Interface to Network Security Functions A. Pastor
Internet-Draft D. Lopez
Intended status: Experimental Telefonica I+D
Expires: August 15, 2019 A. Shaw
ARM
February 11, 2019

Remote Attestation Procedures for Network Security Functions (NSFs) through the I2NSF Security Controller
draft-pastor-i2nsf-nsf-remote-attestation-07

Abstract

This document describes the procedures a client can follow to assess the trust on an external NSF platform and its client-defined configuration through the I2NSF Security Controller. The procedure to assess trustworthiness is based on a remote attestation of the platform and the NSFs running on it performed through a Trusted Platform Module (TPM) invoked by the Security Controller.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on August 15, 2019.

Copyright Notice

Copyright (c) 2019 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 Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.


Table of Contents

1. Introduction

As described in [RFC8192], the use of externally provided NSF implies several additional concerns in security. The most relevant threats associated with a externalized platform are detailed in [RFC8329]. As stated there, mutual authentication between the user and the NSF environment and, more importantly, the attestation of the components in this environment by clients, could address these threats and provide an acceptable level of risk. In particular:

The client can have a proof that their NSFs and policies are correctly (from the client point of view) enforced by the Security Controller. Taking into account the threats identified in [RFC8329], this document first identifies the user expectations regarding remote trust establishment, briefly analyzes Trusted Computing techniques, and finally describes the proposed mechanisms for remote establishment of trust through the Security Controller.

2. Requirements Language

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 [RFC2119].

In this document, these words will appear with that interpretation only when in ALL CAPS. Lower case uses of these words are not to be interpreted as carrying RFC-2119 significance.

3. Establishing Client Trust

From a high-level standpoint, in any I2NSF platform, the client connects and authenticates to the Security Controller, which then initializes the procedures for authentication and authorization (and likely accounting and auditing) to track the loading and unloading of the client's NSFs, addressing the verification of the whole software stack: firmware, (host and guest) OSes, NSFs themselves and, in a virtualized environment, the virtualization system (hypervisors, container frameworks...). Afterwards, user traffic from the client domain goes through the NSF platform that hosts the corresponding NSFs. The user's expectations of the platform behavior are thus twofold:

We will refer to the attestation of these two expectations as the "client-agnostic attestation" and the "client-specific attestation". Trusted Computing techniques play a key role in addressing these expectations.

3.1. First Step: Client-Agnostic Attestation

This is the first interaction between a client and a Security Controller: the client wants to attest that he is connected to a genuine Security Controller before continuing with the authentication. In this context, two properties characterize the genuineness of the Security Controller:

  1. That the identity of the Security Controller is correct
  2. That it will process the client credentials and set up the client NSFs and policies properly.

Once these two properties are proven to the client, the client knows that their credentials will only be used by the Security Controller to set up the execution of their NSFs.

3.2. Second Step: Client-Specific Attestation

From the security enforcement point of view, the client agnostic attestation focuses on the initialization of the execution platform for the NSFs. This second step aims to prove to clients that their security is enforced accordingly with their choices (i.e. NSFs and policies). The attestation can be performed at the initialization of the NSFs, before any user traffic is processed by the NSFs, and optionally during the execution of the NSFs.

Support of static attestation, performed at initialization time, for the execution platform and the NSFs is REQUIRED for a Security Controller managing NSFs, and MUST be performed before any user traffic is redirected through any set of NSFs. The Security Controller MUST provide proof to the client that the instantiated NSFs and policies are the ones chosen.

In addition to the platform and executable component attestation, the infrastructure network topology supporting the NSFs may need to attested, in order to assess the enforcement of the security policies requested by the client. Whilst platform and NSF attestation can be considered sufficient in I2NSF environments in which network elements are connected following a fairly static configuration, the dynamicity brought by networking techniques such as NFV, SDN and SFC make attestation of dynamic topology network topologies a desirable feature in a number of cases. Depending on the level of asurance desired, the client MAY request the Security Controller proof of the network topology connecting the instantiated NSFs.

Additionally to the NSFs instantiation attestation, a continuous attestation of the Security Controller and the NSF execution MAY be required by a client to ensure their security. The sampling periods for the continuous attestation of NSFs an Controller MAY be different.

3.3. Trusted Computing

In a nutshell, Trusted Computing (TC) aims at answering the following question: "As a user or administrator, how can I have some assurance that a computing system is behaving as it should?". The major enterprise level TC initiative is the Trusted Computing Group [TCG], which has been established for more than a decade, that primarily focuses on developing TC for commodity computers (servers, desktops, laptops, etc.).

The overall scheme proposed by TCG for using Trusted Computing is based on a step-by-step extension of trust, called a Chain of Trust. It uses a transitive mechanism: if a user can trust the first execution step and each step correctly attests the next executable software for trustworthiness, then a user can trust the system.

             +-----------+                                    
             |           |    extends PCR                     
             | Platform  +------------------------+           
             |           |                        |           
             +-----^-----+                        |           
                   |                              |           
                   |measures                      |           
             +-----------+                        |           
             | Security  |    extends PCR         |           
             |           +---------------------+  |           
             | Controller|                     |  |           
             +-----^-----+                     |  |           
                   |                           |  |           
                   |measures                 +-v--v----------+
             +-----------+                   |               |
             |           |    extends PCR    |               |
             | Bootloader+-------------------> Root of Trust |
             |           |                   |               |
             +-----^-----+                   |               |
                   |                         +-^--^----------+
                   |measures                   |  |           
             +-----------+                     |  |           
             |           |    extends PCR      |  |           
             | BIOS      +---------------------+  |           
             |           |                        |           
             +-----^-----+                        |           
                   |                              |           
                   |measures                      |           
             +-----------+                        |           
             | Bootblock |    extends PCR         |           
             |  (CRTM)   +------------------------+           
             |           |                                    
             +-----------+                                    
			
			

Figure 1: Applying Trusted Computing

Effectively, during the loading of each piece of software, the integrity of each piece of software is measured and stored inside a log that reflects the different boot stages, as illustrated in the figure above. Later, at the request of a user, the platform can present this log (signed with the unique identity of the platform), which can be checked to prove the platform identity and attest the state of the system. The base element for the extension of the Chain of Trust is called the Core Root of Trust.

The TCG has created a standard for the design and usage of a secure crypto-processor to address the storage of keys, general secrets, identities, and platform integrity measurements: the Trusted Platform Module (TPM). When using a TPM as a root of trust, measurements of the software stack are stored in special on-board Platform Configuration Registers (PCRs) on a discrete TPM. There are normally a small number of PCRs that can be used for storing measurements; however, it is not possible to directly write to a PCR. Instead, measurements must be stored using a process called Extending PCRs.

The extend operation can update a PCR by producing a global hash of the concatenated values of the previous PCR value with the new measurement value. The Extend operation allows for an unlimited number of measurements to be captured in a single PCR, since the size of the value is always the same and it retains a verifiable ordered chain of all the previous measurements.

Attestation of the virtualization platform will thus rely on a process of measuring the booted software and storing a chained log of measurements, typically referred to as Trusted Boot. The user will either validate the signed set of measurements with a trusted third party verifier who will assess whether the software configuration is trusted, or the user can check for themselves against their own set of reference digest values (measurements) that they have obtained a priori, and having already known the public endorsement key of the remote Root of Trust.

Trusted Boot should not be confused with a different mechanism known as "Secure Boot", as they both are designed to solve different problems. Secure Boot is a mechanism for a platform owner to lock a platform to only execute particular software. Software components that do not match the configuration digests will not be loaded or executed. This mechanism is particularly useful in preventing malicious software that attempts to install itself in the boot record (a bootkit) from successfully infecting a platform on reboot. A common standard for implementing Secure Boot is described in [UEFI]. Secure Boot only enforces a particular configuration of software, it does not allow a user to attest or quote for a series of measurements.

3.4. Topology Attestation

There are two methods able to attest the deployment of a topology addressing client requirements on a dynamically controled network infrastructure. The first one asumes the newtork infrastructure is built by means of SDN-enabled fowarding elements, and the second relies on the application of SFC [RFC7665] to build the NSF processing paths. In both cases, a network topology verifier is used.

In the first case, a SDN verifier is introduced, and network forwarding elements required to provide attestation features, as described in the previous section, to provide measures on the enforced SDN configuration. The SDN verifier retrieves from the SDN controller the configuration for the attested network elements, challenges them for their SDN configuration, and assessesit is consistent with the expected SDN configuration retrieved from the SDN controller. The SDN verifier on the network elements leverage a TPM, with the network element implementing a regular measured boot.

The second option considers the application of Proof of Transit (POT) [I-D.ietf-sfc-proof-of-transit] to a SFC-based network, where the NSFs act as service functions. A SFC verifier can inject specific packets requesting POT, and verifying it at the egress of the service path to assess a correct topoloogy is being enforced, by means of the cryptographic proof provided by POT.

4. NSF Attestation Principles

Following the general requirements described in [RFC8329] the Security Controller will become the essential element to implement the measurements described above, relying on a TPM for the Root of Trust.

A mutual authentication of clients and the Security Controller MUST be performed, establishing the desired level of assurance. This level of assurance will determine how stringent are the requirements for authentication (in both directions), and how detailed the collected measurements and their verification will be. Furthermore, the NSF platform MUST run a TPM, able to collect measurements of the platform itself, the Security Controller, and the NSFs being executed. The availability of a network topology verifier is OPTIONAL, though a client MAY require it to fulfill the required level of assurance. The Security Controller MUST make the attestation measurements available to the client, directly or by means of a Trusted Third Party.

As described in [RFC8329], a trusted connection between the client and the Security Controller MUST be established and all traffic to and from the NSF environment MUST flow through this connection

NOTE: The reference to results from WGs such as NEA and SACM is currently under consideration and will be included here.

4.1. Requirements for a Trusted NSF Platform

Although a discrete hardware TPM is RECOMMENDED, relaxed alternatives (such as embedded CPU TPMs, or memory and execution isolation mechanisms) MAY also be applied when the required level of assurance is lower. This reduced level of assurance MUST be communicated to the client by the Security Controller during the initial mutual authentication phase. The Security Controller MUST use a set of capabilities to negotiate the level of assurance with the client.

4.1.1. Trusted Boot

NOTE: This section is derived from the original version of the document, focused on virtual NSFs. Although it seems to be applicable to any modern physical appliance, we must be sure all these considerations are 100% applicable to physical NSFs as well, and provide exceptions when that is not the case. Support from an expert in physical node attestation is required here.

All clients who interact with a Security Controller MUST be able to:

  1. Identify the Security Controller based on the public key of a Root of Trust.
  2. Retrieve a set of measurements of all the base software the Security Controller has booted (i.e. the NSF platform).

This requires that firmware and software MUST be measured before loading, with the resulting value being used to extend the appropriate PCR register. The general usage of PCRs by each software component SHOULD conform to open standards, in order to make verifying attestation reports interoperable, as it is the case of TCG Generic Server Specification [TCGGSS].

The following list describes which PCR registers SHOULD be used during a Trusted Boot process:

A signed audit log of boot measurements should also be provided. The PCR values can also be used as an identity for dynamically decrypting encrypted blobs on the platform (such as encryption keys or configurations that belong to operating system components). Software can choose to submit pieces of data to be encrypted by the Root of Trust (which has its own private asymmetric key and PCR registers) and only have it decrypted based on some criteria. These criteria can be that the platform booted into a particular state (e.g. a set of PCR values). Once the desired criteria are described and the sensitive data is encrypted by the root of trust, the data has been sealed to that platform state. The sealed data will only be decrypted when the platform measurements held in the root of trust match the particular state.

Trusted Boot requires the use of a root of trust for safely storing measurements and secrets. Since the Root of Trust is self-contained and isolated from all the software that is measured, it is able to produce a signed set of platform measurements to a local or remote user. However, Trusted Boot does not provide enforcement of a configuration, since the root of trust is a passive component not in the execution path, and is solely used for safe independent storage of secrets and platform measurements. It will respond to attestation requests with the exact measurements that were made during the software boot process. Sealing and unsealing of sensitive data is also a strong advantage of Trusted Boot, since it prevents leakage of secrets in the event of an untrusted software configuration.

4.1.2. Remote Attestation Service

A service MUST be present for providing signed attestation report (e.g. the measurements) from the Root of Trust (RoT) to the client. In case of failure to communicate with the service, the client MUST assume the service cannot be trusted and seek an alternative Security Controller.

Since some forms of RoT require serialised access (i.e. due to slow access to hardware), latency of getting an attestation report could increase with simultaneous requests. Simultaneous requests could occur if multiple Trusted Third Parties (TTP) request attestation reports at the same time. This MAY be improved through batching of requests, in a special manner. In a typical remote attestation protocol, the client sends a random number ("nonce") to the RoT in order to detect any replay attacks. Therefore, caching of an attestation report does not work, since there is the possibility that it may not be a fresh report. The solution is to batch the nonce for each requestor until the RoT is ready for creating the attestation report. The report will be signed by the embedded identity of the RoT to provide data integrity and authenticity, and the report will include all the nonces of the requestors. Regardless of the number of the number of nonces included, the requestor verifying the attestation report MUST check to see if the requestor's nonce was included in order to detect replay attacks. In addition to the attestation report containing PCRs, an additional report known as an SML (Secure Measurement Log) can be returned to the requestor to provide more information on how to verify the report (e.g. how to reproduce the PCR values). The integrity of the SML is protected by a PCR measurement in the RoT. An example of an open standard for responses is [TCGIRSS]. Further details are discussed in Section 5.2.

As part of initial contact, the Security Controller MAY present a list of external TTPs that the client can use to verify it. However, the client MUST assess whether these external verifiers can be trusted. The client can also choose to ignore or discard the presented verifiers.

If available, the network topology verifier MUST be colocated or integrated with the RoT.

Finally, to prevent malicious relaying of attestation reports from a different host, the authentication material of the secure channel (e.g. TLS, IPSec, etc.) SHOULD be bound to the RoT and verified by the connected client, unless the lowest levels of assurance have been chosen and an explicit warning issued. This is also addressed in Section 5.1.

4.1.3. Secure Boot

Using a mechanism such as Secure Boot helps provide strong prevention of software attacks. Furthermore, in combination with a hardware-based TPM, Secure Boot can provide some resilience to physical attacks (e.g. preventing a class of offline attacks and unauthorized system replacement). For NSF providers, it is RECOMMENDED that Secure Boot is employed wherever possible with an appropriate firmware update mechanism, due to the possible threat of software/firmware modifications in either public places or privately with inside attackers.

5. Remote Attestation Procedures

The establishment of trust with the Security Controller and the NSF platform consists of three main phases, which need to be coordinated by the client:

  1. Trusted channel with the Security Controller. During this phase, the client securely connects to the Security Controller to avoid that any data can be tampered with or modified by an attacker if the network cannot be considered trusted. The establishment of the trusted channel is completed after the next step.
  2. Security Controller attestation. During this phase, the client verifies that the Security Controller components responsible for handling the credentials and for the isolation with respect to other potential clients are behaving correctly. Furthermore, it is verified that the identity of the platform attested is the same of the one presented by the Security Controller during the establishment of the secure connection.
  3. Platform attestation. During this step, which can be repeated periodically until the connection is terminated, the Security Controller verifies the integrity of the elements composing the NSF platform. The components responsible for this task have been already attested during the previous phase.

                                +----------+                   
              3. Attestation    | Trusted  |   3. Attestation 
           +-------------------->  Third   <----------+        
           |                    |  Party   |          |        
           |                    +----------+ +--------+-------+
+----------v-------+                         |  +-----v-----+ |
|      Client      |                         |  | Security  | |
|                  |  1. Trusted channel     |  | Controller| |
| 2. Get Cert      +------+ handshake +--------->           | |
| 3. Attestation   |                         |  +-----------+ |
| 4. Cont.handshake|                         |                |
|                  |                         |                |
|                  |                         |  +---------+   |
|                  |                         |  |   NSF   |   |
|                  |                         |  +---------+   |
+------------------+                         +----------------+

			
		

Figure 2: Steps for remote attestation

In the following each step, as depicted in the above figure, is discussed in more detail.

5.1. Trusted Channel with the Security Controller

A trusted channel is an enhanced version of the secured channel that. It adds the requirement of integrity verification of the contacted endpoint by the other peer during the initial handshake to the functionality of the secured channel. However, simply transmitting the integrity measurements over the channel does not guarantee that the platform verified is the channel endpoint. The public key or the certificate for the secure communication MUST be included as part of the measurements presented by the contacted endpoint during the remote attestation. This way, a malicious platform cannot relay the attestation to another platform as its certificate will not be present in the measurements list of the genuine platform.

In addition, the problem of a potential loss of control of the private key must be addressed (a malicious endpoint could prove the identity of the genuine endpoint). This is done by defining a long-lived Platform Property Certificate. Since this certificate connects the platform identity to the AIK public key, an attacker cannot use a stolen private key without revealing his identity, as it may use the certificate of the genuine endpoint but cannot create a quote with the AIK of the other platform.

Finally, since the platform identity can be verified from the Platform Property Certificate, the information in the certificate to be presented during the establishment of a secure communication is redundant. This allows for the use of self-signed certificates. This would simplify operational procedures in many environments, especially when they are multi-tenant. Thus, in place of certificates signed by trusted CAs, the use of self-signed certificates (which still need to be included in the measurements list) is RECOMMENDED.

The steps required for the establishment of a trusted channel with the Security Controller are as follows:

  1. The client begins the trusted channel handshake with the selected Security Controller.
  2. The certificate of the Security Controller is collected and used for verifying the binding of the attestation result to the contacted endpoint.
  3. The client performs the remote attestation protocol with the Security Controller, either directly or with the help of a Trusted Third Party. The Trusted Third Party MAY perform the verification of attestation quotes on behalf of multiple clients.
  4. If the result of the attestation is positive, the application continues the handshake and establishes the trusted channel. Otherwise, it closes the connection.

5.2. Security Controller Attestation

During the establishment of the trusted channel, the client attests the Security Controller by verifying the identity of the contacted endpoint and its integrity. Initially the Security Controller measures all of the hardware and software components involved in the boot process of the NSF platform, in order to build the chain of trust.

Since a client may not have enough functionality to perform the integrity verification of a Security Controller, the client MAY request the status of a Security Controller to be computed by a Trusted Third Party (TTP). This choice has the additional advantage of preventing an attacker from easily determining the software running at the Security Controller.

If the client directly performs the remote attestation, it executes the following steps:

  1. Ask the Security Controller to generate an integrity report with the format defined in [TCGIRSS].
  2. The Security Controller retrieves the measurements and asks the TPM to sign the PCRs with an Attestation Identity Key (AIK). This signature provides the client with the evidence that the measurements received belong to the Security Controller being attested.
  3. Once the integrity report has been generated it is sent back to the client.
  4. The client first checks if the integrity report is valid by verifying the quote and the certificate associated to the AIK, and then determines if the Security Controller is behaving as expected (i.e. its software has not been compromised and isolation among the clients connected to it is enforced). As part of the verification, the client also checks that the digest of the certificate, received during the trusted channel handshake, is present among measurements.

If the client has limited computation resources, it may contact a TTP which, in turn, attests the Security Controller and returns the result of the integrity evaluation to the client, following the same steps depicted above.

5.3. Platform Attestation

The main outcome of the Security Controller attestation is to detect whether or not it is correctly configuring the operational environment for NSFs to be managed by the connecting client (the NSF platform, or just platform) in a way that any user traffic is processed only by these NSFs that are part of the platform. Platform attestation, instead, evaluates the integrity of the NSFs running on the platform.

Platform attestation does not imply a validation of the mechanisms the Security Controller can apply to select the appropriate NSFs to enforce the Service Policies applicable to specific flows. The selection of these NSFs is supposed to happen independent of the attestation procedures, and trust on the selection process and the translation of policies into function capabilities has to be based on the trust clients have on the Security Controller being attested as the one that was intended to be used. An attestation of the selection and policy mapping procedures constitute an interesting research problem, but it is out of the scope of this document.

The procedures are essentially similar to the ones described in the previous section. This step MAY be applied periodically if the level of assurance selected by the user requires it.

Attesting NSFs, especially if they are running as virtual machines, can become a costly operation, especially if periodic monitoring is required by the requested level of assurance. There are several proposals to make this feasible, from the proposal of virtual TPMs in [VTPM] to the application of Virtual Machine Introspection through an integrity monitor described by [VMIA].

6. Security Considerations

This document is specifically oriented to security and it is considered along the whole text.

7. IANA Considerations

This document requires no IANA actions.

8. Acknowledgments

This work has been partially supported by the European Commission under Horizon 2020 grant agreement no. 700199 "Securing against intruders and other threats through a NFV-enabled environment (SHIELD)". This support does not imply endorsement.

9. References

9.1. Normative References

[I-D.ietf-sfc-proof-of-transit] Brockners, F., Bhandari, S., Dara, S., Pignataro, C., Leddy, J., Youell, S., Mozes, D., Mizrahi, T., Aguado, A. and D. Lopez, "Proof of Transit", Internet-Draft draft-ietf-sfc-proof-of-transit-01, October 2018.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997.
[RFC7665] Halpern, J. and C. Pignataro, "Service Function Chaining (SFC) Architecture", RFC 7665, DOI 10.17487/RFC7665, October 2015.
[RFC8192] Hares, S., Lopez, D., Zarny, M., Jacquenet, C., Kumar, R. and J. Jeong, "Interface to Network Security Functions (I2NSF): Problem Statement and Use Cases", RFC 8192, DOI 10.17487/RFC8192, July 2017.
[RFC8329] Lopez, D., Lopez, E., Dunbar, L., Strassner, J. and R. Kumar, "Framework for Interface to Network Security Functions", RFC 8329, DOI 10.17487/RFC8329, February 2018.
[TCG] "Trusted Computing Group (TCG)"
[TCGGSS] "TCG Generic Server Specification, Version 1.0"
[TCGIRSS] "Infrastructure Work Group Integrity Report Schema Specification, Version 1.0"

9.2. Informative References

[UEFI] "UEFI Specification Version 2.2 (Errata D), Tech. Rep."
[VMIA] Schiffman, J., Vijayakumar, H. and T. Jaeger, "Verifying System Integrity by Proxy"
[VTPM] "vTPM:Virtualizing the Trusted Platform Module"

Authors' Addresses

Antonio Pastor Telefonica I+D Zurbaran, 12 Madrid, 28010 Spain Phone: +34 913 128 778 EMail: antonio.pastorperales@telefonica.com
Diego R. Lopez Telefonica I+D Editor Jose Manuel Lara, 9 (1-B) Seville, 41013 Spain Phone: +34 913 129 041 EMail: diego.r.lopez@telefonica.com
Adrian L. Shaw ARM EMail: als@arm.com