Resource-Oriented Lightweight Indicator ExchangeEMC Corporation1133 Westchester AvenueWhite PlainsNew YorkUSA914-461-3522johnp.field@emc.com
Security
MILE Working GroupThis document defines a resource-oriented approach to cyber security information sharing.
Using this approach, a CSIRT or other stakeholder may share and exchange representations
of cyber security incidents, indicators, and other related information as Web-addressable resources.
The transport protocol binding is specified as HTTP(S) with a MIME media type of Atom+XML.
An appropriate set of link relation types specific to cyber security information sharing is defined.
The resource representations leverage the existing IODEF and RID
specifications as appropriate. Coexistence with deployments that conform to existing
specifications including RID and
Transport of Real-time Inter-network Defense (RID) Messages over HTTP/TLS is
supported via appropriate use of HTTP status codes. This document defines a resource-oriented approach to cyber security information sharing that follows the
REST
architectural style. The resource representations leverage the existing IODEF and RID
specifications as appropriate. The transport protocol binding is specified as HTTP(S) with a
media type of Atom+XML. An appropriate set of link relation types specific to cyber security
information sharing is defined. Using this approach, a CSIRT or other stakeholder may exchange cyber
security incident and/or indicator information as Web-addressable resources.
The goal of this specification is to define a loosely-coupled, agile approach to cyber security
situational awareness. This approach has architectural advantages for some use case scenarios, such as when a CSIRT
or other stakeholder is required to share cyber security information broadly (e.g., at internet scale), or
when an information sharing consortium requires support for asymmetric interactions amongst their stakeholders.
Coexistence with deployments that conform to existing specifications
including RID and Transport of
Real-time Inter-network Defense (RID) Messages over HTTP/TLS is supported via
appropriate use of HTTP status codes. The key words "MUST," "MUST NOT," "REQUIRED," "SHALL," "SHALL NOT," "SHOULD," "SHOULD
NOT," "RECOMMENDED," "MAY," and "OPTIONAL" in this document are to be interpreted as
described in . Definitions for some of the common computer
security-related terminology used in this document can be found in Section 2 of . It is well known that Internet security threats are evolving ever more rapidly, and are
becoming ever more sophisticated than before. The threat actors are frequently distributed
and are not constrained to operating within a fixed, closed consortium.
The technical skills needed to perform effective analysis of a security incident,
or to even recognize an indicator of compromise are already
specialized and relatively scarce. As threats continue to evolve, even an established
network of CSIRT may find that it does not always have all of the skills and knowledge required to
immediately identify and respond to every new incident.
Effective identification of and response to a sophisticated, multi-stage attack frequently depends upon cooperation and
collaboration, not only amongst the defending CSIRTs, but also amongst other stakeholders,
including, potentially, individual end users.
Existing approaches to cyber security information sharing
are based upon message exchange patterns that are point-to-point, and event-driven.
Sometimes, information that may be useful to, and sharable with multiple peers is only made available to peers
after they have specifically requested it. Unfortunately, a sharing peer may not know, a priori, what information
to request from another peer. Sending unsolicited RID reports does provide a mechanism for alerting,
however these reports are again sent point-to-point, and must be reviewed for relevance and then prioritized
for action by the recipient. Thus, distribution of some relevant incident and indicator information may exhibit
significant latency.
In order to appropriately combat the evolving threats, the defending CSIRTs should be enabled to
operate in a more agile manner, sharing selected cyber security information proactively, if and as appropriate.
For example, a CSIRT analyst would benefit by having the ability to search a comprehensive collection of indicators
that has been published by a government agency, or by another member of a sharing consortium. The representation of each
indicator may include links to the related resources, enabling an appropriately authenticated and authorized analyst to
freely navigate the information space of indicators, incidents, and other cyber security domain concepts, as needed.
In general, a more Web-centric sharing approach will enable a more dynamic and agile collaboration amongst a broader, and varying constituency.
The following sections discuss additional specific technical
issues that motivate the development of an alternative approach. The existing approaches to cyber security information sharing are based upon
message-oriented interactions. The following paragraphs explore some of the architectural
constraints associated with message-oriented interactions and consider the relative merits
of an alternative model based on a Resource-oriented architecture for use in some use case
scenarios. In general, message-based integration architectures may be based upon either an
RPC-style or a document-style binding. The message types defined by RID represent an
example of an RPC-style request. This approach imposes implied requirements for
conversational state management on both of the communicating RID endpoint(s). Experience
has shown that this state management frequently becomes the limiting factor with respect
to the runtime scalability of an RPC-style architecture. In
addition, the practical scalability of a peer-to-peer message-based approach will be
limited by the administrative procedures required to manage O(N^2) trust relationships
and at least O(N) policy groups. As long as the number of
CSIRTs participating in an information sharing consortium is limited to a relatively
smaller number of nodes (i.e., O(2^N), where N < 5), these scalability constraints
may not represent a critical concern. However, when there is a requirement to support a
significantly larger number of participating peers, a different architectural approach
will be required. One alternative to the message-based approach that has demonstrated
scalability is the REST architectural style. Applying the REST architectural style to the problem domain of cyber security
information sharing would take the approach of exposing incidents, indicators, and any
other relevant types as simple Web-addressable resources. By using this approach, a
CSIRT or other organization can more quickly and easily share relevant incident and
indicator information with a much larger and potentially more diverse constituency. A
client may leverage virtually any available HTTP user agent in order to make requests of
the service provider. This improved ease of use could enable more rapid adoption and
broader participation, thereby improving security for everyone.
A key interoperability aspect of any RESTful Web service will be the choices regarding
the available resource representations. For example, clients may request that a given
resource representation be returned as either XML or JSON. In order to enable
back-compatibility and interoperability with existing CSIRT implementations,
IODEF is specified for this transport binding as a mandatory to implement (MTI)
data representation for incident and indicator resources.
In addition to the REQUIRED representation, an implementation MAY support additional
representations if and as needed such as IODEF extensions, the RID schema, or other schemas.
For example, an implementation may choose to provide support for returning a
JSON representation of an incident resource.
Finally, an
important principle of the REST architectural style is the use of hypertext links as the
embodiment of application state (HATEOAS). Rather than the server maintaining
conversational state for each client context, the server will instead include a suitable
set of hyperlinks in the resource representation that is returned to the client. In this
way, the server remains stateless with respect to a series of client requests. The
included hyperlinks provide the client with a specific set of permitted state
transitions. Using these links the client may perform an operation, such as updating or
deleting the resource representation. The client may also be provided with hypertext
links that can be used to navigate to any related resource. For example, the resource
representation for an incident object may contain links to the related indicator
resource(s). This document specifies the use of
Atom Syndication Format and
Atom Publishing Protocol as
the mechanism for representing the required hypertext links. In this section we consider a non-normative example use case scenario for
creating a cyber security "mashup". Any CSIRT can enable any
authenticated and authorized client that is a member of the sharing community to quickly and easily navigate through any of the
cyber security information that that provider is willing to share. An authenticated
and authorized analyst may then make HTTP(S) requests to collect incident and indicator
information known at one CSIRT with threat actor data being made available from
another CSIRT. The resulting correlations may yield new insights that enable a more
timely and effective defensive response. Of course, this report may, in turn, be made
available to others as a new Web-addressable resource, reachable via another URL. By
employing the RESTful Web service approach the effectiveness of the collaboration
amongst a consortium of CSIRTs and their stakeholders can be greatly improved. In the store-and-forward, message-based model for information sharing client
authentication is provided via a Public Key Infrastructure (PKI) -based trust and mutually authenticated TLS between the
messaging system endpoints. There is no provision to support authentication of a client by
another means. As a result, participation in the sharing community is limited to those
organizations that have sufficient resources and capabilities to manage a PKI. A CSIRT may apply XML Security to the content of a message, however the
contact information provided within the message body represents a self-asserted identity,
and there is no guarantee that the contact information will be recognized by the peer. As
a result, the audit trail and the granularity of any authorization policies is limited to
the identity of the peer CSIRT organization. A CSIRT implementing
this specification MUST implement server-authenticated TLS. The CSIRT may choose to
authenticate its client users via any suitable authentication scheme that can be
implemented via HTTP(S). A participating CSIRT MAY choose to support more than one
authentication method. Support for use of a Federated Identity approach is RECOMMENDED.
Establishing a specific end user identity prior to processing a request is RECOMMENDED.
Doing so will enable the source system to maintain a more complete audit trail of exactly
what cyber security incident and indicator information has been shared, when, and with
whom. A key aspect of any cyber security information sharing arrangement is assigning the
responsibility for authorization policy enforcement. The authorization policy must be
enforced either at the destination system, or the source system, or both. The following
sections discuss these alternatives in greater detail. The store-and-forward, message-based approach to cyber security information sharing
requires that the origin system delegate authorization policy enforcement to the
destination system. The origin system may leverage XML Encryption and DigitalSignature
to protect the message content. In addition, the origin system assigns a number of
policy-related attribute values, including a "restriction" attribute, before the message
is sent. These labels indicate the sender's expectation for confidentiality enforcement
and appropriate handling at the destination. Section 9.1 of RFC6545 provides specific guidance
to implementers on use of the XML security standards in order to achieve the required levels of
security for the exchange of incident information.
Once the message has been received at the
destination system, the XML encryption and digital signature protections on the message
will be processed, and based upon the pre-established PKI-based trust relationships, the
message content is validated and decrypted. Typical implementations will then pass the
cleartext data to an internal Incident Handling System (IHS) for further review and/or
action by a human operator or analyst. Regardless of where in the deployment
architecture the XML message-level security is being handled, eventually the message
content will be made available as cleartext for handling by human systems analysts and
other operational staff. The authorization policy enforcement
of the message contents must then be provided by the destination IHS. It is the
responsibility of the destination system to honor the intent of the policy restriction
labels assigned by the origin system. Ideally, these policy labels would serve as part
of a distributed Mandatory Access Control scheme. However, in practice a typical IHS
will employ a Discretionary Access Control (DAC) model rather than a MAC model and so
the policy related attributes are defined to represent handling "hints" and provide no
guarantee of enforcement at the destination. As a result,
ensuring that the destination system or counterparty will in fact correctly enforce the
intended authorization policies becomes a key issue when entering into any information
sharing agreements. The origin CSIRT must accept a non-zero risk of information leakage,
and therefore must rely upon legal recourse as a compensating control. Establishing such
legal sharing agreements can be a slow and difficult process, as it assumes a high level
of trust in the peer, with respect to both intent and also technical capabilities. In this model, the required authorization policy enforcements are implemented entirely
within the source system. Enforcing the required authorization policy controls at the
source system eliminates the risk of subsequent information leakage at the destination
system due to inadequate or incomplete implementation of the expected controls. The
destination system is not expected to perform any additional authorization enforcements.
Authorization enforcement at the source system may be based on, e.g. Role-based Access
Controls applied in the context of an established user identity. The source system may
use any appropriate authentication mechanism in order to determine the user identity of
the requestor, including, e.g. federated identity. An analyst or operator at a CSIRT may
request specific information on a given incident or indicator from a peer CSIRT, and the
source system will return a suitable representation of that resource based upon the
specific role of the requestor. A different authenticated user (perhaps from the same
destination CSIRT) may receive a different representation of the same resource, based
upon the source system applying suitable Role-based Access Control policy enforcements
for the second user identity.
Consistent with HTTP a user's
request MAY be denied with a resulting HTTP status code value of 4xx such as 401 Unauthorized,
403 Forbidden, or 404 Not Found, or 405 Method Not Allowed, if and as appropriate.
This section describes the basic use of Atom Syndication Format and
Atom Publishing Protocol as a RESTful transport binding and
dynamic discovery protocol, respectively, for cyber security information sharing.
As described in Atom Publishing Protocol, an Atom Service Document is an XML-based document format that allows a
client to dynamically discover the collections provided by a publisher.
As described in Atom Syndication Format, Atom is an XML-based document format
that describes lists of related information items known as collections, or "feeds". Each feed document contains a collection of zero or more
related information items called "member entries" or "entries".
When applied to the problem domain of cyber security information sharing, an Atom feed
may be used to represent any meaningful collection of information resources such as a set of incidents, or indicators.
Each entry in a feed could then represent an individual incident, or indicator, or some other resource, as appropriate.
Additional feeds could be used to represent other meaningful and useful collections of cyber security resources.
A feed may be categorized, and any feed may contain information from zero or more categories.
The naming scheme and the semantic meaning of the terms used to identify an Atom category are application-defined.
In order to specify a protocol for cyber security information sharing using the REST architectural style
it is necessary to define the set of resources to be modeled, and how these resources are related.
Based on this interface contract, clients will then interact with the REST service by navigating the
modeled entities, and their relationships. The interface contract between the client and the server may
either be statically bound or dynamically bound.
In the statically bound case, the clients have a priori knowledge of the resources that are supported.
In the REST architectural style this static interface contract takes the form of a URL template.
This approach is not appropriate for the cyber security information sharing domain for at least two reasons.
First, there is no standard for a cyber security domain model. While information security practitioners can
generally agree on some of the basic concepts that are important to modeling the cyber security domain --
such as "indicator,” "incident,” or “attacker,” -- there is no single domain model that can been referenced
as the basis for specifying a standardized RESTful URI Template. Second, the use of static URL templates creates
a tighter coupling between the client implementation and the server implementation. Security threats on the
internet are evolving ever more rapidly, and it will never be possible to establish a statically defined resource
model and URL Template. Even if there were an initial agreement on an appropriate URL template, it would eventually
need to change. If and when a CSIRT finds that it needs to change the URL template, then any existing deployed clients
would need to be upgraded.
Thus, rather than attempting to define a fixed set of resources via a URI Template, this document has instead specified
an approach based on dynamic discovery of resources via an Atom Publishing Protocol Service Document. By using this approach,
it is possible to standardize the RESTful usage model, without needing to standardize on the definitions of specific,
strongly-typed resources. A client can dynamically discover what resources are provided by a given CSIRT, and then
navigate that domain model accordingly A specific server implementation may still embody a particular URL template,
however the client does not need a priori knowledge of the format of the links, and the URL itself is effectively opaque
to the client. Clients are not bound to any particular server’s interface.
The following paragraphs provide a number of non-normative examples to illustrate the use of Atom Publishing Protocol for basic
cyber security information sharing service discovery, as well as the use of Atom Syndication Format as a mechanism to
publish cyber security information feeds.
Normative requirements are defined below, in .
This section provides a non-normative example of a client doing service discovery.
An Atom service document enables a client to dynamically discover what feeds a particular publisher makes available.
Thus, a CSIRT may use an Atom service document to enable clients of the CSIRT to determine what specific cyber security information
the CSIRT makes available to the community. The service document could be made available at any well known location, such as via a link
from the CSIRT's home page. One common technique is to include a link in the <HEAD> section of the organization's
home page, as shown below:
This section provides a non-normative example of a client retrieving an incident feed.
Having discovered the available cyber security information sharing feeds, an authenticated and authorized client who is a member of the
private sharing consortium may be interested in receiving the feed of known incidents.
The client may retrieve this feed by performing an HTTP GET operation on the indicated URL.
This section provides a non-normative example of a client retrieving an incident as an Atom entry.
Having retrieved the feed of interest, the client may then decide based on the description and/or category information that one of the
entries in the feed is of further interest. The client may retrieve this incident Entry by performing an HTTP GET operation on the
indicated URL.
Note also that, as described previously, the content of the Atom <category> element is application-defined.
In the present context, the Atom categories have been assigned based on a mapping of the <restriction> and
<purpose> attributes, as defined in the IODEF schema.
In addition, the IODEF <incidentID> element has been judiciously chosen so that the associated name attribute, as well as the
corresponding incidentID value, can be concatenated in order to easily create the corresponding <id> element for the Atom entry.
These and other mappings are normatively defined in , below.
Finally, it should be noted that in order to optimize the client experience, and avoid an additional round trip, a feed provider
may choose to include the entry content inline, as part of the feed document. That is,
an Atom <entry> element within a Feed document may contain an Atom <content> element as a child. In this
case, the client will receive the full content of the entries within the feed. The decision
of whether to include the entry content inline or to include it as a link is a design choice left to the feed provider
(e.g. based upon local environmental factors such as the number of entries contained in a feed, the available network bandwidth, the
available server compute cycles, the expected client usage patterns, etc.).
As noted previously, a key benefit of using the RESTful architectural style is the ability to enable the client to navigate to
related resources through the use of hypermedia links. In the Atom Syndication Format, the type of the related
resource identified in a <link> element is indicated via the "rel" attribute, where the value of this attribute
identifies the kind of related resource available at the corresponding "href" attribute. Thus, in lieu of a well-known
URI template the URI itself is effectively opaque to the client, and therefore the client must understand the semantic
meaning of the "rel" attribute in order to successfully navigate. Broad interoperability may be based upon a sharing
consortium defining a well-known set of Atom Link Relation types. These Link Relation types may either be registered
with IANA, or held in a private registry.
Individual CSIRTs may always define their own link relation types in order to support specific use cases, however support
for a core set of well-known link relation types is encouraged as this will maximize interoperability.
In addition, it may be beneficial to define use case profiles that correspond to specific groupings of supported link
relationship types. In this way, a CSIRT may unambiguously specify the classes of use cases for which a client
can expect to find support.
The following sections provide NON-NORMATIVE examples of link relation usage.
Four distinct cyber security information sharing use case scenarios are described. In each use case, the unique
benefits of adopting a resource-oriented approach to information sharing are illustrated.
It is important to note that these use cases are intended to be a small representative set and is by no means meant
to be an exhaustive list. The intent is to illustrate how the use of link relationship types will enable this
resource-oriented approach to cyber security information sharing to successfully support the complete range of
existing use cases, and also to motivate an initial list of well-defined link relationship types.
This section provides a non-normative example of an incident sharing use case.
In this use case, a member CSIRT shares incident information with another member CSIRT in the same consortium.
The client CSIRT retreives a feed of incidents, and is able to identify one particular entry of interest.
The client then does an HTTP GET on that entry, and the representation of that resource contains link relationships
for both the associated "indicators" and the incident "history", and so on. The client CSIRT recognizes that
some of the indicator and history may be relevant within her local environment, and can respond proactively.
This section provides a non-normative example of a collaborative investigation use case.
In this use case, two member CSIRTs that belong to a closed sharing consortium are collaborating on an incident
investigation. The initiating CSIRT performs an HTTP GET to retrieve the service document of the peer CSIRT,
and determines the collection name to be used for creating a new investigation request. The initiating CSIRT then
POSTs a new incident entry to the appropriate collection URL. The target CSIRT acknowledges the request by responding
with an HTTP status code 201 Created.
This section provides a non-normative example of a search use case.
The following example provides a RESTful alternative to the RID Query messaage, as described in
sections 6.5 and 7.4 of RFC6545. Note that in the RESTful approach described herein there is no requirement to
define a query language specific to RID queries. Instead, CSIRTs may provide support for search operations via
existing search facilities, and advertise these capabilities via an appropriate URL template. Clients
dynamically retrieve the search description document, and invoke specific searches via an instantiated URL template.
An HTTP response body may include a link relationship of type "search."
This link provides a reference to an OpenSearch description document.
This OpenSearch Description Document also contains an example of a <Query> element.
Each <Query> element describes a specific search request that can be made by the client.
Note that the parameters of the <Query> element correspond to the URL template parameters.
In this way, a provider may fully describe the search interface available to the clients.
Section 5.12, below, provides specific NORMATIVE requirements for the use of Open Search.
This section provides a non-normative example of a cyber security data repository use case.
In this use case a client accesses a persistent repository of cyber security data via a RESTful usage model.
Retrieving a feed collection is analogous to an SQL SELECT statement producing a result set.
Retrieving an individual Atom Entry is analogous to a SQL SELECT statement based upon a primary key producing a unique record.
The cyber security data contained in the repository may include different data types, including indicators, incidents, becnmarks,
or any other related resources. In this use case, the repository is queried via HTTP GET, and the results that are returned to
the client may optionally contain URL references to other cyber security resources that are known to be related.
These related resources may also be persisted locally, or they may exist at another (remote) cyber data respository.
Note that the provider of a persistent repostory is not obligated to follow any particular URL template scheme.
The repository available at the hypothetical provider "www.example.com" uses a different URL pattern than the hypothetical
repository available at "www.cyber-agency.gov". When a client de-references a link to resource that is located in a remote
repository the client may be challenged for authentication credentials acceptable to that provider. If the two repository
providers choose to support a federated identity scheme or some other form of single-sign-on technology, then the user experience can
be improved for interactive clients (e.g., a human user at a browser). However, this is not required and is an
implementation choice that is out of scope for this specification.
This section provides the NORMATIVE requirements for using Atom format and Atom Pub as a RESTful binding for
cyber security information sharing.Servers implementing this specification MUST support server-authenticated TLS.
Servers MAY support mutually authenticated TLS.Servers MUST require user authentication.
Servers MAY support more than one client authentication method.
Servers participating in an information sharing consotium and supporting interactive user logins by members of the consortium SHOULD
support client authentication via a federated identity scheme as per SAML 2.0.
Servers MAY support client authenticated TLS. This document does not mandate the use of any specific user authorization mechanisms. However, service implementers SHOULD provide appropriate
authorization checking for all resource accesses, including individual Atom Entries, Atom Feeds, and Atom Service Documents.
Authorization for a resource MAY be adjudicated based on the value(s) of the associated Atom <category> element(s).
When the content model for the Atom <content> element of an Atom Entry contains an <IODEF-Document>, then authorization MUST be adjudicated
based upon the Atom <category> element(s), whose values have been mapped as per .
Any use of the <category> element(s) as an input to an authorization policy decision MUST include both the "scheme" and "term"
attributes contained therein. As described in below, the namespace of the "term" attribute is scoped
by the associated "scheme" attribute.
Member entry resources providing a representation of an incident resource (e.g., as specified in the link relation type)
MUST use the IODEF schema as the content model for the Atom Entry <content> element.
Member Entry resources providing a representation of an indicator resource (e.g., as specified in the link relation type)
MUST use the IODEF schema as the content model for the Atom Entry <content> element.
The resource representation MAY include an appropriate indicator schema type within the <AdditionalData> element of the IODEF Incident class.
Supported indicator schema types SHALL be registered via an IANA table (todo: IANA registration/review).
Member Entry resources providing a representation of a RID report resource (e.g., as specified in the link relation type)
MUST use the RID schema as the content model for the Atom Entry <content> element.
Member Entry resources providing representation of other types, SHOULD use the IODEF schema as the content model
for the Atom Entry <content> element.
If the member entry content model is not IODEF, then the <content> element of the Atom entry MUST
contain an appropriate XML namespace declaration.
The following table defines the HTTP uniform interface methods supported by this specification:
HTTP methodDescriptionGETReturns a representation of an individual member entry resource, or a feed collection. PUTReplaces the current representation of the specified member entry resource with the representation provided in the HTTP
request body.POSTCreates a new instance of a member entry resource. The representation of the new resource is provided in the
HTTP request body.DELETERemoves the indicated member entry resource, or feed collection.HEADReturns metadata about the member entry resource, or feed collection, contained in HTTP response headers.PATCHSupport TBD.Clients MUST be capable of recognizing and prepared to process any standard HTTP status code, as defined
in This specification requires that a CSIRT MUST publish an Atom Service Document that describes the set of cyber security information sharing
feeds that are provided.
The service document SHOULD be discoverable via the CSIRT organization's Web home page or another well-known public resource.
The service document MAY include multiple workspaces.
Any CSIRT providing both public feeds and private consortium feeds MUST
place these different classes of feeds into different workspaces, and provide appropriate descriptions and naming conventions to indicate
the intended audience of each workspace.A CSIRT MAY provide any number of collections within a given Workspace.
It is RECOMMENDED that each collection appear in only a single Workspace.
It is RECOMMENDED that at least one collection be provided that accepts new incident reports from users.
At least one collection MUST provide a feed of incident information for which the content model for the entries uses the IODEF schema.
The title of this collection SHOULD be "Incidents".
Access to the service document MUST be protected via server-authenticated TLS and a server-side certificate.
When deploying a service document for use by a closed consortium, the service document MAY also be digitally signed and/or encrypted,
using XML DigSig and/or XML Encryption, respectively.
This section defines normative requirements for mapping IODEF metadata to corresponding Atom category elements.
(todo: decide between IANA registration of scheme, or use a full URI).
An Atom collection MAY hold entries from one or more categories.
The collection category set MUST contain at least the union of all the member entry categories.
A collection MAY have additional category metadata that are unique to the collection, and not applicable to any individual member entry.
A collection containing IODEF incident content MUST contain at least two <category> elements.
One category MUST be specified with the value of the "scheme" attribute as "restriction".
One category MUST be specified with the value of the "scheme" attribute as "purpose".
The value of the "fixed" attribute for both of these category elements MUST be "yes".
When the category scheme="restriction", the allowable values for the "term" attribute are constrained as per section 3.2 of IODEF,
e.g. public, need-to-know, private, default.
When the category scheme="purpose", the allowable values for the "term" attribute are constrained as per section 3.2 of IODEF,
e.g. traceback, mitigation, reporting, other.
An Atom entry containing IODEF content MUST contain at least two <category> elements.
One category MUST be specified with the value of the "scheme" attribute as "restriction".
One category MUST be specified with the value of the "scheme" attribute as "purpose".
When the category scheme="restriction", the value of the "term" attribute must be exactly one of ( public, need-to-know, private, default).
When the category scheme="purpose", the value of the "term" attribute must be exactly one of (traceback, mitigation, reporting, other).
When the purpose is "other"....
Any member entry MAY have any number of additional categories.
The ID element for an Atom entry SHOULD be established via the concatenation of the value of the name attribute from the IODEF <IncidentID> element and
the corresponding value of the <IncidentID> element.
This requirement ensures a simple and direct one-to-one relationship between an IODEF incident ID and a corresponding Feed entry ID and avoids the need for any system
to maintain a persistent store of these identity mappings.
(todo: Note that this implies a constraint on the IODEF document that is more restrictive than the current IODEF schema.
IODEF section 3.3 requires only that the name be a STRING type.
Here we are stating that name must be an IRI. Possible request to update IODEF to constrain, or to support a new element or attribute).
The <content> element of an Atom <entry> SHOULD include an IODEF document.
The <entry> element SHOULD include an appropriate XML namespace declaration for the IODEF schema.
If the content model of the <entry> element does not follow the IODEF schema, then the <entry> element MUST
include an appropriate XML namespace declaration.
A client MAY ignore content that is not using the IODEF schema.
In addition to the standard Link Relations defined by the Atom specification, this specification defines the following
additional Link Relation terms, which are introduced specifically in support of the
Resource-Oriented Lightweight Indicator Exchange protocol.
NameDescriptionConformanceserviceProvides a link to an atom service document associated with the collection feed.MUSTsearchProvides a link to an associated Open Search document that describes a URL template for search queries.MUSThistoryProvides a link to a collection of zero or more historical entries that are associated with the resource.MUSTincidentsProvides a link to a collection of zero or more instances of actual cyber security event(s) that are associated with the resource.MUSTindicatorsProvides a link to a collection of zero or more instances of cyber security indicators that are associated with the resource.MUSTevidenceProvides a link to a collection of zero or more resources that provides some proof of attribution for an incident. The evidence
may or may not have any identified chain of custody.SHOULDcampaignProvides a link to a collection of zero or more resources that provides a representation of the associated cyber attack campaign.SHOULDattackerProvides a link to a collection of zero or more resources that provides a representation of the attacker.SHOULDvectorProvides a link to a collection of zero or more resources that provides a representation of the method used by the attacker.SHOULDassessmentsProvides a link to a collection of zero or more resources that represent the results of executing a benchmark.SHOULDreportsProvides a link to a collection of zero or more resources that represent RID reports.SHOULDtraceRequestsProvides a link to a collection of zero or more resources that represent RID traceRequests.SHOULDinvestigationRequestsProvides a link to a collection of zero or more resources that represent RID investigationRequests.SHOULD
Unless specifically registered with IANA these short names MUST be fully qualified via concatenation with a base-uri.
An appropriate base-uri could be established via agreement amongst the members of an information sharing consortium.
For example, the rel="indicators" relationship would become rel="http://www.example.org/csirt/incidents/relationships/indicators."
An IODEF document that is carried in an Atom Entry SHOULD NOT contain a <relatedActivity> element.
Instead, the related activity SHOULD be available via a link rel=related.
An IODEF document that is carried in an Atom Entry SHOULD NOT contain a <history> element.
Instead, the related history SHOULD be available via a link rel="history" (todo: or a fully qualified link rek name).
The associated href MAY leverage OpenSearch to specify the required query.
An Atom Entry MAY include additional link relationships not specified here. If a client encounters a link relationship of an unkown type
the client MUST ignore the offending link and continue processing the remaining resource representation as if the offending link element
did not appear.
As described in Authorization Policy Enforcement a RESTful model for cyber security information sharing
requires that all of the required security enforcement for feeds and entries MUST be enforced at the source system, at the point the representation of the
given resource(s) is created. A CSIRT provider SHALL NOT return any feed content or member entry content for which the client identity has not been
specifically authenticated, authorized, and audited.
Sharing communities that have a requirement for forward message security (such that client systems are required to participate in providing message level
security and/or distributed authorization policy enforcement), MUST use the RID schema as the content model for the member entry <content> element.
The Atom feed <updated> element MUST be populated with the current time at the instant the feed representation was generated.
The Atom entry <published> element MUST be populated with the same time value as the <reportTime> element from the IODEF document.
Implementers MUST support OpenSearch 1.1
as the mechanism for describing how clients may form search requests.
Implementers MUST provide a link with a relationship type of "search". This link SHALL return an Open Search Description Document
as defined in OpenSearch 1.1.
Implementers MUST support an OpenSearch 1.1 compliant search URL template that enables a search query via Atom Category, including
the scheme attribute and terms attribute as search parameters.
Implementers SHOULD support search based upon the IODEF AlternativeID class as a search parameter.
Implementers SHOULD support search based upon the four timestamp elements of the IODEF Incident class: <startTime>, <EndTime>,
<DetectTime>, and <ReportTime>.
Implementers MAY support additional search capabilities based upon any of the remaining elements of the IODEF Incident class,
including the <Description> element.
Collections that support use of the RID schema as a content model in the Atom member entry <content> element (e.g. in a report resource
representation reachable via the "report" link relationship) MUST support search operations that include the RID MessageType as a search parameter,
in addition to the aforementioned IODEF schema elements, as contained within the <ReportSchema> element.
Implementers MUST fully qualify all OpenSearch URL template parameter names using the defined IODEF or RID XML namespaces, as appropriate.
The "/" resource MAY be provided for compatibility with existing deployments that are using
Transport of Real-time Inter-network Defense (RID) Messages over
HTTP/TLS. Consistent with RFC6546 errata, a client requesting a GET on "/" MUST
receive an HTTP status code 405 Method Not Allowed. An implementation MAY provide full
support for RFC6546 such that a POST to "/" containing a recognized RID message type just
works. Alternatively, a client requesting a POST to "/" MAY receive an HTTP status code
307 Temporary Redirect. In this case, the location header in the HTTP response will
provide the URL of the appropriate RID endpoint, and the client may repeat the POST method
at the indicated location. This resource could also leverage the new draft by reschke that
proposes HTTP status code 308 (cf: draft-reschke-http-status-308-07.txt). This document defines a resource-oriented approach to lightweight indicator exchange using HTTP, TLS, Atom Syndicate Format,
and Atom Publishing Protocol. As such, implementers must understand the security considerations described in those specifications.
In addition, there are a number of additional security considerations that are unique to this specification.
As described above in the section Authentication of Users,
the approach described herein is based upon all policy enforcements being implemented at the point when a resource representation is created.
As such, CSIRTS sharing cyber security information using this specification must take care to
authenticate their HTTP clients using a suitably strong user authentication mechanism.
Sharing communities that are exchanging information on well-known indicators and incidents for purposes of
public education may choose to rely upon, e.g. HTTP Authentication, or similar.
However, sharing communities that are engaged in sensitive collaborative analysis and/or operational response for indicators and incidents
targeting high value information systems should adopt a suitably stronger user authentication solution, such as TLS client certificates,
or a risk-based or multi-factor approach. In general, trust in the sharing consortium will depend upon the members maintaining
adequate user authentication mechanisms.
Collaborating consortiums may benefit from the adoption of a federated identity solution, such as those based upon
SAML-core and SAML-bind and SAML-prof
for Web-based
authentication and cross-organizational single sign-on. Dependency on a trusted third party identity provider implies that appropriate
care must be exercised to sufficiently secure the Identity provider. Any attacks on the federated identity system would present a risk to
the CISRT, as a relying party.
Potential mitigations include deployment of a federation-aware identity provider that is under the control of the information sharing
consortium, with suitably stringent technical and management controls.
As discussed above in the section Authorization Policy Enforcement, authorization of resource representations
is the responsibility of the source system, i.e. based on the authenticated user identity associated with an HTTP(S) request.
The required authorization policies that are to be enforced must therefore be managed by the security administrators of the source system.
Various authorization architectures would be suitable for this purpose, such as RBAC
and/or ABAC, as embodied in XACML.
In particular, implementers adopting XACML may benefit from the capability to represent their authorization policies in a standardized,
interoperable format.
Additional security requirements such as enforcing message-level security at the destination system could supplement the security enforcements
performed at the source system, however these destination-provided policy enforcements are out of scope for this specification.
Implementers requiring this capability should consider leveraging, e.g. the <RIDPolicy> element in the RID schema.
Refer to RFC6545 section 9 for more information.
When security policies relevant to the source system are to be enforced at both the source and destination systems, implementers must
take care to avoid unintended interactions of the separately enforced policies. Potential risks will include unintended denial of service
and/or unintended information leakage. These problems may be mitigated by avoiding any dependence upon enforcements performed at the
destination system. When distributed enforcement is unavoidable, the usage of a standard language (e.g. XACML) for the expression of
authorization policies will enable the source and destination systems to better coordinate and align their respective policy expressions.
Adoption of the information sharing approach described in this document will enable users to more easily perform correlations across
separate, and potentially unrelated, cyber security information providers.
A client may succeed in assembling a data set that would not have been permitted within the context of the
authorization policies of either provider when considered individually.
Thus, providers may face a risk of an attacker obtaining an access that constitutes an undetected separation of duties (SOD) violation.
It is important to note that this risk is not unique to this specification, and a similar
potential for abuse exists with any other cyber security information sharing protocol. However, the wide availability of tools for
HTTP clients and Atom feed handling implies that the resources and technical skills required for a successful exploit may be less than it
was previously. This risk can be best mitigated through appropriate vetting of the client at account provisioning time.
In addition, any increase in the risk of this type of abuse should be offset by the corresponding increase in effectiveness that
that this specification affords to the defenders.
While it is a goal of this specification to enable more agile cyber security information sharing across a broader and varying constituency,
there is nothing in this specification that necessarily requires this type of deployment. A cyber security information sharing consortium
may chose to adopt this specification while continuing to operate as a gated community with strictly limited membership.
If the values of the newly defined link relations are not fully qualified URIs then we need to register these link types with IANA (e.g. rel="history")
It is possible to adjust this document so that it has no actions for IANA.The following is the "todo" and open issues list:
Need to make a decision on whether new IANA link registrations are required, or whether fully qualified (private) link types are sufficient.Should we require Atom categories that correspond to IODEF Expectation class and/or IODEF Impact class?Should we include specific requirements for Archive and Paging? Perhaps just reference RFC 5005?We need more requirements input on use cases involving RID schema in the Atom member entry content model for link rel=report.An Atom service document will have categories, but this is still coarse-grained, and not visible at the transport protocol level.
Should we include a MIME media type parameter to support negotiation and better document the content model schema contained in a collection, i.e.:
Accept: application/atom+xml;type=entry;content=iodef
Accept: application/atom+xml;type=entry;content=rid
Accept: application/atom+xml;type=entry;content=iodef+openioc
If so, I think these parameters may require media type registration as per RFC4288?The author gratefully acknowledges the valuable contributions of
Tom Maguire, Kathleen Moriarty, and Vijayanand Bharadwaj. These individuals provided detailed review comments on
earlier drafts, and many suggestions that have helped to improve this document .OpenSearch 1.1 draft 5 specification
OpenSearch Community
Assertions and Protocols for the OASIS
Security Assertion Markup Language
(SAML) V2.0 Profiles for the OASIS Security
Assertion Markup Language (SAML)
V2.0Bindings for the OASIS Security
Assertion Markup Language (SAML)
V2.0 XML Encryption Syntax and ProcessingXML-Signature Syntax and ProcessingeXtensible Access Control Markup Language (XACML) Version 3.0Architectural Styles and
the Design of Network-based Software Architectures
University of California, Irvine;
Department of Information and Computer Science
Changes since -00 version, September 5, 2012 to Feb 15, 2013:
Fixed a small number of typographical errors and a few misspellings throughout.Added a number of missing internal cross references to improve readability.Updated the text in the Introduction section for improved brevity and clarity of goal. See: Added new non-normative text describing the use of HTTP 4xx status codes for authorization. See: Added a new non-normative example illustrating a persistent repository use case. See: Added new normative text recommending use of SAML2 for authentication of interactive end users who are members of a sharing consortium. See: Added new normative text describing requirements for user authorization. See: Added non-normative appendix for change tracking. See: Added non-normative appendix describing a suggested approach to a XACML profile. See: As described in above, ROLIE assumes that all authorization policy enforcement is provided at the source server.
The implementation details of the authorization scheme chosen by a ROLIE-compliant provider are out of scope for this specification.
Implementers are free to choose any suitable authorization mechanism that is capable of fulfilling the policy enforcement requirements
relevant to their consortium and/or organization.
It is well known that one of the major barriers to information sharing is ensuring acceptable use of the information shared.
In the case of ROLIE, one way to lower that barrier may be to develop a XACML profile. Use of XACML would allow a ROLIE-compliant
provider to express their information sharing authorization policies in a standards-compliant, and machine-readable format.
This improved interoperability may, in turn, enable more agile interactions in the cyber security sharing community. For example,
a peer CSIRT, or another interested stakeholder such as an auditor, would be able to review and compare CSIRT sharing policies using
appropriate tooling.
The XACML 3.0 standard is based upon the notion that authorization policies are defined in terms of predicate logic expressions written against the
attributes associated with one or more of the following four entities:
SUBJECTACTIONRESOURCEENVIRONMENT
Thus, a suitable approach to a XACML 3.0 profile for ROLIE authorization policies could begin by using the 3-tuple of
[SUBJECT, ACTION, RESOURCE] where:
SUBJECT is the suitably authenticated identity of the requestor.ACTION is the associated HTTP method, GET, PUT, POST, DELETE, HEAD, (PATCH).RESOURCE is an XPath expression that uniquely identifies the instance or type of the ROLIE resource being requested.
Implementers who have a need may also choose to evaluate based upon the additional ENVIRONMENT factors, such as current threat level, and so on.
One could also write policy to consider the CVSS score associated with the resource, or the lifecycle phase of the resource (vulnerability
unverified, confirmed, patch available, etc.), and so on.
Having these policies expressed in a standards-compliant and machine-readable format could improve the agility and effectiveness of a cyber security
information sharing group or consortium, and enable better cyber defenses.
Work-in-Progress. If this aproach finds support in the community then this section (or a new draft, as a seperate document) could provide a more
complete XACML 3.0 compliant example.