Until recently, remote access has typically been characterized by
dial-up users accessing the target network via the Public Switched
Telephone Network (PSTN), with the dial-up connection terminating at
a Network Access Server (NAS) within the target domain. The
protocols facilitating this have usually been PPP-based, and access
control, authorization, and accounting functions have typically been
provided using one or more of a number of available mechanisms,
including RADIUS [RADIUS].
Kelly & Ramamoorthi Informational [Page 2]
RFC 3457 IPsec Remote Access Scenarios January 2003
Note that for such access, it has often been assumed that the
communications infrastructure supporting the ISP connection (the
PSTN) is relatively secure, and poses no significant threats to
communications integrity or confidentiality. Based on this
assumption, connection security has been limited to access control at
the NAS based on username/passphrase pairs. In reality, PSTN dialup
connections have never been impervious to a determined adversary.
The availability of widespread broadband access, in concert with the
desire to reduce the cost of PSTN toll access, have driven the
development of Internet-based remote access mechanisms. In some
cases, PPP-based tunneling mechanisms have been used to provide
remote access by allowing the dial user to first access a local ISP
account, and then tunnel an additional PPP connection over the
Internet into the target network. In the case of broadband users,
such connections are tunneled directly over the Internet. While
these mechanisms have been lacking in terms of security features, the
increasing availability of IPsec renders it possible to provide more
secure remote access to the remote resources via the Internet.
Remote access via the Internet has numerous benefits, financial and
otherwise. However, security is paramount, and this presents strong
incentives for migration from the old dial-up model to a more secure
IPsec-based remote access model. Meeting the security requirements
of various classes of remote access users presents a number of
challenges. It is the aim of this document to explore and enumerate
the requirements of various IPsec remote access scenarios, without
suggesting particular solutions for them.
The keywords MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD,
SHOULD NOT, RECOMMENDED, MAY, and OPTIONAL, when they appear in this
document, are to be interpreted as described in [3].
Reader familiarity with RFCs 2401-2412 is a minimum prerequisite to
understanding the concepts discussed here. Familiarity with concepts
relating to Public Key Infrastructures (PKIs) is also necessary.
Familiarity with RADIUS, PPP, PPTP, L2F, L2TP, and other remote
access support protocols will also be helpful, though not strictly
necessary.
Kelly & Ramamoorthi Informational [Page 3]
RFC 3457 IPsec Remote Access Scenarios January 2003
o Remote Access - this term is used to refer to the case in which
the remote user does not necessarily reside at a fixed location,
i.e., in which the user's IP address is not fixed, and therefore
usually not known prior to connection establishment.
o Secure Remote Access - this term refers to remote access which is
secured using elements of the IPsec protocol suite.
o IPsec Remote Access Client (IRAC)- this term is used to refer to
the remote access user's system.
o IPsec Remote Access Server (IRAS) - this term refers to the device
providing access to the target network. An alternative term is
"Security Gateway".
o Security GateWay (SGW) - this refers to the device providing
access to the target network. An alternative term is IRAS.
o Virtual IP address (VIP) - this term describes an address from a
subnet local to the target network which is assigned to a remote
client, giving the appearance that the remote client actually
resides on the target network.
o Machine-Level Authentication - this term describes the case where
the identity of a machine is verified by virtue of the machine's
possession and application of some combination of authenticators.
For a more complete definition, see section 2.
o User-Level Authentication - this term describes the case where the
identity of a user (as opposed to that of a machine) is verified
by virtue of the user's possession and application of some
combination of authenticators. For a more complete definition,
see section 2.
o NAPT - Network Address/Port Translation
This document, while initially intended to simply outline
requirements for various remote access scenarios, has come to include
somewhat more than this. This document has evolved from discussion
within the IPsec Remote Access (IPSRA) working group. As a result,
it in some respects documents the evolution of this thought process.
While this represents a departure from the typical form of a
Kelly & Ramamoorthi Informational [Page 4]
RFC 3457 IPsec Remote Access Scenarios January 2003
requirements document, the associated historical context should prove
useful in interpreting the conclusions reached in the IPSRA working
group.
The balance of this document is organized as follows: First, there is
a general overview of the basic requirements categories, including
definitions relevant to these categories. Following this is a
section devoted to each remote access scenario. Within each of these
sections there are subsections detailing requirements specific to
that scenario in each of the following areas: endpoint
authentication, remote host configuration, policy configuration,
auditing, and intermediary traversal.
In a very general sense, all secure remote access scenarios have a
similar high-level appearance:
target network
|
| +---+
+-------------+ +-----------+ |---| |
|remote access| Internet | security | | +---+
| client |=============| gateway |--|
| (IRAC) | |(SGW/IRAS) | | +---+
+-------------+ +-----------+ |---| |
| +---+
In all cases, a remote client wishes to securely access resources
either behind a SGW or on an IPsec-protected host, and/or wishes to
provide other (specific) systems with secure access to the client's
own resources. There are numerous details which may differ,
depending on the particular scenario. For example, the IRAC may be
within another corporate network, or connected to an ISP via dialup,
DSL, or CATV media. There may be additional intermediaries between
the remote client and the security gateway, but ultimately, all of
these configurations may be viewed somewhat equivalently from a high
level.
In general, there are several basic categories of requirements
relevant to secure remote access scenarios, including endpoint
authentication, remote host configuration, security policy
configuration, auditing, and intermediary traversal. Endpoint
authentication refers to verification of the identities of the
communication partners (e.g., the IRAC and the IRAS). Remote host
configuration refers to the device configuration parameters of the
IRAC system. Security policy configuration refers to IPsec policy
configuration of both the security gateway and the remote host, and
Kelly & Ramamoorthi Informational [Page 5]
RFC 3457 IPsec Remote Access Scenarios January 2003
might also be termed "access control and authorization
configuration". Auditing refers to the generation and collection of
connection status information which is required for the purpose of
maintaining the overall security and integrity of the connected
networks. Intermediary traversal refers to the ability to pass
secured traffic across intermediaries, some of which may modify the
packets in some manner. Such intermediaries include NAPT and
firewall devices. These various categories are treated in more
detail below.
Before discussing endpoint authentication with respect to remote
access, it is important to distinguish between data source
authentication and end user authentication. Data source
authentication in the IPsec context consists in providing assurance
that a network packet originates from a specific endpoint, typically
a user, host, or application. IPsec offers mechanisms for this via
AH or ESP. End user authentication within the IPsec context consists
in providing assurance that the endpoint is what or who it claims to
be. IPsec currently offers mechanisms for this as part of IKE
[IKE].
While the two types of authentication differ, they are not unrelated.
In fact, data source authentication relies upon endpoint
authentication, because it is possible to inject packets with a
particular IP address into the Internet from many arbitrary
locations. In many instances, we cannot be certain that a packet
actually originates from a particular host, or even from the network
upon which that host resides. To resolve this, one must first
authenticate the particular endpoint somehow, and then bind the
addressing information (e.g., IP address, protocol, port) of this
endpoint into the trust relationship established by the
authentication process.
In the context of secure remote access, the authenticated entity may
be a machine, a user (application), or both. The authentication
methods currently supported by IPsec range from preshared secrets to
various signature and encryption schemes employing private keys and
their corresponding public key certificates. These mechanisms may be
used to authenticate the end user alone, the device alone, or both
the end user and the device. These are each discussed in more detail
below.
Kelly & Ramamoorthi Informational [Page 6]
RFC 3457 IPsec Remote Access Scenarios January 2003
In the case where no user input is required in order for an
authentication credential to be used, the entity authenticated will
primarily be the device in which the credential is stored and the
level of derived assurance regarding this authentication is directly
related to how securely the machine's credential is maintained during
both storage and use. That is, a shared secret or a private key
corresponding to a public key certificate may be either stored within
the device or contained in another device which is securely
accessible by the device (e.g., a smartcard). If the knowledge
required for the use of such authentication credentials is entirely
contained within the subject device (i.e., no user input is
required), then it is problematic to state that such credential usage
authenticates anything other than the subject device.
In some cases, a user may be required to satisfy certain criteria
prior to being given access to stored credentials. In such cases,
the level of user authentication provided by the use of such
credentials is somewhat difficult to derive. If sufficiently strong
access controls exist for the system housing the credential, then
there may be a strong binding between the authorized system user and
the credential. However, at the time the credential is presented,
the IRAS itself has no such assurance. That is, the IRAS in
isolation may have some level of assurance that a particular device
(the one in which the credential resides) is the one from which
access is being attempted, but there is no explicit assurance
regarding the identity of the user of the system. In order for the
IRAS to derive additional assurance regarding the user identity, an
additional user credential of some sort would be required. This is
discussed further below.
In some cases, the user may possess an authentication token
(preshared key, private key, passphrase, etc.), and may provide this
or some derivative of this whenever authentication is required. If
this token or derivative is delivered directly to the other endpoint
without modification by the IRAC system, and if the IRAC system
provides no further credentials of its own, then it is the user alone
which has been authenticated. That is, while there may be some
assurance as to the network address from which the user is
originating packets, there is no assurance as to the particular
machine from which the user is attempting access.
Kelly & Ramamoorthi Informational [Page 7]
RFC 3457 IPsec Remote Access Scenarios January 2003
To authenticate both the user and the system, user input of some sort
is required in addition to a credential which is securely stored upon
the device. In some cases, such user input may be used in order to
"complete" the credential stored on the device (e.g., a private key
is password-encrypted), while in others the user's input is supplied
independently of the stored credential. In the case where the
passphrase is applied to the credential prior to use, the level of
assurance derived from successful application of the credential
varies according to your viewpoint.
From the perspective of a system consisting of user, IRAC, IRAS, and
a collection of system protections and security procedures, it may be
said that the user has been authenticated to an extent which depends
upon the strength of the security procedures and system protections
which are in place. However, from the perspective of the IRAS alone,
there is little assurance with respect to user identity. That is,
schemes requiring that stored credentials be modified by user input
prior to use may only be said to provide user-level authentication
within the context of the larger system, and then, the level of
assurance derived is directly proportional to the weakest security
attribute of the entire system.
When considering remote access from a general perspective,
assumptions regarding the overall system are liable to prove
incorrect. This is because the IRAS and the IRAC may not be within
the same domain of control; extranet scenarios are a good example of
this. Hence, the most desirable joint user/machine authentication
mechanisms in this context are those which provide a high level of
assurance to both the IRAS and the IRAC, independently of the larger
system of which the user, IRAS, and IRAC are a part.
In the general case for remote access, authentication requirements
are typically asymmetric. From the IRAC's perspective, it is
important to ensure that the IRAS at the other end of the connection
is indeed what it seems to be, and not some rogue system masquerading
as the SGW. That is, the IRAC requires machine-level authentication
for the IRAS. This is fairly straightforward, given the
authentication mechanisms supported by IKE and IPsec. Further, this
sort of authentication tends to persist through time, although the
extent of this persistence depends upon the mechanism chosen.
While machine-level authentication for the IRAS is sufficient, this
is not the case for the IRAC. Here, it is often important to know
that the entity at the other end of the connection is one who is
Kelly & Ramamoorthi Informational [Page 8]
RFC 3457 IPsec Remote Access Scenarios January 2003
authorized to access local resources rather than someone who happened
upon an unoccupied but otherwise authorized system, or a malicious
Trojan horse application on that user's system, or some other
unauthorized entity. Authenticating the user presents different
requirements than authenticating the user's machine; this requires
some form of user input, and often the authentication must be
periodically renewed.
In situations where a high level of physical security does not exist,
it is common to require a user-input secret as part of the
authentication process, and then to periodically renew the
authentication. Furthermore, since such circumstances may include
the possibility of the presence of a Trojan horse application on the
IRAC system, one-time passphrase mechanisms are often advisable.
Choosing passphrase mechanisms and renewal intervals which provide an
acceptable level of risk, but which do not annoy the user too much,
may be challenging. It should be obvious that even this approach
offers limited assurance in many cases.
Clearly, there are variable assurance levels which are attainable
with the various endpoint authentication techniques, and none of the
techniques discussed offer absolute assurance. Also, there are
variations in the authentication requirements among different remote
access scenarios. This means there is no "cookie cutter" solution
for this problem, and that individual scenarios must be carefully
examined in order to derive specific requirements for each. These
are examined on a case by case basis below in the detailed scenario
descriptions.
There are a number of currently deployed remote access mechanisms
which were installed prior to the deployment of IPsec. Typically,
these are dialup systems which rely upon RADIUS for user
authentication and accounting, but there are other mechanisms as
well. An ideal IPsec remote access solution might utilize the
components of the underlying framework without modification.
Inasmuch as this is possible, this should be a goal. However, there
may be cases where this simply cannot be accomplished, due to
security and/or other considerations. In such cases, the IPsec
remote access framework should be designed to accommodate migration
from these mechanisms as painlessly as is possible.
In general, proposed IPsec remote access mechanisms should meet the
following goals:
o should provide direct support for legacy user authentication
and accounting systems such as RADIUS
Kelly & Ramamoorthi Informational [Page 9]
RFC 3457 IPsec Remote Access Scenarios January 2003
o should encourage migration from existing low-entropy
password-based systems to more secure authentication systems
o if legacy user authentication support cannot be provided
without some sort of migration, the impact of such migration
should be minimized
o user authentication information must be protected against
eavesdropping and replay (including the user identity)
o single sign-on capability should be provided in configurations
employing load-balancing and/or redundancy
o n-factor authentication mechanisms should be supported
Remote host configuration refers to the network-related device
configuration of the client system. This configuration may be fixed
or dynamic. It may be completely provided by the administrator of
the network upon which the remote user currently resides (e.g., the
ISP), or it may be partially provided by that administrator, with the
balance provided by an entity on the remote corporate network which
the client is accessing. In general, this configuration may include
the following:
o IP address(es)
o Subnet mask(s)
o Broadcast address(es)
o Host name
o Domain name
o Time offset
o Servers (e.g., SMTP, POP, WWW, DNS/NIS, LPR,
syslog, WINS, NTP, etc. )
o Router(s)
o Router discovery options
o Static routes
o MTU
o Default TTL
o Source routing options
o IP Forwarding enable/disable
o PMTU options
o ARP cache timeout
o X Windows options
o NIS options
o NetBIOS options
o Vendor-specific options
o (other options)
Kelly & Ramamoorthi Informational [Page 10]
RFC 3457 IPsec Remote Access Scenarios January 2003
Cases where such configuration is fixed are uninteresting; it is the
cases where specific IRAC configuration occurs as a result of remote
access with which we are concerned. For example, in some cases the
IRAC may be assigned a "virtual address", giving the appearance that
it resides on the target network:
target net
+------------------+ |
| Remote Access | +--------+ | ( ~ ~ ~ ~ ~ )
|+-------+ Client | | | | ( IRAC )
||virtual| | |security| |~~~( virtual )
|| host | |--------|gateway | | ( presence )
|| |<================>| |----| ~ ~ ~ ~ ~
|+-------+ |--------| | |
+------------------+ ^ +--------+ | +--------+
| |---| local |
IPsec tunnel | | host |
with encapsulated | +--------+
traffic inside
In this case, the IRAC system begins with an externally routable
address. An additional target network address is assigned to the
IRAC, and packets containing this assigned address are encapsulated,
with the outer headers containing the IRAC's routable address, and
forwarded to the IRAS through the tunnel. This provides the IRAC
with a virtual presence on the target network via an IPsec tunnel.
Note that the IRAC now has two active addresses: the ISP-assigned
address, and the VIP.
Having obtained this virtual presence on the corporate network, the
IRAC may now require other sorts of topology-related configuration,
e.g., default routers, DNS server(s), etc., just as a dynamically
configured host which physically resides upon the target network
would. It is this sort of configuration with which this requirements
category is concerned.
Security policy configuration refers to IPsec access policies for
both the remote access client and the security gateway. It may be
desirable to configure access policies on connecting IRAC systems
which will protect the target network. For example, since a client
has access to the Internet (via its routable address), other systems
on the Internet also have some level of reciprocal access to the
client. In some cases, it may be desirable to block this Internet
Kelly & Ramamoorthi Informational [Page 11]
RFC 3457 IPsec Remote Access Scenarios January 2003
access (or force it to pass through the tunnel) while the client has
a tunneled connection to the target network. This is a matter of
client security policy configuration.
For the security gateway, it may also be desirable to dynamically
adjust policies based upon the user with which a connection has been
established. For example, say there are two remote users, named
Alice and Bob. We wish to provide Alice with unrestricted access to
the target network, while we wish to restrict Bob's access to
specific segments. One way to accomplish this would be to statically
assign internal "virtual" addresses to each user in a one-to-one
mapping, so that each user always has the same address. Then, a
particular user's access could be controlled via policies based upon
the particular address. However, this does not scale well.
A more scalable solution for remote client access control would be to
dynamically assign IP addresses from a specific pool based upon the
authenticated endpoint identity, with access to specific resources
controlled by address-based policies in the SGW. This is very
similar to the static mapping described above, except that a given
group of users (those with identical access controls) would share a
given pool of IP addresses (those which are granted the required
access), rather than a given user always mapping to a given address.
However, this also has scaling issues, though not as pronounced as
for the static mapping.
Alternatively, an arbitrary address could be assigned to a user, with
the security gateway's policy being dynamically updated based upon
the identity of the remote client (and its assigned virtual address)
to permit access to particular resources. In these cases, the
relevant security policy configuration is specific to the IRAS,
rather than to the IRAC. Both IRAS and IRAC security policy
configuration are encompassed by this requirements category.
Auditing is used here to refer to the collection and reporting of
connection status information by the IRAS, for the purpose of
maintaining the security and integrity of the IRAS protected network.
For remote access, the following auditing information is useful from
a security perspective:
o connection start time
o connection end time
Note that the requirement for a connection-end-time attribute implies
the need for a connection heartbeat mechanism of some sort so that
the IRAS can accurately determine this quantity in cases where the
Kelly & Ramamoorthi Informational [Page 12]
RFC 3457 IPsec Remote Access Scenarios January 2003
IRAC does not explicitly terminate the connection. Also note that
the heartbeat mechanism in this case is always directed from the IRAC
to the IRAS.
In some cases, use of a heartbeat may negatively influence a
connection. For example, if the heartbeat interval is very short,
and the connection is reset after loss of very few heartbeat packets,
there is a possibility that network congestion could lead to
unnecessary connection resets. The heartbeat interval and reset
threshold should be chosen with this in mind, and it should be
possible to adjust these quantities either through configuration or
negotiation.
Intermediary traversal is used here to refer to passing a secured
data stream through an intermediary such as a firewall or NAPT
device. In the case of firewalls, numerous deployed products do not
recognize the IPsec protocol suite, making it difficult (sometimes
impossible) to configure them to pass it through. In such cases, a
mechanism is required for making the data stream appear to be of a
type which the firewall is capable of managing.
In the case of NAPT devices, there are a number of issues with
attempting to pass an encrypted or authenticated data stream. For
example, NAPT devices typically modify the source IP address and
UDP/TCP port of outgoing packets, and the destination IP address and
UDP/TCP port of incoming packets, and in some cases, they modify
additional fields in the data portion of the packet. Such
modifications render the use of the AH protocol impossible. In the
case of ESP, the UDP/TCP port fields are sometimes unreadable and
always unmodifiable, making meaningful translation by the NAPT device
impossible. There are numerous other protocol-field combinations
which suffer similarly. This requirements category is concerned with
these issues.
There are numerous remote access scenarios possible using IPsec.
This section contains a brief summary enumeration of these, followed
by a subsection devoted to each which explores the various
requirements in terms of the categories defined above.
The following scenarios are discussed:
o dialup/dsl/cablemodem telecommuters using their systems to access
remote resources
Kelly & Ramamoorthi Informational [Page 13]
RFC 3457 IPsec Remote Access Scenarios January 2003
o extranet users using local corporate systems to access the remote
company network of a business partner
o extranet users using their own system within another company's
network to access their home corporate network
o extranet users using a business partner's system (located on that
partner's network) to access their home corporate network
o remote users using a borrowed system (e.g., an airport kiosk) to
access target network resources
The telecommuter scenario is one of the more common remote access
scenarios. The convenience and wide availability of Internet access
makes this an attractive option under many circumstances. Users may
access the Internet from the comfort of their homes or hotel rooms,
and using this Internet connection, access the resources of a target
network. In some cases, dialup accounts are used to provide the
initial Internet access, while in others some type of "always-on"
connection such as a DSL or CATV modem is used.
The dialup and always-on cases are very similar, with two significant
differences: address assignment mechanism and connection duration.
In most dialup cases, the IRAC's IP address is dynamically assigned
as part of connection setup, and with fairly high likelihood, it is
different each time the IRAC connects. DSL/CATV users, on the other
hand, often have static IP addresses assigned to them, although
dynamic assignment is on the increase. As for connection duration,
dialup remote access connections are typically short-lived, while
always-on connections may maintain remote access connections for
significantly longer periods of time.
The general configuration in either case looks like this:
corporate net
| +----+
+-----+ +-----+ /---/ Internet +---+ |--| |
|IRAC |---|modem|------|ISP|==========|SGW|--| +----+
+-----+ +-----+ /---/ +---+ |
|
An alternative to this configuration entails placing a security
gateway between the user's system and the modem, in which case this
added SGW becomes the IRAC. This is currently most common in cases
where DSL/CATV connections are used.
Kelly & Ramamoorthi Informational [Page 14]
RFC 3457 IPsec Remote Access Scenarios January 2003
The authentication requirements of this scenario depend in part upon
the general security requirements of the network to which access is
to be provided. Assuming that the corporate SGW is physically
secure, machine authentication for the SGW is sufficient. If this
assumption regarding physical security is incorrect, it is not clear
that stronger authentication for the SGW could be guaranteed, and
derivation of an effective mechanism for that case is beyond the
scope of this document.
For the IRAC, there are numerous threats to the integrity of the user
authentication process. Due to the open nature of common consumer
operating systems, some of these threats are quite difficult to
protect against. For example, it is very difficult to assert, with
any level of certainty, that a single user system which permits the
downloading and running of arbitrary applications from the Internet
has not been compromised, and that a covert application is not
monitoring and interacting with the user's data at any point in time.
However, there are 2 general threats we might realistically hope to
somehow mitigate with appropriate authentication mechanisms if we can
assume that the system has not been compromised in this manner.
First, there is the possibility that a secure connection is
established for a particular user, but that someone other than the
intended user is currently using that connection. Second, there is
the possibility that the user's credential (password, hardware token,
etc.) has been somehow compromised, and is being used by someone
other than the authorized user to gain access.
Mitigation of the first threat, the possibility that someone other
than the authorized user is currently using the connection, requires
periodic renewal of user authentication. It should be clear that
machine authentication will not suffice in this case, and that
requiring periodic re-entry of an unchanging user password (which may
be written on a post-it note which is stuck to the user's monitor)
will have limited effectiveness. Convincing verification of the
continued presence of the authorized user will, in many cases,
require periodic application of a time-variant credential.
Mitigation of the second threat, credential compromise, is difficult,
and depends upon a number of factors. If the IRAC system is running
a highly secure operating system, then a time-variant credential may
again offer some value. A static password is clearly deficient in
this scenario, since it may be subject to either online or offline
guessing, and eventually compromised - which is the threat we are
attempting to mitigate. However, if the IRAC operating system is not
Kelly & Ramamoorthi Informational [Page 15]
RFC 3457 IPsec Remote Access Scenarios January 2003
hardened, the use of a time-variant credential is only effective if
simultaneous access from more than one location is forbidden, and if
the credential generation mechanism is not easily compromised.
A second approach to the credential compromise problem entails using
a PKI-based credential which is stored within a secure container of
some sort, and which requires some user interaction prior to
operation (e.g., a smartcard). If such a credential requires
periodic user interaction to continue operating (e.g., pin re-entry),
this may help to limit the access of an unauthorized user who happens
upon a connected but unattended systems. However, choosing an
acceptable refresh interval is a difficult problem, and if the pin is
not
time-variant, this provides limited additional assurance.
To summarize, the following are the authentication requirements for
the IRAS and IRAC:
IRAS
----
o machine authentication MUST be provided.
IRAC
----
o support for user authentication SHOULD be provided
o support for either user or machine authentication MUST be provided
o support for user authentication MUST be provided if protection
from unauthorized connection use is desired.
o if user authentication is provided for short-lived dialup
connections, periodic renewal MAY occur
o if user authentication is provided for always-on connections,
periodic renewal SHOULD occur
There are 2 possibilities for device configuration in the
telecommuter scenario: either access to the target network is
permitted for the native ISP-assigned address of the telecommuter's
system, or the telecommuter's system is assigned a virtual address
from within the target address space. In the first case, there are
no device configuration requirements which are not already satisfied
by the ISP. However, this case is the exception, rather than the
rule.
The second case is far more common, due to the numerous benefits
derived by providing the IRAC with a virtual presence on the target
Kelly & Ramamoorthi Informational [Page 16]
RFC 3457 IPsec Remote Access Scenarios January 2003
network. For example, the virtual presence allows the client to
receive subnet broadcasts, which permits it to use WINS on the target
network. In addition, if the IRAC tunnels all traffic to the target
network, then the target policy can be applied to Internet traffic
to/from the IRAC.
In this case, the IRAC requires, at minimum, assignment of an IP
address from the target network. Typically, the IRAC requires
anywhere from several more to many more elements of configuration
information, depending upon the corporate network's level of
topological complexity. For a fairly complete list, see section 2.2.
To summarize, the following are the device configuration requirements
for the IRAC:
o support for a virtual IP (VIP) address MAY be provided
o if VIP support is provided, support for all device-related
parameters listed in section 2.2 above SHOULD be provided
o support for address assignment based upon authenticated
identity MAY be provided
o if authenticated address assignment is not supported, an
identity-based dynamic policy update mechanism such as is
described in [ARCH] MUST be supported.
In terms of IRAC policy configuration, the most important issue
pertains to whether the IRAC has direct Internet access enabled (for
browsing, etc.) while a connection to the target network exists.
This is important since the fact that the IRAC has access to sites on
the Internet implies that those sites have some level of reciprocal
access to the IRAC. It may be desirable to completely eliminate this
type of access while a tunnel is active.
Alternatively, the risks may be mitigated somewhat by forcing all
Internet-bound packets leaving the IRAC to first traverse the tunnel
to the target network, where they may be subjected to target network
policy. A second approach which carries a bit less overhead entails
modifying the IRAC's policy configuration to reflect that of the
target network during the time the IRAC is connected. In this case,
traffic is not forced to loop through the target site prior to
exiting or entering the IRAC. This requires some sort of policy
download (or modification) capability as part of the SA establishment
process. A third approach is to provide a configuration variable for
the IRAC which permits specification of "tunnel-all", or "block all
traffic not destined for the target network while the SA is up".
Kelly & Ramamoorthi Informational [Page 17]
RFC 3457 IPsec Remote Access Scenarios January 2003
In terms of IRAS configuration, it may be necessary to dynamically
update the security policy database (SPD) when the remote user
connects. This is because transit selectors must be based upon
network address parameters, but these cannot be known a priori in the
remote access case. As is noted above, this may be avoided by
provision of a mechanism which permits address assignment based upon
authenticated identity.
To summarize, the following are the policy configuration requirements
for the IRAS and IRAC:
IRAS
----
o dynamic policy update mechanism based upon identity and
assigned address MAY be supported.
o if address assignment-based policy update mechanism is not
supported, address assignment based upon authenticated identity
SHOULD be supported.
IRAC
----
o IRAC SHOULD provide ability to configure for "tunnel-all"
and/or "block-all" for traffic not destined for the remote
network to which IPsec remote access is being provided.
o support for dynamic IRAS update of IRAC policy MAY be provided.
For telecommuter sessions, session start/end times must be collected.
Reliable derivation of session end time requires that the IRAC
somehow periodically signify that the connection remains active.
This is implied if the IRAS receives data from the IRAC over the
connection, but in cases where no data is sent for some period of
time, a signaling mechanism is required by which the IRAC indicates
that the connection remains in use.
If the address assigned by the ISP to the IRAC system is globally
routable, and no intermediate devices between the IRAC and the IRAS
perform NAPT operations on the data stream, then there are no
additional requirements. If NAPT operations are performed on the
data stream, some mechanism must be provided in order to render these
modifications transparent to the IPsec implementation.
Kelly & Ramamoorthi Informational [Page 18]
RFC 3457 IPsec Remote Access Scenarios January 2003
Extranets are becoming increasingly common, especially as IPsec
becomes more widely deployed. In this scenario, a user from one
corporation uses a local corporate system to access resources on
another corporation's network. Typically, these corporations are
cooperating on some level, but not to the degree that unbridled
access between the two networks would be acceptable. Hence, this
scenario is characterized by limited access. The general topological
appearance is similar to this:
CORP A CORP B
| |
+----+ | | +-----+
|USER|---| |--| S1 |
+----+ | +------++ ++------+ | +-----+
|---|SGW/FW||===Internet===||SGW/FW|---|
| +------++ ++------+ | +-----+
| SGW-A SGW-B |--| S2 |
| | +-----+
This is purposely simplified in order to illustrate some basic
characteristics without getting bogged down in details. At the edge
of each network is a combination security gateway and firewall
device. These are labeled "SGW-A" and "SGW-B". In this diagram,
corporation B wishes to provide a user from corporation A with access
to servers S1 and/or S2. This may be accomplished in one of several
different ways:
1) an end-to-end SA is formed from USER to S1 or S2
2) a tunnel-mode SA is formed between SGW-A and SGW-B which only
permits traffic between S1/S2 and USER.
3) a tunnel-mode SA is formed between USER and SGW-B which only
permits traffic between S1/S2 and USER.
These various cases are individually discussed with respect to each
requirements category below.
For the corporate extranet scenario, the authentication requirements
vary slightly depending upon the manner in which the connection is
accomplished. If only a particular user is permitted to access
S1/S2, then user-level authentication is required. If connection
types (1) or (3) are used, this may be accomplished in the same
manner as it would be for a telecommuter. If connection type (2) is
Kelly & Ramamoorthi Informational [Page 19]
RFC 3457 IPsec Remote Access Scenarios January 2003
used, one of two things must occur: either SGW-A must provide some
local mechanism for authenticating USER and SGW-B must trust this
mechanism, or SGW-B must have some mechanism for authenticating USER
independently of SGW-A.
If access is permitted for anyone within corporation A, then machine
authentication will suffice. However, this is highly unlikely. A
slightly more likely situation might be one in which access is
permitted to anyone within a particular organizational unit in
corporation A. This case is very similar the single user access case
discussed above, and essentially has the same requirements in terms
of the mechanism required for SGW-A, although machine authentication
might suffice if the organizational unit which is permitted access
has a sufficient level of physical security. Again, this requires
that corporation B trust corporation A in this regard.
To summarize, the following are the authentication requirements for
the IRAS and IRAC:
IRAS
----
o machine authentication MUST be provided.
IRAC
----
o support for either user or machine authentication MUST be
provided
o support for a combination of user and machine authentication
SHOULD be provided
o if user authentication is used, periodic renewal SHOULD occur
It is possible that corporation B would want to assign a virtual
address to USER for the duration of the connection. The only way
this could be accomplished would be if USER were a tunnel endpoint
(e.g., in cases (1) and (3)). It is not clear what benefits, if any,
this would offer.
To summarize, the following are the device configuration requirements
for the IRAC:
o support for a virtual address MAY be provided
o if VIP support is provided, support for all device-related
parameters listed in section 2.2 above SHOULD be supported
Kelly & Ramamoorthi Informational [Page 20]
RFC 3457 IPsec Remote Access Scenarios January 2003
o support for address assignment based upon authenticated
identity SHOULD be supported
o if authenticated address assignment is not supported, an
identity-based dynamic policy update mechanism such as is
described in [ARCH] MUST be supported.
Any of the cases discussed above would present some static policy
configuration requirements. Case (1) would require that SGW-A and
SGW-B permit IPsec traffic to pass between USER and S1/S2. Case (3)
would have similar requirements, except that the IPsec traffic would
be between USER and SGW-B. Case (2) would require that the
appropriate transit traffic be secured between USER and S1/S2.
None of these cases require dynamic policy configuration.
For cases (1) and (3), session start/end times must be collected.
Reliable derivation of session end time requires that the IRAC
somehow periodically signify that the connection remains active.
This is implied if the IRAS receives data from the IRAC over the
connection, but in cases where no data is sent for some period of
time, a signaling mechanism is required by which the IRAC indicates
that the connection remains in use.
For case (2), the type(s) of required auditing data would depend upon
whether traffic from multiple users were aggregated within a single
tunnel or not. If so, the notion of individual connection start/stop
times would be lost. If such measures are desired, this requires
that per-user tunnels be set up between SGW-A and SGW-B, and that
some sort of timeout interval be used to cause tunnel teardown when
traffic does not flow for some interval of time.
If the address assigned by the host network to the IRAC system is
globally routable, and no intermediate devices between the IRAC and
the IRAS perform NAPT operations on the data stream, then there are
no additional requirements in this regard. If NAPT operations are
performed on the data stream, some mechanism must be provided in
order to render these modifications transparent to the IPsec
implementation.
If a firewall situated at the edge of the host network cannot be
configured to pass protocols in the IPsec suite, then some mechanism
must be provided which converts the data stream to one which the
Kelly & Ramamoorthi Informational [Page 21]
RFC 3457 IPsec Remote Access Scenarios January 2003
firewall may be configured to pass. If the firewall can be
configured to pass IPsec protocols, then this must be accomplished
prior to connection establishment.
The use of a laptop while visiting another corporation presents
another increasingly common extranet scenario. In this case, a user
works temporarily within another corporation, perhaps as part of a
service agreement of some sort. The user brings along a CORP-A
laptop which is assigned a CORP-B address either statically or
dynamically, and the user wishes to securely access resources on
CORP-A's network using this laptop. This scenario has the following
appearance:
CORP A CORP B
| |
+----+ | | +--------+
|POP |---| |--| CORP-A |
+----+ | +------++ ++------+ | | laptop |
|---|SGW/FW||===Internet===||SGW/FW|---| +--------+
| +------++ ++------+ |
+----+ | SGW-A SGW-B |
|FTP |---| |
+----+ | |
This is very similar to the telecommuter scenario, but it differs in
several important ways. First, in this case there is often a SGW
and/or firewall at the edge of CORP-B's site. Second, there may be a
significantly increased risk that a long-lived connection could
become accessible to someone other than the intended user.
In most cases, the only acceptable connections from CORP-A's
perspective are between the laptop and either SGW-A or the CORP-A
servers the laptop wishes to access. Most of the considerations
applied to the telecommuter also apply here, and user-level
authentication is required to provide assurance that the user who
initiated the connection is still the active user. As an added
precaution, a combination of user-level and machine-level
authentication may be warranted in some cases. Further, in either
case this authentication should be renewed frequently.
Kelly & Ramamoorthi Informational [Page 22]
RFC 3457 IPsec Remote Access Scenarios January 2003
To summarize, the following are the authentication requirements for
the IRAS and IRAC:
IRAS
----
o machine authentication MUST be provided.
IRAC
----
o support for machine authentication SHOULD be provided
o support for user authentication MUST be provided
o support for a combination of user and machine authentication
SHOULD be provided
o periodic renewal of user authentication MUST occur
The device configuration requirements in this scenario are the same
as for the telecommuter, i.e., the laptop may be assigned a virtual
presence on the corporate network, and if so, will require full
infrastructure configuration.
To summarize, the following are the device configuration requirements
for the IRAC:
o support for a virtual address MAY be provided
o if VIP support is provided, support for all device-related
parameters listed in section 2.2 above SHOULD be supported
o support for address assignment based upon authenticated
identity SHOULD be supported
o if authenticated address assignment is not supported, an
identity-based dynamic policy update mechanism such as is
described in [ARCH] MUST be supported.
The policy configuration requirements in this scenario differ from
those of the telecommuter, in that the laptop cannot be assigned a
policy which requires all traffic to be forwarded to CORP-A via the
tunnel. This is due to the fact that the laptop has a CORP-B
address, and as such, may have traffic destined to CORP-B. If this
traffic were tunneled to CORP-A, there might be no return path to
CORP-B except via the laptop. On the other hand, Internet-bound
traffic could be subjected to this restriction if desired, and/or all
traffic other than that between CORP-A and the laptop could be
blocked for the duration of the connection.
Kelly & Ramamoorthi Informational [Page 23]
RFC 3457 IPsec Remote Access Scenarios January 2003
IRAC
----
o support for IRAS update of IRAC policy MAY be provided.
o if IRAS update of IRAC policy is not supported, IRAC MAY
support IRAS directives to "block-all" for non-tunneled
traffic.
o IRAC SHOULD provide ability to configure for "tunnel-all"
and/or "block-all" for traffic not destined for the remote
network to which IPsec remote access is being provided.
The auditing requirements in this scenario are the same as for the
telecommuter scenario. Session start/end times must be collected.
Reliable derivation of session end time requires that the IRAC
somehow periodically signify that the connection remains active.
This is implied if the IRAS receives data from the IRAC over the
connection, but in cases where no data is sent for some period of
time, a signaling mechanism is required by which the IRAC indicates
that the connection remains in use.
If the address assigned by the host network to the IRAC system is
globally routable, and no intermediate devices between the IRAC and
the IRAS perform NAPT operations on the data stream, then there are
no additional requirements in this regard. If NAPT operations are
performed on the data stream, some mechanism must be provided in
order to render these modifications transparent to the IPsec
implementation.
If a firewall situated at the edge of the host network cannot be
configured to pass protocols in the IPsec suite, then some mechanism
must be provided which converts the data stream to one which the
firewall may be configured to pass. If the firewall can be
configured to pass IPsec protocols, then this must be accomplished
prior to connection establishment.
Kelly & Ramamoorthi Informational [Page 24]
RFC 3457 IPsec Remote Access Scenarios January 2003
This is very similar to the extranet laptop scenario discussed above,
except that a higher degree of trust for CORP-B is required by
CORP-A. This scenario has the following appearance:
CORP A CORP B
| |
+----+ | | +--------+
|POP |---| |--| CORP-B |
+----+ | +------++ ++------+ | |desktop |
|---|SGW/FW||===Internet===||SGW/FW|---| +--------+
| +------++ ++------+ |
+----+ | SGW-A SGW-B |
|FTP |---| |
+----+ | |
The authentication requirements for the desktop extranet scenario are
very similar to those of the extranet laptop scenario discussed
above. The primary difference lies in the authentication type which
may be used, i.e., in the laptop case, CORP-A can derive some
assurance that the connection is coming from one of CORP-A's systems
if a securely stored machine credential is stored on and used by on
the laptop. In the desktop case this is not possible, since CORP-A
does not own the IRAC system.
To summarize, the following are the authentication requirements for
the IRAS and IRAC:
IRAS
----
o machine authentication MUST be provided.
IRAC
----
o support for machine authentication MAY be provided
o support for user authentication MUST be provided
o support for a combination of user and machine authentication
MAY be provided
o periodic renewal of user authentication MUST occur
Kelly & Ramamoorthi Informational [Page 25]
RFC 3457 IPsec Remote Access Scenarios January 2003
The device configuration requirements in this scenario are the same
as for the laptop extranet scenario, i.e., the desktop system may be
assigned a virtual presence on the corporate network, and if so, will
require full infrastructure configuration. However, this seems less
likely than in the laptop scenario, given CORP-A's lack of control
over the software configuration of CORP-B's desktop system.
The policy configuration requirements are quite similar to those of
the extranet laptop, except that in this scenario there is even less
control over CORP-B's desktop than there would be over the laptop.
This means it may not be possible to restrict traffic in any way at
the desktop system.
The auditing requirements in this scenario are the same as for the
telecommuter scenario. Session start/end times must be collected.
Reliable derivation of session end time requires that the IRAC
somehow periodically signify that the connection remains active.
This is implied if the IRAS receives data from the IRAC over the
connection, but in cases where no data is sent for some period of
time, a signaling mechanism is required by which the IRAC indicates
that the connection remains in use.
If the address assigned by the host network to the IRAC system is
globally routable, and no intermediate devices between the IRAC and
the IRAS perform NAPT operations on the data stream, then there are
no additional requirements in this regard. If NAPT operations are
performed on the data stream, some mechanism must be provided in
order to render these modifications transparent to the IPsec
implementation.
If a firewall situated at the edge of the host network cannot be
configured to pass protocols in the IPsec suite, then some mechanism
must be provided which converts the data stream to one which the
firewall may be configured to pass. If the firewall can be
configured to pass IPsec protocols, then this must be accomplished
prior to connection establishment.
Kelly & Ramamoorthi Informational [Page 26]
RFC 3457 IPsec Remote Access Scenarios January 2003
This scenario entails a traveling user connecting to the target
network using a public system owned by someone else. A commonly
cited example is an airport kiosk. This looks very similar to the
extranet desktop scenario, except that in the extranet scenario,
CORP-A might have a trust relationship with CORP-B, whereas in this
scenario, CORP-A may not trust a publicly accessible system. Note
that a trust relationship between CORP-A and the owner of the public
system may exist, but in many cases will not.
There are two variations to this scenario. In the first, no trust
relationship exists between the target network and the borrowed
system. In the second, some trust relationship does exist. In the
case where no trust relationship exists, machine authentication is
out of the question, as it is meaningless in this context. Further,
since such a system could easily capture a passphrase, use of a
static passphrase from such a system would seem to be ill-advised.
If a one-time passphrase were used, this would mitigate the risk of
passphrase capture by the public system. On the other hand, if it is
acknowledged that such capture is a real threat (i.e., the system
itself is malicious), then it must also be recognized that any data
transmitted and received via the resulting session would not be
confidential or reliable with respect to this malicious system, and
that the system could not be trusted to have actually disconnected
when the user walks away. This suggests that accessing non-trivial
information from such a system would be imprudent.
Another possible user authentication option would be a smartcard.
However, many smartcards require a pin or passphrase to "unlock"
them, which requires some level of trust in the kiosk to not record
the pin. Hence, this approach suffers from drawbacks similar to
those of the static passphrase in this regard. The primary
difference would be that the pin/passphrase could not be used alone
for access in the smartcard case.
In cases where a trust relationship with the owner of the public
system exists, the trust level would modulate the risk levels
discussed above. For example, if a sufficient level of trust for the
system owner exists, use of a static passphrase might present no more
risk than if this were permitted from a system owned by the accessed
target. However, the primary benefit of such a trust relationship
would be derived from the ability to authenticate the machine from
Kelly & Ramamoorthi Informational [Page 27]
RFC 3457 IPsec Remote Access Scenarios January 2003
which the user is attempting access. For example, a security policy
requiring that remote access only be permitted with combined
user/machine authentication might be effected, with further control
regarding which machines were allowed.
An additional issue to be dealt with in either case pertains to
verification of the identity of the IRAS. If the IRAC were to be
misdirected somehow, a man in the middle attack could be effected,
with the obtained password being then used for malicious access to
the true IRAS. Note that even a one-time password mechanism offers
little protection in this case. In order to avert such an attack,
the IRAC must possess some certifiable or secret knowledge of the
IRAS prior to attempting to connect. Note that in the case where no
trust relationship exists, this is not possible.
To summarize, the following are the authentication requirements for
the IRAS and IRAC:
IRAS
----
o machine authentication MUST be provided.
IRAC
----
o in cases where no trust relationship exists between the
accessed network and the system owner, sensitive data SHOULD
NOT be transmitted in either direction.
o in cases where a trust relationship exists between the accessed
network and the system owner, machine authentication SHOULD be
supported.
o in cases where a trust relationship exists between the accessed
network and the system owner, a static passphrase MAY be used
in conjunction with machine-level authentication of the IRAC
system.
o frequent renewal of user authentication MUST occur
The auditing requirements in this scenario are the same as for the
telecommuter scenario. Session start/end times must be collected.
Reliable derivation of session end time requires that the IRAC
somehow periodically signify that the connection remains active.
This is implied if the IRAS receives data from the IRAC over the
connection, but in cases where no data is sent for some period of
time, a signaling mechanism is required by which the IRAC indicates
that the connection remains in use.
If the address of the IRAC system is globally routable, and no
intermediate devices between the IRAC and the IRAS perform NAPT
operations on the data stream, then there are no additional
requirements in this regard. If NAPT operations are performed on the
data stream, some mechanism must be provided in order to render these
modifications transparent to the IPsec implementation.
As we examine the various remote access scenarios, a general set of
common requirements emerge. Following is a summary:
o Support for user authentication is required in almost all
scenarios
o Machine authentication for the IRAS is required in all scenarios
o A mechanism for providing device configuration for the IRAC is
required in most scenarios. Such a mechanism must be extensible.
o Machine authentication for IRAC is generally only useful when
combined with user authentication. Combined user and machine
authentication is useful in some scenarios.
o Dynamic IRAC policy configuration is useful in several scenarios.
o Most scenarios require auditing for session start/stop times.
o An intermediary traversal mechanism may be required in any of the
scenarios.
Kelly & Ramamoorthi Informational [Page 29]
RFC 3457 IPsec Remote Access Scenarios January 2003
[ARCH] Kent, S. and R. Atkinson, "Security Architecture for the
Internet Protocol", RFC 2401, November 1998.
[KEYWORDS] Bradner, S., "Key Words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RADIUS] Rigney, C., Rubens, A., Simpson, W. and S. Willens,
"Remote Authentication Dial In User Service (RADIUS)",
RFC 2865, June 2000.
[IKE] Harkins, D. and D. Carrel, "The Internet Key Exchange
(IKE)", RFC 2409, November 1998.
The editors would like to acknowledge the many helpful comments of
Sara Bitan, Steve Kent, Mark Townsley, Bernard Aboba, Mike Horn, and
other members of the ipsra working group who have made helpful
comments on this work.
Scott Kelly
Airespace
110 Nortech Pkwy
San Jose CA 95134 USA
Phone: +1 (408) 941-0500
EMail: scott@hyperthought.com
Sankar Ramamoorthi
Juniper Networks
1194 North Mathilda Ave
Sunnyvale CA 94089-1206 USA
Phone: +1 (408) 936-2630
EMail: sankarr@juniper.net
Kelly & Ramamoorthi Informational [Page 30]
RFC 3457 IPsec Remote Access Scenarios January 2003
Copyright (C) The Internet Society (2003). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.
Kelly & Ramamoorthi Informational [Page 31]