:

   Identifier String:  AUTH_TLS

   Flavor Name:  TLS

   Value:  7

   Description:  Indicates support for RPC-with-TLS

   Reference:  RFC 9289

7.2.  ALPN Identifier for SunRPC

   Following Section 6 of [RFC7301], the following value has been
   allocated in the "TLS Application-Layer Protocol Negotiation (ALPN)
   Protocol IDs" registry.  The "sunrpc" string identifies SunRPC when
   used over TLS.

   Protocol:  SunRPC

   Identification Sequence:  0x73 0x75 0x6e 0x72 0x70 0x63 ("sunrpc")

   Reference:  RFC 9289

7.3.  Object Identifier for PKIX Extended Key Usage

   Per the Specification Required policy defined in Section 4.6 of
   [RFC8126], the following new values have been registered in the "SMI
   Security for PKIX Extended Key Purpose" registry (
   (see Section and Appendix B).

               | Decimal | Description        | Reference |
               | 33      | id-kp-rpcTLSClient | RFC 9289  |
               | 34      | id-kp-rpcTLSServer | RFC 9289  |

                                 Table 1

7.4.  Object Identifier for ASN.1 Module

   Per the Specification Required policy defined in Section 4.6 of
   [RFC8126], the following new value has been registered in the "SMI
   Security for PKIX Module Identifier" registry ( (see
   Appendix B).

             | Decimal | Description            | Reference |
             | 105     | id-mod-rpcWithTLS-2021 | RFC 9289  |

                                 Table 2

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,

   [RFC5056]  Williams, N., "On the Use of Channel Bindings to Secure
              Channels", RFC 5056, DOI 10.17487/RFC5056, November 2007,

   [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,

   [RFC5531]  Thurlow, R., "RPC: Remote Procedure Call Protocol
              Specification Version 2", RFC 5531, DOI 10.17487/RFC5531,
              May 2009, <https://www.rfc-editor.org/info/rfc5531>.

   [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>.

   [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>.

   [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>.

   [RFC8126]  Cotton, M., Leiba, B., and T. Narten, "Guidelines for
              Writing an IANA Considerations Section in RFCs", BCP 26,
              RFC 8126, DOI 10.17487/RFC8126, June 2017,

   [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>.

   [RFC8446]  Rescorla, E., "The Transport Layer Security (TLS) Protocol
              Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018,

   [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,

   [RFC9266]  Whited, S., "Channel Bindings for TLS 1.3", RFC 9266,
              DOI 10.17487/RFC9266, July 2022,

   [X.509]    International Telecommunication Union, "Information
              technology - Open Systems Interconnection - The Directory:
              Public-key and attribute certificate frameworks", ISO/
              IEC 9594-8, ITU-T Recommendation X.509, October 2019.

   [X.680]    ITU-T, "Information technology - Abstract Syntax Notation
              One (ASN.1): Specification of basic notation", ITU-T
              Recommendation X.680, February 2021,

   [X.690]    ITU-T, "Information technology - ASN.1 encoding rules:
              Specification of Basic Encoding Rules (BER), Canonical
              Encoding Rules (CER) and Distinguished Encoding Rules
              (DER)", ITU-T Recommendation X.690, February 2021,

8.2.  Informative References

   [RFC1833]  Srinivasan, R., "Binding Protocols for ONC RPC Version 2",
              RFC 1833, DOI 10.17487/RFC1833, August 1995,

   [RFC2203]  Eisler, M., Chiu, A., and L. Ling, "RPCSEC_GSS Protocol
              Specification", RFC 2203, DOI 10.17487/RFC2203, September
              1997, <https://www.rfc-editor.org/info/rfc2203>.

   [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>.

   [RFC6973]  Cooper, A., Tschofenig, H., Aboba, B., Peterson, J.,
              Morris, J., Hansen, M., and R. Smith, "Privacy
              Considerations for Internet Protocols", RFC 6973,
              DOI 10.17487/RFC6973, July 2013,

   [RFC7258]  Farrell, S. and H. Tschofenig, "Pervasive Monitoring Is an
              Attack", BCP 188, RFC 7258, DOI 10.17487/RFC7258, May
              2014, <https://www.rfc-editor.org/info/rfc7258>.

   [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>.

   [RFC8166]  Lever, C., Ed., Simpson, W., and T. Talpey, "Remote Direct
              Memory Access Transport for Remote Procedure Call Version
              1", RFC 8166, DOI 10.17487/RFC8166, June 2017,

   [RFC8167]  Lever, C., "Bidirectional Remote Procedure Call on RPC-
              over-RDMA Transports", RFC 8167, DOI 10.17487/RFC8167,
              June 2017, <https://www.rfc-editor.org/info/rfc8167>.

   [RFC8899]  Fairhurst, G., Jones, T., Tüxen, M., Rüngeler, I., and T.
              Völker, "Packetization Layer Path MTU Discovery for
              Datagram Transports", RFC 8899, DOI 10.17487/RFC8899,
              September 2020, <https://www.rfc-editor.org/info/rfc8899>.

   [RFC9110]  Fielding, R., Ed., Nottingham, M., Ed., and J. Reschke,
              Ed., "HTTP Semantics", STD 97, RFC 9110,
              DOI 10.17487/RFC9110, June 2022,

Appendix A.  Known Weaknesses of the AUTH_SYS Authentication Flavor

   The ONC RPC protocol, as specified in [RFC5531], provides several
   modes of security, commonly referred to as "authentication flavors".
   Some of these flavors provide much more than an authentication
   service.  We refer to these as authentication flavors, security
   flavors, or simply, flavors.  One of the earliest and most basic
   flavors is AUTH_SYS, also known as AUTH_UNIX.  Appendix A of
   [RFC5531] specifies AUTH_SYS.

   AUTH_SYS assumes that the RPC client and server both use POSIX-style
   user and group identifiers (each user and group can be distinctly
   represented as a 32-bit unsigned integer).  It also assumes that the
   client and server both use the same mapping of user and group to an
   integer.  One user ID, one primary group ID, and up to 16
   supplemental group IDs are associated with each RPC request.  The
   combination of these identifies the entity on the client that is
   making the request.

   A string identifies peers (hosts) in each RPC request.  [RFC5531]
   does not specify any requirements for this string other than that it
   is no longer than 255 octets.  It does not have to be the same from
   request to request.  Also, it does not have to match the DNS hostname
   of the sending host.  For these reasons, even though most
   implementations fill in their hostname in this field, receivers
   typically ignore its content.

   Appendix A of [RFC5531] contains a brief explanation of security

   |  It should be noted that use of this flavor of authentication does
   |  not guarantee any security for the users or providers of a
   |  service, in itself.  The authentication provided by this scheme
   |  can be considered legitimate only when applications using this
   |  scheme and the network can be secured externally, and privileged
   |  transport addresses are used for the communicating end-points (an
   |  example of this is the use of privileged TCP/UDP ports in UNIX
   |  systems -- note that not all systems enforce privileged transport
   |  address mechanisms).

   It should be clear, therefore, that AUTH_SYS by itself (i.e., without
   strong client authentication) offers little to no communication

   1.  It does not protect the confidentiality or integrity of RPC
       requests, users, or payloads, relying instead on "external"

   2.  It does not provide authentication of RPC peer machines, other
       than inclusion of an unprotected domain name.

   3.  The use of 32-bit unsigned integers as user and group identifiers
       is problematic because these data types are not cryptographically
       signed or otherwise verified by any authority.  In addition, the
       mapping of these integers to users and groups has to be
       consistent amongst a server and its cohort of clients.

   4.  Because the user and group ID fields are not integrity protected,
       AUTH_SYS does not provide non-repudiation.

Appendix B.  ASN.1 Module

   The following module adheres to ASN.1 specifications [X.680] and

     { iso(1) identified-organization(3) dod(6) internet(1)
     security(5) mechanisms(5) pkix(7) id-mod(0)
     id-mod-rpcWithTLS-2021(105) }


   -- OID Arc

     { iso(1) identified-organization(3) dod(6) internet(1)
       security(5) mechanisms(5) pkix(7) kp(3) }

   -- Extended Key Usage Values

   id-kp-rpcTLSClient OBJECT IDENTIFIER ::= { id-kp 33 }
   id-kp-rpcTLSServer OBJECT IDENTIFIER ::= { id-kp 34 }



   Special mention goes to Charles Fisher, author of "Encrypting NFSv4
   with Stunnel TLS" <https://www.linuxjournal.com/content/encrypting-
   nfsv4-stunnel-tls>.  His article inspired the mechanism described in
   the current document.

   Many thanks to Benjamin Coddington, Tigran Mkrtchyan, and Rick
   Macklem for their work on prototype implementations and feedback on
   the current document.  Also, thanks to Benjamin Kaduk for his expert
   guidance on the use of PKIX and TLS and to Russ Housley for his ASN.1
   expertise and for providing other proper finishing touches.  In
   addition, the authors thank the other members of the IESG for their
   astute review comments.  These contributors made this a significantly
   better document.

   Thanks to Derrell Piper for numerous suggestions that improved both
   this simple mechanism and the current document's security-related

   Many thanks to Transport Area Director Magnus Westerlund for his
   sharp questions and careful reading of the final revisions of the
   current document.  The text of Section 5.1.2 is mostly his

   The authors are additionally grateful to Bill Baker, David Black,
   Alan DeKok, Lars Eggert, Olga Kornievskaia, Greg Marsden, Alex
   McDonald, Justin Mazzola Paluska, Tom Talpey, Martin Thomson, and
   Nico Williams for their input and support of this work.

   Finally, special thanks to NFSV4 Working Group Chair and document
   shepherd David Noveck, NFSV4 Working Group Chairs Spencer Shepler and
   Brian Pawlowski, and NFSV4 Working Group Secretary Thomas Haynes for
   their guidance and oversight.

Authors' Addresses

   Trond Myklebust
   Hammerspace Inc.
   4300 El Camino Real, Suite 105
   Los Altos, CA 94022
   United States of America
   Email: trond.myklebust@hammerspace.com

   Charles Lever (editor)
   Oracle Corporation
   United States of America
   Email: chuck.lever@oracle.com