Understanding

Certification Path Construction

September 2002

Acknowledgements
Certification path construction is a complex process and is subject to some
degree of trial and error. The certification path construction process has
not been standardized, and there is very little published information
available to help implementers and product evaluators understand the
complexities involved. The purpose of this white paper is to clarify issues
associated with the certification path construction process and to make
recommendations where appropriate.

This White Paper is a

deliverable from the PKI
Forums Technical Group
(TWG). Several member
organizations and individuals
have contributed by providing
content, editorial assistance and
editorial reviews.
Author:

Introduction

Steve Lloyd
PKI Forum
Contributors:

Public Key Infrastructure (PKI) supports a number of security-related services, including

data confidentiality, data integrity, and end-entity authentication. Fundamentally, these
services are based on the proper use of public/private key pairs. The public component of
this key pair is issued in the form of a public key certificate and, in association with the
appropriate algorithm(s), it may be used to verify a digital signature, encrypt data, or
both1.
Before a certificate can be used, it must be validated. In order to validate such a certificate,
a chain of certificates or a certification path between the certificate and an established
point of trust must be established, and every certificate within that path must be checked.
This process is referred to as certification path processing2.

Andrew Nash
Russ Housley
John Linn
Magnus Nystrom
RSA Security
Helen Mullenger
Baltimore Technologies
Jeff Stapleton
KPMG LLP

In general, certification path processing consists of two phases: 1) path construction and
2) path validation described as follows:
1) Path construction involves "building" one or more candidate certification paths. Note
that we use "candidate" here to indicate that although the certificates may chain
together properly, the path itself may not be valid for other reasons such as path
length, name, or certificate policy constraints/restrictions.
2) Path validation includes making sure that each certificate in the path is within its
established validity period, has not been revoked, has integrity, et cetera; and any
constraints levied on part or all of the certification path are honored (e.g., path length
constraints, name constraints, policy constraints). However, some aspects that might
be associated with path validation are sometimes taken into consideration during the
path construction process in order to maximize the chances of finding an acceptable
certification path sooner rather than later. We will revisit these concepts later in this
paper.

It should be recognized that there are several reasons that separate key pairs should be used for
digital signature and confidentiality, including differing requirements associated with key backup/
recovery and long-term handling of keying material, and the ability to use different algorithms for
each (e.g., DSA could be used for the digital signature and RSA could be used for symmetric key
exchange).
This is sometimes called "certificate path processing." We use "certification path processing" in
order to maintain consistency with the terminology found in X.509.

Contents
Introduction

Purpose, Scope and

Assumptions

Basic Concepts

Identifying Appropriate
Certificates

Constructing Certification
Paths

Evaluating Certificate
Paths During the Path
Construction Process

12

Summary

14

References

14

The 4th Edition of X.509 [Ref1] and the Internet Certificate and Certificate Revocation List
Profile as defined in RFC3280 [Ref2] provide the most recent implementation guidelines for
certification path validation. However, both sources are silent when it comes to the certification path construction process.3 In fact, there is very little published literature that addresses
the certification path construction issue. Notable exceptions include [Ref3] and [Ref4].

Purpose, Scope and Assumptions

The purpose of this paper is to

clarify terminology and review issues associated with the certification path construction process and
to make recommendations where
appropriate.

The path construction process can be complex and subject to some degree of trial and error,
and there is very little existing literature that discusses issues associated with the certification
path construction process. Therefore, the purpose of this paper is to clarify terminology and
review issues associated with the certification path construction process and to make recommendations where appropriate. However, this paper is not meant to dictate how vendors
MUST perform path construction, nor are we attempting to describe a complete path construction algorithm. Vendors are free to implement their own path construction logic as they
see fit. Nonetheless, this paper provides useful information that should be taken into consideration when evaluating or implementing certification path construction algorithms.
Although this paper discusses issues related to certification path validation, it is not the
purpose of this paper to address certification path validation per se. The 4th Edition of X.509
and RFC3280 should be consulted for additional information regarding the certification path
validation process.
This paper focuses on Version 3 public key certificates as defined in the 4th Edition of the
X.509 Recommendation [Ref1]. We recognize that other forms of certificates exist, including
the Version 1 and Version 2 public key certificates defined in previous Editions of X.509.
However, we concentrate on Version 3 public key certificates since they are the most common
form of certificate found in most enterprise and government deployments, and many of the
extensions that are only supported with Version 3 are essential in order to control business
relationships within and across PKI domains. This paper discusses certain Version 3 certificate extensions that can be used to help facilitate the certification path construction process.
It should be noted that any discussion associated with these extensions is not applicable in the
context of Version 1 or Version 2 public key certificates.
It is recognized that path construction software can be implemented locally (i.e., coupled with the
client system), remotely (i.e., delegated to an external trusted 3rd party), or a combination of both.
However, where path construction is performed is irrelevant for the purposes of this paper.

This paper focuses on Version 3

public key certificates as defined
in the 4th Edition of the X.509 Recommendation [Ref1]... We concentrate on Version 3 public key
certificates since they are the most
common form of certificate found
in most enterprise and government deployments, and many of
the extensions that are only supported with Version 3 are essential in order to control business relationships within and across PKI
domains.

Basic Concepts
The basic structure of the Version 3 public key certificate is illustrated in Figure 1. Certificates
are issued by Certification Authorities (CAs) to other CAs or to end-entities (e.g., end-users,
devices, Web servers, processes). CAs may also "self-issue" certificates to themselves (this is
discussed in more detail later).
Certificates issued to CAs are known as CA certificates, and certificates issued to end-entities
are referred to as end-entity certificates. The Basic Constraints certificate extension is used to
distinguish between CA certificates and end-entity certificates.
The 4th Edition of X.509 defines a relying party as "a user or agent that relies on the data in a
certificate in making decisions." Stated another way, a relying party "uses" end-entity certificates for some express purpose. For example, a relying party may authenticate a message
originator and verify the integrity of the message by "using" the message originator's corresponding certificate to verify a digital signature associated with the message. A trust anchor
is a CA certificate (or more precisely, the public verification key of a CA) used by a relying
party as the starting point for path validation (which may or may not be the same starting
point for path construction as discussed further below). A relying party may have one or
more trust anchors, and these trust anchors can be derived from a number of sources. For
example, a trust anchor may be the public key of a root CA or it may be the public key of the
CA that issues one or more certificates directly to the relying party.
3

This is simply a statement of fact and is not intended to be a criticism. It is generally agreed within the
industry that path construction algorithms should not be subject to standardization. The purpose of
this paper is to shed light on path construction issues that are not immediately obvious from the existing
standards.

2 Path Construction: September 2002

PKI Forum: Understanding Certification

PKI Forum: Understanding Certification2Path Construction: September 2002

Basic Concepts continued

Figure 1: X.509 Version 3 Public Key Certificate

[Digitally Signed by Issuing CA]

Version

Serial Signature Issuer

Number
(Info)

Validity

Subject

Subject Public
Key Info

Optional
Digital
Issuer
Subject
Unique ID Unique ID Extensions Signature

Possible Extensions
Authority Key
Identifier

A certificate that one CA issues to

another CA is referred to as a crosscertificate. Cross-certification can be
unilateral or bilateral.

Subject Key
Identifier

A certificate that one CA issues to another CA is referred to as a cross-certificate. A

superior CA may issue a cross-certificate to a subordinate CA as commonly found in
hierarchical trust models. This is referred to as unidirectional or unilateral cross-certification. When CAs issue certificates to each other it is known as mutual or bilateral crosscertification. This is commonly found in distributed trust models. We note that there are
many examples in the industry where the term cross-certification is meant to denote the
bilateral case only. However, from a technical perspective cross-certification can also be
unilateral. This is reflected in the 4th Edition of X.509 where a cross-certificate is defined as
follows:
Cross certificate - This is a certificate where the issuer and the subject are different CAs.
CAs issue certificates to other CAs either as a mechanism to authorize the subject CA's
existence (e.g. in a strict hierarchy) or to recognize the existence of the subject CA (e.g.
in a distributed trust model). The cross-certificate structure is used for both of these.
Certification path construction involves discovery of a "chain of certificates" between the
end-entity certificate and a recognized trust anchor. Certification paths can be constructed in the forward direction (i.e., from the end-entity certificate to a recognized trust
anchor) or they can be constructed in the reverse direction (i.e., from a recognized trust
anchor to the end-entity certificate). Which of these methods is best has been the source
of some debate over the past few years. We assert that forward direction path construction is best suited for hierarchical trust models and reverse direction path construction is
best suited for distributed trust models, and this paper will demonstrate that a robust
path construction algorithm must be capable of building paths in both directions. In
fact, it may be appropriate to build portions of a certification path using one method and
other portions of the certification path using the other.
Before we begin to look at the issues surrounding the certification path construction
process, let's first explore some fundamentals behind the notion of "certificate chaining".

We assert that forward direction

path construction is best suited for
hierarchical trust models and reverse
direction path construction is best
suited for distributed trust models,
and this paper will demonstrate that
a robust path construction algorithm
must be capable of building paths
in both directions.

Name Chaining
At the most basic level, a candidate certification path must "name chain" between the
recognized trust anchor and the target certificate (i.e., the end-entity certificate). Working
from the trust anchor to the target certificate, this means that the Subject Name in one
certificate must be the Issuer Name in the next certificate in the path, and so on. Figure 2
helps to illustrate this concept. In this example, the path begins with a self-signed certificate that contains the public key of the trust anchor. The path ends with the end-entity
certificate. All other certificates within the path are referred to as intermediate CA certificates. Note that every certificate in the chain except for the last one is a CA certificate.

PKI Forum: Understanding Certification Path Construction: September 2002

Figure 2: Name Chaining Illustration

Equal

[Self-Signed Certificate]
Issuer = CA0

Subject = CA0

Equal

Issuer = CA0

Name chaining alone may not be

sufficient to determine if the certification path is a legitimate candidate
that should be submitted to the certification path validation process.

Subject = CA1

Equal

Issuer = CA1

[Intermediate CA Certificate]

Subject = CA2

Equal

Issuer = CA2

[Intermediate CA Certificate]

[End-Entity Certificate]

Subject = User1

Although we will discuss this in more detail later, let's quickly review how a certification path
can be constructed using name chaining. (In order to keep things simple, it is assumed that the
Issuer DN's match the corresponding directory entries associated with each CA.) When
building certification paths in the forward direction, we can use the Issuer DN in the end-entity
certificate to retrieve the certificate(s) that have been issued to the CA that issued the end-entity
certificate. As illustrated in Figure 2, Issuer=CA2 in the end-entity certificate will lead to CA2's
certificate. Once we have retrieved CA2's certificate, we can use the Issuer DN in CA2's
certificate to retrieve CA1's certificate. Finally, the Issuer DN in CA1's certificate leads us to
CA0's certificate. The same logic applies when we build certification paths in the reverse
direction (with the order reversed, of course). This is discussed in more detail later.

AKIDs are used to distinguish one

public key from another when a
given Certification Authority (CA) has
multiple signing keys, and SKIDs
provide a means to identify certificates that contain a specific public
key.

If Certification Authorities (CAs) were guaranteed to have only one public/private signing key
pair active at any given time, satisfying the name chaining requirement would be all that is
required to construct a candidate certification path. However, it is possible (even likely) that
CAs will have more than one valid signing key pair at the same time (e.g., to support CA key
rollover). This means that name chaining alone may not be sufficient to determine if the
certification path is a legitimate candidate that should be submitted to the certification path
validation process. This leads us to the notion of "key identifier chaining" as discussed below.

Key Identifier Chaining

The Authority Key Identifier (AKID) and Subject Key Identifier (SKID) are certificate extensions that can be used to help facilitate the certification path construction process. As discussed in X.509 and the Internet Certificate and CRL Profile (RFC3280), AKIDs are used to
distinguish one public key from another when a given Certification Authority (CA) has multiple signing keys, and SKIDs provide a means to identify certificates that contain a specific
public key.4
Similar to "name chaining" between a trust anchor and an end-entity certificate, the SKID of
the first certificate should be the AKID of the next certificate in the path, and so on. Figure 3
helps to illustrate this concept. Note that since the AKID and SKID are certificate extensions,
the concept of key identifier chaining applies only to Version 3 public key certificates.

The reader may be interested in a recently published PKI Forum Implementation Guideline regarding AKIDs/SKIDs in the context of cross-certification. This guideline can be found at http://
www.pkiforum.org/resources.html.

4 Path Construction: September 2002

PKI Forum: Understanding Certification

PKI Forum: Understanding Certification4Path Construction: September 2002

Figure 3: Key Identifier Chaining Illustration

Equal

[Self-Signed Certificate]
AKID = W
(if present)

SKID = W

Equal

[Intermediate CA Certificate]

AKID = W

SKID = X

Equal

AKID = X

[Intermediate CA Certificate]
SKID = Y

Equal

AKID = Y

[End-Entity Certificate]
SKID = Z

AKID/SKID Structure and Use

In accordance with X.509, the AKID can be represented using a keyIdentifier, an
authorityCertIssuer, authoritySerialNumber pair, or both. However, X.509 also states that:
The keyIdentifier form can be used to select CA certificates during path construction.
The authorityCertIssuer, authoritySerialNumber pair can only be used to provide
preference to one certificate over others during path construction.
In addition, the Internet Certificate and CRL Profile (see RFC3280) states that the keyIdentifier
field in the AKID must be included in all certificates generated by conforming CAs (with the
exception of self-signed certificates, in which case the keyIdentifier would be included within
the SKID and, optionally, within the AKID). Further, the Internet Certificate and CRL Profile
states that:

The Internet Certificate and CRL

Profile states that: the
keyIdentifier field of the
Authority Key Identifier extension MUST be included in all
certificates generated by
conforming CAs to facilitate
chain building and the value of
the Subject Key Identifier MUST
be the value placed in the
keyIdentifier field of the
Authority Key Identifier
extension...of certificates issued
by the subject of this certificate.

The keyIdentifier field of the Authority Key Identifier extension MUST be included in all
certificates generated by conforming CAs to facilitate chain building.
In accordance with X.509, the SKID can only be represented using the keyIdentifier (i.e., the
SKID syntax does not include the authorityCertIssuer, authoritySerialNumber pair). Further, the Internet Certificate and CRL Profile states that all CA certificates must include the
SKID extension and:
The value of the Subject Key Identifier MUST be the value placed in the key identifier field
of the Authority Key Identifier extension of certificates issued by the subject of this
certificate.
Thus, a CA conforming to the Internet Certificate and CRL profile must populate the AKID
of all certificates that it issues with the same keyIdentifier that is populated in the SKID of the
certificate used to verify the digital signature on those issued certificates.
Calculation of keyIdentifier Value
There are multiple methods available for calculating the keyIdentifier. The following methods are specifically mentioned in the Internet Certificate and CRL Profile:

the 20 byte SHA-1 value of the subject public key (not including tag, length, and unused bits)
a four bit value 0100 followed by the least significant 60bits of the SHA-1 value of the
subject public key (not including tag, length, and unused bits)
a monotonically increasing sequence of integers

Note that these are recommendations only - the profile does not mandate a particular
algorithm.
PKI Forum: Understanding Certification Path Construction: September 2002

This leads us to note two important considerations. First, while it is possible that the
keyIdentifier could be unique across multiple PKI domains (e.g., when the SHA-1 hash value
of the public key is used), the only guarantee is that the keyIdentifier is unique relative to a
given public key in the context of a given issuer. This means that we cannot rely solely on the
keyIdentifier value to guarantee that we have the right certificate. Second, it is possible that an
issuer may renew a certificate (i.e., the same public key is issued in a new certificate) and the
keyIdentifier value may therefore be the same in more than one certificate5. This is why the
use of the authorityCertIssuer, authoritySerialNumber pair may be needed to select one
certificate over another (i.e., the serial number is guaranteed to be unique for every certificate
issued by a given CA even though the same public key may have been re-issued).
X.509 defines specific directory attributes where CA and end-entity
certificates are to be stored. While
there has been some controversy
over which attributes should be used,
the 4th edition of X.509 mandates
certain requirements and, in some
cases, expected behavior.

Identifying Appropriate Certificates

Now that we have a better understanding surrounding some of the basics, let's turn our
attention to some of the issues associated with locating the appropriate certificates.
In the enterprise context, most PKI vendors have relied on the use of repositories for the
storage and retrieval of certificates. In many instances, this has been supported by an X.500based directory with client-server interaction supported by the Lightweight Directory Access
Protocol (LDAP). X.509 defines specific directory attributes where CA and end-entity certificates are to be stored. While there has been some controversy over which attributes should be
used, the 4th edition of X.509 mandates certain requirements and, in some cases, expected
behavior. The syntax and expected use of these attributes in accordance with X.509 are
described in the paragraphs that follow.
X.509 Certificate Attributes
X.509 defines several certificate attributes. For the purposes of this paper, we are concerned
mostly with the cACertificate attribute and the crossCertificatePair attribute. The
userCertificate attribute used to store end-entity certificates also comes into play, but for
purposes of simplicity we assume that the end-entity certificate is already in-hand before the
certification path construction process begins. We also note that X.509 defines a pkiPath
attribute, and we discuss this briefly at the end of this section.
As specified in X.509:
The cACertificate attribute of a CA's directory entry shall be used to store self-issued certificates (if any) and certificates issued to this CA by CAs in the same realm as this CA. The
definition of realm is purely a matter of local policy.
By definition, self-issued certificates are CA certificates where the Issuer DN and the Subject DN
are the same. A self-signed certificate is a self-issued certificate where the private key used to sign
that certificate corresponds to the public key within that certificate (i.e., the public key within
the certificate is used to verify the digital signature associated with that certificate). X.509
makes the distinction, since a self-issued certificate is not guaranteed to also be a self-signed
certificate. This can be illustrated using CA key rollover as an example. To allow for graceful
transition from the old signing key pair to the new signing key pair, the CA should issue a
certificate that contains the old public key signed by the new private signing key and a certificate
that contains the new public key signed by the old private signing key. Both of these certificates
are self-issued, but neither is self-signed. Note that these are in addition to the two self-signed
certificates (one old, one new).
The cross-certificate pair attribute consists of two elements: 1) issuedToThisCA and 2)
issuedByThisCA. (Prior to the 4th edition of X.509 these elements were referred to as forward
and reverse, respectively.) As specified in X.509:
The issuedToThisCA elements of the crossCertificatePair attribute of a CA's directory entry
shall be used to store all, except self-issued certificates issued to this CA. Optionally, the
issuedByThisCA elements of the crossCertificatePair attribute of a CA's directory entry may
contain a subset of certificates issued by this CA to other CAs. If a CA issues a certificate to
another CA, and the subject CA is not a subordinate to the issuer CA in a hierarchy, then the
issuer CA shall place that certificate in the issuedByThisCA element of crossCertificatePair
attribute of its own directory entry.

The PKI Forum does not necessarily endorse this behavior, we simply note that it is possible and
therefore needs to be considered.
PKI Forum: Understanding Certification Path Construction: September 2002

This leads us to several conclusions:

First, we would expect each CA to populate the cACertificate attribute of its own directory entry with its own self-issued certificates, if any. This will include any self-signed
certificates as well as any certificates issued by the CA in support of CA key update.
Second, we may see certificates issued to a CA that have been issued by other CAs that
belong to the same "realm" to be stored in the cACertificate attribute of its own directory
entry. While realm is purposely undefined, one possible example is a single PKI domain
consisting of a hierarchy of CAs. In this particular case we might expect subordinate CAs
to store the certificates issued to them by their superior CA in the cACertificate attribute
of their own directory entry. However, given that "realm" is undefined and subject to
interpretation, there is no guarantee that this will be implemented consistently by all
vendors.
Third, we would expect to see the issuedToThisCA element of the cross-certificate pair
attribute to be populated with certificates that have been issued to this CA by other CAs.
(This suggests that both the cACertificate and issuedToThisCA might be populated
when the issuing CA is in the same realm as the issued to CA.)
Finally, we would expect the issuedByThisCA attribute to be populated with certificates
issued by this CA to other CAs when those CAs do not belong to the same hierarchy. In
particular, we would expect to see the issuedByThisCA attribute populated with certificates issued to peer CAs as commonly found in a distributed trust model. Although the
issuedByThisCA could be populated in a hierarchical domain, it is not reasonable to
assume that this will always be the case.

Thus, we can expect the following

mandatory behavior for CAs that conform to the 4th edition of X.509:
1.

all self-issued certificates must be

stored in the cACertificate attribute of the issuing CA's directory entry;

2.

all certificates issued to a CA except for self-issued certificates

must be stored in the
issuedToThisCA element of the
cross-certificate pair attribute of
that CA's directory entry; and

3.

all certificates issued by a CA to

a non-subordinate or peer CA
must be stored in the
issuedByThisCA element of the
cross-certificate pair attribute of
the issuing CA's directory entry.

Thus, we can expect the following mandatory behavior for CAs that conform to the 4th
edition of X.509:
1.

all self-issued certificates must be stored in the cACertificate attribute of the issuing
CA's directory entry;

2.

all certificates issued to a CA except for self-issued certificates must be stored in the
issuedToThisCA element of the cross-certificate pair attribute of that CA's directory entry; and

3.

all certificates issued by a CA to a non-subordinate or peer CA must be stored in the

issuedByThisCA element of the cross-certificate pair attribute of the issuing CA's
directory entry.

The rest of the behavior described above is considered optional. While some of these options
can be implemented and used to help in the path construction process, this paper assumes
that these options may not be used and therefore cannot be relied upon.
The 4th Edition of X.509 also defines a pkiPath attribute that can be used to store partial or
complete certification paths. Each path is represented by a sequence of cross-certificates.
X.509 describes the use of this attribute as follows:
This attribute can be stored in the CA directory entry and would contain some certification
paths from that CA to other CAs. This attribute, if used, enables more efficient retrieval of
cross-certificates that form frequently used certification paths. As such there are no specific
requirements for this attribute to be used and the set of values that are stored in the attribute
will likely not represent the complete set of forward certification paths for any given CA.
Since we are unaware of any existing implementations that make use of the pkiPath attribute
in this way, and since X.509 implies that use of this attribute is optional, we do not explore the
use of this attribute further.
Use of Private Certificate Extensions
In many cases, client systems are configured with default IP addresses and/or DNS names of
one or more repositories. These repositories are queried for the necessary certificates. In
addition, search bases are often established so the appropriate certificates can be easily located
and retrieved.

PKI Forum: Understanding Certification Path Construction: September 2002

It remains to be seen how use of

the AIA private extension may
evolve over time... In the future,
we may see vendors use the AIA
to point to additional information
such as a list of superior CAs.

As an alternative, the Internet Certificate and CRL Profile defines a private extension
referred to as the Authority Information Access (AIA) extension. As stated in RFC3280,
the AIA "indicates how to access CA information and services for the issuer of the
certificate in which the extension appears." One possible use of this extension is to point
to a list of "CAs that have issued certificates superior to the CA that issued the certificate
containing this extension[this] is intended to aid certificate users in the selection of a
certification path that terminates at a point trusted by the certificate user." It remains to
be seen how use of the AIA private extension may evolve over time. Currently, Microsoft
(Windows 2000 and Windows XP) uses the AIA to point to the location of the issuing
CA's certificate [Ref5]. In the future, we may see vendors use the AIA to point to
additional information such as a list of superior CAs.
The latest Internet Draft Certificate and CRL Profile also defines a private extension referred
to as the Subject Information Access (SIA) extension. This extension "indicates how to access
information and services for the subject of the certificate in which the extension appears."
One possible use of this extension is to identify where a CA publishes the certificates (and
possibly CRLs) that it issues. This is something that we may see exploited in the future.
Other Options?
It is common practice to send the public key certificate necessary to verify a digital signature
along with digitally signed data6. For example, secure e-mail based on S/MIME supports
this. It is also possible to send partial or complete certification paths to the relying parties via
protocol - perhaps using the pkiPath attribute syntax defined in X.509. In these cases, some
(or perhaps all) of the certification path may not have to be retrieved from external sources.
However, in complex, richly connected PKI domains, it is unlikely that the originator of a
message will know a complete certification path between the relying party's trust anchor and
the originator's certificate. Even if such a path were known, there is no guarantee that this will
be the path that will meet the path validation criteria levied by the relying party (or by any
intermediate CAs). While there may be some cases where we can take advantage of the
knowledge of partial paths, in most cases we will need a robust path construction algorithm
capable of discovering one or more complete certification path(s).

Constructing Certification Paths

While there may be some cases

where we can take advantage of
the knowledge of partial paths, in
most cases we will need a robust
path construction algorithm capable
of discovering one or more complete
certification path(s).

The examples that follow are designed to illustrate some of the issues that can be encountered
when building certification paths. In general, the examples are independent from the mechanism used to retrieve the necessary certificates. Conformance with the 4th edition of X.509 is
assumed.
Simple Example - Strict Hierarchy
Let's start with a relatively simple example. Figure 4 illustrates a strict hierarchy of CAs. The
octagons represent CAs, the arrows represent certificate issuance, and the sectioned rectangles represent certificates. The specific X.509 directory attribute that is used to store each
certificate is identified. Let's examine what happens when User1 sends a digitally signed email to User2. It is assumed that User1's verification certificate is conveyed along with the
message itself.
In this example, User2 is the relying party. Since we are attempting to verify a digital signature
using User1's verification certificate, we need to construct a certification path between User1's
certificate and a trust anchor recognized by User2. In this case, CA0 is the root CA and is, by
definition, the common trust anchor for all users within this strict hierarchy. Basically, User2
wants to know if CA0 has established a trust relationship (either directly or indirectly) with
the issuer of User1's certificate (in this case, CA1). In other words, if the relying party
software is able to resolve the certification path CA0->CA1->User1, then we would have a
candidate certification path that could be submitted to the path validation logic.

Certificates used to verify digital signatures are sometimes referred to as verification certificates.

8 Path Construction: September 2002

PKI Forum: Understanding Certification

Forum: Understanding Certification8Path Construction: September 2002

Building certification paths within strict hierarchies is rather straightforward. Paths are typically constructed in the forward direction (i.e., we start with the target certificate and work our
way to a recognized trust anchor). Thus, the relying party software will start with User1's
verification certificate and work its way to CA0. Since the relying party software knows CA1
is the issuer of User1's certificate (the Issuer DN in User1's certificate is CA1), this is accomplished by retrieving the issuedToThisCA element that is stored in the directory entry for CA1.
We then discover that CA0 is the issuer of CA1's certificate, and we now have a complete
candidate certification path between User1's certificate and a trust anchor recognized by
User2.
Figure 4: Path Construction Illustration Strict Hierarchy

cACertificate

Issuer
CA0

Subject
CA0
CA0

Issuer
Subject
CA0
CA1
issuedToThisCA

CA1

Issuer
Subject
CA0
CA2
issuedToThisCA

CA2

User1
Issuer
CA1

Subject
User1

CA3

Issuer
Subject
CA2
CA4
issuedToThisCA

CA4

userCertificate

User2
Issuer
CA4

Subject
User2

userCertificate

Building certification paths within

strict hierarchies is rather straightforward. Paths are typically constructed in the forward direction (i.e.,
we start with the target certificate
and work our way to a recognized
trust anchor).

More Complex Example -- Single Cross-Certification with Hierarchical and

Distributed Trust Models
We will use Figure 5 to illustrate a more complex path construction process between two PKI
domains that have mutually cross-certified with one another. As before, the octagons represent CAs and the sectioned rectangles represent certificates. The specific X.509 directory
attribute that is used to store each certificate is identified. As in the previous example, the
unidirectional arrows represent certificate issuance. The bi-directional arrows (including the
dashed arrow between CA0 and CA5) represent mutual cross-certification. Thus, the PKI
domain on the left is a hierarchical trust model whereas the PKI domain on the right is a
distributed trust model.
In this example, User3 receives a digitally signed e-mail from User2. User3 is the relying party
and the certificate to be validated is User2's verification certificate. User2's verification certificate accompanies the e-mail sent to User3.
The task at hand is to construct a certification path between User2's verification certificate and
a trust anchor that User3 recognizes (in this case, CA7). We can see that one possibility is to
work our way "out" from the relying party's trust anchor to eventually discover the crosscertificate that CA5 issued to CA0. This is referred to as path construction in the reverse
direction and it produces a partial path CA7->CA5->CA0. This partial path is constructed by
retrieving the issuedByThisCA certificates under CA7's directory entry, which, in turn, leads to
the retrieval of the issuedByThisCA certificates under CA5's directory entry.7

It is recognized that there are actually two paths between CA7 and CA5 (i.e., CA7->CA5 and
CA7->CA6->CA5). However, for the time being we will concentrate on the shorter path. We will
revisit multiple path issues a bit later.

PKI Forum: Understanding Certification Path Construction: September 2002

Figure 5: Path Construction Illustration Single Inter-Domain Cross-Certification

Issuer

Subject

Issuer

Subject

cACertificate

CA0

CA0

cACertificate

CA5

CA5

issuedToThisCA

CA5

CA0

issuedToThisCA

CA0

CA5

issuedByThisCA

CA0

CA5

issuedByThisCA

CA5

CA0

CA1

CA0
CA5

Issuer
Subject
CA0
CA2
issuedToThisCA

CA2

Issuer

Subject

CA7

CA7

issuedToThisCA

CA5

CA7

issuedByThisCA

CA7

CA5

cACertificate

CA6

CA7

User1

CA3

Issuer
Subject
CA2
CA4
issuedToThisCA

CA4

User3
Issuer
CA7

Subject
User3

userCertificate
User2

In a strict hierarchy, working in the

forward direction makes sense,
since we are always guaranteed to
find the issuedToThisCA element
of the cross-certificate pair attribute
populated with the certificate(s) that
have been issued to each subordinate CA.

Issuer
CA4

Subject
User2

userCertificate

Once we "reach" CA0, it may not be possible to continue to construct the path in the reverse
direction, since we cannot depend on the issuedByThisCA element to be populated by CAs
that belong to a hierarchy (recall that population of this element is considered optional in this
case). Even if the issuedByThisCA element is populated by each CA in the hierarchical
domain, searching in the reverse direction from CA0 would be less efficient than searching in
the forward direction from CA4. Specifically, we would encounter two possible paths emanating from CA0 (not counting the link back to CA5), but there is only one possible path leading
from CA4 back up to CA0.
So what do we do next? We now try to construct the rest of the path in the forward direction,
which means we work our way back from the target certificate to CA0. This is accomplished
by looking for the issuedToThisCA element in CA4's directory entry (recall that CA4 is the
issuer of the target certificate). (Optionally, we may look for this information in the cACertificate
attribute, but this attribute is not guaranteed to be populated as discussed earlier.) This leads
us to the discovery of the certificate issued by CA2 to CA4 which, in turn, leads us to the
certificate issued by CA0 to CA2. We now have a complete chain of certificates between the
target certificate and the trust anchor recognized by the relying party given by CA7->CA5>CA0->CA2->CA4->User2. Of course, in this example the order of the path construction
may be reversed. That is, we could have constructed the partial path CA0->CA2->CA4>User2 first, and then we could work our way back from CA7 to CA0.
To complete the picture, let's take a quick look at what happens when User3 sends a digitally
signed e-mail to User2. Now User2 is the relying party, and we need to construct a certification
path between User3's verification certificate and a trust anchor recognized by User2 (in this
case, CA0). Specifically, we need CA0->CA5->CA7->User3 (again ignoring the path through
CA6). Similar to the previous example, we want to work our way "out" from the relying
party's trust anchor. Thus, we can construct the necessary path by retrieving the crosscertificate that CA0 issued to CA5 obtained from the issuedByThisCA attribute in CA0's
directory entry, which in turn leads us to the cross-certificate issued by CA5 to CA7 obtained
from the issuedByThisCA attribute in CA5's directory entry.
Forward versus Reverse Certification Path Construction
In a strict hierarchy, working in the forward direction makes sense, since we are always guaranteed to find the issuedToThisCA element of the cross-certificate pair attribute populated
with the certificate(s) that have been issued to each subordinate CA. In addition, it is typically
more efficient to construct paths in the forward direction within a hierarchy. In some cases,
this will resolve to one and only one possible certification path. In others, a limited number of
alternative paths may be constructed as a result of the existence of multiple certificates per CA.
Use of the keyIdentifier may be used to resolve to a single path in some cases, but this may not

10
PKI Forum: Understanding Certification
10 Path Construction: September 2002

10
PKI Forum: Understanding Certification
Path Construction: September 2002

always be possible (e.g., when Version 1 or 2 certificates are being used or when the same
public key is re-issued and the keyIdentifier value is based on that public key). In any case,
the number of possible certification paths is well bounded when working in the forward
direction within a strict hierarchy.
In the previous two examples above, one might suggest that the certification paths could
have been constructed in the forward direction even in the distributed environment.
Although this may be true in this simple example, this is possible only due to the fact that
there is a single inter-domain cross-certification and a limited number of CAs involved.
When we encounter a distributed trust model, building certification paths in the forward
direction can become much less efficient. This is due to the fact that we may encounter tens
or even hundreds of forward elements associated with a given CA, not just one or two.
This is illustrated in Figure 6, where we have introduced a bridge CA (BCA1) between the two
PKI domains. Although we can easily construct the path from the target certificate to CA0 (as
we saw earlier), this becomes much more complex once we try to continue in the forward
direction working "outward" from CA0. Specifically, BCA1 might be cross-certified with tens
or even hundreds of other CAs (including other bridge CAs), and those CAs may be crosscertified with tens or hundreds more. If we are looking at the issuedToThisCA element only,
we could encounter a large number of CAs that lead away from the path we are seeking.
Attempting to construct every possible path in the forward direction is clearly less practical
under these circumstances. Constructing in the reverse direction also involves trial and error,
but we will always be working with a (partial) path emanating from the relying party's trust
anchor since we are answering the question "who have you issued certificates to" rather
than "who has issued certificates to you." While this may not be enough in itself to
convince everyone that reverse direction is most appropriate in a distributed environment, there are also a number of efficiencies that can be realized when building paths in
the reverse direction, as discussed in [Ref3].

When we encounter a distributed

trust model, building certification
paths in the forward direction can
become much less efficient. This
is due to the fact that we may encounter tens or even hundreds of
forward elements associated with
a given CA, not just one or two.

Figure 6: Path Construction Illustration Fully Distributed

BCA1

CA0

CA1

CA5

CA6

CA2
CA7

User1
User3
CA3

CA4

User2

This leads us to conclude that certification path construction in the forward direction is
optimal for hierarchical trust models, and certification path construction in the reverse direction is optimal for distributed trust models.

PKI Forum: Understanding Certification Path Construction: September 2002

This leads us to conclude that

certification path construction in
the forward direction is optimal
for hierarchical trust models, and
certification path construction in
the reverse direction is optimal
for distributed trust models.

11

Evaluating Certification Paths

During the Path Construction Process
So far, our discussion has not addressed any form of path validation or other helpful criteria
that might be taken into consideration during the path construction process. While it is
possible to ignore any validation criteria during the path construction process, this can lead to
enormous inefficiencies that could otherwise be avoided. This is especially true in complex,
richly interconnected PKI environments.
Let's illustrate this point by revisiting the example we discussed using Figure 5, only now
using the topology represented in Figure 6. As we work our way "out" from User3's trust
anchor, we now encounter the bridge CA (BCA1) instead of the direct cross-certificate
that CA5 issued to CA0. Not counting the cross-certificate BCA1 issued to CA5, we now
have four cross-certificates emanating from BCA1 that come into play, which, in turn,
could lead to dozens more. We would be much better off if we recognized that the crosscertificate BCA1 issued to CA0 is a much more likely candidate to take us closer to where we
need to go. So when we retrieve all the issuedByThisCA entries associated with BCA1's
directory entry, it would greatly improve performance if we somehow knew CA0 is in the path
we are seeking.

Some vendors have developed a

certification path prioritization system in an attempt to submit the
most likely candidate paths to the
path validation logic first.

Since strict hierarchies are typically based on hierarchical naming conventions, we could take
advantage of the fact that the DN of CA0 is very likely a subset of the Relative Distinguished
Names (RDNs) in the target certificate (i.e., User2's certificate) or the RDNs of CA4 (which we
know from the Issuer DN in User2's certificate). In other words, once we discover the crosscertificate issued by CA5 to CA0, we will know we are on the right track, since the RDNs in the
DN of CA0 will match a subset of the RDNs in the Issuer DN and/or Subject DN of User2's
certificate. This does not mean that the other entries will not lead to a legitimate path, but it
does mean that we may want to try CA0 first. Alternatively, we may have constructed the
partial path CA0->CA2->CA4->User2 first, so we would know we have a candidate certification path once we encountered the cross-certificate BCA1 issued to CA0 when working in the
reverse direction.
As mentioned earlier, it is possible to take certain things into consideration in order to maximize the chance that the candidate certification path submitted to the path validation logic will
turn out to be an acceptable path. In fact, some vendors have developed a certification path
prioritization system in an attempt to submit the most likely candidate paths to the path
validation logic first. For example, a freeware Certification Path Library (CPL) that prioritizes
the order in which certificates are considered during the path construction process is available
for download from http://www.entrust.com/entrustcygnacom/products/index.htm. According to the associated Path Development API - Interface Control Document [Ref4], the
prioritization rules are as follows:
1.

Certificates retrieved from the cACertificate attribute should have priority over
certificates retrieved from the crossCertificate attribute

2.

Certificates in which issuer algorithm OID = subject algorithm OID should have
priority

3.

Certificates that assert policies should have priority over certificates that do not.
Among certificates that assert policies, those certificates which match more policies in the initial-acceptable-policy-set should have priority

4.

Certificates with fewer RDN elements in the Issuer DN should have priority

5.

Certificates that match more RDNs between the issuer DN and relying party trust
anchor DN should have priority

6.

Certificates that match more RDNs between the subject DN and the issuer DN
should have priority

7.

Certificates with longer validity periods (furthest notAfter date) should have priority

Note that a number of assumptions (such as breadth versus depth) will have an influence
on prioritization rules, and this information is supplied as an example only. The referenced document is several years old and it is possible that different prioritization rules
may have been selected based on more recent developments, including lessons learned
from some of the recently concluded interoperability initiatives, such as the US Federal
Bridge CA demonstration and the UK Communications-Electronics Security Group

12 Path Construction: September 2002

PKI Forum: Understanding Certification

12
PKI Forum: Understanding Certification
Path Construction: September 2002

(CESG) interoperability initiatives. In any case, vendors may elect to adopt variations on
such a prioritization system. This paper does not make any specific recommendations in
this regard.
It may also be possible to take certain path validation criteria into consideration during
the path construction process. As above, the idea is to help increase the chances that one
or more certification paths will pass the more comprehensive path validation procedure.
However, this allows us to eliminate certain paths from consideration that might otherwise be considered viable candidates.
Building Certification Paths: Forward vs. Reverse [Ref3] identifies several path validation
criteria that might be taken into consideration during the path construction process,
including:

name constraints,
certificate policy processing, and
certificate validation (including revocation status check and digital signature verification).

Taking one or more of these into consideration during the certification path construction
process could possibly eliminate certain paths that might otherwise be considered viable
candidates for submission to the path validation process. There may be other path
validation criteria that a vendor may want to consider as well (e.g., path length constraints). The idea is that you can throw out certificates that will not meet a subset of the
path validation criteria. This helps to promote a "prune as you go" process, particularly
when building paths in the reverse direction.

The idea is that you can throw

out certificates that will not meet
a subset of the path validation
criteria. This helps to promote a
"prune as you go" process, particularly when building paths in
the reverse direction.

However, there are potential drawbacks that may be associated with this. For example,
there are certain things that cannot be determined until the entire candidate path is
constructed (e.g., a require explicit policy requirement could be introduced by the last
intermediate CA certificate in the path). In addition, this pre-supposes that the path
construction logic will be able to process at least a subset of the path validation logic as
outlined in the 4th edition of X.509 and the Internet Certificate and CRL Profile. This may
not be straightforward in the case where a vendor has separated the path construction
and path validation processes. Finally, we could potentially perform certain path validation checking that fails to eliminate a candidate path, even though the full path validation
procedure later determines that some other criterion was not met. This increases the
amount of work during path construction. It may also duplicate work, since many of the
same checks made during path construction may also be made during the path validation process. Nonetheless, it seems clear that certain factors should be taken into consideration during the path construction process. The extent to which these are implemented
are at the discretion of each vendor.
Multiple Paths
Even in our simpler examples, we saw that it is possible to have more than one candidate
path between the relying party's trust anchor and the target end-entity certificate. For
example, in the single cross-certification example (using Figure 5), there were two candidate certification paths between User3's trust anchor and User2's certificate (i.e., CA7>CA5->CA0->CA2->CA4->User2 and CA7->CA6->CA5->CA0->CA2->CA4->User2).
It is possible that multiple candidate paths would meet all of the certification path evaluation criteria equally. In the absence of any other criteria that may be used to prioritize
one candidate path over another, it would make sense to submit the shortest path to the
path validation logic before submitting longer paths.

In the absence of any other

criteria that may be used to
prioritize one candidate path
over another, it would make
sense to submit the shortest path
to the path validation logic
before submitting longer paths.

Detecting Loops
One of the things we need to avoid when constructing certification paths is looping. This
could occur when a series of cross-certificates leads us back to a cross-certificate that is
already part of the candidate certification path. In a richly interconnected environment,
this possibility may often be encountered.
The 4th edition of X.509 makes it clear that a certificate is never permitted to appear more
than once in any given certification path (see Section 10.1, sub-paragraph a). We can
therefore avoid loops by keeping track of all certificates as the path is constructed and
making sure that any certificate already included in the prospective path does not appear
in the path again. If we encounter a duplicate, we can backtrack as necessary. Note that

PKI Forum: Understanding Certification Path Construction: September 2002

13

keeping track of subject DNs of intermediate CAs is not an appropriate duplicate detection mechanism, since a CA may have more than one active public key certificate issued
under the same DN. We must check to make sure the exact same certificate is not used
again, which could be accomplished through direct certificate comparison.

Summary
This paper discusses many of the issues that should be taken into consideration when
developing or evaluating certification path construction algorithms.
A number of observations and/or recommendations have been made throughout this
paper. In summary, a robust path construction algorithm should be able to:

Copyright Statement
This White Paper and all other
materials printed by the PKI
Forum are property of the PKI
Forum and may not be copied
without express consent from the
PKI Forum. 2002 PKI Forum,
Inc.

construct certification paths in both the forward and reverse directions,

search repositories for appropriate certificates in accordance with the 4th edition
of X.509,
recognize the AIA extension and use it to retrieve CA certificates,
use RDN matching to identify more likely candidate certification paths,
give precedence to shorter paths over longer paths (in the absence of any other
overriding criteria), and
detect and avoid duplicate certificates.

In addition, a number of path validation criteria might be taken into consideration during the path construction process, including those discussed in [Ref3].

References
[Ref1] ITU-T Recommendation X.509 Information TechnologyOpen
Systems InterconnectionThe Directory: Public Key and Attribute
Certificate Frameworks, March 2000 (equivalent to ISO/IEC 9594-8,
2000).
[Ref2] Internet X.509 Public Key Infrastructure: Certificate and CRL Profile,
Internet Request for Comments 3280, Housley, R., W. Ford, W. Polk, and
D. Solo., April 2002.

About PKI Forum

The PKI Forum is an international,
not-for-profit, multi-vendor and
end-user alliance whose purpose is
to accelerate the adoption and
use of Public-Key Infrastructure
(PKI).
The PKI Forum advocates cooperation and market awareness to
enable organizations to understand and exploit the value of PKI
in applications relevant to their
business.

[Ref3] Building Certification Paths: Forward vs. Reverse, Yassir Elley, Anne
Anderson, Steve Hanna, Sean Mullen, Radia Perlman, Seth Proctor (See
http://www.isoc.org/isoc/conferences/ndss/01/2001/papers/elley.pdf).
[Ref4] Path Development API Interface Control Document, Cygnacom
Solutions (See http://www.entrust.com/entrustcygnacom/cert/
CPL_1_3_ICD.doc).
[Ref5] Troubleshooting Certificate Status and Revocation, Microsoft Technical
Paper by Brian Komar (See http://www.microsoft.com/technet/treeview/
default.asp?url=/technet/security/prodtech/tshtcrl.asp).

Web: http://www.pkiforum.org
e-Mail: [email protected]

14 Path Construction: September 2002

PKI Forum: Understanding Certification

14Path Construction: September 2002

PKI Forum: Understanding Certification