Network Working Group R. White
Internet-Draft Cisco Systems
Expires: November 20, 2005 May 19, 2005
Architecture and Deployment Considerations for Secure Origin BGP (soBGP)
draft-white-sobgp-architecture-01
Status of this Memo
By submitting this Internet-Draft, each author represents that any
applicable patent or other IPR claims of which he or she is aware
have been or will be disclosed, and any of which he or she becomes
aware will be disclosed, in accordance with Section 6 of BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet-
Drafts.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html.
This Internet-Draft will expire on November 20, 2005.
Copyright Notice
Copyright (C) The Internet Society (2005).
Abstract
There is a great deal of concern over the security of internetworks
built using the Border Gateway Protocol to provide routing
information to autonomous systems connected to the internetwork.
This draft provides an architecture for a secure distributed registry
of routing information to address these concerns. The draft begins
with an overview of the operation of this system, and then follows
with various deployment scenarios, starting with what we believe will
be the most common deployment option.
White Expires November 20, 2005 [Page 1]
Internet-Draft soBGP Architecture & Deployment May 2005
1. Motivation
2. Overview
There are two fundamental pieces of a routing system that need to be
secured:
o Adjacencies between devices running the routing protocol.
o Information carried within the routing protocol.
While security between [BGP] speakers has been addressed in a number
of ways, including cryptographic authentication [BGP-MD5] and
limiting the attack radius through TTL mechanisms [GTSH], security
for the information carried within BGP is not considered a solved
problem.
This draft proposes a possible solution to securing the information
within BGP, using the certificates and protocol extensions proposed
in [SOBGP-BGPTRANSPORT], [SOBGP-CERTIFICATE], and [SOBGP-RADIUS].
There are three major modifications to this draft since the last
edition was published
o An additional short section has been added on the impact of soBGP
on internetwork convergence. This is a rough cut at an analysis
of this work, at this point.
o An additional section on incremental deployment, in terms of
deploying parts of the soBGP system to solve specific problems.
o A section on aggregation considerations has been added.
3. General Theory
soBGP provides a secure registry mechanism against which a BGP
speaker can check:
o The authorization of the AS listed as the originating AS in any
received update to advertise reachability to the prefix listed in
the update.
o The validity of the AS Path contained in the update.
A valid AS Path, in this document, is a path that has the following
attributes:
o Each autonomous system listed in the AS Path is an actual
participant in the internetwork.
o Each pair of autonomous systems listed in the AS Path are actually
interconnected.
White Expires November 20, 2005 [Page 2]
Internet-Draft soBGP Architecture & Deployment May 2005
o Starting from the first autonomous system (the origin AS), and
passing through each autonomous system listed in the AS Path,
actually results in reaching the advertising peer's AS.
As shown in [PATH-CONSIDER], it isn't possible to verify an AS more
than one AS hop away has authorized the advertisement of specific
reachability information based on the AS Path. The concept of
policy, and soBGP's interaction with policy, is considered more fully
in a later section in this draft.
soBGP operates by distributing a set of signed certificates,
described in [SOBGP-CERTIFICATE], containing the information required
to validate the two pieces of information given above. These
certificates MAY be distributed using the mechanisms described in
[SOBGP-BGPTRANSPORT], or some other mechanism. Once these
certificates have been received and processed (signatures validated,
etc, as described in [SOBGP-CERTIFICATE], they form a database
containing:
o A listing of IP address blocks and the AS authorized to originate
them.
o Policies related to specific prefixes and blocks of addresses.
o A list of autonomous systems connected to each autonomous system
within the internetwork. This connection list is used to build a
graph of AS interconnectivity within the internetwork, as
described in the section Building the AS Connectivity Graph,
below.
This effectively forms a secure registry of routing information which
can be used to check the validity of routing information received
from BGP peers. This database is termed the "authorization
database." No assumption about the location of the authorization
database is made within this document.
As BGP updates are processed, a security preference is assigned to
each prefix, as described further in the Security Preference section
of this document. BGP update processing is described in the
Receiving and Processing Updates section of this document.
4. soBGP Operation
Each section below provides detailed information on some aspect of
soBGP operation.
4.1 Building the AS Connectivity Graph
Each ASPolicycert advertised by a member of the internetwork contains
a list of the autonomous systems the advertising AS is connected to,
White Expires November 20, 2005 [Page 3]
Internet-Draft soBGP Architecture & Deployment May 2005
along with possible policy information about that connection. From
this information, a graph of AS connectivity within the internetwork
is built.
Any AS can be used as the starting point for building this graph,
thus multiple disconnected graphs (representing section of the
internetwork running soBGP and providing interconnection information)
are possible. If every AS within the internetwork is providing
interconnection information, one graph can be built containing all
the internetwork's interconnections.
The process of creating this graph is:
o Begin with the local AS, or any AS for which an ASPolicycert is
available.
o Examine the list of connected autonomous systems advertised by the
current AS.
o Examine the ASPolicycert of each AS the current AS is advertising
as connected, and determine if that AS is advertising a connection
back to the current AS. This is termed the two way connectivity
check.
o If the two way connectivity check passes, the connection SHOULD be
added to the interconnection graph, and marked as trustable.
o If the two way connectivity check fails, the connection MAY be
added to the interconnection graph, but marked so a lower security
preference will be assigned to routes containing this AS pair in
their AS Path.
o Apply any policies indicated by either of the two autonomous
systems in their ASPolicycert. This could include, for instance,
noting the connected autonomous system MUST NOT be used for
transiting traffic.
o Repeat this process for each ASPolicycert in the authorization
database.
The resulting graph is called the internetwork graph.
4.2 Validating Routing Information (The Security Preference)
soBGP provides a two tier evaluation of routes. In the first stage,
a BGP speaker evaluating received routing information would discard
all routing information found to be false, or not accurately
representing the internetwork as it exists. Routing information not
meeting this criteria SHOULD be discarded, as indicated in the
processing steps outlined below.
In the second tier, the BGP speaker assigns a Security Preference to
the received routing information, indicating a locally significant
trust level determined by examining the received routing information.
White Expires November 20, 2005 [Page 4]
Internet-Draft soBGP Architecture & Deployment May 2005
The amount by which the Security Preference is increased or decreased
for any operation described in this draft is locally significant to
the autonomous system. This allows the operator provide a finer
granularity of security policy, from dropping routing information
deemed invalid through simply preferring routes the operator deems
"more secure."
The operator MAY configure a lower bound. Routes with Security
Preferences under this lower bound SHOULD be discarded. Any of the
following methods may be used to implement the Security Preference
within an autonomous system:
o Assign the value of the Security Preference to any of the
attributes used in the [BGP] decision process. Care must be taken
with attributes for which the lower value is preferred.
o Use a Cost Community [COST] and its associated methods to consider
the Security Preference at any step in the Decision Process [BGP]
without overloading other attributes. Care must be taken as the
lowest value in a Cost Community is preferred.
Several basic rules apply to all BGP speakers either evaluating the
security level of received routing information, or using the Security
Preference to determine which path to install in the local RIB:
o The method selected to implement the Security Preference MUST be
consistent through the local autonomous system.
o All devices processing routes against soBGP information MUST use
the same mechanisms and values of the Security Preference to
ensure consistent routing within the autonomous system.
o The Security Preference value may be used to select among
different routes for the same prefix; the higher value MUST be
preferred.
The process described below does not rule out additional policies
added locally, or in some future draft. For each route (prefix/
attribute pair) within a given BGP UPDATE message:
o The local authorization database is examined, and the Authcert
with the longest prefix length encompassing the range of addresses
described by the prefix is chosen.
o If there is no entry in the local authorization database which
encompasses the range of addresses described by the prefix, then
the route is said to be unverified. The Security Preference
SHOULD be set to a level indicated by local policy.
o If there is an AS_SET in the AS_PATH, the following process MAY be
followed:
White Expires November 20, 2005 [Page 5]
Internet-Draft soBGP Architecture & Deployment May 2005
* For each AS in the AS_SET, examine the set of PrefixPolicycerts
advertised by that AS.
* If a PrefixPolicycert is found authorizing at least one of the
autonomous systems in the AS_SET to advertise some component of
the prefix, the Security Preference MAY be increased or left at
its current value.
* If a PrefixPolicycert is not found authorizing at least one of
the autonomous systems in the AS_SET to advertise some
component of the prefix, the Security Preference MAY be
decreased or left at its current value.
* If a path exists from the aggregator to each AS listed in the
AS_SET, the Security Preference MAY be increased or left at its
current value.
* If a path does not exist from the aggregator to each AS listed
in the AS_SET, the Security Preference MAY be decreased or left
at its current value.
o If there is an AS_SET in the AS_PATH, it is disregarded in all
further processing. The originator of the AS is considered the
originator of the route.
o The second hop in the AS_PATH attribute is examined.
o
* If the second hop in the AS_PATH is advertised as connected by
the originating AS, the Security Preference for this prefix
SHOULD be increased.
* If the second hop in the AS_PATH is not advertised as connected
by the originating AS, the Security Preference for this prefix
SHOULD be decreased.
* If the second hop in the AS_PATH is not advertised as connected
by the originating AS and the originator's policy indicates the
second hop MUST be validated, the prefix SHOULD be removed from
further consideration.
o The AS_PATH attribute is compared to the internetwork graph.
o
* If a series of two way verified pairwise peerings exists,
beginning with the first AS listed in the AS_PATH, and ending
in the advertising AS, the Security Preference SHOULD be
increased.
* If a series of pairwise peerings exists, beginning with the
first AS listed in the AS_PATH, and ending in the advertising
AS, the Security Preference MAY be increased. This case allows
for the inclusion of one-way advertised AS interconnections in
the graph.
* If the AS_PATH described is not contained within the
internetwork graph, and the originator indicated the AS_PATH
MUST be checked, the prefix SHOULD be removed from further
consideration.
White Expires November 20, 2005 [Page 6]
Internet-Draft soBGP Architecture & Deployment May 2005
* Otherwise, the Security Preference SHOULD be decreased.
o The Authcert chosen at the first step is examined.
o
* If the authorized AS in the Authcert matches the originating AS
in the AS_PATH, the Security Preference SHOULD be increased.
* If the authorized AS in the Authcert does not match the
originating AS in the AS_PATH, the prefix SHOULD be removed
from further consideration.
4.3 Validating Received BGP UPDATES
As BGP UPDATES are received, they MAY be processed at one of several
points:
o Each prefix may be validated according to the process outlined in
Validating Routing Information before they are installed in the
ADJ-RIB-IN.
o Each prefix may be validated according to the process outlined in
Validating Routing Information after they are installed in the
ADJ-RIB-IN, but before they are considered in the BGP Best Path
calculation.
o Each prefix may be validated according to the process outlined in
Validating Routing Information after they are run through the Best
Path algorithm, but before they are installed in the local RIB.
o Routes may be installed in the local RIB, and then validated using
the process outlined in Validating Routing Information. Once
validation is accomplished, adjustments to the local RIB and
routes advertised to BGP peers may need to be adjusted.
4.4 Requirements for Systems Running soBGP
This section describes requirements for autonomous systems running
soBGP, requirements for BGP speakers forming external adjacencies
from within such autonomous systems, and devices exchanging soBGP
certificates.
o Any peering session along the border of an autonomous system
running soBGP SHOULD be authenticated through some means such as
[BGP-MD5], IPsec ([ESP], [AH]), or through some other current,
effective means of protecting BGP sessions from being hijacked, or
otherwise abused.
o Any peering session along which soBGP certificates are exchanged
SHOULD be authenticated through some means such as [BGP-MD5],
IPsec ([ESP, [AH]), or through some other current, effective means
of protecting BGP sessions from being hijacked, or otherwise
abused.
White Expires November 20, 2005 [Page 7]
Internet-Draft soBGP Architecture & Deployment May 2005
o For each received route, the last (most recently added) autonomous
system MUST be compared to the autonomous system of the BGP
speaker advertising the route. If the last (most recently added)
AS in the AS Path does not match the autonomous system of the
transmitting speaker, the route MUST be discarded.
When soBGP is supported, a BGP speaker MUST have access to the
authorization database. Possible methods of access include:
o Have a local copy of this authorization database, and perform the
checks described later in this document against that local
database.
o Pass received routing information to a locally maintained server
for validation against that server's copy of the authorization
database [SOBGP-RADIUS].
o Accept filters built from a copy of the authorization database
contained on a locally maintained server.
5. soBGP Deployment
This section begins by describing what we believe to be the most
practical deployment of this secure registry of routing information.
Following sections describe some other deployment options that may
prove useful in some situations, or may prove to be more practical
than the deployment outlined in this section.
5.1 Deploying soBGP on Distributed Registry Servers
This deployment scenario works within three constraints:
o It may not be not desirable to combine routing and cryptographic
processing of soBGP certificates on the same device.
o The system should be distributed, using as few centralized
resources as possible.
o Trust relationships should be based on existing business and
working relationships, rather than building new relationships
specifically for securing the routing system.
Assume we have a small internetwork, as shown below:
S1 - - - - - - - - - - -S2 - - - -S3
10.1.1.0/24---A---B-----C---D-----E---F
| AS65000 | AS65001 | AS65002
In this network, we assume each AS has an soBGP server locally within
their AS, marked as S1, S2, and S3, above. These servers are
interconnected in a way similar to eBGP peering between AS65000,
AS65001, and AS65002; S1 and S2 are using the mechanisms described in
[SOBGP-BGPTRANSPORT] to distribute the certificates described in
White Expires November 20, 2005 [Page 8]
Internet-Draft soBGP Architecture & Deployment May 2005
[SOBGP-CERTIFICATE] between them.
Each server then processes the certificates as described in [SOBGP-
CERTIFICATE], and either provides a set of filters or a mechanism
through which the eBGP peering routers can authenticate routing
information, such as described in [SOBGP-RADIUS]. This deployment
technique provides BGP route validation that is:
o Fully Distributed: Local server (or set of servers) which builds
the required databases based on received certificates, and
distributes certificates throughout the routing system.
o Locally Controlled: Each local server (or set of server) is
maintained and managed by autonomous systems participating in the
internetwork.
o Based on Existing Business Relationships: Peering autonomous
systems also peer their soBGP servers, so the system uses existing
business relationships to provide the deployment and long term
maintenance of the system.
o Very Little Impact on the Existing Routing System: The current
processing and distribution of routing information through [BGP]
isn't impacted in any way. The only additional requirements on
existing equipment are to compare the routing information to the
database results provided by the local servers (i.e., receiving
and processing filter lists, or through [SOBGP-RADIUS]).
5.2 Certificate Processing on Edge Peering Routers
soBGP can also be deployed entirely within BGP speakers at the edge
of an Autonomous System (AS).
+-(eBGP)-+ +-(eBGP)-+
| | | |
v v v V
A--------B-----C-----D--------E
^ ^
| |
+--(iBGP)---+
In this network, A is sending certificates it has learned from other
sources to B using the mechanisms described in [SOBGP-BGPTRANSPORT].
B is passing these certificates to D via iBGP, and D is passing these
certificates to E via eBGP. Each edge router, B and D, process these
certificates locally, building the databases required to validate
received routing information from them.
White Expires November 20, 2005 [Page 9]
Internet-Draft soBGP Architecture & Deployment May 2005
5.3 Multihoming Deployment
Multihoming presents a special challenge to the deployment of soBGP
within a large scale internetwork.
(---------) (---------)
( AS65401 ) ( AS65402 )
( ) ( )
( ) ( )
(---A---) (---B---)
| |
\ /
\-----+ +-----/
| |
(--C------D--)
( )
( No-AS )
(----------)
Assume No-AS has obtained a block of addresses, 10.1.1.0/24, from
AS65401, and would like to advertise that same block of addresses
through AS65402. Since No-AS has no AS number, it cannot generate
any soBGP certificates, and must rely on its upstream providers to
work out the security impact in some way. The simplest solution
would be, of course, for NOAS to obtain an AS number, and fully
participate in soBGP, but barring that, what other solutions are
there?
AS65401 could issue a certificate allowing AS65402 to originate just
the prefix in question, 10.1.1.0/24, or AS65401 could simply list
AS65402 in the certificate covering 10.1.1.0/24 as an authorized
originator for this address space (as multiple authorized originators
are allowed).
5.4 Proxy Advertisement of Certificates
Note there is no requirement for a given entity which originates
routes into the routing system to actually originate the
corresponding certificates required for the correct origination of
the route to be validated, and the AS Path attached to the route to
be verified.
White Expires November 20, 2005 [Page 10]
Internet-Draft soBGP Architecture & Deployment May 2005
(-----------------)
( Other Third Party )
(---------------)
/ \
/ \
(---------) (---------)
( AS65401 ) ( AS65402 )
( ) ( )
( ) ( )
(---A---) (---B---)
| |
\ /
\-----+ +-----/
| |
(--C------D--)
( )
( AS65403 )
(----------)
In this case, AS65401, AS65402, or some other third part may actually
advertise the certificates necessary for AS65403 to originate
validated routes.
6. Other Considerations
In this section, we move from specific deployment scenarios to other
deployment considerations, such as key generation and protection, and
memory utilization/impact.
6.1 Certificate Generation and Private Key Protection
There is only one private/public key pair per autonomous system;
certificates are generated as determined by local policy and as
required to account for changes in the network. Since the entity's
private key is not used in any part of the operations verifying
received information, or in generating information to transmit to
other devices, these certificates could be generated on some secure
central system in the AS, and the results, containing only public
keys, can be transmitted throughout the network.
Securing the private key of each entity should be relatively easy in
this environment, since the location of the private key can be
carefully constrained; no device other than the system which
generates the required certificates needs use of the private key.
6.2 Impact on Performance and Memory Utilization
Detailed performance and memory utilization characteristics of soBGP
White Expires November 20, 2005 [Page 11]
Internet-Draft soBGP Architecture & Deployment May 2005
will be the subject of future investigation. However, as this is an
important area of consideration, we present some suggested analysis
below. (In other words, this is a guess).
In terms of memory, each device running soBGP will need to store:
o Each of the Entitycerts Received. The maximum number of
Entitycerts within the routing system would be the number
participating autonomous systems multiplied by the number of
outstanding Entitycerts from each autonomous system.
o Each of the ASPolicycerts Received. The number of ASPolicycerts
within the system will probably be similar to the number of
Entitycerts within the system.
o Each of the PrefixPolicycerts Received. The number of
PrefixPolicycerts within the system will depend on the number of
address blocks each participant in the routing system advertises,
and could double during key rollover.
Performance will depend on the cryptographic processing requirements
imposed by the certificate signature methods, as described in [SOBGP-
CERTIFICATE]. However, all of this additional memory and processing
would most likely be required on a distributed soBGP server, rather
than on routers themselves.
The primary impact on routers and routing protocol convergence will
be the memory and processing requirements added from the additional
route filters or processing as required by the deployment technique
used.
6.3 soBGP Impact on Internetwork Convergence
We generally assume that adding a security infrastructure on top of
an operating system will dramatically decrease the performance of
that system. However, much depends on the system being modified,
itself, and how closely to perfectly efficient that system already
performs. We've already examined, in prior sections, the impact of
soBGP on memory and processor utilization in devices running these
extensions, but we've not examined the impact of soBGP on another
aspect of an internetwork's operation, convergence time. In this
section, we will examine some possible side effects of deploying
soBGP using the following small internetwork as an example.
+--B--C--D--+
| |
A---E---F---G---K
| |
+-----H-----+
In this network, assume that:
White Expires November 20, 2005 [Page 12]
Internet-Draft soBGP Architecture & Deployment May 2005
o A prefers the path through {H,G} to K.
o E prefers the path through {F,G} to K.
o B prefers the path through {C,D,G} to K.
In this network, if the link from G to K fails:
o A will first receive a withdraw from H, and begin to prefer the
path through {E,F,G} to K.
o A will then receive a withdraw from E, and begin to prefer the
path through {C,D,E,G} to K.
o A will finally receive a withdraw from C, and remove the route to
K from its local tables.
This processing pattern is well documented through multiple studies
in the operations of [BGP] in large scale internetworks. The most
obvious answer to resolve this problem is for G to include some sort
of information in its withdraw indicating the nature of the failure,
so A can directly remove all paths through the link {G,K} on
receiving the first withdraw. This is more problematic than it
appears, however, because [BGP] is designed for protocol efficiency,
and withdraws are often removed from the internetwork, along with any
information they might contain, at an early point in the convergence
process.
The mechanism soBGP uses to build a graph of the interconnections
between the autonomous systems in the internetwork, however, provide
another place where this sort of direct information about changes in
the topology of the internetwork can be distributed. If this network
were running soBGP, G would be able to reissue its certificate
claiming connectivity to K, or use some specific policy indicator to
note the link {G,K} has failed. On receiving this certificate, all
the autonomous systems could remove all routes with the link {G,K} in
their AS Paths, and the network would converge with much less
distribution and processing of routing information.
We believe there are probably several performance enhancements that
may be gained through the laying of a connectivity graph on top of
the current [BGP] provided view of an internetwork. These types of
efficiency gains may overcome or fully offset the added costs of
deploying soBGP as a security system.
6.4 Aggregation
Aggregation is a diificult problem within any system attempting to
validate routes in an internetwork running BGP. The primary purpose
of aggregation is to remove information from the routing system, and
information removed from the system cannot be validated or verified.
This appears to be a simple observation, but it has a number of far
White Expires November 20, 2005 [Page 13]
Internet-Draft soBGP Architecture & Deployment May 2005
reaching impacts.
( AS1 ) (AS4) (AS5)
10.1.0.0/24----A-----+
|
( AS2 ) | (AS5)
10.1.1.0/24----------B----C
|
( AS3 ) |
10.1.2.0/24----------+
In this small internetwork, B is aggregating 10.1.0.0/24,
10.1.1.0/24, and 10.1.2.0/24, into a single advertisement,
10.1.0.0/22, towards C, using the mechanism described in [BGP]. C
receives 10.1.0.0/22 with the combined communities of 10.1.0.0/24,
10.1.1.0/24, and 10.1.2.0/24, and an AS Path of {(1,2,3,4),5}. What
information can C learn from this advertisement?
o 10.1.0.0/22 is an aggregate, and AS5 claims it has learned some
component of 10.1.0.0/22 through or from AS1, AS2, AS3, or AS4.
o AS5 claims to be able to reach AS1, AS2, AS3, and AS4.
o AS5 claims it can reach any existing destination within
10.1.0.0/22, given no other longer prefix route to a component of
this address space exists.
o AS5 is authorized to advertise 10.1.0.0/22.
Can C validate any of these pieces of information?
6.4.1 The Route is an Aggregate
C assumes 10.1.0.0/22 is an aggregate because of the AS Set contained
in the AS Path; is there any way to validate this claim? If C has
the PrefixPolicycerts for 10.1.0.0/24, 10.1.1.0/24, and 10.1.2.0/24,
it would be able to determine that some of the components of
10.1.0.0/22 are, in fact, owned by at least some of the autonomous
systems listed in the AS Set. This ability, however, is dependant on
C having at least one PrefixPolicycert from AS1, AS2, or AS3, which
depends on all certificates being flooded throughout the
internetwork, regardless of the flow of routing information. If
certificates are blocked when route aggregation or filtering is
performed at B, C will not have these certificates.
Therefore, we can prove a route is actually an aggregate, if we have
at least one PrefixPolicycert from one of the autonomous systems
listed in the AS Set, and that PrefixPolicycert contains at least
some component of the advertised aggregate. C cannot, however,
definitively prove the route is not an aggregate, because it cannot
know if it simply does not have such a certificate, or if the
aggregate isn't really an aggregate.
White Expires November 20, 2005 [Page 14]
Internet-Draft soBGP Architecture & Deployment May 2005
If it is the policy of the internetwork in which soBGP is deployed to
require all certificates to be flooded to all autonomous systems,
soBGP MAY use this test to verify a route is actually an aggregate.
6.4.2 Reachability to the Members of the AS Set
If C has the ASPolicycerts of AS1, AS2, AS3, AS4, and AS5, its local
connectivity graph of the internetwork is going to include these
autonomous systems. In this case, C could verify AS5 can reach each
of the autonomous systems included in the AS Set. Note, however, that
in highly meshed internetworks, reachability will exist between
almost every pair of nodes through multiple paths, so C cannot prove
AS5 should be advertising the received aggregate along this link, or
peering session.
If it is the policy of the internetwork in which soBGP is deployedto
require all certificates to be flooded to all autonomous systems,
soBGP MAY use this test to verify connectivity between an aggregator
and the autonomous systems listed in the AS_SET of a received
aggregate.
6.4.3 Reachability to All Destinations Within the Aggregate
Can C prove AS5 is able to reach every possible destination with
10.1.0.0/22? Assume C has PrefixPolicycerts from AS1, AS2, AS3, AS4,
and AS5. From these PrefixPolicycerts, C could determine AS5 could
be receiving at least three components of 10.1.0.0/22: 10.1.0.0/24,
10.1.1.0/24, and 10.1.2.0/24. However, there is no way to verify AS5
is actually receiving these components. For instance:
o While AS1 may be authorized to advertise 10.1.0.0/24, and may
normally advertise 10.1.0.0/24 to AS4, the AS1 to AS4 link may be
failed, and therefore AS5 may not be receiving this component.
o AS1 may be authorized to advertise 10.1.3.0/24, but may not have
any hosts reachable in that range of addresses, and therefore
might not be advertising it.
In other words, AS5 may not be receiving components of a given
aggregate route because the component doesn't exist, or is not
reachable through AS5. Neither of these conditions can be verified
by C, so AS5's reachability to all the destinations listed within an
aggregate cannot be verified.
6.4.4 Authorization to Advertise an Aggregate
There are at least two ways authorization to advertise an aggregate
could be defined:
White Expires November 20, 2005 [Page 15]
Internet-Draft soBGP Architecture & Deployment May 2005
o The advertising autonomous system is known to have reachability to
all the components contained within the aggregate, and is
authorized to aggregate the address space by each of the owners of
those components.
o The advertising autonomous system is authorized to directly
advertise the address space indicated in the aggregate route.
The first definition fails on two counts:
o There is no way to prove an aggregating autonomous system has
reachability to every possible component of an advertised
aggregate.
o Given the receiver of an aggregate does not know what the complete
list of reachable components within an aggregate actually is,
there is no way for the receiver of an aggregate to know if all
the component's owners have actually authorized the aggregator to
advertise the aggregate.
This leads to the conclusion that authorization to advertise an
aggregate can only exist in the second sense; that if an autonomous
system has authorization to advertise the address block (because it
has a AuthCert authorizing the advertisement of the address block),
then the autonomous system is authorized to advertize the aggregate.
7. Incremental Deployment of soBGP
One of the primary concerns with any security system is the ability
of users to incrementally deploy the system without impacting current
network operations. As the security system is deployed, it should
provide greater security. In theory, the amount of additional
security offered verses the additional work required should be fairly
balanced.
There are two aspects of incremental deployment that need to be
considered:
o The impact of some of the participants in the system deploying the
security system, but not all participants deploy the system.
o The impact of some part of the system being deployed widely, but
not all of the system.
7.1 Not All Connected Networks Participate
The first consideration in incremental deployment of soBGP is asking
what happens if all of the autonomous systems in an internetwork
don't run soBGP. Is there any advantage to partial deployments of
soBGP in this sense?
Throughout this section, we will assume soBGP certificates are
received by all autonomous systems running soBGP, even if they are
White Expires November 20, 2005 [Page 16]
Internet-Draft soBGP Architecture & Deployment May 2005
separated by multiple hops which are not running soBGP. This is not
an unreasonable assumption, since soBGP certificates can be shared in
multiple ways, including multihop BGP sessions across non-
participating autonomous systems.
Assume we have the following small internetwork, what impact will
incrementally deploying soBGP through this network have?
(AS1) (AS2) (AS3) (AS4 )
A-----B-----C-----D---10.1.1.0/24
Assume AS3 and AS4 deploy soBGP, but not AS1 and AS2; is there any
value? When AS3 receives routes from AS4, it can verify AS4 is
authorized to advertise 10.1.1.0/24. Further, any routes AS3
forwards to AS4 from AS1 or AS2 can be validated, to some degree, by
AS4. The AS Path can be checked to make certain AS2 is actually
connected to AS3 (since AS3 is advertising its connectivity to AS3).
If some route is advertised from AS2 showing an AS hop in the middle
of those two autonomous systems, it can be safely discarded by AS4 as
an invalid AS Path.
We can make an alternate assumption, that AS1 and AS4 have deployed
soBGP, while AS2 and AS3 have not. In this case, what gains would be
made by deploying soBGP? Assume Router A receives a route from
Router B with an AS Path of {B,C,D}. If Router A has access to
Router D's certificates, it can:
o Check the origin AS (the first AS in the AS Path, in this case
AS4) is authorized to advertise the address space (in this case
10.1.1.0/24).
o Check the second hop in the AS Path (in this case AS3) is actually
attached to AS4, as advertised by AS4.
o Since Router A knows it is connected to AS2, through B, it can
also validate the last AS listed in the AS Path.
There is some gain, then, in deploying soBGP in both of these
situations. The gain is obviously more in the second scenario than
the first.
7.2 Deploying Parts of soBGP
The second question concerning incremental deploying is if
implementing some part of soBGP, without the remainder, would be
useful. This question is generally placed in the context of
validating the origination authorization of routes, and possibly the
first hop in the AS Path, but not the entire AS Path.
White Expires November 20, 2005 [Page 17]
Internet-Draft soBGP Architecture & Deployment May 2005
o soBGP Authcerts could be advertised or published (for instance, on
a Web page), to provide authorization for each origin AS to
advertise specific address blocks. These certificates could be
self signed, in the most relaxed case, or signed by the entity
authorizing the AS to advertise the address block.
o soBGP PrefixPolicycerts could be advertised or published (for
instance, on a web page), to provide authorization and first hop
checking for received routes. The Authcert within the
PrefixPolicycert contains the information required to validate the
origin's authorization to originate a route. The list of MAY
TRANSIT autonomous systems contained in the PrefixPolicycert would
provide the ability to check the first hop in the AS Path of any
received route.
o soBGP PrefixPolicycerts and ASPolicycerts could be advertised to
provide authorization to advertise a route from within an address
block, and also provide the ability to validate the first hop in
the AS Path. The Authcert, within the PrefixPolicycert, contains
the information required to validate the origin's authorization to
originate a route. The list of connected autonomous systems
within the ASPolicycert provides the information required to
validate the first hop in the AS Path of any received route.
Any of these modes of operation could be mixed with a full deployment
of soBGP, and provides checks for the first hop and origination of
received routes.
8. Policy Interactions with soBGP
Beyond simply securing the information contained within the routing
database [BGP] builds, it's also desirable to have a secure mechanism
for an autonomous system to advertise policy information. For
instance, an autonomous system may not want a specific peer to
transit traffic, or an originator may want routing information to be
advertised only to a specific number of AS hops away from the origin.
The sections below examine some various policies of this type, and
possible solutions within soBGP.
8.1 Indicating Do Not Transit
In the following small internetwork, A would like to enforce a policy
preventing C from transiting traffic from B to A.
A-------B--------D
| |
+---C---+
A may attempt to prevent C from transiting traffic from B to A by
White Expires November 20, 2005 [Page 18]
Internet-Draft soBGP Architecture & Deployment May 2005
advertising its routing information to C in such a way that C cannot
readvertise that routing information to B. The problem with this
approach is that B must assume the lack of specific routing
information from C indicates A has a local policy forbidding C from
transiting traffic to A. Unfortunately, because of the nature of
address space assignment, aggregation, filtering, and other factors,
B cannot make this assumption. For instance, C may receive a
superset of the routing information A is advertising, and advertise
those routes to B instead, in which case A will find there's no
effective way to enforce its policy towards C.
We find, however, that the interconnection graph laid on top of the
routing information transmitted by each autonomous system provides a
point where A may communicate its nontransit policy towards C
directly to B. Using its ASPolicycert, A may indicate B is not a
transit AS, allowing B to mark routes with the AS pair {B,A} in their
AS Path with a lower security preference, or possibly even discarding
such routing information altogether.
This is a simple application of the policies available in the soBGP
certificates; more complex policies may be expressed through similar
means. The certificates described in [SOBGP-CERTIFICATE] are built
so policies may be added in the future, as well.
9. Acknowledgements
A large number of people contributed to this draft either by
contributing text, ideas, or comments; we've tried to include all of
them here (but might have missed a few): James Ng, Tim Gage, Alvaro
Retana, Dave Cook, Brian Weis, Iljitsch van Beijnum, Bora Akyol, Tony
Li, and Sue Hares.
10. Security Considerations
11. IANA Considerations
12. References
12.1 Normative References
[BGP] Rekhter, Y. and T. Li, "A Border Gateway Protocol 4
(BGP-4)", RFC 1771, March 1995.
[SOBGP-BGPTRANSPORT]
Ng, J., "Extensions to BGP Transport soBGP certificates",
draft-ng-sobgp-bgp-extensions-01.txt (work in progress),
April 2004.
White Expires November 20, 2005 [Page 19]
Internet-Draft soBGP Architecture & Deployment May 2005
[SOBGP-CERTIFICATE]
Weiss, B., "Secure Origin BGP (soBGP) certificates",
draft-weis-sobgp-certificates-01.txt (work in progress),
October 2003.
12.2 Informative References
[COST] Retana, A. and R. White, "BGP Custom Decision Process",
draft-retana-bgp-custom-decision-00.txt (work in
progress), October 2002.
[PATH-CONSIDER]
White, R., "Considerations in Validating the Path in
Routing Protocols", draft-white-pathconsiderations-02.txt
(work in progress), April 2004.
[SOBGP-RADIUS]
Lonvick, C., "RADIUS Attributes for soBGP Support",
draft-lonvick-sobgp-radius-04.txt (work in progress),
February 2004.
Author's Address
Russ White, editor
Cisco Systems
Email: riw@cisco.com
White Expires November 20, 2005 [Page 20]
Internet-Draft soBGP Architecture & Deployment May 2005
Intellectual Property Statement
The IETF takes no position regarding the validity or scope of any
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; nor does it represent that it has
made any independent effort to identify any such rights. Information
on the procedures with respect to rights in RFC documents can be
found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementers or users of this
specification can be obtained from the IETF on-line IPR repository at
http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights that may cover technology that may be required to implement
this standard. Please address the information to the IETF at
ietf-ipr@ietf.org.
Disclaimer of Validity
This document and the information contained herein are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIM 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.
Copyright Statement
Copyright (C) The Internet Society (2005). This document is subject
to the rights, licenses and restrictions contained in BCP 78, and
except as set forth therein, the authors retain all their rights.
Acknowledgment
Funding for the RFC Editor function is currently provided by the
Internet Society.
White Expires November 20, 2005 [Page 21]