]> Juniper Networks

kwatsen@juniper.net

Operations NETCONF Working Group call-home This RFC presents NETCONF Call Home and RESTCONF Call Home, which enable a NETCONF or RESTCONF server to initiate a secure connection to a NETCONF or RESTCONF client respectively. This draft contains many placeholder values that need to be replaced with finalized values at the time of publication. This note summarizes all of the substitutions that are needed. Please note that no other RFC Editor instructions are specified anywhere else in this document. Artwork in this document contains placeholder references for this draft. Please apply the following replacement: XXXX --> the assigned RFC value for this draft This document contains references to another draft in progress, both in the Normative References section, as well as in body text throughout. Please update the following reference to reflect its final RFC assignment: draft-ietf-netconf-restconf Artwork in this document contains placeholder values for ports pending IANA assignment from "draft-ietf-netconf-call-home". Please apply the following replacements: PORT-X --> the assigned port value for "netconf-ch-ssh" PORT-Y --> the assigned port value for "netconf-ch-tls" PORT-Z --> the assigned port value for "restconf-ch-tls" The following two Appendix sections are to be removed prior to publication: Appendix A. Change Log Appendix B. Open Issues

This RFC presents NETCONF Call Home and RESTCONF Call Home, which enable a NETCONF or RESTCONF server to initiate a secure connection to a NETCONF or RESTCONF client respectively. NETCONF Call Home supports both of the secure transports used by the NETCONF protocol , SSH and TLS. The NETCONF protocol's binding to SSH is defined in . The NETCONF protocol's binding to TLS is defined in . RESTCONF Call Home only supports TLS, the same as the RESTCONF protocol . The RESTCONF protocol's binding to TLS is defined in . The SSH protocol is defined in . The TLS protocol is defined in . Both the SSH and TLS protocols are layered on top of the TCP protocol, which is defined in . Both NETCONF Call Home and RESTCONF Call Home preserve all but one of the client/server roles in their respective protocol stacks, as compared to client-initiated NETCONF and RESTCONF connections. The one and only role reversal that occurs is at the TCP layer; that is, which peer is the TCP-client and which is the TCP-server. For example, a network element is traditionally the TCP-server. However, when calling home, the network element becomes the TCP-client. The network element's secure transport layer roles (SSH-server, TLS-server) and its application layer roles (NETCONF-server, RESTCONF-server) both remain the same. Having consistency in both the secure transport layer (SSH, TLS) and application layer (NETCONF, RESTCONF) roles conveniently enables deployed network management infrastructure to support call home also. For instance, existing certificate chains and user authentication mechanisms are unaffected by call home.

Call home is generally useful for both the initial deployment and on-going management of networking elements. Here are some scenarios enabled by call home: The network element may proactively call home after being powered on for the first time in order to register itself with its management system. The network element may access the network in a way that dynamically assigns it an IP address, but does not register its assigned IP address to a mapping service (e.g., dynamic DNS). The network element may be deployed behind a firewall that implements network address translation (NAT) for all internal network IP addresses. The network element may be deployed behind a firewall that doesn't allow any management access to the internal network. The network element may be configured in "stealth mode" and thus doesn't have any open ports for the management system to connect to. The operator may prefer to have network elements initiate management connections, believing it is easier to secure one open port in the data center than to have an open port on each network element in the network.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 .

The techniques described in this document are suitable for network management scenarios such as the ones described in . However, these techniques are only defined for NETCONF Call Home and RESTCONF Call Home, as described in this document. The reason for this restriction is that different protocols have different security assumptions. The NETCONF and RESTCONF protocols require clients and servers to verify the identity of the other party. This requirement is specified for the NETCONF protocol in Section 2.2 of , and is specified for the RESTCONF protocol in Sections 2.4 and 2.5 of ). This contrasts with the base SSH and TLS protocols, which do not require programmatic verification of the other party (section 9.3.4 of , section 4 of , and section 7.3 of ). In such circumstances, allowing the SSH/TLS server to contact the SSH/TLS client would open new vulnerabilities. Any use of call home with SSH/TLS for purposes other than NETCONF or RESTCONF will need a thorough contextual risk assessment. A risk assessment for this RFC is in the Security Considerations section ().

This document uses the SSH Transport Layer Protocol with the exception that the statement "The client initiates the connection" made in Section 4 (Connection Setup) does not apply. Assuming the reference to client means "SSH client" and the reference to connection means "TCP connection", this statement doesn't hold true in call home, where the network element is the SSH server and yet still initiates the TCP connection. Security implications related to this change are discussed in Security Considerations ().

Throughout the remainder of this document, the term "NETCONF/RESTCONF" is used as an abbreviation in place of the text "the NETCONF or the RESTCONF". The NETCONF/RESTCONF abbreviation is not intended to require or to imply that a client or server must implement both the NETCONF standard and the RESTCONF standard.

The diagram below illustrates call home from a protocol layering perspective:

| | | | | | 2. SSH/TLS | |<-----------------------------------| | | | | | 3. NETCONF/RESTCONF | |<-----------------------------------| | | Note: arrows point from the "client" to the "server" at each protocol layer ]]>

This diagram makes the following points: The NETCONF/RESTCONF server begins by initiating a TCP connection to the NETCONF/RESTCONF client. Using this TCP connection, the NETCONF/RESTCONF client initiates a SSH/TLS session to the NETCONF/RESTCONF server. Using this SSH/TLS session, the NETCONF/RESTCONF client intiates a NETCONF/RESTCONF session to the NETCONF/RESTCONF server.

The term "client" is defined in , Section 1.1 "client". In the context of network management, the NETCONF/RESTCONF client might be a network management system.

The NETCONF/RESTCONF client listens for TCP connection requests from NETCONF/RESTCONF servers. The client MUST support accepting TCP connections on the IANA-assigned ports defined in , but MAY be configured to listen a defferent port. The NETCONF/RESTCONF client accepts an incoming TCP connection request and a TCP connection is established. Using this TCP connection, the NETCONF/RESTCONF client starts either the SSH-client or the TLS-client protocol. For example, assuming the use of the IANA-assigned ports, the SSH-client protocol is started when the connection is accepted on port PORT-X and the TLS-client protocol is started when the connection is accepted on either port PORT-Y or PORT-Z. If using TLS, the NETCONF/RESTCONF client MUST advertise "peer_allowed_to_send", as defined by . This is required so NETCONF/RESTCONF servers can depend on it being there for call home connections, when keep-alives are needed the most. As part of establishing an SSH or TLS connection, the NETCONF/RESTCONF client MUST validate the server's presented host key or certificate. This validation MAY be accomplished by certificate path validation or by comparing the host key or certificate to a previously trusted or "pinned" value. If a certificate is presented and it contains revocation checking information, the NETCONF/RESTCONF client SHOULD check the revocation status of the certificate. If it is determined that a certificate has been revoked, the client MUST immediately close the connection. If certificate path validation is used, the NETCONF/RESTCONF client MUST ensure that the presented certificate has a valid chain of trust to a preconfigured Issuer and that the presented certificate encodes an "identifier" that the client had awareness of prior to the connection attempt. Clients SHOULD ensure that the Issuer used to authenticate the presented certificate defines the namespace for the identifiers of interest. How identifiers are encoded in certificates MAY be determined by a policy associated with the certificate's Issuer. For instance, a given Issuer may be known to only sign IDevID certificates having a unique identifier (e.g., serial number) in the X.509 certificate's "CommonName" field. After the server's host key or certificate is validated, the SSH or TLS protocol proceeds as normal to establish a SSH or TLS connection. When performing client-authentcation with the NETCONF/RESTCONF server, the NETCONF/RESTCONF client MUST ensure to only use credentials that it had previously associated for the NETCONF/RESTCONF server’s presented host key or server certificate. Once the SSH or TLS connection is established, the NETCONF/RESTCONF client starts either the NETCONF-client or RESTCONF-client protocol. Assuming the use of the IANA-assigned ports, the NETCONF-client protocol is started when the connection is accepted on either port PORT-X or PORT-Y and the RESTCONF-client protocol is started when the connection is accepted on port PORT-Z.

How a NETCONF or RESTCONF client is configured is outside the scope of this document. For instance, such configuration might be used to enable listening for call home connections, configuring trust anchors, or configuring identifiers for expected connections.

The term "server" is defined in , Section 1.1 "server". In the context of network management, the NETCONF/RESTCONF server might be a network element or a device.

The NETCONF/RESTCONF server initiates a TCP connection request to the NETCONF/RESTCONF client. The server MUST support connecting to one of the IANA-assigned ports defined in , but MAY be configured to connect to a different port. Using the IANA-assigned ports, the server connects to port PORT-X for NETCONF over SSH, port PORT-Y for NETCONF over TLS, and port PORT-Z for RESTCONF over TLS. The TCP connection request is accepted and a TCP connection is established. Using this TCP connection, the NETCONF/RESTCONF server starts either the SSH-server or the TLS-server protocol, depending on how it is configured. For example, assuming the use of the IANA-assigned ports, the SSH-server protocol is used after connecting to the remote port PORT-X and the TLS-server protocol is used after connecting to one of the remote ports PORT-Y or PORT-Z. As part of establishing the SSH or TLS connection, the NETCONF/RESTCONF server will send its host key or certificate to the client. If a certificate is sent, the server MUST also send all intermediate certificates leading up to the certificate's trust anchor. How to send a list of certificates is defined for SSH in Section 2.1, and for TLS in Section 7.4.2. Establishing an SSH or TLS session requires server authentication of client credentials in all cases except with RESTCONF, where some client authentication schemes occur after the secure transport connection (TLS) has been established. If transport (SSH or TLS) level client authentication is required, and the client is unable to successfully authenticate itself to the server in an amount of time defined by local policy, the server MUST close the connection. Once the SSH or TLS connection is established, the NETCONF/RESTCONF server starts either the NETCONF-server or RESTCONF-server protocol, depending on how it is configured. Assuming the use of the IANA-assigned ports, the NETCONF-server protocol is used after connecting to remote port PORT-X or PORT-Y, and the RESTCONF-server protocol is used after connecting to remote port PORT-Z. If a persistent connection is desired, the NETCONF/RESTCONF server, as the connection initiator, SHOULD actively test the aliveness of the connection using a keep-alive mechanism. For TLS based connections, the NETCONF/RESTCONF server SHOULD send HeartbeatRequest messages, as defined by . For SSH based connections, per Section 4 of , the NETCONF/RESTCONF server SHOULD send a SSH_MSG_GLOBAL_REQUEST message with a purposely nonexistent "request name" value (e.g., keepalive@ietf.org) and the "want reply" value set to '1'.

How a NETCONF or RESTCONF server is configured is outside the scope of this document. This includes configuration that might be used to specify hostnames, IP addresses, ports, algorithms, or other relevant parameters. That said, a YANG model for configuring NETCONF and RESTCONF servers, including call home, is provided in .

The security considerations described in and , and by extension , , and apply here as well. This RFC deviates from standard SSH and TLS usage by having the SSH/TLS server initiate the underlying TCP connection. This reversal is incongruous with , which says "the client initiates the connection" and also , which says "the client MUST construct a list of acceptable reference identifiers, and MUST do so independently of the identifiers presented by the service." Risks associated with these variances are centered around server authentication and the inability for clients to compare an independently constructed reference idenitifier to one presented by the server. To mitigate these risks, this RFC requires that the NETCONF/RESTCONF client validate the server's SSH host key or certificate, by certificate path validation to a preconfigured Issuer certificate, or by comparing the host key or certificate to a previously trusted or "pinned" value. Furthermore, when a certificate is used, this RFC requires that the client be able to match an identifier encoded in the presented certificate with an identifier the client was preconfigured to expect (e.g., serial number). For cases when the NETCONF/RESTCONF server presents an X.509 certificate, NETCONF/RESTCONF clients should ensure that the Issuer certificate used for certificate path validation is unique to the manufacturer of the server. That is, the certificate should not belong to a 3rd-party certificate authority that might issue intermediate certificates for more than one manufacturer. This is especially important when a client-authentication mechanism passing a shared secret (e.g., a password) to the server is used. Not doing so could otherwise lead to a case where the client sends the shared secret to another server that happens to have the same identity (e.g., serial number) as the server the client was configured to expect. Considerations not associated with server authentication follow next. Internet facing hosts running NETCONF or RESTCONF call home will be fingerprinted via scanning tools such as `zmap` . Both SSH and TLS provide many ways in which a host can be fingerprinted. SSH and TLS servers are fairly mature and able to withstand attacks, but SSH and TLS clients may not be as robust. Implementers and deployments need to ensure that software update mechanisms are provided so that vulnerabilities can be fixed in a timely fashion. An attacker could launch a denial of service (DoS) attack on the NETCONF/RESTCONF client by having it perform computationally expensive operations, before deducing that the attacker doesn't posses a valid key. For instance, in TLS 1.3 , the ClientHello message contains a Key Share value based on an expensive asymmetric key operation. Common precautions mitigating DoS attacks are recommended, such as temporarily blacklisting the source address after a set number of unsuccessful login attempts. When using call home with the RESTCONF protocol, special care is required when using some HTTP authentication schemes, especially the Basic and Digest schemes, which convey a shared key. Implementations and deployments should be sure to review the Security Considerations section in the RFC for any HTTP client authentication scheme used.

This RFC requests that IANA assigns three TCP port numbers in the "Registered Port Numbers" range with the service names "netconf-ch-ssh", "netconf-ch-tls", and "restconf-ch-tls". These ports will be the default ports for NETCONF Call Home and RESTCONF Call Home protocols. Below is the registration template following the rules in .

Contact: IETF Chair Description: NETCONF Call Home (SSH) Reference: RFC XXXX Port Number: PORT-X Service Name: netconf-ch-tls Transport Protocol(s): TCP Assignee: IESG Contact: IETF Chair Description: NETCONF Call Home (TLS) Reference: RFC XXXX Port Number: PORT-Y Service Name: restconf-ch-tls Transport Protocol(s): TCP Assignee: IESG Contact: IETF Chair Description: RESTCONF Call Home (TLS) Reference: RFC XXXX Port Number: PORT-Z ]]>

The author would like to thank the following for lively discussions on list and in the halls (ordered by last name): Andy Bierman, Martin Bjorklund, Mehmet Ersue, Wes Hardaker, Stephen Hanna, David Harrington, Jeffrey Hutzelman, Radek Krejci, Alan Luchuk, Mouse, Russ Mundy, Tom Petch, Juergen Schoenwaelder, Peter Saint-Andre, Joe Touch, Hannes Tschofenig, Sean Turner, and Bert Wijnen.

&rfc2119; &rfc4251; &rfc4252; &rfc4253; &rfc4254; &rfc5246; &rfc6125; &rfc6187; &rfc6241; &rfc6242; &rfc6335; &rfc6520; &rfc7589; YumaWorks Tail-f Systems Juniper Networks &rfc6020; &rfc7616; &rfc7617; IEEE SA-Standards Board RTFM, Inc. Juniper Networks Jacobs University In proceedings of the 22nd USENIX Security Symposium

The term "TCP connection" is now used throughout. The terms "network element" and "management system" are now only used in the Motivation section. Restructured doc a little to create an Introduction section. Fixed reference in Applicability Statement so it would work equally well for SSH and TLS. Fixed reported odd wording and three references.

Added call home support for the RESTCONF protocol. Fixed paragraph 3 of Security Considerations to equally apply to the TLS protocol.

Tried to improve readability (issue #6) Removed "FIXME" in section 1.3 (issue #7) Added RFC Editor notes (issue #8) Removed "TCP session" term (issue #9) Improved language for usage of IANA-assigned ports (issue #10)

Replaced "verify credentials" with "verify identity" (issue #11)

Applied many suggestions from WGLC Removed essay like "Server Identification and Verification" section Added text about keep-alives Added Configuration Data Model section for client protocol Improved Security Considerations section

Addressed comments raised by Alan Luchuk.

replaced "reference identifier" with "identifier" added reference to RFC6125 moved reference to 6020 to Informative section

Added text regarding client authentication Now says client-initiated (not standard) NETCONF/RESTCONF connections Now says server must send all (not any) intermediate certificates Improved wording based on suggestions from Jonathan and Tom

Added dynamic DNS as an example for an IP mapping service Replaced draft-ietf-netconf-rfc5539bis with RFC7589 Recharacterized this draft's relationship to RFC4253

Updates from AD review

Fixed typo introduced in -10

Addresses DISCUSS and COMMENTS from IESG review.

Fixed a couple typos and removed the VeriSign example

Addressed final DISCUSS item from IESG review

All issues with this draft are tracked using GitHub issues. Please see: https://github.com/netconf-wg/call-home/issues to see currently opened issues.