| ANIMA WG | M. Pritikin |
| Internet-Draft | Cisco |
| Intended status: Informational | M. Richardson |
| Expires: September 18, 2016 | SSW |
| M. Behringer | |
| S. Bjarnason | |
| Cisco | |
| March 17, 2016 |
Bootstrapping Key Infrastructures
draft-ietf-anima-bootstrapping-keyinfra-02
This document specifies automated bootstrapping of a key infrastructure (BSKI) using vendor installed IEEE 802.1AR manufacturing installed certificates, in combination with a vendor based service on the Internet. Before being authenticated, a new device has only link-local connectivity, and does not require a routable address. When a vendor provides an Internet based service, devices can be forced to join only specific domains but in limited/disconnected networks or legacy environments we describe a variety of options that allow bootstrapping to proceed.
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 http://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 September 18, 2016.
Copyright (c) 2016 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 (http://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.
To literally "pull yourself up by the bootstraps" is an impossible action. Similarly the secure establishment of a key infrastructure without external help is also an impossibility. Today it is accepted that the initial connections between nodes are insecure, until key distribution is complete, or that domain-specific keying material is pre-provisioned on each new device in a costly and non-scalable manner. This document describes a zero-touch approach to bootstrapping an entity by securing the initial distribution of key material using third-party generic keying material, such as a manufacturer installed IEEE 802.1AR certificate [IDevID], and a corresponding third-party service on the Internet.
The two sides of an association being bootstrapped authenticate each other and then determine appropriate authorization. This process is described as four distinct steps between the existing domain and the new entity being added:
A precise answer to these questions can not be obtained without leveraging some established key infrastructure(s). A complexity that this protocol deals with are dealing with devices from a variety of vendors, and a network infrastructure (the domain) that is operated by parties that do not have any priviledged relationship with the device vendors. The domain's decisions are based on the new entity's authenticated identity, as established by verification of previously installed credentials such as a manufacturer installed IEEE 802.1AR certificate, and verified back-end information such as a configured list of purchased devices or communication with a (unidirectionally) trusted third-party. The new entity's decisions are made according to verified communication with a trusted third-party or in a strictly auditable fashion.
Optimal security is achieved with IEEE 802.1AR certificates on each new entity, accompanied by a third-party Internet based service for verification. Bootstrapping concepts run to completion with less requirements, but are then less secure. A domain can choose to accept lower levels of security when a trusted third-party is not available so that bootstrapping proceeds even at the risk of reduced security. Only the domain can make these decisions based on administrative input and known behavior of the new entity.
The result of bootstrapping is that a domain specific key infrastructure is deployed. Since IEEE 802.1AR PKI certificates are used for identifying the new entity, and the public key of the domain identity is leveraged during communications with an Internet based service, which is itself authenticated using HTTPS, bootstrapping of a domain specific Public Key Infrastructure (PKI) is described. Sufficient agility to support bootstrapping alternative key infrastructures (such as symmetric key solutions) is considered although no such alternate key infrastructure is described.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
The following terms are defined for clarity:
Questions have been posed as to whether this solution is suitable in general for Internet of Things (IoT) networks. In general the answer is no, but the terminology of [RFC7228] is best used to describe the boundaries.
The entire solution described in this document is aimed in general at non-constrained (i.e. class 2+) devices operating on a non-Challenged network. The entire solution described here is not intended to be useable as-is by constrained devices operating on challenged networks (such as 802.15.4 LLNs).
In many target applications, the systems involved are large router platforms with multi-gigabit inter-connections, mounted in controlled access data centers. But this solution is not exclusive to the large, it is intended to scale to thousands of devices located in hostile environments, such as ISP provided CPE devices which are drop-shipped to the end user. The situation where an order is fulfilled from distributed warehouse from a common stock and shipped directly to the target location at the request of the domain owner is explicitly supported. That stock ("SKU") could be provided to a number of potential domain owners, and the eventual domain owner will not know a-priori which device will go to which location.
Specifically, there are protocol aspects described here which might result in congestion collapse or energy-exhaustion of intermediate battery powered routers in an LLN. Those types of networks SHOULD NOT use this solution. These limitations are predominately related to the large credential and key sizes required for device authentication. Defining symmetric key techniques that meet the operational requirements is out-of-scope but the underlying protocol operations (TLS handshake and signing structures) have sufficient algorithm agility to support such techniques when defined.
The imprint protocol described here could, however, be used by non-energy constrained devices joining a non-constrained network (for instance, smart light bulbs are usually mains powered, and speak 802.11). It could also be used by non-constrained devices across a non-energy constrained, but challenged network (such as 802.15.4).
Some aspects are in scope for constrained devices on challenged networks: the certificate contents, and the process by which the four questions above are resolved is in scope. It is simply the actual on-the-wire imprint protocol which is likely inappropriate.
The imprint protocol results in a secure relationship between the domain registrar and the new device. If the new device is sufficiently constrained that the ACE protocol should be leveraged for operation, (see [I-D.ietf-ace-actors]), and the domain registrar is also the Client Authorization Server or the Authorization Server, then it may be appropriate to use this secure channel to exchange ACE tokens.
The logical elements of the bootstrapping framework are described in this section. Figure 1 provides a simplified overview of the components. Each component is logical and may be combined with other components as necessary.
.
.+------------------------+
+--------------Drop Ship-------------->.| Vendor Service |
| .+------------------------+
| .| M anufacturer| |
| .| A uthorized |Ownership|
| .| S igning |Tracker |
| .| A uthority | |
| .+--------------+---------+
| .............. ^
V |
+-------+ ............................................|...
| | . | .
| | . +------------+ +-----------+ | .
| | . | | | | | .
| | . | | | <-------+ .
| | . | Proxy | | Registrar | .
| <--------> <-------> | .
| New | . | | | | .
| Entity| . +------------+ +-----+-----+ .
| | . | .
| | . +-----------------+----------+ .
| | . | Domain Certification | .
| | . | Authority | .
+-------+ . | Management and etc | .
. +----------------------------+ .
. .
................................................
"Domain" components
Figure 1
We assume a multi-vendor network. In such an environment there could be a MASA or Ownership Tracker for each vendor that supports devices following this document's specification, or an integrator could provide a MASA service for all devices. It is unlikely that an integrator could provide Ownership Tracking services for multiple vendors.
This document describes a secure zero-touch approach to bootstrapping a key infrastructure; if certain devices in a network do not support this approach, they can still be bootstrapped manually. Although manual deployment is not scalable and is not a focus of this document the necessary mechanisms are called out in this document to ensure such edge conditions are covered by the architectural and protocol models.
Entities behave in an autonomic fashion. They discover each other and autonomically bootstrap into a key infrastructure delineating the autonomic domain. See [I-D.irtf-nmrg-autonomic-network-definitions] for more information.
This section details the state machine and operational flow for each of the main three entities. The New Entity, the Domain (primarily the Registrar) and the MASA service.
A representative flow is shown in Figure 2:
+--------+ +-------+ +------------+ +------------+
| New | | Proxy | | Domain | | Vendor |
| Entity | | | | Registrar | | Service |
| | | | | | | (Internet |
+--------+ +-------+ +------------+ +------------+
| | | |
|<-RFC3927 IPv4 adr | | |
or|<-RFC4862 IPv6 adr | | |
| | | |
|-------------------->| | |
| optional: mDNS query| | |
| RFC6763/RFC6762 | | |
| | | |
|<--------------------| | |
| mDNS broadcast | | |
| response or periodic| | |
| | | |
|<------------------->|<----------------->| |
| (d)TLS via the Proxy | |
|<--Registrar TLS server authentication---| |
[PROVISIONAL accept of server cert] | |
P---IEEE 802.1AR client authentication--->| |
P | | |
P---Request Audit Token (include nonce)-->| |
P | | |
P | /---> | |
P | | [accept device?] |
P | | [contact Vendor] |
P | | |--New Entity ID---->|
P | | |--Domain ID-------->|
P | | |--optional:nonce--->|
P | | | [extract DomainID]
P | | | |
P | optional: | [update audit log]
P | |can | |
P | |occur | optional: is |
P | |in | an ownership |
P | |advance | voucher available?
P | | | |
P | | |<-device audit log--|
P | | |<-audit token-------|
P | | | |
P | | |<-optional: --------|
P | \----> | ownership voucher |
P | | |
P | [verify audit log or voucher] |
P | | |
P<--Audit token and/or ownership voucher--| |
[verify response ]| | |
[verify provisional cert ]| | |
| | | |
|---------------------------------------->| |
| Continue with RFC7030 enrollment | |
| using now bidirectionally authenticated | |
| TLS session. | | |
| | | |
| | | |
| | | |
Figure 2
A New Entity that has not yet been bootstrapped attempts to find a local domain and join it. A New Entity MUST NOT automatically initiate bootstrapping if it has already been configured.
States of a New Entity are as follows:
+--------------+
| Start |
| |
+------+-------+
|
+------v-------+
| Discover |
+------------> |
| +------+-------+
| |
| +------v-------+
| | Identity |
^------------+ |
| rejected +------+-------+
| |
| +------v-------+
| | Request |
| | Join |
| +------+-------+
| |
| +------v-------+
| | Imprint | Optional
^------------+ <--+Manual input
| Bad Vendor +------+-------+
| response |
| +------v-------+
| | Enroll |
^------------+ |
| Enroll +------+-------+
| Failure |
| +------v-------+
| | Being |
^------------+ Managed |
Factory +--------------+
reset
Figure 3
State descriptions for the New Entity are as follows:
The following sections describe each of these steps in more detail.
The result of discovery is logically communication with a Proxy instead of a Domain Registrar but in such a case the proxy facilitates communication with the actual Domain Registrar in a manner that is transparent to the New Entity. Therefore or clarity a Proxy is always assumed.
To discover the Domain Bootstrap Server the New Entity performs the following actions in this order:
Section 5. The current DNS services returned during each query is maintained until bootstrapping is completed. If bootstrapping fails and the New Entity returns to the Discovery state it picks up where it left off and continues attempting bootstrapping. For example if the first Multicast DNS _bootstrapks._tcp.local response doesn't work then the second and third responses are tried. If these fail the New Entity moves on to normal DNS-based Service Discovery.
Once a Registrar is discovered (technically a communication channel through a Proxy) the New Entity communicates with the Registrar using the bootstrapping protocol defined in
Once all discovered services are attempted the device SHOULD return to Multicast DNS and keep trying. The New Entity may prioritize selection order as appropriate for the anticipated environment.
[[EDNOTE: An appropriate backoff or rate limiting strategy should be defined here such that the device doesn't flood the local network with queries. If the device were to eventually give up -- or at least have too long between attempts -- a power cycle would restart the backoff mechanism.]]
The New Entity identifies itself during the communication protocol handshake. If the client identity is rejected the New Entity repeats the Discovery process using the next proxy or discovery method available.
The bootstrapping protocol server is not authenticated. Thus this connection is provisional and all data received is untrusted until sufficiently validated even though it is over a (D)TLS connection. This is aligned with the existing provisional mode of EST [RFC7030] during s4.1.1 "Bootstrap Distribution of CA Certificates".
All security associations established are between the new device and the Bootstrapping server regardless of proxy operations.
The New Entity POSTs a request to join the domain to the Bootstrapping server. This request contains a New Entity generated nonce and informs the Bootstrapping server which imprint methods the New Entity will accept.
As indicated in EST [RFC7030] the bootstrapping server MAY redirect the client to an alternate server. This is most useful in the case where the New Entity has resorted to a well known vendor URI and is communicating with the vendor's Registrar directly. In this case the New Entity has authenticated the Registrar using the local Implicit Trust Anchor database and can therefore treat the redirect URI as a trusted URI which can also be validated using the Implicit Trust Anchor database. Since client authentication occurs during the TLS handshake the bootstrapping server has sufficient information to apply appropriate policy concerning which server to redirect to.
The nonce ensures the New Entity can verify that responses are specific to this bootstrapping attempt. This minimizes the use of global time and provides a substantial benefit for devices without a valid clock.
The domain trust anchor is received by the New Entity during the bootstrapping protocol methods in the form of either an Audit Token containing the domainID or an explicit ownership voucher. The goal of the imprint state is to securely obtain a copy of this trust anchor without involving human interaction.
The enrollment protocol EST [RFC7030] details a set of non-autonomic bootstrapping methods such as:
This document describes additional autonomic methods:
Since client authentication occurs during the TLS handshake the bootstrapping server has sufficient information to apply appropriate policy concerning which method to use.
An arbitrary basic configuration information package that is signed by the domain can be delivered alongside the Audit Token or ownership validation. This information is signed by the domain private keys and is a one time delivery containing information such as which enrollment server to communicate with and which management system to communicate with. It is intended as a limited basic configuration for these purposes and is not intended to deliver entire final configuration to the device.
If the autonomic methods fail the New Entity returns to discovery state and attempts bootstrapping with the next available discovered Registrar.
As the final step of bootstrapping a Registrar helps to issue a domain specific credential to the New Entity. For simplicity in this document, a Registrar primarily facilitates issuing a credential by acting as an RFC5280 Registration Authority for the Domain Certification Authority.
Enrollment proceeds as described in Enrollment over Secure Transport (EST) [RFC7030]. The New Entity contacts the Registrar using EST as indicated:
Functionality to provide generic "configuration" information is supported. The parsing of this data and any subsequent use of the data, for example communications with a Network Management System is out of scope but is expected to occur after bootstrapping enrollment is complete. This ensures that all communications with management systems which can divulge local security information (e.g. network topology or raw key material) is secured using the local credentials issued during enrollment.
The New Entity uses bootstrapping to join only one domain. Management by multiple domains is out-of-scope of bootstrapping. After the device has successfully joined a domain and is being managed it is plausible that the domain can insert credentials for other domains depending on the device capabilities.
See Section 3.5.
The role of the Proxy is to facilitate communications. The Proxy forwards packets between the New Entity and the Registrar that has been configured on the Proxy. The Proxy does not terminate the (d)TLS handshake.
In order to permit the proxy functionality to be implemented on the maximum variety of devices the chosen mechanism SHOULD use the minimum amount of state on the proxy device. While many devices in the ANIMA target space will be rather large routers, the proxy function is likely to be implemented in the control plane CPU such a device, with available capabilities for the proxy function similar to many class 2 IoT devices.
The document [I-D.richardson-anima-state-for-joinrouter] provides a more extensive analysis of the alternative proxy methods.
The proxy MUST implement an IPIP (protocol 41) encapsulation function for CoAP traffic to the configured UDP port on the registrar. The proxy does not terminate the CoAP DTLS connection. [[EDNOTE: The choice of CoAP as the mandatory to implement protocol rather than HTTP maximizes code reuse on the smallest of devices. Unfortunately this means this document will have to include the EST over CoAP details as additional sections. The alternative is to make 'HTTPS proxy' method the mandatory to implement and provide a less friendly environment for the smallest of devices. This is a decision we'll have to see addressed by the broader team.]]
As a result of the Proxy Discovery process in section Section 3.1.1, the port number exposed by the proxy does not need to be well known, or require an IANA allocation. The address and port of the Registrar will be discovered by the GRASP protocol inside the ACP. For the IPIP encapsulation methods, the port announced by the Proxy MUST be the same as on the registrar.
The IPIP encapsulation allows the proxy to forward traffic which is otherwise not to be forwarded, as the traffic between New Node and Proxy use IPv6 Link Local addresses.
If the Proxy device has more than one interface on which it offers the proxy function, then it must select a unique IP address per interface in order so that the proxy can stateless return the reply packets to the correct link.
The proxy SHOULD also provide one of: an IPIP encapsulation of HTTP traffic on TCP port TBD to the registrar, an HTTP proxy which accepts URLs that reach the Registrar, or a TCP circuit proxy that connects the New Node to the Registrar.
In order to make the HTTP choice above transparent to the New Node, the New Node will always initiate an HTTP connection, and will always send an appropriate CONNECT message to initiate an HTTPS connection to the registrar. [[EDNOTE: The CONNECT syntax is that the New Entity specifies the Registrar server in the CONNECT line. See RFC7231 s4.3.6. We wish the Proxy to override any value with the locally known-to-the-proxy Registrar address.]]
When the Proxy provides a circuit proxy to the Registrar the Registrar MUST accept HTTP connections, and be willing to perform an HTTP proxy (CONNECT) operation to itself, and then initiate HTTPS.
When the Proxy provides a stateless IPIP encapsulation to the Registrar, then the Registrar will have to perform IPIP decapsulation, remembering the originating outer IPIP source address in order to qualify the inner link-local address. Being able to connect a TCP (HTTP) or UDP (CoAP) socket to a link-local address with an encapsulated IPIP header requires API extensions beyond [RFC3542] for UDP use, and requires a form of connection latching (see section 4.1 of [RFC5386] and all of [RFC5660], except that a simple IPIP tunnel is used rather than an IPsec tunnel).
Once a Registrar is established it listens for new entities and determines if they can join the domain. The registrar delivers any necessary authorization information to the new device and facilitates enrollment with the domain PKI.
Registrar behavior is as follows:
Contacted by New Entity
+
|
+-------v----------+
| Entity | fail?
| Authentication +---------+
+-------+----------+ |
| |
+-------v----------+ |
| Entity | fail? |
| Authorization +--------->
+-------+----------+ |
| |
+-------v----------+ |
| Claiming the | fail? |
| Entity +--------->
+-------+----------+ |
| |
+-------v----------+ |
| Log Verification | fail? |
| +--------->
+-------+----------+ |
| |
+-------v----------+ +----v-------+
| Forward | | |
| Audit | | Reject |
| token + config | | Device |
| to the Entity | | |
+------------------+ +------------+
Figure 4
The applicable authentication methods detailed in EST [RFC7030] are:
In a fully automated network all devices must be securely identified and authorized to join the domain.
A Registrar accepts or declines a request to join the domain, based on the authenticated identity presented. Automated acceptance criteria include:
Since all New Entities accept Audit Tokens the Registrar MUST use the vendor provided MASA service to verify that the device's history log does not include unexpected Registrars. If a device had previously registered with another domain, the Registrar of that domain would show in the log.
In order to validate the IEEE 802.1AR device identity the Registrar maintains a database of vendor trust anchors (e.g. vendor root certificates or keyIdentifiers for vendor root public keys). For user interface purposes this database can be mapped to colloquial vendor names. Registrars can be shipped with the trust anchors of a significant number of third-party vendors within the target market.
If a device is accepted into the domain, it is expected request a domain certificate through a certificate enrollment process. The result is a common trust anchor and device certificates for all autonomic devices in a domain (these certificates can subsequently be used to determine the boundaries of the homenet, to authenticate other domain nodes, and to autonomically enable services on the homenet). The authorization performed during this phase MAY be cached for the TLS session and applied to subsequent EST enrollment requests so long as the session lasts.
Claiming an entity establishes an audit log at the MASA server and provides the Registrar with proof, in the form of a MASA authorization token, that the log entry has been inserted. As indicated in Section 3.1.4 a New Entity will only proceed with bootstrapping if a validated MASA authorization token has been received. The New Entity therefore enforces that bootstrapping only occurs if the claim has been logged. There is no requirement for the vendor to definitively know that the device is owned by the Registrar.
Registrar's obtain the Vendor URI via static configuration or by extracting it from the IEEE 802.1AR credential. The imprint method supported by the New Entity is known from the IEEE 802.1AR credential. [[EDNOTE: An appropriate extension for indicating the Vendor URI and imprint method could be defined using the methods described in [I-D.lear-mud-framework]]].
During initial bootstrapping the New Entity provides a nonce specific to the particular bootstrapping attempt. The Registrar SHOULD include this nonce when claiming the New Entity from the MASA service. Claims from an unauthenticated Registrar are only serviced by the MASA resource if a nonce is provided.
The Registrar can claim a New Entity that is not online by forming the request using the entities unique identifier and not including a nonce in the claim request. Audit Tokens obtained in this way do not have a lifetime and they provide a permanent method for the domain to claim the device. Evidence of such a claim is provided in the audit log entries available to any future Registrar. Such claims reduce the ability for future domains to secure bootstrapping and therefore the Registrar MUST be authenticated by the MASA service.
An ownership voucher requires the vendor to definitively know that a device is owned by a specific domain. The method used to "claim" this are out-of-scope. The Registrar simply requests an ownership validation token and the New Entity trusts the response.
The Registrar requests the log information for the new entity from the MASA service. The log is verified to confirm that the following is true to the satisfaction of the Registrar's configured policy:
If any of these criteria are unacceptable to the registrar the entity is rejected. The Registrar MAY be configured to ignore the history of the device but it is RECOMMENDED that this only be configured if hardware assisted NEA [RFC5209] is supported.
The Registrar forwards the received Audit Token to the New Entity. To simplify the message flows an initial configuration package can be delivered at this time which is signed by a representative of the domain.
[[EDNOTE: format TBD. The configuration package signature data must contain the full certificate path sufficient for the new entity to use the domainID information (as a trust anchor) to accept and validate the configuration)]]
The MASA service is provided by the Factory provider on the global Internet. The URI of this service is well known. The URI SHOULD also be provided as an IEEE 802.1AR IDevID X.509 extension (a "MASA Audit Token Distribution Point" extension).
The MASA service provides the following functionalities to Registrars:
A Registrar POSTs a claim message optionally containing the bootstrap nonce to the MASA server.
If a nonce is provided the MASA service responds to all requests. The MASA service verifies the Registrar is representative of the domain and generates a privacy protected log entry before responding with the Audit Token.
If a nonce is not provided then the MASA service MUST authenticate the Registrar as a valid customer. This prevents denial of service attacks.
When determining if a New Entity should be accepted into a domain the Registrar retrieves a copy of the audit log from the MASA service. This contains a list of privacy protected domain identities that have previously claimed the device. Included in the list is an indication of the time the entry was made and if the nonce was included.
As the devices have a common trust anchor, device identity can be securely established, making it possible to automatically deploy services across the domain in a secure manner.
Examples of services:
When a device has joined the domain, it can validate the domain membership of other devices. This makes it possible to create trust boundaries where domain members have higher level of trusted than external devices. Using the autonomic User Interface, specific devices can be grouped into to sub domains and specific trust levels can be implemented between those.
The assumption is that Network Access Control (NAC) completes using the New Entity 802.1AR credentials and results in the device having sufficient connectivity to discovery and communicate with the proxy. Any additional connectivity or quarantine behavior by the NAC infrastructure is out-of-scope. After the devices has completed bootstrapping the mechanism to trigger NAC to re-authenticate the device and provide updated network privileges is also out-of-scope.
This achieves the goal of a bootstrap architecture that can integrate with NAC but does not require NAC within the network where it wasn't previously required. Future optimizations can be achieved by integrating the bootstrapping protocol directly into an initial EAP exchange.
This section describes how an operator interacts with a domain that supports the bootstrapping as described in this document.
This is a one time step by the domain administrator. This is an "off the shelf" CA with the exception that it is designed to work as an integrated part of the security solution. This precludes the use of 3rd party certification authority services that do not provide support for delegation of certificate issuance decisions to a domain managed Registration Authority.
This is a one time step by the domain administrator. One or more devices in the domain are configured take on a Registrar function.
A device can be configured to act as a Registrar or a device can auto-select itself to take on this function, using a detection mechanism to resolve potential conflicts and setup communication with the Domain Certification Authority. Automated Registrar selection is outside scope for this document.
For each New Entity the Registrar is informed of the unique identifier (e.g. serial number) along with the manufacturer's identifying information (e.g. manufacturer root certificate). This can happen in different ways:
None of these approaches require the network to have permanent Internet connectivity. Even when the Internet based MASA service is used, it is possible to pre-fetch the required information from the MASA a priori, for example at time of purchase such that devices can enroll later. This supports use cases where the domain network may be entirely isolated during device deployment.
Additional policy can be stored for future authorization decisions. For example an expected deployment time window or that a certain Proxy must be used.
The approach outlined in this document provides a secure zero-touch method to enroll new devices without any pre-staged configuration. New devices communicate with already enrolled devices of the domain, which proxy between the new device and a Registrar. As a result of this completely automatic operation, all devices obtain a domain based certificate.
The certificate installed in the previous step can be used for all subsequent operations. For example, to determine the boundaries of the domain: If a neighbor has a certificate from the same trust anchor it can be assumed "inside" the same organization; if not, as outside. See also Section 3.5.1. The certificate can also be used to securely establish a connection between devices and central control functions. Also autonomic transactions can use the domain certificates to authenticate and/or encrypt direct interactions between devices. The usage of the domain certificates is outside scope for this document.
For simplicity the bootstrapping protocol is described as extensions to EST [RFC7030].
EST provides a bootstrapping mechanism for new entities that are configured with the URI of the EST server such that the Implicit TA database can be used to authenticate the EST server. Alternatively EST clients can "engage a human user to authorize the CA certificate using out-of-band data such as a CA certificate". EST does not provide a completely automated method of bootstrapping the PKI as both of these methods require some user input (either of the URI or authorizing the CA certificate).
This section details additional EST functionality that support automated bootstrapping of the public key infrastructure. These additions provide for fully automated bootstrapping. These additions are to be optionally supported by the EST server within the same .well-known URI tree as the existing EST URIs.
The "New Entity" is the EST client and the "Registrar" is the EST server.
The extensions for the client are as follows:
These extensions could be implemented as an independent protocol from EST but since the overlap with basic enrollment is extensive, particularly with respect to client authorization, they are presented here as additions to EST.
In order to obtain a validated Audit Token and Audit Log the Registrar contacts the MASA service Service using REST calls:
+-----------+ +----------+ +-----------+ +----------+
| New | | | | | | |
| Entity | | Proxy | | Registrar | | Vendor |
| | | | | | | |
++----------+ +--+-------+ +-----+-----+ +--------+-+
| | | |
| | | |
| (D)TLS hello | | |
Establish +---------------> (D)TLS hello | |
(D)TLS | |---------------> |
connection | (forwarding) | |
| Server Cert <---------------+ |
<---------------+ | |
| Client Cert | | |
+-------------------------------> |
| | | |
HTTP REST | POST /requestaudittoken | |
Data +--------------------nonce------> |
| . | /requestaudittoken
| . +---------------->
| <----------------+
| | /requestauditlog
| +---------------->
| audit token or owner voucher <----------------+
<-------------------------------+ |
| (optional config information) | |
| . | |
| . | |
Figure 5
In some use cases the Registrar may need to contact the Vendor in advanced, for example when the target network is air-gapped. The nonceless request format is provided for this and the resulting flow is slightly different. The security differences associated with not knowing the nonce are discussed below:
+-----------+ +----------+ +-----------+ +----------+
| New | | | | | | |
| Entity | | Proxy | | Registrar | | Vendor |
| | | | | | | |
++----------+ +--+-------+ +-----+-----+ +--------+-+
| | | |
| | | |
| | | /requestaudittoken
| | (nonce +---------------->
| | unknown) <----------------+
| | | /requestauditlog
| | +---------------->
| | <----------------+
| (D)TLS hello | | |
Establish +---------------> (D)TLS hello | |
(D)TLS | |---------------> |
connection | (forwarding) | |
| SerVer Cert <---------------+ |
<---------------+ | |
| Client Cert | | |
+-------------------------------> |
| | | |
HTTP REST | POST /requestaudittoken | |
Data +----------------------nonce----> (discard |
| audit token or owner Voucher | nonce) |
<-------------------------------+ |
| (optional config information) | |
| . | |
| . | |
Figure 6
The Registrar authenticates the client and performs authorization checks to ensure this client is expected to join the domain. This require a common procedure for representing and verifying the identity of the client. The methods detailed in [RFC6125] such as matching DNS Domain Name or Application Service Type are not directly applicable.
Clients presents an IEEE 802.AR certificate complete with subject field identifying the device uniquely in the Distinguished Name serialNumber subfield. The subjectAltName MAY contain a hardwareModuleName as specified in RFC4108. The Registrar extracts this information and compares against a per vendor access control list. (This can be implemented with a single database table so long as the authority key identifier is also maintained and checked to ensure that no two vendors collide in their use of serialNumber's).
When enrollment is complete and a local certificate is issued to the new device the local CA has complete control over the namespace. If this credential is intended for RFC6125 style TLS connections where servers are identified by a server's DNS-ID identity the CA is likely to ensure the dNSName field is populated. For Anima purposes the IEEE 802.1AR serialNumber and hardwareModuleName fields MUST be propagated to the issued certificate.
[[EDNOTE: the above authority key identifier trick works for database lookups and here the inclusion of the DNS name would serve the same purpose. Alternatively an Anima specified domain specific identifier must be indicated.]]
[[EDNOTE: In order to support smaller devices the above section on Proxy behavior introduces mandatory to implement support for CoAP support by the Proxy. This implies similar support by the New Entity and Registrar and means that the EST protocol operation encapsulation into CoAP needs to be described. EST is HTTP based and "CoaP is designed to easily interface with HTTP for integration" [RFC7252] so this section is anticipated to be relatively straightforward. A complexity is that the large message sizes necessary for bootstrapping will require support for [draft-ietf-core-block].]]
When the New Entity reaches the EST section 4.1.1 "Bootstrap Distribution of CA Certificates" state but wishes to proceed in a fully automated fashion it makes a request for a MASA authorization token from the Registrar.
This is done with an HTTPS POST using the operation path value of "/requestaudittoken".
The request format is JSON object containing a nonce.
Request media type: application/auditnonce
Request format: a JSON file with the following:
{"nonce":"<64bit nonce value>", "OwnershipValidation":boolean}
[[EDNOTE: exact format TBD. There is an advantage to having the client sign the nonce (similar to a PKI Certification Signing Request) since this allows the MASA service to confirm the actual device identity. It is not clear that there is a security benefit from this since its the New Entity that verifies the nonce.]]
The Registrar validates the client identity as described in EST [RFC7030] section 3.3.2. The registrar performs authorization as detailed in Section 3.3.2. If authorization is successful the Registrar obtains an Audit Token from the MASA service (see Section 5.2).
The received MASA authorization token is returned to the New Entity.
As indicated in EST [RFC7030] the bootstrapping server can redirect the client to an alternate server. If the New Entity authenticated the Registrar using the well known URI method then the New Entity MUST follow the redirect automatically and authenticate the new Registrar against the redirect URI provided. If the New Entity had not yet authenticated the Registrar because it was discovered and was not a known-to-be-valid URI then the new Registrar must be authenticated using one of the two autonomic methods described in this document.
The Registrar requests the Audit Token from the MASA service using a REST interface. For simplicity this is defined as an optional EST message between the Registrar and an EST server running on the MASA service although the Registrar is not required to make use of any other EST functionality when communicating with the MASA service. (The MASA service MUST properly reject any EST functionality requests it does not wish to service; a requirement that holds for any REST interface).
This is done with an HTTP POST using the operation path value of "/requestaudittoken".
The request format is a JSON object optionally containing the nonce value (as obtained from the bootstrap request) and the IEEE 802.1AR identity of the device as a serial number (the full certificate is not needed and no proof-of-possession information for the device identity is included). The New Entity's serial number is extracted from the IEEE 802.1AR subject name:
{"nonce":"<64bit nonce value>", "serialnumber", "<subjectname/subjectaltname serial number>"}
The Registrar MAY exclude the nonce from the request. Doing so allows the Registrar to request an authorization token when the New Entity is not online, or when the target bootstrapping environment is not on the same network as the MASA server. If a nonce is not provided the MASA server MUST authenticate the client as described in EST [RFC7030] section 3.3.2. The registrar performs authorization as detailed in Section 3.3.2. If authorization is successful the Registrar obtains an Audit Token from the MASA service (see Section 5.4).
The JSON message information is encapsulated in a PKCS7 signed data structure that is signed by the Registrar. The entire certificate chain, up to and including the Domain CA, MUST be included in the PKCS7.
The MASA service checks the internal consistency of the PKCS7 but MAY not authenticate the domain identity information. The domain is not know to the MASA server in advance and a shared trust anchor is not implied. The MASA server MUST verify that the PKCS7 is signed by a Registrar certificate (by checking for the cmc-idRA field) that was issued by a the root certificate included in the PKCS7. This ensures that the Registrar is in fact an authorized Registrar of the unknown domain.
The domain ID (e.g. hash of the public key of the domain) is extracted from the root certificate and is used to populate the MASA authorization token and to update the audit log. The authorization token consists of the nonce, if supplied, the serialnumber and the domain identity:
{"nonce":"<64bit nonce value>", "serialnumber", "<subjectname/subjectaltname serial number>","domainID":}
[[EDNOTE: There is a strong similarity between this and the previous section. Both involve requesting the Audit Token from the upstream element. Because there are differing requirements on the data submitted and the signing of that data they are specified in distinct sections. The design team should have a meeting to discuss how to unify these sections or make the distinctions more clear]]
When the MASA authorization token is returned to the New Entity an arbitrary information package can be signed and delivered along side it. This is signed by the Domain Registrar. The New Entity first verifies the Audit Token and, if it is valid, then uses the domain's TA to validate the Information Package.
[[EDNOTE: The domainID as included in the log and as sent in the authorization token is only a hash of the domain root certificate. This is insufficient for the new entity to move out of the provisional state as it needs a full root certificate to validate the TLS certificate chain. This information package could be used to deliver the full certificate or the full certificate could be included in the authorization token. Lacking either the new entity needs to stay in the provisional state until it performs an RFC7030 /getcacerts to obtain the full certificate chain.]]
[[EDNOTE: The package format to be specified here. Any signed format is viable and ideally one can simply be specified from netconf. The Registar knows the New Entity device type from the 802.1AR credential and so is able to determine the proper format for the configuration.]]
A registrar requests the MASA authorization log from the MASA service using this EST extension.
This is done with an HTTP GET using the operation path value of "/requestMASAlog".
The log data returned is a file consisting of all previous log entries. For example:
"log":[
{"date":"<date/time of the entry>"},
"domainID":"<domainID as extracted from the root
certificate within the PKCS7 of the
audit token request>",
"nonce":"<any nonce if supplied (or NULL)>"},
{"date":"<date/time of the entry>"},
"domainID":"<domainID as extracted from the root
certificate within the PKCS7 of the
audit token request>",
"nonce":"<any nonce if supplied (or NULL)>"},
]
Distribution of a large log is less than ideal. This structure can be optimized as follows: All nonce-less entries for the same domainID can be condensed into the single most recent nonceless entry.
The Registrar uses this log information to make an informed decision regarding the continued bootstrapping of the New Entity. For example if the log includes unexpected domainIDs this is indicative of problematic imprints by the new entity. If unexpected nonce-less entries exist this is indicative of the permanent ability for the unknown domain to trigger a reset of the device and take over management of it. Equipment that is purchased pre-owned can be expected to have an extensive history.
Log entries containing the Domain's ID can be compared against local history logs in search of discrepancies.
[[EDNOTE: certificate transparency style use of merkle tree hash's might offer an alternative log entry method]]
A common requirement of bootstrapping is to support less secure operational modes for support specific use cases. The following sections detail specific ways that the New Entity, Registrar and MASA can be configured to run in a less secure mode for the indicated reasons.
+--------+ +-------+ +------------+ +------------+
| New | | Proxy | | Domain | | Vendor |
| Entity | | | | Registrar | | Service |
| | | | | | | (Internet |
+--------+ +-------+ +------------+ +------------+
Figure 7
Although New Entity can choose to run in less secure modes this is MUST NOT be the default state because it permanently degrades the security for all other uses cases.
The device may have an operational mode where it skips Audit Token or Ownership Voucher validation one time. For example if a physical button is depressed during the bootstrapping operation. This can be useful if the vendor service is unavailable. This behavior SHOULD be available via local configuration or physical presence methods to ensure new entities can always be deployed even when autonomic methods fail. This allows for unsecure imprint.
It is RECOMMENDED that this only be available if hardware assisted NEA [RFC5209] is supported.
The Registrar can choose to accept devices using less secure methods. These methods are RECOMMENDED when low security models are needed as the security decisions are being made by the local administrator:
These modes are not available for devices that require a vendor Ownership Voucher. The methods vendors use to determine which devices are owned by which domains is out-of-scope.
Lower security modes chosen by the MASA service effect all device deployments unless bound to the specific device identities. In which case these modes can be provided as additional features for specific customers. The MASA service can choose to run in less secure modes by:
In order to support a wide variety of use cases, devices can be claimed by a registrar without proving possession of the device in question. This would result in a nonceless, and thus always valid, claim. Or would result in an invalid nonce being associated with a claim. The MASA service is required to authenticate such Registrars but no programmatic method is provided to ensure good behavior by the MASA service. Nonceless entries into the audit log therefore permanently reduce the value of a device because future Registrars, during future bootstrap attempts, would now have to be configured with policy to ignore previously (and potentially unknown) domains.
Future registrars are recommended to take the audit history of a device into account when deciding to join such devices into their network. If the MASA server were to have allowed a significantly large number of claims this might become onerous to the MASA server which must maintain all the extra log entries. Ensuring the Registrar is representative of a valid customer domain even without validating ownership helps to mitigate this.
It is possible for an attacker to send an authorization request to the MASA service directly after the real Registrar obtains an authorization log. If the attacker could also force the bootstrapping protocol to reset there is a theoretical opportunity for the attacker to use the Audit Token to take control of the New Entity but then proceed to enroll with the target domain. Possible prevention mechanisms include:
As indicated in EST [RFC7030] the connection is provisional and untrusted until the server is successfully authorized. If the server provides a redirect response the client MUST follow the redirect but the connection remains provisional. If the client uses a well known URI for contacting a well known Registrar the EST Implicit Trust Anchor database is used as is described in RFC6125 to authenticate the well known URI. In this case the connection is not provisional and RFC6125 methods can be used for each subsequent redirection.
The MASA service could lock a claim and refuse to issue a new token or the MASA service could go offline (for example if a vendor went out of business). This functionality provides benefits such as theft resistance, but it also implies an operational risk to the Domain that Vendor behavior could limit future bootstrapping of the device by the Domain. This can be mitigated by Registrars that request nonce-less authorization tokens.
We would like to thank the various reviewers for their input, in particular Markus Stenberg, Brian Carpenter, Fuyu Eleven.
| [I-D.behringer-homenet-trust-bootstrap] | Behringer, M., Pritikin, M. and S. Bjarnason, "Bootstrapping Trust on a Homenet", Internet-Draft draft-behringer-homenet-trust-bootstrap-02, February 2014. |
| [I-D.ietf-ace-actors] | Gerdes, S., Seitz, L., Selander, G. and C. Bormann, "An architecture for authorization in constrained environments", Internet-Draft draft-ietf-ace-actors-03, March 2016. |
| [I-D.ietf-netconf-zerotouch] | Watsen, K. and M. Abrahamsson, "Zero Touch Provisioning for NETCONF or RESTCONF based Management", Internet-Draft draft-ietf-netconf-zerotouch-07, March 2016. |
| [I-D.irtf-nmrg-autonomic-network-definitions] | Behringer, M., Pritikin, M., Bjarnason, S., Clemm, A., Carpenter, B., Jiang, S. and L. Ciavaglia, "Autonomic Networking - Definitions and Design Goals", Internet-Draft draft-irtf-nmrg-autonomic-network-definitions-07, March 2015. |
| [I-D.lear-mud-framework] | Lear, E., "Manufacturer Usage Description Framework", Internet-Draft draft-lear-mud-framework-00, January 2016. |
| [I-D.richardson-anima-state-for-joinrouter] | Richardson, M., "Considerations for stateful vs stateless join router in ANIMA bootstrap", Internet-Draft draft-richardson-anima-state-for-joinrouter-00, January 2016. |
| [imprinting] | Wikipedia, , "Wikipedia article: Imprinting", July 2015. |
| [pledge] | Dictionary.com, , "Dictionary.com Unabridged", July 2015. |