[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
A request to publish draft-moskowitz-hip-arch-05.txt as an Informational RFC
Dear RFC Editor,
As discussed earlier with Bob Braden, we hereby request
publication of draft-moskowitz-hip-arch-05.txt as an
Informational RFC, as per RFC2026 Section 4.2.3.
The draft has undergone an unofficial Last Call at the
hipsec mailing list <hipsec@honor.trusecure.org>, with
all raised issues resolved, and was announced at the IETF
discuss mailing list on Oct 23. See
http://www1.ietf.org/mail-archive/ietf/Current/msg22714.html
The latter announcement has generated no comments.
While we believe that the document is mature in the sense
that it reflects the current thinking of the people working
with HIP, this is the first time Pekka is submitting a
concrete draft to be published as an RFC. Hence, there are
probably nits that he has missed. Furthermore, Pekka has
requested reviews on the document from Dave Crocker and
Spencer Dawkins, and expects to receive comments in two weeks
or sooner. The idea is that those review comments could be
processed together with the comments from the RFC Editor and/or
from the IESG. If this does not fit in the process, please
advice us, and accept apologies from Pekka.
The draft is enclosed in xml2rfc format.
--Pekka Nikander & Bob Moskowitz
<?xml version="1.0" encoding="iso-8859-1" ?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" [
<!ENTITY % RFC2766 SYSTEM "reference.RFC.2766" >
<!ENTITY % RFC3022 SYSTEM "reference.RFC.3022" >
<!ENTITY % RFC3102 SYSTEM "reference.RFC.3102" >
<!ENTITY % I-D.moskowitz-hip SYSTEM
"reference.I-D.moskowitz-hip">
<!ENTITY % I-D.nikander-hip-mm SYSTEM
"reference.I-D.nikander-hip-mm">
<!ENTITY % I-D.ietf-ipseckey-rr SYSTEM
"reference.I-D.ietf-ipseckey-rr">
<!ENTITY % I-D.irtf-nsrg-report SYSTEM
"reference.I-D.irtf-nsrg-report">
<!ENTITY % I-D.nikander-mobileip-v6-ro-sec SYSTEM
"reference.I-D.nikander-mobileip-v6-ro-sec">
]>
<?rfc toc="yes"?>
<rfc ipr="full2026" docName="draft-moskowitz-hip-arch-05">
<front>
<title>Host Identity Protocol Architecture</title>
<author initials="R." surname="Moskowitz"
fullname="Robert Moskowitz">
<organization>
ICSAlabs, a Division of TruSecure Corporation
</organization>
<address>
<postal>
<street>1000 Bent Creek Blvd, Suite 200</street>
<city>Mechanicsburg</city>
<region>PA</region>
<country>USA</country>
</postal>
<email>rgm@icsalabs.com</email>
</address>
</author>
<author initials="P." surname="Nikander"
fullname="Pekka Nikander">
<organization>Ericsson Research Nomadic Lab</organization>
<address>
<postal>
<street />
<city>JORVAS</city>
<code>FIN-02420</code>
<country>FINLAND</country>
</postal>
<phone>+358 9 299 1</phone>
<email>pekka.nikander@nomadiclab.com</email>
</address>
</author>
<date month="Sep" year="2003" />
<area>Internet</area>
<keyword>Request for Comments</keyword>
<keyword>RFC</keyword>
<keyword>Internet Draft</keyword>
<keyword>I-D</keyword>
<abstract>
<t>This memo describes the reasoning behind a proposed new
namespace, the Host Identity namespace, and a new protocol
layer, the Host Identity Protocol, between the internetworking
and transport layers. Herein are presented the basics of the
current namespaces, strengths and weaknesses, and how a new
namespace will add completeness to them. The roles of this new
namespace in the protocols are defined.</t>
</abstract>
</front>
<middle>
<section title="Introduction">
<t>The Internet has created two global namespaces: Internet
Protocol (IP) addresses and Domain Name Service (DNS) names.
These two namespaces have a set of features and abstractions
that have powered the Internet to what it is today. They also
have a number of weaknesses. Basically, since they are all we
have, we try and do too much with them. Semantic overloading
and functionality extensions have greatly complicated these
namespaces.</t>
<t>The Host Identity namespace fills an important gap between
the IP and DNS namespaces. The Host Identity namespace consist
of Host Identifiers (HI). A Host Identifier is cryptographic in
its nature; it is the public key of an asymmetric key-pair. A
Host Identity is assigned to each host, or technically its
networking kernel or stack. Each host will have at least one
Host Identity and a corresponding Host Identifier, which can
either be public (e.g. published in DNS), or anonymous. Client
systems will tend to have both public and anonymous
Identities.</t>
<t>Although the Host Identities could be used in many
authentication systems, the presented architecture introduces a
new protocol, called the Host Identity Protocol (HIP), and a
cryptographic exchange, called the HIP base exchange <xref
target="I-D.moskowitz-hip"/>. The new protocol provides for
limited forms of trust between systems. It enhances mobility,
multi-homing and dynamic IP renumbering <xref
target="I-D.nikander-hip-mm" />, aids in protocol translation /
transition <xref target="I-D.moskowitz-hip" />, and reduces
certain types of denial-of-service (DoS) attacks <xref
target="I-D.moskowitz-hip" />.</t>
<t>When HIP is used, the actual payload traffic between two HIP
hosts is typically protected with IPsec. The Host Identities
are used to create the needed IPsec Security Associations (SA)
and to authenticate the hosts. The actual payload IP packets do
not differ in any way from standard IPsec protected IP
packets.</t>
</section>
<section title="Background">
<t>The Internet is built from three principle components:
computing platforms, packet transport (i.e. internetworking)
infrastructure, and services (applications). The Internet
exists to service two principal components: people and robotic
processes (silicon based people, if you will). All these
components need to be named in order to interact in a scalable
manner.</t>
<t>There are two principal namespaces in use in the Internet for
these components: IP numbers, and Domain Names. Email, HTTP and
SIP addresses are really only extensions of Domain Names.</t>
<t>IP numbers are a confounding of two namespaces, the names of
the networking interfaces and the names of the locations
('confounding' is a term used in statistics to discuss metrics
that are merged into one with a gain in indexing, but a loss in
informational value). The names of locations should be
understood as denoting routing direction vectors, i.e.,
information that is used to deliver packets to their
destinations.</t>
<t>IP numbers name networking interfaces, and typically only
when the interface is connected to the network. Originally IP
numbers had long-term significance. Today, the vast number of
interfaces use ephemeral and/or non-unique IP numbers. That is,
every time an interface is connected to the network, it is
assigned an IP number.</t>
<t>In the current Internet, the transport layers are coupled to
the IP addresses. Neither can evolve separately from the other.
IPng deliberations were framed by concerns of requiring a TCPng
effort as well. </t>
<t>Domain Names provide hierarchically assigned names for some
computing platforms and some services. Each hierarchy is
delegated from the level above; there is no anonymity in Domain
Names.</t>
<t>Email addresses provide naming for both humans and autonomous
applications. Email addresses are extensions of Domain Names,
only in so far as a named service is responsible for managing a
person's mail. There is some anonymity in Email addresses.</t>
<t>There are three critical deficiencies with the current
namespaces. Firstly, dynamic readdressing cannot be directly
managed. Secondly, anonymity is not provided in a consistent,
trustable manner. Finally, authentication for systems and
datagrams is not provided. All because computing platforms are
not well named with the current namespaces. </t>
<section title="A Desire for a Namespace for Computing Platforms">
<t>An independent namespace for computing platforms could be
used in end-to-end operations independent of the evolution of
the internetworking layer and across the many internetworking
layers. This could support rapid readdressing of the
internetworking layer either from mobility or renumbering.</t>
<t>If the namespace for computing platforms is
cryptographically based, it can also provide authentication
services. If this namespace is locally created without
requiring registration, it can provide anonymity. </t>
<t>Such a namespace (for computing platforms) and the names in
it should have the following characteristics:
<list>
<t>The namespace should be applied to the IP 'kernel'.
The IP kernel is the 'component' between services and the
packet transport infrastructure.</t>
<t>The namespace should fully decouple the internetworking
layer from the higher layers. The names should replace
all occurrences of IP addresses within applications (like
in the TCB). This may require changes to the current
APIs. In the long run, it is probable that some new APIs
are needed.</t>
<t>The introduction of the namespace should not mandate
any administrative infrastructure. Deployment must come
from the bottom up, in a pairwise deployment.</t>
<t>The names should have a fixed length representation,
for easy inclusion in datagrams and programming interfaces
(e.g the TCB).</t>
<t>Using the namespace should be affordable when used in
protocols. This is primarily a packet size issue. There
is also a computational concern in affordability.</t>
<t>The names must be statistically globally unique. 64
bits is inadequate (1% chance of collision in a population
of 640M); thus approximately 100 or more bits should be
used.</t>
<t>The names should have a localized abstraction so that
it can be used in existing protocols and APIs.</t>
<t>It must be possible to create names locally. This can
provide anonymity at the cost of making resolvability very
difficult.
<list style="empty">
<t>Sometimes the names may contain a delegation
component. This is the cost of resolvability.</t>
</list>
</t>
<t>The namespace should provide authentication services.
This is a preferred function.</t>
<t>The names should be long lived, but replaceable at any
time. This impacts access control lists; short lifetimes
will tend to result in tedious list maintenance or require
a namespace infrastructure for central control of access
lists.</t>
</list>
</t>
<t>In this document, such a new namespace is called the Host
Identity namespace. Using Host Identities requires its own
protocol layer, the Host Identity Protocol, between the
internetworking and transport layers. The names are based on
public key cryptography to supply authentication services.
Properly designed, it can deliver all of the above stated
requirements.</t>
</section>
</section>
<section title="Host Identity Namespace">
<t>A name in the Host Identity namespace, a Host Identifier
(HI), represents a statistically globally unique name for naming
any system with an IP stack. This identity is normally
associated, but not limited to, an IP stack. A system can have
multiple identities, some 'well known', some anonymous. A
system may self assert its identity, or may use a third-party
authenticator like DNSSEC, PGP, or X.509 to 'notarize' the
identity assertion. It is expected that the Host Identifiers
will initially be authenticated with DNSSEC and that all
implementations will support DNSSEC as a minimal baseline.</t>
<t>There is a subtle but important difference between Host
Identities and Host Identifiers. An Identity refers to the
abstract entity that is identified. An Identifier, on the other
hand, refers to the concrete bit pattern that is used in the
identification process.</t>
<t>In theory, any name that can claim to be 'statistically
globally unique' may serve as a Host Identifier. However, in
the authors' opinion, a public key of a 'public key pair' makes
the best Host Identifiers. As documented in the <xref
target="I-D.moskowitz-hip">Host Identity Protocol
specification</xref>, a public key based HI can authenticate the
HIP packets and protect them for man-in-the-middle attacks.
Since authenticated datagrams are mandatory to provide much of
HIP's denial-of-service protection, the Diffie-Hellman exchange
in HIP has to be authenticated. Thus, only public key HI and
authenticated HIP messages are supported in practice. In this
document, the non-cryptographic forms of HI and HIP are
presented to complete the theory of HI, but they should not be
implemented as they could produce worse denial-of-service
attacks than the Internet has without Host Identity.</t>
<section title="Host Identifiers">
<t>Host Identity adds two main features to Internet protocols.
The first is a decoupling of the internetworking and transport
layers; see <xref target="sec-architecture" />. This
decoupling will allow for independent evolution of the two
layers. Additionally, it can provide end-to-end services over
multiple internetworking realms. The second feature is host
authentication. Because the Host Identifier is a public key,
this key can be used to authenticate security protocols like
IPsec.</t>
<t>The only completely defined structure of the Host Identity
is that of a public key pair. In this case, the Host Identity
is referred to by its public component, the public key. Thus,
the name representing a Host Identity in the Host Identity
namespace, i.e. the Host Identifier, is the public key. In a
way, the possession of the private key defines the Identity
itself. If the private key is possessed by more than one
node, the Identity can be considered to be a distributed
one.</t>
<t>Architecturally, any other Internet naming convention might
form a usable base for Host Identifiers. However,
non-cryptographic names should only be used in situations of
high trust - low risk. That is any place where host
authentication is not needed (no risk of host spoofing) and no
use of IPsec. The current HIP documents do not specify how to
use any other types of Host Identifiers but public keys.</t>
<t>The actual Host Identities are never directly used in any
Internet protocols. The corresponding Host Identifiers
(public keys) may be stored in various DNS or LDAP directories
as identified elsewhere in this document, and they are passed
in the HIP base exchange. A Host Identity Tag (HIT) is used
in other protocols to represent the Host Identities. Another
representation of the Host Identities, the Local Scope
Identifier (LSI), can also be used in protocols and APIs.</t>
</section>
<section title="Storing Host Identifiers in DNS">
<t>The Host Identifiers should be stored in DNS. The
exception to this is anonymous identities. The HI is stored
in a new RR type, to be defined. This RR type is likely to be
quite similar to the <xref
target="I-D.ietf-ipseckey-rr">IPSECKEY RR</xref>.</t>
<t>Alternatively, or in addition to storing Host Identifiers
in the DNS, they may be stored in various kinds of Public Key
Infrastructure (PKI). Such a practice may allow them to be
used for purposes other than pure host identification.</t>
</section>
<section title="Host Identity Tag (HIT)">
<t>A Host Identity Tag is an 128-bit representation for a Host
Identity. It is created by taking a cryptographic hash over
the corresponding Host Identifier. There are two advantages
of using a hash over using the Host Identifier in protocols.
Firstly, its fixed length makes for easier protocol coding and
also better manages the packet size cost of this technology.
Secondly, it presents the identity in a consistent format to
the protocol independent of the whatever underlying technology
is used.</t>
<t>In the HIP packets, the HITs identify the sender and
recipient of a packet. Consequently, a HIT should be unique
in the whole IP universe. In the extremely rare case that a
single HIT happens to map to more than one Host Identities,
the Host Identifiers (public keys) will make the final
difference. If there is more than one public key for a given
node, the HIT acts as a hint for the correct public key to
use.</t>
</section>
<section title="Local Scope Identifier (LSI)">
<t>An LSI is a 32-bit localized representation for a Host
Identity. The purpose of an LSI is to facilitate using Host
Identities in existing protocols and APIs. LSI's advantage
over HIT is its size; its disadvantage is its local scope.
The generation of LSIs is defined in the <xref
target="I-D.moskowitz-hip">Host Identity Protocol
specification</xref>.</t>
<t>Examples of how LSIs can be used include: as the address in
a FTP command and as the address in a socket call. Thus, LSIs
act as a bridge for Host Identities into old protocols and
APIs.</t>
</section>
</section>
<section anchor="sec-architecture" title="New Stack Architecture">
<t>One way to characterize Host Identity is to compare the
proposed new architecture with the current one. As discussed
above, the IP addresses can be seen to be a confounding of
routing direction vectors and interface names. Using the
terminology from the <xref target="I-D.irtf-nsrg-report">IRTF
Name Space Research Group Report</xref> and, e.g., the
unpublished Internet-Draft <xref
target="chiappa-endpoints">Endpoints and Endpoint Names </xref>
by Noel Chiappa, the IP addresses currently embody the dual role
of locators and endpoint identifiers. That is, each IP address
names a topological location in the Internet, thereby acting as
a routing direction vector, or locator. At the same time, the IP
address names the physical network interface currently located
at the point-of-attachment, thereby acting as a endpoint
name.</t>
<t>In the HIP architecture, the endpoint names and locators are
separated from each other. IP addresses continue to act as
locators. The Host Identifiers take the role of endpoint
identifiers. It is important to understand that the endpoint
names based on Host Identities are slightly different from
interface names; a Host Identity can be simultaneously reachable
through several interfaces.</t>
<t>The difference between the bindings of the logical entities
are illustrated in <xref target="figure-bindings"/>.</t>
<figure anchor="figure-bindings">
<artwork src="draft-moskowitz-hip-arch-1.gif" type="gif">
Process ------ Socket Process ------ Socket
| |
| |
| |
| |
Endpoint | Endpoint --- Host Identity
\ | |
\ | |
\ | |
\ | |
Location --- IP address Location --- IP address
</artwork>
</figure>
<section title="Transport associations and endpoints">
<t>Architecturally, HIP provides for a different binding of
transport layer protocols. That is, the transport layer
associations, i.e., TCP connections and UDP associations, are
no more bound to IP addresses but to Host Identities.</t>
<t>It is possible that a single physical computer hosts
several logical endpoints. With HIP, each of these
endpoints would have a distinct Host Identity. Furthermore,
since the transport associations are bound to Host Identities,
HIP provides for process migration and clustered servers.
That is, if a Host Identity is moved from one physical
computer to another, it is also possible to simultaneously
move all the transport associations without breaking them.
Similarly, if it is possible to distribute the processing of a
single Host Identity over several physical computers, HIP
provides for cluster based services without any changes at the
client endpoint.</t>
</section>
</section>
<section title="End-Host Mobility and Multi-Homing">
<t>HIP decouples the transport from the internetworking layer,
and binds the transport associations to the Host Identities
(through actually either the HIT or LSI). Consequently, HIP can
provide for a degree of internetworking mobility and
multi-homing at a very low infrastructure cost. HIP mobility
includes IP address changes (via any method) to either party.
Thus, a system is considered mobile if its IP address can change
dynamically for any reason like PPP, DHCP, IPv6 prefix
reassignments, or a NAT device remapping its translation.
Likewise, a system is considered multi-homed if it has more than
one globally routable IP address at the same time. HIP allows
these IP addresses to be linked with each other, and if one
address becomes unusable (e.g. due to a network failure),
existing transport associations can be easily moved to another
address.</t>
<t>When a node moves while communication is already on-going,
address changes are rather straightforward. The peer of the
mobile node can just accept a HIP or an integrity protected
IPsec packet from any address and totally ignore the source
address. However, as discussed in <xref target="ssec-flooding"
/> below, a mobile node must send a HIP readdress packet to
inform the peer of the new address(es), and the peer must verify
that the mobile node is reachable through these addresses. This
is especially helpful for those situations where the peer node
is sending data periodically to the mobile node (that is
re-starting a connection after the initial connection).</t>
<section title="Rendezvous server">
<t>Making a contact to a mobile node is slightly more
involved. In order to start the HIP exchange, the initiator
node has to know how to reach the mobile node. Although
Dynamic DNS could be used for this function for infrequently
moving nodes, an alternative to using DNS in this fashion is
to use a piece of new static infrastructure called a HIP
rendezvous server. Instead of registering its current dynamic
address to the DNS server, the mobile node registers the
address(es) of its rendezvous server(s). The mobile node
keeps the rendezvous server(s) continuously updated with its
current IP address(es). A rendezvous server simply forwards
the initial HIP packet from an initiator to the mobile node at
its current location. All further packets flow between the
initiator and the mobile node. There is typically very little
activity on a rendezvous server, address updates and initial
HIP packet forwarding. Thus, one server can support a large
number of potential mobile nodes. The mobile nodes must trust
the rendezvous server to properly maintain their HIT and IP
address mappings.</t>
<t>The rendezvous server is also needed if both of the nodes
are mobile and happen to move at the same time. In that case,
the HIP readdress packets will cross each other in the network
and never reach the peer node. To solve this situation, the
nodes should remember the rendezvous server address, and
re-send the HIP readdress packet to the rendezvous server if
no reply is received.</t>
<t>The mobile node keeps its address current on the rendezvous
server by setting up a HIP association with the rendezvous
server and sending HIP readdress packets to it. A rendezvous
server will permit two mobile systems to use HIP without any
extraneous infrastructure (in addition to the rendezvous
server itself), including DNS if they have a method other than
a DNS query to get each other's HI and HIT.</t>
</section>
<section anchor="ssec-flooding"
title="Protection against Flooding Attacks">
<t>While the idea of informing about address changes by simply
sending packets with a new source address appears appealing,
it is not secure enough. That is, even if HIP does not rely
on the source address for anything (once the base exchange has
been completed), it appears to be necessary to check a mobile
node's reachability at the new address before actually sending
any larger amounts of traffic to the new address.</t>
<t>Blindly accepting new addresses would potentially lead to
flooding Denial-of-Service attacks against third parties <xref
target="I-D.nikander-mobileip-v6-ro-sec" />. In a distributed
flooding attack an attacker opens (anonymous) high volume HIP
connections with a large number of hosts, and then claims to
all of these hosts that it has moved to a target node's IP
address. If the peer hosts were to simply accept the move,
the result would be a packet flood to the target node's
address. To close this attack, HIP includes an address check
mechanism where the reachability of a node is separately
checked at each address before using the address for larger
amounts of traffic.</t>
<t>Whenever HIP is used between two hosts that fully trust
each other, the hosts may optionally decide to skip the
address tests. However, such performance optimization must be
restricted to peers that are known to be trustworthy and
capable of protecting themselves from malicious software.</t>
</section>
</section>
<section anchor="esp" title="HIP and IPsec">
<t>The preferred way of implementing HIP is to use IPsec to
carry the actual data traffic. As of today, the only completely
defined method is to use IPsec Encapsulated Security Payload
(ESP) to carry the data packets. In the future, other ways of
transporting payload data may be developed, including ones that
do not use cryptographic protection.</t>
<t>In practise, the HIP base exchange uses the cryptographic
Host Identifiers to set up a pair of ESP Security Associations
(SAs) to enable ESP in an end-to-end manner. This is
implemented in a way that can span addressing realms.</t>
<t>From a conceptual point of view, the IPsec Security Parameter
Index (SPI) in ESP provides a simple compression of the HITs.
This does require per-HIT-pair SAs (and SPIs), and a decrease of
policy granularity over other Key Management Protocols, such as
IKE and IKEv2. Future HIP extensions may provide for more
granularity and creation of several ESP SAs between a pair of
HITs</t>
<t>Since HIP is designed for host usage, not for gateways, only
ESP transport mode is supported. An ESP SA pair is indexed by
the SPIs and the two HITs (both HITs since a system can have
more than one HIT). The SAs need not to be bound to IP
addresses; all internal control of the SA is by the HITs. Thus,
a host can easily change its address using Mobile IP, DHCP, PPP,
or IPv6 readdressing and still maintain the SAs. Since the
transports are bound to the SA (via an LSI or a HIT), any active
transport is also maintained. Thus, real world conditions like
loss of a PPP connection and its re-establishment or a mobile
handover will not require a HIP negotiation or disruption of
transport services.</t>
<t>Since HIP does not negotiate any SA lifetimes, all lifetimes
are local policy. The only lifetimes a HIP implementation MUST
support are sequence number rollover (for replay protection),
and SA timeout. An SA times out if no packets are received
using that SA. Implementations MAY support lifetimes for the
various ESP transforms.</t>
</section>
<section title="HIP and NATs">
<t>Passing packets between different IP addressing realms
requires changing IP addresses in the packet header. This may
happen, for example, when a packet is passed between the public
Internet and a private address space, or between IPv4 and IPv6
networks. The address translation is usually implemented as
<xref target="RFC3022">Network Address Translation (NAT)</xref>
or <xref target="RFC2766"> NAT Protocol translation
(NAT-PT)</xref>.</t>
<t>In a network environment where the identification is based on
the IP addresses, identifying the communicating nodes is
difficult when NAT is used. With HIP, the transport layer
endpoints are bound to the Host Identities. Thus, a connection
between two hosts can traverse many addressing realm boundaries.
The IP addresses are used only for routing purposes; the IP
addresses may be changed freely during packet traversal.</t>
<t>For a HIP based flow, a NAT or NAT-PT system tracks the
mapping of HITs and the corresponding IPsec SPIs to an IP
address. Many HITs can map to a single IP address on a NAT,
simplifying connections on address poor NAT interfaces. The NAT
can gain much of its knowledge from the HIP packets themselves;
however, some NAT configuration may be necessary.</t>
<t>The NAT systems cannot touch the datagrams within the IPsec
envelope, thus application specific address translation must be
done in the end systems. HIP provides for 'Distributed NAT',
and uses the HIT or the LSI as a place holder for embedded IP
addresses.</t>
<section title="HIP and TCP Checksum">
<t>There is no way for a host to know if any of the IP
addresses in the IP header are the addresses used to calculate
the TCP checksum. That is, it is not feasible to calculate
the TCP checksum using the actual IP addresses in the pseudo
header; the addresses received in the incoming packet are not
necessarily the same as they were on the sending host.
Furthermore, it is not possible to recompute the upper layer
checksums in the NAT/NAT-PT system, since the traffic is IPsec
protected. Consequently, the TCP and UDP checksums are
calculated using the HITs in the place of the IP addresses in
the pseudo header. Furthermore, only the IPv6 pseudo header
format is used. This provides for IPv4 / IPv6 protocol
translation.</t>
</section>
</section>
<section title="HIP Policies">
<t>There are a number of variables that will influence the HIP
exchanges that each host must support. All HIP implementations
should support at least 2 HIs, one to publish in DNS and one for
anonymous usage. Although anonymous HIs will be rarely used as
responder HIs, they are likely be common for initiators.
Support for multiple HIs is recommended.</t>
<t>Many initiators would want to use a different HI for
different responders. The implementations should provide for a
policy of initiator HIT to responder HIT. This policy should
also include preferred transforms and local lifetimes. </t>
<t>Responders would need a similar policy, representing which
hosts they accept HIP exchanges, and the preferred transforms
and local lifetimes.</t>
</section>
<section title="Benefits of HIP">
<t>In the beginning, the network layer protocol (i.e. IP) had
the following four "classic" invariants:
<list>
<t>Non-mutable: The address sent is the address received.</t>
<t>Non-mobile: The address doesn't change during the course
of an "association".</t>
<t>Reversible: A return header can always be formed by
reversing the source and destination addresses.</t>
<t>Omniscient: Each host knows what address a partner host
can use to send packets to it.</t>
</list>
</t>
<t>Actually, the fourth can be inferred from 1 and 3, but it is
worth mentioning for reasons that will be obvious soon if not
already.</t>
<t>In the current "post-classic" world, we are trying
intentionally to get rid of the second invariant (both for
mobility and for multi-homing), and we have been forced to give
up the first and the fourth. <xref target="RFC3102">Realm
Specific IP</xref> is an attempt to reinstate the fourth
invariant without the first invariant. IPv6 is an attempt to
reinstate the first invariant.</t>
<t>Few systems on the Internet have DNS names that are
meaningful to them. That is, if they have a Fully Qualified
Domain Name (FQDN), that typically belongs to a NAT device or a
dial-up server, and does not really identify the system itself
but its current connectivity. FQDN names (and their extensions
as email names) are Application Layer names; more frequently
naming processes than a particular system. This is why many
systems on the internet are not registered in DNS; they do not
have processes of interest to other Internet hosts.</t>
<t>DNS names are indirect references to IP addresses. This only
demonstrates the interrelationship of the networking and
application layers. DNS, as the Internet's only deployed,
distributed, database is also the repository of other
namespaces, due in part to DNSSEC and application specific key
records. Although each namespace can be stretched (IP with v6,
DNS with KEY records), neither can adequately provide for host
authentication or act as a separation between internetworking
and transport layers.</t>
<t>The Host Identity (HI) namespace fills an important gap
between the IP and DNS namespaces. An interesting thing about
the HI is that it actually allows one to give-up all but the 3rd
Network Layer invariant. That is to say, as long as the source
and destination addresses in the network layer protocol are
reversible, then things work ok because HIP takes care of host
identification, and reversibility allows one to get a packet
back to one's partner host. You don't care if the network layer
address changes in transit (mutable) and you don't care what
network layer address the partner is using (non-omniscient).</t>
<t>Since all systems can have a Host Identity, every system can
have an entry in the DNS. The mobility features in HIP make it
attractive to trusted 3rd parties to offer rendezvous
servers.</t>
<section title="HIP's Answers to NSRG questions">
<t>The IRTF Name Space Research Group has posed a number of
evaluating questions in <xref
target="I-D.irtf-nsrg-report">their report</xref>. In this
section, we provide answers to these questions.
<list style="numbers">
<t>How would a stack name improve the overall
functionality of the Internet?
<list style="empty">
<t>At the fundamental level, HI decouples the
internetworking layer from the transport layer,
allowing each to evolve separately. At the same time,
the decoupling makes end-host mobility and
multi-homing easier. It also allows mobility and
multi-homing across the IPv4 and IPv6 networks. HIs
make network renumbering easier. At the conceptual
level, they also make process migration and clustered
servers easier to implement. Furthermore, being
cryptographic in nature, they provide the basis for
solving the security problems related to end-host
mobility and multi-homing.</t>
</list>
</t>
<t>What does a stack name look like?
<list style="empty">
<t>A HI is a cryptographic public key. However,
instead of using the keys directly, most protocols use
a fixed size hash of the public key.</t>
</list>
</t>
<t>What is its lifetime?
<list style="empty">
<t>HIP provides both stable and temporary Host
Identifiers. Stable HIs are typically long lived,
with a lifetime of years or more. The lifetime of
temporary HIs depends on how long the upper layer
connections and applications need them, and can range
from a few seconds to years.</t>
</list>
</t>
<t>Where does it live in the stack?
<list style="empty">
<t>The HIs live between the transport and
internetworking layers.</t>
</list>
</t>
<t>How is it used on the end points
<list style="empty">
<t>The Host Identifiers, in the form of HITs or LSIs,
are used by legacy applications as if they were IP
addresses. Additionally, the Host Identifiers, as
public keys, are used in the built in key agreement
protocol, called the HIP base exchange, to
authenticate the hosts to each other.</t>
</list>
</t>
<t>What administrative infrastructure is needed to support
it?
<list style="empty">
<t>It is possible to use HIP opportunistically,
without any infrastructure. However, to gain full
benefit from HIP, the HIs must be stored in the DNS or
a PKI, and a new infrastructure of rendezvous servers
is needed.</t>
</list>
</t>
<t>If we add an additional layer would it make the address
list in SCTP unnecessary?
<list style="empty">
<t>Yes</t>
</list>
</t>
<t>What additional security benefits would a new naming
scheme offer?
<list style="empty">
<t>HIP reduces dependency on IP addresses, making the
so called address ownership problems easier to solve.
In practice, HIP provides security for end-host
mobility and multi-homing. Furthermore, since HIP
Host Identifiers are public keys, standard public key
certificate infrastructures can be applied on the top
of HIP.</t>
</list>
</t>
<t>What would the resolution mechanisms be, or what
characteristics of a resolution mechanisms would be
required?
<list style="empty">
<t>For most purposes, an approach where DNS names are
resolved simultaneously to HIs and IP addresses is
sufficient. However, if it becomes necessary to
resolve HIs into IP addresses or back to DNS names, a
flat, hash based resolution infrastructure is needed.
Such an infrastructure could be based on the ideas of
Distributed Hash Tables, but would require significant
new development and deployment.</t>
</list>
</t>
</list>
</t>
</section>
</section>
<section title="Security Considerations">
<t>HIP takes advantage of the new Host Identity paradigm to
provide secure authentication of hosts and to provide a fast key
exchange for IPsec. HIP also attempts to limit the exposure of
the host to various denial-of-service (DoS) and
man-in-the-middle (MitM) attacks. In so doing, HIP itself is
subject to its own DoS and MitM attacks that potentially could
be more damaging to a host's ability to conduct business as
usual.</t>
<t>Resource exhausting Denial-of-service attacks take advantage
of the cost of setting up a state for a protocol on the
responder compared to the 'cheapness' on the initiator. HIP
allows a responder to increase the cost of the start of state on
the initiator and makes an effort to reduce the cost to the
responder. This is done by having the responder start the
authenticated Diffie-Hellman exchange instead of the initiator,
making the HIP base exchange 4 packets long. There are more
details on this process in the <xref
target="I-D.moskowitz-hip">Host Identity Protocol
specification</xref>. </t>
<t>HIP optionally supports opportunistic negotiation. That is,
if a host receives a start of transport without a HIP
negotiation, it can attempt to force a HIP exchange before
accepting the connection. This has the potential for DoS
attacks against both hosts. If the method to force the start of
HIP is expensive on either host, the attacker need only spoof a
TCP SYN. This would put both systems into the expensive
operations. HIP avoids this attack by having the responder send
a simple HIP packet that it can pre-build. Since this packet is
fixed and easily replayed, the initiator only reacts to it if it
has just started a connection to the responder.</t>
<t>Man-in-the-middle attacks are difficult to defend against,
without third-party authentication. A skillful MitM could
easily handle all parts of the HIP base exchange, but HIP
indirectly provides the following protection from a MitM attack.
If the responder's HI is retrieved from a signed DNS zone or
secured by some other means, the initiator can use this to
authenticate the signed HIP packets. Likewise, if the
initiator's HI is in a secure DNS zone, the responder can
retrieve it and validate the signed HIP packets. However, since
an initiator may choose to use an anonymous HI, it knowingly
risks a MitM attack. The responder may choose not to accept a
HIP exchange with an anonymous initiator.</t>
<t>In HIP, the Security Association for IPsec is indexed by the
SPI; the source address is always ignored, and the destination
address may be ignored as well. Therefore, HIP enabled IPsec
Encapsulated Security Payload (ESP) is IP address independent.
This might seem to make it easier for an attacker, but ESP with
replay protection is already as well protected as possible, and
the removal of the IP address as a check should not increase the
exposure of IPsec ESP to DoS attacks.</t>
<t>Since not all hosts will ever support HIP, ICMPv4
'Destination Unreachable, Protocol Unreachable' and ICMPv6
'Parameter Problem, Unrecognized Next Header' messages are to be
expected and present a DoS attack. Against an initiator, the
attack would look like the responder does not support HIP, but
shortly after receiving the ICMP message, the initiator would
receive a valid HIP packet. Thus, to protect against this
attack, an initiator should not react to an ICMP message until a
reasonable time has passed, allowing it to get the real
responder's HIP packet. A similar attack against the responder
is more involved.</t>
<t>Another MitM attack is simulating a responder's
administrative rejection of a HIP initiation. This is a simple
ICMP 'Destination Unreachable, Administratively Prohibited'
message. A HIP packet is not used because it would either have
to have unique content, and thus difficult to generate,
resulting in yet another DoS attack, or just as spoofable as the
ICMP message. Like in the previous case, the defense against
this attack is for the initiator to wait a reasonable time
period to get a valid HIP packet. If one does not come, then
the initiator has to assume that the ICMP message is valid.
Since this is the only point in the HIP base exchange where this
ICMP message is appropriate, it can be ignored at any other
point in the exchange.</t>
<section title="HITs used in ACLs">
<t>It is expected that HITs will be used in ACLs. Future
firewalls can use HITs to control egress and ingress to
networks, with an assurance level difficult to achieve today.
As discussed above in <xref target="esp" />, once a HIP
session has been established, the SPI value in an IPsec packet
may be used as an index, indicating the HITs. In practise,
the firewalls can inspect the HIP packets to learn of the
bindings between HITs, SPI values, and IP addresses. They can
even explicitly control IPsec usage, dynamically opening IPsec
ESP only for specific SPI values and IP addresses. The
signatures in the HIP packets allow a capable firewall to make
sure that the HIP exchange is indeed happening between two
known hosts. This may increase firewall security.</t>
<!-- <t>[add here wildcarding]</t> -->
<t>There has been considerable bad experience with distributed
ACLs that contain public key related material, for example,
with SSH. If the owner of the key needs to revoke it for any
reason, the task of finding all locations where the key is
held in an ACL may be impossible. If the reason for the
revocation is due to private key theft, this could be a
serious issue.</t>
<t>A host can keep track of all of its partners that might use
its HIT in an ACL by logging all remote HITs. It should only
be necessary to log responder hosts. With this information,
the host can notify the various hosts about the change to the
HIT. There has been no attempt to develop a secure method
(like in CMP and CMC) to issue the HIT revocation notice.</t>
<t>NATs, however, are transparent to the HIP aware systems by
design. Thus, the host may find it difficult to notify any
NAT that is using a HIT in an ACL. Since most systems will
know of the NATs for their network, there should be a process
by which they can notify these NATs of the change of the HIT.
This is mandatory for systems that function as responders
behind a NAT. In a similar vein, if a host is notified of a
change in a HIT of an initiator, it should notify its NAT of
the change. In this manner, NATs will get updated with the
HIT change.</t>
</section>
<section title="Non-security Considerations">
<t>The definition of the Host Identifier states that the HI
need not be a public key. It implies that the HI could be any
value; for example an FQDN. This document does not describe
how to support such a non-cryptographic HI. A
non-cryptographic HI would still offer the services of the HIT
or LSI for NAT traversal. It would be possible carry the HITs
in HIP packets that had neither privacy nor authentication.
Since such a mode would offer so little additional
functionality for so much addition to the IP kernel, it has
not been defined. Given how little public key cryptography
HIP requires, HIP should only be implemented using public key
Host Identities.</t>
<t>If it is desirable to use HIP in a low security situation
where public key computations are considered expensive, HIP
can be used with very short Diffie-Hellman and Host Identity
keys. Such use makes the participating hosts vulnerable to
MitM and connection hijacking attacks. However, it does not
cause flooding dangers, since the address check mechanism
relies on the routing system and not on cryptographic
strength.</t>
</section>
</section>
<section title="Acknowledgments">
<t>For the people historically involved in the early stages of
HIP, see the Acknowledgements section in the <xref
target="I-D.moskowitz-hip">Host Identity Protocol
specification</xref>.</t>
<t>During the later stages of this document, when the editing
baton was transfered to Pekka Nikander, the comments from the
early implementors and others, including Jari Arkko, Tom
Henderson, Petri Jokela, Miika Komu, Mika Kousa, Andrew
McGregor, Jan Melen, Tim Shepard, Jukka Ylitalo, and Jorma Wall,
were invaluable. </t>
</section>
</middle>
<back>
<references title="References (informative)">
&RFC2766;
&RFC3022;
&RFC3102;
&I-D.moskowitz-hip;
&I-D.ietf-ipseckey-rr;
&I-D.irtf-nsrg-report;
&I-D.nikander-hip-mm;
&I-D.nikander-mobileip-v6-ro-sec;
<reference anchor="chiappa-endpoints">
<front>
<title>Endpoints and Endpoint Names: A Proposed Enhancement
to the Internet Architecture</title>
<author initials="J. N." surname="Chiappa">
<organization />
</author>
<date year="1999" />
</front>
<seriesInfo name="URL"
value="http://users.exis.net/~jnc/tech/endpoints.txt"; />
<format type="txt"
target="http://users.exis.net/~jnc/tech/endpoints.txt"; />
</reference>
</references>
</back>
</rfc>