Network Working Group S. Bellovin
Request for Comments: 5406 Columbia University
BCP: 146 February 2009
Category: Best Current Practice
Guidelines for Specifying the Use of IPsec Version 2
Status of This Memo
This document specifies an Internet Best Current Practices for the
Internet Community, and requests discussion and suggestions for
improvements. Distribution of this memo is unlimited.
Copyright (c) 2009 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 in effect on the date of
publication of this document (http://trustee.ietf.org/license-info).
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document.
This document may contain material from IETF Documents or IETF
Contributions published or made publicly available before November
10, 2008. The person(s) controlling the copyright in some of this
material may not have granted the IETF Trust the right to allow
modifications of such material outside the IETF Standards Process.
Without obtaining an adequate license from the person(s) controlling
the copyright in such materials, this document may not be modified
outside the IETF Standards Process, and derivative works of it may
not be created outside the IETF Standards Process, except to format
it for publication as an RFC or to translate it into languages other
The Security Considerations sections of many Internet Drafts say, in
effect, "just use IPsec". While this is sometimes correct, more
often it will leave users without real, interoperable security
mechanisms. This memo offers some guidance on when IPsec Version 2
should and should not be specified.
The Security Considerations sections of many Internet Drafts say, in
effect, "just use IPsec". While the use of IPsec is sometimes the
correct security solution, more information is needed to provide
interoperable security solutions. In some cases, IPsec is
unavailable in the likely endpoints. If IPsec is unavailable to --
and hence unusable by -- a majority of the users in a particular
protocol environment, then the specification of IPsec is tantamount
to saying "turn off security" within this community. Further, when
IPsec is available, the implementation may not provide the proper
granularity of protection. Finally, if IPsec is available and
appropriate, the document mandating the use of IPsec needs to specify
just how it is to be used.
The goal of this document is to provide guidance to protocol
designers on the specification of IPsec when it is the appropriate
security mechanism. The protocol specification is expected to
provide realistic, interoperable security. Therefore, guidance on
the configuration of the various IPsec databases, such as the
Security Policy Database (SPD), is often required.
This document describes how to specify the use of IPsec Version 2
[RFC2401] including the ESPv2 (Encapsulating Security Payload version
2) [RFC2406], AHv2 (Authentication Header version 2) [RFC2402], and
IKEv1 (Internet Key Exchange version 1) [RFC2409]. A separate
document will describe the IPsec Version 3 suite [RFC4301] [RFC4302]
For further guidance on security considerations (including discussion
of IPsec), see [RFC3552].
NOTE: Many of the arguments below relate to the capabilities of
current implementations of IPsec. These may change over time; this
advice is based on the knowledge available to the IETF at publication
The design of security protocols is a subtle and difficult art. The
cautions here about specifying the use of IPsec should NOT be taken
to mean that you should invent your own new security protocol for
each new application. If IPsec is a bad choice, using another
standardized, well-understood security protocol will almost always
give the best results for both implementation and deployment.
Security protocols are very hard to design; rolling out a new one
will require extensive theoretical and practical work to confirm its
security properties and will incur both delay and uncertainty.
3. The Pieces of IPsec
IPsec is composed of a number of different pieces. These can be used
to provide confidentiality, integrity, and replay protection; though
some of these can be configured manually, generally a key management
component is used. Additionally, the decision about whether and how
to use IPsec is controlled by a policy database of some sort.
3.1. AH and ESP
The Authentication Header (AH) [RFC2402] and the Encapsulating
Security Payload (ESP) [RFC2406] are the over-the-wire security
protocols. Both provide (optional) replay protection. ESP typically
is used to provide confidentiality (encryption), integrity, and
authentication for traffic. ESP also can provide integrity and
authentication without confidentiality, which makes it a good
alternative to AH in most cases where confidentiality is not a
required or desired service. Finally, ESP can be used to provide
confidentiality alone, although this is not recommended [Bell96].
The difference in integrity protection offered by AH is that AH
protects portions of the preceding IP header, including the source
and destination address. However, if ESP is used in tunnel mode (see
Section 3.2) and integrity/authentication is enabled, the IP header
seen by the source and destination hosts is completely protected
AH can also protect those IP options that need to be seen by
intermediate routers, but must be intact and authentic when delivered
to the receiving system. At this time, use (and existence) of such
IP options is extremely rare.
If an application requires such protection, and if the information to
be protected cannot be inferred from the key management process, AH
must be used. (ESP is generally regarded as easier to implement;
however, virtually all IPsec packages support both.) If
confidentiality is required, ESP must be used. It is possible to use
AH in conjunction with ESP, but this combination is rarely required.
All variants of IPsec have problems with NAT boxes -- see [RFC3715]
for details -- but AH is considerably more troublesome. In
environments where there is substantial likelihood that the two
endpoints will be separated by a NAT box -- this includes almost all
services involving user-to-server traffic, as opposed to server-to-
server traffic -- NAT traversal [RFC3948] should be mandated and AH
should be avoided. (Note that [RFC3948] is for ESP only, and cannot
be used for AH.)
3.2. Transport and Tunnel Mode
AH and ESP can both be used in either transport mode or tunnel mode.
In tunnel mode, the IPsec header is followed by an inner IP header.
This is the normal usage for Virtual Private Networks (VPN) and is
generally required whenever either end of the IPsec-protected path is
not the ultimate IP destination, e.g., when IPsec is implemented in a
firewall, router, etc.
Transport mode is preferred for point-to-point communication, though
tunnel mode can also be used for this purpose.
3.3. Key Management
Any cryptographic system requires key management. IPsec provides for
both manual and automatic key management schemes. Manual key
management is easy; however, it doesn't scale very well. Also,
IPsec's replay protection mechanisms are not available if manual key
management is used. The need for automatic key exchange is discussed
in more detail in [RFC4107].
The primary automated key exchange mechanism for IPsec is the
Internet Key Exchange (IKE) [RFC2409]. A new, simpler version of IKE
has been approved [RFC4306], but many existing systems still use
IKEv1. This document does not discuss IKEv2 and IPsecv3. A second
mechanism, Kerberized Internet Negotiation of Keys (KINK) [RFC4430],
has been defined. It, of course, uses Kerberos and is suitable if
and only if a Kerberos infrastructure is available.
If a decision to use IKE is made, the precise mode of operation must
be specified as well. IKE can be used in main mode or aggressive
mode; both support digital signatures, two different ways of using
public key encryption, and shared secrets for authentication.
Shared secret authentication is simpler; however, it doesn't scale as
well in many-to-many communication scenarios since each endpoint must
share a unique secret with every peer with which it can communicate.
Note, though, that using shared secrets in IKE is far preferable to
In most real-world situations where public key modes of IKE are used,
locally issued certificates are employed. That is, the administrator
of the system or network concerned will issue certificates to all
authorized users. These certificates are useful only for IPsec.
It is sometimes possible to use certificates [RFC5280] from an
existing Public Key Infrastructure (PKI) with IKE. In practice, this
is rare. Furthermore, not only is there no global PKI covering most
Internet endpoints, there probably never will be. Designing a
structure that assumes such a PKI is a mistake. In particular,
assuming that an arbitrary node will have an "authentic" certificate,
issued by a mutually trusted third party and vouching for that node's
identity, is wrong. Again, such a PKI does not and probably will not
exist. Public key IKE is generally a good idea, but should almost
always be used with locally issued certificates as opposed to
certificates from an existing PKI.
Note that public key schemes require a substantial amount of
computation. Protocol designers should consider whether or not such
computations are feasible on devices of interest to their clientele.
Using certificates roughly doubles the number of large
exponentiations that must be performed, compared with shared secret
versions of IKE.
Today, even low-powered devices can generally perform enough
computation to set up a limited number of security associations.
Concentration points, such as firewalls or VoIP servers, may require
hardware assists, especially if many peers are expected to create
security associations at about the same time.
Using any automated key management mechanism can be difficult when
trying to protect low-level protocols. For example, even though
[RFC2461] specified the use of IPsec to protect IPv6 Neighbor
Discovery, it was impossible to do key management: nodes couldn't use
IKE because it required IP-level communication, and that isn't
possible before Neighbor Discovery associations are set up.
3.4. Application Programming Interface (API)
It is, in some sense, a misnomer to speak of the API as a part of
IPsec since this piece is missing on many systems. To the extent
that APIs exist, they aren't standardized. The problem is simple:
there is no portable way (and often no way at all) for an application
to request IPsec protection, or to tell if it was used for given
inbound packets or connections.
There are additional problems:
o Applications rarely have access to such APIs. Rather, IPsec is
usually configured by a system or network administrator.
o Applications are unable to verify that IPsec services are being
o Applications are unaware of the specific identities and properties
of the protected channel provided by IPsec. For instance, the
IPsec key management mechanisms may be aware of the identity and
authorization of the peer, but this information cannot be used by
the application nor linked to application-level decisions, such as
access to resources reserved to the entity identified by this
Router- or firewall-based IPsec implementations pose even greater
problems since there is no standardized over-the-wire protocol for
communicating this information from outboard encryptors to hosts.
By contrast, higher-layer security services, such as TLS, are able to
provide the necessary control and assurance.
4. Availability of IPsec in Target Devices
Although IPsec is now widely implemented and is available for current
releases of most host operating systems, it is less available for
embedded systems. Few hubs, network address translators, etc.,
implement it, especially at the low end. It is generally
inappropriate to rely on IPsec when many of the endpoints are in this
Even for host-to-host use, IPsec availability (and experience and
ease of use) has generally been for VPNs. Hosts that support IPsec
for VPN use frequently do not support it on a point-to-point basis,
especially via a stable, well-defined API or user interface.
Finally, few implementations support multiple layers of IPsec. If a
telecommuter is using IPsec in VPN mode to access an organizational
network, he or she may not be able to employ a second level of IPsec
to protect an application connection to a host within the
organization. (We note that such support is, in fact, mandated by
Case 4 of Section 4.5 of [RFC2401]. Nevertheless, it is not widely
available.) The likelihood of such deployment scenarios should be
taken into account when deciding whether or not to mandate IPsec.
[RFC2401] describes many different forms of endpoint identifier.
These include source addresses (both IPv4 and IPv6), host names
(possibly as embedded in X.500 certificates), and user IDs (again,
possibly as embedded in a certificate). Not all forms of identifier
are available on all implementations; in particular, user-granularity
identification is not common. This is especially a concern for
multi-user systems, where it may not be possible to use different
certificates to distinguish between traffic from two different users.
Again, we note that the ability to provide fine-grained protection,
such as keying each connection separately and with per-user
credentials, was one of the original design goals of IPsec.
Nevertheless, only a few platforms support it. Indeed, some
implementations do not even support using port numbers when deciding
whether or not to apply IPsec protection.
6. Selectors and the SPD
Section 4.4 of [RFC2401] describes the Security Policy Database (SPD)
and "selectors" used to decide what traffic should be protected by
IPsec. Choices include source and destination addresses (or address
ranges), protocol numbers (i.e., 6 for TCP and 17 for UDP), and port
numbers for TCP and UDP. Protocols whose protection requirements
cannot be described in such terms are poorer candidates for IPsec; in
particular, it becomes impossible to apply protection at any finer
grain than "destination host". Thus, traffic embedded in a Layer 2
Tunneling Protocol (L2TP) [RFC2661] session cannot be protected
selectively by IPsec above the L2TP layer, because IPsec has no
selectors defined that let it peer into the L2TP packet to find the
TCP port numbers. Similarly, the Stream Control Transmission
Protocol (SCTP) [RFC4960] did not exist when [RFC2401] was written;
thus, protecting individual SCTP applications on the basis of port
number could not be done until a new document was written [RFC3554]
that defined new selectors for IPsec, and implementations appeared.
Furthermore, in a world that runs to a large extent on dynamically
assigned addresses and often uses dynamically assigned port numbers
as well, an all-or-nothing policy for VPNs can work well; other
policies, however, can be difficult to create in any usable form.
The granularity of protection available may have side effects. If
certain traffic between a pair of machines is protected by IPsec,
does the implementation permit other traffic to be unprotected or
protected by different policies? Alternatively, if the
implementation is such that it is only capable of protecting all
traffic or none, does the device have sufficient CPU capacity to
encrypt everything? Note that some low-end devices may have limited
secure storage capacity for keys, etc.
Implementation issues are also a concern here. As before, too many
vendors have not implemented the full specification; too many IPsec
implementations are not capable of using port numbers in their
selectors. Protection of traffic between two hosts is thus on an
all-or-nothing basis when these non-compliant implementations are
7. Broadcast and Multicast
Although the designers of IPsec tried to leave room for protection of
multicast traffic, a complete design wasn't finished until much
later. As such, many IPsec implementations do not support multicast.
[RFC5374] describes extensions to IPsec to support it. Other
relevant documents include [RFC3830], [RFC3547], and [RFC4535].
Because of the delay, protocol designers who use multicast should
consider the availability of these extensions in target platforms of
8. Specifying IPsec
Despite all of the caveats given above, it may still be appropriate
to use IPsec in particular situations. The range of choices makes it
mandatory to define precisely how IPsec is to be used. Authors of
standards documents that rely on IPsec must specify the following:
a. What selectors should the initiator of the conversation (the
client, in client-server architectures) use? What addresses,
port numbers, etc., are to be used?
b. What IPsec protocol is to be used: AH or ESP? What mode is to be
employed: transport mode or tunnel mode?
c. What form of key management is appropriate?
d. What form of identification should be used? Choices include IP
address, DNS name with or without a user name, and X.500
e. If the application server will switch user IDs (i.e., it is a
login service of some sort) and user name identification is used,
is a new security association negotiated that utilizes a user-
granularity certificate? If so, when?
f. What form of authentication should be used? Choices include pre-
shared secrets and certificates.
g. How are the participants authorized to perform the operations
that they request? For instance, are all devices with a
certificate from a particular source allowed to use any
application with IPsec or access any resource? (This problem can
appear with any security service, of course.)
h. Which of the many variants of IKE must be supported? Main mode?
Note that there are two different versions of IKE: IKE and IKEv2.
IKEv2 is simpler and cleaner, but is not yet widely available.
You must specify which version of IKE you require.
i. Is suitable IPsec support available in likely configurations of
the products that would have to employ IPsec?
Let us now work through an example based on these guidelines. We
will use the Border Gateway Protocol (BGP) [RFC4271] to show how to
evaluate and specify the use of IPsec for transmission security,
rather than the mechanism described in [RFC2385]. Note carefully
that we are not saying that IPsec is an appropriate choice here.
Rather, we are demonstrating the necessary examination and
specification process. Also note that the deeper security issues
raised by BGP are not addressed by IPsec or any other transmission
security mechanism; see [Kent00a] and [Kent00b] for more details.
Selectors BGP runs between manually configured pairs of hosts
on TCP port 179. The appropriate selector would be
the pair of BGP speakers, for that port only. Note
that the router's "loopback address" is almost
certainly the address to use.
Mode Transport mode would be the proper choice if IPsec
were used. The information being communicated is
generally not confidential, so encryption need not
be used. Either AH or ESP can be used; if ESP is
used, the sender's IP address would need to be
checked against the IP address asserted in the key
management exchange. (This check is mandated by
[RFC2401].) For the sake of interoperability,
either AH or ESP would need to be specified as
mandatory to implement.
Key Management To permit replay detection, an automated key
management system should be used, most likely IKE.
Again, the RFC author should pick one.
Security Policy Connections should be accepted only from the
designated peer. (Note that this restriction
applies only to BGP. If the router -- or any IPsec
host -- runs multiple services with different
security needs, each such service requires its own
Authentication Given the number of BGP-speaking routers used
internally by large ISPs, it is likely that shared
key mechanisms are inadequate. Consequently,
certificate-based IKE must be supported. However,
shared secret mode is reasonable on peering links or
(perhaps) on links between ISPs and customers.
Whatever scheme is used, it must tie back to a
source IP address or Autonomous System (AS) number
in some fashion, since other BGP policies are
expressed in these terms. If certificates are used,
would they use IP addresses or AS numbers? Which?
Availability For this scenario, availability is the crucial
question. Do likely BGP speakers -- both backbone
routers and access routers -- support the profile of
IPsec described above? Will use of IPsec, with its
attendant expensive cryptographic operations, raise
the issue of new denial-of-service attacks? The
working group and the IESG must make these
determinations before deciding to use IPsec to
10. Security Considerations
IPsec provides transmission security and simple access control only.
There are many other dimensions to protocol security that are beyond
the scope of this memo, including most notably availability. For
example, using IPsec does little to defend against denial-of-service
attacks; in some situations, i.e., on CPU-limited systems, it may
contribute to the attacks. Within its scope, the security of any
resulting protocol depends heavily on the accuracy of the analysis
that resulted in a decision to use IPsec.
Ran Atkinson, Lakshminath Dondeti, Barbara Fraser, Paul Hoffman, Russ
Housley, Stephen Kent, Eric Fleischman, assorted members of the IESG,
and a plethora of others have made many useful suggestions.
12.1. Normative References
[RFC2401] Kent, S. and R. Atkinson, "Security Architecture for the
Internet Protocol", RFC 2401, November 1998.
[RFC2402] Kent, S. and R. Atkinson, "IP Authentication Header",
RFC 2402, November 1998.
[RFC2406] Kent, S. and R. Atkinson, "IP Encapsulating Security
Payload (ESP)", RFC 2406, November 1998.
[RFC2409] Harkins, D. and D. Carrel, "The Internet Key Exchange
(IKE)", RFC 2409, November 1998.
[RFC3554] Bellovin, S., Ioannidis, J., Keromytis, A., and R.
Stewart, "On the Use of Stream Control Transmission
Protocol (SCTP) with IPsec", RFC 3554, July 2003.
[RFC3948] Huttunen, A., Swander, B., Volpe, V., DiBurro, L., and M.
Stenberg, "UDP Encapsulation of IPsec ESP Packets",
RFC 3948, January 2005.
[RFC4107] Bellovin, S. and R. Housley, "Guidelines for Cryptographic
Key Management", BCP 107, RFC 4107, June 2005.
[RFC5280] Cooper, D., Santesson, S., Farrell, S., Boeyen, S.,
Housley, R., and W. Polk, "Internet X.509 Public Key
Infrastructure Certificate and Certificate Revocation List
(CRL) Profile", RFC 5280, May 2008.
[RFC5374] Weis, B., Gross, G., and D. Ignjatic, "Multicast
Extensions to the Security Architecture for the Internet
Protocol", RFC 5374, November 2008.
12.2. Informative References
[Bell96] Bellovin, S., "Problem Areas for the IP Security
Protocols", Proc. Sixth Usenix Security Symposium, pp.
[Kent00a] Kent, S., Lynn, C., and K. Seo, "Secure Border Gateway
Protocol (Secure-BGP)", IEEE Journal on Selected Areas in
Communications, 18:4, pp. 582-592, 2000.
[Kent00b] Kent, S., Lynn, C., Mikkelson, J., and K. Seo, "Secure
Border Gateway Protocol (Secure-BGP) -- Real World
Performance and Deployment Issues", Proc. Network and
Distributed System Security Symposium (NDSS), 2000.
[RFC2385] Heffernan, A., "Protection of BGP Sessions via the TCP MD5
Signature Option", RFC 2385, August 1998.
[RFC2461] Narten, T., Nordmark, E., and W. Simpson, "Neighbor
Discovery for IP Version 6 (IPv6)", RFC 2461,
[RFC2661] Townsley, W., Valencia, A., Rubens, A., Pall, G., Zorn,
G., and B. Palter, "Layer Two Tunneling Protocol "L2TP"",
RFC 2661, August 1999.
[RFC3547] Baugher, M., Weis, B., Hardjono, T., and H. Harney, "The
Group Domain of Interpretation", RFC 3547, July 2003.
[RFC3552] Rescorla, E. and B. Korver, "Guidelines for Writing RFC
Text on Security Considerations", BCP 72, RFC 3552,
[RFC3715] Aboba, B. and W. Dixon, "IPsec-Network Address Translation
(NAT) Compatibility Requirements", RFC 3715, March 2004.
[RFC3830] Arkko, J., Carrara, E., Lindholm, F., Naslund, M., and K.
Norrman, "MIKEY: Multimedia Internet KEYing", RFC 3830,
[RFC4271] Rekhter, Y., Li, T., and S. Hares, "A Border Gateway
Protocol 4 (BGP-4)", RFC 4271, January 2006.
[RFC4301] Kent, S. and K. Seo, "Security Architecture for the
Internet Protocol", RFC 4301, December 2005.
[RFC4302] Kent, S., "IP Authentication Header", RFC 4302,
[RFC4303] Kent, S., "IP Encapsulating Security Payload (ESP)",
RFC 4303, December 2005.
[RFC4306] Kaufman, C., "Internet Key Exchange (IKEv2) Protocol",
RFC 4306, December 2005.
[RFC4430] Sakane, S., Kamada, K., Thomas, M., and J. Vilhuber,
"Kerberized Internet Negotiation of Keys (KINK)",
RFC 4430, March 2006.
[RFC4535] Harney, H., Meth, U., Colegrove, A., and G. Gross,
"GSAKMP: Group Secure Association Key Management
Protocol", RFC 4535, June 2006.
[RFC4960] Stewart, R., "Stream Control Transmission Protocol",
RFC 4960, September 2007.
Steven M. Bellovin
1214 Amsterdam Avenue
New York, NY 10027
Phone: +1 212 939 7149