| I2RS WG | D. Migault, Ed. |
| Internet-Draft | J. Halpern |
| Intended status: Informational | Ericsson |
| Expires: January 7, 2016 | S. Hares |
| Huawei | |
| July 6, 2015 |
I2RS Environment Security Requirements
draft-mglt-i2rs-security-environment-reqs-00
This document provides environment security requirements for the I2RS architecture. Environment security requirements are independent of the protocol used for I2RS. As a result, the requirements provided in this document are intended to provide good security practise so I2RS can be securely deployed and operated.
These security requirements are designated as environment security requirements as opposed to the protocol security requirements described in [I-D.hares-i2rs-auth-trans]. The reason to have separate document is that protocol security requirements are intended to help the design of the I2RS protocol.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on January 7, 2016.
Copyright (c) 2015 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This document addresses security considerations for the I2RS architecture. These requirements are also designated as environment security requirements. These security requirements are independent from the I2RS protocol used, and as such do not address requirements the I2RS protocol is expected to meet. The security requirement provided in this document are intended to provide guidance and security principles to guarantee the stability of the I2RS architecture. This documents provides an analysis of the security issues of the I2RS architecture beyond those already listed in [I-D.ietf-i2rs-architecture].
On the other hand, security requirements for the I2RS protocol design are described in a separate document [I-D.hares-i2rs-auth-trans].
Even though I2RS is mostly concerned by the interface between the I2RS Client and the I2RS Agent, the security recommendations must consider the entire I2RS architecture, specifying where security functions may be hosted, and what should be met so to address any new attack vectors exposed by deploying this architecture. In other words, security has to be considered globally over the complete I2RS architecture and not only on the interfaces.
I2RS architecture depicted in [I-D.ietf-i2rs-architecture] describes the I2RS components and their interactions to provide a programmatic interface for the routing system. I2RS components as well as their interactions have not yet been considered in conventional routing systems. As such it introduces a need to interface with the routing system designated as I2RS plane in this document.
This document is built as follows. Section 3 describes how the I2RS plane can be contained or isolated from existing management plane, control plane and forwarding plane. The remaining sections of the document focuses on the security within the I2RS plane. Section 4 analyzes how the I2RS Authentication Authorization and Access Control (I2RS AAA) can be deployed throughout the I2RS plane in order to only grant access to the routing system resources to authorized components with the authorized privileges. This also includes providing a robust communication system between the components. Then, Section 5 details how I2RS keeps applications isolated one from another and do not affect the I2RS components. Applications may be independent, with different scopes, owned by different tenants. In addition, they modify the routing system that may be in an automatic way.
The reader is expected to be familiar with the [I-D.ietf-i2rs-architecture]. The document provides a list of environment security requirements. Motivations are placed before the requirements are announced.
[QUESTION: Some suggested to use system instead of plane. Which is the more appropriate terminology?]
Isolating the I2RS plane from other network plane, such as the control plane, is foundational to the security of the I2RS environment. Clearly differentiating I2RS components from the rest of the network protects the I2RS components from vulnerabilities in other parts of the network, and protect other systems vital to the health of the network from vulnerabilities in the I2RS plane. Separating the I2RS plane from other network control and forwarding planes is similar to the best common practice of containerizing software into modules, and defense in depth in the larger world of network security.
That said the I2RS plane cannot be considered as completely isolated from other planes, and interactions should be identified and controlled. Follows a brief description on how the I2RS plane positions itself in regard to the other planes. The description is indicative, and may not be exhaustive.
The I2RS plane and the management plane both interact with several common elements on forwarding and packet processing devices. [I-D.ietf-i2rs-architecture] describes several of these interaction points such as the local configuration, the static system state, routing, and signalling. Because of this potential overlaps, a routing resource may be accessed by different means (APIs, applications) and different planes. To keep these overlaps under control, one could either control the access to these resources with northbound APIs for example. Northbound APIs are provided to limit the scope of the applications toward the routing resources. In our case, the northbound API may be provided for the I2RS applications by the I2RS Client as well as to the management plane. In case conflicting overlaps cannot be avoided, and routing resource can be accessed by both the management plane and the I2RS plane, then, they should be resolved in a deterministic way.
On the northbound side, there must be clear protections against the I2RS system "infecting" the management system with bad information, or the management system "infecting" the I2RS system with bad information. The primary protection in this space is going to need to be validation rules on the speed of information flow, value limits on the data presented, and other protections of this type.
On the conflicting side/issues, there should be clear rules about which plan's commands win in the case of conflict in order to prevent attacks where the two systems can be forced to deadlock.
Applications hosted on I2RS Client belongs to the I2RS plane, but remains hard to remain constrained into the I2RS plane, and even within the I2RS plane to have a limited scope.
Applications using I2RS are part of the I2RS plane but may also interact with other components outside the I2RS plane. A common example may be an application uses I2RS to configure the network according to security or monitored events. As these events are monitored on the forwarding plane and not the I2RS plane, the application breaks plane isolation.
In addition, applications may communicate with multiple I2RS Clients; as such, any given application may have a broader view of the current and potential states of the network and the I2RS plane itself. Because of this, any individual application could be an effective attack vector against the operation of the network, the I2RS plane, or any plane with which the I2RS plane interacts. There is little the I2RS plane can do to validate applications with which it interacts, other than to provide some broad general validations against common misconfigurations or errors. As with the separation between the management plane and the I2RS plane, this should minimally take the form of limits on information accepted, limits on the rate at which information is accepted, and rudimentary checks against intentionally formed routing loops or injecting information that would cause the control plane to fail to converge. Other forms of protection may be necessary.
The network control plane consists of the processes and protocols that discover topology, advertise reachability, and determine the shortest path between any location on the network and any destination. It is not anticipated there will be any interaction between the on-the-wire signalling used by the control plane. However, in some situations the I2RS system could modify information in the local databases of the control plane. This is not normally recommended, as it can bypass the normal loop free, loop free alternate, and convergence properties of the control plane. However, if the I2RS system does directly inject information into these tables, the I2RS system should ensure that loop free routing is preserved, including loop free alternates, tunnelled interfaces, virtual overlays, and other such constructions. Any information injected into the control plane directly could cause the control plane to fail to converge, resulting in a complete network outage.
To isolate I2RS transactions from other planes, it is recommended that:
When the I2RS Agent performs an action on a routing element, the action is performed via process(es) associated to a system user . In a typical UNIX system, the user is designated with a user id (uid) and belong to groups designated by group ids (gid). These users are dependent of the routing element's operation system and are designated I2RS System Users. Some implementation may use a I2RS System User for the I2RS Agent that proxies the different I2RS Client, other implementations may use I2RS System User for each different I2RS Clients.
I2RS resource may be shared with the management plane and the control plane. It is hardly possible to prevent interactions between the planes. I2RS routing system resource management is limited to the I2RS plane. As such, update of I2RS routing system outside of the I2RS plane may be remain unnoticed unless explicitly notified to the I2RS plane. Such notification is expected to trigger synchronization of the I2RS resource state within each I2RS component. This guarantees that I2RS resource are maintained in a coherent state among the I2RS plane. In addition, depending on the I2RS resource that is updated as well as the origin of the modification performed, the I2RS Authentication Authorization and Access Control policies (I2RS AAA) may be impacted. More especially, a I2RS Client is more likely to update an I2RS resources that has been updated by itself, then by the management plane for example.
This section details the I2RS Authentication and Authorization Access Policy (I2RS AAA) associated to the routing system resources. These policies only apply within the I2RS plane for I2RS users.
Applications access to routing system resource via numerous intermediaries nodes. The application communicates with an I2RS Client. In some cases, the I2RS Client is only associated to a single application, but the I2RS Client may also act as a broker. The I2RS Client, then, communicates with the I2RS Agent that may eventually access the resource.
The I2RS Client broker approach provides scalability to the I2RS architecture as it avoids that each Application be registered to the I2RS Agent. Similarly, the I2RS AAA should be able to scale numerous applications.
This results in a layered and hierarchical I2RS AAA. An application will be able to access a routing system resource only if both the I2RS Client is granted access by the I2RS Agent AAA and the application is granted access by the I2RS Client AAA.
In order to limit the number of access request that result in an error, each component should be able to retrieve the global I2RS AAA policies that applies to it. This subset of rules is designated as the "I2RS AAA component's subset policies". As they are subject to changes, a dynamic synchronization mechanism should be provided. This requirements is expressed by various sub requirements. This may be considered as a protocol security requirement when the I2RS Client and the I2RS Agent are involved. However, for completeness of the security requirements over the I2RS environment, they are are still listed below.
Similarly, for the application
I2RS AAA should be appropriately be balanced between the I2RS Client and the I2RS Agent which can be illustrated by two extreme cases:
In order to keep the I2RS AAA architecture as distributed as possible,
The I2RS Agent AAA restricts the routing system resource access to authorized components. Possible access policies may be none, read or write. The component represents the one originating the access request. The origin of the query is always an application. However, the I2RS Agent may not be able to authenticate the application as the I2RS Client may act as a broker. Similarly, multiple I2RS Agents may be used, and different access privilege may be provided depending on the I2RS Agent used. As a result, the origin of the query may be represented in multiple ways, and each way be may associated to a specific AAA. In some cases, the origin of the I2RS query is only represented by the I2RS Client, and the I2RS Agent does not have any means to associate the request to an application. In some cases, the I2RS Agent may identify the application by the I2RS Client or via other means. In addition, there is not a single way to represent an I2RS Client, and multiple identities may be used (FQDN, public key, certificates)
The I2RS Agent AAA may evolve over time as resource may also be updated outside the I2RS plane. Similarly, a given resource may be accessed by multiple I2RS users within the I2RS plane. Although this is considered as an error, depending on the I2RS Client that performed the update, the I2RS may accept or refuse to overwrite the routing system resource.
The I2RS Client AAA works similarly to the I2RS Agent AAA. The main difference is that components are applications. As a result,
In case, no authentication mechanisms have being provided between the I2RS Client and the application, then I2RS Client may not act as broker, and be instead dedicated to a single application. By doing so, application authentication may rely on the I2RS authentication mechanisms between the I2RS Client and the I2RS Agent. On the other hand, although this is not recommended, the I2RS AAA is only enforced by the I2RS Agent AAA.
I2RS AAA enforcement should not remain local, and the security domain resulting from this enforcement must be extended throughout the network. More specifically I2RS AAA policies enforced on one point remain reliable for another point as long as the communication between the two points is reliable too. This means communications should remain:
These characteristics are mostly the goal of a security transport layer. As such:
Communication is considered available if and only if all components are available as well as the communication channel itself. In order to maintain it available here are the considered aspects:
Protocols used to communicate between components should not provide means that would result in a component's resource exhaustion.
If non secure transport layer is used, when possible, protocols that do not implement any mechanisms to check the origin reachability should be avoided (like UDP). Instead, if possible, protocols like TCP or SCTP with origin reachability verification should be preferred.
Anti DoS mechanisms should also be considered at other layers including the application layer. In our case the application layer may be the I2RS protocols itself or the applications that are using the I2RS protocol. More specifically, it should be avoided to perform actions that generate heavy computation on a component. At least the component should be able to post-pone and re-schedule the action. Similarly, DoS by amplification should be avoided, and special attention should be given to small access request that generate massive network traffic without any control. An example of asymmetric dialogue could be the subscription of information streams like prefix announcement from OSPF. In addition, some service may also provide the ability to redirect these streams to a third party. In the case of information stream, registration by an I2RS Client may provide the possibility to redirect the stream on a shared directory, so it can be accessed by multiple I2RS Clients, while not flooding the network. In this case, special attention should be provided so the shared directory can agree based on its available resources the service subscription by the I2RS Client. Otherwise, the shared directory may become overloaded.
Components should be able to control the computing resource they allocate to each other components, or each actions. Based on available resource, requests should be differed, or returned an error.
One alternative way to mitigate a DoS attack or event is to limit the damages when resource exhaustion happens. This can be done by appropriately group or ungroup applications. For example, critical applications may not share their I2RS Client with multiple other Applications. This limits the probability of I2RS Client failure for the critical application. Similarly, I2RS Agent may also be selective regarding their I2RS Client as well as to the scope of their routing system resources. In fact some, some I2RS Client may be less trusted than others and some routing system resource access may be more sensitive than the others. Note that trust of an I2RS Client is orthogonal to authentication and rather involves, for example, the quality of the hosted Applications.
Even though this should be considered, it does not address the high availability issue. In order to reduce the impact of a single I2RS Client failure, remote applications may load balance their access request against multiple I2RS Clients. Non remote I2RS Client or I2RS Agent are bound the system hosting the application or to the routing element. This makes high availability be provided by the system, and thus implementation dependent.
Section 2.2 of [I-D.hares-i2rs-auth-trans] provides requirements to establish a secure communication between the I2RS Agent and the I2RS Client. These requirements can be generalized to any I2RS communications within the I2RS plane. This may include for example a remote application connected to the I2RS Client.
A key aspect of the I2RS architecture is the network oriented application. As these application are supposed to be independent, controlled by independent and various tenants. In addition to independent logic, these applications may be malicious. Then, these applications introduce also programmability which results in fast network settings.
The I2RS architecture should remain robust to these applications and make sure an application does not impact the other applications. This section discusses both security aspects related to programmability as well as application isolation in the I2RS architecture.
I2RS provides a programmatic interface in and out of the Internet routing system. This feature, in addition to the global network view provided by the centralized architecture comes with a few advantages in term of security.
The use of automation reduces configuration errors. In addition, this interface enables fast network reconfiguration. Agility provides a key advantage in term of deployment as side effect configuration may be easily addressed. Finally, it also provides facilities to monitor and mitigate an attack when the network is under attack.
On the other hand programmability also comes with a few drawbacks. First, applications can belong to multiple tenants with different objectives. This absence of coordination may result in unstable routing configurations such as oscillations between network configurations, and creation of loops for example. A typical example would be an application monitoring a state and changing its state. If another application performs the reverse operation, the routing system may become unstable. Data and application isolation is expected to prevent such situations to happen, however, to guarantee the network stability, constant monitoring and error detection are recommended to be activated.
Requirements for robustness to Dos Attacks have been addressed in the Communication channel section [I-D.ietf-i2rs-architecture].
The I2RS interface is used by application to interact with the routing states. As the I2RS Agent is shared between multiple applications, one application can prevent an application by performing DoS or DDoS attacks on the I2RS Agent or on the network. DoS attack targeting the I2RS Agent would consist in providing requests that keep the I2RS Agent busy for a long time. This may involve heavy computation by the I2RS Agent for example to blocking operations like disk access. In addition, DoS attacks targeting the network may use specific commands like monitoring stream over the network. Then, DoS attack may be also targeting the application directly by performing reflection attacks. Such an attack could be performed by indicating the target application as the target for some information like the listing of the RIB. Reflection may be performed at various levels and can be based on the use of UDP or at the service level like redirection of information to a specific repository.
Requirements for Application Control have been addressed in the I2RS plane isolation as well as in the trusted Communication Channel sections.
Applications use the I2RS interface in order to update the routing system. These updates may be driven by behavior on the forwarding plane or any external behaviors. In this case, correlating observation to the I2RS traffic may enable to derive the application logic. Once the application logic has been derived, a malicious application may generate traffic or any event in the network in order to activate the alternate application.
The whole document is about security.
We would like to thanks Russ White for its review and editorial contributions.
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, March 1997. |
| [I-D.ietf-i2rs-architecture] | Atlas, A., Halpern, J., Hares, S., Ward, D. and T. Nadeau, "An Architecture for the Interface to the Routing System", Internet-Draft draft-ietf-i2rs-architecture-09, March 2015. |
| [I-D.hares-i2rs-auth-trans] | Hares, S., "I2RS Security Related Requirements", Internet-Draft draft-hares-i2rs-auth-trans-03, June 2015. |