Internet Engineering Task Force (IETF) S. Kent
Request for Comments: 7132 BBN
Category: Informational A. Chi
ISSN: 2070-1721 UNC-CH
February 2014 Threat Model for BGP Path Security
This document describes a threat model for the context in which
External Border Gateway Protocol (EBGP) path security mechanisms will
be developed. The threat model includes an analysis of the Resource
Public Key Infrastructure (RPKI) and focuses on the ability of an
Autonomous System (AS) to verify the authenticity of the AS path info
received in a BGP update. We use the term "PATHSEC" to refer to any
BGP path security technology that makes use of the RPKI. PATHSEC
will secure BGP, consistent with the inter-AS security focus of the
The document characterizes classes of potential adversaries that are
considered to be threats and examines classes of attacks that might
be launched against PATHSEC. It does not revisit attacks against
unprotected BGP, as that topic has already been addressed in the
BGP-4 standard. It concludes with a brief discussion of residual
Status of This Memo
This document is not an Internet Standards Track specification; it is
published for informational purposes.
This document is a product of the Internet Engineering Task Force
(IETF). It represents the consensus of the IETF community. It has
received public review and has been approved for publication by the
Internet Engineering Steering Group (IESG). Not all documents
approved by the IESG are a candidate for any level of Internet
Standard; see Section 2 of RFC 5741.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
Copyright (c) 2014 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 22. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 43. Threat Characterization . . . . . . . . . . . . . . . . . . . 64. Attack Characterization . . . . . . . . . . . . . . . . . . . 84.1. Active Wiretapping of Sessions between Routers . . . . . 84.2. Attacks on a BGP Router . . . . . . . . . . . . . . . . . 94.3. Attacks on Network Operator Management Computers (Non-CA
Computers) . . . . . . . . . . . . . . . . . . . . . . . 114.4. Attacks on a Repository Publication Point . . . . . . . . 124.5. Attacks on an RPKI CA . . . . . . . . . . . . . . . . . . 145. Residual Vulnerabilities . . . . . . . . . . . . . . . . . . 166. Security Considerations . . . . . . . . . . . . . . . . . . . 187. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 188. Informative References . . . . . . . . . . . . . . . . . . . 181. Introduction
This document describes the security context in which PATHSEC is
intended to operate. The term "PATHSEC" (for path security) refers
to any design used to preserve the integrity and authenticity of the
AS_PATH attribute carried in a BGP update message [RFC4271]. The
security context used throughout this document is established by the
Secure Inter-Domain Routing (SIDR) working group charter [SIDR-CH].
The charter requires that solutions that afford PATHSEC make use of
the Resource Public Key Infrastructure (RPKI) [RFC6480]. It also
calls for protecting only the information required to verify that a
received route traversed the Autonomous Systems (ASes) in question,
and that the Network Layer Reachability Information (NLRI) in the
route is what was advertised.
Thus, the goal of PATHSEC is to enable a BGP speaker to verify that
the ASes enumerated in this path attribute represent the sequence of
ASes that the NLRI traversed. The term "PATHSEC" is thus consistent
with the goal described above. (Other SIDR documents use the term
"BGPSEC" to refer to a specific design; we avoid use of that term
This document discusses classes of potential adversaries that are
considered to be threats, and classes of attacks that might be
launched against PATHSEC. Because PATHSEC will rely on the RPKI,
threats and attacks against the RPKI are included. This model also
takes into consideration classes of attacks that are enabled by the
use of PATHSEC (e.g., based on use of the RPKI).
The motivation for developing PATHSEC, i.e., residual security
concerns for BGP, is well described in several documents, including
"BGP Security Vulnerabilities Analysis" [RFC4272] and "Design and
Analysis of the Secure Border Gateway Protocol (S-BGP)" [Kent2000].
All of these documents note that BGP does not include mechanisms that
allow an AS to verify the legitimacy and authenticity of BGP route
advertisements. (BGP now mandates support for mechanisms to secure
peer-to-peer communication, i.e., for the links that connect BGP
routers. There are several secure protocol options to address this
security concern, e.g., IPsec [RFC4301] and TCP Authentication Option
(TCP-AO) [RFC5925]. This document briefly notes the need to address
this aspect of BGP security, but focuses on application layer BGP
security issues that must be addressed by PATHSEC.)
RFC 4272 [RFC4272] succinctly notes:
BGP speakers themselves can inject bogus routing information,
either by masquerading as any other legitimate BGP speaker, or by
distributing unauthorized routing information as themselves.
Historically, misconfigured and faulty routers have been
responsible for widespread disruptions in the Internet. The
legitimate BGP peers have the context and information to produce
believable, yet bogus, routing information, and therefore have the
opportunity to cause great damage. The cryptographic protections
of [TCPMD5] and operational protections cannot exclude the bogus
information arising from a legitimate peer. The risk of
disruptions caused by legitimate BGP speakers is real and cannot
PATHSEC is intended to address the concerns cited above, to provide
significantly improved path security, which builds upon the route
origination validation capability offered by use of the RPKI
[RFC6810]. Specifically, the RPKI enables relying parties (RPs) to
determine if the origin AS for a path was authorized to advertise the
prefix contained in a BGP update message. This security feature is
enabled by the use of two types of digitally signed data: a PKI
[RFC6487] that associates one or more prefixes with the public key(s)
of an address space holder, and Route Origin Authorizations (ROAs)
[RFC6482] that allow a prefix holder to specify one or more ASes that
are authorized to originate routes for a prefix.
The security model adopted for PATHSEC does not assume an "oracle"
that can see all of the BGP inputs and outputs associated with every
AS or every BGP router. Instead, the model is based on a local
notion of what constitutes legitimate, authorized behavior by the BGP
routers associated with an AS. This is an AS-centric model of secure
operation, consistent with the AS-centric model that BGP employs for
routing. This model forms the basis for the discussion that follows.
This document begins with a brief set of definitions relevant to the
subsequent sections. It then discusses classes of adversaries that
are perceived as viable threats against routing in the public
Internet. It continues to explore a range of attacks that might be
effected by these adversaries against both path security and the
infrastructure upon which PATHSEC relies. It concludes with a brief
review of residual vulnerabilities, i.e., vulnerabilities that are
not addressed by use of the RPKI and that appear likely to be outside
the scope of PATHSEC mechanisms.
The following security and routing terminology definitions are
employed in this document.
Adversary: An adversary is an entity (e.g., a person or an
organization) that is perceived as malicious, relative to the
security policy of a system. The decision to characterize an
entity as an adversary is made by those responsible for the
security of a system. Often, one describes classes of adversaries
with similar capabilities or motivations rather than specific
individuals or organizations.
Attack: An attack is an action that attempts to violate the security
policy of a system, e.g., by exploiting a vulnerability. There is
often a many-to-one mapping of attacks to vulnerabilities because
many different attacks may be used to exploit a vulnerability.
Autonomous System (AS): An AS is a set of one or more IP networks
operated by a single administrative entity.
AS Number (ASN): An ASN is a 2- or 4-byte number issued by a
registry to identify an AS in BGP.
Certification Authority (CA): An entity that issues digital
certificates (e.g., X.509 certificates) and vouches for the
binding between the data items in a certificate.
Countermeasure: A countermeasure is a procedure or technique that
thwarts an attack, preventing it from being successful. Often,
countermeasures are specific to attacks or classes of attacks.
Border Gateway Protocol (BGP): A path vector protocol used to convey
"reachability" information among ASes in support of inter-domain
False (Route) Origination: If a network operator originates a route
for a prefix that the operator does not hold (and that has not
been authorized to originate by the prefix holder), this is termed
false route origination.
Internet Service Provider (ISP): An organization managing (and
typically selling) Internet services to other organizations or
Internet Number Resources (INRs): IPv4 or IPv6 address space and
Internet Registry: An organization that manages the allocation or
distribution of INRs. This encompasses the Internet Assigned
Number Authority (IANA), Regional Internet Registries (RIRs),
National Internet Registries (NIRs), and Local Internet Registries
(LIRs) (network operators).
Man in the Middle (MITM): A MITM is an entity that is able to
examine and modify traffic between two (or more) parties on a
Network Operator: An entity that manages an AS and thus emits (E)BGP
updates, e.g., an ISP.
Network Operations Center (NOC): A network operator employs a set of
equipment and a staff to manage a network, typically on a 24/7
basis. The equipment and staff are often referred to as the NOC
for the network.
Prefix: A prefix is an IP address and a mask used to specify a set
of addresses that are grouped together for purposes of routing.
Public Key Infrastructure (PKI): A PKI is a collection of hardware,
software, people, policies, and procedures used to create, manage,
distribute, store, and revoke digital certificates.
Relying Parties (RPs): An RP is an entity that makes use of signed
products from a PKI, i.e., it relies on signed data that is
verified using certificates and Certificate Revocation Lists
(CRLs) from a PKI.
RPKI Repository System: The RPKI repository system consists of a
distributed set of loosely synchronized databases.
Resource PKI (RPKI): A PKI operated by the entities that manage INRs
and that issue X.509 certificates (and CRLs) that attest to the
holdings of INRs.
RPKI Signed Object: An RPKI signed object is a data object
encapsulated with Cryptographic Message Syntax (CMS) that complies
with the format and semantics defined in [RFC6488].
Route: In the Internet, a route is a prefix and an associated
sequence of ASNs that indicates a path via which traffic destined
for the prefix can be directed. (The route includes the origin
Route Leak: A route leak is said to occur when AS-A advertises
routes that it has received from AS-B to the neighbors of AS-A,
but AS-A is not viewed as a transit provider for the prefixes in
Threat: A threat is a motivated, capable adversary. An adversary
that is not motivated to launch an attack is not a threat. An
adversary that is motivated but not capable of launching an attack
also is not a threat.
Vulnerability: A vulnerability is a flaw or weakness in a system's
design, implementation, or operation and management that could be
exploited to violate the security policy of a system.
3. Threat Characterization
As noted in Section 2 above, a threat is defined as a motivated,
capable adversary. The following classes of threats represent
classes of adversaries viewed as relevant to this environment.
Network Operators: A network operator may be a threat. An
operator may be motivated to cause BGP routers it controls to emit
update messages with inaccurate routing info, e.g., to cause
traffic to flow via paths that are economically advantageous for
the operator. Such updates might cause traffic to flow via paths
that would otherwise be rejected as less advantageous by other
network operators. Because an operator controls the BGP routers
in its network, it is in a position to modify their operation in
arbitrary ways. Routers managed by a network operator are
vehicles for mounting MITM attacks on both control and data plane
traffic. If an operator participates in the RPKI, it will have at
least one CA resource certificate and may be able to generate an
arbitrary number of subordinate CA certificates and ROAs. It will
be authorized to populate (and may even host) its own repository
publication point. If it implements PATHSEC, and if PATHSEC makes
use of certificates associated with routers or ASes, it will have
the ability to issue such certificates for itself. If PATHSEC
digitally signs updates, it will be able to do so in a fashion
that will be accepted by PATHSEC-enabled neighbors.
Hackers: Hackers are considered a threat. A hacker might assume
control of network management computers and routers controlled by
operators, including operators that implement PATHSEC. In such
cases, hackers would be able to act as rogue network operators
(see above). It is assumed that hackers generally do not have the
capability to effect MITM attacks on most links between networks
(links used to transmit BGP and subscriber traffic). A hacker
might be recruited, without his/her knowledge, by criminals or by
nations, to act on their behalf. Hackers may be motivated by a
desire for "bragging rights", for profit, or to express support
for a cause ("hacktivists" [Sam04]). We view hackers as possibly
distinct from criminals in that the former are presumed to effect
attacks only remotely (not via a physical presence associated with
a target) and not necessarily for monetary gain. Some hackers may
commit criminal acts (depending on the jurisdiction), and thus
there is a potential for overlap between this adversary group and
Criminals: Criminals may be a threat. Criminals might persuade
(via threats or extortion) a network operator to act as a rogue
operator (see above) and thus be able to effect a wide range of
attacks. Criminals might persuade the staff of a
telecommunications provider to enable MITM attacks on links
between routers. Motivations for criminals may include the
ability to extort money from network operators or network operator
clients, e.g., by adversely affecting routing for these network
operators or their clients. Criminals also may wish to manipulate
routing to conceal the sources of spam, DoS attacks, or other
Registries: Any registry in the RPKI could be a threat. Staff at
the registry are capable of manipulating repository content or
mismanaging the RPKI certificates that they issue. These actions
could adversely affect a network operator or a client of a network
operator. The staff could be motivated to do this based on
political pressure from the nation in which the registry operates
(see below) or due to criminal influence (see above).
Nations: A nation may be a threat. A nation may control one or
more network operators that operate in the nation, and thus can
cause them to act as rogue network operators. A nation may have a
technical active wiretapping capability (e.g., within its
territory) that enables it to effect MITM attacks on inter-network
traffic. (This capability may be facilitated by control or
influence over a telecommunications provider operating within the
nation.) It may have an ability to attack and take control of
routers or management network computers of network operators in
other countries. A nation may control a registry (e.g., an RIR)
that operates within its territory and might force that registry
to act in a rogue capacity. National threat motivations include
the desire to control the flow of traffic to/from the nation or to
divert traffic destined for other nations (for passive or active
wiretapping, including DoS).
4. Attack Characterization
This section describes classes of attacks that may be effected
against Internet routing (relative to the context described in
Section 1). Attacks are classified based on the target of the
attack, an element of the routing system, or the routing security
infrastructure on which PATHSEC relies. In general, attacks of
interest are ones that attempt to violate the integrity or
authenticity of BGP traffic or that violate the authorizations
associated with entities participating in the RPKI. Attacks that
violate the implied confidentiality of routing traffic, e.g., passive
wiretapping attacks, are not considered a requirement for BGP
security (see [RFC4272]).
4.1. Active Wiretapping of Sessions between Routers
An adversary may attack the BGP (TCP) session that connects a pair of
BGP speakers. An active attack against a BGP (TCP) session can be
effected by directing traffic to a BGP speaker from some remote
point, or by being positioned as a MITM on the link that carries BGP
session traffic. Remote attacks can be effected by any adversary. A
MITM attack requires access to the link. Modern transport networks
may be as complex as the packet networks that utilize them for inter-
AS links. Thus, these transport networks may present significant
attack surfaces. Nonetheless, only some classes of adversaries are
assumed to be capable of MITM attacks against a BGP session. MITM
attacks may be directed against BGP and PATHSEC-protected BGP, or
against TCP or IP. Such attacks include replay of selected BGP
messages, selective modification of BGP messages, and DoS attacks
against BGP routers. [RFC4272] describes several countermeasures for
such attacks, and thus this document does not further address such
4.2. Attacks on a BGP Router
An adversary may attack a BGP router, whether or not it implements
PATHSEC. Any adversary that controls routers legitimately, or that
can assume control of a router, is assumed to be able to effect the
types of attacks described below. Note that any router behavior that
can be ascribed to a local routing policy decision is not considered
to be an attack. This is because such behavior could be explained as
a result of local policy settings and thus is beyond the scope of
what PATHSEC can detect as unauthorized behavior. Thus, for example,
a router may fail to propagate some or all route withdrawals or
effect "route leaks". (These behaviors are not precluded by the
specification for BGP and might be the result of a local policy that
is not publicly disclosed. As a result, they are not considered
attacks. See Section 5 for additional discussion.)
Attacks on a router are equivalent to active wiretapping attacks (in
the most general sense) that manipulate (forge, tamper with, or
suppress) data contained in BGP updates. The list below illustrates
attacks of this type.
AS Insertion: A router might insert one or more ASNs, other than
its own ASN, into an update message. This violates the BGP spec
and thus is considered an attack.
False (Route) Origination: A router might originate a route for a
prefix when the AS that the router represents is not authorized to
originate routes for that prefix. This is an attack, but it is
addressed by the use of the RPKI [RFC6480].
Secure Path Downgrade: A router might remove AS_PATH data from a
PATHSEC-protected update that it receives when forwarding this
update to a PATHSEC-enabled neighbor. This behavior violates the
PATHSEC security goals and thus is considered an attack.
Invalid AS_PATH Data Insertion: A router might emit a PATHSEC-
protected update with "bad" data (such as a signature), i.e.,
PATHSEC data that cannot be validated by other PATHSEC routers.
Such behavior is assumed to violate the PATHSEC goals and thus is
considered an attack.
Stale Path Announcement: If PATHSEC-secured announcements can
expire, such an announcement may be propagated with PATHSEC data
that is "expired". This behavior would violate the PATHSEC goals
and is considered a type of replay attack.
Premature Path Announcement Expiration: If a PATHSEC-secured
announcement has an associated expiration time, a router might
emit a PATHSEC-secured announcement with an expiry time that is
very short. Unless the PATHSEC protocol specification mandates a
minimum expiry time, this is not an attack. However, if such a
time is mandated, this behavior becomes an attack. BGP speakers
along a path generally cannot determine if an expiry time is
"suspiciously short" since they cannot know how long a route may
have been held by an earlier AS, prior to being released.
MITM Attack: A cryptographic key used for point-to-point security
(e.g., TCP-AO, TLS, or IPsec) between two BGP routers might be
compromised (e.g., by extraction from a router). This would
enable an adversary to effect MITM attacks on the link(s) where
the key is used. Use of specific security mechanisms to protect
inter-router links between ASes is outside the scope of PATHSEC.
Compromised Router Private Key: If PATHSEC mechanisms employ
public key cryptography, e.g., to digitally sign data in an
update, then a private key associated with a router or an AS might
be compromised by an attack against the router. An adversary with
access to this key would be able to generate updates that appear
to have passed through the AS that this router represents. Such
updates might be injected on a link between the compromised router
and its neighbors if that link is accessible to the adversary. If
the adversary controls another network, it could use this key to
forge signatures that appear to come from the AS or router(s) in
question, with some constraints. So, for example, an adversary
that controls another AS could use a compromised router/AS key to
issue PATHSEC-signed data that includes the targeted router/AS.
(Neighbors of the adversary's AS ought not accept a route that
purports to emanate directly from the targeted AS. So, an
adversary could take a legitimate, protected route that passes
through the compromised AS, add itself as the next hop, and then
forward the resulting route to neighbors.)
Withdrawal Suppression Attack: A PATHSEC-protected update may be
signed and announced, and later withdrawn. An adversary
controlling intermediate routers could fail to propagate the
withdrawal. BGP is already vulnerable to behavior of this sort,
so withdrawal suppression is not characterized as an attack under
the assumptions upon which this mode is based (i.e., no oracle).
4.3. Attacks on Network Operator Management Computers (Non-CA
An adversary may choose to attack computers used by a network
operator to manage its network, especially its routers. Such attacks
might be effected by an adversary who has compromised the security of
these computers. This might be effected via remote attacks,
extortion of network operations staff, etc. If an adversary
compromises NOC computers, he can execute any management function
that authorized network operations the staff would have performed.
Thus, the adversary could modify the local routing policy to change
preferences, to black-hole certain routes, etc. This type of
behavior cannot be externally detected as an attack. Externally,
this appears as a form of rogue operator behavior. (Such behavior
might be perceived as accidental or malicious by other operators.)
If a network operator participates in the RPKI, an adversary could
manipulate the RP tools that extract data from the RPKI, causing the
output of these tools to be corrupted in various ways. For example,
an attack of this sort could cause the operator to view valid routes
as not validated, which could alter its routing behavior.
If an adversary invoked the tool used to manage the repository
publication point for this operator, it could delete any objects
stored there (certificates, CRLs, manifests, ROAs, or subordinate CA
certificates). This could affect the routing status of entities that
have allocations/assignments from this network operator (e.g., by
deleting their CA certificates).
An adversary could invoke the tool used to request certificate
revocation, causing router certificates, ROAs, or subordinate CA
certificates to be revoked. An attack of this sort could affect not
only this operator but also any operators that receive allocations/
assignments from it, e.g., because their CA certificates were
If an operator is PATHSEC-enabled, an attack of this sort could cause
the affected operator to be viewed as not PATHSEC-enabled, possibly
making routes it emits less preferable to other operators.
If an adversary invoked a tool used to request ROAs, it could
effectively reallocate some of the prefixes allocated/assigned to the
network operator (e.g., by modifying the origin AS in ROAs). This
might cause other PATHSEC-enabled networks to view the affected
network as no longer originating routes for these prefixes. Multi-
homed subscribers of this operator who received an allocation from
the operator might find that their traffic was routed via other
If the network operator is PATHSEC-enabled, and makes use of
certificates associated with routers/ASes, an adversary could invoke
a tool used to request such certificates. The adversary could then
replace valid certificates for routers/ASes with ones that might be
rejected by PATHSEC-enabled neighbors.
4.4. Attacks on a Repository Publication Point
A critical element of the RPKI is the repository system. An
adversary might attack a repository, or a publication point within a
repository, to adversely affect routing.
This section considers only those attacks that can be launched by any
adversary who controls a computer hosting one or more repository
publication points, without access to the cryptographic keys needed
to generate valid RPKI-signed products. Such attacks might be
effected by an insider or an external threat. Because all repository
objects are digitally signed, attacks of this sort translate into DoS
attacks against the RPKI RPs. There are a few distinct forms of such
attacks, as described below.
Note first that the RPKI calls for RPs to cache the data they acquire
and verify from the repository system [RFC6480][RFC6481]. Attacks
that delete signed products, insert products with "bad" signatures,
tamper with object signatures, or replace newer objects with older
(valid) ones, can be detected by RPs (with a few exceptions). RPs
are expected to make use of local caches. If repository publication
points are unavailable or the retrieved data is corrupted, an RP can
revert to using the cached data. This behavior helps insulate RPs
from the immediate effects of DoS attacks on publication points.
Each RPKI data object has an associated date on which it expires or
is considered stale (certificates expire and CRLs become stale).
When an RP uses cached data, how to deal with stale or expired data
is a local decision. It is common in PKIs to make use of stale
certificate revocation status data when fresher data is not
available. Use of expired certificates is less common, although not
unknown. Each RP will decide, locally, whether to continue to make
use of or ignore cached RPKI objects that are stale or expired.
If an adversary inserts an object into a publication point, and the
object has a "bad" signature, the object will not be accepted and
used by RPs.
If an adversary modifies any signed product at a publication point,
the signature on the product will fail, causing RPs to not accept it.
This is equivalent to deleting the object, in many respects.
If an adversary deletes one or more CA certificates, ROAs, or the CRL
for a publication point, the manifest for that publication point will
allow an RP to detect this attack. An RP can continue to use the
last valid instance of the deleted object (as a local policy option),
thus minimizing the impact of such an attack.
If an adversary deletes a manifest (and does not replace it with an
older instance), RPs are able to detect this action. Such behavior
should result in the CA (or publication point maintainer) being
notified of the problem. An RP can continue to use the last valid
instance of the deleted manifest (a local policy option), thus
minimizing the impact of such an attack.
If an adversary deletes newly added CA certificates or ROAs, and
replaces the current manifest with the previous manifest, the
manifest (and the CRL that it matches) will be "stale" (see
[RFC6486]). This alerts an RP that there may be a problem. The RP
should use the information from a Ghostbuster Record [RFC6493] to
contact the entity responsible for the publication point and request
a remedy to the problem (e.g., republish the missing CA certificates
and/or ROAs). An RP cannot know the content of the new certificates
or ROAs that are not present, but it can continue to use what it has
cached. An attack of this sort will, at least temporarily, cause RPs
to be unaware of the newly published objects. INRs associated with
these objects will be treated as unauthenticated.
If a CA revokes a CA certificate or a ROA (via deleting the
corresponding End Entity (EE) certificate), and the adversary tries
to reinstate that CA certificate or ROA, the adversary would have to
rollback the CRL and the manifest to undo this action by the CA. As
above, this would make the CRL and manifest stale, and this is
detectable by RPs. An RP cannot know which CA certificates or ROAs
were deleted. Depending on local policy, the RP might use the cached
instances of the affected objects and thus be tricked into making
decisions based on these revoked objects. Here too, the goal is that
the CA will be notified of the problem (by RPs) and will remedy the
In the attack scenarios above, when a CRL or manifest is described as
stale, this means that the next issue date for the CRL or manifest
has passed. Until the next issue date, an RP will not detect the
attack. Thus, it behooves CAs to select CRL/manifest lifetimes (the
two are linked) that represent an acceptable trade-off between risk
and operational burdens.
Attacks effected by adversaries that are legitimate managers of
publication points can have much greater effects and are discussed
below under attacks on or by CAs.
4.5. Attacks on an RPKI CA
Every entity to which INRs have been allocated/assigned is a CA in
the RPKI. Each CA is nominally responsible for managing the
repository publication point for the set of signed products that it
generates. (An INR holder may choose to outsource the operation of
the RPKI CA function and the associated publication point. In such
cases, the organization operating on behalf of the INR holder becomes
the CA from an operational and security perspective. The following
discussion does not distinguish such outsourced CA operations.)
Note that attacks attributable to a CA may be the result of malice by
the CA (i.e., the CA is the adversary), or they may result from a
compromise of the CA.
All of the adversaries listed in Section 2 are presumed to be capable
of launching attacks against the computers used to perform CA
functions. Some adversaries might effect an attack on a CA by
violating personnel or physical security controls as well. The
distinction between the CA as an adversary versus the CA as an attack
victim is important. Only in the latter case should one expect the
CA to remedy problems caused by an attack once the attack has been
detected. (If a CA does not take such action, the effects are the
same as if the CA is an adversary.)
Note that most of the attacks described below do not require
disclosure of a CA's private key to an adversary. If the adversary
can gain control of the computer used to issue certificates, it can
effect these attacks, even though the private key for the CA remains
"secure" (i.e., not disclosed to unauthorized parties). However, if
the CA is not the adversary, and if the CA's private key is not
compromised, then recovery from these attacks is much easier. This
motivates use of hardware security modules to protect CA keys, at
least for higher tiers in the RPKI.
An attack by a CA can result in revocation or replacement of any of
the certificates that the CA has issued. Revocation of a certificate
should cause RPs to delete the (formerly) valid certificate (and
associated signed object, in the case of a revoked EE certificate)
that they have cached. This would cause repository objects (e.g., CA
certificates and ROAs) that are verified under that certificate to be
considered invalid, transitively. As a result, RPs would not
consider any ROAs or PATHSEC-protected updates to be valid based on
these certificates, which would make routes dependent on them less
preferred. Because a CA that revokes a certificate is authorized to
do so, this sort of attack cannot be detected, intrinsically, by most
RPs. However, the entities affected by the revocation or replacement
of CA certificates can be expected to detect the attack and contact
the CA to effect remediation. If the CA was not the adversary, it
should be able to issue new certificates and restore the publication
An adversary that controls the CA for a publication point can publish
signed products that create more subtle types of DoS attacks against
RPs. For example, such an attacker could create subordinate CA
certificates with Subject Information Access (SIA) pointers that lead
RPs on a "wild goose chase" looking for additional publication points
and signed products. An attacker could publish certificates with
very brief validity intervals or CRLs and manifests that become
"stale" very quickly. This sort of attack would cause RPs to access
repositories more frequently, and that might interfere with
legitimate accesses by other RPs.
An attacker with this capability could create very large numbers of
ROAs to be processed (with prefixes that are consistent with the
allocation for the CA) and correspondingly large manifests. An
attacker could create very deep subtrees with many ROAs per
publication point, etc. All of these types of DoS attacks against
RPs are feasible within the syntactic and semantic constraints
established for RPKI certificates, CRLs, and signed objects.
An attack that results in revocation and replacement (e.g., key
rollover or certificate renewal) of a CA certificate would cause RPs
to replace the old, valid certificate with the new one. This new
certificate might contain a public key that does not correspond to
the private key held by the certificate subject. That would cause
objects signed by that subject to be rejected as invalid, and prevent
the affected subject from being able to sign new objects. As above,
RPs would not consider any ROAs issued under the affected CA
certificate to be valid, and updates based on router certificates
issued by the affected CA would be rejected. This would make routes
dependent on these signed products less preferred. However, the
constraints imposed by the use of extensions detailed in [RFC3779]
prevent a compromised CA from issuing (valid) certificates with INRs
outside the scope of the CA, thus limiting the impact of the attack.
An adversary that controls a CA could issue CA certificates with
overlapping INRs to different entities when no transfer of INRs is
intended. This could cause confusion for RPs as conflicting ROAs
could be issued by the distinct (subordinate) CAs.
An adversary could replace a CA certificate, use the corresponding
private key to issue new signed products, and then publish them at a
publication point controlled by the attacker. This would effectively
transfer the affected INRs to the adversary or to a third party of
his choosing. The result would be to cause RPs to view the entity
that controls the private key in question as the legitimate INR
holder. Again, the constraints imposed by the use of the extensions
in RFC 3779 prevent a compromised CA from issuing (valid)
certificates with INRs outside the scope of the CA, thus limiting the
impact of the attack.
Finally, an entity that manages a repository publication point can
inadvertently act as an attacker (an example of Walt Kelly's most
famous "Pogo" quote [Kelly70]). For example, a CA might fail to
replace its own certificate in a timely fashion (well before it
expires). It might fail to issue its CRL and manifest prior to
expiration, creating stale instances of these products that cause
concern for RPs. A CA with many subordinate CAs (e.g., an RIR or
NIR) might fail to distribute the expiration times for the CA
certificates that it issues. A network with many ROAs might do the
same for the EE certificates associated with the ROAs it generates.
A CA could rollover its key but fail to reissue subordinate CA
certificates under its new key. Poor planning with regard to rekey
intervals for managed CAs could impose undue burdens for RPs, despite
a lack of malicious intent. All of these examples of mismanagement
could adversely affect RPs, despite the absence of malicious intent.
5. Residual Vulnerabilities
The RPKI, upon which PATHSEC relies, has several residual
vulnerabilities that were discussed in the preceding text (Sections
4.4 and 4.5). These vulnerabilities are of two principle forms:
o The RPKI repository system may be attacked in ways that make its
contents unavailable, not current, or inconsistent. The principle
defense against most forms of DoS attacks is the use of a local
cache by each RP. The local cache ensures availability of
previously acquired RPKI data in the event that a repository is
inaccessible or if the repository contents are deleted
(maliciously). Nonetheless, the system cannot ensure that every
RP will always have access to up-to-date RPKI data. An RP, when
it detects a problem with acquired repository data, has two
1. The RP may choose to make use of its local cache, employing
local configuration settings that tolerate expired or stale
objects. (Such behavior is, nominally, always within the
purview of an RP in PKI.) Using cached, expired, or stale
data subjects the RP to attacks that take advantage of the
RP's ignorance of changes to this data.
2. The RP may chose to purge expired objects. Purging expired
objects removes the security information associated with the
real-world INRs to which the objects refer. This is
equivalent to the affected INRs not having been afforded
protection via the RPKI. Since use of the RPKI (and PATHSEC)
is voluntary, there may always be a set of INRs that are not
protected by these mechanisms. Thus, purging moves the
affected INRs to the set of non-participating INR holders.
This more conservative response enables an attacker to move
INRs from the protected set to the unprotected set.
o Any CA in the RPKI may misbehave within the bounds of the INRs
allocated to it, e.g., it may issue certificates with duplicate
resource allocations or revoke certificates inappropriately. This
vulnerability is intrinsic in any PKI, but its impact is limited
in the RPKI because of the use of extensions in RFC 3779. It is
anticipated that RPs will deal with such misbehavior through
administrative means once it is detected.
PATHSEC has a separate set of residual vulnerabilities:
o It has been stated that "route leaks" are viewed as a routing
security problem by many operators. However, BGP itself does not
include semantics that preclude what many perceive as route leaks,
and there is no definition of the term in any RFC. This makes it
inappropriate to address route leaks in this document.
Additionally, route leaks are outside the scope of PATHSEC,
consistent with the security context noted in Section 1 of this
document. If, at a later time, the SIDR security context is
revised to include route leaks, and an appropriate definition
exists, this document should be revised.
o PATHSEC is not required to protect all attributes associated with
an AS_PATH, even though some of these attributes may be employed
as inputs to routing decisions. Thus, attacks that modify (or
strip) these other attributes are not prevented/detected by
PATHSEC. As noted in Section 1, the SIDR security context calls
for protecting only the information needed to verify that a
received route traversed the ASes in question, and that the NLRI
in the route is what was advertised. (The AS_PATH data also may
have traversed ASes within a confederation that are not
represented. However, these ASes are not externally visible and
thus do not influence route selection, so their omission in this
context is not a security concern.) Thus, protection of other
attributes is outside the scope of this document, as described in
Section 1. If, at a later time, the SIDR security context is
revised to include protection of additional BGP attributes, this
document should be revised.
o PATHSEC cannot ensure that an AS will withdraw a route when the AS
no longer has a route for a prefix, as noted in Section 4.2.
PATHSEC may incorporate features to limit the lifetime of an
advertisement. Such lifetime limits provide an upper bound on the
time that the failure to withdraw a route will remain effective.
6. Security Considerations
A threat model is, by definition, a security-centric document.
Unlike a protocol description, a threat model does not create
security problems nor does it purport to address security problems.
This model postulates a set of threats (i.e., motivated, capable
adversaries) and examines classes of attacks that these threats are
capable of effecting, based on the motivations ascribed to the
threats. It describes the impact of these types of attacks on
PATHSEC, including the RPKI on which PATHSEC relies. It describes
how the design of the RPKI (and the PATHSEC design goals) address
classes of attacks, where applicable. It also notes residual
The authors with to thank the members of the SIDR working group for
the extensive feedback provided during the development of this
8. Informative References
[Kelly70] Kelly, W., "We Have Met The Enemy and He Is Us: Pogo Earth
Day Poster", April 1970.
Kent, S., Lynn, C., and K. Seo, "Design and Analysis of
the Secure Border Gateway Protocol (S-BGP)", IEEE DISCEX
Conference, June 2000.
[RFC3779] Lynn, C., Kent, S., and K. Seo, "X.509 Extensions for IP
Addresses and AS Identifiers", RFC 3779, June 2004.
[RFC4271] Rekhter, Y., Li, T., and S. Hares, "A Border Gateway
Protocol 4 (BGP-4)", RFC 4271, January 2006.
[RFC4272] Murphy, S., "BGP Security Vulnerabilities Analysis", RFC
4272, January 2006.
[RFC4301] Kent, S. and K. Seo, "Security Architecture for the
Internet Protocol", RFC 4301, December 2005.
10 Moulton St.
Cambridge, MA 02138
University of North Carolina - Chapel Hill
c/o Department of Computer Science
CB 3175, Sitterson Hall
Chapel Hill, NC 27599